Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Journalof Crrminal Jwtice. Vol. X, pp. l-26 (lY80). ~Kl~7-?357~XOiOllXH~I-?

6$O2 00/O
PergamonPress.Printed I” U.S.A Copyright Q IYXO PergamonPressLtd

DEMOGRAPHICALLY DISAGGREGATED PROJECTIONS


OF PRISON POPULATIONS

ALFRED BLUM STEIN, JACQUELINE COHEN. and HAROLD D. M ILLER

Urban Systems Institute


Carnegie-Mellon University
Pittsburgh. Pennsylvania 15213

ABSTRACT

The need for improved long-run projections of prison populations hus increased in recent years because of
record-high numbers of inmates and severe overcrowding in .stute und federal prisons, and becuuse of the growing
importance of changing demographic factors in influencing corrections populations. A model is developed for
projecting: general population demographics; demographic- and offense-specific arrest rates, imprisonment proh-
abilities, and times served; and then the size and composition of prison populations. Model purameters ure
estimated for Pennsylvania and are shown to be sensitive to demographic fuctors, particularly age and race.
Projections of future arrests, prison commitments, and prison population are derseloped for Pennsyh,ania using
projections of demographic changes in the state’s populution. Arrests ure expected to peak in 1980. prison
commitments are expected to peak in 198.5. and prison populutions are expected to peuk in 1990, with the
subsequent declines reflecting the maturution of the postwur baby boom children out of the highly crime-prone
ages and, somewhat luter, out of the highly prison-prone uges.

I. INTRODUCTION Since the choice of a strategy for dealing with


the overcrowding problem depends critically on
The 1970’s have witnessed unparalleled in- the degree to which the problem will continue
creases in the inmate populations of both state into the future. projections of future prison
and federal prisons. The total number of pris- populations are necessary for planning purposes.
oners in the United States increased by over 30 Mere trend analyses of prison populations are
percent from 1973 to 1977, reaching record-high clearly insufficient for this purpose, however,
levels of 253,816 in 1975 and 292,325 in 1977 since they cannot capture the fluctuations (some-
(Figure 1). (USDOJ. LEAA, 1976. 1978) These times large, as shown in Figure 1) that occur in
increases have caused overcrowding in many prison populations. Even sophisticated time-
prisons, forcing a number of states to resort to series techniques that try to capture these fluctua-
using temporary facilities to house the increased tions are not very satisfying since they do not
inmate populations. (Wilson, 1977) If the trend in explicitly take account of the underlying factors
prison population growth continues, many other influencing the growth and decline of prison
states will face inmate populations that exceed populations. This shortcoming not only reduces
their prisons’ current capacities. confidence in the projections but it also makes it
2 ALFRED BLUMSTEIN. JACQUELINE COHEN. and HAROLD D. MILLER

300.00(. I_

27o.wc )_

240.13OCI_

2 10 .OO(I_

180.00( 1,

150,00( 1,

120.000

/ zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
YO.000

lY25 1035 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA


104 5 lYS5 1Y65 1975

Figure 1. Number of sentenced prisoners in state and federal institutions at year end. lY25-lY77
Demographically Disaggregated Projections of Prison Populations 3

impossible to explore policy alternatives for studies have found that much of the increase in
reducing prison populations when they do reach crime over the past two decades has been due to
excessive levels. shifts in the demographic mix of the population
Similar criticisms apply to attempts to project (Sagi and Wellford, 1968; President’s Commis-
prison populations that are based on statistical sion, 1968; Ferdinand, 1970; Wellford, 1973;
relationships between prison receptions or prison Blumstein and Nagin, 1975; Fox, 1976). These
population and indicator variables such as unem- studies showed that the large number of children
ployment. (Robinson, Smith, and Wolf, 1977) born in the 1950’s_the postwar baby boom-
While the models may have good explanatory were moving into the high crime-prone ages (15-
power in terms of past data, estimating future 25 years) during the 1960’s and 1970’s, thus
values of the dependent variables (i.e., prison causing increases in crime. Since this large cohort
receptions or population) requires projections of will soon be aging out of the same age-range,
the independent variables, and that is likely to be demographers and other researchers are predict-
fully as difficult as projecting prison population ing decreases in crime in the near future (Toby,
itself. Even if an indicator variable such as 1977; Fox, 1976).” Use of demographically based
unemployment were reliable and could be pro- projections of crime in the system flow models is
jected accurately, it would still be of only very thus likely to result in more accurate projections
limited usefulness as a policy variable. of prison receptions and population.
On the other hand, the decisions made within There is an equally important shortcoming in
the criminal justice system clearly do affect prison the models themselves, however. Each of the
populations and are certainly relevant policy models tends to predict that court commitments
variables. A number of previous researchers have and prison population will increase or decrease
tried to represent the effects of the various with the number of crimes. But the pattern
components of the criminal justice system (e.g., displayed in the 1960’s was completely different:
police. courts) on each other, and. in particular, A steadily increasing number of crimes, accompa-
on the prisons. Blumstein and Larson’ (1969) nied by a generally decreasing prison population
represented the criminal justice system as a flow (see Figure 2) (USDOJ, LEAA, 1976, 1977, and
process from crime commission to corrections, 1978) that did not begin to increase until 1969.
with individual parameters representing the pro- Even though a demographically based projection
cessing decisions at various points in the system. of crime would have correctly predicted the
Their model simulated the changes in flows increase in crime in the 1960’s, each of the flow
through the criminal justice system and in the size models would have predicted a corresponding
of prison populations, and allowed the effects of increase in prison population over that period.4
policy changes to be estimated. Stollmack (1973) The phenomenon of decreasing prison popula-
developed an analytical model relating prison tions during a period of rapidly rising crime might
populations to commitment rates and time served also be attributed to the changing demographic
in prison and used it to project institutional composition of the population. Even though the
populations in the District of Columbia. A recent baby-boom cohort was moving into the high-
report prepared for the U.S. Congress by Abt crime ages in the 1960’s, and hence causing
Associates (1977) simulates the criminal justice substantially more crime, they were still juveniles
system using the methodologies of system dynam- and were not likely to go to prison for their
ics and uses the resulting model to project prison crimes, even if caught. Even when these individu-
populations in federal and state institutions.’ als became adults, they would typically be first-
In application of these models, the input time offenders (since records of juvenile offenses
projections have typically been based on assump- are usually sealed when the individual reaches
tions of constancy or linear extrapolation ‘of adulthood), and first offenders are rarely sent to
trends in crime rate. Such an approach is inade- prison. Only when a sizeable portion of the
quate, however. during a period of changes in the offenders in the cohort were old enough to have
demographic composition of the population since developed criminal records would there be any
such changes have been shown to be important significant increases in prison commitments and
determinants of changes in crime rates. Several in prison populations.
3 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ALFRED BLUMS’I‘EIN. JAC‘OUELINE COHEN. ad HAKOI>D I>. hflLLEK

Sentenced Prisoners Total Index of Crime.


300.000 in State and Federal U.S.
Institutions at Year End

350.000

200.000

1Y60 196.5 1970 1975

Figure 2. Crimes and prison populations in the U.S.. 1960-1977

Consequently, just as the changing age mix of We focus on estimating the effects of these
the population interacting with age-specific differ- demographic shifts through the use of a model of
ences in crime rates seems to have accounted for the criminal justice system. This model is similar
much of the increased crime in the past fifteen to the three identified earlier. but it accounts for
years, a similar interaction wtih age-specific the differences in crime rates and imprisonment
differences in imprisonment probabilities proba- probabilities across different demographic
bly accounts to a large extent for the pattern of groups. The model begins with a demographically
change in prison populations experienced in the disaggregated projection of the population in a
sixties and seventies. Thus, even if crime starts to jurisdiction and produces projections of arrests.
decline in the future because of the continued court cases, commitments to prison, and prison
aging of the postwar cohort. it might be expected populations. Long-term projections produced in
that prison populations would continue to in- this way should be more reliable than would be
crease, since the baby-boom cohort will remain in possible using existing techniques. and should
the somewhat older prison-prone ages after they also provide additional insights into the process
begin moving out of the high crime-prone ages. by which prison populations are generated.
Demographically Dlsaggregated Projections of Prison Populations 5

We illustrate the application of the model by developing long-term projections of prison popu-
estimating demographic projections and model lation, the early processing delays can reasonably
parameters for the state of Pennsylvania. Penn- be ignored but the time served in prison must be
sylvania’s prison population demonstrated the considered explicitly.
behavior typical of many other states. of the
federal system, and of the United States as a Projection of Prison Commitments
whole-i.e., decreases through most of the
1960’s followed by a sharp upturn in the 1970’s. The aggregate number of court commitments to
Since the Pennsylvania prison population is rap- prison in a given year for a particular jurisdiction
idly approaching the capacity of its existing can be represented as a product of flow rates and
institutions, the model’s projection of future branching ratios as follows:
prison population in Pennsylvania is not only
illustrative of the use of the model, but has
c, = N, . a, . c,’ ’ ~5 c: (!I
important policy implications as well. where: C, = the total number of commitments
(i.e., individuals convicted and sen-
tenced) to prison in year t;
II PROJECTION METHODOLOGY
N, = the size of the population of the
The social processes that produce prison popu- jurisdiction in year t;
lations are extremely complex and involve both
a, = the arrest rate for the jurisdiction in
the crime-committing behavior of the general
year t (i.e., the number of arrests
population and the response of the various parts
made in the jurisdiction in year t
of the criminal justice system to criminals. The
divided by N,);5
basic elements in that process can be represented
by a flow diagram like the one in Figure 3. The c,I = the probability of an arrestee in
“flow” is composed of the offenders who commit year t being indicted in the criminal
crimes and are processed by the criminal justice court in the jurisdiction;
system; the “stages” of the diagram represent the
processing of offenders by the different segments c;! = the probability of an individual in-
of the criminal justice system; and the “branches” dicted in the jurisdiction in year t
being convicted;
leading from one stage to another (or out of the
system) represent the decisions made in those and c;’ = zyxwvutsrqponmlkjihgfedcbaZYXWVUT
the probability of a convicted of-
stages about the offenders passing through them fender in the jurisdiction in year I
(Blumstein and Larson, 1969). being sentenced to prison.
There is a natural partition of the generation of
prison populations into two basic subprocesses: This expression is nothing more than a mathe-
1. The production of the flow of court commit- matical representation of the first five stages of
ments from the general population; and Figure 3: N, is the initial flow into the system, and
2. The generation stock of prison popu- a,, c:, c:, c;’represent the proportions
of the zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA of the flow at
lations from those committed to prison by each stage that continue to the succeeding stage.
the courts. The formulation in equation 1, of course,
Aside from the inherent differences in the con- aggregates across all demographic groups and
cepts of “flows” and “stocks,” this partition is types of offenses. To capture the effects of
also appropriate because of the nature of process- demographic shifts, this model can be applied
ing times in the system: even though offenders do separately to each demographic group, i.e., esti-
not pass through any of the stages instantane- mates of N,, a,, ci, <, c;’ can be developed for
ously, the processing delays in the early stages each subgroup and then equation 1 used to
(i.e.. the times from arrest to trial and to project C, for that group. The demographic
sentencing) are much shorter (on the order of disaggregation can be done by age, race, and sex;
weeks or months) than the time served by inmates it is also desirable to disaggregate further by
in prison (which is on the order of years). Thus, in offense type, since individuals committing differ-
ALFRED BLUMSTEIN. JACQUELINE COHEN. and HAROLD D. MILLER

w
in I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
1

z I I
Demographically Disaggregated Projections of Prison Populations 7

ent offenses are processed differently in the where P,_, = the prison population at the begin-
criminal justice system.6 Thus, we have ning of year t;
c lll,S”= N,,,X . %mo . c:mJ . d*m . C”,arso(2) x, = the number of P,_, remaining at
the end of year t; and
where G,, = the number of commitments to
prison in year t of individuals of T = the average time served in prison.’
age zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
a, race r, and sex s, and who
Assuming that prison receptions arrive accord-
committed offense o; ing to a Poisson process, and, again, that time
and the other variables are defined similarly. served is exponentially distributed, we have
Then, the total number of prison commitments (Stollmack, 1973: 145-47):
in year t (C,) can be calculated as the sum: Y, = C, * T. (l-e-“r) (8)
C, = c c c c CW1. (3) where Y, = the number of inmates admitted to
” , > D
prison during year t and still in
In this way, the effects of a changing demographic prison at the end of year t;
mix in the population or of a changing offense mix
C, = the number of court commitments to
among criminals can be projected.
Of course, equation 2 can also be used to prison (i.e., prison receptions) dur-
project future arrests (A,), indictments (D,), and ing the year (from equation 3); and
convictions (S,), viz., T = the average time served in prison.
A, = c c c c W I,,, . 4 m> Combining equations 7 and 8 gives the total
(1 , s 0
prison population at the end of year t:
P, = X, + Y, = C, . T . (1 -eml’r)
+ P,_, . em”T. (9)
(6)
As with equation 1, equation 9 can be disaggre-
gated by demographic group and offense type:
P ,m,n= Ctnrso . T,,x,, . (1 -em “h)
Projection of Prison Population
+ P, ,__,,u,(o . e~1~7u~~,. (10)
The generation of prison populations from
prison commitments can be viewed as a queueing where P,,,, = the number of individuals in prison
process: offenders are received in prison (enter the at the end of year t who were
queue), serve their sentences (wait in the queue), admitted at age a,“’ are of race r
and are then released (leave the queue). This and sex s, and who committed
queueing process transforms the “flow” of court offense 0;
commitments into the “stock” of prison popula- C ,iilT” = the number of court commitments,
tion by means of the time served in prison. At the as defined in equation 2; and
end of any year, the individuals in prison fall into
T“,E” = the average time served by indi-
one of two groups: those who were in prison at the
viduals who committed offense o,
beginning of the year and those who were received
are of race r, sex s, and were of age
during the year. It is convenient to model each of
a at commitment;
these groups separately, i.e., the reduction during
the year of the initial prison population, and the and the disaggregated estimates P_ can be
accumulation of new prison population from summed to calculate the total prison population
inmates received during the year.’ P,:
By assuming that the distribution of time served
P, = c c c c P,,,,, . (11)
in prison is exponential (Stollmack, 1973:145), 0 r s 0
the reduction of the initial population can be
Thus, with initial values for prison population
represented as follows: ’
and the projection of court commitments devel-
X = P,_,,- ‘J (7) oped using equation 2, equations 10 and 11 can be
use d to pro je c t p riso n po pula tio ns fo r a n! future tions were made to the year 2000. starting from
ye
. ’ll-
‘ s. 1970 census data on Pennsylvania population. All
of the data and projections were disaggregated by
age (O-14. 15-19, 20-24. 25-34, 35-44, 45-54.
III. PROJECTION OF POPUl.ATION and 55-t years). race (white and nonwhite). and
sex (male and female).
Changes in the sile and demographic composi- The projections of Pennsylvania population
tion of a jurisdiction’s population arc c a use d h> indicated that significant changes will occur in
three basic factors: births. deaths, and migration Pennsylvania’s population in the future. While
into a nd o ut o f the jurisdiction. While complete the number of whites in the state is projected to
precision is never possible in an) projection. the increase slightly (from 10.75 million in 1970 to
rclati\,e stability o\‘er time of the biological a nd 10.Y7 million in 1985) and then begin declining.
socioeconomic conditions that influence thrsc the nonwhite population is projected to continue
factors leads to an expectation of I-eaaonable increasing from 1 .OS million in 1Y70 to I.58
accuracy with the use of contemporary data. million in the year 2000, for a total increase of
Obviously. the larger the populaticjn being con- almost 60 percent. This racial difference is due to
sidered. the less likely that large or sudden the consistently higher (30-40 percent) birthrates
changes will occur in these basic factor\ a nd of nonwhites throughout the lY6O’s and into the
significantly affect the sire or composition of the 1970’s. These higher birth rates resulted in a
population-in the future. higher proportion of children and teenagers
The model used here for projecting future among nonwhites than among whites in the early
population incorporates each of those three fa t- seventies-in 1970. 33 percent of the nonwhite
tars and is b a se d o n the m o de l use d b y the population was under 20. while only 36 percent of
Pennsylvania Department of Education in its the white population was under 20. Hence. in the
population projections (PA Dept. of Education. 1980’s. a higher proportion of the nonwhite
1Y75: also, Barclay. lY5X: Co?;. 1Y70). The model population will be in the childbearing ages, and
is initialized with census population data disaggre- their relatively higher birth rates will cause the
gatcd by age. race. and sex. Age-, race-. a nd total nonwhite population to continue to increase
sex-specitic mortality rates and net migration in size.
rates are then applied to each subgroup to Most age groups (for both whites and non-
calculate the number of deaths and migrations in whites) are projected to show a “peaking” phe-
each year. The number of births is calculated as nomenon, i.e.. they will increase in size for
the product of the number of women in the several years. reach a maximum size, and then
childbearing ages (15-U) in each race with begin declining. The peak years are later for the
race-specific fertility rates (i.e.. the average num - older age groups, and the nonwhite age groups
ber of children per woman per year) in each year. generally peak five years later than their white
Each demographic subgroup is then “aged” into counterparts. Thus. the baby boom shows up as a
an older subgroup and the calculation is repeated. “hump” passing through the population: as the
large number of children born in the 19.50’s passes
through each age group. that group increases in
size, reaches a peak size when the last of the
Population projections were nude for Pennsyl- “baby boom” births are in that age group. and
vania using this model together with historical then declines in size as the large cohort moves on
data on births, deaths. and estimated migrations to the next age group.
in the state. The birth data confirmed that the The peaking phenomenon is displayed in Fig-
high birth rates of the 1950’s had continued to ure 4 for the two most crime-prone age groups.
decline throughout the 1960’s in Pennsylvania 15-lY years and 20-24 years. As shown in Figure
(decreasing from an average of 4.1 children per 4, the white 15-to-19-year-old group (which ac-
nonwhite woman in 1960 to 2.4 children in lY76. counted for 28 percent of arrests in Pennsylvania
and from 3.0 children per white woman in lY60 to in 1076) peaks in 1975. the white 20-to-24year-
1.7 children in lY76).” The population projec- old and nonwhite lS-to-lY-year-old groups (which
Demographically Disaggrcgated Projections of Prison Populations

Nonwhites
Ic-- <.‘....* 20-24 year olds
,*..**** ..* \ .‘**....,
\N ***.... ***a.* . . . . . .;s._.rr.s
15-19 year olds--Z ----.--

lY70 197s 1980 1985 1990 19% 2000

Figure 4. Population projections of crime-prone age groups in Pennsylvania.

together accounted for 2.5 percent of Pennsylva- IV. PROJECTION OF ARREST RATES AND
nia arrests in 1976) peak in 1980, and the ARRESTS
nonwhite 20-to-24-year-old group (which ac-
counted for 6 percent of all Pennsylvania arrests Arrest marks the first stage in an individual’s
in 1976) peaks in 1985. Thus, from 1970 through processing by the criminal justice system. The
1975, all of these crime-prone groups were “arrest rate” in year t (u,,,~,~,)for a given offense
increasing in size. Aside from any changes in and demographic group is calculated as the
crime or arrest rates, such demographic changes number of arrests for that offense of individuals
alone would have caused increased crime and in that demographic group in year t, divided by
arrests over the 1970-1975 period. Since these the total number of individuals in that demo-
same groups are projected to start decreasing graphic group in year t. The arrest rates for a
again by 1980, a reverse demographic effect can particular demographic group are determined by
be expected to reduce crime and arrests. four factors:
10 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ALFRED BLUMSTEIN. JACQUELINE COHEN. and HAROLD D. MILLER

1. The proportion of individuals in the group vided by the Pennsylvania State Police. which
who commit crimes; began a statewide UCR program in lY74. The
2. The frequency with which those individuals data were disaggregated into the age. race, and
commit crimes; sex categories described in the previous section.
3. The types of crimes they commit; zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFE
a nd as well as twelve crime groups (murder and
4. The likelihood that the police arrest some- voluntary manslaughter; robbery; burglary;
one in the group (falsely or properly) in drugs; larceny. auto theft and stolen property;
response to a particular crime. aggravated assault; simple assaults. fraud, and
Thus. arrest rates could change over time because weapons; arson and forgery; rape; sex offenses;
of changes in any of these factors. drunk driving; and all others),” making for a total
While demographic-specific arrest zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
rates for a of 7 x 2 x 2 x 12 = 336 separate subgroups.
jurisdiction are not usually published, the FBI’s These arrest data were then combined with the
Uniform Crime Reporting (UCR) program col- population estimates developed for 1Y70- lY76
lects data on the offense type, age. race, and sex of (see Section III) to calculate arrest rates. The
arrestees, allowing population data to be used to arrest rates (Table 1) showed the anticipated
calculate arrest rates. Unfortunately, the UCR differences between demographic groups-non-
data do not provide a complete demographic whites had higher arrest rates than whites. men
breakdown of arrests. Standard UCR reporting had higher arrest rates than women. and 15-21
forms provide offense by sex by age, and offense year olds had higher arrest rates than other age
by sex by race breakdowns. but not the complete groups.
partition--offense by sex by age by race (e.g., the To estimate the complete set of demographic
data show how many 15year-old male robbery arrest rates. the iterative proportional fitting
arrestees and how many white male robbery technique was applied to the data using two
arrestees there are in a given year, but not how alternative assumptions about the age-race
many 15year-old white male robbery arrestees interaction:
there are). Since future age changes in the 1, An assumption of no interaction in numbers
population may be very different for different of arrests (i.e., the ratio of white to non-
races (as demonstrated for Pennsylvania in the white arrests is the same in each age group):
previous section), it is desirable to estimate the 2. An assumption of no interaction in arrest
complete breakdown from the data available. rates (i.e., the ratio of white to nonwhite
This can be done using the technique of arrest rates is the same in each age group).
“iterative proportional fitting” (Bishop, Fien- The two estimations were compared by combin-
berg. and Holland. 1975). This technique esti- ing (using equation 4) the 1976 arrest rates
mates the complete demographic breakdown of generated by each assumption with the popula-
arrests in such a way that the marginal values of tion projections from Section III. Since the
the estimated arrests (i.e., the offense by sex by resulting total projections of future arrests dif-
age and offense by sex by race totals) equal the fered very little,” the arrest rates generated using
corresponding figures provided in the UCR data. the second assumption were used in subsequent
The calculation requires an assumption about the analyses.
interaction between race and age by specifying Examination of the arrest rates revealed that
the differences in the age-specific ratios of racial substantial changes had taken place between 1970
arrests (i.e., the ratios of white to nonwhite and 1976. The overall arrest rate increased by
arrests in each age group) (Bishop, Fienberg. and almost 60 percent, with larger percentage in-
Holland, 1975). creases for women. for whites, and for teenagers.
These changes were not always the result of a
Arrest Rate Estimation for Pennsylvania smooth trend, however. Over half the increase in
the total arrest rate occurred between lY73 and
Arrest data were gathered for Pennsylvania for 1974. The same pattern-an unusually high in-
the years 1970-1976; the 1970-1973 data were crease between 1973 and 1974-generally held in
supplied by the Uniform Crime Reporting section the subgroups that showed large increases over
of the FBI, and the 1974-1976 data were pro- the seven-year period.”
Demographically Disaggregated Projections of Prison Populations zyxwvutsrqponmlkjihgfedcbaZY
11

TABLE 1

PENNSYLVANIA ARREST RATES, 1970-1976

Arrest Rates per 1000 Population


Percentage
Change
Subgroup 1970 1971 1972 1973 1974 1975 1976 1970-1976

Total 20.95 21.55 23.63 23.74 30.76 33.31 33.16 58

Male 38.36 39.37 43.26 43.25 55.78 60.25 59.65 56


Female 4.85 5.11 5.54 5.80 7.79 8.60 8.89 83

White 13.39 12.97 15.76 15.08 22.60 24.78 25.08 87


Nonwhite 98.40 108.22 102.09 109.10 110.31 115.61 110.14 12

O-14 years 9.51 10.90 13.04 13.81 18.62 18.57 17.15 80


15-19 years 71.08 73.56 88.55 94.44 135.30 147.28 147.13 107
20-24 years 48.58 50.38 49.78 46.61 59.51 65.64 67.18 38
25-34 years 26.32 25.81 22.40 21.77 26.19 31.87 34.10 30
35-44 years 19.17 18.66 19.78 18.49 20.75 22.24 22.16 16
45-54 years 14.06 13.97 14.54 13.30 14.07 14.38 14.19 1
55-t years 6.78 6.16 7.29 6.49 6.52 5.84 5.26 -22

Murder 0.059 0.062 0.063 0.060 0.066 0.068 0.064 8


Rape 0.072 0.066 0.099 0.091 0.117 0.127 0.127 76
Robbery 0.389 0.446 0.499 0.457 0.560 0.681 0.587 51
Burglary 1.165 1.010 1.235 1.240 1.754 1.938 1.731 49
Agg. Assault 0.438 0.436 0.444 0.491 .0686 0.745 0.745 70

Larceny, Auto
Theft, Stolen
Property 2.475 2.368 2.260 2.403 3.886 4.601 4.731 91

Thus, it appeared that a shift had occurred arrest rates were constrained to lie within 30
between 1973 and 1974 in many subgroups; the percent of the 1976 values,” to preclude unrea-
arrest rates in those subgroups seemed either to sonable extrapolations. zyxwvutsrqponmlkjihgfedcbaZYXW
continue an old trend after 1974 but at a higher
level,” or to establish a new trend after 1974. To Arrest Projections for Pennsy lvania
project the future values of arrest rates, a
heuristic rule was chosen to distinguish those Using equation 4 to combine the projected
demographic and offense subgroups in which such arrest rates with the population projections for
a shift had occurred.‘h In those subgroups where it Pennsylvania results in the projection of total
was determined that a shift had occurred, the arrests shown in Figure 5. (As an indication of the
1974-1976 arrest rates were linearly extrapolated separate effects of the changing composition of
to the year 2000, and in the remaining cases, the the population and of the trends in arrest rates,
arrest rates for all seven years were used in the Figure 5 also contains-as the dotted curve-a
extrapolation. Because the projections in each projection of total arrests in which arrest rates are
subgroup were based on so few points, the future held constant at the averages of the 1974-76
12 AI.FKED BLUMSTEIIU. JACQUELINE COHEN. and 1IAKOL.D II. hflLl.EK

_--- Projection Using Arrest Rate Trends

........ .. Projection Using Constant 1074- lY76


Average Arrest Rates

1970 1Y75 1Y80 1Y8.5 IYYO 1YY5 2000

Figure 5. Projections of total arrests in Pennsylvania. 1970-2000.

values.) As is apparent from the figure. total


arrests arc projected to increase until 1980 (rcach-
ing :I level 15 percent higher than in 1976) and
then to decline. This peaking phenomenon is due
to the peaking of the crime-prone subgroup< in
the pclpulation. as described earlier.
Table 3 shows the distribution of arrests over
time for several of the crime types and age
groups. and also the year in which the peak
number of arrests is projected to occur in each
category. Several crime types. such as aggravated
assault. rape. and drunk driving, peak later than
IWO
Dcmopraphically Dxqgreyated Prelection\ of Prison Populations 13

TABLE 2

DISAGGREGA-IECI~ANGES IN PFNNSYLVANIAARRESTS, 10752000

Percentage of Total Arrests in Year in Which


Maximum Number
1975 1YX5 2000 of Arrests Occur

Murder 0.2 0.2 2000


Burglary 5.8 4.6 4.4 1980
Drugs 4.5 4.1 3.x 19x3
Drunk Driving 4.3 5.4 5.x 1990
Rape 0.3 0.4 0.3 19YO
Aggravated
Assault 2.2 2.5 2.6 lY90

O-14 years 13.1 x.2 9.0 lY74


15-19 years 41.4 34.5 34.7 1980
20-24 years 17.2 19.1 14.4 1Y80
25-34 years 12.2 22.4 1X.4 1YYO
35-44 years 6.8 8.6 13.0 2000
Whites 6X.0 65.5 62.0 1980
Males 86.5 85.X X6.1 1YXO

groups in 19X.5. While the same type of phenome- important variables in the model, since they
non occurs with the white population, it is not reflect the discretion exercised by magistrates,
strong enough to reverse the downward trend in prosecutors, and judges. By the same token.
white arrests caused by the aging of the baby- however, these are perhaps the most difficult
boom cohort. The different white and nonwhite parameters to project into the future, since it is
effects cause total arrests to begin increasing difficult to predict when a new “policy” (e.g., a
slightly in 1995 (Figure 5) and cause the nonwhite toughening of sentences) might be introduced or
proportion of total arrests to increase from 32 how it might be implemented. Of particular
percent in 1976 to 38 percent in 2000. importance is how the parameters may change as
the system adapts to increased caseloads in the
courts or to overcrowding in the prisons.
V. COURT PROCESSING AND FUTURE
PRISON COMMITMENTS Estimation of Court Parameters for Pennsylvania

The processing of arrestees by the judicial Data for Pennsylvania were obtained from the
system is represented by three separate parame- Criminal Justice Statistics Division of the Penn-
ters: sylvania Governor’s Justice Commission. The
1. The probability of an arrest resulting in a data consisted of records of all cases disposed of
criminal court case being filed (c’); in Pennsylvania Criminal Courts Ix from 1970 to
2. The probability of a criminal court case 1975. Each record described the processing of a
resulting in a conviction (c’): and case by the judicial system: whether the case was
3. The probability of a convicted offender dismissed or diverted prior to trial; whether the
being sentenced to prison (c’). defendant pleaded guilty or went to trial; whether
In a policy sense, these are likely to be the most a conviction or acquittal was obtained; and
ALFRED BLUMSI‘EIN. JACOLIELINE C‘OllEN. and 11AROLD 1). ~lll.l.tR

1970 lY75 lYX0 IYXS IYYO lYY5 3( ItIt)

Figure 6. Projected arrests in Pennsylvania. 1Y70-2000 (by race).

finally, the sentence imposed on convicted of- could not be calculated directly from this same
fenders. The age. race, and sex of the individuals data. While some aggregate data on the process-
were recorded, as well as the offense charged. ing of arrestees prior to indictment was available.
These records were processed to calculate three no demographic-specific data could be obtained.
sets of figures: Thus, it was necessary to estimate c),,,, as the ratio
1. Thetotal number of cases disposed of. M),,,,,, of the number of cases disposed of (M I,,,,) in year
in each year t, from age group rr, race zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
r. and t, in each subgroup, to the number of arrests in
crime type 0;“’ that subgroup in year f. This method is not exact
2. The total number of defendants convicted. since M ’ is the number of cases disposed of (i.e..
@,,,, in each year 1, from age group I(. race completed, either through dismissal, acquittal, or
r, and crime type o; and conviction) in a given year, not the number of
3. The total number of defendants sentenced cases filed in that year,” although it is a good
to prison .“I IV:,,,,,. in each year t, from age approximation if the backlog of cases is stable.
group a, race r, and crime type o. Examination of aggregate court processing statis-
The parameter -cz,,,, was then estimated
. .
by tics indicated that, while the courts had an
M~~,JW,,,~ and CL,, was estimated by zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
M ;,,,,,lM y c,,,, increasing backlog from 1970 to 1972, the backlog
for all t, o, a, and r. was reduced in 1973 and seemed to stabilize in
Unfortunately. the data did not include any 1974 and 1975. Also, the numbers of filings and
record of the processing of arrestees for whom dispositions were very close in 1974 and lY7S.
court cases were not filed: hence the parameter c’ Thus, the average of the 1974 and 1975 ratios
DemographicallyDisaggregatedProjections of Prison Populations 15

TABLE 3

ESTIM ATED PROBABILITIESOF COURT CASE FILING GIVEN ARREST IN PENNSYLVANIA

Subgroup Arrests in 1975 Percentage Reaching Court (c’)

O-14 years 52,404 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK


0
15-19 years 166,059 7
20-24 years 69,025 40
25-34 years 48,775 51
35-44 years 27,255 38
45-54 years 21,266 33
55+ years 16,399 26

Murder 813 100


Rape 1,375 42
Robbery 8,196 44
Burglary 23,340 34
Aggravated Assault 8,971 38

Robbery: O-14 years 817 0


15-19 years 3,706 11
20-24 years 2,048 81
25-34 years 1,254 100
35-44 years 282 61
45-54 years 74 78
55+ years 16 56

of dispositions to arrests (in each crime type and Both of the parameters c? and ti showed strong
demographic group) was chosen to represent the trends between 1970 and 1975. As shown in Table
value of c1 (in that subgroup) for the whole period 4, the probability of a court case resulting in a
1970-1975. These values of c1 were then assumed conviction (c’) decreased from about 60 percent
to remain constant through the year 2000. in 1970 and 1971 to 43 percent in 1974 and 1975.
As shown in Table 3, the estimated values of c1 The reason for this large decline was the introduc-
show the anticipated age effect. Arrestees under tion of the Accelerated Rehabilitative Disposition
15 were never processed by the adult criminal (ARD) program in Pennsylvania in 1972. ARD
court system, and only 7 percent of the permits defendants to be placed in a probationary
15-to-19-year-old group were processed by the status without an adjudication of guilt; the
courts, since most of the arrestees in this group charges are then dropped if the defendant suc-
were also juveniles (under 18). Because those two cessfully completes the program.
age groups so predominate in arrests, less than As shown in Table 5, the probability of a
half of arrestees in offenses such as robbery and conviction resulting in a prison sentence (c”)
burglary are processed by the adult courts. As nearly doubled between 1970 and 1975, increas-
arrestees become older in the future, an increas- ing from 5.5 percent to 10 percent. These trends
ing zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
proportion of total arrests will be processed by in c2 and c3 raise concern about the future values
the criminal courts. Thus, a future decrease in of the parameters, i.e., whether the trends will
total arrests will not necessarily lead to a decrease continue, stabilize, or reverse themselves. How-
in the caseload of the criminal courts. ever, when the product of the two parameters is
16 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ALFRED BLUMSTEIN. JACQUELINE COHEN. and HAROLD D. MILLER

TABLE 4

PROBABILITIES OF CONVICTION GIVEN INDICTMENT (k’), 1970-1975

1970 1971 1972 1973 1974 1975

Number of Dispositions 58.332 56,458 72,138 76,102 83,049 85,409


Percentage Convicted (2) 59 60 54 48 43 43

TABLE 5

PROBABILITIES OF PRISON GIVEN CONVIVTION (c’). 1970-1975

1970 1971 1972 1973 1974 1975

Number of Convictions 34.406 34.07 I 38,964 36,472 35.616 36,525


Percentage Committed
to Prison (c’) 5.5 5.6 7.1 x.2 8.1 10.0

TABLE 6

PROBABILITIES OF PRISON GIVEN INDICTMENT (c’ x c’), 1970-1975

1970 1971 1972 1973 1974 1975

Number of Dispositions 58.332 56.458 72,138 76,102 83.049 85.409


Percentage Committed
to Prison (c’ X c’) 3.2 3.4 3.9 3.9 3.5 4.3

examined, i.e., the probability of a prison sen- the 1975 value in the future in each crime type
tence in any court case filing, that probability and demographic group.
does appear to be highly stable, as shown in Table As shown in Table 7. the combined parameter
6.22 This indicates that the increase in the propor- 2 . c3 shows only a small increase with age
tion of convicted offenders receiving prison sen- between the ages of 15 and 35 and then decreases
tences (c”) was not due to judges becoming considerably after the age of 35. This effect is
“tougher,” but simply due to the early diversion partially due to a lower probability of conviction
to ARD of the defendants who would otherwise (c’) after age 35. but is primarily a result of a
have been convicted but would not have received much lower probability of prison given conviction
prison sentences. Thus, in the projections. it was (c’) for older offenders. The probability of prison
assumed that the product 2 . c’ would remain at is low even in the most serious crime types; as
Demographically Disaggregated Projections of Prison Populations zyxwvutsrqponmlkjihgfedcbaZYX
17

TABLE 7

PROBABILITIES OF CONVICTION AND IMPRISONMENTAFTER INDICTMENT (1975 VALUES)

Probability of Prison Probability of Conviction Probability of Prison


Given Indictment Given Indictment Given Conviction
Subgroup (c’ x c’) (2) (3)

15-19 years 0.044 0.452 0.097


20-24 years 0.049 0.462 0.105
25-34 years 0.052 0.429 0.121
35-44 years 0.027 0.376 0.072
45-54 years 0.015 0.369 0.041
55+ years 0.014 0.358 0.039

Murder 0.46 0.65 0.71


Rape 0.21 0.38 0.54
Robbery 0.19 0.53 0.36
Burglary 0.09 0.51 0.17
Aggravated
Assault 0.05 0.39 0.13

Burglary:
15-19 years 0.088 0.558 0.157
20-24 years 0.088 0.526 0.167
25-34 years 0.090 0.477 0.189
35-44 years 0.079 0.437 0.181
45-54 years 0.045 0.348 0.129
55+ years 0.082 0.328 0.249

shown in Table 7, less than half of the murder however, the peak occurs in 1985, which is five
defendants are sentenced to prison, and less than years later than the peak in arrests. This shift
one-fourth of rape and robbery defendants re- results from the strong differences in the parame-
ceive prison sentences. This reflects the combined ter c’ between age groups-as the average age of
effects of low conviction probabilities and low arrestees increases, the overall proportion of
prison probabilities given conviction. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
arrestees being indicted, convicted, and sen-
tenced to prison also increases. Thus, even
though total arrests begin to decline in 1980, the
Projections of Court Commitments for
number of court commitments to prison will
Pennsylvania
continue to increase until 1985, when arrests in
the prison-prone ages (20-34) begin to decline.
Using equation 3 to combine the estimated
values of c’ and the 1975 values of the product c?
c’ with the arrest projections described previously
yields the projection of future court commitments VI. PRISON POPULATION PROJECTION
shown in Figure 7. The projection of commit-
ments shows the same peaking phenomenon as The final parameter to be considered is the
was displayed in projected arrests; in this case, average time served in prison. The time that an
18 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ALFRED HLUMSTEIN. JACQUELINE COHEN. and HAROLD D. MILLER

6,000

1970 197s 1980 1985 1990 1995 2000

Figure 7. Projection of court commitments to prison in Pennsylvania, 1970-2000.

inmate spends in prison depends on two primary remain constant, the parole board might respond
factors: to increasing prison populations by reducing time
1. The sentence length imposed by the commit- served. The nature of these responses will not be
ting court; immediately clear from contemporary data since
2. The decisions of the parole board as to the the time served by inmates currently being
time of release (which, typically, must be released reflects sentencing decisions made sev-
after the inmate has served the minimum eral years earlier.
sentence, but no later than the expiration of
the maximum sentence).” Time Served in Pennsylvania
Because these two factors do not act indepen-
dently, projection of future time served is diffi- The Pennsylvania Bureau of Correction pro-
cult: if sentence lengths increase, inmates might vided data on all offenders who were committed
be paroled sooner; even if sentence lengths to prison by criminal courts and who were
Demographically Disaggregated Projections of Prison Populations 19 zyxwvuts

TABLE 8

AGE DIFFERENCES IN TIME SERVED IN PENNSYLVANIA

Crime Type Age Group Average Time Served (years)

All Offenses 15-19 years 2.2


20-24 years 2.1
25-34 years 2.2
35-44 years 2.5
45-54 years 2.4
55+ years 1.8

Robbery 15-19 years 2.5


20-24 years 2.6
25-34 years 2.6
35-44 years 2.7
45-54 years 3.4
55+ years 3.4

Rape 15-19 years 2.7


20-24 years 3.4
25-34 years 3.2
35-44 years 2.6
45-54 years 2.3
55+ years 3.1

received in or released from state prisons between Projection of Future Prison Population in
1971 and 1976. The data were in the form of Pennsylvania
individual records and indicated each offender’s
sentence, date of reception in prison, and date of Using equation 10 to combine the time-served
release from prison, as well as the age (at estimates with the projection of court commit-
reception), race, sex, and offense of the individ- ments developed in Section V results in the
ual. The offender’s time served in prison was projection of prison population shown in Figure
calculated as the difference between the reception 8.” This projection has features similar to the
and release dates.IJ projections of arrests and court commitments,
The data indicated that there had been a slight viz., prison population is expected to continue
increasing trend in the average minimum sen- increasing, reach a peak, and then decline. The
tence of offenders between 1971 and 1975, peak in prison population, however, occurs in
although average minimum sentences for murder 1990, five years later than the peak in court
and rape showed decreases from 1974 to 1975. No commitments. The peak is delayed for several
such trends were apparent in time served, how- reasons:
ever. Time served for most offenses seemed to be 1. It takes time for prison population to build
fluctuating about a stable mean, even within up from court commitments, so that the
individual demographic subgroups. Thus, it was large number of offenders committed during
assumed that time served would remain at the the peak commitment years will remain in
average of the 1975 and 1976 values in the future. prison after commitments begin to decline.
There were no consistent age patterns in time 2. As can be seen from Figure 7, the rate of
served, as is apparent from Table 8. decrease of court commitments is much
20 ALFKED BLUMS'I‘EIN..lACOUEI.INE (‘OHEN. ;~ntl
IiAKOLL) [).h1ll.l
CIK

1977 Total Capacity

1970 1975 lY80 10x5 1YYO I YY5 2000

Figure 8. Projected prison population in Pennsylvania. lY70-3000.

slower near the peak than was the rate of of prison population than of court commit-
increase prior to the peak (by a factor of ments, thus delaying the peak.
about five). Thus, there will be more recep- Since no actual prison population data were
tions than releases for several years after used in developing the prison population projec-
198.5, causing prison population to continue tions. a comparison of the actual and estimated
increasing. values of prison population between 1970 and
3. Commitments for many of the more serious 1977 serves as a useful validation of the parameter
offenses peak somewhat later than do com- estimations. Figure 9 shows the actual December
mitments as a whole. For example, commit- 31 population in Pennsylvania state prisons from
ments for murder peak in lYY4, and commit- 1970 to 1977.” as well as the population values
ments for aggravated assault peak in 1YYO. produced by the projection model for those years.
Since these offenses have relatively longer As is apparent, the estimated values match the
times served, they form a larger proportion actual data very closely. and the correlation
Demographically D~saggregatcd Projections of Prison Populations zyxwvutsrqponmlkjihgfedcbaZ
21

Inm a te s

I_
7O O U

6500

6000 \

I_

c
5000 Estim a te d

4500

4000

3500

I I I I I I I
1070 lY71 1Y72 1973 lY74 lY7S 1Y76 1Y77

Fig ure 9. Ac tua l a nd e stim a te d p riso n po pula tio n in Pe nnsylva nia . 1Y70- lY77.
22 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ALFRED BLUMSTEIN, JACQUELINE COHEN. and HAROLD D. MILLER

Inmates

Nonwhite
*_*..... l *** . . . . . . . . *. . . . . . . .
.-
,..A -
/ M.a. -.
\
/ ,.= \
\
/
/ .=* .I
, .a* N +White
\
/ .=*
, .**
, /
/ *:**
..
. . .’

1Y70 1975 IYXO IYXS zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK


1YYO lYY5 7000

Figure 10. Projected prison populations in Pennsylvania. 1970-2000 (by race)

between the two sets of data is 0.888. The largest (Figure 10). This is due to the continued increase
error occurs in 1970;” eliminating the 1970 values in arrests projected for nonwhites (see Section
gives a correlation coefficient of 0.9Y4 for the IV). This difference will cause the nonwhite
remaining points. proportion of prison population to increase from
The projections indicate that the number of 48 percent in 1976 to 55 percent in 2000, as shown
white inmates will peak in 19X6. while the in Table 9. Table Y also indicates that the
nonwhite inmate population will peak in 19Y6 proportion of inmates under the age of 25 will
Demographically Disaggregated Projections of Prison Populations 23 zyxwvuts

TABLE 9

DISAGGREGATE CHANGES IN PRISON POPULATIONS. 1975-2000

Percentage Percentage Percentage Year of


of Total of Total of Total Maximum
Subgroup Inmates in 1975 Inmates in 1990 Inmates in 2000 Population

White 52 49 45 1986
Nonwhite 48 51 55 1996

15-19 years 13 7 7 1980


20-24 years 35 25 23 1984
25-34 years 36 54 49 1991
35-44 years 10 10 15 2000
45-54 years 4 2 3 2000

Murder 23 21 23 1998
Rape 5 6 6 1990
Robbery 21 20 20 1989
Burglary 17 14 13 1986
Aggravated
Assault 3 3 1990

decrease from 48 percent to 30 percent by the aging of the genera1 population. Since the baby
year 2000, and that the mix of offenses among boom was a nationwide phenomenon, similar
inmates will change very little over the twenty- results should hold for many other states. These
five-year period. findings should inhibit some of the pressures to
The projection indicates that prison population build more prisons to house the currently increas-
will exceed the stated current total capacity of the ing prison populations. By the time new prisons
prison system” in 1979, to peak at over 10,200 are planned, designed, and constructed, much of
inmates (25 percent over the 1977 capacity), and the need for the extra capacity is likely to have
will remain above current capacity until after the disappeared.
end of the century. As prisons fill up or become
overcrowded, however, adaptation in the crimi-
nal justice system (through increased plea bar- VII. SUMMARY
gaining, decreased severity of sentences, earlier
release on parole, etc.) is likely to occur to some The demographically disaggregated projection
extent. This would be reflected in decreases in the methodology described here can provide useful
parameters of the model presented here (i.e., c’. short and long-range projections of future arrests,
c2, c3, T) and could tend to reduce prison court workloads, and prison populations for use
population below the levels projected using con- in planning for the criminal justice system. It was
stant parameter values. shown in the case of Pennsylvania that arrest rates
An important implication of the prison popula- and imprisonment probabilities vary considerably
tion projection developed here. then, is that even across demographic groups. It was also shown
if no adaptation in the criminal justice system that significant changes will occur in the demo-
takes place, prison populations in Pennsylvania graphic composition of the general population in
should ultimately start to decrease because of the the future. Thus, projections developed in the
2-I ALFRED BLUMSIEIN, IAC‘OUELINE C‘OllEN. ,md HAKOLD I) MILLER

fashion described here should provide a signifi- ’ Three other projectron zyxwvutsrqponmlkjihgf
model5 were a la o use d to prolect
cant increase in accuracy over models that do not prison population in the Abt report: these model5 altcr-
nately a55umed that the total prison population would
account for the changing composition of the remain constant, that the total numbers of admissions and
population or for demographic differences in releases would remain constant. and that the number of
processing by the criminal justice system. admissions and average time served would remain constant.

The model can also serve as an important tool ’ See. e.g.. Toby (lY77). Fox (lY7h) proJects that the violent
crime rate will decrease in the lY)xO’s but begin increasrng
in policy analysis for the criminal justice system.
again in the 1990‘s: he projects that the property crime rate
By changing appropriate parameters in the mod- will continue to increase. His prolectrons of crime ratca.
el. simulated experiments can be performed to however. also depend on a project&n of the consumer price
index and hcncc are not based wlcly on demographic
estimate the effects of various policy alternatives
changes.
(such as mandatory sentencing laws. pretrial
’ In validations using actual crirncs and other paramctcrs for
diversion programs. etc.) on the workloads in the the federal prison system and for the Iowa pri5ons. the Abt
criminal justice system. The model can provide Associate5 model failed to show the dccrcascs In prison
estimates of both the current and future effects of population actually experienced during the mid-lY6O‘s
(Abt. iY77: Tech. App.. 70-73)
a policy change.
~The etfects of individual crimrnal propensity and of police
An important area for future research is investi-
arrest effectiveness are reflected in the single parameter u.
gating how the processing of offenders by the Thus, there is no attempt to scparote any trends in
criminal justice system (and hence the parameters crime-committing behavior from changes in the willingness
or ability of the police to make arrest\.
of the model) changes in response to increases or
decreases in the workload at various points in the ’ Obviously. additional disaggrcgation can bc done to an!
level desirable and feasible. For example, urban area< tend
system. For example, if prisons become o\er- to have higher crime rates but lower imprisonment proh-
crowded, to what extent will parole boards abilities than rural area5; if a partrculnr jurisdiction has
urban and rural regrons with distinct police and court
release inmates sooner, judges sentence fcwcr
systems. the parameters should also be disaggregated
offenders to prison, or prosecutors divert more geographically in order to project the etfect5 of urban-rural
defendants prior to trial’? Conversely. does the migration or of differential growth rates.

existence of unused capacity in the prison system The time Interval used need not be a year, but can he longer
increase an offender’s chances of being sentenced or shorter depending on data availability

to prison or the time to be served? These ’ The assumption that time served is distrrbuted exponcn-
tially is fairly common (XC. c.g.. Av-Itzhak and Shinnar,
questions about “adaptation” and “feedback” 197.1: Stollmack, lY73). Although there has been little
within the criminal justice system are only begin- research done to te5t the validity of this assumption.
ning to be explored. preliminary Investigations by the authors into data from
Pennsylvania prisons have shown time served to follow an
exponential assumption quite closely. Also, a Rand study ol
California prironers found that the drstribution of time
ACKNOWLEDGEMENTS served can be represented by mixtures of exponential and
translated exponentral diarributions: this is not necessaril>
Thi5 study was partially support& by Grant No. S.P.C‘. inconsistent with the awunption used here since different
#56878X from the Pennsvlvania Governor‘s Justice <‘om- values ot time served are used for dilferent crime types and
mission. The authors would like to exprc\\ thclr apprecw demographic groups (Chaiken. 1Y77).
tion to the numerous individual5 u ho pro\ i&d awstancc As presented, the model u5cs a constant value of time
during the course 01 the project. espcciall) PhIHIp J, served in each offense tvpc and demographic subgroup.
Rennmger, Director of the Criminal Justice Statistics The model could be modrhed to account for changing time
Division of the Governor’5 JustIce C‘ommlssion: John H served: however, the type of modification needed depends
Mease. Chief of the Criminal Analvsl\ Unit of the C‘rimtnal on the cause of the changing time served. If the change IS
Justice Statistics Diwqion: and Jc;hn Senwr. Educational due to a change in parole behawor (i.e.. releasmg individu-
Research Associate with the Pennsylvania Department 01 als from their sentences earlier or later), then the same
Education. The authors would also hke to thank David functional form of the model can he used. hut a different
Arnett. Albert Brooks. Deborah Kahn. Darryl White. and value of time served used in each year. However, if the
Gary Wigmore for assistance in data collcctlon and anal~w changing tlmc served is due to changing sentence length.
during the course of the research. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
then the changes are properly associated with each yearly
cohort of rtwy~r~on.s. not of releases. This would require
having a separate equatron (5imllar to cquatrons 7 and 8) to
N OT ES
represent the growth and decline of the prison population
subset generated by each such cohort, and would add
’ This model ha5 heen tran5lated Into an Interactlvc Nov.
considerable complexity to the calculations.
simulation (Belkin. Blumateln. and Glass, lY71) that allow
policy analysts to te5t the effects of \*arous parametric The aging of prisoner5 over time while lncarceratcd i\
changes on workload5 in the crlmlnal justlcc s!\tcm. ignored.
Demographically Disaggrcgatcd Projections of Prison Population5 25

” It was assumed that future birth rates would remain at the tie\) and hence the size of the jail population is largely a
1976 levels in each race. Since most of the individuals who function of court case inventories and bail decisions.
will affect prison population over the next 15-25 years have Although the size of the zyxwvutsrqponmlkjihgfedcbaZYX
sentenced population in the jails
already been born, the accuracy of this assumption will not could be projected much as for the state prisons. projection
significantly affect the projection of prison population. of the total jail population would hc considerably more
Details on the model, data, and assumptions used in complex; thus, the populations of the county jails were not
making the population projections are available from the projected in these analyses.
authors.
Since cases take a finite amount of time to process (IS
I2 The crime types were formed by grouping together crimes percent of the cases disposed in 1975 had taken more than a
with similar legislative penalties and with empirically year to process), many of the disposed cases in a given year
similar imprisonment probabilities and sentence lengths. as result from arrests in that year. but some also result from
determined from Pennsylvania statutes and data. arrests in earlier years. If the increasing caseload between
1970 and 1975 had caused an increasing backlog of cases,
” A third estimation of arrest rates was made using the actual then the parameter c’ would be underestimated in every
interaction in Pittsburgh arrest rates (where a complete year. On the other hand. if the backlog were held stable (by
demographic breakdown in arrests was available); the faster processing, for example) in each offense type and
resulting projections of future arrests also differed only demographic subgroup. then c’ would be unhiased in this
minimally from those obtained assuming no arrest-rate regard.
interaction.
The slight trend apparent in Table 8 seems due primarily to
” Since 1974 was the first year in which the Pennsylvania the changing demographic mix of defendants since prison
State Police operated the state UCR program. this led to a probabilities within crime types and demographic groups
concern that an increase in the number of agencies are much flatter over time.
reporting arrests had occurred. thus causing the increase in
the arrest rate in 1974. However, the U.S. index crime rate Only an offender’s first release is considered here. i.e.. no
as reported by the FBI showed the same pattern of increase account is taken of additional time that parole violators may
(see Figure 2), so it was felt that any reporting increase in serve.
1974 was likely to have contributed only minimally to the This does not include pretrial detention time, which is
arrest rate increases. normally credited towards an offender’s sentence. This time
” In many cases. separate linear regressions of the 1970- 1973 is typically spent in county jails. however. not in state
points and the 1974-1976 points produced similar slopes prisons.
but quite different intercepts. No count of actual prison populations, diaygregated by
crime type, age, and race, was available t<<er<e as ai
” The rule used was to project a value for the 1974 arrest rate
on the basis of the 1970-1973 points using simple linear initial value (P,,I for the model (eauation 10). The prison
regression. If the actual 1974 point was outside a 90 percent population in i970 was estimated by m&iplying the
confidence interval around the projection point, a trend number of court commitments in 1970 by the time-served
change was determined to have occurred. parameter for each crime type and demographic group.
i.e., for I = 1970, P,O,, = C,,,,, T,,,. (This product would
” If a* was the extrapolated arrest rate value in year t. and aTo represent the size of prison populations in that year if both
was’the actual 1976 arrest rate, then the arrest rate value commitments and time served had previously been con-
used in year t. 8,, is given by: stant.)
2, = Max[Min (a:. 1.3 X uTh). 0.7 X a,,] The actual December 31 prison population data for 1Y70 to
1976 were taken from Statistical Analy sis and Correclional
” Juvenile arrestees are processed by the juvenile courts. and
Populution Data, 1976 (Penna. Bureau of Correction.
by order of the Pennsylvania Attorney General, they
1977). The 1977 value in Figure 13 is from U.S. Department
cannot be sentenced to prison; therefore, they are not
of Justice, Law Enforcement Assistance Administration
considered in the analysis.
(1978).
I9 The calculations were not disaggregated by sex after the
The first half of 1971 showed a temporary reversal of a
arrest stage. Women made up only about 3 percent of
three-year increase in prison populations which had fol-
court commitments to prison and of prison populations. as
lowed an eight-year decline (see Figure 3). Since no data
well as a relatively small proportion of court cases; thus,
prior to 1970 on court processing or time served was
separate estimations of imprisonment probabilities for
available, the reasons for this sudden change could not be
women would be highly unreliable due to the small
determined nor could they be accounted for in the
samples involved. Since the arrest projections showed a
projection model.
stable male/female ratio in arrests through the year 2000,
the aggregation of males and females after arrest will not The Pennsylvania Bureau of Correction reported that there
affect the projections. were 8.179 total cells in the state prisons in 1977. of which
7,539 were immediately usable. This does not include the
“I There are two basic types of imprisonment facilities for capacities of the community service centers and group
convicted offenders in Pennsylvania-prisons operated by homes, which housed about 250 inmates at the end of 1Y76
the Pennsylvania Bureau of Correction and jails operated (Penna. Bureau of Correction. 1977).
by the individual counties. Since the state prisons are
operated as a single system. they are considered separately
here. While the population of the county jails in Pennsylva- REFERENCES
nia was almost equal (6,500 in 1975) to that of state prisons,
two-thirds of the county jail populations are composed of Abt Associates. Inc. (1977). Prison population and policy
detentioners (persons being held for trial or other authori- choices, Volume 1, Preliminary report IO Congress. Wash-
26 ALFRED BLUMSTEIN. JACQUELINE COHEN, and HAROLD D. MILLER

ington. DC: Law Enforcement Assistance Administration. President’s Commission on Law Enforcement and Admin-
September. istration of Justice (1967). Task force report: Crime and its
impucr-An assessment. Washington, DC: U.S. Govcrn-
Avi-Itzhak. B., and Shinnar, R. (1973). Quantitative ment Printing Office. Appendix D. pp. 207-10.
models in crime control. Journal of criminal justice l(3):
185-217. Robinson. W.H.; Smith, P.: and Wolf. J. (1977). Prison
population and costs-Illustrative projecrions to 1980. Con-
Barclay, G. (1958). Techniques ofpopulation analysis. New
gressional research service, Library of Congress. April 23.
York: Wiley.
Sagi. P., and Wellford. C. (1968). Age composition and
Belkin. J.: Blumstein. A.; and Glass. W. (1971). JUSSIM.
patterns of change in criminal statistics. Journal ofcriminul
an interactive computer program for analysis of criminal
luw~, criminology, and police science 59:29936.
justice systems. Urban Systems Institute. Carnegie-Mellon
University, Pittsburgh. PA. Stollmack, S. (1973). Predicting inmate populations from
arrest. court disposition and recidivism rates. Journal of
Bishop, Y.: Fienberg, S.; and Holland, P. (1975). Discrete
reseurch in crime and delinquency lO(2): 131-62.
multivariate analysis: Theory and practice. Cambridge, MA:
MIT Press. Toby. J. (1977). A prospect of less crime in the 1980’s. The
New York Time.s. October 26.
Blumstein, A., and Larson. R. (1969). Models of a total
criminal justice system. Operations reseurch 17: 199-232. U.S. Department of Justice. Federal Bureau of Investiga-
Blumstein, A.. and Nagin, D. (1975). Analysis of arrest tion (1976). Uniform crime reporfs. 197-y. Washington, DC:
rates for trends in criminality. Socio-economic planning
U.S. Government Printing Office.
.cciences 9~221-27. -( 1977). Uniform crime reporrs, lY76. Washington.
Chaiken. J. (1977). Distribution of time spent in prison. DC: U.S. Government Printing Office.
Rand working nole (June).
U.S. Department of Justice, Law Enforcement Assistance
Cox. P. (1970). Demography. New York: Cambridge Administration ( lY76). Prlso~irrr in sfufe andfederal insmu-
University Press. lions on December 31. IY7.5. Washington, DC: U.S.
Government Printing Oflice.
Ferdinand, T. (1970). Demographic shifts and criminality.
British journal of criminology 10: 169975. ~ (lY77). Sourcebook of criminal justice .starisncs.
ZY76. Washington. DC: U.S. Government Printing Office.
Fox, J. (1976). An econometric analysis of crime data.
Ph.D. dissertation. University of Pennsylvania. ~ ( 1978). Prisoners iti sfute und fedrrul institutions on
December 3/. JY77. Washington, DC: U.S. Government
Pennsylvania Bureau of Correction (1977). Statistical anaiy- Printing Office.
sis and correcrionul population data. 1976. Harrisburg, PA.
Wellford. C. (1973). Age composition and increase in
Pennsylvania Department of Education (1975). Population recorded crime. Criminology 1 I:hl-70.
projections for Pennsylvania and counties, 1970-2000
Harrisburg, PA. Wilson, R. (1977). U.S. prison population sets another
record. Corrections muguzine 3:3-X!.

You might also like