Results Table 1 shows means and standard deviations of items measured at T1 and T2, and T1–T2 change scores. Observed ranges for self-reported speeding were 0–4, and dangerousness 1–5. All variables were approximately normally distributed with adequate variance, although a ceiling effect may exist in the danger at 80km/h variable. This variable showed a mean of 4.69 at T1, and 4.35 at T2. A logarithmic transformation was used to correct a negative skew. This was transformed by creating T1 and T2 binomial variables, combining the categories ‘all the time’, and ‘most of the time’, and comparing them with other categories. The untransformed variable appears in Table 1 but the transformed variable was used for all inferential tests.Table 1 also shows that self-reported speeding and perceived danger at 70 and 80km/h decreased significantly from T1 to T2. The change in self-reported speeding was negatively associated with changes in perceived danger at 70 (rð253Þ ¼ 2:17, p , :05), and 80km/h (rð253Þ ¼ 20:13, p , :05), but not changes in risk beliefs (rð253Þ ¼ :06, p ¼ :346). These negative associations may appear incongruous, as scores on all variables decreased from T1 to T2. This can be resolved by the fact that the effect size (h2) of the reduction of self-reported speeding was greater than those of perceived danger at 70 and 80km/h.Preparatory analyses A discriminant function analysis was performed to detect differences on T1 variables between participants included and excluded at T2. The function was significant (l ¼ :981, x2 ð6Þ ¼ 14:90, p , :05). Follow-up t tests showed that non-retained participants were significantly older (Mean ¼ 46.51, SD ¼ 16:81) than those retained (Mean ¼ 42.00, SD ¼ 13:49; tð644:24Þ ¼ 3:95, p , :01). Of the participants, 47 reported being unaware of the media/enforcement campaign. Four 2 £ 2 mixed model ANOVAs were conducted, using the distinction between those who reported being aware and those unaware of the campaign to predict T1 and T2 observations of speeding, danger at 70 and 80km/h, and risk beliefs. There were neither significant overall differences between aware and unaware participants, (speeding, Fð1;253Þ ¼ 0:66, h2 ¼ :001; danger at 70, Fð1;253Þ ¼ 0:03, h2 ¼ :000; danger at 80, Fð1;253Þ ¼ 0:03, h2 ¼ :000), nor any interactions with T1–T2 differences (speeding, Fð1;253Þ ¼ 0:54, h2 ¼ :002; danger at 70, Fð1;253Þ ¼ 0:10, h2 ¼ :001; danger at 80, Fð1;253Þ ¼ 0:06, h2 ¼ :000). This suggests that patterns of responses amongst unaware participants did not differ from those who reported being aware of the campaign. T1 to T2 correlations were computed to examine temporal consistency of the measures over this period. Speeding (rð253Þ ¼ :51, p , :001) and risk beliefs (rð253Þ ¼ :66, p , :001) showed moderate correlations. Smaller correlations were observed for danger at 70km/h (rð253Þ ¼ :43, p , :001), and danger at 80km/h (x2 ð1Þ ¼ 36:07, f ¼ :37, p , :001).Cross-sectional analyses Table 2 shows cross-sectional Pearson and point-biserial correlation coefficients between all variables. The three risk perception variables were moderately intercorrelated, providing some evidence of criterion validity. Consistent with the hypothesis, speeding was negatively correlated with all risk perception items at both T1 and T2.