.

// MODELOS PARA VARIABLES DEPENDIENTES ORDINALES EN STATA

.
. // Regression Models for Categorical Dependent Variables - 2nd Edition
. // Chapter 5 - Models for Ordinal Outcomes
. // Long and Freese 27jul2005
. // Comentarios en espaol: Javier Aparicio - 3nov2008
. // (Ver DO FILE al final de este documento)
------------------------------------------------------------------------------log: D:\MyDocs\Metodos08\Long&FreeseSpost9\ch5_ordinal.log
log type: text
opened on: 30 Nov 2008, 20:45:08
. version 9
. * para que stata use sintaxis de version 9 (en caso de tener una versin
posterior)
. set scheme s2manual
// modo monocromtico para grficas
. set more off
// para que el output del DO file no haga pausa al
correr
. * Ejemplo de variable ordinal:
. * Actitudes frente a las madres que trabajan
. * Var dep: Esta de acuerdo o no con la frase:
. * "Las mamas que trabajan pueden mantener relaciones calidas con sus hijos"
.
. sysuse ordwarm2, clear
(77 & 89 General Social Survey)
. desc warm yr89 male white age ed prst
storage display
value
variable name
type
format
label
variable label
------------------------------------------------------------------------------warm
byte
%10.0g
sd2sa
Mom can have warm relations
with child
yr89
byte
%10.0g
yrlbl
Survey year: 1=1989 0=1977
male
byte
%10.0g
sexlbl
Gender: 1=male 0=female
white
byte
%10.0g
race2lbl
Race: 1=white 0=not white
age
byte
%10.0g
Age in years
ed
byte
%10.0g
Years of education
prst
byte
%10.0g
Occupational prestige
. sum warm yr89 male white age ed prst
Variable |
Obs
Mean
Std. Dev.
Min
Max
-------------+-------------------------------------------------------warm |
2293
2.607501
.9282156
1
4
yr89 |
2293
.3986044
.4897178
0
1
male |
2293
.4648932
.4988748
0
1
white |
2293
.8765809
.3289894
0
1
age |
2293
44.93546
16.77903
18
89
-------------+-------------------------------------------------------ed |
2293
12.21805
3.160827
0
20
prst |
2293
39.58526
14.49226
12
82
.
. tab warm
. *(Working moms can have warm relations with children)
Mom can |
have warm |
relations |
with child |
Freq.
Percent
Cum.
------------+-----------------------------------

1SD |
297
12.95
12.95
2D |
723
31.53
44.48
3A |
856
37.33
81.81
4SA |
417
18.19
100.00
------------+----------------------------------Total |
2,293
100.00
.
.
.
.
.

* Var. Dep.: Est de acuerdo o no con la frase?

* "Las mams que trabajan pueden mantener relaciones calidas con sus hijos"
// Un modelo de regresin logstico ordinal
ologit warm male white age ed prst, nolog

Ordered logistic regression

Log likelihood = -2866.5752

// muestra completa
Number of obs
LR chi2(5)
Prob > chi2
Pseudo R2

=
=
=
=

2293
258.39
0.0000
0.0431

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------male | -.7499562
.0783083
-9.58
0.000
-.9034377
-.5964747
white | -.4127267
.1180436
-3.50
0.000
-.6440879
-.1813655
age | -.0206757
.0024598
-8.41
0.000
-.0254967
-.0158547
ed |
.0789344
.0158731
4.97
0.000
.0478238
.1100451
prst |
.0066996
.0032863
2.04
0.041
.0002585
.0131408
-------------+---------------------------------------------------------------/cut1 | -2.466022
.2386009
-2.933671
-1.998373
/cut2 | -.6564482
.2329347
-1.112992
-.1999045
/cut3 |
1.215503
.2335542
.7577454
1.673261
-----------------------------------------------------------------------------. * la opcion nolog omite el output de las iteraciones
. ologit warm male white age ed prst if yr89==1, nolog
Ordered logistic regression
Log likelihood = -1083.2871

// muestra restringida

Number of obs
LR chi2(5)
Prob > chi2
Pseudo R2

=
=
=
=

914
91.02
0.0000
0.0403

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------male | -.7589487
.1265685
-6.00
0.000
-1.007018
-.510879
white | -.3579802
.18615
-1.92
0.054
-.7228275
.0068671
age | -.0227585
.0038285
-5.94
0.000
-.0302621
-.0152548
ed |
.0421346
.025821
1.63
0.103
-.0084737
.0927429
prst |
.0094713
.0049673
1.91
0.057
-.0002645
.019207
-------------+---------------------------------------------------------------/cut1 | -3.528722
.4055
-4.323487
-2.733957
/cut2 | -1.380973
.3850552
-2.135668
-.6262787
/cut3 |
.6297378
.3821996
-.1193596
1.378835
-----------------------------------------------------------------------------.
. * comparando ORDINAL LOGIT con ORDINAL PROBIT
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. estimates store ologit

// guarda los estimadores ologit

. oprobit warm yr89 male white age ed prst, nolog

Ordered probit regression
Log likelihood =

-2848.611

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
294.32
0.0000
0.0491

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.3188147
.0468519
6.80
0.000
.2269867
.4106427
male | -.4170287
.0455459
-9.16
0.000
-.5062971
-.3277603
white | -.2265002
.0694773
-3.26
0.001
-.3626733
-.0903272
age | -.0122213
.0014427
-8.47
0.000
-.0150489
-.0093937
ed |
.0387234
.0093241
4.15
0.000
.0204485
.0569982
prst |
.003283
.001925
1.71
0.088
-.0004899
.0070559
-------------+---------------------------------------------------------------/cut1 | -1.428578
.1387742
-1.700571
-1.156586
/cut2 | -.3605589
.1369219
-.6289209
-.092197
/cut3 |
.7681637
.1370564
.4995381
1.036789
-----------------------------------------------------------------------------. estimates store oprobit

// guarda los estimadores oprobit

. estimates table ologit oprobit, b(%9.3f) t label varwidth(30)

-------------------------------------------------------Variable | ologit
oprobit
-------------------------------+-----------------------warm
|
Survey year: 1=1989 0=1977 |
0.524
0.319
|
6.56
6.80
Gender: 1=male 0=female |
-0.733
-0.417
|
-9.34
-9.16
Race: 1=white 0=not white |
-0.391
-0.227
|
-3.30
-3.26
Age in years |
-0.022
-0.012
|
-8.78
-8.47
Years of education |
0.067
0.039
|
4.20
4.15
Occupational prestige |
0.006
0.003
|
1.84
1.71
-------------------------------+-----------------------cut1
|
Constant |
-2.465
-1.429
|
-10.32
-10.29

-------------------------------+-----------------------cut2
|
Constant |
-0.631
-0.361
|
-2.70
-2.63
-------------------------------+-----------------------cut3
|
Constant |
1.262
0.768
|
5.39
5.60
-------------------------------------------------------legend: b/t
. * crea una tabla con los modelos previos con tres decimales, t-stats y
etiquetas de variables de hasta 30 caracteres
.
. // Pruebas de hiptesis para coeficientes individuales
. * Wald test
. ologit warm male yr89 white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. test male
( 1)

// test similar al z-score de la tabla

[warm]male = 0
chi2( 1) =
Prob > chi2 =

87.30
0.0000

.
. display "z*z=" -9.343*-9.343
z*z=87.291649
. return list
scalars:
r(drop)
r(chi2)
r(df)
r(p)

=
=
=
=

0
87.30028866314545
1
9.32342950928e-21

. display "chi2=" r(chi2) "; sqrt(chi2)= " sqrt(r(chi2))

chi2=87.300289; sqrt(chi2)= 9.3434623
. * LR (likelihood ratio) test
. ologit warm yr89 male white age ed prst, nolog // modelo "completo"

Ordered logistic regression

Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. estimates store fullmod
. * estimate constrained model
. ologit warm yr89 white age ed prst, nolog
Ordered logistic regression
Log likelihood =

-2889.278

// un modelo restringido o anidado

Number of obs
LR chi2(5)
Prob > chi2
Pseudo R2

=
=
=
=

2293
212.98
0.0000
0.0355

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5486813
.0796222
6.89
0.000
.3926246
.704738
white | -.4248955
.1184223
-3.59
0.000
-.656999
-.192792
age | -.0197197
.0024473
-8.06
0.000
-.0245164
-.014923
ed |
.0659969
.0159375
4.14
0.000
.03476
.0972337
prst |
.0046902
.003286
1.43
0.153
-.0017502
.0111306
-------------+---------------------------------------------------------------/cut1 | -2.066166
.2337297
-2.524268
-1.608064
/cut2 | -.2697128
.2290783
-.718698
.1792725
/cut3 |
1.566067
.2312319
1.112861
2.019273
-----------------------------------------------------------------------------. estimates store nestedmod
. lrtest fullmod nestedmod
Likelihood-ratio test
(Assumption: nestedmod nested in fullmod)
.
.
.
.
.

LR chi2(1) =
Prob > chi2 =

88.73
0.0000

* Compara el likelihood de ambos modelos

* Ho: el modelo anidado es igual de verosmil que el modelo completo
// Pruebas de hiptesis para varios coeficientes
ologit warm yr89 male white age ed prst, nolog

Ordered logistic regression

Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

------------------------------------------------------------------------------

warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. * Wald test
. test age white male
. // Ho: las 3 variables son "conjuntamente insignificantes"
( 1)
( 2)
( 3)

[warm]age = 0
[warm]white = 0
[warm]male = 0
chi2( 3) =
Prob > chi2 =

166.62
0.0000

. * LR test
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

// modelo completo
Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. estimates store fmodel
. ologit warm yr89 ed prst, nolog
Ordered logistic regression
Log likelihood = -2930.7047

// modelo restringido
Number of obs
LR chi2(3)
Prob > chi2
Pseudo R2

=
=
=
=

2293
130.13
0.0000
0.0217

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5160795
.0792245
6.51
0.000
.3608024
.6713566
ed |
.1064872
.014975
7.11
0.000
.0771367
.1358377

prst |
-.001567
.003212
-0.49
0.626
-.0078624
.0047283
-------------+---------------------------------------------------------------/cut1 | -.5241726
.158709
-.8352366
-.2131086
/cut2 |
1.22832
.1587648
.9171465
1.539493
/cut3 |
3.016659
.1689551
2.685513
3.347805
-----------------------------------------------------------------------------. estimates store nmodel
. lrtest fmodel nmodel
Likelihood-ratio test
(Assumption: nmodel nested in fmodel)
.
.
.
.
.
.

LR chi2(3) =
Prob > chi2 =

171.58
0.0000

* Compara el likelihood de ambos modelos

* Ho: el modelo anidado es igual de verosmil que el modelo completo
// Medidas de bondad de ajuste usando FITSTAT
ologit warm yr89 male white age ed prst, nolog

Ordered logistic regression

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

Log likelihood = -2844.9123

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. fitstat
Measures of Fit for ologit of warm
Log-Lik Intercept Only:
D(2284):
McFadden's R2:
ML (Cox-Snell) R2:
McKelvey & Zavoina's R2:
Variance of y*:
Count R2:
AIC:
BIC:
BIC used by Stata:
.
.
.
.
.
.

-2995.770
5689.825
0.050
0.123
0.127
3.768
0.432
2.489
-11982.891
5759.463

Log-Lik Full Model:

LR(6):
Prob > LR:
McFadden's Adj R2:
Cragg-Uhler(Nagelkerke) R2:
Variance of error:
Adj Count R2:
AIC*n:
BIC':
AIC used by Stata:

-2844.912
301.716
0.000
0.047
0.133
3.290
0.093
5707.825
-255.291
5707.825

// Modelos OLOGIT y el supuesto de las "regresiones paralelas"

* Los modelos ologit asumen que cada variable X tiene el mismo
* efecto (positivo o negativo) en cada una de las categoras ordinales
* de la variable dependiente: X afecta igual la prob(outcome1)

.
.
.
.
.

* que prob(outomce2), etc., una vez que ajustamos por cada intercepto
* o cutpoint.
* generamos datos para ilustrar el supuesto de regresiones paralelas
clear

. set obs 41
obs was 0, now 41
. gen n = 41-_n
. replace n = (n-20)/2.5
(41 real changes made)
.
.
.
.
.
.
.

gen p1 = exp((-1*n)-3.5)/(1+exp((-1*n)-3.5))
gen p2 = exp((-1*n))/(1+exp((-1*n)))
gen p3 = exp((-1*n)+2)/(1+exp((-1*n)+2))
gen x = (n+8)*(9.375)
label var p1 "Pr(y<=1 | x)"
label var p2 "Pr(y<=2 | x)"
label var p3 "Pr(y<=3 | x)"

.2

Pr(y<=m)
.4
.6

.8

. graph twoway (line p1 p2 p3 x) , ylabel(0(.2)1) xlabel(0(50)150) ///

>
ytitle("Pr(y<=m)") xtitle("x") ///
>
ysize(2.6558) xsize(4.0413)

50
Pr(y<=1 | x)
Pr(y<=3 | x)

100

150

Pr(y<=2 | x)

.
. graph export 05parallel.emf, replace
(file D:\Metodos08\05parallel.emf written in Enhanced Metafile format)
.
. * Una prueba LR aproximada del supuesto de regresiones paralelas
. * (proporcional odds assumption)
.
. * nota: omodel es un comando extra de Stata: findit omodel
.
. sysuse ordwarm2, clear
(77 & 89 General Social Survey)
. omodel logit warm yr89 male white age ed prst nolog
Ordered logit estimates

Number of obs
LR chi2(6)
Prob > chi2

=
=
=

2293
301.72
0.0000

Log likelihood = -2844.9123

Pseudo R2

0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------_cut1 | -2.465362
.2389126
(Ancillary parameters)
_cut2 |
-.630904
.2333155
_cut3 |
1.261854
.2340179
-----------------------------------------------------------------------------Approximate likelihood-ratio test of proportionality of odds
across response categories:
chi2(12) =
48.91
Prob > chi2 =
0.0000
. * Una prueba de Wald del supuesto
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

Log likelihood = -2844.9123

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. brant, detail
Estimated coefficients from j-1 binary regressions
yr89
male
white
age
ed
prst
_cons

y>1
.9647422
-.30536425
-.55265759
-.0164704
.10479624
-.00141118
1.8584045

y>2
.56540626
-.69054232
-.31427081
-.02533448
.05285265
.00953216
.73032873

y>3
.31907316
-1.0837888
-.39299842
-.01859051
.05755466
.00553043
-1.0245168

Brant Test of Parallel Regression Assumption

Variable |
chi2
p>chi2
df
-------------+-------------------------All |
49.18
0.000
12

-------------+-------------------------yr89 |
13.01
0.001
2
male |
22.24
0.000
2
white |
1.27
0.531
2
age |
7.38
0.025
2
ed |
4.31
0.116
2
prst |
4.33
0.115
2
---------------------------------------A significant test statistic provides evidence that the parallel
regression assumption has been violated.
.
. /*
> brant performs a Brant test of the parallel regression assumption (also
> called the proportional odds assumption) after ologit. The test compares
> slope coefficients of the J-1 binary logits implied by the ordered regression
> model. Stata reports both the results of an omnibus test for the entire
> model and tests of the assumption for each of the independent variables in
> the model.
> */
.
. // Identificando residuales y outliers con predict despus de ologit
.
. sysuse ordwarm2, clear
(77 & 89 General Social Survey)
.
.
.
.

drop warmlt2 warmlt3 warmlt4

gen warmlt2 = (warm<2) if warm <. // omite recodificar los missing values
gen warmlt3 = (warm<3) if warm <.
gen warmlt4 = (warm<4) if warm <.

.
. * regresin logit cuando warm < 2
. logit warmlt2 yr89 male white age ed prst, nolog
Logistic regression
Log likelihood = -819.61992

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
128.58
0.0000
0.0727

-----------------------------------------------------------------------------warmlt2 |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 | -.9647422
.1542064
-6.26
0.000
-1.266981
-.6625033
male |
.3053643
.1291546
2.36
0.018
.052226
.5585025
white |
.5526576
.2305396
2.40
0.017
.1008082
1.004507
age |
.0164704
.0040571
4.06
0.000
.0085187
.0244221
ed | -.1047962
.0253348
-4.14
0.000
-.1544516
-.0551409
prst |
.0014112
.0056702
0.25
0.803
-.0097023
.0125246
_cons | -1.858405
.3958164
-4.70
0.000
-2.63419
-1.082619
-----------------------------------------------------------------------------. predict rstd_lt2, rs

// guarda los "residuales estandarizados" de Pearson

.
. * regresin logit cuando warm < 3
. logit warmlt3 yr89 male white age ed prst, nolog
Logistic regression
Log likelihood = -1449.7863

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
251.23
0.0000
0.0797

10

-----------------------------------------------------------------------------warmlt3 |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 | -.5654063
.0928433
-6.09
0.000
-.7473757
-.3834368
male |
.6905423
.0898786
7.68
0.000
.5143834
.8667012
white |
.3142708
.1405978
2.24
0.025
.0387042
.5898374
age |
.0253345
.0028644
8.84
0.000
.0197203
.0309486
ed | -.0528527
.0184571
-2.86
0.004
-.0890279
-.0166774
prst | -.0095322
.0038184
-2.50
0.013
-.0170162
-.0020482
_cons | -.7303287
.269163
-2.71
0.007
-1.257878
-.202779
-----------------------------------------------------------------------------. predict rstd_lt3, rs
. * regresin logit cuando < 4
. logit warmlt4 yr89 male white age ed prst, nolog
Logistic regression

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

Log likelihood = -1011.9542

=
=
=
=

2293
150.77
0.0000
0.0693

-----------------------------------------------------------------------------warmlt4 |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 | -.3190732
.1140756
-2.80
0.005
-.5426572
-.0954891
male |
1.083789
.1220668
8.88
0.000
.8445422
1.323035
white |
.3929984
.1577582
2.49
0.013
.083798
.7021989
age |
.0185905
.0037659
4.94
0.000
.0112096
.0259715
ed | -.0575547
.0253812
-2.27
0.023
-.1073008
-.0078085
prst | -.0055304
.0048413
-1.14
0.253
-.0150193
.0039584
_cons |
1.024517
.3463123
2.96
0.003
.3457571
1.703276
-----------------------------------------------------------------------------. predict rstd_lt4, rs
* graficando los residuales cuando warm < 3
sort prst
// ordena los datos de acuerdo a prestigio
gen index = _n // genera un ndice de 1 al nmero de obs.
// (para el eje horizontal de la grfica)
graph twoway scatter rstd_lt3 index, yline(0) ylabel(-4(2)4) ///
xtitle("Observation Number") xlabel(0(500)2293) ///
ysize(2.6558) xsize(4.0413) msymbol(Oh)

-4

standardized Pearson residual

-2
0
2

.
.
.
.
.
.
>
>

2500

500

1000
1500
Observation Number

2000

. graph export 05residplot.emf, replace

(file D:\Metodos08\05residplot.emf written in Enhanced Metafile format)

11

.
.
. // Calculando efectos estandarizados con LISTCOEF
.
. sysuse ordwarm2, clear
(77 & 89 General Social Survey)
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. listcoef, std help
ologit (N=2293): Unstandardized and Standardized Estimates
Observed SD: .9282156
Latent SD: 1.9410634
------------------------------------------------------------------------------warm |
b
z
P>|z|
bStdX
bStdY
bStdXY
SDofX
-------------+----------------------------------------------------------------yr89 |
0.52390
6.557
0.000
0.2566
0.2699
0.1322
0.4897
male | -0.73330
-9.343
0.000 -0.3658 -0.3778 -0.1885
0.4989
white | -0.39116
-3.304
0.001 -0.1287 -0.2015 -0.0663
0.3290
age | -0.02167
-8.778
0.000 -0.3635 -0.0112 -0.1873
16.7790
ed |
0.06717
4.205
0.000
0.2123
0.0346
0.1094
3.1608
prst |
0.00607
1.844
0.065
0.0880
0.0031
0.0453
14.4923
------------------------------------------------------------------------------b = raw coefficient
z = z-score for test of b=0
P>|z| = p-value for z-test
bStdX = x-standardized coefficient
bStdY = y-standardized coefficient
bStdXY = fully standardized coefficient
SDofX = standard deviation of X
.
. // Graficando probabilidades predichas con predict
.
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression

Number of obs
LR chi2(6)
Prob > chi2

=
=
=

2293
301.72
0.0000

12

Log likelihood = -2844.9123

Pseudo R2

0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. predict SDwarm Dwarm Awarm SAwarm
(option pr assumed; predicted probabilities)
.
.
.
.
.

* guarda las predicciones de los 4 resultados posibles de warm

label var SDwarm "Pr(SD)" // prob(Strongly Disagree)
label var Dwarm "Pr(D)"
// prob(Strongly Disagree)
label var Awarm "Pr(A)"
// prob(Strongly Disagree)
label var SAwarm "Pr(SA)" // prob(Strongly Agree)

.25

.5

.75

.
. dotplot SDwarm Dwarm Awarm SAwarm, ylabel(0(.25).75) ///
>
ysize(2.0124) xsize(3.039)

Pr(SD)

Pr(D)

Pr(A)

Pr(SA)

. graph export 05pred.emf, replace

(file D:\Metodos08\05pred.emf written in Enhanced Metafile format)
.
. // Calculando probabilidades predichas para casos particulares con PRVALUE
.
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766

13

white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------.
. * Un obrero de 64 aos encuestado en 1977
. prvalue, x(yr89=0 male=1 prst=20 age=64 ed=16) rest(mean)
ologit: Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
x=

yr89
0

0.2317
0.4221
0.2723
0.0739
male
1

95% Conf. Interval

[ 0.1776,
0.2857]
[ 0.3942,
0.4500]
[ 0.2249,
0.3198]
[ 0.0523,
0.0954]

white
.8765809

age
64

ed
16

prst
20

.
. * Una joven de 1989 muy educada y con un empleo muy prestigioso
. prvalue, x(yr89=1 male=0 prst=80 age=30 ed=24) rest(mean)
ologit: Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
x=

yr89
1

0.0164
0.0781
0.3147
0.5908
male
0

95% Conf. Interval

[ 0.0106,
0.0222]
[ 0.0554,
0.1008]
[ 0.2636,
0.3658]
[ 0.5143,
0.6673]

white
.8765809

age
30

ed
24

prst
80

.
. * Un individuo promedio encuestado en 1977
. prvalue, x(yr89=0) rest(mean) brief
. * la opcin brief presenta un output resumido de prvalue
ologit: Predictions for warm
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):

0.1336
0.3577
0.3737
0.1349

95% Conf. Interval

[ 0.1176,
0.1496]
[ 0.3348,
0.3806]
[ 0.3517,
0.3957]
[ 0.1195,
0.1504]

.
. * Un individuo promedio encuestado en 1989
. prvalue, x(yr89=1) rest(mean) brief
ologit: Predictions for warm

14

Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):

0.0837
0.2802
0.4277
0.2085

95% Conf. Interval

[ 0.0711,
0.0963]
[ 0.2571,
0.3032]
[ 0.4046,
0.4507]
[ 0.1855,
0.2315]

.
. // Tablas de probabilidades predichas con PRTAB
.
. prtab yr89 male, novarlbl
. * Tabula las diferentes combinaciones de yr89 y male
. * en una tabla 2x2 (porque ambas son dummies)
ologit: Predicted probabilities for warm
Predicted probability of outcome 1 (1SD)
-------------------------|
male
yr89 | Women
Men
----------+--------------1977 | 0.0989 0.1859
1989 | 0.0610 0.1191
-------------------------Predicted probability of outcome 2 (2D)
-------------------------|
male
yr89 | Women
Men
----------+--------------1977 | 0.3083 0.4026
1989 | 0.2282 0.3394
-------------------------Predicted probability of outcome 3 (3A)
-------------------------|
male
yr89 | Women
Men
----------+--------------1977 | 0.4129 0.3162
1989 | 0.4406 0.3904
-------------------------Predicted probability of outcome 4 (4SA)
-------------------------|
male
yr89 | Women
Men
----------+--------------1977 | 0.1799 0.0953
1989 | 0.2703 0.1510
-------------------------x=

yr89
.39860445

male
.46489315

white
.8765809

age
44.935456

ed
12.218055

prst
39.585259

.
. * prtab no incluye intervalos de confianza, para obtenerlos
. * hay que usar prvalue:
.
. prvalue, x(yr89=0 male=1)
ologit: Predictions for warm

15

Confidence intervals by delta method

Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
yr89
0

x=

0.1859
0.4026
0.3162
0.0953
male
1

95% Conf. Interval

[ 0.1631,
0.2088]
[ 0.3777,
0.4275]
[ 0.2925,
0.3400]
[ 0.0814,
0.1092]

white
.8765809

age
44.935456

ed
12.218055

prst
39.585259

. prvalue, x(yr89=1 male=0)

ologit: Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
yr89
1

x=

0.0610
0.2282
0.4406
0.2703
male
0

95% Conf. Interval

[ 0.0504,
0.0716]
[ 0.2046,
0.2517]
[ 0.4176,
0.4635]
[ 0.2406,
0.2999]

white
.8765809

age
44.935456

ed
12.218055

prst
39.585259

.
. // Cambios discretos con Intervalos de confianza
.
. * Hombres vs. mujeres
. prvalue , x(male=0 yr89=1) rest(mean) save // guarda la estimacin
ologit: Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
yr89
1

x=

0.0610
0.2282
0.4406
0.2703
male
0

95% Conf. Interval

[ 0.0504,
0.0716]
[ 0.2046,
0.2517]
[ 0.4176,
0.4635]
[ 0.2406,
0.2999]

white
.8765809

age
44.935456

. prvalue , x(male=1 yr89=1) rest(mean) dif

.
ologit: Change in Predictions for warm

ed
12.218055

prst
39.585259

// calcula la diferencia con los

// resultados guardados

Confidence intervals by delta method

Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
Current=
Saved=
Diff=

yr89
1
1
0

Current
0.1191
0.3394
0.3904
0.1510
male
1
0
1

Saved
0.0610
0.2282
0.4406
0.2703

Change
0.0581
0.1113
-0.0502
-0.1192

white
.8765809
.8765809
0

95% CI for Change

[ 0.0438,
0.0725]
[ 0.0876,
0.1350]
[-0.0679, -0.0324]
[-0.1448, -0.0937]

age
44.935456
44.935456
0

ed
12.218055
12.218055
0

prst
39.585259
39.585259
0

16

. * Efecto de aumentar la edad en una desviacin estndar

. * centrada en torno a la edad promedio (44.93)
.
. display 44.935 - (.5*16.779)
36.5455
. display 44.935 + (.5*16.779)
53.3245
.
. prvalue, x(male=0 yr89=1 age=36.5455) rest(mean) save
ologit: Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):
yr89
1

x=

0.0514
0.2019
0.4392
0.3076
male
0

95% Conf. Interval

[ 0.0418,
0.0609]
[ 0.1788,
0.2250]
[ 0.4160,
0.4623]
[ 0.2744,
0.3408]

white
.8765809

age
36.5455

ed
12.218055

prst
39.585259

. prvalue, x(male=0 yr89=1 age=53.3245) rest(mean) dif

ologit: Change in Predictions for warm
Confidence intervals by delta method
Pr(y=1SD|x):
Pr(y=2D|x):
Pr(y=3A|x):
Pr(y=4SA|x):

Current
0.0723
0.2556
0.4362
0.2359

yr89
1
1
0

Current=
Saved=
Diff=

male
0
0
0

Saved
0.0514
0.2019
0.4392
0.3076

Change
0.0209
0.0537
-0.0030
-0.0716

white
.8765809
.8765809
0

95% CI for Change

[ 0.0155,
0.0262]
[ 0.0414,
0.0661]
[-0.0118,
0.0058]
[-0.0883, -0.0549]

age
53.3245
36.5455
16.779

ed
12.218055
12.218055
0

prst
39.585259
39.585259
0

.
. // Graficando una serie de probabilidades predichas con PRGEN
.
. prgen age, from(20) to(80) generate(w89) x(male=0 yr89=1) ncases(13)
. * Probabilidades predichas para edades de 20 a 80 aos, usando
. * mujeres encuestadas en 1989 como lnea de base.
. * Las 13 predicciones se guardan en una serie de variables con el prefijo
"w89"
ologit: Predicted values as age varies from 20 to 80.
x=

yr89
1

male
0

white
.8765809

age
44.935456

ed
12.218055

prst
39.585259

. desc w89*
storage display
value
variable name
type
format
label
variable label
------------------------------------------------------------------------------w89x
float %9.0g
Age in years

17

w89p1
w89p2
w89p3
w89p4
w89s1
w89s2
w89s3
w89s4
.
.
.
.
.
.
.

label
label
label
label
label
label
label

float
float
float
float
float
float
float
float
var
var
var
var
var
var
var

w89p1
w89p2
w89p3
w89p4
w89s1
w89s2
w89s3

%9.0g
%9.0g
%9.0g
%9.0g
%9.0g
%9.0g
%9.0g
%9.0g

pr(1SD)=Pr(1)
pr(2D)=Pr(2)
pr(3A)=Pr(3)
pr(4SA)=Pr(4)
pr(y<=1)
pr(y<=2)
pr(y<=3)
pr(y<=4)

"SD" // prob(outcome1)
"D"
// prob(outcome2)
"A"
// prob(outcome3)
"SA" // prob(outcome4)
"SD"
"SD or D" // probabilidad acumulada
"SD, D or A" // probabilidad acumulada

. * Paso 1: Grfica de probabilidades predichas

. graph twoway connected w89p1 w89p2 w89p3 w89p4 w89x, ///
>
title("Panel A: Predicted Probabilities") ///
>
xtitle("Age") xlabel(20(10)80) ylabel(0(.25).50) ///
>
yscale(noline) ylabel("") xline(44.93) ///
>
ytitle("") name(tmp1, replace)
Panel A: Predicted Probabilities

20

30

40

50
Age
SD
A

60

70

80

D
SA

.
. * Paso 2: Grfica de probabilidades acumuladas
. graph twoway connected w89s1 w89s2 w89s3 w89x, ///
>
title("Panel B: Cumulative Probabilities") ///
>
xtitle("Age") xlabel(20(10)80) ylabel(0(.25)1) xline(44.93) ///
>
yscale(noline) ylabel("") name(tmp2, replace) ///
>
ytitle("")

18

Panel B: Cumulative Probabilities

20

30

40

50
Age
SD
SD, D or A

60

70

80

SD or D

.
. * Paso 3: combinar las grficas guardadas en memoria
. graph combine tmp1 tmp2, col(1) iscale(*.9) imargin(small) ///
>
ysize(4.31) xsize(3.287)
Panel A: Predicted Probabilities

20

30

40

50
Age

60

SD

SA

70

80

70

80

Panel B: Cumulative Probabilities

20

30

40

50
Age
SD

60
SD or D

SD, D or A

.
. * Paso 4: Guardar grfica en formato "enhanced metafile"
. graph export 05prgen.emf, replace
(file D:\Metodos08\05prgen.emf written in Enhanced Metafile format)
.
.
. // Calculando cambios en probabilidades predichas con PRCHANGE
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression

Number of obs

2293

19

LR chi2(6)
Prob > chi2
Pseudo R2

Log likelihood = -2844.9123

=
=
=

301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------.
. * Efectos marginales usando prchange (comando Spost)
. prchange age, x(male=0 yr89=1) rest(mean) help
ologit: Changes in Probabilities for warm
age
Avg|Chg|
.16441458
.00222661
.0373125
.00222662

Min->Max
-+1/2
-+sd/2
MargEfct
Pr(y|x)
x=
sd(x)=

1SD
.06099996
yr89
1
.489718

1SD
.10941909
.00124099
.0208976
.00124098
2D
.22815652

male
0
.498875

2D
.21941006
.00321223
.05372739
.00321226

3A
.44057754

white
.876581
.328989

3A
-.05462247
-.0001803
-.00300205
-.00018032

4SA
-.27420671
-.00427291
-.07162295
-.00427292

age
44.9355
16.779

4SA
.27026597
ed
12.2181
3.16083

prst
39.5853
14.4923

Pr(y|x): probability of observing each y for specified x values

Avg|Chg|: average of absolute value of the change across categories
Min->Max: change in predicted probability as x changes from its minimum to
its maximum
0->1: change in predicted probability as x changes from 0 to 1
-+1/2: change in predicted probability as x changes from 1/2 unit below
base value to 1/2 unit above
-+sd/2: change in predicted probability as x changes from 1/2 standard
dev below base to 1/2 standard dev above
MargEfct: the partial derivative of the predicted probability/rate with
respect to a given independent variable
.
. * Efectos marginales usando mfx (comando stata)
. mfx compute, at(male=0 yr89=1) predict(outcome(1))
warning: no value assigned in at() for variables white age ed prst;
means used for white age ed prst
Marginal effects after ologit
y = Pr(warm==1) (predict, outcome(1))
= .06099996
-----------------------------------------------------------------------------variable |
dy/dx
Std. Err.
z
P>|z| [
95% C.I.
]
X

20

---------+-------------------------------------------------------------------yr89*| -.0378526
.00601
-6.30
0.000 -.049633 -.026072
1
male*|
.0581355
.00731
7.95
0.000
.043803 .072468
0
white*|
.0197511
.0055
3.59
0.000
.008972
.03053
.876581
age |
.001241
.00016
7.69
0.000
.000925 .001557
44.9355
ed | -.0038476
.00097
-3.96
0.000 -.005754 -.001941
12.2181
prst | -.0003478
.00019
-1.83
0.068 -.000721 .000025
39.5853
-----------------------------------------------------------------------------(*) dy/dx is for discrete change of dummy variable from 0 to 1
.
. * Efectos marginales de cambios discretos con prchange
. prchange male age prst, x(male=0 yr89=1) rest(mean) help
ologit: Changes in Probabilities for warm
male
0->1

Avg|Chg|
.08469636

1SD
.05813552

2D
.11125721

3A
-.05015317

4SA
-.11923955

Min->Max
-+1/2
-+sd/2
MargEfct

Avg|Chg|
.16441458
.00222661
.0373125
.00222662

1SD
.10941909
.00124099
.0208976
.00124098

2D
.21941006
.00321223
.05372739
.00321226

3A
-.05462247
-.0001803
-.00300205
-.00018032

4SA
-.27420671
-.00427291
-.07162295
-.00427292

Avg|Chg|
.04278038
.00062411
.00904405
.00062411

1SD
-.02352008
-.00034784
-.00504204
-.00034784

2D
-.06204067
-.00090037
-.01304607
-.00090038

3A
.00013945
.00005054
.00073212
.00005054

4SA
.08542132
.00119767
.01735598
.00119767

age

prst
Min->Max
-+1/2
-+sd/2
MargEfct
Pr(y|x)
x=
sd(x)=

1SD
.06099996
yr89
1
.489718

2D
.22815652

male
0
.498875

3A
.44057754

white
.876581
.328989

age
44.9355
16.779

4SA
.27026597
ed
12.2181
3.16083

prst
39.5853
14.4923

Pr(y|x): probability of observing each y for specified x values

Avg|Chg|: average of absolute value of the change across categories
Min->Max: change in predicted probability as x changes from its minimum to
its maximum
0->1: change in predicted probability as x changes from 0 to 1
-+1/2: change in predicted probability as x changes from 1/2 unit below
base value to 1/2 unit above
-+sd/2: change in predicted probability as x changes from 1/2 standard
dev below base to 1/2 standard dev above
MargEfct: the partial derivative of the predicted probability/rate with
respect to a given independent variable
.
. * Efecto marginal de un aumento de 10 aos en edad
. * (usando la opcin delta de prchange)
. quietly ologit warm yr89 male white age ed prst, nolog
. prchange age, x(male=0 yr89=1) rest(mean) delta(10) help
ologit: Changes in Probabilities for warm
(Note: d = 10)

21

age
Avg|Chg|
.16441458
.02225603
.0373125
.00222662

Min->Max
-+d/2
-+sd/2
MargEfct
Pr(y|x)
x=
sd(x)=

1SD
.06099996

1SD
.10941909
.01242571
.0208976
.00124098

yr89
1
.489718

2D
.22815652

male
0
.498875

2D
.21941006
.03208634
.05372739
.00321226

3A
.44057754

white
.876581
.328989

3A
-.05462247
-.00179818
-.00300205
-.00018032

4SA
-.27420671
-.04271388
-.07162295
-.00427292

age
44.9355
16.779

4SA
.27026597
ed
12.2181
3.16083

prst
39.5853
14.4923

Pr(y|x): probability of observing each y for specified x values

Avg|Chg|: average of absolute value of the change across categories
Min->Max: change in predicted probability as x changes from its minimum to
its maximum
0->1: change in predicted probability as x changes from 0 to 1
-+d/2: change in predicted probability as x changes from 5 units
below base value to 5 units above
-+sd/2: change in predicted probability as x changes from 1/2 standard
dev below base to 1/2 standard dev above
MargEfct: the partial derivative of the predicted probability/rate with
respect to a given independent variable
.
.
. // Mostrar odds ratios con LISTCOEF
.
. ologit warm yr89 male white age ed prst, nolog
Ordered logistic regression
Log likelihood = -2844.9123

Number of obs
LR chi2(6)
Prob > chi2
Pseudo R2

=
=
=
=

2293
301.72
0.0000
0.0504

-----------------------------------------------------------------------------warm |
Coef.
Std. Err.
z
P>|z|
[95% Conf. Interval]
-------------+---------------------------------------------------------------yr89 |
.5239025
.0798988
6.56
0.000
.3673037
.6805013
male | -.7332997
.0784827
-9.34
0.000
-.8871229
-.5794766
white | -.3911595
.1183808
-3.30
0.001
-.6231815
-.1591374
age | -.0216655
.0024683
-8.78
0.000
-.0265032
-.0168278
ed |
.0671728
.015975
4.20
0.000
.0358624
.0984831
prst |
.0060727
.0032929
1.84
0.065
-.0003813
.0125267
-------------+---------------------------------------------------------------/cut1 | -2.465362
.2389126
-2.933622
-1.997102
/cut2 |
-.630904
.2333155
-1.088194
-.173614
/cut3 |
1.261854
.2340179
.8031873
1.720521
-----------------------------------------------------------------------------. listcoef male age, help
ologit (N=2293): Factor Change in Odds
Odds of: >m vs <=m
---------------------------------------------------------------------warm |
b
z
P>|z|
e^b
e^bStdX
SDofX
-------------+-------------------------------------------------------male | -0.73330
-9.343
0.000
0.4803
0.6936
0.4989

22

age | -0.02167
-8.778
0.000
0.9786
0.6952
16.7790
---------------------------------------------------------------------b = raw coefficient
z = z-score for test of b=0
P>|z| = p-value for z-test
e^b = exp(b) = factor change in odds for unit increase in X
e^bStdX = exp(b*SD of X) = change in odds for SD increase in X
SDofX = standard deviation of X
. listcoef male age, percent help
ologit (N=2293): Percentage Change in Odds
Odds of: >m vs <=m
---------------------------------------------------------------------warm |
b
z
P>|z|
%
%StdX
SDofX
-------------+-------------------------------------------------------male | -0.73330
-9.343
0.000
-52.0
-30.6
0.4989
age | -0.02167
-8.778
0.000
-2.1
-30.5
16.7790
---------------------------------------------------------------------b = raw coefficient
z = z-score for test of b=0
P>|z| = p-value for z-test
% = percent change in odds for unit increase in X
%StdX = percent change in odds for SD increase in X
SDofX = standard deviation of X
. listcoef male, reverse help
ologit (N=2293): Factor Change in Odds
Odds of: <=m vs >m
---------------------------------------------------------------------warm |
b
z
P>|z|
e^b
e^bStdX
SDofX
-------------+-------------------------------------------------------male | -0.73330
-9.343
0.000
2.0819
1.4417
0.4989
---------------------------------------------------------------------b = raw coefficient
z = z-score for test of b=0
P>|z| = p-value for z-test
e^b = exp(b) = factor change in odds for unit increase in X
e^bStdX = exp(b*SD of X) = change in odds for SD increase in X
SDofX = standard deviation of X
.
. log close
log:
log type:
closed on:

D:\MyDocs\Metodos08\Long&FreeseSpost9\ch5_ordinal.log
text
30 Nov 2008, 20:45:37

23

DO FILE
capture log close
log using ch5_ordinal, replace
* crea una bitcora con formato de texto plano
//

MODELOS PARA VARIABLES DEPENDIENTES ORDINALES

//
//
//
//

Regression Models for Categorical Dependent Variables - 2nd Edition

Chapter 5 - Models for Ordinal Outcomes
Long and Freese - 27Jul2005
Comentarios en espaol: Javier Aparicio - 3nov2008

version 9
* para que stata use sintaxis de version 9 (en caso de tener una versin
posterior)
set scheme s2manual
// modo monocromtico para grficas
set more off
// para que el output del DO file no haga pausa al correr
*
*
*
*

Ejemplo de variable ordinal:

Actitudes frente a las madres que trabajan
Var dep: Esta de acuerdo o no con la frase:
"Las mamas que trabajan pueden mantener relaciones calidas con sus hijos"

sysuse ordwarm2, clear

desc warm yr89 male white age ed prst
sum warm yr89 male white age ed prst
tab warm
* Var. Dep.: Est de acuerdo o no con la frase?
* "Las mams que trabajan pueden mantener relaciones calidas con sus hijos"
//

Un modelo de regresin logstico ordinal

ologit warm male white age ed prst, nolog

// muestra completa
* la opcion nolog omite el output de las iteraciones
ologit warm male white age ed prst if yr89==1, nolog

// muestra restringida

* comparando ORDINAL LOGIT con ORDINAL PROBIT

ologit warm yr89 male white age ed prst, nolog
estimates store ologit // guarda los estimadores ologit
oprobit warm yr89 male white age ed prst, nolog
estimates store oprobit // guarda los estimadores oprobit
estimates table ologit oprobit, b(%9.3f) t label varwidth(30)
* crea una tabla con los modelos previos con tres decimales, t-stats y etiquetas
de variables de hasta 30 caracteres
//

Pruebas de hiptesis para coeficientes individuales

* Wald test
ologit warm male yr89 white age ed prst, nolog
test male
// test similar al z-score de la tabla

24

display "z*z=" -9.343*-9.343

return list
di "chi2=" r(chi2) "; sqrt(chi2)= " sqrt(r(chi2))
* LR (likelihood ratio) test
ologit warm yr89 male white age ed prst, nolog // modelo "completo"
estimates store fullmod
* estimate constrained model
ologit warm yr89 white age ed prst, nolog
estimates store nestedmod

// un modelo restringido o anidado

lrtest fullmod nestedmod

* Compara el likelihood de ambos modelos
* Ho: el modelo anidado es igual de verosmil que el modelo completo
//

Pruebas de hiptesis para varios coeficientes

ologit warm yr89 male white age ed prst, nolog

* Wald test
test age white male
// Ho: las 3 variables son "conjuntamente insignificantes"
* LR test
ologit warm yr89 male white age ed prst, nolog
estimates store fmodel
ologit warm yr89 ed prst, nolog
estimates store nmodel

// modelo completo

// modelo restringido

lrtest fmodel nmodel

* Compara el likelihood de ambos modelos
* Ho: el modelo anidado es igual de verosmil que el modelo completo
//

Medidas de bondad de ajuste usando FITSTAT

ologit warm yr89 male white age ed prst, nolog

fitstat
// Modelos OLOGIT y el supuesto de las "regresiones paralelas"
* Los modelos ologit asumen que cada variable X tiene el mismo
* efecto (positivo o negativo) en cada una de las categoras ordinales
* de la variable dependiente: X afecta igual la prob(outcome1)
* que prob(outomce2), etc., una vez que ajustamos por cada intercepto
* o cutpoint.
* generamos datos para ilustrar el supuesto de regresiones paralelas
clear
set obs 41
gen n = 41-_n
replace n = (n-20)/2.5
gen p1 = exp((-1*n)-3.5)/(1+exp((-1*n)-3.5))
gen p2 = exp((-1*n))/(1+exp((-1*n)))
gen p3 = exp((-1*n)+2)/(1+exp((-1*n)+2))
gen x = (n+8)*(9.375)
label var p1 "Pr(y<=1 | x)"
label var p2 "Pr(y<=2 | x)"
label var p3 "Pr(y<=3 | x)"
graph twoway (line p1 p2 p3 x) , ylabel(0(.2)1) xlabel(0(50)150) ///
ytitle("Pr(y<=m)") xtitle("x") ///

25

ysize(2.6558) xsize(4.0413)
graph export 05parallel.emf, replace
* Una prueba LR aproximada del supuesto de regresiones paralelas
* (proporcional odds assumption)
* nota: omodel es un comando extra de Stata: findit omodel
sysuse ordwarm2, clear
omodel logit warm yr89 male white age ed prst
* Una prueba de Wald del supuesto
ologit warm yr89 male white age ed prst, nolog
brant, detail
/*
brant performs a Brant test of the parallel regression assumption (also
called the proportional odds assumption) after ologit. The test compares
slope coefficients of the J-1 binary logits implied by the ordered regression
model. Stata reports both the results of an omnibus test for the entire
model and tests of the assumption for each of the independent variables in
the model.
*/
//

Identificando residuales y outliers con predict despus de ologit

sysuse ordwarm2, clear

drop warmlt2 warmlt3 warmlt4
gen warmlt2 = (warm<2) if warm <. // omite recodificar los missing values
gen warmlt3 = (warm<3) if warm <.
gen warmlt4 = (warm<4) if warm <.
* regresin logit cuando warm < 2
logit warmlt2 yr89 male white age ed prst, nolog
predict rstd_lt2, rs // guarda los "residuales estandarizados" de Pearson
* regresin logit cuando warm < 3
logit warmlt3 yr89 male white age ed prst, nolog
predict rstd_lt3, rs
* regresin logit cuando < 4
logit warmlt4 yr89 male white age ed prst, nolog
predict rstd_lt4, rs
* graficando los residuales cuando warm < 3
sort prst
gen index = _n

// ordena los datos de acuerdo a prestigio

// genera un ndice de 1 al nmero de obs.
// (para el eje horizontal de la grfica)

graph twoway scatter rstd_lt3 index, yline(0) ylabel(-4(2)4) ///

xtitle("Observation Number") xlabel(0(500)2293) ///
ysize(2.6558) xsize(4.0413) msymbol(Oh)
graph export 05residplot.emf, replace

//

Calculando efectos estandarizados con LISTCOEF

sysuse ordwarm2, clear

ologit warm yr89 male white age ed prst, nolog

26

listcoef, std help

//

Graficando probabilidades predichas con predict

ologit warm yr89 male white age ed prst, nolog

predict SDwarm Dwarm Awarm SAwarm
* guarda las predicciones de los 4 resultados posibles de warm
label
label
label
label

var
var
var
var

SDwarm "Pr(SD)"
Dwarm "Pr(D)"
Awarm "Pr(A)"
SAwarm "Pr(SA)"

//
//
//
//

prob(Strongly
prob(Strongly
prob(Strongly
prob(Strongly

Disagree)
Disagree)
Disagree)
Agree)

dotplot SDwarm Dwarm Awarm SAwarm, ylabel(0(.25).75) ///

ysize(2.0124) xsize(3.039)
graph export 05pred.emf, replace

// Calculando probabilidades predichas para casos particulares con PRVALUE

ologit warm yr89 male white age ed prst, nolog
* Un obrero de 64 aos encuestado en 1977
prvalue, x(yr89=0 male=1 prst=20 age=64 ed=16) rest(mean)
* Una joven de 1989 muy educada y con un empleo muy prestigioso
prvalue, x(yr89=1 male=0 prst=80 age=30 ed=24) rest(mean)
* Un individuo promedio encuestado en 1977
prvalue, x(yr89=0) rest(mean) brief
* la opcin brief presenta un output resumido de prvalue
* Un individuo promedio encuestado en 1989
prvalue, x(yr89=1) rest(mean) brief
//

Tablas de probabilidades predichas con PRTAB

prtab yr89 male, novarlbl

* Tabula las diferentes combinaciones de yr89 y male
* en una tabla 2x2 (porque ambas son dummies)
* prtab no incluye intervalos de confianza, para obtenerlos
* hay que usar prvalue:
prvalue, x(yr89=0 male=1)
prvalue, x(yr89=1 male=0)
//

Cambios discretos con Intervalos de confianza

* Hombres vs. mujeres

prvalue , x(male=0 yr89=1) rest(mean) save // guarda la estimacin
prvalue , x(male=1 yr89=1) rest(mean) dif // calcula la diferencia con los
// resultados guardados
* Efecto de aumentar la edad en una desviacin estndar
* centrada en torno a la edad promedio (44.93)
display 44.935 - (.5*16.779)
display 44.935 + (.5*16.779)

27

prvalue, x(male=0 yr89=1 age=36.5455) rest(mean) save

prvalue, x(male=0 yr89=1 age=53.3245) rest(mean) dif
//

Graficando una serie de probabilidades predichas con PRGEN

sysuse ordwarm2, clear

ologit warm yr89 male white age ed prst, nolog
prgen age, from(20) to(80) generate(w89) x(male=0 yr89=1) ncases(13)
* Probabilidades predichas para edades de 20 a 80 aos, usando
* mujeres encuestadas en 1989 como lnea de base.
* Las 13 predicciones se guardan en una serie de variables con el prefijo "w89"
desc w89*
label var
label var
label var
label var
label var
label var
label var

w89p1
w89p2
w89p3
w89p4
w89s1
w89s2
w89s3

"SD" // prob(outcome1)
"D"
// prob(outcome2)
"A"
// prob(outcome3)
"SA" // prob(outcome4)
"SD"
"SD or D" // probabilidad acumulada
"SD, D or A" // probabilidad acumulada

* Paso 1: Grfica de probabilidades predichas

graph twoway connected w89p1 w89p2 w89p3 w89p4 w89x, ///
title("Panel A: Predicted Probabilities") ///
xtitle("Age") xlabel(20(10)80) ylabel(0(.25).50) ///
yscale(noline) ylabel("") xline(44.93) ///
ytitle("") name(tmp1, replace)
* Paso 2: Grfica de probabilidades acumuladas
graph twoway connected w89s1 w89s2 w89s3 w89x, ///
title("Panel B: Cumulative Probabilities") ///
xtitle("Age") xlabel(20(10)80) ylabel(0(.25)1) xline(44.93) ///
yscale(noline) ylabel("") name(tmp2, replace) ///
ytitle("")
* Paso 3: combinar las grficas guardadas en memoria
graph combine tmp1 tmp2, col(1) iscale(*.9) imargin(small) ///
ysize(4.31) xsize(3.287)
* Paso 4: Guardar grfica en formato "enhanced metafile"
graph export 05prgen.emf, replace
//

Calculando cambios en probabilidades predichas con PRCHANGE

ologit warm yr89 male white age ed prst, nolog

* Efectos marginales usando prchange (comando Spost)
prchange age, x(male=0 yr89=1) rest(mean) help
* Efectos marginales usando mfx (comando stata)
mfx compute, at(male=0 yr89=1) predict(outcome(1))
* Efectos marginales de cambios discretos con prchange
prchange male age prst, x(male=0 yr89=1) rest(mean) help
* Efecto marginal de un aumento de 10 aos en edad
* (usando la opcin delta de prchange)

28

quietly ologit warm yr89 male white age ed prst, nolog

prchange age, x(male=0 yr89=1) rest(mean) delta(10) help
//

Mostrar odds ratios con LISTCOEF

ologit warm yr89 male white age ed prst, nolog

listcoef male age, help
listcoef male age, percent help
listcoef male, reverse help
log close

29