Smith, R. E., & Bayen,
U. J. (2004). A multinomial model of event-based
prospective memory. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 30, 756-777.
Prospective
memory is remembering to do something in the future. The current article presents and tests a
multinomial model designed to explain some facets of prospective memory.
The model
is designed to separate the prospective and retrospective components of PM to
evaluate the experimental influence on each.
This particular model is based heavily on the process and memory process
(PAM) theory of PM proposed by Smith (2003).
PAM proposes that processes used for the prospective component are
active and not automatic. The current
model is based on a setup that has a ongoing task with
two choices and two types of trials.
Experiment One
Methods
This first
experiment manipulated the prospective task importance through two groups. The first group, the Prospective Memory
Importance (PMI) group, was told that the prospective task was more
important. The Color-Matching Importance
(CMI) group was told that the ongoing task was more important. The ongoing color-matching task involved
responding to the color of a word by reporting if that color had been
previously presented.
Results and Discussion
The PMI
group showed a higher prospective hit rate, than did the CMI group (Table
1). There was no interaction with the
trial type. The model showed a
significant fit. The parameter estimate
of P was significantly greater in the PMI group than the CMI group, as would
follow from the above results. This was
expected, as the importance on the prospective task should affect the
processing for said task. For the ongoing
task, reaction times increased significantly for the PMI group over the CMI
group, while both group showed a slowing versus a non-prospective control. Additionally there was a significant positive
correlation between response time and PM accuracy in both the CMI and PMI
conditions. These two results suggest
that there is some type of processing going on in the PM tasks.
Experiment Two
Methods
This
experiment followed the same basic setup, and added a target similarity
manipulation. This manipulation used
targets that were similar to the nontarget words or targets that were dissimilar
to the nontarget words.
Results and Discussion
There was a significant
difference in PM accuracy for the similarity type, with similar target
facilitating PM accuracy. As in
experiment one, PM accuracy was higher in the PMI condition than in the CMI
condition. Again the model showed a good
fit to the data. The model correctly
predicted an increase in processing in the similar target condition (P), and it
also predicted an increase in contribution of the retrospective component in
the dissimilar condition. The response times were affected by both the task
importance and the similarity of targets.
Experiment Three
Methods
The
importance instruction was dropped for this experiment and a study time
manipulation for the prospective targets was added. The targets were studied for either 5 seconds
or 20 seconds each (between subjects).
The same ongoing task was used as in experiments one and two.
Results and Discussion
Those in the 20 second condition
were better able to identify prospective targets on the test. There was also a significant difference in
the same direction in a posttest recall of the targets. Once again, the model fit the data
significantly for a small effect size, and as predicted, the time manipulation
only affected the M parameter in the model.
There were no significant effects on the ongoing task in this
experiment.
Experiment Four
Methods
This final
experiment used the same manipulation as experiment three, but utilized a new
methodology. Additionally, there were no
response time limitation placed on the subjects to allow for concerns over the
PAM explanation to be addressed (i.e. PAM is only used when participants must
respond quickly).
Results and Discussion
The long
encoding group did show significantly higher PM accuracy and target recall than
the short encoding group. A new model
was added to account for the additional response available in the ongoing
task. Both models fit the data. In both models the M parameter was affected,
but no other parameter was affected, just as in experiment three. Likewise, both models showed better
target/nontarget discrimination in the long encoding group than the short
encoding group. Comparisons
between experiments three and four showed similar results.
General Discussion
Since this model was based on PAM theory, it might be inherently flawed if there are automatized processes going on in prospective memory. To check this idea, the P parameter was set to a constant of 1. In this case the model only fit in the specific case of the PIM similar target group in experiment two. This leads to the conclusion that the PM tasks in these experiments were not being completed through some automatic process. However, the argument for the other side could be that these particular designs only create PM tasks that need monitoring for completion. This type of modeling might help to further analyze the issue of monitoring in prospective memory.