6.

GESTORES ESPONTÁNEOS DE FONDOS DE ALTO

RIESGO

El riesgo es una realidad fundamental en la vida de las personas

pobres, que a menudo tienen pequeños comercios o granjas, o trabajan
como jornaleros sin estar asegurados y sin empleos estables. Con estas
vidas, un mal tropiezo puede tener consecuencias desastrosas.
En el verano de 2008, Ibu Tina vivía con su madre discapacitada, sus
dos hermanos y sus cuatro hijos, con edades comprendidas entre los
tres y los diecinueve años, en una casa diminuta de Cica Das, el gran
suburbio urbano de Bandung, Indonesia. Los tres hijos pequeños iban a
la escuela —al menos oficialmente—, pero el mayor la había dejado. Los
dos hermanos solteros de Ibu Tina, un trabajador de la construcción
con contratos diarios como jornalero y un taxista, evitaban que la
familia se arruinase, pero nunca había suficiente dinero para las
matrículas escolares, comida, ropa para los niños y cuidados para su
madre anciana.
Sin embargo, la vida de Ibu Tina no había sido siempre así. Cuando
era joven trabajó en una fábrica textil. Después de casarse se incorporó
al negocio textil de su marido. Tenían cuatro empleados y el negocio iba
bien. Sus problemas comenzaron cuando alguien a quien conocían por
razones de trabajo y en quien confiaban les dio un cheque falso por
valor de 20 millones de rupias (3750 dólares PPC). Acudieron a la
policía, que les pidió 2,5 millones de rupias como soborno solo para
empezar a investigar el caso. Después de cobrar, la policía consiguió
arrestar al defraudador. Tras pasar una semana en la cárcel, quedó libre
después de prometer que devolvería lo que debía. Una vez que hubo
reembolsado 4 millones de rupias a Ibu Tina (de los que la policía pidió
2) y tras prometer que devolvería el resto más adelante, el defraudador
desapareció y nunca más se volvió a saber de él. Ibu Tina y su marido
habían pagado 4,5 millones de rupias en sobornos para recuperar 4.
 Durante los tres o cuatro años siguientes trabajaron duramente para
intentar que el negocio se recuperase y, finalmente, se las arreglaron
para conseguir un préstamo de 15 millones de rupias (2800 dólares PPC)
de PUKK, un programa gubernamental de préstamos. Utilizaron el
dinero para iniciar un negocio de intermediación comercial de tejidos.
Uno de sus primeros pedidos importantes fue de pantalones.
Compraron los pantalones a las fábricas textiles y los plancharon y
empaquetaron para la venta, pero entonces los comerciantes minoristas
se volvieron atrás, dejándolos con miles de pantalones que nadie quiso.
La secuencia de desastres supuso un gran estrés para el matrimonio
que, tras el segundo contratiempo, se separó. Ibu Tina se mudó junto a
su madre, y se llevó a los cuatro niños y montones de pantalones.
Cuando la conocimos todavía estaba intentando recuperarse del trauma
y decía que, en realidad, no tenía la energía necesaria para empezar de
nuevo. Pensaba que cuando se encontrase mejor podría abrir una
pequeña tienda de comestibles en una parte de la casa de su madre y que
quizá podría vender algunos de los pantalones durante la fiesta
musulmana de Idur Fitri.
Para empeorar las cosas, su hija mayor necesitaba mucha atención ya
que cuatro años antes, cuando tenía quince, había sido raptada por un
vagabundo que vivía cerca de su casa. La había dejado en libertad unos
días después, pero la chica estaba traumatizada por el suceso y desde
entonces no había salido de casa, incapaz de ir ni a la escuela ni a
trabajar.
¿Tenía Ibu Tina una especial mala suerte? Hasta cierto punto sí.
Pensaba que el secuestro de su hija había sido un accidente anormal
(aunque eso estaba relacionado con el hecho de que residían cerca de las
vías de ferrocarril, donde vivían muchos vagabundos), pero también
creía firmemente que sus desgracias en los negocios eran sintomáticas
de las vidas de los pequeños comerciantes.

LOS RIESGOS DE SER POBRE

Un amigo nuestro del mundo de las altas finanzas siempre dice que los
pobres son como gestores de hedge funds o fondos de alto riesgo,
porque viven con cantidades enormes de riesgo. La única diferencia se
encuentra en sus niveles de renta. De hecho, nuestro amigo subestima
tremendamente el caso, pues ningún gestor de fondos de alto riesgo es
responsable del cien por cien de sus pérdidas, al contrario de lo que les
ocurre a casi todos los dueños de pequeños negocios y a los pequeños
agricultores. Además, para conseguir todo el capital para sus negocios,
los pobres a menudo tienen que recurrir a la «riqueza» acumulada por
sus familias o a préstamos de algún tipo, circunstancias a las que nunca
se enfrenta la mayoría de los gestores de fondos de alto riesgo.
Una gran proporción de personas pobres son pequeños
comerciantes o agricultores. En nuestra base de datos de dieciocho
países, un promedio del 50 por ciento de los pobres de zonas urbanas
tienen algún negocio no agrícola, mientras que la proporción de los
pobres de zonas rurales que dicen tener un negocio agrícola varía entre
el 25 y el 98 por ciento (la única excepción es Suráfrica, donde la
población negra fue excluida históricamente de la agricultura).
Adicionalmente, una fracción significativa de estos hogares también
lleva un negocio no agrícola. Además, la mayor parte de la tierra que
cultivan los pobres no es de regadío. Esto hace que los ingresos
agrícolas dependan mucho del tiempo, y una sequía, o incluso un
retraso de las lluvias, pueden causar la reducción de la cosecha en las
tierras de secano, con lo que podría esfumarse la mitad de los ingresos
del año.
Los dueños de negocios o de granjas no son los únicos cuyos
ingresos están expuestos al riesgo. Para los pobres, la otra forma de
empleo principal es el trabajo como jornaleros; entre los más pobres de
las zonas rurales, más de la mitad de quienes trabajan lo hacen a cambio
de un jornal diario. En las zonas urbanas la proporción se acerca al 40
por ciento. Cuando los jornaleros tienen suerte, encuentran trabajos
que duran varias semanas, o incluso varios meses, en obras de
construcción o en granjas; pero a menudo el empleo puede ser nada más
que para unos días o para un par de semanas. Nunca saben si tendrán
trabajo cuando acaben el empleo actual. Si la empresa tiene algún
problema, estos oficios son los primeros que desaparecen. Pak Solhin,
presentado en el capítulo segundo, no tardó en perder su puesto de
trabajo cuando el precio de los fertilizantes y del gasóleo subió y las
explotaciones agrícolas recortaron el número de trabajadores. Por eso,
los jornaleros tienden a trabajar menos días al año en comparación con
el resto de trabajadores y muchos de ellos apenas trabajan unos pocos
días al cabo de un año. Una encuesta hecha en Gujarat, en la India,
descubrió que los jornaleros trabajaban un promedio de 254 días al año
(mientras los asalariados trabajaban 354 y los autónomos 338) y que la
tercera parte que menos trabajaba lo hacía solamente 137 días[1].
Las grandes catástrofes atraen por sí solas la atención de los medios
de comunicación. Así ocurrió con la sequía de Bangladesh de 1974
(cuando los ingresos cayeron a la mitad en términos de paridad de
poder de compra y cuando, según algunas estimaciones, murieron hasta
un millón de personas)[2] y con las crisis alimentarias de África o la
sequía de 2005-2006 en Níger. Pero incluso en temporadas normales,
los ingresos agrícolas varían tremendamente de un año a otro. En
Bangladesh, en un año cualquiera considerado normal, los salarios
agrícolas podrían situarse un 18 por ciento por encima o por debajo de
sus niveles medios[3]. Cuanto más pobre es un país, mayor es la
inestabilidad. Por ejemplo, la variabilidad de los ingresos agrícolas en la
India multiplica por 21 la variabilidad de Estados Unidos[4], lo que no
debe sorprendernos, pues los agricultores estadounidenses están
asegurados, reciben subvenciones y se benefician de los programas
habituales de seguridad social; no necesitan despedir a sus trabajadores
o recortar los salarios cuando tienen una mala cosecha.
Como si los caprichos de los elementos no fueran suficientemente
malos, los precios agrícolas fluctúan enormemente. Entre 2005 y 2008
se produjo un incremento sin precedente del precio de los alimentos.
Durante la crisis financiera global se desplomaron, pero durante los
últimos dos años crecieron de nuevo, recuperando los niveles anteriores
a la crisis. En principio, los altos precios de los alimentos deberían
favorecer a los productores (los pobres de las zonas rurales) y
perjudicar a los consumidores (los pobres urbanos). Sin embargo, en el
verano de 2008, un año de récord en los precios tanto de alimentos
como de fertilizantes, todas las personas con las que hablamos en países
como la India e Indonesia pensaban que les estaba tocando la peor
parte: los agricultores pensaban que sus costes habían crecido más que
sus precios; los jornaleros se quejaban al no poder encontrar trabajo
porque los agricultores estaban ahorrando; al mismo tiempo, la gente
que vivía en las ciudades tenía dificultades para pagar la comida. El
problema no solo eran los precios, sino también la incertidumbre. Por
ejemplo, los agricultores, que pagaban precios elevados por los
fertilizantes, no tenían la seguridad de que el precio de sus productos
continuase siendo alto cuando llegara el tiempo de la cosecha.
Para las personas pobres, el riesgo no se limita a los ingresos o a la
comida; la salud, que se ha abordado en un capítulo anterior, es una
fuente fundamental de riesgo. También está la violencia política, la
delincuencia (como en el caso de la hija de Ibu Tina) y la corrupción.

El riesgo en la vida diaria de los pobres es de tal magnitud que,

paradójicamente, sucesos que tendrían tintes catastróficos en los países
ricos con frecuencia apenas ni les afectan. En febrero de 2009, Robert
Zoellick, el presidente del Banco Mundial, advirtió a los líderes
mundiales: «La crisis económica mundial [propagada por la caída de
Lehman Brothers en septiembre de 2008] amenaza con convertirse en
una crisis humana en muchos países en desarrollo, salvo que puedan
tomar medidas específicas dirigidas a proteger a las personas vulnerables
de sus comunidades. Mientras una gran parte del mundo está
preocupada por los rescates bancarios y por las medidas de estímulo, no
deberíamos olvidarnos de que las personas pobres de los países en
desarrollo están mucho más expuestas si sus economías se
tambalean»[5]. La nota del Banco Mundial sobre esta cuestión añadía
que, por culpa de la caída de la demanda global, los pobres perderían
mercados para sus productos agrícolas, empleos temporales en obras de
construcción y puestos de trabajo en las fábricas. Bajo la presión
simultánea de la caída de ingresos fiscales y del desplome en la ayuda
internacional, se recortarían los presupuestos de los gobiernos para
colegios, centros de salud y programas de emergencia.
En enero de 2009 viajamos con Somini Sengupta, la entonces
corresponsal del New York Times en la India, a Maldah, un distrito rural
de Bengala Occidental. Ella quería escribir un reportaje sobre las
consecuencias de la crisis global para los pobres. Sengupta, que creció
en California pero que habla bengalí perfectamente, sabía que muchos
jornaleros de la construcción de Delhi eran de Maldah y que las obras
estaban ralentizándose. Así, fuimos de pueblo en pueblo preguntando a
los hombres jóvenes acerca de sus experiencias de la emigración.
 Todo el mundo conocía a alguien que había emigrado. Incluso
muchos de los emigrantes estaban allí, pasando en casa el mes de
Muharram, que secundan muchos indios musulmanes. A todos les
encantaba hablar de sus experiencias migratorias. Las madres nos
hablaban de ciudades lejanas, al sur o al norte de la India, sitios como
Ludhiana, Coimbatore y Baroda, donde sus hijos y sobrinos vivían y
trabajaban ahora. Lógicamente, había algunas tragedias —una madre
habló de un hijo suyo que había muerto en Delhi de una enfermedad
misteriosa—, pero el tono era abrumadoramente optimista. «¿Hay
trabajo en la ciudad?», preguntaba Sengupta. Sí, mucho. «¿Ha oído
hablar de despidos?». No, no hay despidos en Mumbai, las cosas van
estupendamente. Y así sucesivamente. Fuimos a la estación de
ferrocarril para ver si alguien había vuelto después de perder su empleo
y allí saludamos a tres hombres que iban de camino a Mumbai. Uno de
ellos nunca había estado allí, mientras que los otros dos, veteranos, le
aseguraban que encontrar trabajo no era ningún problema. Al final,
Sengupta nunca escribió el reportaje sobre cómo había afectado la crisis
global a la gente pobre.
Lo importante no es que se perdieran o no puestos de trabajo en la
construcción por culpa de la crisis —algunos seguro que se habían
perdido—, sino que, para la mayoría de estos hombres, lo importante
en aquellos momentos era la oportunidad. Todavía podían encontrarse
trabajos, empleos cuya remuneración superaba el doble de lo que
podían ganar en el pueblo. En comparación con lo que habían pasado —
la ansiedad habitual de no encontrar ningún trabajo, la espera
aparentemente interminable a que llegaran las lluvias—, la vida de un
trabajador de la construcción emigrante todavía parecía bastante
atractiva.
Evidentemente, la crisis global incrementó el riesgo para los pobres,
pero fue poco lo que añadió al riesgo total con el que tienen que vivir a
diario, incluso cuando en el Banco Mundial no hay crisis de la que
preocuparse. Durante la crisis de 1998 en Indonesia, la rupia perdió el
75 por ciento de su valor, los precios de los alimentos crecieron un 250
por ciento y el PIB se redujo en un 12 por ciento, pero los productores
de arroz, que tienden a estar entre la población más pobre, en este caso
salieron ganando en términos de poder adquisitivo[6]. Quienes
empeoraron fueron los empleados públicos y otras personas con
ingresos fijos. Incluso en 1997-1998, el año de la gran crisis financiera
en Tailandia, cuando la economía decreció en un 10 por ciento, dos
terceras partes de las casi mil personas encuestadas dijeron que la razón
principal de la caída de sus ingresos había sido una sequía[7]. Solo el 26
por ciento mencionó la pérdida de empleo y, seguramente, no todos los
puestos destruidos lo fueron como consecuencia de la crisis. Para la
mayor parte de los pobres, una vez más, parece que las cosas no fueron
mucho peor comparadas con cualquier otro año, precisamente porque
su situación siempre es bastante mala. Los pobres se enfrentaban a
problemas a los que están muy acostumbrados, pues todos los años
tienen la impresión de estar en medio de una enorme crisis financiera.
La gente pobre no solo tiene vidas más arriesgadas que la gente
menos pobre, sino que un determinado tropiezo de idéntica magnitud
los daña más. En primer lugar, reducir el consumo es más doloroso para
alguien que ya consume muy poco. Cuando se necesita ajustar el
consumo en un hogar que no es tan pobre, sus miembros pueden
reducir algo las llamadas de teléfono móvil, comprar carne con menos
frecuencia o mandar a los hijos a un internado menos caro. Aunque no
sea fácil. Pero para los pobres un ajuste del consumo puede significar el
recorte de gastos esenciales. En la encuesta que hicimos en la zona rural
de Udaipur, los adultos del 45 por ciento de los hogares más pobres
habían tenido que reducir la cantidad de comida en algún momento del
último año. Y las personas pobres detestan reducir la comida: los
encuestados que se habían visto obligados a comer menos afirmaron
sentirse mucho menos felices que quienes no habían necesitado hacerlo.
En segundo lugar, cuando la relación entre los ingresos actuales y
los ingresos futuros tiene forma de S, el efecto sobre los pobres de un
determinado perjuicio puede llegar a ser mucho peor que la infelicidad
temporal. En la Figura 1 se ha representado la relación entre los
ingresos actuales y los ingresos futuros de Ibu Tina, la empresaria
indonesia.
 FIGURA 1. El efecto de un shock en la fortuna de Ibu Tina

En el capítulo primero se explicó que la posibilidad de la trampa de

la pobreza se produce cuando las inversiones tienen un rendimiento
relativamente bajo para aquellos que pueden invertir poco, y es mayor
para quien pueda invertir lo suficiente. Ibu Tina estaba claramente en la
segunda situación. En su caso, la relación entre ingresos actuales y
futuros tenía forma de S porque su negocio necesitaba un tamaño
mínimo para ser rentable (en el capítulo noveno se verá que esta es una
característica central de los negocios que llevan los pobres, por lo que
su caso no es único). Antes del fraude, ella y su marido tenían cuatro
empleados y el dinero suficiente para comprar materias primas y poner
a sus máquinas y a sus trabajadores a producir prendas de vestir. El
negocio era rentable. Después, todo lo que podían hacer era comprar
pantalones ya hechos y empaquetarlos, una actividad mucho menos
rentable (o en absoluto rentable). Antes del desastre del cheque
rechazado, Ibu Tina y su marido estaban fuera de la zona de la trampa
de pobreza. Si se analiza su recorrido a lo largo del tiempo, se observa
que estaban en una trayectoria que los llevaría finalmente a obtener
unos ingresos decentes. Pero el fraude acabó con todos sus activos, lo
que tuvo como efecto que fueran a parar a la zona de la trampa de
pobreza. A partir de ese momento ganaban tan poco dinero que se iban
empobreciendo cada vez más y, cuando la conocimos, Ibu Tina vivía de
la caridad de sus hermanos. Un mal tropiezo en este mundo en forma
de S puede tener consecuencias permanentes. Cuando la relación entre
los ingresos actuales y los ingresos futuros tiene forma de S, una familia
puede precipitarse desde el camino que lleva a la clase media hasta una
situación de pobreza permanente.
Este proceso a menudo está reforzado por otro de tipo psicológico.
La pérdida de esperanzas y la sensación de que no hay una salida fácil
hace mucho más difícil que uno tenga el autocontrol necesario para
recuperar el camino perdido. Esto se vio en el capítulo segundo con Pak
Solhin, que había sido trabajador agrícola y ahora es pescador, y
también con Ibu Tina. No parecían tener las condiciones psicológicas
necesarias para recomponerse y volver a empezar. En Udaipur
conocimos a un hombre que, respondiendo a una pregunta estándar de
una encuesta, afirmó que había estado tan «preocupado, tenso o
ansioso» que durante más de un mes sus actividades normales como
dormir, trabajar y comer se habían visto afectadas. Le preguntamos la
razón y nos contestó que su camello había muerto y que había estado
llorando y nervioso desde entonces. Continuamos preguntándole, quizá
de forma algo ingenua, si había hecho algo para tratar esta depresión
(como hablar con un amigo, con un médico o con un curandero).
Pareció irritado: «He perdido el camello. Es lógico que esté triste. No
se puede hacer nada».
También pueden estar funcionando otros procesos psicológicos.
Enfrentarse al riesgo —no solamente al riesgo relativo a los ingresos,
sino también al peligro de muerte o de enfermedad— hace que nos
preocupemos y la preocupación genera estrés y depresión. Los síntomas
de depresión están mucho más extendidos entre las personas pobres. El
estrés hace que sea más difícil concentrarse, lo que a su vez nos
convierte en menos productivos. Cabe subrayar que hay una asociación
fuerte entre la pobreza y el nivel de cortisol, un indicador de estrés, que
produce el cuerpo. Del mismo modo, los niveles de cortisol bajan
cuando los hogares reciben algún tipo de ayuda. Por ejemplo, se ha
observado que los hijos de los beneficiarios de PROGRESA, el programa
mexicano de transferencias monetarias, tienen niveles
significativamente más bajos de cortisol que otros niños cuyas madres
no se beneficiaron del programa. Esto es importante porque el cortisol
supone un daño directo a la capacidad cognitiva y de toma de
decisiones; la liberación de cortisol inducida por el estrés afecta a zonas
del cerebro como el córtex prefrontal, las amígdalas y el hipotálamo,
que son esenciales para el funcionamiento cognitivo; el córtex
prefrontal es especialmente importante para la supresión de respuestas
impulsivas. Por tanto, es lógico que las personas que participan en
experimentos y se someten artificialmente a condiciones de estrés en un
laboratorio muestren una menor probabilidad de tomar la decisión
económica racional cuando tienen que decidir entre varias
alternativas[8].

LA COBERTURA

¿Qué pueden hacer las personas pobres para enfrentarse a estos riesgos?
Cuando se sufre una disminución de salarios o de ingresos, una
reacción natural es intentar trabajar más. Sin embargo, puede resultar
autodestructivo. Si todos los jornaleros quieren trabajar más cuando
vienen malos tiempos —por ejemplo, cuando hay una sequía o cuando
sube el precio de otros factores de producción—, compiten entre ellos,
lo que reduce los salarios. El problema se agudiza si no pueden
encontrar trabajo fuera del pueblo. El resultado es que el efecto sobre
los salarios de un mismo tipo de sequía es relativamente más negativo
en los pueblos de la India que se encuentran más aislados, donde es más
difícil para los trabajadores salir a buscar trabajo fuera. En esos lugares,
trabajar más no es necesariamente una forma efectiva de superar una
caída de los ingresos[9].
Cuando trabajar más no es una buena alternativa tras una caída de
ingresos, a menudo la mejor opción es intentar limitar el grado de
exposición al riesgo mediante la creación de una cartera diversificada,
como un gestor de fondos de alto riesgo. Es evidente que los pobres
ponen mucho ingenio en ello, con la única diferencia de que diversifican
actividades en vez de instrumentos financieros. Un hecho llamativo de
los pobres es el gran número de ocupaciones a las que se dedican los
miembros de una misma familia. En una encuesta realizada en
veintisiete pueblos de Bengala Occidental, incluso en los hogares que
decían cultivar un terreno, sus miembros dedicaban solamente el 40 por
ciento de su tiempo a la agricultura[10]. En esta encuesta, la familia tipo
tenía a tres de sus miembros trabajando en siete ocupaciones diferentes.
En términos generales, aunque la mayoría de las familias rurales están
vinculadas a la agricultura, es raro que esta sea su única ocupación.
Puede tratarse de una forma de reducción del riesgo —si una actividad
decae, otras pueden mantenerse—, aunque, como se verá más adelante,
también hay otras razones que explican esta diversificación.
Cultivar diversos terrenos en lugares diferentes del pueblo, en vez
de llevar una única tierra más grande, también permite diversificar algo
el riesgo. Cuando una plaga afecta a una parte del pueblo, es posible que
otras zonas puedan librarse de ella. Cuando falta la lluvia, es más
probable que puedan mantenerse las cosechas de las tierras más
próximas a aguas subterráneas. Resulta más sorprendente aún que zonas
distintas de un mismo pueblo puedan tener microclimas diferentes,
determinados por la pendiente, la altitud y la humedad.
La emigración temporal también puede interpretarse en este mismo
sentido. Es relativamente poco frecuente que una familia entera se
desplace a la ciudad. Lo habitual es que la mayoría se quede cuando
emigran algunos de sus miembros (que en la India y en México son
sobre todo los hombres y los adolescentes, mientras en China, Filipinas
o Tailandia son las hijas mayores). Así se consigue que el futuro de la
familia no dependa solamente del empleo de quien está en la ciudad, y
permite también mantener los contactos que la familia tiene en el
pueblo, lo que a menudo resulta útil, como se verá más adelante.
Otra forma de limitar el riesgo es actuar de forma muy
conservadora en la gestión de sus negocios o de sus tierras. Por
ejemplo, es posible que conozcan la existencia de una nueva variedad de
su cosecha principal, más productiva, pero que decidan no adoptarla.
Una ventaja de continuar con la tecnología tradicional es que a los
agricultores les basta con guardar semillas de la última cosecha para
plantar la siguiente, con lo que no necesitan comprar nuevas, que con
frecuencia significaría mucho dinero. Incluso en el caso de que las
nuevas semillas permitan recuperar con creces la inversión si las cosas
van bien, siempre hay una pequeña probabilidad de que la cosecha
fracase —por ejemplo, si no llegan las lluvias—, con lo que el agricultor
perdería la inversión adicional.
 La familia también se utiliza para repartir el riesgo mediante
fórmulas originales. En la India, los hogares agrícolas usan el
matrimonio como forma de diversificar la cartera de riesgos de su
extensa familia. Cuando una mujer se traslada al pueblo de su familia
política, después de la boda, se crea un vínculo entre el hogar del que
procede y el hogar al que se incorpora y cada familia puede recurrir a la
otra en caso de apuro[11]. Los hogares agrícolas tienden a casar a sus
hijas en pueblos que estén lo suficientemente cerca como para que se
mantenga la relación, pero lo suficientemente lejos como para que haya
un patrón climatológico ligeramente diferente. De esta forma, si llueve
en un pueblo pero no en el otro, pueden ayudarse mutuamente. Otra
forma de comprar seguridad es tener muchos hijos. Como se ha visto
anteriormente, Pak Sudarno tuvo nueve hijos para asegurarse de que al
menos uno de ellos se encargaría de cuidarle.

Todas estas formas utilizadas por los pobres para afrontar el riesgo
suelen tener costes elevados. Esto se aprecia especialmente en la
agricultura. En la India, los agricultores pobres utilizan los recursos de
forma más conservadora pero menos eficiente cuando viven en zonas
donde la lluvia es más irregular[12]. Las tasas de beneficio de los
agricultores pobres crecen hasta un 35 por ciento cuando viven en
zonas donde el patrón anual de precipitaciones es muy predecible.
Además, este tipo de riesgo solo afecta a los pobres. Para los
agricultores más ricos no se observa relación alguna entre las tasas de
beneficio y la variabilidad de las precipitaciones, lo que probablemente
se debe a que pueden permitirse perder alguna cosecha y, por
consiguiente, están dispuestos a asumir riesgos.
Otra estrategia adoptada por los agricultores pobres es firmar
contratos de aparcería que establecen que el dueño de la tierra paga una
parte del coste de producción y se atribuye una parte de la cosecha.
Esto limita la exposición al riesgo del agricultor a costa de reducir sus
incentivos; puesto que el dueño se llevará, por ejemplo, la mitad de toda
la producción, el agricultor tiene un estímulo menor para trabajar
intensamente. Un estudio hecho en la India mostró que los agricultores
dedican hasta un 20 por ciento menos de esfuerzo personal a la tierra
explotada en régimen de aparcería que a la de tierras donde tienen
derecho a toda la cosecha[13]. Como consecuencia, estos terrenos se
explotan de forma menos intensiva y menos eficiente.
Tener ocupaciones múltiples, como les ocurre a muchas personas
pobres, también resulta ineficiente. Es difícil llegar a ser especialista en
alguna cosa sin especializarse en ella. Las mujeres que se ocupan de tres
negocios diferentes, al igual que los hombres que no pueden
comprometerse con un puesto fijo en la ciudad porque quieren volver al
pueblo cada cierto número de semanas, no aprovechan la oportunidad
de adquirir habilidades y experiencia en sus ocupaciones principales. Al
dejar pasar estas oportunidades también dejan pasar los beneficios de
especializarse en aquello que realmente se les da bien.
Por tanto, el riesgo que soportan los pobres no les pasa factura
solamente cuando llega un mal momento. El miedo a que pueda ocurrir
algo malo tiene como consecuencia el deterioro de la capacidad de las
personas pobres para desarrollar plenamente su potencial.

Ayudarse mutuamente

Otra forma de gestionar el riesgo, que además es potencialmente

mucho mejor, consiste en que los vecinos se ayuden unos a otros. La
mayoría de la gente vive en pueblos o en comunidades y tienen acceso a
una red extensa de personas que los conocen bien: la familia numerosa y
la comunidad creada en torno a la religión o a la etnia. Aunque es
posible que algunos problemas afecten a todos los miembros de la red
—por ejemplo, un monzón dañino—, otros son más específicos. Si las
personas a quienes les va bien en ese momento pueden ayudar a aquellas
que están pasándolo mal, a cambio de una ayuda similar cuando se
invierten los papeles, todos pueden mejorar. Ayudarse unos a otros no
tiene que ser por caridad.
Un trabajo de Christopher Udry muestra tanto la potencia como
los límites que tiene este tipo de seguros informales. A lo largo de todo
un año, que pasó en la zona rural de Nigeria, Udry consiguió que sus
vecinos registraran cada regalo o cada préstamo informal que se dieron
unos a otros, así como los términos que regían la forma en que
devolvían esos préstamos[14]. También les preguntaba cada mes si les
había ocurrido algo malo. Descubrió que en cualquier momento la
familia media tenía una relación deudora o acreedora con un promedio
de otras 2,5 familias. Además, los términos de los préstamos se
ajustaban dependiendo de las circunstancias tanto de los prestamistas
como de los prestatarios. Cuando el prestatario tenía un problema,
devolvía una cantidad menor (con frecuencia, menor incluso que la
cuantía del préstamo original) pero cuando el que tenía una mala racha
era el prestamista, el prestatario llegaba a devolver una cantidad superior
a lo que debía. La densa red de relaciones mutuas de concesión y
obtención de préstamos tuvo una gran importancia para reducir el
riesgo que afrontaba cada individuo. Sin embargo, existían límites en lo
que se podía alcanzar con esta solidaridad informal. Las familias que
experimentaban alguna crisis reducían su nivel de consumo, incluso
cuando la suma de los ingresos totales de todos los miembros de la red
no hubiera cambiado.
Muchas investigaciones sobre los seguros informales han analizado
este fenómeno en lugares diversos, desde Costa de Marfil a Tailandia,
llegando a las mismas conclusiones: mientras que las redes de
solidaridad tradicional ayudan en momentos de crisis, el seguro
informal que proveen dista mucho de ser perfecto. Si el riesgo estuviese
asegurado de forma adecuada, una familia podría mantener siempre un
nivel de consumo más o menos igual, que vendría dado por su capacidad
media de obtención de ingresos. Así, en los buenos tiempos podrían
ayudar a otros, mientras que en los malos tiempos otros les ayudarían a
ellos. Pero no es lo que sucede habitualmente. Los problemas de salud,
concretamente, están muy mal asegurados. En Indonesia el consumo se
reduce en un 20 por ciento cuando un miembro del hogar padece una
enfermedad grave[15].
Un estudio hecho en Filipinas descubrió que la solidaridad entre los
vecinos del mismo pueblo apenas funciona en casos de enfermedades
que, siendo graves, no son fatales[16]. Cuando una familia tiene una
mala cosecha o cuando alguien pierde su empleo, otras familias del
pueblo acuden al rescate. La familia afectada recibe regalos, préstamos
sin interés y otras formas diversas de ayuda. Pero cuando los individuos
sufren un problema de salud no ocurre nada de esto. La familia se tiene
que hacer cargo de ello por sí sola.
La falta de seguros para problemas de salud resulta muy llamativa,
dado que las familias se ayudan en otros casos. En un capítulo anterior
se habló de Ibu Emptat, una mujer a quien conocimos en un pequeño
pueblo de Java y cuyo marido había tenido un problema en los ojos. Su
hijo se había visto obligado a dejar los estudios porque no podían
hacerse cargo del medicamento que necesitaba para el asma. Ibu Emptat
había recibido un préstamo de 10 000 rupias (18,75 dólares PPC) del
prestamista local para pagar la cura de los ojos de su marido y, cuando la
conocimos, la suma de su deuda más los intereses superaba el millón de
rupias. Estaba muy preocupada porque el prestamista amenazaba con
quedarse con todo lo que tenían. Sin embargo, durante la entrevista nos
contó que una de sus hijas le acababa de regalar una televisión. Se
acababa de comprar una nueva por unas 800 000 rupias (150 dólares
PPC) y decidió regalar la vieja —que aún estaba en muy buen estado— a
sus padres. Esto nos sorprendió un poco; ¿no hubiese sido más
adecuado que la hija conservase la vieja televisión y les diese el dinero a
los padres para que pudieran pagar al prestamista? Le preguntamos a
Ibu Emptat si no le podía ayudar con la deuda alguno de sus hijos y ella,
moviendo la cabeza, respondió que los hijos tenían sus propios
problemas y tenían que hacerse cargo de sus familias, dando a entender
que no debíamos cuestionar el regalo. Veía como algo normal que nadie
le ofreciese ayuda para sus gastos sanitarios.
¿Qué impide que la gente haga algo más para ayudarse unos a otros?
¿Por qué algunas formas de riesgo no están cubiertas o no están bien
cubiertas?
Hay buenas razones por las que no estamos dispuestos a ofrecer
ayuda incondicional a nuestros amigos y vecinos. Por una parte,
podemos temer que la garantía de ayuda haga sentir la tentación de
descuidarse —esto es lo que los aseguradores llaman riesgo moral—. O
que la gente pueda reclamar ayuda incluso cuando no la necesita. O
sencillamente que la promesa de ayuda recíproca pueda no cumplirse:
yo te ayudo, pero cuando llega el momento de que tú me ayudes a mí,
estás demasiado ocupado para hacerlo.
Todas estas explicaciones justificarían la preferencia por moderar un
poco la ayuda que ofrecemos, pero no la falta de ayuda para quienes
acaban de caer gravemente enfermos, pues la enfermedad no suele ser
producto de una elección. Por otra parte, cabe la posibilidad de que el
enfoque adoptado por muchos economistas para el seguro informal, es
decir, pensar en situaciones donde ayudamos a los demás porque
podemos necesitar su ayuda en el futuro, no sea capaz de explicar todo
lo que pasa. Podría ocurrir que ayudemos a nuestros vecinos in extremis
incluso cuando no tengamos expectativas de llegar a encontrarnos en su
posición —por ejemplo, cuando resulte inmoral dejar que los vecinos
mueran de hambre—. El libro de Betsy Hartmann y Jim Boyce sobre la
vida rural en Bangladesh a mediados de los setenta describe a dos
familias vecinas, una hinduista y otra musulmana, que no eran
especialmente cercanas[17]. La familia hinduista perdió a la persona que
constituía la fuente principal de ingresos y empezó a pasar hambre;
desesperada, la mujer empezó a saltar de vez en cuando la verja que
separaba las dos casas para entrar en el jardín de la otra familia y robar
algunas hojas comestibles. Hartmann descubrió que la familia
musulmana se dio cuenta de lo que pasaba pero decidió ignorarlo. El
hombre dijo: «Sé que no es una mala persona. Si yo estuviera en su
lugar, es probable que también robase. Cuando desaparecen pequeñas
cosas intento no enfadarme. Pienso que la persona que las cogió pasa
más hambre que yo».
Que la gente pueda ayudarse entre sí en los malos tiempos por un
sentido de obligación moral, más que por esperar necesariamente que se
les ayude a ellos en el futuro, contribuye a explicar por qué las redes
informales no están preparadas para hacerse cargo de los problemas de
salud. Incluso cuando un hogar muy pobre, pero que tiene suficiente
para comer, ve cómo el vecino no lo tiene, reparte lo suyo. Pero ayudar
a otros a pagar, por ejemplo, los gastos hospitalarios, exige ir más allá de
ese acto básico de compartir. Se necesitaría que contribuyesen muchos
hogares, porque la hospitalización puede resultar muy cara. Por
consiguiente, tiene sentido excluir los problemas costosos de salud del
imperativo moral de ayudar a los vecinos necesitados porque, de otro
modo, para tener éxito, se requeriría un contrato social mucho más
elaborado.
Esta perspectiva del seguro como un deber moral de ayuda a quien
la necesita explica la razón de que, en los pueblos nigerianos, los vecinos
se ayudasen unos a otros de forma individual, en lugar de contribuir
todos a una caja común, a pesar del hecho de que compartir el riesgo de
esta otra forma sería más eficiente. También puede explicar por qué la
hija de Ibu Emptat le dio una televisión a su madre, pero no se hizo
cargo de sus gastos sanitarios. No quiso ser la hija que se
responsabilizase de la atención médica de sus padres ni quiso
presuponer la generosidad de sus hermanos. Por tanto, eligió hacer algo
bueno para los padres sin morder más de lo que podía masticar.

¿DÓNDE ESTÁN LAS COMPAÑÍAS DE SEGUROS PARA LOS

POBRES?

Puesto que el coste del riesgo es muy elevado y el seguro que se puede
conseguir a través de canales informales tiene muchas limitaciones, cabe
preguntarse por qué los pobres no acceden más a los seguros oficiales,
es decir, a los ofrecidos por las compañías aseguradoras. Cualquier tipo
de seguro formal es una rareza entre los pobres. Determinados
contratos de seguro que son productos estándar entre los agricultores
de países ricos, como los seguros de salud, los seguros contra el mal
tiempo o contra la muerte del ganado, prácticamente no existen en el
mundo en desarrollo.
Ahora que todo el mundo conoce los microcréditos, los seguros
para los pobres parecen un objetivo evidente, una oportunidad para los
capitalistas creativos llenos de buenas intenciones (un editorial de la
revista Forbes lo llamó «un mercado natural no penetrado»)[18]. Los
pobres se enfrentan a un riesgo enorme y deberían estar dispuestos a
pagar una prima razonable para asegurar sus vidas, su salud, su ganado o
su cosecha. Con miles de millones de pobres a la espera de contratar
seguros, incluso si se obtiene un beneficio minúsculo por cada póliza,
podría generarse ya un gran proyecto empresarial y, a la vez, les serviría
a estos de gran ayuda. Lo único que falta es alguien que organice este
mercado. Por eso, algunas organizaciones internacionales (como el
Banco Mundial) y grandes fundaciones (como la Fundación Gates) han
invertido cientos de millones de dólares para fomentar el desarrollo de
los seguros para los pobres.
Las dificultades obvias que conlleva ofrecer seguros son numerosas
y no afectan solo a los pobres. Se trata de problemas básicos que se
amplifican en los países pobres, donde es más difícil regular a las
aseguradoras y supervisar a los asegurados. Ya se ha mencionado el
«riesgo moral», que implica cambios de conducta en las personas
(como, por ejemplo, cultivar con menos cuidado, gastar más dinero en
sanidad y cosas así) cuando saben que no tendrán que hacerse cargo de
todas las consecuencias. Veamos los problemas que acompañan a la
oferta de seguros sanitarios. Ya se ha dicho que los pobres visitan
habitualmente a diversos especialistas en temas de salud. ¿Qué harán si
las visitas empiezan a ser gratuitas? Y los médicos, ¿no tendrán
tentaciones de pedir pruebas y recetar medicinas innecesarias,
especialmente si tienen también un laboratorio (como ocurre con
frecuencia en Estados Unidos y en la India) o si reciben comisión de la
farmacia? Parece que todo lleva a lo mismo, pues los pacientes quieren
ver acción, por lo que preferirán médicos proclives a recetar, mientras
que estos pueden ganar más dinero si recetan más. En suma, la oferta de
seguros sanitarios basados en reembolsar los gastos de asistencia
ambulatoria en países donde la asistencia médica, en el mejor de los
casos, está regulada deficientemente y donde cualquiera puede
establecer una consulta «médica», parece un primer paso hacia la
quiebra.
Otra cuestión es la «selección adversa». Si el seguro no es
obligatorio, las personas con posibilidades de tener algún problema en
el futuro pueden ser también más propensas a contratarlo; lo que sería
correcto, siempre y cuando la aseguradora también lo supiera, porque
entonces lo podría incluir en el cálculo de la prima. Pero si la
aseguradora no puede identificar a quienes compran el seguro porque
necesitan atención médica en ese momento, solo le queda la opción de
aumentar la prima de todo el mundo. No obstante, elevar el precio
empeora las cosas, pues desanima a quienes saben que probablemente
no necesitarán el seguro y, con ello, se agrava el problema original. Esta
es la misma razón por la que, en Estados Unidos, a quienes no pueden
tener un seguro sanitario a través de su empresa les resulta muy difícil
poder contratarlo a precios razonables. Y es también la causa de que los
programas de seguros sanitarios asequibles suelen ser también
obligatorios, pues si todos deben participar, la aseguradora no se queda
solo con las personas de riesgo alto.
Un tercer problema es el fraude descarado. ¿Qué impide a un
hospital presentarle a la aseguradora facturas falsas o cobrarle al
paciente una cantidad de dinero muy superior al coste del tratamiento?
Y si un ganadero ha contratado un seguro para su búfalo de agua, ¿qué
le impide decir que el búfalo ha muerto? Nachiket Mor y Bindu
Ananth, de la Fundación ICICI, son las dos personas más
comprometidas con el diseño de unos mejores servicios financieros para
los pobres dentro del sector de las finanzas de la India. Con un gran
sentido del humor, nos contaron su fracaso absoluto para ofrecer, hace
ya muchos años, seguros de ganado por primera vez. Después de que
todos los individuos asegurados en la primera fase afirmasen haber
perdido su ganado, decidieron establecer como exigencia para
demostrar la muerte del animal que el dueño mostrase la oreja de la vaca
muerta. El resultado fue que surgió un sólido mercado de orejas de
vaca, pues a cualquier vaca que moría —asegurada o no— le cortaban la
oreja y después la vendían a quienes tenían la suya asegurada para que
pudieran cobrar la póliza sin perder su vaca. En el verano de 2009
fuimos a una presentación de Nandan Nilekani, fundador y antiguo
consejero delegado del gigante indio del software, Infosys, a quien el
gobierno indio había encargado la creación de un sistema de
identificación personal que permitiese darle a cada ciudadano del país su
documento nacional de identidad. Nilekani explicó su plan, señalando
que diez huellas dactilares y una fotografía del iris de los ojos son
suficientes para identificar a alguien sin lugar a duda. Mor le escuchaba
concentrado y, cuando Nilekani hizo una pausa, dijo: «Es una pena que
las vacas no tengan dedos».
Algunos riesgos deberían ser más fáciles de asegurar que otros. Por
ejemplo, veamos el caso de la meteorología. Los agricultores deberían
valorar una política que les pagara una cantidad fija, basada en la prima
que contrataron, cuando la lluvia caída en la estación meteorológica más
cercana no alcanzara un determinado nivel crítico. Dado que nadie
controla el tiempo atmosférico y que no procede hacer ninguna
valoración sobre lo que debería hacerse (al contrario que en la atención
médica, donde alguien debe decidir qué pruebas o qué tratamientos se
necesitan), no existe margen para el riesgo moral ni para el fraude.
En el ámbito de la atención sanitaria, asegurar los problemas serios
de salud, como enfermedades graves o accidentes, parece mucho más
sencillo que la cobertura de la asistencia ambulatoria. Nadie quiere
someterse a una operación o a un tratamiento de quimioterapia sin
necesidad y los tratamientos son fácilmente verificables. Sigue habiendo
riesgo de tratamientos excesivos, pero la aseguradora puede limitar lo
que pagará por cada tratamiento. Sin embargo, persiste el gran problema
de la selección, pues la compañía no quiere tener como clientes
solamente a los enfermos.
El secreto para evitar la selección adversa está en comenzar con un
gran grupo de gente cuya relación sea ajena a cuestiones de salud —los
empleados de una gran empresa, los clientes de servicios de
microcréditos, los comunistas con carné, etcétera— y en tratar de
asegurarlos a todos.
Por esta razón, muchas instituciones de microfinanza (IMF)
pensaron en ofrecer seguros médicos. Tienen un conjunto amplio de
clientes o prestatarios a los que podrían ofrecer seguros. Dado que
estos clientes, que suelen tener tasas de morosidad muy bajas, podrían
dejar de pagar si sufrían problemas graves de salud, contratar seguros
médicos con ellos también serviría, de alguna forma, para asegurar a
estas instituciones de microfinanza. Además, el cobro de las primas
sería fácil, pues el personal que gestiona los préstamos ya se reúne con
los clientes semanalmente. De hecho, la prima se podría incorporar al
préstamo.
En 2007, SKS Microfinance, que entonces era la mayor institución de
microfinanza de la India, creó Swayam Shakti, un programa piloto de
seguros sanitarios que ofrecía coberturas por maternidad,
hospitalización y accidentes. Para evitar la selección adversa, el seguro
debía ser obligatorio para los grupos a quienes se les ofrecía. Con el
objeto de evitar el posible fraude, la cuantía de cada prestación estaba
limitada y se aleccionó seriamente a los clientes para que utilizaran los
hospitales de la red con la que SKS había alcanzado acuerdos a largo
plazo. Con el fin de hacer más atractivos a estos hospitales, los
asegurados que acudiesen a ellos no tendrían que pagar por adelantado
si se trataba de una enfermedad incluida en la póliza de seguro. SKS
pagaría directamente a los hospitales.
Cuando SKS introdujo el producto por primera vez, intentó hacerlo
obligatorio para todos sus clientes, pero protestaron y la empresa
decidió que solo fuera obligatorio en la primera renovación del crédito.
Como consecuencia, algunos clientes decidieron no renovarlo y SKS
empezó a perder clientes en las zonas donde estaban ofreciendo el
seguro. Unos meses después, las tasas de renovación de créditos de SKS
cayeron desde un 60 por ciento hasta cerca del 50 por ciento. La
consejera delegada de una institución de la competencia nos preguntó
sobre nuestro trabajo con SKS y, cuando le dijimos que estábamos
evaluando el impacto de la oferta de seguros médicos obligatorios a los
clientes de microcréditos, se rio y dijo: «Yo sé cuál es el efecto. En
todos los sitios donde SKS ha hecho este producto obligatorio, nuestros
clientes han aumentado. ¡La gente está dejando SKS para venir a nuestra
empresa!». Cerca de una cuarta parte de los clientes que querían
continuar pidiendo créditos a SKS, pero evitar el seguro médico,
encontraron una vía de escape legal: empezaron a pagar el crédito
anticipadamente, antes de que terminase el año de vigencia de la póliza.
De este modo, cuando renovaban su crédito, técnicamente estaban
cubiertos todavía por el seguro y, por tanto, no tenían que pagar la
nueva póliza. Ante esta resistencia, SKS decidió que el producto fuera
voluntario. Pero un seguro voluntario contratado solamente por
algunos clientes es susceptible de generar selección adversa y riesgo
moral. Las pólizas por cliente se dispararon y la empresa
ICICI Lombard, en cuya representación ofrecía el producto SKS,
concluyó que estaba perdiendo dinero y pidió a SKS que dejase de
asegurar a clientes nuevos. Otras organizaciones que intentaron poner
en marcha iniciativas similares han tropezado con obstáculos muy
parecidos en cuanto a la resistencia de los clientes a contratar
obligatoriamente el seguro.
Los seguros médicos vinculados a microcréditos no son los únicos
que han tenido problemas. Un grupo de investigadores entre los que
está Robert Townsend, compañero nuestro en el MIT, intentó medir los
efectos de un sistema muy sencillo de cobertura por riesgos
meteorológicos. Muy parecido al descrito anteriormente, el sistema
consiste en el pago de una cierta cantidad de dinero cuando la lluvia no
alcanza un mínimo determinado[19]. El producto se comercializó en dos
regiones de la India, Gujarat y Andhra Pradesh, secas y propensas a
sufrir sequías. En ambos casos se vendía a través de una organización de
microcréditos muy conocida y prestigiosa. La organización probó varias
formas de oferta y presentó el seguro a los agricultores. En general, la
contratación fue muy escasa, pues en el mejor de los casos solo lo
compró el 20 por ciento de los agricultores, y ese nivel solo se alcanzó
cuando alguien de esa IMF fue vendiendo puerta a puerta. Además,
incluso entre quienes contrataron el seguro, las cantidades vendidas
fueron muy pequeñas, pues la mayoría compraron pólizas que cubrían
solamente entre un 2 y un 3 por ciento de lo que serían sus pérdidas en
caso de falta de agua de lluvia.

¿Por qué no quieren seguros los pobres?

Una primera explicación a la escasa demanda de seguros puede ser que

el gobierno haya arruinado el mercado. Este es el argumento tradicional
de los wallahs de la demanda: cuando no funcionan los mercados, lo
más probable es que la culpa sea de una sobreoferta del gobierno o de
las instituciones internacionales. El argumento concreto viene a decir
que cuando ocurre una desgracia esas almas amables acuden al socorro
y, como resultado, la gente ya no necesita seguros.
Es cierto que en años de monzones malos los estados de la India
compiten para ser designados «zonas afectadas por la sequía» porque
eso les abre la puerta a la ayuda gubernamental. Se crean empleos en
obras públicas, se distribuye comida y así sucesivamente. Pero hay que
dejar claro que esto no es más que una pequeña parte de las necesidades
de los pobres. Por un lado, el gobierno interviene solamente ante
desastres de gran magnitud y no cuando muere un búfalo o cuando un
coche atropella a alguien. Y en la mayoría de los casos, las ayudas tras
un desastre, cuando llegan finalmente a sus destinatarios pobres,
resultan muy insuficientes.
También es posible que los pobres no comprendan del todo bien el
concepto de seguro. Ciertamente, un contrato de seguro es diferente a
la mayoría de las transacciones que suelen hacer. Se trata de pagar un
dinero confiando en que uno nunca se vea obligado a utilizarlo. Cuando
hablamos con clientes de SKS conocimos a mucha gente que se había
disgustado al ver que no les reembolsaban sus pólizas de seguro médico,
a pesar de que no habían pasado ningún parte durante el año anterior.
Seguramente se podría explicar mejor el concepto de seguro, pero es
difícil imaginarse que personas que habían encontrado una ingeniosa vía
de escape legal en el sistema de SKS no hayan podido entender su
principio básico. Townsend, como parte de su esfuerzo para vender las
coberturas para riesgos meteorológicos, llevó a cabo un ejercicio para
saber si la gente entiende cómo funciona el seguro. Durante la visita al
agricultor, el comercial leía en voz alta una breve descripción de una
cobertura hipotética (seguro contra cambios de temperatura), después
describía varias situaciones hipotéticas y le preguntaba al potencial
cliente en qué casos se le abonaría la póliza. Tres cuartas partes de las
respuestas fueron acertadas. No está claro que un ciudadano medio de
Estados Unidos o de España lo hiciese mucho mejor. Por consiguiente,
no es extraño que el intento de explicar mejor el seguro por riegos
meteorológicos haya carecido de efectos sobre la disposición de los
agricultores a contratarlo[20].
Fueron capaces de entender el concepto básico del seguro y su
funcionamiento, pero no estaban interesados en comprarlo. No
obstante, algunos detalles relativamente menores les hicieron dudar en
su elección. Una simple visita a su casa, sin un esfuerzo de marketing
especial, hace que se cuadruplique la proporción de personas que
contratan la cobertura por riesgos meteorológicos. En Filipinas, los
hogares que fueron seleccionados al azar para responder a una encuesta
inicial con muchas preguntas sobre salud acabaron siendo mucho más
propensos a suscribir el seguro médico que otro grupo de hogares
comparables en los que no se había realizado la encuesta. Posiblemente,
responder a esas preguntas sobre posibles problemas de salud les había
servido de recordatorio de las cosas que pueden pasar[21].
Y si se están jugando tanto, ¿por qué las personas pobres no son
más entusiastas ante las ventajas de estar aseguradas, sin necesidad de
estos pequeños empujones?
El problema fundamental, en nuestra opinión, es que por culpa de
los problemas mencionados anteriormente el tipo de seguro que puede
ofrecer el mercado solamente los cubre frente a escenarios de
catástrofe, lo que, a su vez, suscita varias cuestiones. En primer lugar, la
credibilidad siempre es problemática con los seguros. Dado que el
contrato exige el pago adelantado por parte del hogar y que el futuro
reembolso depende de la aseguradora, el cliente debe confiar totalmente
en ella. En el caso de la cobertura frente a riesgos meteorológicos, el
equipo que comercializaba el producto a veces fue acompañado de
Basix, una organización bien conocida por los agricultores, y otras veces
fue por su cuenta. Descubrieron que la presencia de un miembro de
Basix incidía bastante en las ventas, lo que da a entender que la
confianza es clave.
Por desgracia, la falta de credibilidad puede ser un problema
endémico debido a la naturaleza de estos productos y a la forma en que
reaccionan las aseguradoras a cualquier posibilidad de fraude. En el
invierno de 2009 visitamos algunos de los clientes de SKS que habían
decidido no renovar sus seguros médicos. Una mujer dijo que había
tomado esa decisión después de que SKS se negase a reembolsarle
cuando fue al hospital con una infección estomacal. La póliza solo
cubría daños catastróficos y una infección estomacal no estaba incluida,
por muy dolorosa que fuera. Pero no quedó claro que la mujer
entendiese la diferencia —al fin y al cabo, había ido al hospital y la
habían atendido—. También nos habló de una mujer que pertenecía a
otro grupo (al igual que hace la mayoría de las IMF, SKS organiza a sus
clientes por grupos) cuyo marido murió de una infección grave, no sin
antes haberse tenido que gastar bastante dinero en médicos y
tratamientos. Tras su fallecimiento, la viuda envió las facturas a la
aseguradora y esta rehusó reembolsárselas argumentando que no había
tenido que pasar ninguna noche hospitalizado. Un grupo de mujeres,
horrorizadas con este caso, decidieron dejar de pagar la prima. Desde
un punto de vista estrictamente legal, la aseguradora estaba en su
derecho de rehusar el pago pero, por otra parte, ¿qué puede resultar
más catastrófico?
Los seguros frente a riesgos meteorológicos comparten muchos de
estos problemas. La sequía puede haber destruido la cosecha y los
agricultores pueden estar hambrientos, pero si la lluvia superó el
mínimo establecido para la temporada de lluvias, nadie recibirá pago
alguno en la zona. El problema es que hay muchos microclimas y
cualquier año en el que el promedio de lluvias esté justo por encima del
límite, muchos agricultores individuales deben afrontar condiciones
similares a las de sequía, aunque solamente sea por las leyes de la
probabilidad. Para los afectados no va a ser fácil aceptar el veredicto de
la estación meteorológica, especialmente en un contexto que no es
ajeno a la corrupción.
La segunda cuestión viene dada por el problema de la inconsistencia
temporal, que ya se ha tratado en el capítulo dedicado a la salud. Al
tomar la decisión sobre si comprar o no el seguro, se piensa en el
presente, que es cuando se debe pagar la prima, mientras que la
contraprestación, en su caso, tendría lugar en el futuro. Ya se ha visto
que este es un razonamiento que los seres humanos hacen bastante mal.
El problema se agrava cuando el seguro cubre una catástrofe. La
contraprestación no solamente se recibiría en el futuro, sino en un
futuro especialmente desagradable en el que nadie quiere pensar. Pensar
poco en que puedan ocurrir sucesos de este tipo puede ser una reacción
natural de protección y podría explicar por qué la gente es más
propensa a contratar seguros después de que se le fuerce a pensar en
ello, por ejemplo, al responder a una encuesta.

Todas estas consideraciones evidencian que los microseguros no van a

convertirse en la próxima oportunidad de mercado con miles de
millones de clientes. Parece haber razones de peso por las que casi nadie
está satisfecho con las coberturas que ofrece el mercado. Por otra parte,
las personas pobres tienen niveles de riesgo claramente inaceptables.
Por consiguiente, existe un margen claro para la acción del
gobierno, lo que no significa que tenga que ser el sustituto de un
mercado de seguros privado, pero si se quiere que haya alguna
posibilidad de que surja un verdadero mercado, el gobierno
probablemente tenga que intervenir. Las empresas privadas podrían
seguir vendiendo exactamente el mismo tipo de productos que están
ofreciendo ahora (riesgos catastróficos con límites estrictos, coberturas
para riesgos meteorológicos indexados, etcétera). Pero, por ahora, el
gobierno debería pagar una parte de las primas de seguro de los pobres.
Ya hay pruebas de que puede funcionar. En Ghana, cuando se les
ofreció a los agricultores cobertura frente a los riesgos meteorológicos
con una subvención importante en la prima, casi todos la contrataron.
Como el miedo a las crisis lleva a los pobres a desarrollar estrategias
costosas de diversificación, la subvención de los seguros podría ser
rentable para que los pobres aumentaran sus ingresos. En Ghana,
comparando los agricultores que recibieron los seguros baratos con los
que no, los primeros se mostraron más propensos a usar fertilizante en
sus cosechas y, gracias a ello, acabaron en mejor situación. Por ejemplo,
la probabilidad de que tuvieran que suprimir alguna comida fue mucho
menor[22]. Es posible que la subvención pudiese desaparecer con el
tiempo, a medida que la gente viera cómo funciona el seguro y que el
mercado fuera creciendo. Pero, incluso si no fuese posible, dadas las
enormes ventajas potenciales que podrían lograrse si los pobres no
tuvieran que ser los gestores de los fondos de alto riesgo de sus propias
vidas, este parece un muy buen ámbito de actuación en el que usar
fondos públicos para promover el bien común.
Journal of Economic Literature 2012, 50(4), 1051–1079
http://dx.doi.org/10.1257/jel.50.4.1051

Women Empowerment and

Economic Development
Esther Duflo*

Women empowerment and economic development are closely related: in one direction,
development alone can play a major role in driving down inequality between men
and women; in the other direction, empowering women may benefit development.
Does this imply that pushing just one of these two levers would set a virtuous circle
in motion? This paper reviews the literature on both sides of the empowerment–
development nexus, and argues that the interrelationships are probably too weak
to be self-sustaining, and that continuous policy commitment to equality for
its own sake may be needed to bring about equality between men and women.
( JEL I14, I24, I32, I38, J13, J16, O15)

1.  Introduction above the age of 60. Stark as the excess

mortality is, it still does not capture the

T he persistence of gender inequality is

most starkly brought home in the phe-
nomenon of “missing women.” The term
was coined by Amartya Sen in a now clas-
sic article in the New York Review of Books
(Sen 1990) to capture the fact that the pro-
portion of women is lower than what would
be expected if girls and women throughout
the developing world were born and died at
the same rate, relative to boys and men, as
they do in sub-Saharan Africa. Today, it is
estimated that 6 million women are miss-
ing every year (World Bank 2011) Of these,
23 percent are never born, 10 percent are
missing in early childhood, 21 percent in
the reproductive years, and 38 percent
* Massachusetts Institute of Technology.
fact that throughout their lives, even before
birth, women in developing countries are
treated differently than their brothers,
­lagging behind men in many domains. For
each missing woman, there are many more
women who fail to get an education, a job,
or a political responsibility that they would
have obtained if they had been men.
Table 1 summarizes some indicators of
the relative position of women and men
circa 1990 and circa 2009 in poor countries.
Both the relative deprivation of women,
and the extent to which there have been
improvements over the last twenty years, are
apparent in a number of spheres. In access
to education: in low and moderate income
countries, the enrollment rate for girls in
secondary school was 34 percent in 2010,
while that for boys was 41 percent. Twenty

1051
1052 Journal of Economic Literature, Vol. L (December 2012)

Table 1
Some Indicators of the Welfare of Men and Women

1990 2009 or 2010

Boys/men Girls/women Boys/men Girls/women
Primary enrollment rate (gross) (1) 81(1991) 67(1991) 107 100
  [low income countries]
Secondary enrollment rate (gross) (2) 30 (1991) 22 (1991) 41 34
  [low income countries]

Labor force participation (3) 82% 50% 78% 52%

(1980) (1980)
Gender ratio at birth, China (4) 1.11 1 1.18 1
Gender ration at birth, India (5) 1.06 1 1.09 1

Maternal mortality (per 100,000 live births) (6) x 850 x 580

  [low income countries]

Life expectancy (7)

  [low income countries] 51 54 56 59
  [high income countries] 72 79 77 83

Sources:
(1) World Bank (2011).
(2) World Bank (2011).
(3) World Bank (2011).
(4) People’s Republic of China Population Census Office (1990).
(5) Indian Registrar General and Census Commissioner (1990).
(6) World Bank (2011).
(7) World Bank (2011).

years before, the rates were respectively
22 percent and 30 percent. Meanwhile
primary school enrollment has become
­ ­constituted just 19.4 percent of the mem-
nearly universal for both boys and girls. In
labor market opportunities: women are less
likely to work, they earn less than men for
similar work, and are more likely to be in
poverty even when they work. Women spend
almost twice as much time on housework,
almost five times as much time on child care,
and about half as much time on market work
as men do (Berniell and Sánchez-Páramo
2011). In political representation: women
bers of lower and upper houses of parlia-
ments in July 2011 (Inter-parliamentary
Union 2011). In legal rights: women in
many countries still lack independent rights
to own land, manage property, conduct
business, or even travel without their hus-
band’s consent. Twenty-one of the 63 coun-
tries studied by Htun and Weldon (2011)
 Duflo: Women Empowerment and Economic Development
have unequal inheritance rights for men
and women.
There is a bidirectional relationship
between economic development and wom-
en’s empowerment defined as improving
the ability of women to access the constitu-
ents of development—in particular health,
education, earning opportunities, rights,
and political participation. In one direction,
development alone can play a major role in
driving down inequality between men and
women; in the other direction, continuing
discrimination against women can, as Sen
has forcefully argued, hinder development.
Empowerment can, in other words, acceler-
ate development.
Policymakers and social scientists have
tended to focus on one or the other of
these two relationships. Those focus-
ing on the first have argued that gender
equality improves when poverty declines.
Policymakers should therefore focus on
creating the conditions for economic
growth and prosperity, while seeking, of
course, to maintain a level playing field for
both genders, but without adopting specific
strategies targeted at improving the condi-
tion of women.
In contrast, many emphasize the sec-
ond relationship, from empowerment to
development. The Secretary General of the
United Nations, Kofi Annan, for example,
has argued that achieving gender equality
is a “prerequisite” to achieving the other
Millennium Development Goals (MDGs),
including eliminating poverty, reducing
infant mortality, achieving universal edu-
cation, and eliminating the gender gap in
education by 2015 (United Nations 2005).
In its report, “Engendering Development,”
the World Bank (2001) calls for policies
to address gender imbalance in “rights,
resources, and voice,” and recommends
that institutional structures be overhauled
to promote equality, and that specific mea-
sures, such as girls’ scholarships and ­quotas
1053
for women in parliament, be adopted.
These measures are justified, according to
the report, not only because they promote
equity, but also because they are necessary
to accelerate development. Interestingly,
the 2012 World Development Report
(World Bank 2011) adopts a much more
nuanced message. While it emphasizes the
“business case” for women empowerment,
it mainly takes it as given that the equality
between women and men is a desirable goal
in itself, and policies should aim to achieve
that goal.
This paper reviews the evidence on both
sides of the empowerment–development
relationship. It first shows that poverty
and lack of opportunity breed inequality
between men and women, so that when
economic development reduces poverty,
the condition of women improves on two
counts: first, when poverty is reduced, the
condition of everyone, including women,
improves, and second, gender inequality
declines as poverty declines, so the condi-
tion of women improves more than that of
men with development. Economic devel-
opment, however, is not enough to bring
about complete equality between men
and women. Policy action is still neces-
sary to achieve equality between genders.
Such policy action would be unambigu-
ously justified if empowerment of women
also stimulates further development, start-
ing a virtuous cycle. This essay argues
that empowering women does indeed
change society’s choices in important ways,
although the usual depiction of women
as always making the best decisions for
long-term development is somewhat exag-
gerated. The conclusion here is a more bal-
anced, somewhat more pessimistic picture
of the potential for women’s empowerment
and economic development to mutually
reinforce each other than that offered by
the more strident voices on either side of
the debate.
1054 Journal of Economic Literature, Vol. L (December 2012)
2.  Can Economic Development Cause
Women’s Empowerment?
Gender inequality is often greater among
the poor, both within and across countries.
For example, while the gender gap in pri-
mary and secondary gross enrollment has
rapidly gone down between 1991 and 2009
worldwide, it is still wider in poor countries
(7 percentage points for primary enrollment,
13 percentage points for secondary enroll-
ment) than in middle income countries (3
percentage points for primary enrollment, 2
percentage points for secondary enrollment)
and rich countries (0 percentage points for
primary, 1 percentage point for secondary).
And within countries, gaps between boys
and girls persist in poorer and more iso-
lated communities (World Bank 2011). The
participation of women in the labor market
has grown by 15 percent in East Asia and
Latin America between 1971 and 1995, a
rate faster than that for men, and the gender
gap in wages has narrowed as well. The life
expectancy of women has increased by 20–25
years in developing countries over the past
fifty years (World Bank 2011), while male life
expectancy did not improve as much.
Is it the case that as countries develop,
women’s empowerment will follow naturally,
and there is therefore no need for specific
policies targeted at improving the condition
of women? Is it sufficient to fight poverty
and to create the conditions for economic
growth in poor countries? Recent research
suggests that economic growth, by reduc-
ing poverty and increasing opportunity, can
indeed have an important positive impact on
gender equality.
2.1 Relaxing the Grip of Poverty through
Economic Development
The first way by which economic devel-
opment reduces inequality is by relaxing
the constraints poor households face, thus
reducing the frequency at which they are
placed in the position to make life or death
choices. Because these tragic choices are
often resolved at the expense of women’s
well-being, increasing the resources avail-
able to families, as economic development
does, reduces the excess vulnerability of
women.
Even in the countries where the prefer-
ence for boys is strongest, the evidence that
girls systematically receive less care than
boys under normal circumstances is not as
clear-cut as one might guess. It is, of course,
very difficult to observe whether, for exam-
ple, girls are given less to eat than boys, since
households under observation are likely to
change their behavior. Also, when asked to
keep diaries of how much each member con-
sumes over a given period, households may
misreport the portions given to boys and
girls. Finally, households may give the same
amount to boys and girls, but girls’ needs may
be different. Deaton (1989, 1997) proposed
an ingenious way to use household consump-
tion data to indirectly estimate whether girls
are given less to eat than boys. He noted that,
when a child is born, the household, in effect,
becomes poorer, since there is now one more
person to feed—one who will remain unpro-
ductive for a long while. This means that the
adult members of the household need to cut
down on their own consumption to make
room for the new expenditures. Observing
the extent to which the household consump-
tion of “adult goods,” such as cigarettes, alco-
hol, or adult clothing drops when a child is
born, provides us with an indirect estimate
of the “cost” of the child. If families expend
fewer resources on girls, for example, if girls
are given less to eat than boys, then the adults
will cut their consumption of adult goods by
a smaller amount when they have an extra
girl than when they have an extra boy. Yet,
both in Cote d’Ivoire, where Deaton first
conducted the research, and in Pakistan,
where he repeated the analysis (Deaton
1997), there is no evidence that households
 Duflo: Women Empowerment and Economic Development
reduce consumption of adult goods less for
a girl than for a boy. The lack of evident dis-
crimination in everyday life is also apparent,
for example, in immunization rates. The
vaccination differential between boys and
girls in India is small or nonexistent (accord-
ing to the National Family Health surveys,
girls have received 4.55 shots, and boys 4.79
shots). Rajasthan is one of the states in India
with the lowest female to male ratio and the
lowest immunization rate among states in
India. Yet, Banerjee et al. (2010) find no dif-
ference in immunization rates between boys
and girls: on average, boys had received 1.19
immunizations, and girls 1.18.
This is not to say that girls are not treated
differently than boys. The differential treat-
ment is observed when either children or
parents face extreme circumstances. India
has one of the largest discrepancies in gen-
der-specific mortality rates. One reason for
this is that that girls are treated differently
when ill: for example, a study found that,
in the poor neighborhoods of New Delhi,
girls are more than twice as likely to die of
diarrhea (Khanna et al. 2003). This may not
only cause excess mortality, but a progressive
deterioration in the nutritional status of girls
compared to boys, even if girls eat as much
as boys. If poor households are less likely
to spend money on a girl’s illness than on a
boy’s illness, then improved access to health
services, through either health insurance for
the entire family or free medical care for the
poor, would disproportionately help girls,
even if parents do not change their behavior
toward them.
Differential treatment of girls and boys
also becomes acute when the household
itself is facing a crisis. In India, the excessive
mortality rate of girls, relative to boys, spikes
during droughts. When they cannot afford to
feed everyone, families disproportionately
sacrifice the welfare of girls (Rose 1999).
Research conducted in rural Tanzania shows
explicitly how the vulnerability of women
1055
increases when households face a crisis.
When the harvest is bad, due to droughts
or floods, and food is scarce, the murder of
“witches” (almost always old women) is twice
as likely to occur as in normal years (Miguel
2005). If crises throw poor households into
circumstances in which they are more likely
to discriminate against vulnerable women,
then increasing the ability of poor house-
holds to weather crises would disproportion-
ately help women.
These two examples suggest that just
reducing the grip of poverty on these house-
holds or helping them to deal with crises
could improve the welfare of women of all
ages. Ali et al. (2011) show that, in seven
countries around the world (including
Afghanistan and India), there is no difference
in the ways that girls and boys are treated
by health practitioners once they reach the
facility. If facilities were closer and easier to
reach, parents would likely be bringing boys
and girls more often, and they would then
be equally treated. Some also argue that,
mechanically, improvements in water and
sanitation conditions, as well as the better
control of infection, may also be more ben-
eficial to girls than to boys, because boys are
more susceptible to congenital deformation,
and their disadvantage compared to girls, is
lower for infectious diseases (Drevenstedt et
al. 2008).
As households become richer, they will
also be less likely to face choices at the
margin of subsistence. Indeed, in India,
the study by Rose (1999) shows that house-
holds that can buffer their consumption in
a bad year—those, for example, with assets
to sell—do not show the dramatic increase
in relative mortality of girls during droughts.
This first suggests that providing house-
holds with insurance against risk may reduce
both child mortality and the mortality gap
between boys and girls, and second, that
households that are a little richer are less
likely to be thrown into desperation by bad
1056 Journal of Economic Literature, Vol. L (December 2012)
rainfall. Reducing poverty, it seems, even
without targeting women, will dispropor-
tionately help women. Pension remittances
in South Africa offer suggestive evidence of
how a nontargeted transfer can improve the
plight of women. At the end of apartheid in
the early 1990s, old-age pension programs,
previously limited to whites, were expanded
to cover South Africans of all races. Since the
introduction of the program, witch killings in
rural Northern Province have dropped dra-
matically (Singer 2000). It is, of course, dif-
ficult to definitively establish causality given
the many other political and social changes
that occurred in South Africa during the
same period.
Economic development reduces pov-
erty. It increases the ability—distinct from
will—of households to withstand crises and
the ability of governments to insure their
poorest citizens against sickness and hunger.
Thus, by reducing the vulnerability of poor
households to risk, economic development,
even without specifically targeting women,
disproportionately improves their well-being
2.2 Economic Development, Fertility,
and Maternal Mortality
Other than pre-birth and in early child-
hood, women are most likely to be missing
relative to men in childbearing years. This,
of course, is not the result of active discrimi-
nation, but due to the fact that women carry
children and give birth, and that in itself is a
dangerous activity. Other than directly affect-
ing the welfare of women (and not men) in a
significant way, maternal mortality is poten-
tially a source of lower parental investment in
childhood: if parents expect girls to be much
more likely to die as young women than boys,
they may be more inclined to invest in boys.
Jayachandran and Lleras-Muney (2009) find
that the sudden reduction in maternal mor-
tality in Sri Lanka (brought about by effec-
tive public policy) led to a convergence in
the education level of boys and girls. They
estimate that every year of increase in life
expectancy leads to an increase in years of
education of girls (relative to boys) of 0.11 of
a year. This creates two ways for economic
development to potentially improve the
relative welfare of women: by reducing the
chance that they die at each childbirth, and
because economic development goes hand
in hand with a reduction in fertility.
In sub-Saharian Africa, a woman faces a
1 in 31 chance of dying from complications
due to pregnancy or childbirth, while in
the developed regions the risk is 1 in 4,300
(World Bank 2011). While maternal mortal-
ity depends on a number of factors, it is less
likely to occur if individuals are richer (and
are able to deliver in a high quality facility)
and if health systems function better. The
diffusion of technological innovation matters
as well: Jayachandran, Lleras-Muney, and
Smith (2010) find that the introduction of
sulfa drugs in the United States led to a 24 to
36 percent reduction in maternal mortality
in the 1930s.
Overall, fertility decreases with income
and education and age at first birth increases.
Age at first birth is a risk factor in maternal
mortality: very young mothers are more
likely to die or have complications in child-
birth. Fertility has declined dramatically in
most developing countries over the last two
decades, with few countries now having a
total fertility rate of 6 or higher. That in itself
has had a mechanical impact in reducing the
number of missing women by decreasing the
number of chances a woman has to die dur-
ing her childbearing years.
2.3 Giving Women Hope by Expanding
Their Opportunities
The fact that women have fewer oppor-
tunities in the labor market may contribute
to their unequal treatment in the house-
hold. Parents have lower aspirations for their
daughters than for their sons, and female
teenagers themselves have lower aspirations.
 Duflo: Women Empowerment and Economic Development
For example, in West Bengal, Beaman et al.
(2011) find that, in places where no woman
had ever been the local leader, 86 percent of
parents wanted their daughters to be either
a housewife or whatever their in-laws would
decide for her (the corresponding fraction
was below 1 percent for the boys).
If women do not work outside the home,
there may be a perception that they do
not need to be as strong and healthy and
that they do not need a formal education.
Interviews of parents in five states in north
India conducted for the Public Report on
Basic Education (PROBE) in India (The
PROBE Team 1999), found that as high as
10 percent of them believed that it was not
important for girls to be educated—only up
to 1 percent believed the same for boys. Fifty-
seven percent wanted their sons to study “as
far as possible,” while only 28 percent wanted
the same for their daughters. Many parents
believed that educating girls is not necessary,
since girls are only expected to marry and
take care of their households. In the Beaman
et al. (2011) study, 32 percent of the parents
reported that they wanted their teenage boys
to graduate from secondary school or college,
and only 18 percent wished the same for their
girls.
If part of the motivation for educating chil-
dren is to enhance their employment oppor-
tunities, then improving the opportunities
available to women in the labor market would
provide a strong catalyst for the treatment of
women to change for the better. Economic
development leads to a change in the nature
of work that is more conducive to women’s
work. This happened in the United States dur-
ing 1930–50, where expansion of the demand
for clerical workers increased the availability
of jobs that even respectable women could
keep once married (Goldin 2006). More
recently, this has happened not only in China
and Mexico, with the rise in factory work suit-
able for women, but also in India with the rise
in service jobs due to outsourcing.
1057
It has been shown in several contexts that
parents and children are sensitive to the
perceived returns to education: those who
believe that education is more worthwhile
invest more in school (avoiding dropping out,
being absent less often, or working harder
toward exams).1 Three recent studies, one in
China and two in India, show that increased
opportunities for women in the labor market
do indeed translate into better outcomes for
women. The introduction of the Household
Production Responsibility System as part of
post-Mao agricultural reforms in rural China
allowed farming households to grow cash
crops instead of staple cereals. This led to a
substantial boost in the production of tea in
areas that were suitable for growing it. Given
their smaller stature, particularly in terms
of their height and the size of their hands,
women have a comparative advantage over
men in the production of tea. In regions
unsuitable for growing tea, the production of
cash crops disproportionately produced by
men rather than women increased following
the reform. Qian (2008) shows that the num-
ber of missing women, which is particularly
high in China, decreased in tea producing
regions compared to other regions. For the
same increase in total household income, an
increase in female income of 7 U.S. dollars
per month (10 percent) translates into a 1
percentage point increase in the survival rate
for girls. It is striking that this is true even
in China, where it is generally believed that
cultural factors and the “one-child” policy
are very strong determinants of the prefer-
ence for boys.
The entry of India into the world econ-
omy provides another example of economic
development leading to improved gender
equality, even reversing the fortunes of
boys and girls as the traditional segregation
of schooling options turned against boys.
1 See Jensen (2010b), Nguyen (2008), and Attanasio
and Kaufmann (2009).
1058 Journal of Economic Literature, Vol. L (December 2012)
Munshi and Rosenzweig (2006) study the
choice of language instruction in Mumbai
over several decades. In universities, instruc-
tion is in English, but in primary and sec-
ondary schools, parents can choose either
English or the local language, Marathi, as the
medium of instruction for their child. When
India liberalized its economy in the 1990s,
and its software and service industries grew,
the economic returns to education in English
increased dramatically. The new sectors,
such as outsourced telemarketing, also pro-
vided labor market opportunities for women
who had traditionally been shut out of the
labor market, leading to a rapid increase in
English-based education for both boys and
girls. Among the lower castes, the increase
was much faster for girls than for boys: the
proportion of lower-caste girls instructed in
English almost caught up to that of upper-
caste girls, but for the boys, the increase
was not any faster compared to other castes.
Among the lower castes, girls are now more
likely to be educated in English than boys
are. The reason is that members of low-
caste households have traditionally relied on
the caste network to find jobs, and choosing
instruction in English for the child was con-
strued as an attempt to break out from the
foray of the caste. This spirit persists to some
extent, locking boys into education in Marathi
and then less lucrative jobs. This constraint
does not apply to girls, who traditionally
did not participate in the labor market. In
other words, girls have no tradition of rely-
ing on the caste (old-boy) network and are
free from the group expectations that bind
the boys. Girls can be educated in English
and therefore be in a better position to take
advantage of marketplace opportunities as
they arise. A quiet revolution is happening,
even if households are not fully aware of the
consequences of their individual choices.
An experiment conducted by Jensen
(2010a) provides another clear demon-
stration of how an increase in available
opportunities can benefit girls more than
boys. Jensen teamed up with Business Process
Outsourcing centers (BPOs) centers to orga-
nize recruiting sessions in randomly selected
villages in rural areas where recruiters would
typically not go in three states in Northern
India. In some villages, only women were
recruited, and in some villages, both men and
women were recruited. Not surprisingly, com-
pared to other randomly chosen villages that
did not see any such recruiting efforts, there
was an increase in the employment of young
women in BPOs in these villages. Much more
remarkably, given that this was in the part of
India probably most notorious for discrimi-
nation against women, three years after the
recruiting started, girls age five to eleven were
about 5 percentage points more likely to be
enrolled in school in the villages where there
was recruiting. They also weighed more, sug-
gesting that parents were taking better care
of them. The effects on boys was much more
muted, even in villages where they were
explicitly recruited. Jensen and Miller (2010)
show that this low effect is the combination
of a negative effect of boys whom parents
wanted (at baseline) to keep them on the farm
and positive for those whom they wanted to
send away. Here again, the facts that girls are
not expected to continue to be the ones to
preserve the family tradition actually turned
out to be to their advantage.
All these examples show that gender-
blind policies that improve the economic
welfare of households can improve gender
equality, and that diversifying the economy
and increasing women’s options in the labor
market can cause households to adjust
their behavior, moving them toward gender
equality.
2.4 Freeing Up Women’s Time
A key source of inequality between
women and men stems from the way they are
expected to spend their time. Berniell and
Sánchez-Páramo (2011) conducted a study
 Duflo: Women Empowerment and Economic Development
of time use for the World Development
Report 2012, using data for 23 countries
from the Multinational Use Study, and 12
countries that they added to make the data
set more representative of poor countries.
A clear, unsurprising pattern emerges: at all
level of incomes, women do the majority of
housework and care and, correspondingly,
spend less time in market work. The dif-
ference ranges from 30 percent more time
spent on housework by women than men
in Cambodia to six times more in Guinea,
and from 70 percent more time for child
care in Sweden to ten times more in Iraq.
These differences have an impact on wom-
en’s ability to participate in market work, be
fully engaged in their career, etc. This is one
reason why women in developing countries
are more likely to be involved in informal
work (such as running a small business) that
may not be the most productive use of their
time. For example, they may run a shop from
their house in order to be able to mind the
children at the same time, instead of taking
on productive employment. This reduces
the chance that they have an autonomous
income, which, as we will see below, has
clear implications for their bargaining power
within the household.
In this context, economic development
can lead to the empowerment of women by
freeing their time (which can then be used
for market activities or for other things).
Greenwood, Seshadri, and Yorukoglu
(2005) argue that the diffusion of appli-
ances in the United States between 1930
and 1950 was a key driver of the increase
in the labor market participation of women
during that period and beyond. Dinkelman
(2010) exploits the logistical feasibility of
the roll out of electrification in South Africa
to study the impact of access to electricity
on female and male labor supply. She finds
that electrification led to an increase of 9.5
percentage points in female employment
(without any change in male employment)
1059
and argues that this increase was due to
time freed for women in home production.
Devoto et al. (2011) find no impact on time
devoted to market activities when house-
holds became connected to piped water in
Morocco, but a significant increase in lei-
sure and reduction in stress levels and intra-
households conflicts.
Another source of gain of time with eco-
nomic development is the reduction of fertil-
ity, already mentioned, as well as the change
in the timing of this fertility. Young women
who marry or have children typically drop
out of school or college and are less likely to
work (Field and Ambrus 2008, Miller 2010,
Duflo, Dupas, and Kremer 2011). They tend
to marry partners who are older than them.
All of these lead to a reduction in women’s
autonomy and capacities. In the United
States, Goldin and Katz (2002) show how
the availability of the pill to young, unmar-
ried women in the 1960s increased the age
at marriage and the length of study of young
women.
2.5 Economic Development and
Women’s Rights
Empirically, there is a strong correlation
between economic development and wom-
en’s legal rights, in areas as diverse as prop-
erty rights, access to land, access to bank
loans, violence against women, abortion
policy, etc. Doepke and Tertilt (2009) show
a robust negative correlation of 0.4 or higher
across countries between the lack of rights
and GDP per capita. Historically, the expan-
sion of economic rights to women in the
United States and Europe preceded their
access to political rights (Doepke and Tertilt
2009; Fernandez 2009). While it is of course
impossible to infer causality from the data,
two lines of argument suggest why economic
growth could lead men to willingly surren-
der economic rights to their wives. Doepke
and Tertilt (2009) argue that, when the
importance of human capital in the economy
1060 Journal of Economic Literature, Vol. L (December 2012)
increases (with technological progress), men
start to be willing to surrender some rights
to women to ensure that children get bet-
ter educated (the argument requires that
women care more about children’s human
capital and that bargaining power matter for
household decisions, which we will discuss
below). The trade-off is between their utility
today and the utility of their offspring (chil-
dren, grandchildren, and future generations).
Fernandez’s (2009) argument is slightly dif-
ferent, although related: she argues that as
fertility declines, fathers’ interest as hus-
bands (who would like to have all the rights),
start to conflict with their interest as fathers
(who would like to protect their daugh-
ters against their future son-in-laws). With
economic growth and a decline in fertility,
the balance starts tilting towards the latter
interest, and women’s economic rights are
expanded. Doepke and Tertilt provide some
historical narrative that is consistent with
their theory, and Fernandez some sugges-
tive evidence from the granting of rights to
women in the United States that states that
had a faster reduction in fertility were more
likely to expand women’s rights. Empirically,
there remains much more to be done, and
this is a very interesting area of research.
These two papers provide convincing theo-
retical arguments, however, that economic
growth can lead to a progression in women’s
rights. In fact, Doepke and Tertilt explicitly
conclude that institutions such as the World
Bank that are interested in women’s rights
would be well inspired to focus on programs
favorable to economic development (such as
education policy), rather than push for direct
legislative changes for women’s rights.
Is that how it works, then: as countries
develop, empowerment of women will follow
naturally? Is there a reason to design policies
specifically targeted toward improving the
condition of women? Or is it sufficient for
improving women’s condition to fight poverty
and to create the conditions for economic
growth in poor countries? In a word, will
economic development be enough?
2.6 But Will Economic Development Be
Enough?
There is evidence that growth will not
be enough to overcome discrimination in
the home and in a number of domains. Sex
ratios remain skewed in favor of boys. In
China, despite rapid economic growth (and
the reforms described above), the sex ratio at
birth has worsened continuously since 1970,
with an acceleration in the 1990s from about
53 percent of boys among all (reported)
births to about 57 percent of boys among all
(reported) births. Within Asia, the sex ratio
at birth in South Korea and Taiwan, both
rich countries, is similar to that in China and
India. The gap between girls and boys is clos-
ing for primary and secondary schooling, but
for tertiary education, the ratio of females to
males has not improved overall, even though
participation has risen for both boys and
girls. In the labor market, even in developed
countries, women who are equally qualified
continue to earn less than men at all levels of
qualification. Legal rights, particularly prop-
erty rights, of women remain different from
that of men in many countries, even as econo-
mies grow. Compared to economic opportu-
nities, education, and legal rights, the gender
gap in political participation has narrowed
the least between 1995 and 2005. As of July
2011, only 26 countries in the world had met
the target (set by the U.N. Economic and
Social Council in 1990) of having 30 percent
or more women in national legislative seats;
the proportion of seats held by women in sin-
gle or lower houses of parliament was only
19.4 percent globally, up from 15.9 in 2005,
13.5 percent in 2000, and 9 percent in 1987
(Inter-parliamentary Union 2011).
The persistent difference in sex ratios
at birth illustrates the fact that economic
development, and the availability of new
technologies, can have perverse effects on
 Duflo: Women Empowerment and Economic Development
gender equality if it decreases the cost of
discriminating against girls. High differ-
ences in reported sex ratios at birth between
girls and boys are the result of unreported
birth—infanticide—and increasingly, from
sex-selective abortion. Sex-selective abor-
tion shows how the wider availability of new
technologies and the increased well-being
of households resulting from economic
development have led to an increase in a
particularly egregious form of discrimina-
tion. This is not limited to China: the 2001
census in India revealed a reversal of the
trend in the sex ratios, particularly in the
most prosperous states in the north of India.
By the time of the 2011 census, they had
dropped to their lowest levels since inde-
pendence. Economic calculus plays a role
here as well. Advertisements for clinics
practicing sex-selection in Mumbai used to
read, “Better pay Rs 500 now than Rs 50,000
later” (Desai 1994). The Rs 50,000 refers to
the dowry that the parents would need to
pay when a girl is married. With the cost of
sex identification and abortion becoming so
low with new technologies, many more par-
ents may prefer to abort girls rather than
to raise and marry them. In Taiwan, Lin,
Qian, and Liu (2008) document that the
sex ratio at birth (i.e., the fraction of boys
among live births) in Taiwan increased from
0.515 in 1980 to 0.54 by 1990. All of the
increase is accounted for by increases in
the sex ratio for higher parity. They argue
that the legalization of abortion in Taiwan
in 1985/86 (when prenatal sex determina-
tion was already available) is responsible.
Even if increased opportunities for women
reduce the dowry, there is little chance that
they will bring it to such a low enough level
so as to make it worthwhile to let a girl live
in the face of such a low cost for abortion.
Sex selection does not appear to disappear
even in the United States. When a blood test
that can determine the gender of the baby
after seven weeks of pregnancy became
1061
commercially available in the United States,
there were reports of concerns that it could
be used for sex-selective abortion (Belluck
2011). These concerns are almost surely
justified: even without this test, there is evi-
dence of sex-selective abortion in some eth-
nic groups in the United States and Canada
today: in particular, since 1980, there is an
abnormal ratio of boys to girls among chil-
dren at higher parity in Chinese, Korean,
Asian American, and Indian-American fami-
lies (Abrevaya 2009; Almond and Edlund
2008; Almond, Edlund, and Milligan 2009).
Almond and Eldund report, for example,
that for third children, if there is no previ-
ous boy, boys will outnumber girls by 50 per-
cent. Using linked records from California,
Abrevaya shows that Asian Indian mothers
were significantly more likely to have a ter-
minated pregnancy and to give birth to a boy
when they have previously only given birth
to girls. Eventually, in societies where there
is a male preference, the pressure on the
marriage markets due to the rising sex ratio
(it is estimated that there are twenty million
more boys under 20 than girls in China, and
25 million more in India (Edlund and Lee
2009)) may provide some incentive for the
sex ratio to revert. Edlund and Lee (2009)
provide evidence that the sex ratio is nor-
malizing in Korea, after having exploded
with the introduction of sonography.
The disparity in earnings at all levels of
qualification even in richer countries illus-
trates the persisting bias against women.
This is not the place for a detailed review on
gender discrimination in developed coun-
tries: Goldin (2006) provides a masterful
summary of the long-run evaluation of the
place of women in society. Gaps in wages,
occupation, education, and labor force par-
ticipation have shrunk considerably in the
United States from 1930 to 1990, in part due
to some of the phenomena discussed earlier
(a rise in the demand for clerical work, the
resulting larger investment in education,
1062 Journal of Economic Literature, Vol. L (December 2012)
household appliances, the diffusion of the
pill), but the convergence seemed to level
off after 1990. Differences in participation
and in wages do persist, particularly at the
top, in part due to biases toward women, as
well as different cultural attitudes of what is
expected of women.
Ample research by psychologists shows
that in developed countries, there is a wide-
spread “implicit” bias, shared by both men
and women, associating men with career and
the sciences and women with family and lib-
eral arts For example, implicit associations
tests (e.g., Greenwald et al. 2002; see also
www.implicit.org) ask participants to sort
two series of names to the left or the right of
the screen. For example, female first names
must be placed on the right, and male first
name on the left. In the second series (alter-
nating with the first), words evoking career
must be placed on the left, and words evok-
ing family must be placed on the right [or
the opposite]. Because it is easier for people
to put things on the same side if they “go
together,” the difference in the time spent
filling the stereotypical block (e.g., women
and family are on the same side) and the
nonstereotypical block (women and career
are on the same side) indicate an implicit
bias. Both women and men are more likely
to associate women with family and men
with careers. Both women and men are more
likely to associate women with liberal arts
and men with science.
This bias has persisted despite the wide-
spread participation of women in these aca-
demic disciplines and the labor markets in
these countries. These biases affect women’s
rewards for participating in the labor mar-
ket or for getting a higher education both
directly and indirectly, by persuading girls
that they are not cut out for particular jobs,
or just not as good as men. Psychologists
have shown this effect, known as the “ste-
reotype threat,” to be very powerful. When
female and male students, ­ recognized
for being good at math, are given a diffi-
cult math test in college, women do worse
than men. When they are given the same
test after being told, “You may have heard
that girls are less good than boys at math,
but this is not true for this particular test,”
however, female students do just as well as
males (Spencer, Steele, and Quinn 1999).
The explanation for this phenomenon is
that girls have accepted and internalized
the bias that they are not as good at math,
and they give up when the going gets tough.
When they are told that this “fact” does not
apply to that particular test, they know to
continue to try hard. As long as these biases
persist, gender equality will be hindered
even if the technological conditions for
an even playing field are met. Women are
also negotiating less and less aggressively
than men at hiring and during the promo-
tion stage, and are less willing to compete
(Babcock and Laschever 2003; Gneezy,
Niederle, and Rustichini 2003).
Likewise, while a number of factors con-
tinue to hinder the parliamentary repre-
sentation of women—the type of electoral
system present in a country, the role and
discipline of the political parties, the lack of
previous political experience of women—
the widespread perception that women
are not competent leaders is probably the
strongest barrier to greater participation
of women in policy making. It persists
in developed countries as well: a series
of experiments have shown that, holding
performance constant, women leaders are
evaluated more negatively than male lead-
ers. These studies typically either provide a
written description of leadership situations,
varying the sex of the leader, or use trained
actors to lead, allowing the experimenters
to control the degree of success the leader
achieves (Swim et al. 1989). The surveys
find the bias is most pronounced when the
leadership role is typically considered a
male role.
 Duflo: Women Empowerment and Economic Development
Similar evidence coming from India sug-
gests that citizens tend to give lower per-
formance marks to female leaders than to
male leaders, given the same information
about them, are also less likely to associate
women with leadership than men, and are
generally not feeling positive toward females
in politics. Beaman et al. (2009) investigate
attitudes in rural West Bengal using a vari-
ety of methods. In one experiment, they
ask respondents to listen to a short political
speech, read out either by a male actor or
a female actor. Among people who had not
had the experience of a female leader, ran-
domly selected respondents who heard the
“male” speech are significantly more likely
to give it favorable marks than those who
heard the female speech. In another part of
the study, they conduct “implicit association
tests,” and show a strong association between
women and the domestic sphere. Finally,
respondents (particularly male) seemed
quite comfortable acknowledging that they
strongly preferred a male as leader rather
than a woman (this is in a country that has
a female head of the majority party and a
female president—since then, the state of
West Bengal has also elected a female as the
head of the state). It seems that there is a
significant cultural barrier to recognizing
women as competent policy makers.
Evidence such as this provides support
for the idea of “reservations” or quotas for
women in policy-making positions. Since
perceptions are biased, in the absence of
affirmative action of some sort, it would be
very difficult for women to break into politics.
Indeed, in most of the seventeen countries
where the target of 30 percent of women in
parliament has been achieved, some kind of
affirmative action measure has been in place.
If one wants to achieve balanced gender rep-
resentation rapidly, it seems clear that affir-
mative action will be needed.
More generally, economic development
alone will probably not be enough to bring
1063
about equality between women and men in
the foreseeable future, and policies will be
required to accelerate this process.
Yet, the gains from policies that target
women come, to some extent, at the expense
of men. This much is evident in politics. Any
position that a woman gets through a quota
is a position that a man does not get. The
trade-offs are not always as explicit, but can
be very stark indeed. For example, specific
measures to improve access of girls to school,
such as scholarships for girls or latrines in
school, are an expensive way to get more
girls into schools. Given that enrollment is
already high, many scholarships go to girls
who would have gone to school anyway, mak-
ing the cost per additional girl induced to go
to school very high. This means that within
the very limited budget of most develop-
ing countries, the transfers to girls come
at the direct expenses of boys. The money
spent on scholarships is not spent on other
things that may help both boys and girls,
such as hiring new teachers, or deworming
everyone. The trade-offs are real: compara-
tive cost effectiveness research (comparing
across different experiments in a variety
of contexts) suggest that the cost per extra
child enrolled of additional scholarships may
be much larger than that of deworming, or
informing parents of the returns to educa-
tion (Dhaliwal et al. 2011).
Thus, policies that explicitly favor women
need to be justified, not just in terms of being
necessary to bring about gender equality,
but in terms of gender equality itself being
desirable and worth the cost it implies. The
second part of this essay explores the com-
mon justification that the trade-off between
the interests of various people seen in the
short run is transitory; in the long run, there
is no trade-off between helping women more
and helping everyone, because increasing
the share of resources going to women will
increase the amount of resources so much
that everyone will be better off.
1064 Journal of Economic Literature, Vol. L (December 2012)
3.  Can Women’s Empowerment Cause
Economic Development?
There are two rationales for supporting
active policies to promote women. The first is
that equity is valuable in and of itself: women
are currently worse-off than men, and this
inequality between genders is repulsive in
its own right. For example, in the United
Nation’s 2005 report on the Millennium
Development Goals (MDG), Kofi Annan,
the Secretary General of the United Nations,
writes: “The full participation of women to
all levels of decision-making is a basic human
right.” The second, a central argument in the
discourse of policymakers, is that women
play a fundamental role in development. The
gender gap in education, political participa-
tion, and employment opportunities should
therefore be reduced not only because it is
equitable to do so, but also because it will
have beneficial consequences on many
other society-wide outcomes. It should be
done, in other words, to increase efficiency.
Accordingly, in the same report, Kofi Annan
argues that gender equality is in fact a “pre-
requisite” to achieving the other MDGs.
The stance that women empowerment is
desirable for efficiency shapes both the pol-
icy debate and the resultant economic poli-
cies the world over. Micro-credit schemes,
for example, have been directed almost
exclusively at women, because, it is argued,
women invest the money in goods and ser-
vices that improve the well-being of families,
in goods that are conducive to development.
Similarly, most conditional cash transfer
benefit programs in developing countries,
such as PROGRESA/Opportunidades in
Mexico, direct the transfer to women, not
men. Transfers conditioned on school enroll-
ment are often higher for girls or even
positive only for girls. Measures to enforce
women’s access to political positions through
quotas have been instituted in 87 countries,
including India, where a 1993 constitutional
amendment required that one-third of rural
village council seats and village presidencies
be reserved for women.
Former World Bank President, James
Wolfensohn, addressing the Fourth UN
Conference on Women, said:
Education for girls has a catalytic effect on
every dimension of development: lower child
and maternal mortality rates; increased edu-
cational attainment by daughters and sons;
higher productivity; and improved environ-
mental management. Together, these can
mean faster economic growth and, equally
important, wider distribution of the fruits of
growth. . . . More education for girls will also
enable more and more women to attain lead-
ership positions at all levels of society: from
health clinics in the villages to parliaments in
the capitals. This, in turn, will change the way
societies will deal with problems and raise the
quality of global decision-making.
Wolfensohn makes four claims. First of
all, that education of girls would change out-
comes for their children and the rest of us.
We might call this a “reduced form” argu-
ment: broad “empowerment” of women
could change outcomes. Education could
have an effect, for example, through, for
example, by improving their understanding
of how to raise children, use contraception,
and manage their homes. His second claim is
that empowerment of women in a narrower
sense (power or the ability to influence deci-
sion making) would also change outcomes.
The third claim is that these changes would
be positive. The comment also raises a fourth
issue, assuming that increasing decision mak-
ing of women would indeed lead to different
(and better) outcomes, that is what policy
instruments are available to policymakers to
achieve these changes. Do they need to wait
for it to come from improvement in their
human capital (the quote suggests it would
be a consequence of their improved educa-
tion), or can policy interventions, such as
changes in the law, change in electoral rules,
changes in the rule governing programs,
or other similar “top down” interventions
change effective power?
 Duflo: Women Empowerment and Economic Development
The remainder of this section investigates
these four issues. As will become apparent,
the answers are often interlinked: it is often
a change in the rules governing local politics,
or particular programs, for example, which
allow us to identify shifts in power and their
resulting outcomes. Moreover, there is no
simple measure of “power” (unlike educa-
tion, for example), so that change in power
is often proxied by changes in outcomes.
Nevertheless, these four themes serve
as a useful organizing framework for the
literature.
3.1 Women Empowerment and Changes in
Family Outcomes
A substantial literature has studied these
issues and found clear evidence of a correla-
tion between mothers’ education and earn-
ings, and child welfare, particularly child
health. Moreover, the correlation with moth-
ers’ education and earnings is almost always
found to be stronger than the correspond-
ing correlation with fathers’ education and
earnings. However, correlations are often
misleading. In this instance, there are two
fundamental problems with the interpreta-
tion of the results. First, a woman’s educa-
tion, earnings, or political participation may
be correlated with unobserved dimensions
of her ability, family, or community back-
ground. To the extent that these unobserved
factors directly determine child health, the
correlation does not indicate the causal
effect of raising a woman’s education, earn-
ings, or political participation. It is important
to recognize that the very fact that women
are typically less likely to get an education,
earn an income, and participate in political
decisions is likely to make this bias stronger
for women than for men. Likewise, chil-
dren might do better in countries or regions
where women’s political participation is
greater because these places are otherwise
more favorable environments. Second, the
comparison between the coefficients of
1065
husbands’ and wives’ education or earnings
might be obscured by a correlation between
wives’ education or earnings and unobserved
characteristics of husbands for two reasons:
On the one hand, more educated or richer
women may be able to marry men who care
more about their children. On the other
hand, the observation that, after controlling
for total resources, income in the hands of
women is associated with more desirable
outcomes than income in the hands of men
may reflect unobserved attributes of a man
that directly influence child outcomes. For
example, if he is progressive enough to allow
his wife to seek employment, then this same
progressive attitude may make him treat his
children better.
To get around these problems, researchers
have exploited specific circumstances where
the distribution of power, education, or earn-
ings of women and men changed for reasons
that had nothing to do with their individual
choices. These situations allow us to approxi-
mate what would happen if women’s empow-
erment policies were put in place and applied
to everyone, not only to very special families.
A first dimension of empowerment (that
mentioned by Wolfensohn) is education.
Education can have a range of benefits within
the households: if more educated women
command higher outside wages, and it is eas-
ier for them to get a job, then investing more
in educating women, rather than in men,
may indeed have a more positive impact on
child health than when the investments are
spread evenly if women’s bargaining power
in the household leads to better outcomes
for children. In addition, if women are the
primary caregivers for young children, then
more education may help them provide bet-
ter care. Therefore, many believe that a spe-
cial effort is needed to educate girls, and that
educating girls would have tremendous spill-
over effects. Unfortunately, the evidence for
this is not as strong as is commonly believed.
Although an association between education
1066 Journal of Economic Literature, Vol. L (December 2012)
of mothers and the outcomes of their chil-
dren has been found over and over again, the
evidence, in most cases, suffers from obvi-
ous biases: educated girls come from richer
families and marry richer, more educated,
more progressive husbands. As such, it is, in
general, difficult to account for all of these
factors, and few of the studies have tried to
do so. Breierova and Duflo (2004) use the
massive expansion of school construction in
Indonesia in the 1970s to generate variation
in the differences in schooling between hus-
bands and wives based only on their region
of birth and their dates of birth to get around
this problem. When they correct for the
fact that more-educated women differ from
less-educated ones, they still find that, con-
ditional on the household’s average educa-
tion level, households have fewer children
when the wife is more educated. However,
they did not find a lower infant mortality.
While there is no doubt that education has
a positive effect on child mortality, it is not
clear that girls’ education is much more criti-
cal than boys’ education. Chou et al. (2010)
use the same empirical strategy to study the
impact of father and mother education on
child health in Taiwan (which introduced
compulsory junior secondary schooling in
1968 and introduced it progressively in dif-
ferent regions). They find strong and positive
effects of both father and mother education
on child survival, and one effect is not sig-
nificantly greater than the next. While more
needs to be learned about this, the automatic
presumption that female education is more
important than male education for child
mortality and for other children outcomes
may need to be revised: it seems that both
matter.
3.2 Women as Decisionmakers
3.2.1 . . . within the Household
Micro-credit schemes or welfare agencies
that restrict credit or transfers to women on
the grounds that the money will be put to use
germane to development implicitly recog-
nize that women are not entirely powerless.
If women were powerless, then the money
would be immediately appropriated by their
spouses, and we would see no impact of dis-
tributing the money to women rather than to
men. Conversely, if households were harmo-
nious entities where everyone had the same
preferences and desires, then the nominal
ownership of money would not matter within
the household. It would all go to a common
pool and channeled toward the best uses for
the families.
In practice, there are good reasons to think
that real families are somewhere between
these two extremes—neither ruled by a dic-
tator, nor without discord. Instead, all house-
hold members have different preferences,
and different ideas about many things, from
how many children to have to how best to
spend household income, and each idea
is given a different weight in the ultimate
decision depending on each member’s infor-
mation set and bargaining power. In demo-
graphic and health surveys, for example,
women typically report wanting fewer chil-
dren than their husbands (Becker 1999).
A consequence of this difference is that if
women can hide their contraceptive choice
from their husbands, final fertility may be
lower. Ashraf, Field, and Lee (2010) provide
a clear demonstration of this phenomenon
in Zambia. In an experiment, they provided
836 married women in Lusaka, Zambia, with
a voucher guaranteeing free and immediate
access to a range of modern contraceptives
through a private appointment with a family-
planning nurse. Some women received the
voucher in private, while others received the
voucher in the presence of their husbands.
This made a huge difference: compared
to cases where husbands were involved,
women who were seen alone were 23 per-
cent more likely to visit a family-planning
nurse, 38 percent more likely to ask for a
 Duflo: Women Empowerment and Economic Development
relatively concealable form of contraception
(injectable contraceptives or contraceptive
implants), and 57 percent less likely to report
a unwanted birth nine to fourteen months
later. It is important to note that this does
not mean that allowing women to conceal
contraception would be a desirable policy:
it is quite possible that, once husbands do
learn that this option exists, they become
quite suspicious of their wives. They could,
for example, prevent them from going to the
health clinic altogether, delaying important
care. But the paper is clearly demonstrating
that the lack of congruence in preferences
and the ability (and willingness) to conceal
information is a critical element in under-
standing household decisions. Ashraf (2009)
presents lab-experiment evidence that also
demonstrates the importance of information:
men who need to decide how to allocate a
prize do it very differently when they must
share the information on what they decided
with their wives and when they must negoti-
ate with them before taking the decision.
The need to keep things hidden from, or at
least not immediately available to, their hus-
bands appears to lead women to forgo conve-
nience to keep things quiet, and some times
to act in a way that lowers overall efficiency.
Anderson and Baland (2002) argue, for
example, that the rotating savings and credit
associations (ROSCAs) popular especially
among women in many parts of Africa are
primarily a way for women to keep savings
away from their husbands.2 Schaner (2011)
provides striking evidence of this phenom-
enon in Kenya. As part of an experiment, she
opened bank accounts for men and women
in Kenya. After having opened the accounts,
2 Other arguments have been made for ROSCAs.
Besley, Coate, and Loury (1994) show that for every person
except the last one in the group, ROSCAs allow members
to have the good faster than if they saved on their own,
increasing utility even if they have to forgo interest rates.
Gugerty (2007) and Basu (2011) propose that ROSCAs are
a way for hyperbolic discounters to commit to save.
1067
some randomly selected individuals received
a free ATM card, which cut the cost charged
by the bank for withdrawals (in addition to
making it more convenient). Men behaved as
expected: those who got the ATM card made
more active use of the accounts: they saved
more, and withdrew and deposited money
more often. But the effect was opposite for
women: those who got the ATM card used
the accounts less. Further analysis showed
that this effect was concentrated among
women with low bargaining power. It thus
seems that women were less willing to use
the account if it was easy for their husbands
to get the money out.
In this context, the share of the house-
hold budget contributed by a member may
well have an important impact on how much
weight is given to the expenditures he/she
favors. When women are decisionmakers
within the household, how much they bring
to the table can have an important impact
on ultimate choices. Empirically, a large lit-
erature has tested whether income in the
hands of women of a household has a dif-
ferent impact on intra-household allocation
than income in the hands of the men. The
evidence suggests that, compared to income
or assets in the hands of men, income or
assets in the hands of women is associated
with larger improvements in child health
(e.g., Thomas 1990), and larger expenditure
shares of household nutrients, health, and
housing (e.g., Thomas 1993). These stud-
ies may in general, however, suffer from the
problem that families where women earn
a larger share of the income are different
from those where women do not. Thomas
(1990, 1993) uses unearned income to cir-
cumvent the issue that earned income is
endogenous (and corresponds to the labor
supply decisions of men and women). But
this only partially addresses the problem,
because unearned income is often the prod-
uct of assets (so women with more unearned
income may have come to the marriage with
1068 Journal of Economic Literature, Vol. L (December 2012)
more wealth, and therefore are different).
Moreover, marriage decisions further com-
plicate the interpretation of these results, as
men who marry richer women may be more
likely to have tastes similar to theirs.
Duflo (2003) takes the strategy of using
unearned income a step further, and uses
the rapid expansion of the Old Age Pension
Program in South Africa to address the
remaining identification issues. At the end
of apartheid, the government committed to
achieving parity of benefits and eligibility
requirements between whites and blacks.
This was achieved mostly by increasing the
benefits received by blacks, which grew
very rapidly between 1990 and 1993. In
1993, 80 percent of black women above
age 60 and 77 percent of black men above
age 65 received the pension. The maximum
benefit of 370 rands per month—about 3
U.S. dollars per day—was equal to one-half
of the minimum wage, and about twice the
median per capita income in rural areas.
Due to traditional living arrangements,
nearly one-third of black children under
the age of 5 currently live with a pension
recipient. Children who live with a pension
recipient tend to come from relatively dis-
advantaged backgrounds, and so they tend
to be smaller than other children their age.
To estimate the effect of receiving a pension
on the anthropometric status of children,
Duflo exploits the fact that height reflects
accumulated investments in child nutri-
tion. The larger the proportion of life dur-
ing which a child was well-nourished, the
taller she will be given her age. Expansion
of the program in the early 1990s increased
the likelihood of receiving a pension among
qualified elderly people, and the ben-
efits became substantially larger. Thus,
to the extent that the pension resulted in
improved nutrition, children born after the
expansion were more likely to have spent a
larger fraction of their lives well-nourished
if they lived with a pension recipient.
Indeed, Duflo finds that girls who live with
a grandmother who receives the pension are
heavier than those who live with a grand-
mother who is not quite old enough to receive
the pension. Moreover, when she looks at
height, Duflo finds that older girls, who were
born before the pension was in effect, are
smaller when they live with a pension recipi-
ent (male or female) than when they live
with a nonrecipient. However, among young
girls, who have lived their lives since the pen-
sion system was put in place, those who live
with a grandmother who receives the pen-
sion are taller than those who live without
one. This suggests that pensions received
by women do translate into better nutri-
tion for girls. While the weight of the older
girls catches up immediately, deficit in early
nutrition continues to be seen in height even
after good nutrition has resumed, and this is
why the older girls remain smaller. We esti-
mate that for girls, living with a grandmother
who receives the pension is enough to bridge
half the gap between the size of children in
the United States and in South Africa. In
contrast, no such effect is found when the
pension is received by a man and no corre-
sponding effects were found for boys (note
that boys were as delayed as girls in terms of
their stature before getting the pension, so
this difference does not appear to be due to
the fact that only girls need to catch up).
3.2.2 . . . on the Farm: Women and
Property Rights
We observe many instances in which
women retain property rights over a por-
tion of the assets they bring into the house-
hold (we have already seen an example
in Indonesia). In Africa, it is common for
women and men to retain separate property
rights over productive assets, particularly
land. For example, in Burkina Faso, women
retain ownership of their individual plots
within the marriage. Household members
work on each other’s plots, and inputs such
 Duflo: Women Empowerment and Economic Development
as fertilizer may be purchased collectively.
The production is also pooled together for
consumption by all household members. Of
course, as we saw above, household mem-
bers may have different opinions about how
these pooled resources should be used, and
various factors, including how much land
they own, will influence the final consump-
tion choices of the household. A well-func-
tioning family, however, would first try to
maximize the size of the pie before thinking
about how to divide it. All inputs, including
labor, seeds, and fertilizer, should be allo-
cated to all plots so as to maximize the overall
productivity of the landholding. Udry (1996)
shows that, in practice, this is far from being
the case. Many more inputs are being used
on the men’s plots, so that, after controlling
for the year, the characteristics of the plots,
and the type of crops produced, men’s plots
are much more productive than women’s
plots, mostly because much more fertilizer
is applied to these plots. Because the effec-
tiveness of using fertilizer declines steeply
with how much is used, it would make much
more sense to use a little on both men’s
and women’s plots. This, however, is not
what households seem to be doing in most
cases. Overall, household production might
increase by 6 percent just by reallocating the
same amount of fertilizer and labor. This is
not negligible in an environment where peo-
ple are so poor.
So why would they not do this? One answer
may be that households decide how to allo-
cate resources as a function of how much food
everyone produced in a day, without taking
into account the fact that a household mem-
ber may have helped out another to generate
this production. Each household member
prefers a bigger share of a smaller pie and, as
a result, each works mostly on their own plot
without regard to maximizing overall house-
hold production, and they do not sufficiently
help other household members with time
or inputs. Men, being richer and therefore
1069
less credit constrained, can afford fertilizer,
which they use on their own plot, but not on
the woman’s plot. Given this imperfection in
negotiating within the household, a solution
to increase household productivity would
be for the woman to “rent” her land to her
husband, who would then make the efficient
production choice. One reason why this does
not happen is because property rights are
weak: if the husband works on the wife’s land
for too long, he will eventually be recognized
as “owning” it, and will stop compensating
her. Anticipating this, women are not ready
to surrender control to their husbands even
for compensation, and both members con-
tinue to produce in their own corners. This is
an example where the fact that women have
insecure property rights leads to sheer waste,
and literally makes families poorer.
Weak property rights for women also lead
to an inefficiently low level of investment in
Ghana. Farmers in Ghana let their land lie
fallow for a few years to replenish its nutri-
ents. This is a form of investment in that
farmers forgo production for a few seasons
for higher post-fallow production. The risk,
of course, is that someone seeing the fallow-
ing land may decide to start producing on it.
Where property rights are secure, this does
not happen. Property rights on agricultural
land in Ghana, however, are anything but
secure. In fact, a farmer only has, at best,
usufructuary rights: the land belongs to both
the village and the lineage, so that even if a
piece of land is attributed to someone by the
lineage, there is always the possibility that
the village will reallocate it to someone else.
The more powerful someone is within either
the village or the lineage, the more likely it is
that they can fight opposing claims on their
land, and thus can choose to fallow without
risk of being expropriated. Goldstein and
Udry (2005) show that indeed, farmers who
have a position of power within either the vil-
lage or the lineage are more likely to fallow
their land, and fallow it longer. As a result,
1070 Journal of Economic Literature, Vol. L (December 2012)
they have more productive land. Women
are less likely to hold a position of power.
As a result, they fallow less often, and this
leads to a lower productivity for their land.
Again, the lower position of women directly
results in inefficient investment, and this is
not corrected by the transfer of property
rights within the household, which might
secure the claim of the household vis-à-vis
the lineage or the village, but at the expense
of weakening women’s positions within their
own households.
3.2.3 . . . and within the Community:
Women as Policymakers
We have seen that women and men have
different preferences, and that the house-
hold does not efficiently bargain to choose
the actions that maximize the household’s
utility, suggesting that women and men will
have different policy preferences. First,
women will prefer policies that better
reflect their own priorities. Since they are
particularly concerned about child health
and nutrition, they should prefer policies
that will help them achieve these objectives.
Second, women should be in favor of poli-
cies that will increase their bargaining power
within the household, that is, policies that
improve their situation in cases of divorce,
and policies that increase their productivity
in everyday work or improve their chances
to access the labor market. In practice, we
do see that women and men have different
preferences over policies. Chattopadhyay
and Duflo (2004) collect data on the types
of complaints brought by both men and
women to the local village councils in two
districts in the states of West Bengal and
Rajasthan, India. Village councils admin-
ister local public goods in rural India. The
types of formal complaints differ by gender.
For public works, women in West Bengal
complained most about drinking water and
roads, and much less abut education and
irrigation. Men complained most about
roads and irrigation and less about drinking
water. In Rajasthan, over half the complaints
made by women were about drinking water,
while very few complained about education.
Drinking water was also a big concern for
men in Rajasthan, but it was not as predomi-
nant an issue as for women. In contrast, men
complained much more than women about
roads and education. The pattern of com-
plaints dovetails with gender responsibilities
of men and women within the household and
their employment opportunities. Women
are responsible for collecting drinking water
and, in West Bengal, they are the main ben-
eficiaries of road works as they provide the
bulk of labor on roads. In Rajasthan, both
men and women work on roads, but men
use the roads more as they travel more fre-
quently in search of work.
This implies that giving women the right to
vote makes a difference. And indeed, Miller
(2008) shows that the introduction of women
suffrage in the United States was associated
with a decline in infant mortality. In most
countries however, women and men now
have the right to vote, although they do not
get elected. Thus, the next step to increase
women’s political power is to directly affect
their representation through quotas or reser-
vation policies.
To understand the effect of having
women as policymakers, Chattopadhyay and
Duflo (2004) study the reservation policy for
women in India mentioned above. A consti-
tutional amendment required states to both
devolve power over expenditure for local
public goods to rural village councils, and to
reserve a third of all council seats and coun-
cil presidencies for women. As a result, the
political representation and participation of
women has increased. A comparison of the
type of public goods provided in reserved
and unreserved village councils shows that
the gender of the council president impacts
investments. Women invest more in infra-
structure that is directly relevant to the
 Duflo: Women Empowerment and Economic Development
expressed development priorities of women.
In West Bengal, where women complained
more often than men about water and roads,
reserved councils invested more in water
and roads. In Rajasthan, where women com-
plained more often about drinking water but
less about roads, reserved councils invested
more in water and less in roads. This result is
in sharp contrast to the commonly held view
that positions reserved for women usually
go to the wife of the previous male leader
who effectively continues to rule through
his wife. The study also found that women
in reserved seats were less experienced, had
less ambition and lower prospects in poli-
tics, and were of lower socioeconomic status.
There is no evidence, however, that these
differences drove the results. It appears that
only the gender of the local leader explains
the different investment decisions in vil-
lages reserved for women. In other words, a
world run by women would look decidedly
different. Women leaders do seem to better
represent the needs of women. This is true
even in an environment (rural Rajasthan)
where women traditionally have very little
power, female literacy is very low, and where
many believe women leaders simply imple-
ment the wishes of their husbands. In fact,
these women are changing the realities on
the ground.
3.3 Empowering Women: What Policy
Instruments?
The South African situation, where a radi-
cal change in the political situation led to
the establishment of a permanent transfer
program of reasonably large magnitude for
adults who live with children, is relatively
rare. Absent these rare situations, are there
things policymakers can do to shift power
to women in their community or in their
families? Or do they need to rely, once again,
on a natural process of development (per-
haps accelerated a little bit by working on
improving access to education for girls or by
1071
encouraging an economy where women have
better access to market opportunities)?
The Chattopadhyay and Duflo (2004)
study in India suggests that it is indeed pos-
sible to shift power within the community
via a change in the electoral rules. Even if
women and men were to have different
preferences it is not sufficient to ensure
that women policymakers, particularly those
selected through quotas, would act any dif-
ferently than men. In fact, in a perfectly
democratic system, they would not, since
both men and women would try to do what
their constituents expect them to do in order
to be reelected. Additionally, even if we
thought women leaders did a better job of
understanding women’s needs, it would not
necessarily follow that policies designed to
guarantee greater representation by women
would be an effective way to achieve this
representation. For example, if women are
politically weak, they may find it difficult to
influence actual decisions. In India, the res-
ervation policy was (and is still, to a large
extent) met with a large measure of suspi-
cion: women would not be powerful leaders;
they would be eclipsed by their husbands; or
worst, local elite would take advantage of the
policy to reascertain their power against the
Panchayat. And yet the quota policy made a
real difference in actual outcomes.
Furthermore, this difference appears to be
durable, beyond the initial push, by affecting
attitudes. The study by Beaman et al. just
mentioned takes advantage of India’s reser-
vation policy, which specifies that, at each
election, one-third of the villages are ran-
domly selected and must elect a woman at the
head of the local council. They compare the
results of the implicit association test associ-
ating women and politics and the responses
to the speeches. They find that, in men’s
response to the speeches, all the bias against
women is erased (and in fact it becomes a
bias in favor of women). The bias against
women and leadership is also diminished.
1072 Journal of Economic Literature, Vol. L (December 2012)
Although, interestingly, the response to the
explicit question is not affected, people do
vote more for women after the seats lose
their reserved status. Beaman et al. (2011)
examine another form of attitude: the aspi-
rations that parents have for their children,
and that teenagers have for themselves. We
have already noted the gaps between boys
and girls. This gap is reduced by 0.15 stan-
dard deviations (one-fifth) in villages where
positions have been reserved for females for
ten years. Moreover, even in the absence of
any additional investment in education infra-
structure, this led into an increase in edu-
cational attainment for teenage girls, and a
reduction in the gaps with boys. There are
thus real effects on future gender equality
of giving more power to women then they
would naturally be getting, over and above
their ability to directly influence outcomes.
To what extent can similar “top–down”
policies affect the balance of bargaining
within the household? One avenue is the
legal environment. A woman’s options out-
side the marriage are also important for
bargaining within the marriage. If a woman
has no options outside of the marriage, and
both she and her husband know that divorce
is not possible or that in the case of divorce
she would be without any means of support,
then the husband can impose decisions on
the household—he can threaten his wife
with being left in a much worse position if
she does not yield. This suggests that the
institutional environment, such as the type of
divorce laws that exist, and the ease of remar-
rying or of finding other ways of supporting
oneself in case of divorce would affect deci-
sion making within households, even when
they do not end up divorcing. In Indonesia,
each spouse retains ownership of the
resources he/she brought into the marriage.
Thomas, Contreras, and Frankenberg (2002)
show that, even controlling for the house-
hold resources today, the resources each
spouse brought into the marriage continue
to affect decision making within households.
Specifically, children are in better health if
their mother brought more resources into
the marriage. Field (2007) finds that when
squatters in Peru were given a property title
for their land, fertility declined, but only in
areas where the prescription to put both the
woman’s and man’s name on the title was
effectively enforced.
In many developing countries, women are
very poorly protected in the case of divorce,
and stand to lose assets and the custody of
their children. Even when divorce laws exist,
it is frowned upon in society, and remarry-
ing is difficult. Surprisingly, there is little
research on the effect of these institutions
in developing countries. Work from the
United States suggests that the institutions
themselves matter for the outcomes within
the family. Chiappori, Fortin, and Lacroix
(2002) show that the rights of each spouse
in case of divorce (which vary by state and
have evolved differently in different states
over time) affect the labor supply decisions
among married couples. Angrist (2002)
shows that the female children of American
immigrants have a higher marriage rate and
lower labor supply (a sign of higher bargain-
ing power, in the United States) when there
were more males among the recent immi-
grants of their age cohorts. The reason is that
migrants tend to marry within their ethnic
group. Immigration-induced sex ratios that
are not balanced across sexes thus increase
the likelihood of remarriage of the sex that is
in short supply in case of a divorce, and thus
increases their bargaining power within the
household.
This suggests that policies seeking to
increase women’s welfare in case of divorce or
to increase women’s access to the labor mar-
ket may impact outcomes within the house-
hold, in particular child health. But would
these effects extend to temporary transfer
programs that do not profoundly change
the nature of the relative status of men and
 Duflo: Women Empowerment and Economic Development
women? For example, do conditional cash
transfers given to a mother for a period of a
few years have greater positive effects than if
they had been given to a father? The answer
is not obvious a priori. One might expect the
household to recognize that the transfers are
temporary. In this case, the transfer should
have no real effect on the recipient’s bargain-
ing power in the decision making, since the
household knows that very soon it will revert
to the prior situation, and that the woman
would be punished if she tried to extract
too much while her power lasted. In other
words, instead of trying to extract as much as
possible at any point in time, the household
members should insure each other, and tem-
porary transfers would not matter.
In practice, however, we observe that
household members do not seem able to pro-
vide efficient insurance to each other. They
may face the same constraints that informal
insurance between households is subject to
(moral hazard, hidden information, limited
commitment, etc.). In Côte d’Ivoire, women
and men grow different crops and are there-
fore affected differentially by the same rain-
fall. Some years are good for men, and some
years are good for women. One would expect
that the consumption in the family would not
be affected by the temporary, rain-induced
fluctuations in the share of income contrib-
uted by each partner. Yet Duflo and Udry
(2004) find that it is: in years when the
production of women’s crops is higher, the
household spends a bigger share of its bud-
get on food and also on private goods for
women. In years when the production of
male’s crops is higher, the household spends
a bigger share of the budget on alcohol and
tobacco and male private goods. Robinson
(2011) has a similar finding from an experi-
ment in Kenya, where husbands and wives
were given small transfers over a period of a
few weeks.
And indeed, the fact that conditional trans-
fer programs are targeted to women, rather
1073
than to men, seem to make a difference, even
when they are both temporary and small.
This was first demonstrated in a nice study by
Lundberg, Pollak, and Wales (1997), which
takes advantage of a change in the alloca-
tion of child benefits in the United Kingdom
from a tax credit to a direct payment in the
name of the mother. This transfer “from the
wallet to the purse” was associated with an
increase in the consumption of women’s and
children’s clothing in households with chil-
dren. Surprisingly, despite the fact that most
of the recent CCT are targeted to women,
there is very little evidence on whether this
is a factor in their effectiveness: since most of
them are just targeted to women, it is essen-
tially impossible to distinguish the impact of
the gender of the recipient from that of the
effect of the increased income, or the condi-
tionality.3 An exception is a recent study by
Benhassine et al. (2011), which exploits the
randomization of the gender of the recipi-
ent of a small conditional cash transfer in
Morocco. They found that, despite the fact
that the transfer was very small (it repre-
sented only about 3 percent to 4 percent of
beneficiary’s total consumption), households
reported spending it differently when the
woman was a recipient, and it had a some-
what larger effect on the education of young
children and girls when women received it.
It seems unlikely that the program directly
affected their bargaining power, and indeed
the study does not find any evidence that
other dimensions of women empowerment
have changed. However, they seem to have
been more likely to be in charge to spend
that money, and they were more likely to
spend it on education.
There are two important lessons from
these findings. First, even if they do not
bring about radical changes in women’s
3 Although some studies try to do this by using a func-
tional form for the impact of income on consumption
patterns.
1074 Journal of Economic Literature, Vol. L (December 2012)
status or in the way women are perceived in
society, policies targeted toward women can
have immediate consequences. Even smaller
changes that just give women control over a
small stream of resources affect at least the
way that this money is spent. Increasing
women’s control over resources, even in
the short run, will improve their say within
the household, which will not only increase
their welfare, but as research seems to have
shown repeatedly, child nutrition and health
as well. The second lesson is that families
do not function very well, since they are not
able to provide each other even basic neces-
sities. This means that we cannot rely on
the family to correct imbalances in society
in, for example, women’s property rights. If
the family is functioning well, the fact that
women have weak property rights could be
solved by transferring property from women
to men within the household, with the men
able to compensate the women for this. If a
woman cannot rely on her husband to con-
tinue to compensate her over time, however,
she will be reluctant to do so, and will pre-
fer to retain ownership of the property, even
if it is with weak ownership. In turn, this
may have consequences for the efficiency
of households’ production, may slow down
investment and growth, and may contribute
to keeping households in poverty.
3.4 Empowering Women: Is It Free Lunch
for Development Policy?
Women and men have different prefer-
ences, and accordingly, use their rights over
the permanent transfer to express those
preferences. However, the fact that women
have different preferences does not mean
that those preferences are always benevo-
lent, that they always favor “good things,”
favorable to development. In the South
Africa case, giving money to women helps
girls, but not boys. Edmonds (2006) uses
the same reform in South Africa to study the
impact of the pension program on education.
He compares school enrollment of adoles-
cents (ages 13 to 17) in families where there
is an elderly member who is eligible for the
pension and in families where there is an
ineligible elderly member. He finds that,
compared to those in families with ineligi-
ble elders, children are more likely to be in
school when they live with an eligible man
than with an eligible woman. Here again, we
find evidence that the identity of the income
holder matters. In this case, however, it is
when men receive the pension that they
make the decision favorable to well-being
and development.
In the case of women’s power, in the first
cycle, women shift the allocation of public
goods in a way that is unambiguously posi-
tive: while women invest more in drinking
water, they invest less in schools in West
Bengal, and less in roads in Rajasthan. This is
not surprising since they operate within a set
budget, and whatever resources are used in
one area need to come from another (in the
second cycle however, in West Bengal, they
seem to be doing more pretty much across
the board, though still disproportionately on
water). While this reform was clearly good
for women (whose preferences were now
taken into account), it is less clear whether it
is overall welfare improving or not: in order
to answer this question, one would need to
decide whether water was more important
than schools or roads. There is no obvious
way to do this calculation: in West Bengal,
we tested all water wells, and found most
of them to be completely clean. The extra
investment in drinking water infrastruc-
ture may thus have been primarily a matter
of convenience for women. Beaman et al.
(2009) collect detailed data on health out-
comes in households in these villages. They
find no impact of women’s leadership on
measure of infant health, but it does reduce
the time teenage girls spend collecting water.
This is not to say that this convenience is not
important. Devoto et al. (2011) found a large
 Duflo: Women Empowerment and Economic Development
effect of the connection to the water system
on self-reported happiness. How one values
this convenience, versus educating children
or better roads, is not something that econo-
mists are well placed to evaluate. A coherent
argument could be made that if women had
not had power for a long time, and thus the
investment in drinking water had been lag-
ging behind, it is probably efficient to invest
more there. But this reminds us that there is
no magic bullet, and that trade-offs remain
unavoidable.
That said, we have seen along the way
several examples where giving women more
power would have been efficiency enhanc-
ing: this was the case in Udry’s study in
Burkina Faso, where lower investments on
the women’s fields makes households inef-
ficient, or in Goldstein and Udry’s study in
Ghana, where because of weak property
rights, women were not investing. In the
West Bengal study, though the static gains
may depend on a policymaker’s objective
function, there seems to be more unam-
biguous dynamic gains, through the changes
in attitudes toward female policymakers
(allowing voters to choose among a larger
pool of candidates) and teenage girls (which
leads to a change in the households’ willing-
ness to invest in their human capital, with-
out reduction of the investments in boys).
However, policymakers often are too quick
to conclude that and allocation of resources
toward women, rather than men, would
always be efficiency enhancing. The exam-
ple of capital for microenterprise is a useful
cautionary tale.
A strong claim by microcredit organi-
zations (who disproportionately lend to
women, rather than to men) is that because
women are more likely to be shut off from
formal or informal sources of credit than
men, the returns to investing in their busi-
nesses should be larger than the returns to
investing in women. The problem is that if
women face other forms of unequal access to
1075
markets than men (from insecure property
rights, to constraints on their mobility or on
their time due to the obligation of child rear-
ing), access to inputs or cash alone may actu-
ally be more productive for women than for
men. Karlan and Zinman (2011) find no evi-
dence that microcredit loans given to women
have larger effects than those given to men.
Even more troubling are the results of De
Mel, McKenzie, and Woodruff (2009) that
directly test the proposition that the mar-
ginal returns to investing in female-owned
enterprises is larger than that of investing
in male-owned enterprises. They provided
cash or in-kind grants to randomly selected
businesses in Sri Lanka, and compared their
performance to that of a comparison group.
While they found a very large effect of the
grants given to men (suggesting returns to
capital of about 5 percent per month), the
effect of the grant on profit was essentially
zero for women. Fafchamps et al. (2011) rep-
licate the experiment in Ghana, with similar
results: the in-kind grants had some effect
on the profit of female-owned businesses,
but only for those businesses that were quite
profitable to start with. And for women, like
in Sri Lanka, the cash grants had no effect on
business profits whatsoever.
There are a number of potential expla-
nations for these results: women work in
different sectors, which may be more com-
petitive, and have less scope for expansion
and increases in profit; they may also be
less committed to their businesses (or their
husband may be less committed to letting
them grow it . . . ) if this is mainly seen as
a way to slightly increase household income
while keeping her busy. Finally, women
may know less about how to run a busi-
ness. The latter hypothesis has motivated
a large effort worldwide to combine micro-
finance services with business training,
particularly by the organization Freedom
From Hunger. Addressing one constraint
(credit) is not enough, the theory goes; the
1076 Journal of Economic Literature, Vol. L (December 2012)
gaps in education also need to be addressed.
Randomized impact evaluations tend to sug-
gest that the effects of such business train-
ing are quite limited (Karlan and Valdivia
2011; Drexler, Fischer, and Schoar 2010).
Field, Jayachandran, and Pande (2010)
shed interesting light on why that may be
the case, and in particular on how an entre-
preneurship opportunity (which could in
principle be training or finance) interacts
with existing, slow moving social norms. In
collaboration with SEWA, a microfinance
and support organization for self-employed
women, they conduct a randomized evalua-
tion of a relatively standard business training
program in Ahmedabad, Gujarat. They find
that the training had a positive impact on
business income for Upper Caste women,
but not for Lower Caste women (the least
restricted group in their sample) or Muslim
women (the most restricted group). They
interpret these differences as stemming
from a nonmonotonic effect of social norms
on the ability to acquire and apply business
knowledge. Upper Caste women might have
had more to learn than Lower Caste women
because they are less likely to be very active
participants in the market place before the
training. But Muslim women, whose mobil-
ity was even more restricted, were not able
to put that knowledge to use. This study pro-
vides a striking illustration of the trade-off
faced by policy: the fact that women face
multiple constraints means that relaxing just
one may not improve outcomes. In this case,
some policies that explicitly target women
(for training or for financial resources) may
actually not be effective or cost effective.
4.  Conclusion
Women’s empowerment and economic
development are closely interrelated. While
development itself will bring about wom-
en’s empowerment, empowering women
will bring about changes in decision mak-
ing, which will have a direct impact on
development. Contrary to what is claimed
by some of the more optimistic policymak-
ers, it is, however, not clear that a one-time
impulsion of women’s rights will spark a vir-
tuous circle, with women’s empowerment
and development mutually reinforcing each
other and women eventually being equal
partners in richer societies.
On the one hand, economic development
alone is insufficient to ensure significant
progress in important dimensions of women’s
empowerment, in particular, significant prog-
ress in decision-making ability in the face of
pervasive stereotypes against women’s abil-
ity. On the other hand, women’s empower-
ment leads to improvement in some aspects
of children’s welfare (health and nutrition, in
particular), but at the expense of some oth-
ers (education).
This suggests that neither economic devel-
opment nor women’s empowerment is the
magic bullet it is sometimes made out to be.
In order to bring about equity between men
and women, in my view a very desirable goal
in and of itself, it will be necessary to continue
to take policy actions that favor women at the
expense of men, and it may be necessary to
continue doing so for a very long time. While
this may result in some collateral benefits,
those benefits may or may not be sufficient
to compensate for the cost of the distortions
associated with such redistribution. This
measure of realism needs to temper the posi-
tions of policymakers on both sides of the
development/empowerment debate.
References
Abrevaya, Jason. 2009. “Are There Missing Girls in the
United States? Evidence from Birth Data.” Ameri-
can Economic Journal: Applied Economics 1 (2):
1–34.
Ali, Rabia, Jishnu Das, Damien de Walque, Kenneth
L. Leonard, Mattias Lundberg, and David Peters.
2011. “Patterns of Health Care Interactions in Seven
Low and Middle-Income Countries.” Background
Paper for the World Development Report 2012.
4 Measuring Corruption

Measuring corruption is the primary challenge in the empirical literature. With-

out robust measures, the theories cannot be tested, the magnitudes of corruption
cannot be quantified, anticorruption policies cannot be assessed, and so forth.

However, measuring corruption is challenging, and even today, relatively few

studies are able to credibly describe the extent of the problem. As Bardhan

(1997: p. 1320) notes in his review of corruption, “our approach in this paper is

primarily analytical and speculative, given the inherent difficulties of collecting

(and hence nonexistence of) good empirical data on the subject of corruption.”

The measurement challenges are driven by several problems. First, the very

fact that corruption is illegal makes people reluctant to talk about it for fear of

getting into trouble and, possibly, for shame. Second, the existing literature—

because of the theory it draws on and how it defines corruption—takes a trans-

actional view of corruption. Measurement means finding out who bribed who

and by how much, which is inherently harder to quantify. Third, the traditional

narrow focus on monetary transactions also makes it more difficult.19 When a

government official benefits by stealing “time”—she decides not to show up for

work—random spot checks can be very revealing (e.g. Chaudhuri et al. 2006).

Finally, measurement systems will evoke responses that make the measurement

system less reliable. If the government has a monitoring system in place, people

adjust to it and find ways around it. As a result, these measurement systems

will underestimate corruption.

Despite these difficulties, many early attempts at measuring corruption relied

on rich qualitative data and were occasionally backed up by numbers. These

studies gave the first real evidence about the channels through which corrup-
19 There are exceptions. For example, Tran (2008) gathers a comprehensive set of internal

bribery records from a firm in Asia to document the cost of bribe payments over time.

41
tion occurred and possible methods to eliminate it. Wade (1982) in particular

provides a detailed description of how irrigation engineers reap revenue from

the distribution of water and contracts in a village in South India. The most

fascinating aspect of this study was the documentation of a fairly formal system

in which the engineers redistributed revenue to superior officers and politicians.

To obtain transfers to lucrative posts, the junior officers paid bribes based on

expectations about the amount of bribe money that can be collected from the

post. Thus, the value of the bribe payment in the transfer process was higher

for jobs that had greater potential for bribe extraction. In essence, the ability

to take bribes in a job induces bureaucrats to bribe others to get it. This highly

detailed study gives a glimpse into the pervasiveness of corruption in this area

and the mechanics of how it operates. It also illustrates how corruption is in-

terconnected throughout the entire organization and raises the possibility that

rather than trying to remove one aspect of corruption, it may be necessary to

invoke coordinated policies to stamp it out throughout the system. However,

like all case studies, the study raises questions of generalizability. Is there as
much corruption in other contexts? Under what set of circumstances do these

systems come into being?

Other early studies focused on anticorruption policies. For example, Klitgaard

(1988) provides several case studies of successful elimination of corruption, such

as in the Hong Kong Police Force, Singapore Excise Department, and the Bureau

of Internal Revenue of the Philippines. In all these cases, the levers used are

intuitive from an agency theory perspective—more intense or better monitoring,

replacing individual actors, and so forth. They also all seem to involve a person

at the top of each institution who was eager to implement these changes. On the

one hand, these cases represent a vindication of an agency theory of corruption.

On the other hand, they raise the more fundamental question: if these levers

42
for eliminating corruption are within the choice sets of governments, why are

they not implemented? Although they spark hope that corruption can be fought,

these examples leave lingering questions about why conditions were ripe for these

interventions, but not for those elsewhere. Is what we observe due to particular

institutional factors in these settings or to more generalizable features of how

governments function? What particular combinations of institutions, policies, or

conditions would lead to similar steps being taken elsewhere? Should we expect

the same consequences of similar anticorruption policies in different settings?

To address these inherent problems of case studies, the next attempts to measure

corruption tried to provide consistent measures of corruption across countries.

However, given the difficulty of inducing people to talk about corruption, these

measures focus on collecting perceptions of corruption rather than on the actual

bribes paid or the actual theft of resources. This perceptions based approach

has been carried out at quite a large scale, generating interesting cross-country
and cross-time datasets. The Economist Intelligence Unit created one of the

first such datasets.20 The data collection effort consisted of factor assessment

reports that were filled in by their network of correspondents and analysts. The

data are then aggregated into risk factors for about 70 countries. The report

included, for example, a question where the respondents rate “the degree to

which business transactions (in that country) involve corruption or questionable

payments” on a scale of one to ten, where a high value implies good institutions

(Mauro 1995: p.684). Other cross-country measures on subjective perceptions

of corruption followed, including the Control of Corruption measures in the

World Bank Governance Indicators (a description of the measures can be found

in Kaufmann et al. 2004), and measures by Transparency International.21 Each

20 These data also called the Business International Indicators.
21 Papers that use perception-based measures of corruption in cross-country regressions in-
clude Mauro (1995), Knack and Keefer (1995), LaPorta et al. (1999), Rauch and Evans (2000),
Treisman (2000), Fisman and Gatti (2002), Adsera et al. (2003), Fredriksson and Svensson
(2003), and Persson et al. (2003).

43
of these indicators uses a different methodology with its own advantages and

disadvantages, which we do not discuss here due to space constraints.22 The real

advantage of such data is their breadth, which allows one to run large correlation

studies. Mauro (1995) is an often-cited example of this kind of study. He uses

the Economist Intelligence Unit measures in a cross-country growth regression

equation to study the relationships between economic growth, corruption, and

red tape.

While these perception-based measures of corruption provided evidence on which

countries tend to report more or less corruption, many have pointed out their

limitations. First, as Rose-Ackerman (1999) noted, it is unclear what the cor-

ruption indices actually mean and what a particular rank implies about the

type and level of corruption in a country. For example, in the Transparency

International Corruption Perceptions Index for 2008, Brazil, Burkina Faso, Mo-

rocco, Saudi Arabia, and Thailand all have the same index value. However,
the value does not tell us what the form of corruption in these countries en-

tails, nor does it indicate whether the types of corruption observed in these

very different countries have different efficiency implications. For example, in

the theoretical framework developed above, bribe taking can actually promote

efficiency, if the problem is that the government is committed, for political or

ideological reasons, to set the price cap below what it should be. Moreover,

corruption often emerges as a result of government interventions designed to

deal with some other distortion (see Section 3). These countries may have very

different problems—why would the gap between l and yL or between h and yH

be the same in Saudi Arabia and Burkina Faso, given their very different levels

of per capita wealth? Or, corruption may be a result of the government’s at-

tempt to fight some other form of misbehavior by its bureaucrats—for example,

22 Svensson (2005) provides a thorough description of the differences among the most com-

mon cross-country indicators of corruption.

44
in Case I (Section 3.3.1), absent a price cap, the bureaucrat might choose the

monopoly outcome and many slots may be wasted. A price cap might move the

outcome to more visible malfeasance by the bureaucrats (red tape and bribe

taking), but less misallocation and less inefficiency overall. Greater corruption

in one country could simply be a reflection of a greater willingness to fight cor-

ruption in that country. Because they do not supply information about the

sources of corruption, these corruption indices actually tell us little about what

types of governance interventions would help deal with these problems, or even

whether we should reward or praise governments that have less corruption by

these measures.

Second, perceptions may indicate little about the actual reality of situations

because they are influenced by the way we see everything else. For example,

perhaps when the economy is good, we perceive less corruption because we are

more satisfied with the government. Olken (2009), for example, compares the
perception of corruption in a roads project with actual missing expenditures. He

finds that although there is real information in perceptions, reported corruption

is not particularly responsive to actual corruption. For example, “increasing the

missing expenditures measure by 10 percent is associated with just a 0.8 percent

increase in the probability a villager believes that there is any corruption in the

project” (p.951). He also finds that the bias in perceptions is correlated with

demographic characteristics, implying that perceptions of a nonrandom sample

of the population may not adequately reflect real corruption levels. This bias is

particularly problematic, as many perception measures are not necessarily based

on random samples. For example, the measures from the Economist Intelligence

Unit are based on the perceptions of foreign analysts, who may have different

perceptions of corruption than the average citizen of the country. Finally, and

perhaps probably most importantly for our purposes, these data are most useful

45
for cross-country or cross-geography analysis. They are less useful for testing

micro theories of corruption.

Newer methods for measuring corruption have been developed to solve the small-

sample problem and move to more concrete measures of corruption. The first set

of methods focus on refining survey and data collection techniques to improve

the ability to assemble data on self-reported bribes and service delivery quality.

For example, Svensson (2003) analyzes a dataset that provides information on

bribes paid by firms in Uganda. To encourage truth telling in the survey, it was

conducted by a trusted employers’ association; it also asked carefully worded,

hypothetical questions to measure corruption.23 Hunt (2007) uses the Interna-

tional Crime Victims Surveys and Peruvian Household surveys, both of which

contain information on bribes to public officials if the individual has been the

victim of a crime.24 Other studies collect prices paid for services that should be

free. For example, Banerjee et al. (2004) collects fees paid at government health
centers in India (that should mostly be free), while Atanassova et al. (2008)

collect data on prices paid and quantities received from the public distribution

system in India and compare them to the official prices for these commodities.

The main benefit of these methods is a move to measures that have actual mean-

ings. For example, using the measure in Banerjee et al. (2004), we can estimate

the bribe amounts paid at the health centers and use this information to under-

stand how the bribes affect the allocation of health services. We can also use

it as an outcome measure to study the effectiveness of anticorruption policies

in government health centers. However, these types of measures are limited if

individuals are underreporting bribes, and particularly if this underreporting is

23 For example, “Many business people have told us that firms are often required to make

informal payments to public officials to deal with customs, taxes, licenses, regulations, and
services, etc. Can you estimate what a firm in your line of business and similar size and
characteristics typically pays each year?”
24 Mocan (2008) also use the ICVS to determine what characteristics were associated with

greater corruption.

46
biased by corruption levels. Moreover, most of the time, these measures are

often limited to petty corruption, because it is difficult to ask individuals about

the larger bribes that they may have paid. Indeed, in many of these cases the

reports are assumed to be reliable precisely because the agents do not know

how much they should have paid, and therefore, do not see themselves as pay-

ing bribes. It is also clear that people might be more willing to reveal bribes

that they paid in settings where the good they receive is abundant (so that they

are not displacing anyone else by paying the bribe).

The second method is the use of physical audits of governmental processes. For

example, Chaudhury et al. (2006) conducted a multicountry study of teacher

and health worker absence, where they performed spot checks to determine

whether bureaucrats were showing up for work.25 Similarly, Bertrand et al.

(2007) followed individuals through the process of obtaining a driver’s license

in India, and recorded all extra-legal payments made and the rules that were
broken in exchange for these extra-legal payments. Barron and Olken (2007)

designed a study in which surveyors accompanied truck drivers on 304 trips

along their regular routes in two Indonesian provinces. The surveyors observed

the illegal payments the truck drivers made to the traffic police, military officers,

and attendants at weighing stations.

One of the key challenges to the audit studies is whether the observed outcomes

actually reflect corruption rather than some less intentional form of bureaucratic

ineffectiveness, because often there is no smoking gun (bribe overtly paid, job

left entirely unattended, etc.). For example, Bertrand et al. (2007) find that

there is a misallocation of licenses—people who cannot drive are able to ob-

tain them. Could this be due to an “overloaded” bureaucrat who does not have

time to screen license candidates or due to an “incompetent” bureaucrat who

25 Other such studies include Duflo et al. (2008), who measure teacher absence, and Banerjee

et al. (2007), who measure absenteeism among medical providers.

47
cannot distinguish between a good and bad driver? To understand this issue,

Bertrand et al. (2007) collect detailed quantitative and qualitative data on how

the bureaucrats behave during the licensing process. They document extreme

behaviors (e.g. simply never administering a driving test) which would be hard

to label as “incompetence.” Similarly, Duflo et al. (2008) measure teacher absen-

teeism in rural India using audit methods. Does the fact that teachers often do

not come to school imply that they are consciously breaking the rules for private

gain? Or are lives of these teachers so difficult that they just cannot make it to

school often enough, despite trying as hard as they can? The research design

provides information that allows them answer these questions. Specifically, they

evaluate a program that monitors the teachers and provides incentives to the

teachers based on their attendance. They find that teachers are very responsive

to the incentives. That teachers respond to the incentives so strongly implies

that the teachers were previously ignoring the rules and were not incapable of

attending.

We refer to the third technique as “cross-checking”. The idea behind it is to

compare official records of some outcome with an independently collected mea-

sure of the same outcome. One example of cross-checking is to compare how

much money was released to the bureaucrat with how much the ultimate benefi-

ciaries of the funding report have received. For example, Reinikka and Svensson

(2005) compare data from records on central government disbursements and a

public expenditure tracking survey to measure dissipation in a school capita-

tion grant in Uganda. They find that the average school received only about

20 percent of central government spending on the program. Fisman and Wei

(2004) compare Hong Kong’s reported exports to China at the product level

with China’s reported imports from Hong Kong to understand the extent of

tax evasion. Another way to conduct a cross-check would be to collect records

48
from the bureaucrat documenting how the government resources were spent in

achieving a task and then compare them with an objective measure of how much

it should have cost to conduct the task. The difference between the two mea-

sures is, then, the estimate of how much was “stolen.” Olken (2007) uses this

method. He calculates corruption in the context of road projects by comparing

the actual expenditures reported with an independent measure of what it should

have cost to build a road of that particular quality. To obtain the independent

measure, he sampled each road to determine the materials and labor used, and

then used local prices to cost these items. He finds that, on average, about 24

percent of expenditures across the project villages were missing.26

As with any other method, this one has both strengths and limitations. Its

innovation lies in the fact that it creatively allows the measurement of dissipated

government resources without asking the actors involved if they have paid or

received an actual bribe, reducing the chance of underreporting or misreporting.

Thus, it often allows us to move past petty corruption and perhaps look at larger

scale corruption. However, it is difficult to understand whether the dissipated

resources are actually corruption or simply mismeasurement in the indicators or

even just a sign of bureaucrat incompetence. For example, in Olken (2007), it

remains possible (though unlikely) that road quality is mismeasured or that the

bureaucrats are not good at building roads. It is possible (though again unlikely)

that the missing resources indicate that the bureaucrats are trying to reallocate

funds to better uses. For example, in Reinikka and Svensson (2005), it is possible

(although again unlikely) that the resources that should have gone to the schools

capitation program were actually spent on services that the community deemed

more important, and did not end up in the pockets of government officials.
26 Other examples of cross-checking in the developing world include the Hsieh and Moretti

(2006) estimate of bribes in Iraq’s Food for Oil Program; Olken (2006) and Atanassova et al.
(2008) measure theft in food distribution programs using the same method; and Di Tella and
Schargrodsky (2003) use it to measure corruption in hospital procurements.

49
Would this result have necessarily been bad?

One way to get around these concerns is to look for correlations (motivated by

theory) between the extent of dissipation and some other variable. For exam-

ple, to show that the differences in reported shipments is corruption and not

just mismeasurement in the shipments, Fisman and Wei (2004: p.471) docu-

ment that the differences are “negatively correlated with tax rates on closely

related products, suggesting that evasion takes place partly through misclassifi-

cation of imports from higher-taxed categories to lower-taxed ones, in addition

to underreporting the value of imports.27

In addition to these methods recently used in the literature, there are several

innovative methods being explored in current studies. Although some of this

work is not yet published, such methods will surely contribute to the tools
available for studying corruption. For example, Banerjee and Pande (2009)

attempt to use second-hand data on which politicians have gotten rich since

they entered politics to identify those who have profited from corruption.28 They

find that this measure correlates strongly with other, more direct (perceptual),

measures of corruption (e.g. the answer to the question "do you think the

politician used his office for private gain?"). Cai et al. (2009) use predictions of

auction theory to argue that certain types of land auctions in China are used to

promote collusion between the auctioneer and the participants, for their mutual

benefit.
27 Duggan and Levitt (2002) provide an interesting example of cross-checking in sumo

wrestling. They basically show that a wrestler has a higher probability of winning than
expected when the match is key to his rank. To distinguish match throwing from effort, they
use theory as a guide: the effect goes away when there is greater media scrutiny, suggesting
that it is not effort. Moreover, the next time the same two wrestlers meet, the opponent is
more likely to win, suggesting that throwing future matches is a form of the bribe paid for
winning a key match. Similarly, Atanassova et al. (2008) cross-check receipt of a BPL card
(which in India identifies someone to be poor and allows them access to a set of redistributive
programs) against actual income levels. They correlate the error rate with features such as
caste of the recipient, their place in certain social networks etc., and argue that the correlations
are what a simple theory of corruption would predict.
28 Di Tella and Weinschelbaum (2007) provide a theoretical framework for thinking about

unexplained wealth as an indicator for corruption.

50
All these methods pick up the direct correlates of corruption, such as bribes and

absence from work, rather than its more indirect ramifications. In some cases,

such as teacher absenteeism, the direct consequence may be the most important.

However, the point of our framework is to argue that in many situations, the

bribe may just be the tip of the iceberg, with the more serious repercussions

showing up in terms of misallocation and red tape. The next section discusses

an example of empirical research that tries to capture the bureaucrat’s entire

decision process and its various ramifications.

5 Understanding the Structure of Corruption

Most of the empirical research has been based on measuring the extent of bribery

or shirking and on how incentives affect these behaviors. Bertrand et al. (2007)

differs from much of the empirical research by focusing on the entire resource

allocation problem faced by the bureaucrat and therefore looking beyond bribe

taking as the measure of welfare. The basic strategy of the paper is to experi-

mentally vary the underlying types a bureaucrat faces and use the bureaucrat’s

responses to infer their chosen allocative mechanism.

Specifically, Bertrand et al. (2007) compare three randomly chosen groups of

license candidates. The first group was told to obtain a license as usual, the

second group was given a large incentive to obtain the license in the minimum

legal time allowed (30 days), and a third group was offered free driving lessons.

In our model the second group represents a situation where h and l have both

increased by the same amount. The third group represents a situation where
some of those who were low types will now become high types.

The driver’s license case corresponds to our Case IV (see Section 3.3.3). To

reduce the number of possible cases, assume that yH = h < l = yL (which, as

51
that we advocate in this paper.

6 Combating corruption

There is a large and growing empirical literature that studies the effect of ef-

forts to fight corruption. For example, Fisman and Miguel (2007) find that an

increase in punishments for parking violations in New York City reduced the

violations among the set of diplomats, who were most likely to violate the rules.

Using experimental techniques, Olken (2007) finds that theft in road projects

is greatly reduced by raising the probability of being caught. Banerjee et al.

(2007) and Duflo et al. (2008) find that strengthening incentives reduces ab-

senteeism. Using a natural experiment in Buenos Aires hospitals, Di Tella and

Schargrodsky (2003) find effects of both more stringent monitoring and higher

wages on procurement prices.

Several more recent papers on this topic have also tried to go a step further and

understand whether a reduction in corruption due to monitoring and incen-

tives improves the final allocation of services. In the Di Tella and Schargrodsky

(2003) study, less bribe taking means better procurement prices, which is the

outcome of interest. Olken (2007) looks at the effect of auditing not only on

theft but also on the quality of the roads that were built. Duflo et al. (2008)

study whether incentives can create distortions due to multitasking. In other

words, they are concerned that to complete the task as specified by the incen-

tives, the agent reduces his effort along other dimensions. Specifically, they ask

whether providing incentives for teachers to attend school will cause the teach-
ers to compensate by teaching less. To answer this question, they measure not

only teacher attendance as the final outcome but also the learning levels of the

children. They find that the multitasking problem is certainly not large enough

56
to outweigh the benefits of better incentives.

6.1 Institutional Structures for Monitoring

Corruption exists because there are not enough monitoring and incentives to

eliminate it. What then determines the extent of monitoring and incentives?

One challenge of looking at this issue empirically is essentially methodologi-

cal. What should we assume about the extent to which these rules are the

product of optimization by the government? Governments may make rules to

combat malfeasance by government officials, but how well does it understand

the consequences of these rules for corruption?

There are two possible approaches to answering these questions. One is what

might be called the political economy approach. This approach is taken in the

theory section here (Section 3). We assume some preferences for the government
and figure out how the rules and the nature and extent of corruption should vary

as a function of the underlying economic environment if the government were

optimizing based on those preferences. We could then look for evidence for the

comparative static implications of that model and jointly test the model and

our assumption about what is being optimized.

The alternative is to assume that the forces of political economy, although im-

portant, leave a significant amount undetermined, and as a result, changes in

rules can often arise as pure organizational innovations, without changes in the

fundamentals. This approach leads naturally to an experimental approach to

studying the effects of the rules.

Banerjee et al. (2001) implement a version of the first approach in the context

of the governance of localized sugar cooperatives. They assume that the coop-

erative maximizes a weighted average of the profits of the various principals of

these cooperatives —the different types of member farmers who grow the cane

57
that the cooperative turns into sugar —taking into account the desire of the

management of the cooperative to siphon off as much of the profits as possible.

Banerjee et al. (2001) then generate a set of comparative statics predictions

that map the pricing of cane and the productivity of the cooperative onto the

underlying mix of farmers in the cooperative. These implications are then tested

and seem to be broadly confirmed.

However, corruption in these cooperatives is essentially private sector corrup-

tion, embodied primarily by the underpricing of cane. We have yet to come

across a paper that combines this political economy approach with the more

complex manifestations of corruption that were identified above.

More importantly, in many instances, theories of political economy are simply

too incomplete to be useful guides to what rules are actually chosen. The objec-
tive of a specific government at a specific point in time is some complex product

of its long-term goals and its short-term compulsions. Moreover, the way it

chooses rules must take into account the compulsions of all future governments.

Although there is an interesting and growing literature on this subject, it is not

clear that it is ready to be taken to the data.

It is also not clear how much governments understand about the consequences

of the various policy choices they make or about the policy options available

to them. A more evolutionary approach to policy change, where changes hap-

pen because political actors are trying to solve some "local" problem but the

solutions often have unanticipated and often global consequences, may be more

descriptively accurate. Certainly this approach fits better with the kinds of

stories that one hears about how change came to China.

An advantage of this second approach is that it permits us to think of policy

changes as organizational innovations, that are therefore at least initially ex-

ogenous in a way that technological innovations are usually thought of as being

58
exogenous. That is, the assumption is that the need to solve problems is a

product of various forces of society, but the adoption of a particular solution at

a particular point in time, is less so.30 The approach also makes it clear that

governments might choose bad rules (rules that go against its own objectives),

because it does not understand the consequences of its choices.

Several recent empirical exercises start from this point of view. For example,

Besley et al. (2005) find that, in South India, there exists a relationship between

holding village meetings (i.e., more community participation in the process)

and better allocations of Below Poverty Line cards, which provide privileged

access to subsidies and government services. Bjorkman and Svensson (2009)

study decentralization in an experimental context. Rather than imposing all

centralized rules on health centers, community meetings are held to decide the

most important rules that health centers should follow, and the mechanisms for

the community to monitor the health centers. They find huge impacts: infant
mortality rates were cut by one-third. However, Banerjee et al. (2008) evaluate

a similar decentralization program in India and find that it performs no better

than the civil service based system of monitoring teachers.

It is difficult to make much of these vastly different empirical findings, given

that we do not have a particularly good theory of how decentralization affects

corruption and the distortions associated with corruption.31 How does decen-

tralization change the kinds of rules that are optimal and the way in which they

are violated? How does the exact nature of decentralization factor into all this?

As it is, the presumption behind the empirical literature is that decentraliza-

tion is a shift of control rights into the hands of those who have more local

information. The basic notion is that the community now has more information

and therefore can limit the extent of malfeasance by the bureaucrat. We do see
30 Banerjee (2002) discusses many of these methodological issues in greater detail.
31 Bardhan and Mookherjee (2006) provide a rare exception.

59
some evidence that the easy availability of information matters. In the driver’s

license experiment, Bertrand et al. (2007) report that there are two obstacles

that bribes cannot get around: one is the requirement of showing some proof

of address and the other is the requirement of waiting at least 30 days after

making the initial application for a learner’s permit. Neither of these seem as

important as being able to drive, especially in India, where the driver’s license

is not always accepted as an identification. However, violations of these rules

are easy to observe, whereas the inability to drive properly is something that

requires another test to verify. Therefore, these are the rules that are enforced.

However, the answer to the question of what rules are violated in what way

must also depend on who exercises which type of control rights and what infor-

mation reaches whom, which all turns on the exact model of decentralization

adopted. More generally, theoretical work mapping the effects of alternative

organizational forms on the choice of rules and corruption outcomes must be a

high priority if this literature is to make progress.

7 Conclusion

Where should the literature go next? We have already discussed a number of

the gaps in the literature. In Section 6, we discussed the need to more tightly

model “learning the system.” Here we discuss other important gaps.

7.1 Corruption and Competition

Thinking about organizational forms naturally leads to the role of competition

in reducing corruption, as emphasized by Rose-Ackerman (1978). The way we

modeled corruption takes as given the idea that the assignment of the appli-

cants to the bureaucrats is random. This effectively places the applicants and

60