PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Memory. Author manuscript; available in PMC 2011 April 1.
Published in final edited form as:
Memory. 2010 April; 18(3): 272–283.
Published online 2010 February 25. doi:  10.1080/09658211003601506
PMCID: PMC2858259
NIHMSID: NIHMS181409

The Bias for a Recognition Judgment Depends on the Response Emitted in a Prior Recognition Judgment

Abstract

On each trial of the experimental procedure, the participant read a list of words and made successive recognition judgments to multiple test words. The bias for a given recognition judgment was more conservative if the judgment followed a correct positive response to a target than if it followed a correct negative response to a lure. Similar results were not observed for successive semantic recognition judgments. The bias shift was greater when the study list was short than when the list was long. The results suggest that participants in a recognition task have a sense of the size of the set of targets that might possibly be presented on the next trial, and that, under conditions in which a word can only be presented once during the test phase, their bias becomes more conservative after a positive response to a target because the set is depleted.

Keywords: recognition, response bias, sequential effects

We study memory by observing its behavioral manifestations. We infer that a person remembers a previous experience if the person’s behavior implies this to be the case. Certain means of studying memory are appealing because the behavioral indices of remembering are relatively transparent. In the recognition task, for example, which will be of interest here, the participant makes a simple response to indicate whether or not the test item has been encountered previously. Several kinds of recognition task have been explored in past work. In the semantic recognition task, the participant indicates whether the test item has been encountered over the course of a lifetime. In the episodic recognition task, which will be the primary focus here, the participant indicates whether the test item has been encountered during an earlier phase of the experiment.

Notwithstanding the simplicity of the recognition response, the behavior of a participant in a recognition task is probably not governed solely by mnemonic factors. Assuming that the test items vary in the degree to which they evoke evidence of prior encounter, how does the participant decide, for a given item, whether to emit a positive or negative response? The decision must reflect, not only the mnemonic evidence evoked by the item, but also a certain bias, where this is understood as a greater or lesser tendency to make a positive recognition decision/response to an item that evokes a given degree of mnemonic evidence. In the present study we explored a phenomenon that may reflect a pattern of bias in episodic recognition.

Numerous factors probably influence bias in episodic recognition. Several of these factors have been more or less conclusively identified in previous work. Bias becomes more liberal with increases in the benefit of a correct positive response and decreases in the cost of an incorrect positive response (Healy & Kubovy, 1978; Smith, 1969; Zimmerman & Kimble, 1973). Bias becomes more liberal with increases in the proportion of test items that are targets (i.e. that belong to the memory set) (Estes & Maddox, 1995; Healy & Kubovy, 1978). Finally, bias becomes more conservative with increases in the similarity of test items that are lures (i.e. that do not belong to the memory set) to test items that are targets (Brown & Steyvers, 2005, Brown, Steyvers, & Hemmer, 2007).

Additional factors may influence bias in episodic recognition. Of particular interest here is sequential context, which appears to influence bias in several tasks that, like the recognition task, require a simple binary response. According to some proposals, sequential context influences bias in detection tasks. Data from such tasks reveal both positive and negative sequential effects, where, a positive sequential effect occurs if positive responses are facilitated following positive responses and a negative sequential effect occurs if positive responses are facilitated following negative responses. In some accounts, these sequential effects reflect patterns of bias. For example, Treisman and Williams (1984) propose that the response in a psychophysical task is based on a criterion, which reflects, on one hand, a long term reference value, and, on the other hand, traces of recent responses and stimuli. Traces of the participant’s most recent responses provide the best evidence of the current state of the world and give rise to the positive sequential effect. Traces of the stimuli that have been most recently presented stabilize the criterion and give rise to the negative sequential effect.

Sequential context also may also influence bias in two-choice reaction time tasks. Data from such tasks reveal both positive and negative sequential effects, where a positive sequential effect occurs, for example, when a right-handed response is facilitated if the right-handed stimulus occurred on the previous trial, and a negative sequential effect occurs if the right-handed response is facilitated when the left-handed stimulus occurred on the previous trial. The polarity of the sequential effect in a given context depends on the temporal interval between the emission of the response on the previous trial and the presentation of the stimulus for the current trial (the RSI), such that positive sequential effects are more likely with short RSI’s and negative sequential effects are more likely with long RSI’s (Kirby, 1980). Whereas positive sequential effects in the two-choice task have generally been attributed to a sensory process, negative sequential effects have been attributed to bias effects. Participants are said to be predisposed toward alternating responses, such that, for example, the bias for the left handed response increases following a right handed response. This predisposition is thought to reflect the fact that participants expect more alternations than repetitions in a sequence - a version of the gamblers’ fallacy (Soetens, 1998).

Less is known about the effect of sequential context on bias in recognition tasks. In this study, we explored a phenomenon that may reflect an effect of sequential context on bias in episodic recognition. The phenomenon occurs in a task in which the participant studies a list of items and then makes recognition judgments, in succession, to multiple items. Analyses of the hit- and false-alarm-rate data for this task suggest that the bias for the current test item is more conservative if the previous item was a target than if the previous item was a lure.

Here we report three studies exploring this effect of sequential context on bias in recognition memory. An initial experiment demonstrated the bias shift phenomenon. Two further experiments sought support for an account according to which the phenomenon reflects an inference regarding the number of targets that might potentially be presented as the test item. This account proposes that, upon apprehending the preceding item to be a target, the participant infers that fewer targets remain that might be presented as the test item; the participant’s bias consequently becomes more conservative.

Experiment 1

In this experiment we sought to show that, when a participant makes recognition judgments to words in succession, the participant takes the list status of the preceding word into consideration in setting the bias for the recognition judgment to the current test word. The experiment consisted of a series of 72 trials, with each trial comprising a study phase, in which a list of 12 words was presented, and a test phase, in which recognition judgments were made with respect to a single pair of words – a preceding word and a test word. In the Preceding Word Target and Test Word Target conditions, the preceding word and test word, respectively, were targets. In the Preceding Word Lure and Test Word Lure conditions, the preceding word and test word, respectively, were lures. The goal of the experiment was to show that the bias for the recognition judgment to the test word depended on the list status of the preceding word.

We used a response signal procedure in an attempt to clarify and strengthen the effect of sequential context. The goal in using this procedure was to hold response time more or less constant and to concentrate the effect of sequential context in the hit- and false-alarm-rate data. The goal was not to observe responses based on early states of processing as has sometimes been the case with this procedure. Thus, we set the response interval to match the mean response time that we had observed in a preliminary version of the task, in which participants were instructed to respond to the test word as quickly as possible, without sacrificing accuracy.

Method

Participants

The participants were 106 students from psychology classes at the George Washington University. They participated in fulfillment of a course requirement.

Design

Preceding Word (Target/Lure), and Test Word (Target/Lure) were manipulated within participants. 18 preceding word - test word pairs were assigned to each of the four conditions that occurred when Preceding Word and Test Word were crossed.

Materials

Half of the 18 lists for each condition were composed of proper nouns and half were composed of common nouns. The proper and common noun lists were created by sampling without replacement from sets of 612 proper and 612 common nouns, respectively. On average, the words in the proper and common noun sets were composed of, respectively, 5.65 and 4.67 letters, and 2.02 and 1.33 syllables. Their Kucera and Francis (1967) frequencies were, respectively, 23.13 and 72.15.

For a given trial in the Preceding Word Target and Test Word Target conditions, the preceding word and the test word, respectively, were sampled from the list for that trial. To guard against primacy effects, the test word was never the first, second or third word in the list. For a given trial in the Preceding Word Lure and Test Word Lure conditions, the preceding word and the test word, respectively, were sampled without replacement from either the proper or common noun word set, as was appropriate given the type of words in the list for that trial.

Procedure

The lists were presented on a computer monitor. The participant started the presentation of each list by pressing the space bar of the computer. The list was presented word by word, with each successive word appearing alone in the middle of the screen, in lower-case letters, for 1000 ms. To guard against recency effects, a message appeared at the top of the screen, after the last word of the list disappeared, instructing the participant to count backwards from a randomly-selected number. The message remained on the screen for 3000 ms. After this interval had elapsed, the message disappeared and a message appeared instructing the participant to press the space bar for the test. When the participant pressed the space bar, the preceding word appeared, in upper-case letters, at the top of the screen. The word remained on the screen until the participant responded, at which point the test word appeared, in upper-case letters, where the preceding word had been. Seven hundred ms after the test word appeared, a row of asterisks appeared below the test word. The asterisks remained on the screen, along with the test word, until the participant responded. The participant pressed the “B” and “N” keys to indicate positive and negative responses, respectively. The participant was instructed to respond to the preceding word as quickly as possible, without sacrificing accuracy, and to respond to the test word when the asterisks appeared.

After the participant responded to the test word, the test word and asterisks disappeared and feedback was presented on response speed: if the participant responded before the asterisks appeared, the message “Too Fast” was presented; if the participant responded more than 250 ms after the asterisks appeared, the message “Too Slow” was presented. The response-speed feedback remained on the screen until the participant pressed the space bar of the computer. After the response speed feedback, feedback was presented regarding the accuracy of the responses to the preceding word and the test word: if the response to the preceding word and/or the test word was incorrect, a message was presented to this effect. This message remained on the screen until the participant pressed the space bar of the computer. We presented feedback on response accuracy because such feedback has been shown in other contexts to be crucial to shifts in bias (Estes & Maddox, 1995; Rhodes & Jacoby, 2007).

Results and discussion

Five participants with particularly low levels of accuracy (more then 66% errors in at least one of the conditions) were removed from the sample (the criterion was taken from our previous work in this area and was used throughout the project; excluding these low-performing participants does not substantially change the pattern of the results). In the remaining sample, recognition responses to the preceding word were correct 80% of the time. The same results were observed with proper and common noun lists, so the results for the two list types were collapsed.

We wished to show that the participant took the list status of the preceding word into consideration in setting bias for the recognition judgment to the test word. We reasoned that it was easiest to show this for the case in which the participant correctly apprehended the list status of the preceding word. Our best evidence regarding the participant’s apprehension of the preceding word’s list status was the recognition response that the participant emitted to this word. We consequently confined our analysis of the data for the test word to responses that followed correct responses to the preceding word1. We analyzed the hit- and false-alarm-rate data for these responses as a function of the list status of the preceding word and the test word. Figure 1 presents the results of this analysis.

Figure 1
Experiment 1: Hit- and false-alarm rates for responses to the test word following correct responses to the preceding word, broken down as a function of the list status of the preceding word and the test word.

We used the equal-variance Gaussian signal detection model to compute measures of d’ and c for the different levels of Preceding Word (We used the correction procedure of Snodgrass and Corwin (1988) to compute hits and false alarm rates throughout the project.) We observed no differences in d’ across the Preceding Word Target and Preceding Word Lure conditions, F(1,100) < 1. In contrast, we found that c was higher in the Preceding Word Target than in the Preceding Word Lure condition: .19 vs. -.13, F(1,100) = 9.69, MSe = .54, g = .432

The results of the experiment imply that a participant in our recognition task takes the list status of the preceding word into consideration in setting the bias for the recognition judgment to the current test word. These results suggest that the bias for the participant’s recognition judgment to the test word is more conservative when the preceding word is a target than when it is a lure. Similar results were observed in two generalizing experiments. On each of the 8 trials of the first experiment, participants read a list of 32 words and made recognition judgments to a series of 32 preceding word – test word pairs. On each of the 36 trials of the second experiment, participants read a list of 12 words and made recognition judgments to an unstructured series of 12 words (the words in the test series were grouped for purposes of analysis into preceding word – test word pairs). Additional detail regarding these generalization experiments is available from the authors.

Experiment 2

The results of Experiment 1 suggest that the bias for the recognition judgment to the current test item in an episodic recognition task depends on the list status of the preceding item. Why does bias vary in this way? According to the Set Depletion Hypothesis, the phenomenon reflects an (unconscious) calculation regarding the likelihood of the test item’s being a target. The calculation starts from the observation that items are never presented more than once during the test phase of the recognition task. Given this observation, the participant infers, upon apprehending the preceding item to be a target, that fewer items from the list remain that might possibly be presented as the test item, and that the likelihood of the test item’s being a target is accordingly reduced. The participant consequently shifts to a more conservative bias. The key idea here is that the likelihood assessment is based on a mental representation of the set of potential targets. When the participant apprehends the preceding item to be a target, this item is removed from the set. The inferred likelihood of the test word’s being a target is therefore reduced.

Experiment 2 sought support for the Set Depletion Hypothesis. The experiment asked whether the present bias shift occurs in semantic recognition. According to the hypothesis, the effect should not be observed in semantic recognition, because the set of potential targets is too large to be depleted. Previous work does not tell us definitively whether the present bias shift occurs in semantic recognition. Although previous studies have explored sequential effects in lexical decision, the results of these studies do not speak directly to the question at issue here - in some cases because a response was not required to the preceding item (Ratcliff & McKoon, 1995), in other cases because the sequential context was more complex than in the present study (McNamara, 1992), and in other cases because factors other than sequential context were also manipulated (Perea & Carreiras, 2003). Using the same basic approach as Experiment 1, Experiment 2 asked whether the present bias shift occurs in semantic recognition.

The stimulus materials for the experiment were names of people who were famous (W. Shakespeare), names of people who were slightly famous (L. Cranach), and names of people who were not famous (T. Grinder). On each of the 72 trials in the Episodic condition, participants read a 12-item list consisting primarily of slightly famous names and then made recognition judgments to a preceding name and a test name, with the preceding name and test name being famous on half of the trials and non-famous on half of the trials. In the Preceding Name Target and Test Name Target conditions, the preceding name and the test name, respectively, were present in the list. In the Preceding Name Lure and Test Name Lure conditions, the preceding name and the test name, respectively, were absent from the list. On each of the 72 trials in the Semantic condition, participants indicated, for a preceding name and a test name, whether it was associated with a famous person. In the Preceding Name Target and Test Name Target conditions, the preceding name and the test name, respectively, were famous. In the Preceding Name Lure and Test Name Lure conditions, the preceding name and the test name, respectively, were non-famous.

We expected the bias for the recognition judgment to the test name to depend on the list status of the preceding name in the Episodic condition. The goal of the experiment was to find out whether the bias for the recognition judgment to the test name depended on the polarity of the preceding name in the Semantic condition.

Method

Participants

The 70 participants (35 in the Episodic and 35 in the Semantic condition) were drawn from the same population as the participants of Experiment 1.

Design

Task (Episodic/Semantic) was manipulated between participants. Preceding Name (Target/Lure) and Test Name (Target/Lure) were manipulated within participants. In addition, within the Episodic condition, Preceding/Test Name Type (Famous/Non-famous) was manipulated within participants. For a given participant in the Episodic condition, 9 trials were assigned to each of the eight conditions that occurred when the two levels of Preceding Name, Test Name, and Preceding/Test Name Type were crossed. For a given participant in the Semantic condition, 18 trials were assigned to each of the four conditions that occurred when the two levels of Preceding Name and Test Name were crossed.

Materials

At the base of the materials set were 72 famous names, 792 slightly famous names, and 72 non-famous names. All names consisted of a first initial and a last name. All last names were two syllables long. The materials for a given trial in the Episodic condition were constructed as follows: 1) Depending on the assignment of the trial, the preceding name and test name were sampled without replacement from the famous or non-famous name set. 2) The preceding name and test name were placed in the list for the trial if the trial was assigned to the Preceding Name Target and Test Name Target condition, respectively. To guard against primacy and recency effects, the test name was not placed in the first or last three places in the list. 3) The rest of the words in the list were sampled without replacement from the set of slightly famous names.

The materials for a given trial in the Semantic condition were constructed as follows: The preceding name and test name were sampled from the list of famous names if the trial was assigned, respectively, to the Preceding Name Target and Test Name Target conditions and from the list of non-famous names if the trial was assigned, respectively, to the Preceding Name Lure and Test Name Lure conditions.

Procedure

The procedure for the Episodic condition was the same as for Experiment 1 except that participants did not count backward before being tested on the list for a given trial (this change was made so that the procedures for the Episodic and Semantic conditions would be as similar as possible). The procedure for the Semantic condition was the same as for the Episodic condition except that no list was presented and participants were instructed to indicate, for each name, whether it belonged to a famous person.

Results and discussion

We will consider the results for the Episodic and Semantic conditions in turn. Five participants in the Episodic condition did not reach the accuracy criterion and were removed from the sample. In the remaining sample, recognition responses to the preceding name were correct 74% of the time. Figure 2 presents the hit- and false-alarm-rate data in this sample for responses that followed correct responses to the preceding name. The same results were observed with famous and non-famous test names, so the results for the two types of test name were collapsed.

Figure 2
Experiment 2, Episodic Condition: Hit- and false-alarm rates for responses to the test name following correct responses to the preceding name, broken down as a function of the list status of the preceding name and the test name.

Whereas d’ did not differ in the Preceding Name Target and Preceding Name Lure conditions, F(1,29) = 1.45, MSe = .824, c was greater in the Preceding Name Target (.60) than in the Preceding Name Lure (-.11) condition, F(1,29) = 4.96, MSe = 1.52, g = .48.

Five participants in the Semantic condition did not reach the accuracy criterion and were removed from the sample. In the remaining sample, recognition responses to the preceding name were correct 76% of the time. Figure 3 presents the hit- and false-alarm-rate data in this sample for responses that followed correct responses to the preceding name. Neither d’, F(1,29) < 1, nor c, F(1,29)< 1, differed in the Preceding Name Target and Preceding Name Lure conditions.

Figure 3
Experiment 2, Semantic Condition: Hit- and false-alarm rates for responses to the test name following correct responses to the preceding name, broken down as a function of the list status of the preceding name and the test name.

The results for the Episodic condition imply that the bias for the participant’s recognition judgment to the current test item was more conservative when the preceding item was a target than when it was a lure. In contrast, the results for the Semantic condition imply no such dependency. These results suggest that the present bias shift phenomenon does not occur in semantic recognition. These results support the Set Depletion Hypothesis.

Experiment 3

Experiment 3 sought further support for the Set Depletion Hypothesis. The experiment compared recognition judgments with respect to short and long lists. In the Short List condition, the experiment comprised 24 trials, on each of which the study list comprised 4 words and the test list comprised 2 preceding word – test word pairs. In the Long List condition, the experiment comprised 8 trials, on each of which the study list comprised 32 words and the test list comprised 32 preceding word – test word pairs. Words were never repeated during the test phase of a trial. The list status of the preceding word and the test word was manipulated as in the earlier experiments.

The Set Depletion Hypothesis predicted that the effect of the list status of the preceding word would be larger for the Short list than for the Long list. To see the reason for this prediction, consider the inference that can be drawn, under the assumption that words are never repeated during the test phase of a trial, when the preceding word is a target and the participant responds correctly to it. In the Short List condition, one of the four members of the memory set is subsequently unavailable for presentation as test word. So the chance of choosing a word from the set, at random, that can be presented as test word is reduced by ¼. In the Long List condition, one of the 32 members of the memory set is subsequently unavailable for presentation as test word. So the chance of choosing a word from the set, at random, that can be presented as test word is reduced by 1/32. In sum, following a correct positive response to the preceding word, the chance of choosing a word that can be presented as the test word is reduced more in the Short List than the Long List condition. Thus, the Set Depletion Hypothesis predicted that the effect of the list status of the preceding word would be larger for the Short list than for the Long list condition.

The Set Depletion Hypothesis also predicted that, within the Long List condition, the effect of list status might be larger in the second than in the first half of the test phase. To see the reason for this prediction, notice that the participant may see the memory set as becoming smaller over the course of the test phase, because increasing numbers of members are unavailable for presentation as preceding word or test word, having already been presented. This re-evaluation of memory set size is more likely in the Long List condition than in the Short List condition, because the test phase has greater temporal extension in the Long List condition. If the participant sees the memory set as shrinking in this way, consider, again, the inference that can be drawn when the preceding word is a target and the participant responds correctly to it. During the first half of the test phase, the participant assumes that the memory set contains 32 words. The chance of choosing a word from the set, at random, that can be presented as the test word is reduced by 1/32 every time the preceding word is a target and the participant responds correctly to it, by the argument given earlier. During the second half of the test phase, the participant assumes, on the basis of the experience that equal numbers of targets and lures are presented, that the memory set contains 16 words; that is, only 16 un-presented words remain in the memory set. The chance of choosing a word from the set, at random, that can be presented as the test word is reduced by 1/16 every time the preceding word is a target and the participant responds correctly to it. Thus, the Set Depletion Hypothesis predicted that the effect of the list status of the preceding word might be larger in the second than in the first half of the test phase.

Method

Participants

The 80 participants were (40 in the Short and 40 in the Long List condition) drawn from the same population as the participants of Experiment 1.

Design

List Length (Short/Long) was manipulated between participants. Preceding Word (Target/Lure) and Test Word (Target/Lure) were manipulated within participants. For a given participant in the Short List condition, 12 preceding word – test word pairs were assigned to each of the four conditions that occurred when the two levels of Preceding Word and Test Word were crossed. For a given participant in the Long List condition, 64 preceding word – test word pairs were assigned to each of the same four conditions.

Materials

The lists were created by sampling without replacement from a set of 1076 common nouns. On average, the words in the set were composed of 5.15 letters and 1.52 syllables. Their Kucera and Francis (1967) frequency was 65.10. In other respects, the materials were constructed in the same manner as for Experiment 1.

Procedure

The procedure was the same as for Experiment 1 except in the following respects; During the study phase of a given trial, each word in the list was presented for 1600 rather than 1000 ms and followed by an ISI of 400 ms. Between the study and test phase of a given trial, participants counted backwards from 100 for 56 seconds in the Short List condition and for 3 seconds in the Long List condition (participants were required to count for a longer period of time in the Short List condition to equate the average time interval between study and test in the two conditions). During the test phase of the trial, each preceding word - test word pair was preceded by the message “Next pair”, which remained on the screen for 1000 ms.

Results and discussion

No participants failed to reach the accuracy criterion. Recognition responses to the preceding word were correct 91% of the time in the Short List condition and 72% of the time in the Long List condition. Figure 4 presents the hit- and false-alarm-rate data for responses that followed correct responses to the preceding word.

Figure 4
Experiment 3: Hit- and false-alarm rates for responses to the test word following correct responses to the preceding word, broken down as a function of the list status of the preceding word and the test word, and list length.

Whereas d’ was greater in the Short (4.12) than in the Long List (1.44) condition, F(1,78) = 97.57, MSe = 2.88, g = 2.13, d’ did not differ in the Preceding Word Target and Preceding Word Lure conditions, F(1,78) < 1. The effects of Preceding Word and List Length did not interact in the d’ data, F(1,78) < 1. Whereas c did not differ in the Short and Long List conditions, F(1,78) < 1, c was greater in the Preceding Word Target (.43) than in the Preceding Word Lure (-.11) condition, F(1,78) = 14.03, MSe = .84, g = .58. The effects of Preceding Word and List Length interacted in the c data, F(1,78) = 4.14, MSe = .836, g = .64, with the effect of Preceding Word being greater in the Short List (.56 vs. -.29) than in the Long List condition (.31 vs. .06).

Figure 5 presents the hit- and false-alarm-rate data for the Long List condition, broken down as a function of whether the test word appeared in the first or second half of the test phase. Whereas d’ was greater in the first (1.68) than the second half (1.40) data, F(1,39) = 5.99, MSe = .49, g = .28, d’ did not differ in the Preceding Word Target and Preceding Word Lure conditions, F(1,39) = 2.33, MSe = .32. The effects of Preceding Word and Half did not interact in the d’ data, F(1,39) < 1. Whereas c did not differ in the first and second halves of the test series, F(1,39) < 1, c was greater in the Preceding Word Target (.39) than in the Preceding Word Lure (.03) condition, F(1,39) = 8.28, MSe = .58, g = .44. The effects of Preceding Word and Half did not interact in the c data, F(1,39) < 1, although the average difference between the value of c in the Preceding Word Target and Preceding Word Lure conditions was greater for the Second Half (.44 vs. .01) than for the First Half (.34 vs. .06). The data for the Short List condition were not examined as a function of Half because several of the participants were missing data for one or more cells in the analysis.

Figure 5
Experiment 3: Hit- and false-alarm rates for responses to the test word following correct responses to the preceding word, broken down as a function of the list status of the preceding word and the test word, and whether the test word appeared in the ...

As in the earlier experiments, the results imply that the bias for the recognition judgment to the current test word was more conservative when the preceding word was a target than when it was a lure. Of greater importance, the effect of the list status of the preceding word was greater for short lists than for long lists. These results support the Set Depletion Hypothesis. Within the Long List condition, the average difference between the value of c in the Preceding Word Target and Preceding Word Lure conditions was greater in the second half than the first half of the test phase. However, the difference between the two halves was not statistically significant. Thus, the prediction of the Set Depletion Hypothesis was not confirmed in this case. We must remember, though, that the prediction was contingent on participants seeing the memory set as shrinking over the course of the test phase3. If a participant saw the memory set as remaining the same size throughout the test phase, the effect of the list status of the preceding word would have been the same in both halves of the test phase. The non-significant difference between the size of the bias shift in the first and second halves probably reflects the fact that only some subjects saw the memory set as shrinking over the course of the test phase.

General Discussion

Our results suggest that, when episodic recognition judgments occur in succession, in tasks such as we have used, participants take the list status of the preceding item into consideration in setting the bias for the recognition judgment to the current test item. Specifically, our results suggest that the bias for the recognition judgment to the test item is more conservative when the preceding item is a target than when it is a lure. We observed evidence of this pattern in the hit- and false-alarm-rate data for all of our experiments. The hit- and false-alarm-rate results were in no case compromised by a speed-accuracy tradeoff. The effect sizes in the data that implied this pattern were in the small to medium range. Finally, our results suggest that the present bias shift phenomenon is not general to recognition memory. We observed no evidence of the phenomenon in a semantic recognition task.

We acknowledge that, whereas our results are consistent with a bias account, they are also consistent with another account according to which performance in our experiments reflected a signal detection process under which 1) the decision criterion for the judgment to the test item remained constant throughout the experiments, and 2) the target and lure distributions shifted the same distance to the left when the participant correctly apprehended the preceding item to be a target (Dosher, 1991; MacMillan & Creelman,1991; Wixted & Stretch, 2000). This distribution shift account is very difficult to differentiate from the bias account. We suggest, however, that the distribution shift account is less plausible than the bias account. It is not difficult to come up with reasons why the bias for the recognition judgment to the test item might depend on the list status of the preceding item (one possible account has been presented here). It is more difficult to come up with reasons why the positions of the target and lure distributions might depend on the list status of the preceding item.

Granting that our results reflect a dependency of bias on sequential context, what is the functional basis of this dependency? We have suggested that the dependency reflects a calculation regarding the likelihood of the test item’s being a target. According to our account, participants infer that this likelihood is reduced after they apprehend the preceding item to be a target, because fewer items from the list remain that might possibly be presented as the test item.

Our account differs crucially from accounts that have been proposed for sequential dependencies in psychophysical and reaction time tasks, under which the bias for the current response is more closely tied to concrete properties of previous responses and the stimuli to which they were emitted (Mozer, Kinoshita, & Shettel, 2006; Petrov & Anderson, 2005; Treisman & Williams, 1984). Under the account that we have proposed, the bias for the current response is tied to an abstract property of the previous item – whether or not the item belongs to the memory set. Two sorts of evidence are available regarding this property: 1) the response that the respondent makes to the item (which, of course reflects the consultation of memory), and 2) the feedback that the respondent receives for this response. In the present study, feedback for the response to the preceding item was not presented until after the response to the test item was emitted. Thus, the feedback could not be used to infer the list status of the preceding item in setting the bias for the test item. Stronger bias shifting effects may be observed if feedback is presented immediately after the response to the preceding item.

Under the proposed account, the present results constitute a case in which bias was determined anew for each item in a recognition task. Increasing attention has been paid recently to the conditions under which this sort of item-level bias setting occurs. Initial reports suggested that such bias setting was rare. Thus, Stretch and Wixted (1998) and Morrel, Gaitan, and Wixted (2002) observed no evidence of item-level bias setting when different classes of item were associated with different amounts of learning during the study phase of a recognition task. More recently, several conditions have been identified under which item-level bias setting appears to occur. Dobbins and Kroll (2005) reported results suggesting that bias can be set at the item level when the items in the memory set have different levels of pre-experimental familiarity. Singer and Wixted (2006) reported results suggesting that bias can be set at the item level when different categories of item are associated with different retention intervals. Rhodes and Jacoby (2007) reported results suggesting that bias can be set at the item level when different testing contexts are associated with different prior probabilities that the test item belongs to the memory set. The present results suggest a further extension of the range of conditions under which item-level bias setting occurs (see also Bruno, Higham, & Perfect, 2009).

Discussions of performance in simple cognitive tasks have often been framed in terms of the concept of optimality. Interest has focused on the criteria for optimality in such tasks (Edwards, 1965, Maddox & Bohil, 1998), on devising models that can produce optimal performance (Bogacz, Brown, Moehlis, Holmes, & Cohen, 2006; McMillen & Holmes, 2006), and on cases in which humans approximate optimal performance (Bogacz et al., 2006). The present phenomenon may not be one of those cases. Assuming the validity of the set depletion account, bias becomes more conservative when the participant apprehends the preceding item to be a target because the set of possible targets is presumed to have become smaller. But the change in the size of the set of potential targets is quite minimal. In Experiment 1 and the Episodic condition of Experiment 2, the memory set of twelve words was depleted by one word when the preceding word was apprehended to be a target. In the manner of thinking discussed in the introduction to Experiment 3, the chance of choosing a word from the set, at random, that could be presented as the test word was reduced by 1/12. In the Short and Long List conditions of Experiment 3, the corresponding reductions were 1/4 and 1/32. These are not large reductions. Yet we observed an average difference of .29 in c between the Preceding Word Target and Preceding Word Lure conditions across the study. In comparison, Estes and Maddox (1995) observed an average difference of .265 in c across conditions in which the base rates of targets were .33 and .67. The possibility arises, then, that the present phenomenon is yet another instance in which human behavior departs from what would be expected under a strictly rational analysis.

In conclusion, our results suggest that, when two episodic recognition judgments occur in succession, the bias for the recognition judgment to the current test item reflects the list status of the preceding test item. Our results suggest that this phenomenon involves a calculation regarding the likelihood of the test item’s being a target.

Acknowledgments

This research was supported in part by NIH grant MH 66189. We thank Tim Perfect, Mathew Rhodes, and an anonymous reviewer for suggesting Experiments 2 and 3.

Footnotes

1We could, in principle, have ignored whether the participant responded correctly to the preceding word, and simply examined performance as a function of the participant’s response to the preceding word. We have found, however, that participants sometimes report being aware of having made an error upon responding incorrectly. Thus, when the participant responded incorrectly to the preceding word, the participant’s response may not have accurately reflected the participant’s sense of the preceding word’s list status.

2We observed parallel results in the response-time data. That is, the effects of Preceding Word and Test Word interacted in these data; when the test word was a target, response time was longer when the preceding word was a target than when the preceding word was a lure; when the test word was a lure, response time was longer when the preceding word was a lure than when the preceding word was a target. We observed a similar pattern in the first generalizing experiment for Experiment 1. There were no significant effects in the response-time data for the second generalizing experiment for Experiment 1, or in Experiments 2 or 3. Further information regarding the response time analyses is available from the authors.

3Seeing this shrinkage is distinct from seeing the members of the set as being unavailable consequent to previous presentation, which is what is predicted under the Set Depletion Hypothesis. For example, some participants may have seen the 32-element set as remaining the same size throughout the test phase, and seen each set member that was presented throughout the test phase as reducing the number of candidates for future presentation by 1/32, with the result that the bias shift remained constant in size for these participants throughout the test phase. Alternatively, some participants may have seen the set as shrinking; for example, the participants may have seen the set as being reduced in size from 32 to 16 during the second half of the test phase. In this case, whereas the participants saw each set member that was presented during the first half of the test phase as reducing the number of candidates for future presentation by 1/32, the participants would have seen each member that was presented during the second half as reducing the number of future candidates by 1/16, with the result that the bias shift would have been larger for these participants during the second than the first half of the test phase.

Contributor Information

Stephen Dopkins, The George Washington University.

Jesse Sargent, Washington University.

Catherine Trinh Ngo, Baylor College of Medicine.

References

  • Bogacz R, Brown E, Moehlis J, Holmes P, Cohen JD. The physics of optimal decision making: A formal analysis of models of performance in two-alternative forced-choice tasks. Psychological Review. 2006;113:700–765. [PubMed]
  • Brown S, Steyvers M. The dynamics of experimentally induced criterion shifts. Journal of Experimental Psychology: Learning, Memory, & Cognition. 2005;31:578–599. [PubMed]
  • Brown S, Steyvers M, Hemmer P. Modeling experimentally induced strategy shifts. Psychological Science. 2007;18:40–45. [PubMed]
  • Bruno D, Higham PA, Perfect TJ. Global subjective memorability and the strength-based mirror effect in recognition memory. Memory & Cognition. 2009;37:807–818. [PubMed]
  • Dobbins IG, Kroll NEA. Distinctiveness and the recognition mirror effect: Evidence for an item-based criterion placement heuristic. Journal of Experimental Psychology: Learning, Memory, & Cognition. 2005;31:1186–1198. [PubMed]
  • Dopkins S, Ngo CT. Inhibition of verbal memory retrieval as a consequence of prior retrieval. Journal of Memory and Language. 2002;46:606–621.
  • Dopkins S, Ngo CT, Sargent J. Exploring a recognition-induced recognition decrement. Memory & Cognition. 2006;34:839–853. [PMC free article] [PubMed]
  • Dosher BA. Bias and discrimination in cuing of memory: A weighted decisions model. In: Hockley WE, Lewandowsky S, editors. Relating Theory and Data: Essays on Human Memory in Honor of Bennet B Murdock. Hillsdale, New Jersey: Lawrence Erlbaum Associates; 1991. pp. 249–278.
  • Edwards W. Optimal strategies for seeking information: Models for statistics, choice reaction times, and human information processing. Journal of Mathematical Psychology. 1965;2:312–329.
  • Estes WK, Maddox WT. Interactions of stimulus attributes, base rates, and feedback in recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1995;31:1075–1095. [PubMed]
  • Gernsbacher MA. Mechanisms that improve referential access. Cognition. 1989;32:99–156. [PMC free article] [PubMed]
  • Gillund G, Shiffrin RM. A retrieval model for both recognition and recall. Psychological Review. 1984;91:1–67. [PubMed]
  • Healy AF, Kubovy M. The effects of payoffs and prior probabilities on indices of performance and cutoff locations in recognition memory. Memory & Cognition. 1978;6:544–553. [PubMed]
  • Hirshman E. Decision processes in recognition memory: Criterion shifts and the list-strength paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1995;21:302–313. [PubMed]
  • Kirby N. Sequential effects in choice reaction time. In: Welford AT, editor. Reaction Times. London: Academic Press; 1980. pp. 127–172.
  • Kucera H, Francis WN. Computational analysis of present-day American English. Providence: Brown University; 1967.
  • Luo CR. Enhanced feeling of recognition: Effects of identifying and manipulating test items on recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1993;13:405–413.
  • MacMillan NA, Creelman CD. Detection theory: a user’s guide. New York: Cambridge University Press; 1991.
  • Maddox WT, Bohil CJ. Base-rate and payoff effects in multidimensional perceptual categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1998;24:1459–1482. [PubMed]
  • McMillen T, Holmes P. The dynamics of choice among multiple alternatives. Journal of Mathematical Psychology. 2006;50:30–57.
  • McNamara TP. Theories of priming: I. Associative distance and lag. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1992;18:1173–1190.
  • Morrell HER, Gaitan S, Wixted JT. On the nature of the decision axis in signal-detection-based models of recognition memory. Journal of Experimental Psychology: Learning, Memory, &Cognition. 2002;28:1095–1110. [PubMed]
  • Mozer MC, Kinoshita S, Shettel M. Sequential dependencies in human behavior offer insights into cognitive control. In: Gray WD, editor. Integrated Models of Cognitive Systems. New York: Oxford; 2007. pp. 180–193.
  • Niewiadomski MW, Hockley WE. Interrupting recognition memory: Tests of familiarity-based accounts of the revelation effect. Memory & Cognition. 2001;29:1130–1138. [PubMed]
  • Perea M, Carreiras M. Sequential effects in the lexical decision task: The role of the item frequency of the previous trial. Quarterly Journal of Experimental Psychology. 2003;56A:385–401. [PubMed]
  • Petrov AA, Anderson JR. The dynamics of scaling: A memory-based anchor model of category rating and absolute identification. Psychological Review. 2005;112:383–416. [PubMed]
  • Ratcliff R, McKoon G. Sequential effects in lexical decision: Tests of compound –cure retrieval theory. Journal of Experimental Psychology: Learning, Memory, and Cognition. 1995;21:1380–1388. [PubMed]
  • Rhodes MG, Jacoby LL. On the dynamic nature of response criterion in recognition memory: Effects of base rate, awareness, and feedback. Journal of Experimental Psychology: Learning, Memory, & Cognition. 2007;33:305–320. [PubMed]
  • Singer M, Wixted JT. Effect of delay on recognition decisions: Evidence for a criterion shift. Memory & Cognition. 2006;34:125–137. [PubMed]
  • Smith PT. The non-monotonicity of the psychometric function in recognition memory. Perception & Psychophysics. 1969;5:329–337.
  • Snodgrass JG, Corwin J. Pragmatics of measuring recognition memory: Applications to dementia and amnesia. Journal of Experimental Psychology: General. 1988;117(1):34–50. [PubMed]
  • Soetens E. Localizing sequential effects in serial choice reaction time with information reduction procedure. Journal of Experimental Psychology: Human Perception and Performance. 1998;24:547–568.
  • Stretch V, Wixted J. On the difference between strength-based and frequency-based mirror effects in recognition memory. Journal of Experimental Psychology: Learning, Memory, & Cognition. 1998;24:1379–1396. [PubMed]
  • Treisman M, Williams TC. A theory of criterion setting with an application to sequential dependencies. Psychological Review. 1984;91:68–111.
  • Verde MF, Rotello CM. Does familiarity change in the revelation effect? Journal of Experimental Psychology: Learning, Memory, & Cognition. 2003;29:739–746. [PubMed]
  • Verde MF, Rotello CM. ROC curves show that the revelation effect is not a single phenomenon. Psychonomic Bulletin & Review. 2004;11:560–566. [PubMed]
  • Watkins MJ, Peynircioglu ZF. The revelation effect: When disguising test items induces recognition. Journal of Experimental Psychology: Learning, Memory, & Cognition. 1990;16:1012–1020. [PubMed]
  • Westerman DL, Greene RL. The revelation that the revelation effect is not due to revelation. Journal of Experimental Psychology: Learning, Memory, & Cognition. 1998;24:377–386. [PubMed]
  • Wixted JT, Stretch V. The case against a criterion-shift account of false memory. Psychological Review. 2000;107:368–376. [PubMed]
  • Zimmerman J, Kimble GA. Effects of incentive on false recognition. Journal of Experimental Psychology. 1973;97:264–266.