Scottish jury research: findings from a mock jury study

Annex H - Statistical tests

The following kinds of statistical tests were used in analysis of the quantitative data included in this report:

• Observed differences between jury-level outcomes, where these can be expressed as percentages (e.g. % giving different verdicts), were tested using CHI-Squared and/or Fisher's exact. Fisher's exact is a more precise test that is used when the analysis includes cell sizes under 5 cases (as they often were at jury level, given there were 64 juries).
• Differences in means (e.g. mean scores on questions on a scale at juror level, or mean length of deliberations at jury level) were tested using 2 way ANOVAs.
• Differences in juror-level responses to questions in the pre- and post-deliberation questionnaires were tested using either a CHI-squared test or a 2-way T-test. CHI-squared was used when testing in general whether there was any variation in response to, for example, a 5 point agree-disagree question by condition. T-tests were used to check whether there was a significant difference on a specific response category (e.g. % agreeing), or when the number of possible responses was very high (e.g. reasons for choosing a particular verdict) so it made sense only to test those that could be significantly different (rather than testing every single category).
• Logistic regression analysis was used in a number of cases to control for interactions between variables and establish which factors were most significantly associated with any difference observed. For example, logistic regression was used to examine which factors were most strongly associated with individual jurors favouring a guilty verdict post-deliberation.