Post-Secondary Correctional Education and Recidivism: A Meta-Analysis of Research Conducted 1990-1999

Journal of Correctional Education, Jun 2004 by Chappell, Cathryn A

Chi-square test for variation across studies

Although a matter of debate, particularly by Hunter and Schmidt (1990), a chi-square test was run to test if the observed variance in the studies is greater than expected by chance. This is a test for homogeneity and can be used to test for moderators. If the chi-square is significant, there may be true variation across the studies, or it may be the result of the operation of moderator variables. Therefore, if the chi-square is significant, it does support a positive relationship between post-secondary correctional education and the reduction of recidivism in this study.

Confidence Intervals

Confidence intervals estimate the extent that sampling error remains in the summary statistics. It is a range of values that the mean size is likely to be if other studies were taken from the population (Arthur, et. al, 2001). Depending on the outcome of the chi-square (significant implies that a participant in PSCE is less likely to recidivate than a non-participant) one of two computations for the standard error of the mean correlation around correlation estimates will be used. Once again, it should be noted that this meta-analysis is not testing for causation, it is a correlation which tests for relationships.

Selecting and testing for potential moderators

A moderator is a variable that affects the relationship between the independent and dependent variables. It can account for, or helps explain, more variance than would otherwise be the case. Since the possibility of moderators exist, individual subsets of the studies in the overall meta-analysis will be tested for their influence. A moderator variable is typically identified by a corrected variance that has a lower average in the subsets than for the data as a whole, and a corrected mean r that varies from subset to subset. Three theoretical moderators have been identified by the researcher and are discussed in the results section. Subset 4 is not so much a test for moderators as it is a separate metaanalysis.

Three studies were quasi-experimental and included control groups. In their reports, comparison data from general statistics as well as the control group, were included. For each of these studies, two sets of data were produced: one using the general recidivism rate for that state or institution and another using the control group recidivism rate. The first set was included in the primary meta-analysis and a second, smaller meta-analysis was run on the second set of data (Subset 4). It is hypothesized that the control group meta-analysis will control for limitations such as selection bias, and will possibly show less of a relationship between PSCE and recidivism reduction.

Limitations

There are several known limitations of this study that are beyond the control of the researcher. First, the relatively few (15) studies appropriate and available for this 10-year meta-analysis were a disappointment to the researcher. Initial searches and reviews had indicated that a larger sample would fit the criteria. As noted earlier, several factors contributed to this limitation. Many studies that included post-secondary correctional education in their investigation of recidivism did not separate the data between secondary and post-secondary. Several studies failed to note the population and statistics they were using for comparison purposes. Other studies reported relationships but did not supply the necessary data for inclusion.