© 2004 by Oxford University Press
A re-examination of the 1966 Kalven–Zeisel study of judge–jury agreements and disagreements and their causes
George Washington University, 2140 Pennsylvania Avenue N.W., Washington D.C. 20052, USA
The classic study, The American Jury, by Kalven and Zeisel showed that trial judges agreed with juries in criminal cases about 75% to 80% of the time. Kalven and Zeisel observed that fewer disagreements occurred in cases where the evidence was clear than when it was close. They found that the juries were more lenient and called these normal disagreements. They also explored the possible factors that determined the normal and other disagreements. Kalven and Zeisel quantified the effect of a factor using a measure now referred to as the population attributable risk. Here, the factor attributable risk (FAP) measure is adopted as it focuses on those cases where that factor is present. The FAP is the measure relied on in product liability and related tort cases to estimate the fraction of cases in the exposed population that are attributable to the exposure under study. The advent of readily available computer packages enables us to use logistic models to reanalyse the data and estimate the effect each factor has on the odds of a disagreement. Thus, while the original authors showed that the presence/absence of a prior record plays a key role in creating a disagreement between the judge and the jury, now one can estimate the increased odds of a disagreement it generates controlling for other variables. From this quite general logistic model it is seen that the strength that ‘absence of a prior record’ had in the original study is partly due to the fact that it is a significant factor in determining whether the jury views the defendant sympathetically. The weight given to the various explanatory variables also depends on the seriousness of the crime. In serious crimes, both a superior defence counsel and a prior record are influential determinants of whether there is a disagreement. For lesser crimes, however, a jury's sympathetic view of the defendant is the main factor creating a disagreement between the judge and jury. The re-analysis demonstrates the insight that logistic regression and related generalized linear models can provide as well as the importance of utilizing the most appropriate measure of attributable risk to evaluate the role of the various explanatory factors.
Keywords: accuracy of verdicts; judge–jury agreement; classification errors; measures of effect
Received 11 August 2003. Revised 29 June 2004. Accepted 3 September 2004.
* E-mail: jlgast{at}gwu.edu. Joseph Gastwirth is Professor of Statistics and Economics, Department of Statisitcs, George Washington University, Washington D.C. 20052. Michael Sinclair is Director of Statistical Analysis, Office of Federal Contract Compliance Programmes, U.S. Department of Labor. Washington D.C., 20210. This research was supported in part by grant from the National Science Foundation.