April 19, 2007
A systematic examination of prison growth
John Pfaff has a fascinating new article that examines the theories and the empirical literature on the forces driving prison growth in the US over the past three decades. The paper, entitled "The Growth of Prisons: Toward a Second Generation Approach" is available at this link. Here is the abstract:
Over the past three decades, the US prison population has soared from 300,000 inmates to 1.5 million. In recent years, many scholars have devised rigorous empirical models to try to determine what forces have been most responsible for this impressive growth. This article reviews these studies and finds that all suffer from important shortcomings that limit the extent to which they accurately identify causal mechanisms. The problems are both technical and conceptual. Technically, most studies either fail to control for several significant empirical defects ― such as endogeneity, omitted variable bias, and colinearity ― or so do unconvincingly. Conceptually there are several issues. In some instances, for example, it is unclear whether the variable chosen to test a particular causal theory is an effective or accurate proxy; in others, the theory itself does not appear to be formulated correctly. This article sets forth the problems with the current studies and suggests technical and conceptual improvements for future work.
April 19, 2007 at 11:08 AM | Permalink
TrackBack URL for this entry:
Listed below are links to weblogs that reference A systematic examination of prison growth:
In other words it is difficult to construct a model when most of the variables are out of control.
Posted by: John Neff | Apr 19, 2007 4:43:25 PM
Is the prison incarceration rate for violent crime is related to the violent crime rate? If you plot the incarceration rate for prisoners (prison inmates per 100,000 persons) where the most serious charge is a violent crime versus time between 1980 and 2003 you will get a steadily rising curve. If you plot the violent crime rate versus time for the same years you get a curve that first rises and then falls which means that you have two values of the incarceration rate for the same value of the crime rate. If there is a relation it not a simple one.
The results are similar when you plot incarceration rates for property crimes and property crime rates versus time. If this were a physical system we would say this is evidence of hysteresis but a prison system is a social construct so hysteresis is not a valid descriptive term.
The empirical models used to model prison populations are multivariate statistical models that use sociological and criminal justice system data. I use criminal justice system data extensively and whenever I can I apply validity tests with failure a common outcome. I think the criminal justice system data quality is too low for even a well formulated model to provide credible predictions. It may be possible to improve the quality of future data but it does not appear to be possible to fix the problems caused by the older missing and suspect data.
On the other hand an experienced analyst who understands a particular prison system and is aware of the data quality issues can make useful short term forecasts (about five years). The same analyst can also make a useful estimate what will happen to the prison population if a sentencing or parole policy is altered. The primary audience for such forecasts is the governor, legislative committee members and reporters assigned to cover criminal justice issues.
I don't know what it is like in other states but in Iowa the legislature will order that certain types of data should be collected and reported by the cities, counties and the courts but they do not provide funds to collect, analyze and store the data. As a consequence incomplete data is collected stored for awhile and then thrown out because they lack space.
Posted by: John Neff | Apr 21, 2007 9:47:03 PM