Evaluation Review

President Carter proclaimed a war on fraud and error in the food stamp program as part of a more general concern with the mismanagement of social programs. He succeeded in reversing the increase in the number of households classified as receiving both too many and too few food stamps, and he also increased the percentage of eligible households that actually received stamps. An even more visible war against fraud and abuse was pursued in the first Reagan administration. This resulted in a decrease in the number of cases being paid too much, but no such decrease in the number paid too little. The percentage of eligible households receiving food stamps also decreased, and there was a rapid increase in the number of fraud cases detected and prosecuted. However, conflicts emerged in the first Reagan term between federal and state governments about who should bear the financial costs of the war on error and fraud. The second Reagan term resulted in few changes in error and fraud compared to what was found at the end of his first term, and the level of rhetoric about "fraud and abuse" decreased markedly. However, the data available on fraud and abuse deal only with detected frauds. Hence we cannot say with any certainty whether fraud actually decreased m the food stamp program during either the Carter or the Reagan years.

In January 1987, Kenosha County, Wisconsin implemented its work experience and job training program, a prototype job opportunities and basic skills training (JOBS) program. Kenosha's program requires participation, has broad coverage, and is relatively expensive. This article reports findings on the impacts the program has on aid to families with dependent children (AFDC) spells. The evaluation used administrative welfare data and an event history analysis. Findings suggest that Kenosha's program has no affect on AFDC-Basic spell lengths, has a positive impact on AFDC-Basic cases headed by never-married black teenagers, and has a negative impact on A FDC-Unemployed Parent cases.

Evaluations of the impacts of government employment and training programs on workers have generally followed a strategy whereby the labor market outcomes of workers who receive program services are compared with the outcomes of workers who do not receive program services. Differences in outcomes are then attributed to the effects of receiving services. This article examines the problem of measuring the impacts of reemployment bonuses when treatment group members and control group members interact in the labor market. It is shown that the conventional measure of bonus impacts — the difference in average time spent unemployed of the treatment group and control group — may significantly overestimate the net impact of the bonus offer, and that measuring the net impact correctly requires the presence of another control group that is not affected by the bonus program. It is also shown that measured impacts from previous bonus experiments may have been due entirely to displacement, but the designs of these experiments were not statistically adequate to detect displacement. The implication is that the results of previous bonus experiments should be treated with considerable caution.

Graphics are widely believed to be useful ways to communicate statistical information and stimulate data analysis. Yet we often use only the most simple graphics, bar and pie charts, or tables to present data, especially with the evaluation's audcences. Little research is available to assess the effectiveness of graphical displays with evaluation audiences. Two types of graphical displays, a modified box plot and a multivariate STAR graph were compared with tables containing the same information in an experiment with five groups that comprise the audience for evaluation results. For the univariate display, the audiences for an educational evaluation were 67% accurate in interpreting evaluation results when using a graph, compared to 63% accuracy for a table. For the multivariate display, the audiences made accurate comparisons 70% of the time, but these results were bettered by those using a table (84%). However, an individual item analysis offers some guidance for designing multivariate graphics with nontech nical audiences that may be used to improve these displays.

The random assignment of indivcduals to treatment conditions is often corrupted in field studies. The typical result is nonequivalence between the treatment groups and a corresponding reduc tion in the study's credibilety. The corruption of the assignment process seems to be particularly common in studies involving trickle processing, wherein the research participants appear and are assigned to treatments one at a time over an extended period Several procedures for safeguarding the integrity of random assignment with trickle processing have been proposed. The present article describes a computer-based, alternative method

Issues related to the reliability and validity of self-reported behavior within a deviant population are examined using data obtained from narcotics addicts in two face-to-face interviews conducted 10 years apart. The same measures of behavioral self-report for an overlapping period of 4-5 years were collected at each interview and were analyzed within a test-retest design. Agreement between measures obtained at the two recall points was reasonably good except for low-frequency behaviors and for measures of income and monetary costs. The correlation coefficients of the intervariable relationships measured at each interview increased as the constituent item reliabilities increased. Urinalyses conducted at each interview generally validated recent self-reported narcotics use.

A pretest-posttest control group design was employed to assess the leadership and problem-solving skills of participants in a statewide community leadership development program. Quantitative data and analysis revealed that significant changes had resulted from the program. Qualitative data collected after the conclusion of the program revealed how program participants became further involved in their communcties in a variety of ways and on several local issues. The need for evaluation of community leadership development programs to document community impacts is then discussed

Analysis of several data sets suggests that when "don't know" is used as an option in an evaluation of others, the number of omitted items may correlate negatively with the summated evaluation. When present, this variation of social desirability or nonrandom tendency toward item nonresponse has the potential to bias results. Measures are suggested to both detect and control for this tendency.