SEVERITY OF PUNISHMENT
JEFFREY GROGGER*
297
Economic Inquiry
Vol.XXIX, April 1991,297-309 @WesternEconomic Association International
298 ECONOMIC INQUIRY
who were not matched, roughly half cipal dependent variable, NARR86, is the
lacked Social Security numbers, while number of times the individual was ar-
earnings records were simply missing rested in 1986. About 27 percent of the
from the wage files for the rest. While the sample was arrested at least once in 1986.
final merged sample members tended to Conditional on arrest, the mean number
have slightly more arrests on average than of arrests was about 1.5. NFARR86 and
the original sample members, and there NPARR86 give the number of felony and
were proportionately fewer Hispanics, the property crime arrests in 1986, respec-
departure from randomness does not tively. Eighteen and 9 percent of the sam-
seem too severe. ple were charged with these types of
Forty-eight percent of the sample were crimes, respectively. Conditional means
arrested only once in the seven- to nine- for both types of arrest were the same at
year period between their eighteenth 1.3.
birthdays and the end of 1986, and 21 As in all studies of crime, we are re-
percent reported earnings well in excess of stricted to analyzing noisy indicators
national per capita income. On the other rather than a direct measure of criminal
hand, several individuals were arrested activity itself. Arrests, of course, are not a
more than thirty times and were fre- measure only of criminal activity, but also
quently incarcerated during that period. of the response of law enforcement agen-
The age-arrest profile of the sample, cies to prevent such behavior. In the dis-
shown in Table I, is similar to others cussion below, I will generally interpret
reported in the criminal justice literature, the arrest measures as proxies for the
rising sharply during the teens and declin- individual's true level of criminal activity,
ing thereafter. Age- earnings profiles, also although I do point out particular in-
given in Table I, exhibit the same steep stances where alternative interpretations
increase as is usually reported for this age would lead one to draw differing substan-
group in the labor economics literature. tive conclusions.
These univariate relationships, then, The variable PCNV is the proportion of
would seem to indicate a role for labor adult arrests prior to 1986 which resulted
market policies as a tool to reduce crime. in convictions, that is, the number of adult
To generate the sample used in this convictions divided by the number of
analysis, all individuals were excluded for adult arrests. This is the estimator of the
whom no arrest, court, or earnings activity individual's expected certainty of punish-
was recorded in any of the years 1984 to ment. The next measure, AVGSEN, esti-
1986. This criterion, which resulted in the mates the expected severity of punish-
exclusion of 951 of the 14929 original sam- m e n t ~Two
. ~ measures were initially exam-
ple members, is an admittedly coarse ined, the average length of prison senten-
means to account for migration of the ces served since age eighteen, in months,
sample members, which is unobservable and the length of the most recent sentence.
in the data. It should be noted, however, Since results were very similar, only equa-
that the erroneous inclusion of individuals tions containing the first measure are re-
who had in fact left California (hence
whose earnings and criminal activity are
unobserved) would lead to a bias of the
estimates reported below toward zero, 3. In the criminology literature, these variables are
said to provide measures of specific deterrence, that
weakening the reported results relative to is, of the effect of a given individual's criminal justice
the true magnitudes. history on his current behavior. An analysis of general
Table I1 presents summary statistics of deterrence, or the effect of community law enforcement
efforts on individuals' behavior, is provided by
the variables used in the study. The prin- Tauchen et al. [1988].
300 ECONOMIC INQUIRY
TABLE I
Arrests and Earnings, by Age
Age 18 19 20 21 22 23 24 25 26
Arrests per
100 persons 23 45 39 38 36 36 36 31 31
Earnings per
Capita 2656 3694 4538 5147 5714 6458 7079 8235 8475
11980 dollars)
ported below. Six percent of the sample leased from prison, whichever was more
spent some time in prison, among whom recent.
the average sentence served was about
111. THE EMPIRICAL MODEL
thirteen months. The variable PTIME86
gives the number of months the individual I specify the conditional expectations of
spent in prison during 1986 and will pro- arrests as a (generally non-linear) function
vide an estimate of the incapacitative ef- of the explanatory variables described
fect of incarceration. TOTTIME measures above, as well as binary, mutually exclu-
the total amount of time the individual has sive race- ethnicity indicators BLACK and
spent in prison since age eighteen and can HISPANIC and a binary variable BORN60
be considered an indicator of criminal indicating 1960 birth. The simplest models
human capital as discussed by Myers of criminal activity, such as those of Becker
[1983] or Grogger [1989]. Among those [1968] or Ehrlich [1973], would yield the
with any prison time, the average time intuitive predictions that conviction prob-
served was eighteen months. abilities and expected sentences should
The next three variables are measures enter the equation with negative signs, as
of labor market opportunities. The first, should contemporaneous prison time for
QEMP86, is a measure of employment, obvious reasons. Total prison time is ex-
giving the number of quarters in 1986 pected to enter positively. Such simple
during which some positive earnings were theoretical models would also predict neg-
reported. INC86 is reported earnings for ative coefficients for the employment and
1986, expressed in hundreds of 1980 dol- earnings variables. Intuitively, one would
lars. It should be noted that only earnings expect the duration of the current jobless
from establishments which contribute to spell to enter the equation with positive
the unemployment insurance system are sign, indicating the greater attractiveness
included. For the time period covered, this of crime after longer periods of unemploy-
includes virtually all non-federal employ- ment.
ers. Not included, however, are cash or
any other under-the-table payments. DU- IV. ESTIMATION
RATION is a measure of the duration of The non-negative integer nature of the
the current spell of unemployment. For all dependent variable makes one of the so-
individuals who reported zero earnings in called count regression methods the natu-
the first quarter of 1986, DURATION gives ral choice in estimating the above model,
the number of quarters since the individ- as discussed by Hausman, Hall, and
ual last had positive earnings or was re- Griliches [1984]. Based on the Poisson
GROGGER: CERTAINTY VS. SEVERITY OF PUNISHMENT 301
TABLE I1
Summary Statistics
Standard
Variable Mean Variance Deviation Min Max
~~ ~
TABLE I11
Estimates of Basic Model Parameters
Negative Binomial Non-Linear Instrumental
Variable Estimator Variables Estimator
possible power. Under the null hypothe- cients of all other regressors are virtually
sis, H has an asymptotic 13 distribution. identical across estimation method. The
Table 111presents the results of negative coefficients of the two suspect variables do
binomial and non- linear instrumental vary somewhat, although by an amount
variables estimation and the Hausman test insufficient to warrant rejection of the null
statistic for the exogeneity of employment hypothesis. The results discussed in the
and earnings. Note first that the coeffi- next section are therefore based on the
304 ECONOMIC INQUIRY
activity to employment and earnings. The arrests, and arrests for property and non-
point estimate of the coefficient of the property crime^.^ Consider first the felony
employment variable is negative, but not and non-felony arrests. Comparing esti-
at all significant. The income coefficient, mation results for these models with each
however, is negative and highly signifi- other and with the estimates for all arrests
cant, indicating that a $100 increase in presented in Table 111, we see that in-
earnings reduces arrests by just under 1 creases in the probability of conviction
percent, on average. This effect is very seem to be more effective in deterring
large; a one-standard deviation rise in serious felony crimes than less serious
income is associated with an average de- crimes. Sanction severity, on the other
crease in arrests of 48 percent. The non- hand, is about equally (in)effective in de-
employment duration coefficient indicates terring both types of criminal activity. One
that increasing the length of a current also sees, by comparing the estimates of
jobless spell by one calendar quarter in- the coefficients of PTIME, that most of the
creases arrests on average by about 2 per- crimes prevented by the incapacitative ef-
cent. A one- standard deviation increase of fects of prison are of the less serious non-
4.6 quarters results in an average increase felony type. The estimates of criminal
of about 7 percent. human capital effects are roughly the
Finally, the demographic indicator vari- same for both models.
ables indicate that blacks and Hispanics The estimates of employment effects are
on average were arrested 66 and 52 per- quite different for these two types of
cent more often than whites, respectively, crimes. An additional quarter of employ-
and that members of the older cohort were ment is associated with a 4 percent de-
arrested about 14 percent fewer times on
average. These numbers therefore reflect
race and age effects very commonly found 4. Note that the conviction probability and aver-
in studies of crime. agy sentence variables are still computed from
These estimates provide evidence that individuals' entire criminal history, rather than felony
or property arrests, convictions, and sentences only.
both criminal justice sanctions and eco- This was done for a number of reasons. First, data
nomic factors play a considerable role in limitations preclude one from ascribing prison terms
to the arrest or conviction from which they resulted.
determining the extent of an individual's Next, when a felony conviction probability variable,
criminal activity, at least as proxied by constructed from felony arrests and felony convic-
official arrest statistics. While these results tions, was used in the felony arrest equations, quali-
tative results were similar to those shown, but the co-
are quite useful in resolving some of the efficient was less significant. This might indicate that
conflicting evidence stemming from previ- felony conviction rates are an inefficient estimator of
the individuals' expected probability of conviction, ex-
ous research, the size of the data set lets cluding important information known to the offender
me examine these effects in yet more de- about the likelihood of conviction. When a property-
tail. In particular, it is of interest to deter- crime conviction probability variable, constructed
from property arrests and property convictions, was
mine whether different types of criminal used in the equation for property arrests, it entered
activity are influenced differently by crim- with a perverse positive sign. Again, this brings into
question the validity of using only a part of the avail-
inal justice and economic factors, and able information set in attempting to estimate
whether the responsiveness of criminal individuals' expected conviction probabilities. Fur-
thermore, mandatory prison terms for burglary were
activity in general differs among different introduced in California in the early 1980s. This is
demographic groups. I therefore turn first likely to have led police to "overcharge" lesser of-
to an analysis of felony and property ar- fenses, and there is evidence that such mandatory sen-
tencing schemes lead prosecutors to reduce conviction
rests, followed by estimates of the general charges to avoid overly harsh sentences for many rel-
model disaggregated by race. atively minor crimes (see, e.g, Blumstein et al. [1983,
24-30]). In this case, restricting an estimator of convic-
Table IV presents parameter estimates tion probabilities to property offenses may well lead
for models of felony arrests, non-felony to inconsistent and misleading results.
306 ECONOMIC INQUIRY
TABLE IV
Negative Binomial Estimates for Models of Felony and Property Arrests
Felony Non-Felony Property Non-Property
Variable Arrests Arrests Arrests Arrests
-- - -
PCNV = probability of conviction; AVGSEN = average prison sentence length; TOTTIME total
- -
prison time; PTIME86 time in prison in 1986; QEMP86 quarters employed in 1986; INC86 =
1986 income; DURATION duration of current jobless spell; BLACK 1 if black; HISPANIC = 1
if Hispanic; BORN60 1 if born in 1960.
crease in felonies, but a 6 percent increase two crime types are pooled, one erron-
in non-felonies. Both effects are at least eously concludes that employment has no
marginally significant as well. This differ- effect on criminal activity.
ence in employment effects explains the Income effects, in contrast, are roughly
small and insignificant coefficient in the the same across crime types, while felony
aggregate model; apparently, relatively crimes are more sensitive to the duration
minor criminal activity complements em- of a spell of joblessness. The coefficients of
ployment, while employment and serious BLACK and HISPANIC indicate that blacks
crime are substitute activities. When these are even more overrepresented in the se-
GROCGER: CERTAINTY VS. SEVERITY OF PUNISHMENT 307
rious crime category than is generally eral, the results for blacks accord most
true, while more arrests of Hispanics are closely with the intuitive predictions of
for non-felony offenses. simple economic models of criminal activ-
Turning now to the results for property ity, whites accord the least, and Hispanics
and non-property crimes, we see that non- lie in between.
property arrests are much more respon- The certainty of punishment measure is
sive to higher conviction probabilities negative and significant across all three
than property arrests, but that sanction groups, but is strongest among whites,
severity matters much more for the prop- and weakest among Hispanics. In con-
erty offenses than the non-property cate- trast, the severity of punishment measure
gory. Incapacitative effects are larger for is negative for both blacks and Hispanics,
non-property offenses as well, while in- and significant for blacks, but is positive
creases in criminal human capital lead to and marginally significant for whites. The
slightly greater increases in property of- incapacitative effects of prison are at best
fenses. marginally significant among whites, but
Again we see mixed results for the for blacks indicate a proportionate reduc-
effect of employment. The results for tion in arrests in excess of the proportion
property crimes indicate that property of time excluded from the general popu-
crimes and employment are complemen- lation. The positive effect of past prison
tary, as was seen for non-felony arrests time on current arrests is strongest among
above. Together, these findings suggest blacks, and weakest among whites.
that relatively minor property crimes are For whites and Hispanics, we again see
undertaken to supplement income from evidence of complementarity between my
conventional employment. Another possi- measure of crime and conventional em-
bility is that these positive relationships ployment, although for Hispanics this ef-
reflect employee theft from the workplace, fect is insignificant. Earnings effects are
although the data do not allow this hy- negative and significant for all groups,
pothesis to be tested directly. Property and slightly greater in absolute value
arrests are seen to be roughly twice as among Hispanics than among whites and
responsive to changes in income as are blacks. The effect of longer spells of job-
arrests overall, and three times as respon- lessness is greatest among whites, but neg-
sive as non-property arrests. The coeffi- ative and marginally significant among
cients of the demographic variables indi- Hispanics. Finally, negative age effects
cate that blacks are arrested much more vary by race, being strongest among
often for property crimes than whites or whites, and weakest among blacks.
Hispanics, and that Hispanics are much
more likely to be arrested for a non-prop-
erty crime than for a property offense. The VI. CONCLUSIONS
age coefficients indicate a slightly slower ‘This paper provides estimates of deter-
”ageing-out” effect for property offenses rent, incapacitative, and criminal human
than for other crimes. capital effects, as well as measures of the
The distribution of arrests is well responsiveness of one indicator of crimi-
known to differ widely by race; it is there- nal activity, arrests, to employment and
fore of interest to examine whether the earnings variables. The estimates were de-
response of this measure of criminal activ- rived from a large sample of arrestees for
ity to criminal justice and labor market whom matching earnings records could be
factors varies according to race as well. located. Sample members’ criminal histo-
Table V presents estimates for models of ries range from slight to severe.
all arrests disaggregated by race. In gen- In contrast to several previous studies,
308 ECONOMIC INQUIRY
TABLE V
Negative Binomial Estimates of Model of All Arrests, by Race
Variable White Black Hispanic
~~ ~
=
- - - -
PCNV = probability of conviction; AVGSEN = average prison sentence length; TOTjrIME
-
total prison time; PTIME86 = time in prison in 1986; QEMP86 quarters employed in 1986;
INC86 1986 income; DURATION duration of current jobless spell; BLACK 1 if black;
HISPANIC 1 if Hispanic; BORNGO = 1 if born in 1960.
I found consistent evidence indicating the These findings, I believe, call into question
importance of both criminal justice sanc- the economic rationality of a sanctioning
tions and labor market activity in deter- strategy based on increasingly lengthy
mining the individual's level of criminal prison terms as a means of reducing crime.
activity. The results point to large deter- Results on the effect of contemporane-
rent effects emanating from increased cer- ous employment on the individual's crim-
tainty of punishment, and much smaller, inal activities are mixed, but among
and generally insignificant effects, stem- blacks, at least, there is evidence that
ming from increased severity of sanction. greater employment is associated with
Additionally, I found evidence of a size- lower levels of arrest. Income is seen to
able criminogenic effect of imprisonment. have a strong negative effect, and longer
GROGCER: CERTAINTY VS. SEVERITY OF PUNISHMENT 309