Measuring Deterrence - Imperfect Data and Estimation Bias

• NCJ Number: 74133
• Author(s): S B Long
• Date Published: 1980
• Length: 264 pages
• Grant Number(s): 79-NI-AX-0007; SOC-7825039
• Sponsoring Agency: National Science Foundation; National Institute of Justice (NIJ); US Dept of Justice NIJ Pub
• Publication Format: Dissertation/Thesis
• Publication Type: Thesis/Dissertation
• Language: English
• Country: United States of America
• Note: University of Washington - doctoral dissertation

Annotation

This study investigates the implications of the use of imperfect data in deterrence research; limits of current macrostatistical approaches for testing deterrence theory are emphasized.

Abstract

Parameter estimates for 'deterrence effects' are shown to be quite sensitive to moderate levels of random or constant error in measuring offense or penalty rates. Because of error bias, the magnitude, statistical significance, and stability of parameter estimates provide inadequate guidance on substantive significance. In fact, the consistency of deterrence findings across studies is enhanced by the presence of such errors in measurement. Preliminary analysis indicates that estimation errors can arise in both single equation and simultaneous equation models. The investigation of error bias uses a variety of approaches, including computer simulations, reanalysis of prior deterrence studies on index crimes, and results with new sources of data. Procedures such as sensitivity analysis and diagnostic plots are suggested to aid in uncovering potential error bias. In addition, diagnostic plots also indicate whether the relationship between violation and punishment rates is consistent with behavioral assumptions underlying deterrence theory. As part of the study, the deterrence model developed by Erlich and revised by Forst is reestimated using data on Federal income tax violations. This analysis also highlights several other problems in correctly specifying the structural equation models used to estimate deterrence effects. First, the association of tax violation rates and income at the State level is found to be opposite in direction from the relationship at the individual level, making it inadvisable to infer individual level behavior on the basis of observed aggregate relationships. Second, the failure of the current deterrence models to control differences in seriousness of offenses across jurisdictions is also shown to lead to specification bias. It is concluded that since errors of measurement present serious problems in estimating deterrence effects with present models, deterrence estimates should not be accepted until sensitivity and other diagnostic checks are made. Tables, figures, footnotes, and approximately 275 references are included in the doctoral dissertation. (Author abstract modified).