Effects of pretrial juror bias, strength of evidence and deliberation process on juror decisions: new validity evidence of the Juror Bias Scale scores

Abstract

This study examines the validity of the Juror Bias Scale scores in relation to the mock juror decisions reached in two real life homicide cases before and after the deliberation process. The judicial cases used varied in terms of the ambiguity of the evidence presented at both trials. The WLS methodology for statistical modelling of categorical data was used to analyse data. The findings indicated that the Juror Bias Scale scores successfully predict the verdicts and other related questions before and after deliberations in the case with ambiguous evidence. Furthermore, deliberations caused a generalisation effect on the pretrial juror bias in such a case, and enhanced the differences between defense-biased and prosecution-biased jurors in the verdicts delivered after deliberations. The implications of these findings are discussed in relation to the use of pretrial juror bias questionnaires in jury selection.

Keywords:

• Juror Bias Scale
• Validity evidence
• Jury deliberation effects
• Jury selection
• Juror decision making
• Pretrial juror bias

Notes

¹Prior to the completion of the present study, an initial phase of translation and adaptation of the scale to Spanish standards, evaluating the models proposed by Kassin and Wrightsman (1983) and Myers and Lecci (1988) for its factor structure was carried out. The results obtained indicated a bidimensional structure with two closely related factors (Probability or Guilt and Reasonable Doubt), suggesting the viability of the use of a single score as a bias measure along the lines indicated by the authors of the scale.

²To classify the subjects we used the same procedure followed by the authors of the original scale (Kassin and Wrightsman (1983)). In that study the median of the scale was 51.50.

³In Spain the juries are composed by nine persons for all cases. The participants were assigned to the groups trying to maintain a certain variability with regard to their studies and gender, so that none of the juries would consist totally of men or women or of the same discipline.

⁴The decision rule adopted was that established by the Spanish Jury Law, in which both the voting on the facts (stipulated in the questions relating to the verdict object) and the final vote on guilt or not guilty are based on the presumption of innocence. Thus, to declare the facts proven, when they are favourable to the defendant, five to nine votes are required and when they are unfavourable seven to nine. The same rule is followed to declare the guilt or innocence of the defendant, so that to consider the defendant guilty a minimum of seven votes are needed in favour of a guilty verdict, while to declare him/her innocent five votes are sufficient. Such a distribution of votes is compulsory, which is called “qualified majority”, for each question and for the verdict, so that if they are not achieved in the first vote, this will be repeated as often as is necessary to obtain a “qualified majority”.

⁵Since the variable “bias tendency” has one degree of freedom, its effect can be estimated by the direct comparison between both categories (D-biased vs. P-biased) at each level of the repeated-measures factor. Thus, the comparison between the two groups can be proved by the quotient between the estimator and its typical error, which is equivalent to the Wald statistic.

⁶For this purpose, a model in which the interaction effect was replaced by this specific effect was fitted. The results indicated that the distribution of guilty marginal proportions vary significantly throughout the times in which the verdicts were reached by the jurors, both in the D-biased (χ²(3) = 69.91; p < 0.001) and in the P-biased subjects (χ²(3) = 15.69; p < 0.001). Subsequently, contrasts two-by-two were carried out between the parameters estimated by the fitted model for the differential effects of some of the repeated-measures factor levels.

⁷Group marginals ≥ 25 and at least two n _ij  ≥ 1 (Stokes, Davis and Koch 1995).

⁸The main advantage in using this type of response variable as opposed to the one used previously, is that it allows us to work with only one response variable for each experimental condition, whereas using marginal proportions we would have to use three response variables for each condition, since we have four response categories. Therefore, the response variable is defined as the average score obtained by each group at each level of the variable with repeated measures.

⁹Therefore, a main effect model was adjusted for the data, which confirmed the previous result. The residual goodness of fit test indicated that this model was adequate [W(3) = 0.91, P(χ² ≥ 0.91) = 0.824].

¹⁰The analysis of variance of the saturated model for this variable showed that both the main factor with repeated measures (χ²(3) = 88.58, p < 0.001) and the main fact of the “bias tendency” variable (χ²(1) = 4.11, p = 0.0427) were statistically significant. Following this, a main effects model was fitted. When this model was fitted the main effect of the “bias tendency” variable ceased to be statistically significant (χ²(1) = 3.53, p = 0.0674), which was to be expected given the p value that this effect had taken in the saturated model and the average scores of the groups in the different experimental conditions. The main effect of the factor with repeated measures was confirmed as the only one statistically significant (χ²(3) = 97.15, p < 0.001). The residual goodness of fit test indicated that the model was adequate [W(3) = 2.65, P(χ² ≥ 2.65) = 0.4489].