Number of questions.

A count was taken of the number of questions asked in the QU phase. Questions which were repeated (if the participant had not understood the question initially) were only scored once and questions clarifying a piece of information given by the participant were not scored (e.g. "So the gun was silver?" when the participant had just provided this information). Where the interviewer was using imagery instructions to focus the interviewee on a particular area (e.g. to describe the gunman's face) these instructions were coded separately as "CI probes". These were often used by some of the interviewers in conjunction with open questions and follow-up questions (closed questions). The following example would be scored as 1 CI probe (the gunman's face), one open question. "Now I want you to concentrate and picture the gunman's face in your mind. You might prefer to close your eyes. Have you got a clear image of the gunman's face? Can you describe the gunman's face?". A closed question was used to get more specific information when a partial description had already been elicited from a prior open question.

The number of CI probes used in the questioning phase of each interview (especially CI) was tallied. (None of the SI or untrained interviews were found to use imagery or context reinstatement spontaneously.) The use of the other instructions (for example, report everything, transfer control) were noted as either being present or absent from the interviews.

Results

Recall Data

A series of one way analyses of variance (ANOVA) was performed on the recall data in order to compare performance across CI (n=22), SI (n=23) and UI (n=21) conditions[4]. The dependent measures were the total number of correct details, errors and confabulations elicited per interviewee. The amount of correct, incorrect and confabulated details will be reported first followed by the proportion of recall that was correct (i.e. accuracy).

(insert table one about here)

(a) Total amount recalled.

As predicted from earlier research on the CI, there was a significant difference between conditions in the total number of correct details recalled (F (2,63)=4.30, p<.05). Post-hoc analyses showed that significantly more correct information was elicited in the CI condition than the UI condition (Fisher PLSD= 11.21, p<.01). Contrary to expectation, however, there was no significant difference between CI and SI condition (Fisher PLSD= 7.19, p=.07). There was also no significant difference between the SI and UI (Fisher PLSD= 3.28, p>.05).

(b) Errors.

The errors followed a similar pattern. There was a significant effect of interview condition (F (2,63)=6.27, p<.01). Post-hoc comparisons showed more errors were produced in the CI condition as compared to the UI (Fisher PLSD= 3.08, p<.001) and more errors in the SI as compared to the UI (Fisher PLSD= 2.01, p<.05). Again the CI and SI did not differ significantly Fisher PLSD= 1.06, p>.05).

(c) Confabulations.

Finally, there was a significant effect of interview condition on confabulations (F (2,63)= 6.14, p<.01). As with the errors, there were more confabulations in the CI condition as compared to the UI (Fisher PLSD= 3.08, p<.01) and more confabulations in the SI as compared to the UI (Fisher PLSD= 2.72, p<.01). There were no significant differences between the CI and SI (p>.05).

(d) Accuracy rates.

The accuracy rates (i.e. number of correct details divided by total number of details) for the CI, SI and untrained group were 84%, 84% and 92% respectively. There was a significant difference between the groups in these accuracy rates (F (2, 63)= 9.85, p<.001). Post-hoc analyses showed the CI and SI interviews were significantly less accurate than the UI interviews (Fisher PLSD= .076, p<.001 and Fisher PLSD= .079, p<.00 respectively) but not different from each other.

To summarise the results so far, performance in the CI and SI conditions did not differ significantly from each other in terms of the number of correct details, errors and confabulations. The CI and SI both produced more errors and confabulations than the UI, and the CI more correct recall than the UI. This is the first study conducted in our laboratory which shows an increase in confabulations with the CI, although a similar effect has been noted in a recent study conducted in a different laboratory (Gwyer, Clifford & Dritschel, 1995). The next section compares performance across interview categories under free recall conditions (see table 1).

(e) Recall in different phases.

In the free recall phase there were no significant differences between conditions in total correct details or errors (Fs <1). There were however significant differences in the total number of confabulations (F (2,63) = 5.75, p<.01) with more confabulations in the CI as compared to the UI (Fisher PLSD = .72, p<.01) and in the SI as compared to the UI (Fisher PLSD=.74, p<.01). The CI and SI were not significantly different. This was unexpected and surprising as there were no questions asked during the free recall phase. This argues that the increased number of confabulations cannot be due to (inappropriate) questioning. It should be pointed out however, that accuracy rates were high in free recall and the number of confabulations was relatively small (approximately 0.5-1.5 confabulations for every 30 units of correct information).

In the questioning phase there was a significant effect of interview condition for total correct details (F (2, 63) = 5.35, p<.01). Post-hoc analyses showed significantly more correct information in the CI as compared to the UI group (Fisher PLSD= 4.32, p<.01). The post-hoc test did not reveal a significant difference between the CI and SI conditions or SI and UI conditions (p>.05). Looking at the number of errors in the QU phase, there was again significant effect of interview condition (F (2, 63)= 8.09, p<.01). Post-hoc analyses showed more errors in the CI as compared to the UI (Fisher PLSD= 1.44, p<.01) and in the SI as compared to the UI (Fisher PLSD=1.43, p<.01). CI and SI did not differ significantly. There was also an effect for confabulations (F (2, 63)= 5.07, p<.01) with more confabulations in the CI as compared to the UI (Fisher PLSD= 1.67, p<.05) and in the SI as compared to the UI (Fisher PLSD=1.65, p<.01). Again, the CI and SI were not different (p>.05). The increase in confabulations in the CI and SI as compared to the UI were unexpected (cf. Memon, Wark, Holley, Bull and Koehnken, 1996b).

(f) Information type.

Breaking down the total correct details into person, action, objects and location information, there were some significant differences between groups in person details. There were significantly more total person items correctly reported in the CI group as compared to the untrained group (F (2, 63)= 4.01, p<.05, Fisher PLSD= 3.91, p<.01). There were no significant CI/ SI differences in the total correct person information. There were significantly more total person errors in the CI group as compared to the UI group (F (2, 63)= 4.01, p<.05, Fisher PLSD= 1.25, p<.01). Again there were no CI vs. SI differences (cf. Memon et al, in press; Milne et al, 1995). Finally, there was an effect of interview type on person confabulations F (2, 63)= 4.81, p<.05). Post-hoc analyses showed a significant difference between CI and UI (Fisher PLSD= 1.34, p<.01) and the SI and UI (Fisher PLSD= 1.33, p<.01) but not the CI and SI (p>.05). These data are presented in Table 2

(insert table 2 about here)