The finding that children's accuracy decreased during the second interview tells us that children made changes to their direct examination reports. The final phase of our analysis was to examine these changes. Children were considered to have changed a direct examination report if, in the second interview, they received a score of 2 (direct contradiction) or 1 (partial contradiction) for a given question.
The majority of 5- and 6-year-olds (66%) and 42% of 9- and 10-year-olds made at least one change to their earlier responses. All four original responses were changed by 24% of younger children and 10% of older children. A 2 (Age) × 2 (Delay) × 2 (Second Interview Format) between-subjects ANOVA revealed that children in the cross-examination condition made nearly 10 times as many changes (M= 2.42, SE= .15) as children who received a repeated direct examination interview (M= .26, SE= .06), F(1, 177) = 203.28, p < .0001, f= 1.07.
The three-way between-subjects ANOVA also revealed main effects of age and delay, qualified by a significant Age × Delay interaction, F(1, 177) = 7.89, p= .006, f= .21. At the short delay, older (M= 1.04, SE= .21) and younger children (M= 1.24, SE= .24) made similar numbers of changes, t(88) = .65, p= .52, d= .14, power = .75. At the long delay, however, younger children (M= 2.20, SE= .24) made more changes than older children (M= 1.11, SE= .20), t(93) = 3.47, p= .001, d= .72. Delay did not influence the number of changes that older children made, t(101) =−.24, p= .81, d= .05, power = .80, but younger children made more changes after a long delay than after a short delay, t(80) =−2.81, p= .006, d= .44.
Because children's response changes could be directed towards or away from the truth, our next step was to examine the changes as a function of original response accuracy. The proportion of correct direct examination responses changed was examined in a 2 (Age) × 2 (Delay) × 2 (Second Interview Format) between-subjects ANOVA. As expected, given that children in the cross-examination condition changed nearly 10 times as many of their direct examination responses as children in the repeated direct examination condition, a main effect of second interview format was revealed. Children in the cross-examination condition changed a greater proportion of their correct responses (M= .60, SE= .04) than children in the repeated direct examination condition (M= .06, SE= .02), F(1, 177) = 201.69, p < .0001, f= 1.07. Significant main effects of age and delay were qualified by an Age x Delay interaction, F(1, 177) = 10.07, p= .002, f= .24. At the short delay, younger (M= .30, SE= .06) and older (M= .28, SE= .05) children changed a similar proportion of correct responses, t(88) = .27, p= .79, d= .06, power = .75. At the long delay, however, younger children (M= .53, SE= .06) changed a greater proportion of correct responses than older children (M= .26, SE= .05), t(93) = 3.50, p= .001, d= .73. The proportion of correct responses changed by older children did not differ as a function of delay, t(101) = .23, p= .82, d= .05, power = .80; younger children changed a greater proportion of correct responses after a long delay than after a short delay, t(80) =−2.72, p= .008, d= .42.
Of course, children could also change incorrect direct examination responses during the second interview. The proportion of incorrect direct examination responses that children changed was therefore submitted to a similar ANOVA. Children in the cross-examination condition changed a greater proportion of incorrect responses (M= .77, SE= .08) than children in the repeated direct examination condition (M= .26, SE= .10), F(1, 37) = 10.89, p= .002, f= .54. No other significant main effects nor interactions were revealed (non-significant Fs ≤ 3.49, ps = .07–.86, fs = .03–.31, power = .11–.36), although there was a trend for younger children (M= .65, SE= .09) to change a greater proportion of incorrect responses than older children (M= .36, SE= .13), F(1, 37) = 3.49, p= .07, f= .31, power = .33. In conducting these analyses, we were only able to consider children who had made errors in their direct examination reports (n= 45); as such, power values were low.
Were incorrect responses more likely to be changed than correct ones? To answer this question, separate 2 (Direct Examination Response: inaccurate, accurate) x 2 (Second Interview Response: change, no change) chi-square contingency tests were conducted for younger and older children, as a function of delay and second interview format. Given that jurors may look unfavourably on witnesses who lack confidence (Brennan & Brennan, 1988; Brewer & Burke, 2002; Park, 2003), I don't know responses during the direct examination interview were considered to be inaccurate for the purpose of these analyses. Recall that there were very few of these types of responses for both 5- and 6-year-olds (n= 4, 1.2%) and 9- and 10-year-olds (n= 3, 0.7%). For younger children undergoing cross-examination, accurate responses were just as likely as inaccurate responses to be changed, regardless of whether the delay was short (χ2 (1, N= 84) = .58, Fisher's exact p= .50, w= .08, power = .78) or long (χ2 (1, N= 84) = .70, Fisher's exact p= .68, w= .09, power = .78). Similar findings were observed for the repeated direct examination interview (short delay, χ2 (1, N= 80) = 1.20, Fisher's exact p= .33, w= .12, power = .76; long delay, χ2 (1, N= 80) = .25, Fisher's exact p= .69, w= .06, power = .76). Older children showed the same pattern of findings for the cross-examination interview (short delay, χ2 (1, N= 100) = .001, Fisher's exact p= 1.00, w= .003, power = .85; long delay, χ2 (1, N= 108) = .09, Fisher's exact p= 1.00, w= .03, power = .85), and the repeated direct examination interview after a long delay, χ2 (1, N= 108) = .03, Fisher's exact p= 1.00, w= .02, power = .87. None of the older children changed their responses when re-interviewed with the direct examination questions at the short delay. Given that children in some jurisdictions are actively encouraged to say I don't know when they are uncertain of an answer (e.g., Morgan Libeau et al., 2003), the data were reanalysed with I don't know responses coded as accurate. No changes in the direction of findings were observed. Taken together, these findings suggest that children's accurate responses were just as susceptible to change following a second interview as inaccurate responses. Neither of the second interview formats appeared to differentially target inaccurate over accurate responses.
In summary, children performed well in the direct examination interview, but made numerous changes to their original responses during the second interview, to the detriment of their accuracy. While age and delay played a role in the number of response changes that children made and their overall accuracy, the format of the second interview exerted the greatest effect. The cross-examination interview was considerably more detrimental to children's reports than the repeated direct examination interview.