This study examines the phenomenon of imagination inflation.  The term was first used in a study by Garry, Manning, Loftus & Sherman (1996).  In that original study participants filled out a questionnaire called the Life Events Inventory (LEI).  It asked participants to rate their confidence that various events happened during their childhood (e.g. found a $10 bill in a parking lot, etc.).  A couple of weeks later participants engaged in an imagination session in which they imagined some of the events from the LEI.  Participants then took the LEI a second time under the cover story that the investigator had lost their original questionnaire.  Garry et al. reported that participants scores were more likely to increase than to decrease for the imagined events.

Garry et al. reported data only for those events that were initially given low confidence ratings.  This makes sense in that the study was aimed at the effect of imagination on confidence for events that did not occur.  However, Pezdek and Eddy contend that the results could be merely due to regression towards the mean.  They point to two pieces of empirical evidence from Garry et al.'s own study:

• First, they show that LEI ratings were higher at time two for both imagined and non-imagined events.  Figure 1 shows that the imagined events appeared somewhat more likely to increase than the non-imagined events, but Pezdek and Eddy argue that inferential tests were not performed to confirm this.
• Second, they obtained the original data from Garry et al. and used it to show that if one looks at events that were initially given high LEI ratings, the imagined events appeared to be less likely to increase than the non-imagined events.  This is shown in Figure 2.

Methods

• A group of elderly subjects and college students completed an initial LEI.
• Four of the events on the LEI were critical events.
• Two weeks later participants imagined two of the critical events and did not imagine the other two.
• After the imagination session participants completed the LEI a second time.

Items With Initially Low LEI scores.  Pezdek and Eddy first analyzed the results exactly the way Garry et al. did.  That is, by looking only at those events initially given low LEI ratings and indicating the proportion that increased, decreased or stayed the same.  Keep in mind, it is the items that were initially given that low ratings that Pezdek and Eddy believe could be subject to regression towards the mean. So both Pezdek and Eddy and Garry et al. would predict that the LEI scores should be more likely to increase than to decrease.

• Most LEI ratings did not change.  However when they did change they tended to increase.  This was true for both the imagined items and the non-imagined items (although it looks like the imagined items were more likely to increase than the non-imagined)

• The most common thing to happen was for LEI ratings to decrease from time 1 to time 2 (consistent with regression towards the mean).

Score extremity.  Pezdek and Eddy performed a more fine grained analysis by looking at those items with initial scores of 1-2, 3-4, 5-6, and 7-8.  The regression argument holds that the more extreme scores (1-2 and 7-8) should show larger changes from time 1 to time 2.  This prediction was confirmed.

Tests of Significance.  In Garry et al.'s original study they used an item level analysis as their inferential test.  Remember that each critical item was imagined by some subjects and not imagined by other subjects.  And in Garry et al.'s original study there were 8 critical items.  So what Garry et al. did was for each item:

• Find the cases where the item was initially given a low rating (1-4)
• Determined the proportion of participants whose LEI score increased when the item was imagined.
• Determined the proportion of participants whose LEI score increased when the item was not imagined.
• So for each of the eight items there was a score for when the item was imagined and when it wasn't imagined.  Garry et al. performed a paired t test on these scores.  The result was that the proportion of participants whose scores increased was significantly higher when the items were imagined than when they were not imagined.

Pezdek and Eddy acknowledge that this suggests that imagination inflation is going on and not merely regression towards the mean.  However, they argue that an analysis that looks at the size of the change produces different results.

Size of the Change.  Pezdek and Eddy next looked at the size of the change not just whether or not there was a change.  First they looked at the items initially given the lower ratings.  Here's what they found:

• Imagined events increased by more than non-imagined events.
• Although there was an increase the means for the imagined events tended to stay on the same side of the scale (i.e. 1-4).
• Most subjects whose scores changed only changed by a little bit.
• Changes of 3 or more points was more likely for non-imagined than imagined events.
• Figure 3 shows that for items initially rated 1-4 ratings tended to increase for all three types of items and that for items initially rated 5-8 ratings tended to decrease.

Regression

In what Pezdek and Eddy say is their strongest test they use a regression analysis.  This analysis was done only for subjects whose first LEI score was below the mean LEI score.   Here's the logic of their argument.

• First, look at the items in the control condition and figure out what the prediction line is relating LEI scores at time 1 and time 2.
• Next compare this prediction line for the control items with the observed data in the imagined condition.
• If imagination inflation is going on then the prediction line should tend to under-estimate the observed data (Because the increased LEI ratings aren't merely due to regression toward the mean).
• If all that is going on is regression toward the mean then the prediction line should be just as likely to over-estimate as to under-estimate the observed data.
• Pezdek and Eddy looked at the mean residual difference (the difference between the prediction line and the observed values) and found that it was not significantly different from zero.  Hence, they concluded that all that's going on is regression toward the mean.

Pezdek and Eddy argue that the regression toward the mean account ant the imagination inflation account can be distinguished in four ways.

• First they argue that imagination should increase LEI ratings for items that are initially given high ratings as well as items that are initially given low ratings.  But this does not appear to happen.
• Second, they argue that there should be an interaction between condition (imagined, not imagined) and time (time 1 and time 2).  But the ANOVA did not reveal such an interaction.
• Third, they argue that the change in LEI scores should be greater for older adults than younger adults, but this was not the case.
• Fourth, in the regression analysis, there should be an effect of imagination over and above what you'd predict based on regression towards the mean.  But the mean residuals were not significantly greater than zero.