Está en la página 1de 15

Achievements to Date and Goals for the Future: New Looks at Criminal Careers

Criminal Incapacitation by William Spelman; Analysing Offending: Data, Models, and


Interpretation by Roger Tarling
Review by: Jeffrey A. Roth
Journal of Quantitative Criminology, Vol. 11, No. 1 (March 1995), pp. 97-110
Published by: Springer
Stable URL: http://www.jstor.org/stable/41954123 .
Accessed: 18/06/2014 20:10
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Journal of Quantitative
Criminology.

http://www.jstor.org

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Journal
Vol.11,No.1,1995
Criminology,
ofQuantitative

Review Essay
Achievements
New Looks

to Date
at Criminal

and Goals

for the Future:

Careers

A. Roth1
Jeffrey
CriminalIncapacitation.By WilliamSpelman.PlenumPress,New York and
London, 1994,338 pp.
Data, Models, and Interpretation.
By Roger Tarling.
AnalysingOffending:
Her Majesty'sStationeryOffice,London, 1993,203 pp.
Criminal Incapacitation(Spelman, 1994) and AnalysingOffending
(Tarling, 1993) revisitthe two principalquestionsasked by the National
Research Council's (NRC) Panel on Research on Criminal Careers
[Blumsteinet al., 1986 (hereinafterthe Careers Panel)]: What is known
about individuals'criminalcareers?What does that knowledgeimplyfor
thedual roots
criminaljusticepolicy?The pairingof thosequestionsreflects
of the careerparadigmin policy analysisand researchmethodology.The
National Instituteof Justicehad requestedthe NRC to createthe Careers
Panel to addressa policyissue: thevalidityof publishedclaimsthata "selecwithhigherpretiveincapacitation"policy- sentencing
convictedoffenders
dicted futureoffendingfrequenciesto longer prison terms than other
- could reducecrimewithoutincreasingtheprisonpopuconvictedoffenders
lation(Greenwood,1982,pp. xix,79). Duringits work,however,thepanel
also took on a researchquestion: Does existingresearchconductedor interpretedfroma careerperspectiveyieldinsightsabout criminalbehaviorthat
had not been obtainedfromstudiesframedin otherways,such as analyses
of aggregatedata or recidivismpredictionstudies?
The CareersPanel called foran ambitiousresearchagenda to improve
the measurementof individualoffending
frequency(A) and career length
and officialarrestrecords,and to learnmore about
( T ), usingself-reports
careeronset,termination
or dropout(r), and the
the processesunderlying
TheUrbanInstitute,
DC.
'StatePolicy
Center,
Washington,
97
0748-451
Plenum
1995
8/95/0300-0097507.50/0
Publishing
Corporation

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

98

Roth

timepath of A. In theyearsbetweenthepanel reportand publicationof the


books under review,criminalcareersresearchsaw threeprimarylines of
development.First,severalauthorsapplied survivalmodels- typicallyproportionalhazards, exponential,or Weibull models- to officiairecordsto
partitionvariationin A and r and to improveestimatesof thosedimensions
and theircovariationwithhypothesizedcausal factors(e.g., Schmidtand
Witte,1988; Rhodes, 1989; Visheret al ., 1991). Second, an acrimonious
debate occurredover the meritsof longitudinalstudiesof individualsas a
means of studyingoffending
over the lifecourse and the careerparadigm
as a frameworkforinterpreting
findingsfromlongitudinaland otherdata
on offendingover time (see, e.g., Gottfredsonand Hirschi, 1986, 1988;
Blumsteinet ai, 1988a,b; Hagan and Palloni, 1988; Land, 1992). EmpirZ)a, the fractionof a population
ically,most testsfoundthat partitioning
thatparticipatesin at least one crimeby age a, and A producedat best a
in fitoversimplerparameterizations
of offending
fairlymodestimprovement
et
al
Rowe
and
.,
1990;
Smith,
1990;
sequences (e.g.,
Nagin
Nagin and
Land, 1993; but see Blumsteinet al , 1988b,foran exception).Third,as the
culminationof a 5-yearplanningeffort(Tonry et al ., 1991), the Harvard
School of PublicHealthrecently
launchedtheProgramon Human Develop- a largeprospectivelongitudinalstudyof
mentin Chicago Neighborhoods
the kindrecommendedby the CareersPanel.
Tarling'sand Spelman's books are the most comprehensiveworksin
the careerframeworksince the Career Panel's report.Tarlingupdates the
Career Panel's reviews(Visher and Roth, 1986; Cohen, 1986) of findings
about the career dimensions,and he replicatesseveral analyses reported
thereusingBritishdata bases maintainedby the Home Office.He presents
the resultsof analysesof A and T usingproportionalhazard, exponential,
and Weibullmodels.His concludingchaptersreportestimatesoftheincapacitativeeffectof currentBritishincarcerationpolicy (i.e., the percentageof
potentialcrimespreventedbyisolatingoffenders
throughincarceration)and
theelasticityof theincapacitationeffect
withrespectto changesin theincarcerationrate.These estimatesare computedin theusual way,bysubstituting
hisestimatesof A,5, and T intoformulasderivedbyAvi-Itzhakand Shinnar
(1973) and Shinnarand Shinnar(1975); theseformulashave also been used
by Greenwood(1982), theCareersPanel, and others(e.g., Cohen, 1978) to
estimateincapacitativeeffects.
Tarling'sreaderswill findsynopsesof European and recentcriminal
careers researchnot covered in the CriminalCareers panel report.And
readerswho do not regularlysee Home Officeresearchreportswill learn
about severalBritishdata bases describingarrestsof cohortsof newborns,
sentencedoffenders,
and parolees,whichcould proveusefulin cross-national
of
criminal
careersand recidivismpatterns.
comparisons

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Careers
Criminal
Research

99

is cost-effective.
His
Spelman'scentralquestionis whetherincarceration
for
the
Shinnar
but
that
is
also
model,
departurepoint
question
answering
he extendsit in threeimportantways. He replacesseveralparameterswith
distributions
and relationshipsthathe estimatesfromthedatabase used by
Greenwood: self-reported
values of A froma sample of inmatessurveyed
in
the
Rand
Corporation 1978 (Petersonet al ., 1982). To cope withthese
by
additionalcomplexities,he uses simulationsinsteadof algebraicformulas
to estimateincapacitationeffects.
And he extendsthedomainoftheincapacitationmodel fromprisonsentencesto otherpolicy instruments,
including
sanctionssuch
police and prosecutionresourceallocationsand intermediate
as part-time
incarceration
and intensivesupervision,
whichalso have partial
effects.
incapacitation
Spelman's readerswill findall the ingredientsof a thoroughpolicy
analysis: a well-defined
policyquestiontranslatedinto a model withknowa selectiveliterature
reviewand creativeanalysisto produce
able parameters,
that
narrow
therangesof uncertainty
around
"best
acknowledged
guesses"
of
thoseparameters,liberaluse of sensitivity
to
estimate
the
effects
analyses
in
errors
the
best
of
the
model
to
a
wide
range
plausible
guesses,applications
of policy options,and a concludingdiscussionof omittedconsiderations
In thecourseof refining
thatcould substantially
altertheanalyticalfindings.
the incapacitationmodel, Spelman reportsnew findingsabout criminal
others,and offersseveral
careers,challengessome old ones whileconfirming
methodologicalinsightsabout analyzingcriminalcareers and estimating
incapacitationeffects.
Researchersare likelyto welcomeSpelman's answerto the cost-effectivenessquestionmorewarmlythan will legislatorsand othersin searchof
unambiguouspolicy advice. Appropriately,he concludes that substantial
of incarceraeffects
progresshas been made in estimatingtheincapacitative
tion but thatits cost-effectiveness
depends on magnitudesabout whichwe
effect
stillknowverylittle: theindirectcostsof crime,thegeneraldeterrence
of incarceration,
of incarceration
on thesubsequentcriminal
and theeffects
career.
For both authors,criminalcareerfindingsremaina matterof description,not causal understanding.
Althoughcausal conjecturesare sprinkled
throughoutboth books, neithercontainsformaltestsof causal hypotheses
about theprimarycareerdimensions,and neitherauthorentersthedebates
about themeritsofthecareerparadigmand prospectivelongitudinalstudies.
Both authorsofferestimatesof A, with Spelman's based on the selfreportsof theRand sampleand Tarling'sbased on theconvictionrecordsof
two birthcohortsand a parolee sample.Spelmangenerallyfollowsprevious
data butdepartsfromprevious
analystsin "massaging"theRand self-report
practicein his treatmentof specificoffendersubpopulations.His review

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

100

Roth

(pp. 70-79) argues persuasivelythat inherentcriminal justice system


accountsformuchof the sizable increasesin estimatedA as one
selectivity
proceedsfromthe entirecriminalpopulationto samplesof arrestees,then
to prison inmates.One implicationis that policy analystsshould use A
estimatesspecificto thepopulationaffectedby thepolicybeinganalyzedobtainedby studyinga representative
sampleof thatpopulationifpossible
or by reweighting
an available nonrepresentative
sample if necessaryand
appropriate.
For his own analyses,Spelmanreweights
the Rand sample,whichwas
selectedto representan incomingcohort of prison inmates,to represent
adult "activeoffenders."
His conceptualdefinition
of "active" is hardto pin
on
down, because different
wordingsappear
pages 23, 25, 68, 74, and 76.
the
definition
seems
to be adults arrestedat least
however,
Operationally,
twicein theirlifetimes
FBI indexcrimesat least as
others
who
commit
plus
but
arrest
and
frequently escape
imprisonment
by chance. In any event,
from
common
which
Spelman departs
practice(in
everyonewho commits
at least one crimeduringa reference
is
as "active") by first
defined
period
the
of
his
narrower
"active"
behavior
analyzing
population,thenexplicitly
his
estimates
for
minor
crimes
thatare unlikelyto
adjusting incapacitation
be avertedby imprisoningadults because theyare committedby juveniles
or byveryoccasionaladultoffenders
who are unlikelyeverto be imprisoned.
He estimatesthemean A forhis activeoffenders
as 30-50 FBI indexcrimes
which
he
for
later
simulationsas the mean
per year (p. 76),
parameterizes
of a lognormaldistribution.
fromdistributions
of theannualTarlingderiveshis A distributions
ized individualconvictionfrequencyfor an "active offender"subsample
definedsimilarly
to Spelman's.He thenfactorsup theconvictiondistribution
a
constant
by
'/q, with q estimatedas the aggregateconvictionrate for
crimes reportedto police, includingthose committedby non-"active"
offenders.
The upwardbias fromincludingthosecrimesin the estimateof
A is morethanoffsetby largedownwardbiases due to the failureto adjust
forcrimesnot reportedto police and formultipleconvictionspercrimethat
can occur when multiple offenderscommit a single crime. Rough
adjustmentsforboththesefactors(assumingTarling'sestimatedaverageof
1.5 offenders
per crimeand a 50% rateof crimereportingto police) would
raiseTarling'sestimatedrangeformean A to 3-11. This rangeis consistent
withothers'estimatesforarresteesamplesbutis low comparedto American
as Tarling's and
samples of "active" offendersdefinedas restrictively
Spelman's.
Both authorsalso examineoffenders'
mixesof crimetypes.Following
commonpractice,Tarling(pp. 123-125) approachesthematterusingarrest
transitionmatrices,and he replicatesa familiarfinding:a "statistically

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Criminal
Careers
Research

101

. .but not particularlystrong"degreeof offensespecialization.


significant.
As Cohen (1986, pp. 367-374) noted, however,thereare two principal
in interpreting
arresttransitionmatricesin termsof behavior.
difficulties
The firstis thatarrestsare a small and almostcertainlynonrepresentative
is theecologsampleof crimes.The second,whichSpelmanalso highlights,
to ascerare
needed
ical fallacy.Data on individuals'crime-type
sequences
tain what combinationof perfectspecialists,partial specialists,perfect
"randomwalkers,"and desistersfromeach crimetypeaccountsfora particular transitionmatrix. The Rand data permit Spelman to replicate
Chaiken and Chaiken's (1982) individual-level
analysisrelatingcrime-type
mix and A, thento estimatea regressionrelationshipbetweenthe A's for
personaland propertycrime.Insteadof usingthisrelationshipin estimating
however,he combinestheseparatecrimetypesintoa
incapacitationeffects,
index.
seriousness-weighted
Both authorsalso addressdropoutfromthe criminalcareer.Tarling's
estimatesof severalmodels of reciChapter6 reportsmaximum-likelihood
of
British
a
30-month
divismduring
parolees.The modelspartition
followup
thatare familiarto readers
Aand r,thedropoutrate,usingparameterizations
of thisjournal: proportionalhazards, exponential,and Weibull distributions.Because of computationaldifficulties
reportedby others,he does not
attemptto estimateany split-populationsurvivalmodels with covariates
(see, e.g., Schmidtand Witte,1988; Rhodes, 1989). In Chapter7, usinga
stochasticprocessmodel of the officialrecordof the entirecareer,Tarling
estimatesof 0.446 formean fi,
derivesand calculatesmaximum-likelihood
the individualannual frequencyof court appearances,and 7.5 years for
mean T. Introducingcovariates,Tarling findsthat fj,increaseswith the
numberof previouscourt appearances. Like most previoussimultaneous
estimatesof A or j, and T, Tarling'sare obtained under the assumptions
that T is distributed
exponentially(or, equivalently,thatthedropoutrater
is constantthroughoutthe career), that A and T are distributedindependently,and thatA is constantthroughoutthecareer.
Spelman's treatmentof dropoutinvolvesinnovationsthat replacethe
firsttwo of theseassumptionswithrelationshipsestimatedfromthe Rand
between
data. The firstreplacementresolvesa decade-longincompatibility
empiricalfindingsabout r and the adjustmentfactorthat Avi-Itzhakand
Shinnar(1973) had derivedto adjust incapacitationestimatesfortheprison
Theiradjustment
whosecareershad terminated.
space occupiedby offenders
factorhad been derivedunderthe assumptionof a constantr withrespect
to age. However,firstBlumsteinand Cohen, withHsieh (1982), analyzing
arrestdata, and now Spelman (pp. 148-154), using the Rand self-reports,
founda "bathtub-shaped"timepath of r over the age range 18-50. For
simulationpurposes,Spelmanapproximatesthe"bathtub"byconcatenating

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Roth

102

a declining"weedthreeWeibull distributionsrepresenting,
respectively,
out" rate for amateursduringtheir20s, a rising"burnout" rate of hardcore persistersstartinglate in their40s, and a low constantrate of "stressthecareer.Withthisconcatenateddistribution
related"dropoutsthroughout
insteadof a constantr,Spelman'ssimulationsyielda revisedcareertermination adjustmentwith decreasingmarginalwastage of prison space: Each
additionalyear of timeservedwastes less than the precedingyear because
more of the population servingit lies in the age range of low dropout
rates.
Replacementof the second assumption,independenceof A and T, is
of selective
motivatedby the argumentthat estimatesof the effectiveness
if
have
career
will
be
biased
offenders
different
incapacitation
high-A
lengths
To testthispossibilityand developa corrective
thanotheroffenders.
adjustof r foroffenders
at five
ment,Spelman beginsby estimatingdistributions
A
across
the
Rand
inmates'
distribution.
He
adjusts
percentilepointsspread
thosedistributions
usinga procedurethathe describesas "Rube Goldbergesque" and thatis fartoo intricateto summarizehere.However,itsfunda- constantoverall A distributionsover the 3
mental requiredassumption
- is fairlywell
in
theRand inmates'self-reports
of
described
years offending
in
the
book
data
(Table 4.5, p. 163). Spelmanestisupportedby
presented
are 2 to 3 years
mates that the careersof both high-and low-Aoffenders
than
of
moderate-A
This
raises
an etiological
those
offenders. finding
longer
question (about which he speculates) and providesanotherpiece of the
simulationmodel forlateruse.
In Chapter5, Spelmanlays a shakierfoundationforreplacinganother
constantin the incapacitationformula,q9 with a learningfunction.His
q tendsto fallover timeas he or she learns
hypothesisis thatan offender's
fromexperience(measuredas crimescompleted)and to riseas police learn
fromexperience(measuredby arrestsof that offender).If, as he reports,
offenderlearningdominates,incapacitationmodels based on an identical
and constantq for all offendersoverestimatethe crimecontroleffectof
withtheirlargernumbersofprevilongersentencesbecausehigh-Aoffenders,
ous crimes,disproportionately
escape arrest.
WhileSpelman'sanalysisofq drawsattentionto a potentially
important
to
issue, his estimateof the net learningfunction,Eq. (5.11), is difficult
accept at face value. Unbiasednessand consistencyof the estimatesrest
on nonstochasticpolice and offenderlearningfunctionsand two untested
windowin
assumptions: constantindividualq9s duringthe 3-yearreference
theRand survey(whichare incompatiblewiththesteepnetlearningcurves
he estimates)and statistically
measurement
errorsin self-reports
independent
of arrestsand crimes.Also, because the data cover only the 3 reference
prewindowlearning,which
years,theestimatedlearningcurvecannotreflect

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Criminal
Careers
Research

103

shoulddiffer
acrosssampledoffenders
at different
stagesoftheir
significantly
careers.
Aftertheirchaptersdescribingcriminalcareers,both authorsreport
estimatesof the collectiveincapacitationeffectsof currentpolicy and of
changesin policy.Using thestandardAvi-Itzhakand Shinnar(1973) model
withthe correctionfor career termination,
Tarling estimatesthat current
6 to 7% of reportedseriouscrime
Britishincarcerationpolicyis preventing
and thatfurther
prisonpopulationincreasesof 16 to 20% would be needed
- a similarestimateto that of the
to achievea further1% crimereduction
Criminal Careers Panel for the United States (Blumsteinet al., 1986,
pp. 148-149).
This familiaranalysisis the departurepoint forSpelman's analysisof
in whichpoint estimatesof the key parametersare
incapacitationeffects,
the
distributions
and ranges developed in earlierchapters.
with
replaced
all thosecomplexitieswould be intractable.ThereFormulasincorporating
fore,Spelman obtains his estimatesthroughsimulation,using the BoxBehnkenfactorialdesignto selectvalues fromthe estimateddistributions
and the ranges around the "best guesses". His simulationsindicatethat
about 8% of all activeoffenders,
currentpolicyis fairlyselective,imprisoning
of
crimes.His simulationstake
he
commit
about
21%
index
who, estimates,
themarginalincarcerof
the
fact
that
because
of
account
systemselectivity,
A
A
the
for
all
active
offenders.
offender's
exceeds
ated
Therefore,
average
of
are
than
those
his elasticityestimates
Tarlingand the Criminal
higher
CareersPanel- onlyabout a 6% increasein prisonpopulationis neededto
reducecrimeby 1%. AttachingCohen's (1988) estimatesof directcosts of
to theseelasticities,
crimeand incarceration
Spelmanestimatesthatincarceriftheindirectbenefitof avertinga crimeis at least 2.6
ation is cost-effective
timesthedirectcost per crime.Using Cohen and colleagues' (1994) revised
costestimatesforrobberyand assaultwould reducethebreakeventhreshold
forthosecrimesto about 2.0.
of alternativeselectiveincaIn Chapter7, Spelmanestimatestheeffects
pacitation policies, which allocate more police/prosecutoror prison
than to others.Proresourcesto predictedhigh-Aand/orlong-T offenders
ducingtheseestimatesrequiresaddingfeaturesto thesimulationmodelthat
describethe selectivepolicies in the followingterms:predictionaccuracy
(measuredbytetrachoric
r2),selectionrate(i.e., thefractionselectedas highrisk),disparityratio (i.e., theratioof therespectivesentencelengthsforthe
predictedhigh- and low-riskcategories),and a productionfunctionfor
resourcesin increasingtheexpectedtermof incarceration
police-prosecutor
by increasingtheprobabilityof arrestand conviction.To describethesimulation resultsconcisely,he uses a regressionanalysis of crime-reduction
of both the populationand the policies.
on characteristics
effectiveness

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Roth

104

The simuiationsreportedin Chapters7 and 8 generatea richarrayof


estimatesand insightsabout alternativeselectiveincapacitationpolicies
underfixedresources.Given available predictiontechnology(tetrachoricr2
between0.2 and 0.4) and fixedcriminaljusticeresources,theoptimalselectivesentencing
policy(a selectionrateof about 12% and a disparityratioof
would
reduce
crimeby about 4%; this enhanced effectiveness
would
20)
even if the indirectbenefitsare as littleas
make incarcerationcost-effective
2.1 timesthedirectbenefits.If ethicalconsiderationsprevented
judges from
five
the
sentences
to
more
than
times
offenders
sentencing
predictedhigh-risk
forotheroffenders,
60 to 65% of thecrimereductionbenefitwould stillbe
achieved. The optimal selectivepolicingand prosecutionpolicy withfixed
- increasingthe probabilityof incarcerationby a factorof about
resources
3 to 5% of offenders
and decreasingit forthe rest30 forthe highest-risk
would reducecrimeby about 6%, whichis morethantheoptimalsentencing
programsare less
policy.However,because thebestfeasiblerepeatoffender
than the optimal program,theycan be expectedto achieve only
efficient
about halfthisreduction.
Most of theresponsecurvesgeneratedby thesimulationsare fairlyflat,
diminishesslowlyas disparityand
so thatthepredictedincapacitationeffect
selectionratios divergefromtheiroptimums.For the same reason, while
optimal predictiveaccuracy always exceeds currentlyachievable levels,
increasethe predictedincapacitagreateraccuracywould not substantially
effects
could more
deterrence
tioneffect.Whileplausiblecounterproductive
than offsetthe crimecontrolbenefitsof selectiveincapacitationpolicies,as
Cook (1986) found,Spelman offersarguments(but no evidence)that the
likelydeterrenceeffectsare smallenoughto be ignored.Finally,depending
on the indirectbenefitsof crimecontroland on how intensiveand regular
or one of thoseintermediate
probationaffectq and A,eitherimprisonment
sanctions may be the most cost-effective
punishmentfor the marginal
offender.
of theanalysesreportedin bothbooks
Severalproblemsand limitations
have alreadybeen pointedout. Of Spelman's,all but thespecification
problemsin theanalysisofq seemto be fairlywellhandledbysensitivity
analyses.
One simulationresult(p. 157) that surprisesthis readerbut not Spelman
towardthatof coin-flipping,
impliesthatas predictionaccuracydeteriorates
ofpersons
increasingly
disparatesentencesshouldbe givento largerfractions
classifiedas dangerous- a puzzlingconclusiongiventherationaleforselective incapacitation.One cost of the analyticalpower of simulationis the
of peeringinto the black box to findthe reason for surprising
difficulty
results.
Spelman's discussionson pages 172 and 291 illustratethe danger of
misinterpreting
incapacitationeffectsexpressedin percentageratherthan

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Criminal
Careers
Research

105

absolute terms.To illustratethe pointthat"the betterthe systemis doing


already,thebetterit can do by increasingitsrelianceon generalincapacitaon
tion," he examinesthe reductionsin the percentageof activeoffenders
thestreetthata 10% increasein prisonspace would achieve.The percentage
reductionis, ofcourse,greaterwhen40% ratherthan20% of activeoffenders
like an exceptionto the
are already in prison- which looks superficially
But
the
of
returns.
percentagesare misleadingbecause
principle diminishing
in theformercase the 10% increaserequirestwiceas manycellsand removes
froma base thatis only 75% as large as in the lattercase.
activeoffenders
as
Moreover, Spelmannoteselsewhere,analysisof crimesprevented(which
is of greaterpublic interest)insteadof offendersremovedwould need to
alreadyoffthe street,
adjust forsystemselectivity:With40% of offenders
on average,thanifonly20%
thenewcellswould hold less prolificoffenders,
werealreadyin prison.
Both books occasionallysound "out of tune" whentheydiscusspublic
policy or opinion. If debatersover futurecrime legislationwant to pay
serious attentionto Spelman's findings(as this reader hopes theywill),
crimeindex will be an unpopular substitutefor
his seriousness-weighted
disaggregatedmurder,robbery,and assault counts. Policy commentators
of incarwillwishthatthebook containedestimatesof thecost-effectiveness
is
understandable
the
omission
although
ceratingconvicteddrugoffenders,
in view of the conceptualand measurementproblemsassociatedwithdrug
careers.In 1994 it feelsstrangeto read Spelman's comments
distribution
presumingthat the public is unwillingto increase spendingon prisons
Tarling'scomment(p. 161) that"interest[in
(p. 237,forexample).Similarly,
the United States] in incapacitationseemsto be waningfollowingresearch
evidencepointingout thelimitationsof sucha policy"seemsto describethe
stateof public interestin the research,not in the policy.
These books suggest both applied and basic directionsfor future
researchon criminalcareers.On the applied side, Spelman'sincorporation
of a productionfunctionforq into incapacitationanalysisbringsto mind
thatin economics,the productionfunctionplays one of the roles that the
: as a deviceforsummarizcriminalcareerconcepthas playedin criminology
ing a host of complex relationshipsin a formthat generatespredictions
about how outputsvary when inputsare changed. Justas incapacitation
and criminalcareermodels predicthow changesin prisonpopulationwill
affectcrimegivenA and T, a productionfunctionforelectricity
generation
predictshow changesin fuelinputwill affecta powerplant's outputgiven
Econometric
ofitsturbogenerators.
thesizeand technologicalcharacteristics
willnotyieldknowledge
no matterhow sophisticated,
productionfunctions,
ifa managerneedsoperatingrules
about fundamental
physics.Nevertheless,
across multiplepower plants of
for balancing load variationsefficiently

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

106

Roth

to derivethemfromproducvariedtechnologies,it is morestraightforward
tion and cost functionsthan directlyfromthe principlesof physicsthat
- eventhoughtherelevantprinciplesof physicsare
underliethosefunctions
more preciseand settledthanthe relevantprinciplesof criminology.
Spelman's work pushes the familiar incapacitationmodel in the
directionof a productionfunctionin whichtheoutputis crimesavertedand
the inputs are resourcesdevoted to imprisonment,
policing,prosecution,
sanctions.Its preintensivepostreleasesupervision,and otherintermediate
dictionswillbecomemoreaccurateas futureresearchimprovesmeasuresof
the distributionsof A, q, and the numberof co-offenders
per crime. Its
as
will
more
and
flexible
distribution
become
crimes,
specific
drug
predictions
X distributions
and crime-specific
are
individual-level
transitions,
crime-type
into
the
model.
And
the
model
could
become
useful
for
comincorporated
paringa broaderrangeof policyoptionsas betterestimatesemergeof how
theyaffectparametersof the criminalcareer and incapacitationmodels.
Examples of such relationshipsinclude the effectsof plea bargainingon
ofcommunity
convictionprobabilityand sentencelength;theeffects
policing
shelterson q, the probabilityof arrest; the postand threatened-witness
releaseresponsesof A and r to in-prisonsubstanceabuse treatment;
and the
of
to
various
interventions.
Given
such
data
(and
preventive
responses Da
to use
presumablysome ambitiousbut manageablesoftwareenhancements
them),Spelman'ssimulationmodelwouldofferone ofthefewavailabletools
of incarceration
withthatof preventive,
forcomparingthecost-effectiveness
- an issue that mightreceive
and proceduralinterventions
rehabilitative,
moreweightin policydebatesifthecomparisonsweremorereadilyavailable.
estimatesof how
These extensionswill requirenew or reinterpreted
policingand prosecutionpracticesaffectcase outcomesand how criminal
In addition,
interventions.
and rehabilitative
behaviorrespondsto preventive
that
of cost-effectiveness
empiricalworkis neededon thethreedeterminants
are not incorporatedeven in Spelman'sextendedmodel: generaldeterrence
the indirectcosts of crime,and the effectsof incarcerationon the
effects,
subsequentcriminalcareer.Therefore,merelycontinuingtheextensionsof
applied researchbegun in Spelman's book definesambitiouslong-range
goals forfutureempiricalresearchin thecriminalcareerframework.
As mentionedearlier,questionshave been raised about the scientific
value of the criminalcareer frameworkand the prospectivelongitudinal
studiesit has stimulated.This reviewis not theplace to add to thatdebate.
However,the two books implicitly
point to at least fourfuturedirections
thecase fortheparadigm
criminal
careers
research
that
for
mightstrengthen
studies
for
and
prospectivelongitudinal
by exploitingtheirpotentialmore
fully.
First,analyses of individuals'criminalcareers,like analysesof legalsectorcareers,may not demonstratetheirfullpoweruntiltheyincorporate

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Criminal
Careers
Research

107

more descriptorsof community-level


processes.Because the available data
of largesocial units,testsof theincrebases generallyincludeno descriptors
mentalexplanatorypower of the career model (e.g., Rowe et al ., 1990;
Nagin and Smith,1990; Nagin and Land, 1993) have notconsideredspecificationsin which,forexample,one's careeronsetis influencedprimarily
by
individualattributesand family
/peergroup processesin childhood,while
one's A as an adult is more heavilyinfluencedby collectiveprocessesthat
and refracted
havebeenselected,filtered,
throughpersonaland familyconditionsat thattime.Potentiallyrelevantcausal processesare likelyto involve
suchcollectiveentitiesas a neighborhood'sdensityofcriminalopportunities,
social networksthatconnectindividualsto legal-sectoremployment
opporor an "oppositionalculture"thatvaluesthepersonalstyleand skills
tunities,
value
thatlead to successin illegalmarkets(Anderson,1990). The scientific
bases
in
data
social
units
of
of includingdescriptors larger
longitudinal
in
less
of
behavior
lies, course,
resolvingcompetitions
describingindividual
theoriesthatintegrate
thanin testingand refining
betweenparameterizations
individual-and social-levelprocesses.
in
and officialrecordsof offending
Second, includingboth self-reports
of
research
opportunitiesbeyond
longitudinaldatabases opens up a range
and cross-validatingestimatesof careerdimensions.Examples
triangulating
about q fromtheorderof magniincludereducingtherangeof uncertainty
tude thatexistsnow (Spelman, 178-179); estimatinghow arrest,incarceration, and intermediatesanctions interactwith other influencesto alter
offenders'subsequentcriminaland legitimatecareers; and studyinghow
theirown behaviorand thatof thecriminal
individuals'arrestrecordsreflect
justicesystem.
Third, because the advantages claimed for prospectivelongitudinal
databases relateto thecareeras a dynamicprocess,thoseadvantagescannot
be fullydemonstrated
byanalyzingthemwithcomparativestaticmodelsand
techniques.What is needed,forexample,are fewertabulationsof aggregate
statisticsabout careersand moreanalysesof overlappingsubcareersin multiplecrimetypes(e.g., Chaikenand Chaiken, 1982; Lehoczky,1986). More
generally,greateruse is neededof available techniquesforanalyzingmultirecordedin a longitudinaldatabase
on eventsand trajectories
levelinfluences
(see, e.g., Strenioet al., 1983; Stiratelliet al ., 1984; Brykand Raudenbush,
1992).
Fourth,linkinga varietyof "small science" projectsto a multilevel
longitudinaldata collectionprogramcould broaden the rangeof scientific
perspectivesinvolved,spread thefixedcosts of longitudinaldata collection
over more objectives,and enrichthe largerenterprisein a varietyof ways.
Examples of such projectsinclude intensiveobservationand analysis of
about developmento helpinterpret
quantitativefindings
familyfunctioning
data collection(and perhaps case control
tal trajectories,supplementary

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

108

Roth

analyses)to develophypothesesforexplainingoutliercases,and observation


and analysesof researchsubjects'encounterswithformalinstitutions
such
as welfareand housingauthoritiesto discover"soft" and "hard" riskand
protectivefactorsin such encounters.
In short,Tarlingoffersa comprehensive
surveyof thekindsof findings
thatthe quantitativecriminalcareersperspectivehas generatedin its early
the careermodel and
years.Spelmanhas takenimportantstepsin refining
For
who
its
domain.
those
would
use the careermodel
expanding policy
and prospectivelongitudinaldata to addressmorebasic questions,thechalintocareermodels
lengesnow may be to incorporatesocial-levelinfluences
and databases,to makemorediverseuses of databases assembledto analyze
criminalcareers,to analyze longitudinaldatabases usingtechniquesdevelofenriching
oped to modeldynamicprocesses,and to considerthepossibility
with
smaller
data
collection
related
studiesthat
programs
largelongitudinal
of
other
scientific
traditions.
out
grow
ACKNOWLEDGMENTS
The author is gratefulto an anonymousreviewerand, especially,to
JohnLaub, forcommentson an earlydraft.
REFERENCES
E. (1990).Streetwise:
Race, Class
inan Urban
, andChange
, University
Anderson,
Community
ofChicago
Press,
Chicago.
incrime
R. (1973).Quantitative
models
control.
J.Crim.
Just.
Avi-Itzhak,
B.,andShinnar,
1:185-217.
P. (1982).TheDuration
Criminal
Careers
Blumstein,
A.,andCohen,
J.,withHsieh,
ofAdult
totheNational
Institute
ofJustice,
(FinalReport
University,
Pittsburgh.
Carnegie-Mellon
and"Career
Careers
C. (eds.)(1986).Criminal
Blumstein,
J.,Roth,
J.,andVisher,
A.,Cohen,
DC.
Criminals
," National
Press,
Academy
Washington,
research:
Itsvalue
D. P. (1988a).Criminal
career
Blumstein,
A,.,Cohen,
J.,andFarrington,
26: 1-36.
forcriminology.
Criminology
career
andcriminal
D. P. (1988b).Longitudinal
J.,andFarrington,
Blumstein,
A.,Cohen,
26: 57-74.
research
: further
clarifications.
Criminology
andMeta
Models:
Linear
A.S.,andRaudenbush,
S. W.(1992).Hierarchical
Applications
Bryk,
Park,CA.
, Sage,Newbury
Analysis
Techniques
Behavior
M. (1982).Varieties
, RandCorp.,Santa
Chaiken,
J.,andChaiken,
of Criminal
CA.
Monica,
A critical
review
oftheliterature.
J.(1978).Theincapacitative
effect
ofimprisonment:
Cohen,
EstimatandIncapacitation:
InBlumstein,
D. (eds.),Deterrence
A.,Cohen,
J.,andNagin,
Rates
Sanctions
onCrime
Press,
,National
Washington,
Academy
ingtheEffects
ofCriminal
DC.
seriousrates
andoffense
Individual
careers:
J.(1986).Research
oncriminal
Cohen,
frequency
and
Careers
C. (eds.),Criminal
ness.In Blumstein,
A.,Cohen,
J.,Roth,J.,andVisher,
"Career
CriminalsVol./,National
DC, pp.292-418.
Press,
Washington,
Academy

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Criminal
Careers
Research

109

A study
M. (1988).Pain,suffering,
andjuryawards:
ofthecostofcrime
tovictims.
Cohen,
LawSoc.Rev.22:537-555.
T. R.,andRossman,
S. B. (1994).Thecostsandconsequences
ofviolent
Cohen,
M.,Miller,
intheU.S.InReiss,
A.J.,Jr.,
andRoth,
J.A.(eds.),Understanding
andPreventbehavior
andResponses
, Vol.IV.Consequences
, National
Press,
ingViolence
Academy
Washington,
DC.
inanadaptive
framework.
effects
considered
choice
Cook,P.J.(1986).Criminal
incapacitation
Rational
In Cornish,
D. B.,andClarke,
R. V. (eds.),TheReasoning
Criminal:
Choice
onOffending
NewYork,pp.202-216.
, SpringerVerlag,
Perspectives
M. R.,andHirschi,
T. (1986).Thetruevalueoflambda
wouldappearto be
Gottfredson,
oncareer
criminal
selective
cohort
zero: Anessay
studies,
criminals,
careers,
incapacitation,
24:213-233.
andrelated
Criminology
topics.
T. (1988).Science,
andthecareer
M. R.,andHirschi,
Gottfredson,
paradigm.
public
policy,
26:37-56.
Criminology
T. (1990).A General
M. R.,andHirschi,
, Stanford
Gottfredson,
University
Theory
ofCrime
CA.
Press,
Stanford,
Rand
A.(1982).Selective
,Report
R-2815-NIJ,
Greenwood,
P.,with
Abrahamse,
Incapacitation
CA.
Corp.,SantaMonica,
insocialandlegalcontext.
LawSoc.Rev.27:255-262.
J.(1993).Introduction:
Crime
Hagan,
inthelifecourse:Reconceiving
A. (1988).Crime
as socialevents
a
Hagan,J.,andPalloni,
26:87-100.
criminological
controversy.
Criminology
careers:
Somesuggestions
formoving
the
Land,K. C. (1990).Modelsofcriminal
beyond
current
debate.
30: 149-155.
Criminology
In
J.P. (1986).Random
models
ofcriminal
careers.
Lehoczky,
parameter
stochastic-process
and"Career
C. (eds.),Criminal
Careers
A.,Cohen,
Blumstein,
J.,Roth,J.,andVisher,
Criminals
", Vol.2,National
DC, pp.380-404.
Press,
Academy
Washington,
D. S.,andLand,K. C. (1993).Age,criminal
andpopulation
careers,
Nagin,
heterogeneity:
andestimation
ofa nonparametric,
mixed
Poisson
model.
31:
Specification
Criminology
327-362.
D. S.,andSmith,
D. A. (1990).Participation
inandfrequency
ofdelinquent
behavior:
Nagin,
A testforstructural
differences.
J.Quant.
Crimol.
6: 335-356.
P. (1982).Survey
andJailInmates:
Peterson,
M.,Chaiken,
J.,Ebener,
P.,andHonig,
ofPrison
andMethods
CA.
, a RandNote.RandCorp.,SantaMonica,
Background
W.M. (1989).Thecriminal
career:
Estimates
oftheduration
andfrequency
ofcrime
Rhodes,
commission.
J.Quant.
Criminol.
5: 3-32.
W. A. (1990).A latent
traitapproach
to
Rowe,D. C., Osgood,D. W.,andNicewander,
criminal
careers.
28: 237-270.
unifying
Criminology
A. (1988).Predicting
Recidivism
Survival
Models
Schmidt,
P., andWitte,
, SpringerUsing
NewYork.
Verlag,
S. (1975).Theeffect
ofthecriminal
onthecontrol
of
Shinnar,
R.,andShinnar,
justice
system
A quantitative
crime:
LawSoc.Rev.9: 581-612.
approach.
J.H. (1984).Random-effects
models
forserial
observations
Stiratelli,
R.,Laird,
N.,andWare,
with
Biometrics
40:961-971.
binary
response.
J.L. F., Weisberg,
H. I., andBryk,
A. S. (1983).Empirical
of
Strenio,
Bayesestimation
individual
curve
andtheir
tocovariates.
Biometrics
39:
growth
parameters
relationship
71-86.
L. E., andFarrington,
D. P. (1991).Human
andCriminal
M.,Ohlin,
Tonry,
Development
NewYork.
Behavior,
Springer-Verlag,
C. A.(1986).TheRandinmate
a reanalysis.
InBlumstein,
Visher,
A.,Cohen,
J.,Roth,
survey:
C. (eds.),Criminal
Careers
and"CareerCriminals
J.,andVisher,
", Vol.//,National
Press,
DC, pp.161-211.
Academy
Washington,

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

Roth

110

In Blumstein,
inCriminal
Careers.
C. A.,andRoth,J.A. (1986).Participation
A.,
Visher,
and"Career
CriminalsVol.
Careers
C. (eds.),Criminal
J.,Roth,
Cohen,
J.,andVisher,
DC, pp.211-291.
/,National
Press,
Academy
Washington,
ofserious
therecidivism
C. A.,Lattimore,
P. K.,andLinster,
R. L. (1991).Predicting
Visher,
29: 329-366.
offenders
survival
models.
Criminology
using
youthful

This content downloaded from 185.2.32.90 on Wed, 18 Jun 2014 20:11:00 PM


All use subject to JSTOR Terms and Conditions

También podría gustarte