Population Projections in Practice

The previous discussion demonstrates how assumptions about future conditions are critical to the results of projection models. The most effective projection models allow decisionmakers to consider a wide range of policy choices and to incorporate those choices into a series of different models so that their effect on future populations can be seen. (A brief history of corrections population projection methods is presented below and a summary of commonly used projection models follows.) If used in this way, population projections can be flexible tools for understanding the ramifications of various policy choices and the use of confinement resources. Projection models, however, should not be offered to policymakers as a simplistic mechanism for predicting future corrections populations.

A Brief History of Corrections Population Projection Methods

Beginning in the early 1970’s, corrections researchers began to develop increasingly sophisticated methods for projecting adult prison populations. Their methods drew largely from the fields of demography and operations research. Since the 1970’s, population projection models and the data available for those models have improved considerably. The fundamentals of population projections, however, are still based on the work of a few original innovators.

In 1973, Stephen Stollmack published one of the first “mathematical flow” models for projecting prison populations. The model used an input-output analysis of the corrections system. It incorporated data about how offenders “flowed” through the stages of the justice process—for example, from arrest to indictment, conviction, and incarceration. Prison populations were projected by relating flows to “stocks” (or the starting point of a prison population) and by incorporating information on the average length of time individuals stay in prison. The model even allowed for limited evaluations of policy changes (for example, the impact of policies that change length of stay can be built into the model and their impacts can be assessed by seeing how the prison population is affected).

Stollmack’s model took population projections beyond traditional statistical models (e.g., time series and regression). Statistical models projected future populations by linear extrapolation of trends in prior populations. Statistical models continue to be used today because they allow forecasters to make projections without having to assemble a great deal of data about case processing. With statistical models, however, forecasters cannot disaggregate projections for subpopulations, nor can they analyze the impact of policy changes that affect only certain types of offenders.

In addition, statistical models are effective only when data are available for extended periods, and they can be difficult to interpret for nontechnical audiences.

In 1980, Alfred Blumstein and his colleagues continued the development of mathematical flow models by making two enhancements to the Stollmack model (Blumstein, Cohen, and Miller, 1980). First, they disaggregated population projections by racial and crime categories. Second, instead of assuming a constant rate of admissions into the population, their model projected admissions as age-specific proportions of the general population. They developed these proportions with census projections and historical data on prison admissions. Their innovation acknowledged that rates of crime, arrest, and incarceration varied among groups in the general population. Population projections were calculated as a weighted sum of the separate projections for each subpopulation.

Arnold Barnett (1987) introduced another refinement to mathematical flow models based on the concept of “criminal careers.” Barnett’s model began with age-specific probabilities that nonincarcerated offenders are actively involved in crime. His model estimated the incarceration rate for offenders based on several factors—age, criminal activity, and the expected rate of desistance. The probability of criminal activity could be revised within the model to account for policy changes, and the impact of these changes could be factored directly into projections of prison populations.

While Blumstein and his colleagues and Barnett were improving Stollmack’s mathematical flow model, other researchers were developing an entirely different approach to population projections. This second approach would become known as “microsimulation.” By the end of the 1990’s, 24 States and the Federal Bureau of Prisons were using some form of microsimulation to project prison populations (Sabol, 1999).

Microsimulation models project prison populations by simulating what happens to individual offenders as they are processed by the justice system and enter and leave prison. Early microsimulation models began by estimating the length of time individual offenders were likely to remain in prison. For each prison admission, a path (or “trace vector”) is mapped. Future prison populations are projected by adding together the number of individuals remaining in prison at any given point in the future. The California Department of Corrections developed one of the first functional microsimulation models in the early 1970’s (Chaiken and Carlson, 1988).

In the early 1980’s, the National Council on Crime and Delinquency drew from the experiences of California when it developed its “Prophet” model (National Council on Crime and Delinquency, n.d.). The Prophet model was constructed on the concept of “ID groups”—subpopulations of offenders categorized according to how they were likely to be handled in the justice system. Each group could be modeled through various decision points in the criminal justice system, and lengths of stay were estimated using sentencing variables or data on time served by previous cohorts of released offenders. Incarcerated populations were projected by estimating the number of offenders in each ID group who were expected to be in prison at certain points in the future.

Unfortunately, many State and local agencies are still unable to produce the detailed data necessary to make full use of microsimulation models. In practice, most jurisdictions continue to use grouped data rather than individual-level data in their population projections. Whenever grouped data are used, microsimulation models function essentially as disaggregated flow models.

Note: Much of this history is drawn from Sabol (1999).

Because projection models are unable to account for all of the details involved in the juvenile justice process, they will never be foolproof. Moreover, until State and local agencies are able to support significant expansions in their data collection and analysis capabilities, it is unlikely that any projection model will ever represent the true diversity of the juvenile population. For this reason, juvenile justice agencies should resist the temptation to rely on any single prediction of future demand for space. Instead, they should invest in an extended process of “forecasting.”

Models Commonly Used To Project Corrections Populations

Projecting corrections populations is often incorrectly understood as an effort to “get the right number.” This assumes that a projection is inferior if it produces a number that turns out to be different from actual need or if a projection becomes irrelevant after a change in policy. It is more appropriate to view projections as conditional statements of a future corrections popula tion that will hold true only if current assumptions about the factors that generated past populations persist into the future. A comprehensive forecasting effort should include not only population projections but also policy debates and analyses to understand why actual populations depart from projections and to demonstrate the role of policy in shaping demands for space.
Type of ModelMethod or ApproachComments
Microsimulation
  • Projects the movement of individual entities through the justice system using detailed information about real individuals who have gone through the system or are still in process.

  • Permits users to aggregate information at the end of a simulation into whatever categories are needed.

  • Offers the greatest flexibility/power in projecting populations under various policy assumptions.

  • Requires extensive data about individual offenders.

  • Most State and local jurisdictions are not able to meet the data requirements.

  • For national-level projections, data requirements for microsimulation will likely never be met.
Disaggregated flow
  • Uses rates of flow between the stages of the justice system (e.g., odds of adjudication after arrest, odds of incarceration after adjudication).

  • Rates can be entered and then altered for various subpopulations for repeated projections over time.

  • Generates projections based on the movement of groups through the justice system.

  • Next to microsimulation, offers the most flexibility for anticipating future conditions.

  • Requires grouped data only.
Statistical Uses methods such as time series or multiple regression to project populations based on changes in other, related variables.

  • Requires less data but does not provide much flexibility for modeling future policy changes.

  • Generates projections based on past values of the variable to be projected and their relationship to other factors.

  • May require the values of independent or causal variables to be projected as well.
Mathematical May involve various methods, ranging from simple growth-rate projections to more sophisticated stochastic models.
  • Requires minimal data but is very inflexible.

  • Projections are generated by adding a constant to existing populations or by multiplying populations by calculated growth rates.

  • Assumes future conditions will be the same as past conditions.

  • May include parameters that relate inflow to outflow or that model length of stay in corrections.

Forecasting Rather Than Predicting

Forecasting is different from predicting, although both strategies involve statistical projections of corrections populations.

Forecasting relies on reflection instead of speculation. In a prediction context, researchers focus on the future. They use data about the past to speculate about the future, and they encourage policymakers to act on their statistical vision of the future. In a forecasting context, researchers focus on the recent past. They use data to understand how the recent past turned out to be different from previous expectations. By identifying and examining these differences, policymakers and other professionals increase their understanding of the factors that are likely to influence future trends, but they do not place undue faith in anyone’s ability to predict those trends accurately.

A forecasting approach also encourages decisionmakers to review their assumptions about their own policies and practices on a regular basis. Some agencies may engage in a forecasting process on an annual or even semiannual schedule. They conduct repeated projections of their corrections populations and compare actual developments with their previous expectations of demand for bedspace. Administrators and policymakers use the occasion of each forecasting exercise to review their assumptions about their system and how it uses bedspace. In such an environment, population projections can be used to encourage sound policy and practice decisions. (See “Forecasting Juvenile Corrections Populations in Oregon” for a description of one agency’s approach to integrating forecasts into its policy process.)

Forecasting Juvenile Corrections Populations in Oregon

The Oregon Youth Authority obtains twice-yearly forecasts of the number of young offenders likely to be in its “close custody” programs 10 years into the future. (Close custody refers to youth housed in the State’s MacLaren and Hillcrest facilities and also those in “accountability camps,” “work study camps,” and Oregon’s Juvenile Intake Center.) Forecasts are generated by Oregon’s Office of Economic Analysis using models developed by the office and overseen by an interdisciplinary advisory committee. Members of the committee include researchers from a local university, court and probation officials, and the Director of the Oregon Youth Authority.

Each forecast incorporates the most recent data on intake trends, arrest trends, and future population growth for Oregon youth ages 12 through 17. Separate models are used to forecast important subpopulations within the juvenile offender population, including youth affected by Oregon’s “Ballot Measure 11,” which automatically transfers certain categories of offenders to the criminal court.

The forecasts are provided to policymakers and other officials in the State to foster discussions about recent trends and their effect on future corrections populations. The Office of Economic Analysis advises officials that each “forecast is not what the population will be, but what the population would be if current practices and policies were applied to future conditions” (Oregon Youth Authority Close Custody Population Forecast: Biennial Review of Methodology, page 2 ).

Source: Oregon Youth Authority Close Custody Population Forecast (April 2000), a biennial series, and Oregon Youth Authority Close Custody Population Forecast: Biennial Review of Methodology (June 1998). Salem, OR: Oregon Office of Economic Analysis.

No single projection exercise should drive policy and budgetary decisions. Every projection should be used in conjunction with policy debates about the type of programs a jurisdiction wishes to support. Decisionmakers can use a forecasting process to reflect on current policies and practices and to ask critical questions about their use of bedspace: If current trends continue, which type of offenders will be committed to secure confinement and which will be placed in community-based programs? What type of offenders will stay the longest in secure facilities? Which facilities will see the largest increases in daily populations or length of stay? Which areas of the State will experience the greatest changes in expected demand? Projections of future custody populations can be powerful learning tools that serve the twin goals of making communities more secure and providing appropriate treatment programs for youth.

    Differences Between Predicting and Forecasting

 PredictingForecasting

    Focus Future Recent past

    Goal Accurately predict the future Examine recent developments and their relevance for the future

    Methods Statistical projections Statistical projections, policy discussions, program reviews

    Personnel Involved Analysts Policymakers, administrators, practitioners, analysts

    Frequency As needed Regularly

    Definition of Success Accuracy Utility/learning

Forecasting and the Policy Process

The juvenile justice process has many unique features that need to be accounted for in projection methodologies. These features include a wide use of diversion, great discretion at all levels, and the juvenile court’s ability to base dispositions on not only the public safety but also on the best interests of the juvenile. Because juvenile court dispositions are sometimes for indeterminate periods of time, lengths of stay are often linked not only to the severity of the offense but also to a youth’s progress in treatment programs and the availability of space. As a result, juvenile detention and corrections systems have much less stable information on which to build forecasts than criminal justice agencies.

Researchers must encourage policymakers and administrators to understand that no projection methodology will ever be able to model the complexity of the decisionmaking processes that lead juvenile offenders to be placed in secure facilities or that determine how long juveniles will stay in those facilities. It will always be necessary for decisionmakers to review the results of a projection model and consider its value for policy and practice. However simple it may appear at first, estimating a jurisdiction’s future need for detention and corrections space requires an extensive examination of the justice system and of the processes used to select juvenile offenders for placement.

Methodology
Decomposition Methods
A statistical flow model is used in this analysis to decompose changes in the national juvenile commitment population between 1993 and 1997. The model segments the overall change in the commitment population into offense-specific groups (person, property, drug, and public order). Within each group, the model decomposes the overall change in the commitment population into the portions of total change that can be attributed to the following factors:

  • Changes in the number of juvenile court referrals.

  • Changes in the number of referred cases that result in adjudication.

  • Changes in the number of adjudicated cases that result in residential placement.

  • Expected length of stay in residential placement (using a stock/flow estimate of length of stay).

The offense-specific changes in these components of growth are then aggregated to obtain the total change in the juvenile commitment population over the period of analysis.

The population change model used in this Bulletin follows the approach of Abrahamse’s (1997) method for assessing change in prison populations. The number of juveniles committed to residential placements at the end of a year is defined as follows:

POPULATION = REFERRALS x ADJUDICATION
x PLACEMENT x LENGTH OF STAY

Where each element is defined as follows:

POPULATION = the juvenile population committed to residential placement facilities.
REFERRALS = the total number of delinquency cases referred to the juvenile court system.
ADJUDICATION = the proportion of referred cases that results in adjudication.
PLACEMENT = the proportion of adjudicated cases that results in commitment to residential placement facilities.
LENGTH OF STAY = the expected length of stay, estimated by a “stock/flow” ratio (see discussion).

The amount of change in the juvenile commitment population between 1993 and 1997 is a function of

the offense-specific changes in each individual component of change as measured in the above model. Thus, the difference in the population is a “weighted sum” of differences in each component, where the weights equal the offense-specific contribution to change in the population. The decomposition of change is applied separately to each offense group, and each of the offense-specific changes in the juvenile commitment population can be summed to obtain the total change in the population between 1993 and 1997.

Projection Calculation
Using data for the 1993–97 period, a mathematical flow model is used to project the juvenile commitment population for the years 1998 through 2002. The model follows the approach developed by Stollmack (1973) to project prison populations. The analysis uses the following equation to project the juvenile committed population for each year, from 1998 to 2002:

P(t) = A(t)xLOS(t)+[P(t–1)–(A(t)xLOS(t))]x exp[–1/
LOS(t)]

Where each element is defined as follows:

P(t–1) = the population in the previous year (t–1).
A(t) = admissions or commitments to residential placement during the year.
LOS(t) = the estimated length of stay in commitment.
t = the time unit for flows (in this example, years).

This model requires three data inputs for each time period: the starting population, which is the population from the previous time period [P(t–1)]; admissions during time t; and length of stay. The projection scenarios described in this Bulletin use the 1997 juvenile commitment population as the initial starting population and assume that admissions either remained at 1997 levels throughout the 1998–2002 period or that they increased each year based on applying the average annual changes observed from 1993 to 1997. Similarly, average length of stay is either assumed to remain at 1997 levels or projected for each year based on the average annual change observed from 1993 to 1997.

As with the decomposition model, the projection models presented in this Bulletin were apportioned into offense-specific components (person, property, drug, and public order) and then summed to obtain the total populations projected for each year from 1998 to 2002. Since data on the number of committed youth released from residential placement were not available for all years in this analysis, the model presented in this Bulletin must assume that admissions and releases were in equilibrium.

An effective forecasting process should take into account the important role played by each jurisdiction’s policy preferences and professional practices. Forecasting should include at least three general areas of activity:

  • First, decisionmakers should have regular access to extensive data about trends in juvenile crime and juvenile justice processing within their jurisdictions, and they should use that information to project the size of future detention and corrections populations.

  • Second, they should develop a thorough understanding of their jurisdiction’s policies and practices regarding the use of secure confinement for juvenile offenders, including how the diversity and depth of juvenile justice resources are related to the need for secure space.

  • Third, they should host a rotating series of strategy meetings with a variety of audiences from the juvenile justice system and the larger community. These meetings should focus on the relationships among the availability of juvenile justice program resources, recent trends in the use of those resources, and projections of future confinement populations.

The validity of any projection model rests on the reasonableness of its assumptions and the persistence of these assumptions into the future. When projections fail to anticipate future conditions, forecasters should seek to explain why actual populations differ from projected populations. Decisionmakers then have the opportunity to learn about the effects of practice and policy actions that were not included in the projection.

The success of a forecasting process is not determined by its predictive accuracy. A projection that turns out to be wrong (or one that produces population estimates that deviate from actual future populations) is not necessarily an invalid projection. An invalid projection is one in which the differences between a projected population and the actual population cannot be explained. A projection that turns out to be inaccurate as a prediction may still be a useful projection if analysts are able to explain which critical assumptions were violated and what impact these violations had on corrections populations.



Previous Contents Next

Anticipating Space Needs in Juvenile Detention and Correctional Facilities Juvenile Justice Bulletin March 2001