This study examines when prior knowledge becomes available during retrieval and how episodic memory and prior knowledge interact in source monitoring

Source monitoring relies on characteristic qualities of episodes (eg. Perceptual) & prior knowledge (source bias/schemas) to make recognition judgements. Bayen et. al (2001) illustrated in their multinomial model that participants can use the schemas in source monitoring as a bias (guessing) as to whether an item was presented at test. The current study examines if episodic or source bias occurs at separate times and rely on different processes.

Multinomial Models of Source Monitoring: Source monitoring tasks depend on recognition of old/new items, attributing a source for that item, and various response biases. Since all of these processes contribute to a response, these different processes are able to be observed using a multinomial processing tree (MPT). MPT models allow examination of the item-recognition parameters, source memory parameters, and guessing bias parameters. The current study uses the 2HT (two high threshold) model because it allows one to examine the independent measures of old-new recognition and also source for a memory. (See figure 1). In this study, Spaniol & Bayen performed time-course analyses based on these parameters for multiple response times to examine the retrieval and use of schematic knowledge in source monitoring.

Techniques used to manipulate response times for parameters in MPT model:

Response-Signal Technique (speed accuracy tradeoff): This study looks at the time-course function for old/new recognition and source guessing bias. Spaniol & Bayen want to look at the shape of the time-course function for schema based source guessing bias. In other words, they are looking at how the amount of response time influences old/new item recognition and bias (guessing) for response by manipulating response times using 7 lags.

Probability-Matching Theory (looks at how often participants use bias (guessing) in source monitoring tasks): PMT looks at the how participants adjust their source bias to the perceived ratio of target items that were presented by the source for which they are expected to target items that were presented by the source for which they are somewhat unexpected.

Experiment 1 examines how response time influences bias in source monitoring and provides evidence for whether schematic knowledge is automatically (heuristically) or systematically (controlled) used upon retrieval. They manipulate source, expectancy for items, & response-signal lag for parameter estimates of item memory and source bias. Spaniol & Bayen predicted that with more response time participants would have more accurate item memory and that the time-course curve of source bias would show similar form as item memory. In other words as item memory increases, bias or guessing should decrease. If the source bias intercepts is smaller or equal to the item recognition, schematic information can be viewed as automatic, and if reversed effects occur it can be assumed that schematic information is controlled.

Experiment 1

Participants:

Only right handed, native English speakers who had never taken a general psychology course and who were not color blind were chosen for a total of 18 subjects.

Design: 2(expectancy of source A and Source B items) x 2(actual source) x 7 (response-signal lag) manipulated within subjects.

Materials: Participants given 8 study-test blocks. In each block the item was presented by Source A or Source B which were identified by descriptors by showing a picture of the source and the name of the source (sportscaster and weatherman) which

were similar thematically (television news). At test the participants asked to identify the source the item was presented. Items chosen to study were either strongly associated with source A and weakly associated with Source B, or vice versa & low semantic ambiguity and low semantic and orthographic similarity to other experimental items.

Procedure: Participants were ran separately and completed three sessions. 20-30 hours passed between each session. In the first session, participants completed practice task and two study-test blocks, and at the second and third session completed three study-test blocks. Each study phase had one presentation of an 84-word study list. Participants were shown two sources and each source said a word and participants were told to try and remember the words as best as they could. Half the words were presented with the expected source and the other half were presented with the unexpected source. After studying the words and sources, participants were given a task to eliminate recency effects. Immediately after the task participants were tested.

At test the screen reinstated the sources. (Phil is a weatherman and Jack is a sports reporter). When a word appeared, participants told to respond which source the word had been presented with or if it was neither source. Participants were told to respond as quickly, but as accurately as possible. Again, participants were told that if their response was made in perfect time they would be rewarded financially. Participants were told if they were to fast or slow when answering.

Results:

Multinomial Modeling

Participants who did not respond within 75-500 ms were eliminated (4%). Spaniol & Bayen pooled the data by coding responses according to whether the source was expected, somewhat unexpected, or new. The 2HT model was not identifiable so they fitted Sub model 4. Model 4 makes 3 assumptions.

1. Item memory is equal for items presented by both sources and new items.

2. Source memory is equal for items shown by both sources.

3. The probability of guessing for either Source A or B is equal for recognized and unrecognized items.

Time-Course Modeling

Spaniol & Bayen averaged the response time for each signal lag and used the following exponential growth model to determine the relationship between response time and item memory and the relationship between response time and source bias for each individual.

There was great individual variety in parameter estimates and goodness of fit. See figure 2. Ten of the participants did not show an increase in source bias over time, whereas 7 participants demonstrated significant increase of source bias over time. Results indicated that retrieval of prior knowledge seen in source bias is slower than episodic information. Since the data from this experiment includes a small sample, experiment 2 increase the sample size to see if the difference of time retrieval between prior knowledge and episodic memory is reliable. Some participants tried to be accurate and others did not as so Experiment 2 manipulates accuracy and memory performance to look at response bias.

Experiment 2

Experiment 2 manipulates response time, number of study presentations, and accuracy payoffs. Since more presentations of studied items should increase episodic memory and encouraging subjects to be accurate, a stronger relationship should ensue between bias and episodic memory.

There were 3 conditions in this experiment:

Condition 1. Same as Experiment 1 (subjects paid for time response, not accuracy).

Condition 2. Same as Experiment 1 but study items were presented twice.

Condition 3. Items presented twice and rewards were only given if subjects answered within the given time & were accurate.

Participants: 18 total participants were in this study.

Design: Spaniol & Bayen added a between subjects factor to manipulate memory performance. There was study presentation without emphasis on accuracy; two presentations without emphasizing accuracy; Two study presentations and emphasis on accuracy; & source expectancy for items, actual source of items, and response signal lag as within subject variables.

Materials and Procedure: Same materials as in Experiment 1. Group 2 & 3 differed from Group 1 in procedures (group 1 was the same as in Experiment 1).

Group 2: Each study word presented twice in study phase of each block. No reward for accuracy, but rewarded for response times

Group 3: Each study word shown twice as in group two but participants were rewarded based on correct response time and accuracy.

Results: Responses that did not occur within 75-500ms were excluded (4%). In 8 cases, results did not fit the model but were included in results because of chance.  Individual responses varied as in Experiment 1. Again the exponential model was shown  To be superior to the simple linear model when describing the time course of source bias in 7 participants.

Comparing Item Memory and Source Bias Time Course:

Spaniol and Bayen pooled data from 12 participants who showed source bias separately for each signal lag. Item memory was above baseline at the shortest lag suggesting that  item memory occurs before source bias.

General Discussion:

When participants had low memory for lists they responded using source bias which is consitant with probability matching hypothesis.

Episodic retrieval occurs before use of semantic retrieval (source bias) as illustrated with the multinomial model parameters..