Westerberg, C. E., & Marsolek, C. J. (2003). Sensitivity reductions in false recognition: A measure of false memories with stronger theoretical implications. Journal of Experimental Psychology: Learning, Memory and Cognition, 29, 747-759.
When presented with a list composed of words that all relate to a central non presented theme, participants falsely recall and recognize the word that represents that theme at very high rates. Arguably, signal detection analysis has been used successfully to measure the ability of participants to discriminate between different classes of items, and the bias that participants have to respond old. Of course, it must be stressed that certain assumptions are made by researchers when using signal detection analysis to measure recognition memory performance. First, the researcher must assume that responses on a recognition memory test are based on a single dimension of memory strength. Second, the researcher assumes different types of items are distributed along this memory strength dimension. Third, there is typically an assumption that a single process is used to judge a recognition item as new or old based on it memory strength. Additionally, it is often assumed that each type of item is normally distributed across this dimension and that the variability in each of these distributions is equal. In signal detection theory there are multiple items that are used to measure memory performance. There are targets and lures. A measure of sensitivity informs a person of how easily participants can distinguish between old and new items. When sensitivity is high, participants were very effective at distinguishing between targets and lures. Behaviorally, this is when the difference between HITS for targets and FALSE ALARMS to lures is high. Conceptually, there is an assumption that this difference is due to the distance of these two distributions along the memory strength dimension.
|
Use the graphic taken from the Westerberg article to illustrate this point. The right curve represents targets and the left curve represents lures. The distance between these two distributions is used as a measure of sensitivity. As the two curves move further apart there is less overlap. Take a moment to think about what the overlap between the two curves represents. Also, there is response criterion placed in the figure. As depicted in the figure any recognition item that has a level of strength that is greater than or equal to the response criterion will |
|
be accepted as old. Any item that has a response criterion that is less than the response criterion will be rejected as new. The placement of the response criterion in relationship to the intersect between the two curves determines a person’s bias. If the criterion is placed to the left of the intersection of the two curves, will people accept many items or only a few items? If the criterion is places to the right of the intersection of the two curves, will people accept many items or only a few items? You can also think about these questions in terms of how moving the response criterion in relationship to the intersection will affect the acceptance of targets and lures.
|
Miller & Wolford (1999) observed there to be different biases for the different item types. Critical lures had a more liberal bias, related lures had more a conservative bias, and unrelated lures has the most conservative bias. Westerberg and Marsolek point out that the resulting debate has not allowed for a better theoretical understanding of the DRM effect, because multiple theories can account for the very same data. There are two general types of theories acknowledged in this article, decision-based models and storage-based models |
|
Researchers advocating decision based models assume that participants are actually setting different criteria for the item types. Memory strength is equivalent for all three items, but the response criterion is differentially set depending upon the item type. Notice in the figure above that there is only one new item curve depicted. In actuality it is three identically shaped curves superimposed onto one another. Yet notice that there are three different response criteria in this figure.
|
Researchers advocating storage-based models assume that the distributions of the different items to fall at different points along the memory strength continuum as depicted in the figure to the right. This is because of the level of activation that the different types of items experience at study. Remember that by an activation account all of the DRM list items are activating the critical lure. Not to mention an IAR account. |
|
Westerberg and Marsolek discuss various reasons for why there might be different levels of sensitivity associated with the different types of items. Most notably, they suggest that the global matching component of MINERVA2 would predict decreased sensitivity for distinguishing old and new critical lures, because of the amount of feature overlap that these items have with items stored in memory.
Experiment 1
Participants were presented with 32 DRM lists and 2 lists composed of completely unrelated words (auditory presentation). The test was compose of unrelated words some were old and some were new, related words some were old and some were new, and critical words some were old and some were new. Unrelated words were taken from the 2 lists composed of completely unrelated words. The related words were simply DRM list item. And critical words were the critical non-presented word of the DRM list. Participants made old new and confidence judgments for each of the recognition test items.
The results demonstrated participants to better able to discriminate old unrelated item from new unrelated items and old related (i.e., DRM lists items) from new related items better than old critical items from new critical items. Also you can see in Figure 6, that participants were more biased to respond old to critical items regardless of whether they were old or new than either of the of the other two item types.
Experiment 2 and 3
The same types of item and study presentation were used as in the first Experiment 1. These experiments differed from the first in that a forced choice recognition memory test was used. Participants were presented with one old item and one new item in a pair and they had to decide which of the two had been presented at study. Experiment 3 differed from the second experiment in that the unrelated items were matched to related items in terms of word frequency.
Results
As can be seen in Figure 8 Experiment 2 replicated the results of the first experiment. The third experiment replicated the general findings of the first two experiments. These researchers note that sensitivity for related items tended to be greater than sensitivity for unrelated items in both Experiments 1, 2 and 3.
General Discussion
Neither of the types of models described in the introduction can account for the change in sensitivity observed for the different types of items. First, the decision-based models predict bias to change and not the position of the distribution. Yet, the differences in sensitivity would suggest that the distributions for the three types of items were located at different points along the memory strength continuum. Second, the storage based models cannot account for the trend for related words (i.e., DRM list items) to have higher levels of sensitivity associated with them than the unrelated words.