Forecasts and Simulations: Two Different Analysis Tools

Because the terms forecast and simulation are often confused or used interchangeably, this paper describes the two different policy analysis functions that these methods address. A forecast is a population projection based on historic data and past growth rates which may be modified to reflect assumptions about changes in underlying trends or forces such as a jurisdiction's general population. A simulation model replicates the flow of individuals or cases through a system so that the potential impact of changes in procedures or resources can be estimated. Although these two analysis tools are closely related and are often used to complement each other, very different methods and data sets are used to produce either of these estimates.

A forecast can be produced in a relatively short amount of time since it is mainly a mathematical/statistical operation which applies various growth rates to a base population for a specified number of time periods. After obtaining the necessary historical data and determining the assumed growth rates, an analyst can calculate future populations for any number of possible future scenarios ranging from no change in recent growth rates to radical shifts due to any contributing growth factor such as demographics or available resources.

However, the reliability of this forecast depends on two major assumptions. The first is that the relationship between whatever underlying factors and the overall growth rate are perfectly understood. Unfortunately, changes in complex systems may involve a number of actions and reactions which do not always result in the expected outcomes. A simulation can sometimes help with documenting the impact of alternatives and identifying the interactions of changes within a system. The second assumption is that all of the factors contributing to an overall growth rate will occur as planned. For example, while an overall annual prison population growth rate of 3% may be considered reasonable for developing a forecast, the actual results depend upon the expectations for average sentence lengths, release rates, and other contributory factors being met to produce this outcome. The forecast is only telling you the likely populations if the assumed conditions and causes should actually occur. This is why an on-going monitoring effort is important for assessing the accuracy of any forecast.

A simulation model requires more time and testing because it must accurately reflect the actual processes or case flows occurring in your corrections system. This includes identifying all possible sources of new admissions and readmissions into the system and each significant decision point that could affect the retention or release of an individual. The length of time between decision points must also be specified in order to know when these would occur for a given time period. The information requirements grow geometrically as the number of subpopulation characteristics (e.g., individuals by offense type, by risk classifications, by gender, etc.) increase since each is a separate component of an overall system. Once a valid simulation model is developed, it can be used for assessing various alternatives which affect how fast and in what manner an individual moves through the corrections system.

For example, a population forecast may help in determining the impact on the corrections population when increased arrests and prosecutions are being considered. This could estimate how much inmate admissions will increase in future years assuming that the rate of arrests resulting in convictions stays constant or changes in a predictable manner. On the other hand, a simulation model would be used for analyzing the effect of longer sentences. Because of this policy change it would be expected that future populations will be held longer in the corrections system. Although the net effect of both policy changes would be an increased need for prison facilities, the way that they affect the corrections systems is different. Therefore, different approaches are needed for estimating the potential impact of various policy alternatives.

Regarding the prior forecasting example, there also may be additional refinements that are needed for accurate estimates of future populations. If the police and prosecutors will be concentrating on drug offenders specifically which means this group can be expected to have a growth rate substantially different from other offender groups, then separate forecasts may be necessary. This is because as drug offenders increase as a proportion of the total population, using average population figures which do not apply to this group will result in erroneous estimates. This will be true for any major category of offender for which there is concern that their growth rates or other factors will be significantly different from general averages. Similarly if proposed policies only apply to certain subpopulations, then the simulation model should be based on data for these specific subgroups to insure greater accuracy and reliability.

When criminal justice policies or other conditions radically depart from past circumstances, a forecast based on historical trends will have little validity other than providing a baseline for future population estimates. Usually a combination of forecasting and simulation tools are used in these situations to draw on applicable past data and to model likely future outcomes based on new policies, procedures, or other changes. This requires a sufficiently detailed and extensive database of information to forecast or simulate the many factors which ultimately result in the total corrections population.

One approach is to develop several alternative projections which use a range of growth rates or other assumptions. However, these should only be considered as illustrative of possible future outcomes if all assumptions are valid and all expectations are met for a particular scenario. As was previously mentioned, relying on these estimates would have to be supported by a constant monitoring effort to detect any deviations occurring in the presumed conditions underlying a forecast. Otherwise, there is a strong likelihood that there will be some surprises in prison facility needs since assumptions about the future are often wrong and operating conditions do change.