Abstract
In this study, the number of semantic associates in Deese–Roediger–McDermott (DRM) lists was varied from 4 to 14 in a modified Sternberg paradigm. The false alarm (FA) and correct rejection (CR) reaction time (RT)/memory-set size (MSS) functions of critical lures showed a cross-over interaction at approximately MSS 7, suggesting a reversal of the relative dominance between these two responses to the critical lure at this point and also indicating the location of the boundary between the sub- and supraspan MSS. For the subspan lists, FA to critical lures was slower than CR, suggesting a slow, strategic mechanism driving the false memory. Conversely, for the supraspan lists, critical lure FA was faster than its CR, suggesting a spontaneous mechanism driving the false memory. Results of two experiments showed that an automatic, fast, and a slow, controlled process could be error-prone or error-corrective, depending on the length of the DRM memory list. Thus there is a dual retrieval process in false memory as in true memory. The findings can be explained by both the activation/monitoring and the fuzzy-trace theories.
Keywords
The Subjective Experiences of False Memory and an Objective Test
Roediger and McDermott (1995) revived and expanded a lab approach using semantic associates to create false memory that Deese (1959) first initiated. This method of studying false memory is now known as the Deese–Roediger–McDermott (DRM) paradigm. In this paradigm, subjects study a list of words associated with a theme, but not the thematic word itself. For example, the list of words can be bed, rest, awake, tired, dream, wake, snooze, blanket, doze, slumber, snore, nap, peace, yawn, drowsy, dead, lay, all associated with the thematic word sleep, which, however, is itself not presented. Subjects falsely recall or recognize the nonpresented thematic word or critical lure at about the same level as a list word located in the middle of the list (Roediger & McDermott, 1995). Moreover, subjects feel very confident about studying the critical lure and report vivid, compelling sensory, perceptual details for the critical lure as if it were subjectively no different from true memory (Hicks & Marsh, 1999; Lampinen, Neuschatz, & Payne, 1997; Lampinen, Neuschatz, & Payne, 1999; Neuschatz, Payne, Lampinen, & Toglia, 2001; Payne, Elie, Blackwell, & Neuschatz, 1996; Reed, 1996; Robinson & Roediger, 1997; Roediger & McDermott, 1995). These were subjective verbal reports of one's memory of what had happened. As indicated by Nisbett and Wilson (1977), verbal reports often are not an accurate representation of the true underlying cognitive processes. One objective test that can be used to probe the aspects of the critical-lure memory that may be beyond the subjects’ conscious access is a memory probe task similar to the one used by Sternberg (1966, 1969, 1975) to study the relationship between recognition time of a test probe and the number of items held in memory.
The relevance of this recognition memory paradigm to the purpose of the present study is its robust finding that the test probe's recognition reaction time (RT) increases linearly with the number of items held in memory, at least within the range six items. Subjects in the Sternberg paradigm study a variable number of items (e.g., letters, digits, or words), typically from 1 to 6, and subsequently take a recognition memory test in which they respond “yes” or “no” to a probe (Sternberg, 1966, 1969, 1975). In this study, we modified some procedural features of the paradigm that we deemed non-crucial in determining this basic relationship to suit the purpose of the present experiments. 1 Our focal question is what the shapes of the RT functions of false alarm (FA) and correct rejection (CR) to the critical lure look like in comparison with those of the studied words and the noncritical distractors.
If false recognition memory is like true recognition memory, then the shape of the response time (RT)/memory-set size (MSS) function should be the same as or similar to that of true recognition memory. If the memory of the critical lure is not like that of a memory-set word, then a different RT/MSS function for the critical lure should be observed. The linear increase in RT of a memory-set word with the increase of MSS is attributed either to a linear serial search of the memory array (Sternberg, 1966, 1969) or to a parallel search with limited total memory resource or activation (Anderson, 1983; Townsend, 1971, 1990). According to the latter view, since the total activation is fixed, when it is divided among a greater number of items, the activation per item decreases, causing the recognition for a probe to slow down (Anderson, 1983; Townsend, 1990). The increase in the recognition RT as a function of the MSS is referred to as the MSS effect and is used as a diagnostic test index for the memory property of the critical lure in this study.
We are interested in comparing the RT/MSS functions of the FA and CR to critical lures with the corresponding RT functions of the studied critical items and the semantic associates on the study lists. For instance, does the false recognition RT for the critical lure increase linearly with the MSS just as that of the studied words does? Specifically, will the recognition of the critical lure slow down, or at least not speed up, as does a studied word when the number of words in the list increases? If the critical lure is rehearsed as other list words are during learning, its RT profile should be the same as or similar to that of a studied word. In an attempt to answer the question of whether subjects rehearse the critical lure, Goodwin, Meissner, and Ericsson (2001) and Goodwin (2007, Experiment 1) conducted a “thinking aloud” protocol experiment and revealed that subjects indeed often rehearsed the critical lure during encoding and that the level of rehearsing the critical lure was highly correlated with the likelihood of falsely recalling the critical lure later. If the thematic word is rehearsed during the encoding stage as list words, it should be searched for in a Sternberg task just as a word in the memory set and therefore yield a linearly increasing RT/MSS function. That is, if it has the memory properties of an externally encoded event, the critical lure should behave no differently from other memory-set words in the Sternberg paradigm. If the critical lure is not searched for in the memory set, then the Sternberg RT profile of the FA response to the critical lure will be different from that of a memory-set word.
From the point of view of the false memory being a product of internally generated activation by an activation spreading mechanism (Roediger, Balota, & Watson, 2001; Underwood, 1965), the more semantic associates subjects encode, the higher should the summed implicit activation be for the critical lure. Consistent with this concept, Robinson and Roediger (1997) found that when the number of semantic associates in the list increased, the rate of false recall and of FA to the critical lure increased, which they attributed to the increase in the activation level of the critical lure resulting from the additional semantic associates in the list. Although Robinson and Roediger (1997) and Roediger, Watson, McDermott, and Gallo (2001) measured the activation level of the critical lure by the rate of false recall and false recognition, activation level is traditionally measured by RT, with a shorter RT in recall or recognition indicating a higher level of activation and a longer RT a lower level of activation (Anderson, 1983; Balota & Paul, 1996; Jou, 2008; Jou, Matus, Aldridge, Rogers, & Zimmerman, 2004; Meade, Watson, Balota, & Roediger, 2007). Thus, if the memory activation of the critical lure increases with the number of semantic associates in memory set, then a FA response to the critical lure should grow faster as the length of the DRM list increases. Therefore, depending on whether the critical lure behaves no differently from an externally encoded (memory-set) word, or like an internally generated word through activation spreading from semantic associates (Underwood, 1965), the critical lure FA can be predicted to produce either an increasing or a decreasing RT/MSS function. By testing out the critical lure memory in a modified Sternberg paradigm, we hope to answer the question of whether the recognition of the critical lure speeds up (due to increased spread activation from a larger number of semantic associates) or slows down (if it has the same attributes as a memory-set word) with the increase in the DRM list length.
Coane, McBride, Raulerson, and Jordan (2007) used the Sternberg paradigm to study the DRM false recognition memory. They aimed to compare the FA rate and RT of the critical lure across three incremental MSSs: 3, 5, and 7. However, due to a relatively small sample size and an insufficient number of FA responses to the critical lure in the shortest list condition (i.e., too many missing data for MSS 3), they ended up comparing only MSS 5 with MSS 7. Most relevant to the present purpose, they found that both the FA RT and the FA rate of the critical lure increased significantly from MSS 5 to 7. In terms of activation, an increase in RT (a slowdown in response) indicates a decrease in the activation level, but an increase in FA rate indicates an increase in the activation level of the critical lure. From our point of view, the increased RT from MSS 5 to 7 suggests that the critical lure behaves just like a memory-set item: As the number of memory-set items increases, the recognition time of the memory-set item increases. Conversely, the increased critical-lure FA rate, in our view, suggests that the critical lure is not like a member in the memory set. Rather, it behaves like DRM false memory should: As the number of semantic associates on the list increases, the implicit activation level of the critical lure increases (Robinson & Roediger, 1997; Underwood, 1965). Hence, from the activation standpoint, Coane et al.'s critical-lure FA rate and RT results are incompatible. However, they interpreted the critical-lure FA RT increase from MSS 5 to 7 as the result of increased monitoring with increased list length, albeit a failed one in this case.
We were curious about the nature of the relationship between the critical-lure FA rate and its RT as a function of MSS as to whether they are positively or inversely related. We conducted the present experiments to see whether we would obtain the same results as did Coane at al. (2007) or different results. We also made substantial changes from Coane et al.'s in design and methods in the present experiments. Among the changes, we started from MSS 4 and extended it beyond 7 to 14 to investigate the dynamic interaction between the critical lure's FA and CR across the traditional sub- and supramemory-span boundary of 7. Another important difference between our study and Coane at al.'s was that they studied the short-term version of the DRM false memory whereas we studied the long-term DRM false memory (we explain this distinction in the next section).
Our hypothesis is that if the FA rate of the critical lure increases with the increase in MSS, then its FA RT should decrease. However, if the critical lure memory is just like that of a studied word, then its false recognition rate should decrease (or at least not increase), and the RT should increase (or at least not decrease) with the increase in MSS.
Memory-Span Limit and False Memory
Duration and capacity limitation has long been recognized as a fundamental property of attention and memory, which influences both storage and retrieval in important ways (Baddeley, 1994; Broadbent, 1975; Cowan, 2001; Cowan, Morey, Chen, Gilchrist, & Saults, 2008; Ebbinghaus, 1885/1964; Mandler, 1975; Miller, 1956; Slamecka, 1985). Surprisingly, however, little research has been done to examine how memory-span or limitation affects DRM false memory retrieval. The question of how span size affects DRM false-memory generation and retrieval is therefore another major focus of the present study.
Regarding memory limitation, a distinction needs to be made between a time-based and capacity-based memory limitation. According to Cowan (1999; Cowan et al., 2008), these are two separable types of limits, with the former being a duration limit of memory activation, and the latter a capacity limit at any one point in time. The duration or time limit refers to the duration limit of activation (without reactivation via rehearsal) before the activation fades away. The capacity limit, on the other hand, refers to the maximum number of unrelated items that can be held in memory at any one point in time. In our experiments, subjects performed a 20-s interpolated task after studying the list of words before taking the recognition test under all list-length conditions, thus making the retrieval from both short and long lists supposedly from the long-term memory in the time-based dimension. Therefore, our span manipulation was not of short- versus long-term memory in the time-based (retention-interval) sense. The retrieval both from the short and long lists was from long-term memory given the 20-s filled retention interval between studying and testing. The list-length manipulation in this study was therefore one of a capacity-based limitation.
In a standard Sternberg paradigm, memory list length varies from 1 to 6 items, and the RT/MSS function obtained is linear. However, studies using MSSs beyond 6 indicated either a bilinear (Burrows & Okada, 1975; Okada & Burrows, 1978) or a logarithmic (Corballis, Katz, & Schwartz, 1980), rather than a linear, RT/MSS function. The bilinear RT functions showed a deflection point occurring at about MSS 7 separating an earlier steeper slope and a later shallower slope. A logarithmic function displayed a smooth, gradually decreasing slope as the MSS increased. These findings are taken to suggest distinct retrieval modes for the subspan and the supraspan memories. That is, a serial exhaustive search mode is employed in a subspan memory retrieval whereas a direct access or a parallel search may be the predominant mode used in a supraspan memory retrieval (Burrows & Okada, 1975; Jou, 2001; Jou & Aldridge, 1999 2 ). But the questions of how the different retrieving modes can affect the generation and retrieval of false memory has not been as fully investigated as that of true memory.
Span limits can be affected by the contexts and materials used in an experiment, and are found to vary from one study to another (Baddeley, 1994; Broadbent, 1975; Cowan, 2001; Cowan et al., 2008; Mandler, 1975). In this study, what is the boundary between sub- and supraspan for the generation and retrieving of false memory is an empirical question. We think that we can determine it by examining the change in the shapes of the RT functions of the critical lure's FA and CR in relation to each other as the MSS increases. We assume that for the short lists, because the items’ verbatim forms are more easily available in memory, this should weaken the tendency to make a FA response to the critical lure, but at the same time enhance the tendency of a CR response to the critical lure. The relative strengths of the two responses can be measured both by response rate and RT. The stronger response should produce a higher response rate and a shorter RT, and the weaker response produce a lower response rate and a longer RT. As the list grows longer, the verbatim memory of the items becomes weaker, and at the same time, the gist memory grows stronger. As a result, the dynamic relation of the two responses should be reversed at some point. The point at which this reversal occurs will give an indication of where the boundary between the sub- and supraspan sizes is located along the length of the list.
The Dual Process in False-Memory Retrieval
The prevalent dual-process view holds that implicit spreading activation, which is the source of false memory, is automatic and fast (Jacoby, 1991; Underwood, 1965), but monitoring, which is the basis of CR, is an effortful, slow, controlled process (D. G. Gallo & Roediger, 2002; Jacoby, 1991; Roediger, Watson, et al., 2001). We expect that increasing the length of the DRM lists should decrease the RT of FA but increase the RT of CR to the critical lure and that the relative-strength relationship between the two responses to the critical lure will reverse at a certain point along the length of the memory list.
To foreshadow our results, our data suggested that there may be two different mechanisms that can give rise to false memory: an automatic, fast process (Jacoby, 1991) and a slow, strategic, and constructive process (Brainerd, Wright, Reyna, & Payne, 2002; Jou, 2008). If the activation of the critical lure is caused by automatic activation spreading, and the monitoring process by a controlled, strategic process that involves a determination of the source of the memory, then a FA to the critical lure should be faster than a CR to it. This is what the activation/monitoring and the dual-process theories predict. If both a fast, automatic, familiarity-based process and a slow, deliberate process can produce false memory, then it would pose a challenge to the above theories. We discuss the theoretical implications of our findings in the General Discussion.
Experiment 1
There are several purposes in conducting this experiment. First, we attempted to find out the shape of the RT/MSS function for the critical lure in a modified Sternberg paradigm. Specially, how does the RT function of the critical lure change as the number of semantic associates on the DRM list increases? The crucial question is whether the RT/MSS function increases or decreases. Second, how do the trend of the FA RT function and that of the CR of critical lure change in relation to each other over the whole range of the list length? Third, what do the RT/MSS functions for the studied critical words and list words look like? Only one probe was presented for a learned list of words to prevent possible confounding from contextual factors when multiple probes are tested on a list of learned words (Jou, 2014). As Jou (2014) showed, this single probe test procedure will elevate the RT due to a task-switching cost (Mayr & Kliegl, 2000; Monsell, 2003; Monsell, Sumner, & Waters, 2003; Rogers & Monsell, 1995). A control condition using unrelated word lists was employed to serve as a baseline condition to see how the RT functions of the DRM words could differ from that of the semantically unrelated words.
Method
Subjects
Eighty-one undergraduate introductory psychology students participated in the unrelated-word (control) condition, and 181 students participated in the DRM (semantic associates list) condition, for course credit. Because the probability of making a false alarm response to the critical lure in the shorter MSS conditions was relatively low, and also because only a single probe was tested for each list of words and the highly variable RT, a larger number of subjects than usual was needed to produce enough data to allow for the detection of reliable trends for the critical lure and other words’ RT functions.
Materials and Design
Semantic relatedness of word lists was varied between subjects to keep the experiment from running too long. Words on a list were semantically unrelated in the control condition. The words in a list were semantic associates in the related (DRM) condition. For the control condition, 36 lists of 15 words each of semantically unrelated words were created by selecting words from Kucera and Francis's (1967) word frequency norms in the frequency range of 100 to 150 occurrences per million words. Care was taken to see that the words within a list were semantically unrelated and that no word was repeated across the unrelated lists and the DRM lists. Furthermore, words in an unrelated list corresponding to the same numbered DRM list were carefully examined to make sure that those words were not related to the thematic topic in the corresponding DRM list. The 36 word lists in the DRM condition were adopted from Stadler, Roediger, and McDermott's (1999) DRM paradigm word norms with 15 words in each list. 3 For half of the lists, the critical word was studied, and for the other half, it was not studied. Across the subjects, the lists for which the critical words were studied and not studied were counterbalanced. For each subject, 9 of the 36 lists had their critical words presented both at study and at test (studied critical word probe), 9 lists had their critical words not presented at study but presented as probes at test (critical lure probe), 9 lists had a studied list word presented at test (studied list word probe), and 9 lists had an unrelated distractor word presented at test with this unrelated word chosen randomly from the 15 words of the counterpart list in the control condition. The use of this unrelated distractor was intended to make the probes by and large in keeping with the DRM paradigm in which two of the three distractor probes for a learned list are semantically unrelated, and the third is the critical lure.
The design for the unrelated (control) condition was basically the same. However, because all the words were unrelated, the “critical presented words” in the unrelated condition were simply presented list words, and the “critical lures” were simply nonpresented distractors. Thus, the presented “critical words” and the presented list words in the control condition were just presented list words, and the “critical lures” and the unrelated distractors were just nonpresented distractors. Therefore, the control condition had only two types probes: studied words and nonstudied distractors.
Six MSSs were used: 4, 6, 8, 10, 12, and 14. The 36 lists were rotated across the 6 MSSs across subjects. The presented words were randomly selected from the 15 words in a list for each subject, except for the critical word, which was designated for each list. We adopted Sternberg's varied-set testing procedure in which only one probe was tested for a learned list of words (Sternberg, 1966, 1969) to avoid potential confounds in a fixed-set procedure in which multiple probes are tested for a learned list of words (see Jou, 2014, for a detailed explanation). 4
Procedure
At the study phase, each word in the randomly selected memory set was presented for 2 s with a 1-s interword interval of blank screen. After the words were displayed, there was a 20-s counting backward interpolated task in which the subject started counting down by 3 at a time from a random three-digit number. This was done to weaken verbatim memory and increase false memory somewhat to avoid insufficient instances of false recognition for a data analysis. When subjects made an error, they were required to go back to recount from the previous number. They were encouraged to count at a fast pace. At the end of the counting, a single probe word was displayed for recognition for each list of words. Subjects were instructed to respond fast but not to sacrifice accuracy. There were 9 blocks of test trials with 4 trials in each block. In the unrelated condition, two trials of the 4-trial block were made of studied probes, and the other two trials were made of nonstudied distractors. The 4 types of trials in a block were presented in a random order in both the control and the DRM condition. The “/” and “z” keys were designated as response keys, one for a positive and one for a negative response, with the response-to-key mapping counterbalanced across subjects. At the end of each test trial and before the presentation of the next memory-set words, subjects could take a break if they so chose.
Results and Discussion
The mean hit rate and CR rate for the control condition as a function of MSS are presented in Figure 1. There was a significant linearly decreasing trend for the hit rate but a nonsignificant decreasing trend for the CR rate as a function of MSS. Neither response showed a significant quadratic trend. The trend test statistics are presented in Table 1.
Mean hit rate of targets and correct rejection (CR) rate of distractors as a function of memory-set size (MSS) of the control (unrelated-word list) condition of Experiment 1. To view this figure in colour, please visit the online version of this Journal. Trend test results for mean hit, FA, and CR rates as a function of test probe, word list relatedness condition, and experiment. Note: FA = false alarm; CR = correct rejection; DRM = Deese–Roediger–McDermott; ns = not significant.
To lessen the impact of RT outliers, individual subjects’ median RT scores in the response (“yes” vs. “no”) by MSS (6 memory list sizes) cell were used as the input data points in the calculation of the cell means used for the analysis of variance (ANOVA) and trend analysis of the RT data. The mean RTs of hits and CRs as a function of MSS and response are presented in Figure 2. There was a significant linearly increasing trend for the hits and a marginally significant increasing trend for the CRs as a function of MSS (for RT test result statistics, see Table 2). The quadratic trends for the two RT functions were both nonsignificant.
Mean reaction time (RT) of hits for targets and correct rejection (CR) for distractors as a function of memory-set size (MSS) of the control (unrelated-word list) condition of Experiment 1. To view this figure in colour, please visit the online version of this Journal. Trend test results for mean RTs as a function of test probe, list word relatedness condition, and experiment. Note: FA = false alarm; CR = correct rejection; DRM = Deese–Roediger–McDermott; ns = not significant.
An ANOVA was performed for these two RT functions. The mean RT of the hit responses (2427 ms) was significantly longer than that of the CR responses (2272 ms), F(1, 159) = 5.79, MSE = 1,931,357, p = .017. The MSS main effect was marginally significant, F(5, 805) = 1.95, MSE = 2,125,871, p = .08. The interaction of response with MSS was not significant, F < 1. The faster negative than the positive response was not typical in the Sternberg paradigm. An explanation is that in the standard Sternberg's experiments, 9 digits or a small set of alphabetic letters (high frequency items of a small vocabulary, as Wickens, Moody, & Vidulich, 1985, called it) is repeatedly and inconsistently used as targets and distractors. This creates a great amount of interference. For that reason, subjects have to rely heavily on the detailed contextual information to determine the memory-set membership of the probe (i.e., to resort to the recollective process) for both positive and negative probes. In our control condition, unrelated words were used, and they were not repeatedly and inconsistently assigned to the positive and negative sets. For this reason, subjects did not have to resort to the recollective retrieval process. A quick, direct familiarity check involving limited search was sufficient for rejecting a negative probe based on a lack of a familiarity feeling (although this process is not as useful for recognizing a positive probe) resulting in faster negative responses than positive responses (Atkinson & Juola, 1973; Banks & Atkinson, 1974; Jou, 2014; Roediger, Knight, & Kantowitz, 1977, also obtained faster negative than positive responses).
For the DRM word list condition, there were four types of probe words: studied critical words, unrelated distractors, studied list words, and the critical lure. The mean hit rate for the studied critical words, the CR rate for the unrelated distractors, the hit rate for the studied list words, and the FA and the CR rates for the critical lure as a function of MSS are presented in Figure 3.
Mean hit rate of studied critical words, correct rejection (CR) rate of unrelated distractors, hit rate of studied list words, and false alarm (FA) and CR rates of the critical lure as a function of memory-set size (MSS) of the Deese–Roediger–McDermott (DRM; semantic associates word list) condition of Experiment 1. To view this figure in colour, please visit the online version of this Journal.
Overall mean hit rate, FA rate, and CR rate as a function of test probe and experiment.
Note: HR = hit rate; FAR = false alarm rate; CRR = correct rejection rate; wd = word; distr = distractor.
The trend test result statistics of the five response rates are presented in Table 1. As is apparent by a visual inspection of Figure 3, the hit rate for the studied critical word did not show a decreasing trend over the MSS, suggesting a privileged memory status that the studied critical word enjoyed. This result was consistent with the corresponding findings in recall showing that the studied critical words were recalled on average at an earlier serial position (Barnhardt, Choi, Gerkens, & Smith, 2006; Jou, 2008) and with a shorter recall latency (Jou, 2008). Nor did the CR rate of the unrelated distractor show a decreasing trend. The hit rate of the studied list words did show a significant linearly decreasing trend, and a marginally significant quadratic trend (see Table 1). Because the FA and CR rates of the critical lure were complements of each other (the two rates added up to 1.0), the test results were identical except with the trends reversed.
The rate of false recognition of the critical lure showed a significant linear increase. This result was consistent with the Robinson and Roediger (1997) finding of an increase in critical lure FA rate with the increase in the number of semantic associates and with Coane et al.'s (2007) finding of a critical lure FA rate increase from MSS 5 to MSS 7. In contrast, the critical lure CR rate showed a significant decrease as a function of MSS.
The mean RTs of the above five types of responses as a function of the MSS for the DRM condition are presented in Figure 4 and the trend test results in Table 2.
Mean reaction time (RT) of hits of studied critical words, correct rejection (CR) of unrelated distractors, hits of studied list words, and false alarm (FA) and CR of the critical lure as a function of memory-set size (MSS) of the Deese–Roediger–McDermott (DRM; semantic associates word list) condition of Experiment 1. To view this figure in colour, please visit the online version of this Journal.
A visual inspection of the RT functions of the hits for the studied critical words and the studied list words indicated that there was a rise from MSS 4 to 10 and then a fall from 10 to 14 (see Figure 4). Consistent with the visual observation, for the studied critical words, the linear trend was significant, and the quadratic trend was very close to significance (see Table 2). For the studied list words, both the linear and the quadratic trends were highly significant. For the RT function of the CR to the unrelated distractor, neither the linear nor the quadratic trend was significant.
For the FA to the critical lure, the linearly decreasing trend was significant although the quadratic trend was not. This RT linear decreasing trend with the increase in MSS was the opposite of Coane et al.'s (2007) finding of an increase in the FA RT from MSS 5 to MSS 7 for the critical lure. We discuss this discrepancy in detail in General Discussion. For the CR to the critical lure, the linearly increasing trend was significant, but the quadratic trend was not. An ANOVA was conducted on these two RT functions to confirm the visual observation that there was an interaction between response (FA vs. CR) and MSS. The results showed that the mean RT of the CR (3664 ms) was significantly longer than that of the FA (3218 ms), 5 F(1, 344) = 5.75, MSE = 9,160,468, p = .017. The MSS main effect was not significant, F = 1.22. The Response by MSS interaction was highly significant, with F(5, 775) = 8.79, MSE = 4,387,231, p < .0001.
Several findings were significant. The first important finding was that the RT of the critical lure's FA function was decreasing rather than increasing. If it was rehearsed with the other list words before the recognition test (i.e., if the critical lure became a memory-set word), it should have produced an increasing (or at least no decreasing trend) instead of a decreasing RT/MSS function. Nowhere along the 6 MSSs continuum did the critical lure false recognition RT function show any signs of even slightly resembling an upward trend of a list word. This suggested that the critical lure is not like a memory-set word, and that an internally generated memory is different in its RT profile from a memory of an externally encoded event. Thus, despite the vivid, detailed, compelling subjective experiences reported by the subjects about their memory of the critical lure, the RT/MSS profile of the FA response to the critical lure gave no indication of the critical lure being actually among the memory-set words.
Of special interest is the cross-over of the FA and CR RT functions of the critical lure, which indicated increasing strength of the FA response (an increasing response rate accompanied by a decreasing RT) and declining strength of the CR response (a decreasing response rate accompanied by an increasing RT) to the critical lure as the list length increased, and the location on the MSS scale at which the relative strengths of these two responses were reversed. That reverse point was located at MSS 7 (although it was not an actual MSS used in this experiment). Thus, we can characterize the recognition of the critical lure in MSS 4 and 6 as CR dominant, and that in MSSs larger than 7 as FA dominant. The reverse of the dominance relationship between these two responses can be attributed to the transition from a verbatim dominant memory mode below the interpolated MSS 7 to a gist dominant mode past MSS 7. Therefore, based on the cross-over point being indicated at MSS 7, we can locate the boundary between sub- and supraspan memory at MSS 7 for the present purpose.
The supraspan part of the critical lure RT function clearly showed an increase in the critical lure activation (as demonstrated by a speeding up of the FA response to the critical lure) as the total number of associates increased, consistent with the finding of an increase in the rate of false recall or recognition as the number of semantic associates on the list was increased, as reported in Robinson and Roediger (1997). The CR response was the result of a successful monitoring. It was much slower than the FAs (at least for the supraspan section), suggesting the slow and effortful nature of this process. And the larger the number of associates in the list was, the slower the execution of the monitoring process grew, which makes sense since successful monitoring probably involves a thorough search of the whole array of the memory-set items.
The subspan section of the MSSs showed an RT function pattern for the FA and CR of the critical lure opposite to that in the supraspan section and told a different story (as the significant interaction already indicated). Within the subspan section, the CR was significantly faster (3056 ms) than the FA response (3754 ms), F(1, 250) = 5.80, MSE = 4,890,924, p = .017, corroborating the significant interaction. This suggests that within this MSS range, the CR was, relative to the FA, a more spontaneous, faster response, whereas the FA might be the product of some controlled, strategic process. Interestingly, this seemingly slow, controlled response led to a wrong response—that is, a FA response to the critical lure. A possible explanation for this slower FA than CR response was that subjects might have made a deliberate inferential judgment or a constructive synthesis on the semantic theme of the associates (Brainerd et al., 2002) based on the portion of the 15-word list and overridden the initial, faster but correct response driven by the verbatim memory. For the supraspan section, the FA response to the critical lure (3093 ms) was significantly faster than the CR response (3595 ms) to the critical lure, F(1, 325) = 4.46, MSE = 4,607,210, p = .035, again corroborating the response by MSS interaction. Thus, the two RT functions of the critical lure demonstrated how the two processes, the automatic spreading of activation and the metaknowledge judgment on the make-up of the list, can both produce false memory, but through possibly opposite mechanisms. This also mirrors Jou's (2008) finding in recall that there was an early and a late batch of false recall in the DRM paradigm revealed by two distinguishable distributions of false recalls along the recall temporal course. He attributed the late batch of false recall to a slow, inferential, constructive process performed at the retrieval stage.
The reversed patterns of the RT function of the false recognition of the critical lure in relation to its CR across the sub- and supraspan lists is consistent as well with the idea of a dual retrieval process in sub- and supraspan veridical memories discussed earlier. These two processes by which false memory can arise were discussed in the literature (Brainerd & Reyna, 1998; Miller & Wolford, 1999; Roediger, Balota, et al., 2001; Underwood, 1965). Our data seemed to suggest that the deliberate inferential judgment about the theme word may be the mechanism generating false recognition in the subspan list conditions, whereas a more automatic, faster activation spreading process might be responsible for generating false recognition in the supraspan list conditions. Again, our data suggest that there is a fast and a slow mechanism of generating false recognition memory with the former operative in the supra- and the latter in the subspan memory set sizes.
Finally, the bowed RT functions of the studied critical words and list words suggested that when the MSS exceeded 10, subjects might have relied progressively more on a strategy using semantic categorization to make a decision on the probe, rather than relied on the memory of the individual word for a decision, consistent with Jones and Anderson's (1982, 1987) findings. The fact that the controlled condition did not yield a bowed RT curve supported this hypothesis. Thus, the DRM word list produced RT/MSS profiles in this modified Sternberg paradigm both for false and for veridical memories that were different from the typical RT/MSS functions.
Experiment 2
The shape of the RT functions of the studied words (the studied critical words and list words) in Experiment 1 deviated from the pattern of the RT functions typically found in this paradigm. We believe that the bow-shaped functions of the studied words were the result of subjects making a decision by increasingly relying on semantic category of the probe past MSS 10 (e.g., “Is this a word related to ‘doctor’ or not? If yes, respond yes; if no, respond no”) rather than on the memory of the individual word (Jones & Anderson, 1982, 1987). In the typical Sternberg paradigm, all probes are categorically homogeneous (e.g., all digits or letters). Thus, the first purpose of Experiment 2 was to make all the probes semantically related to the same theme to prevent the use of the semantic categorization strategy (since all the probes are related to the same theme, subjects have to reply on memory of the individual words). Therefore, in Experiment 2, the only unrelated distractor probe in each block of four trials in Experiment 1 was replaced with a nonstudied semantic associate, making the two targets and the two distractors in a block semantically (categorically) homogeneous. We expect that the anomalous bowedness of the RT functions of the studied words will be eliminated as a result of this change in the make-up of the distractors. The second purpose of Experiment 2 was to find out whether we could replicate the FA and CR RT functions for the critical lure under the circumstances where all the test materials were semantically related, and (if as predicted) the true memory RT functions showed the normal linearly increasing trend as a function of MSS.
Method
Subjects
Three hundred and eleven 6 introductory psychology students participated in this experiment for fulfilling a research participation requirement for the course.
Design, Materials, and Procedure
The design, materials, and procedure were the same as those in Experiment 1 with two exceptions. First, the unrelated distractor probe in the four-trial block was replaced by a semantic associate distractor. The semantic associate distractor was randomly chosen from the remaining words in the 15-word DRM list after the memory-set words were selected. Second, the control condition was not used.
Results and Discussion
The mean rates of hits for the studied critical words and of the CR for the related distractor, the hit rate for the studied list words, and the FA and CR rates for the critical lure as a function of MSS are presented in Figure 5.
Mean hit rate of studied critical words, correct rejection (CR) rate of related distractors, hit rate of studied list words, and false alarm (FA) and CR rates of the critical lure as a function of memory-set size (MSS) of Experiment 2. To view this figure in colour, please visit the online version of this Journal.
The FA and CR rates of the critical lure showed a significant linear and a quadratic trend (again, the two test results were identical, see Table 1). The trend test results for the above response-rate functions are presented in Table 1. The overall mean FA rate for the critical lure did not change across the two experiments (both were .34, see Table 3).
The hit rate for the studied critical words showed neither a significant linear nor a significant quadratic trend, consistent with the corresponding results in Experiment 1. The hit rate of the studied list words showed a significant linearly decreasing trend, also consistent with the corresponding finding in Experiment 1. However, unlike the flat CR rate function of the unrelated distractor in Experiment 1, the CR rate of the related distractor showed a significant linear decrease over the MSS.
As shown in Table 3, the hit rate for the studied words and the CR rate of the distractors decreased from Experiment 1 to Experiment 2, indicating that the recognition discrimination for these probes was poorer when the distractor was semantically related than when it was unrelated.
The mean RTs of the above five responses as a function of MSS are presented in Figure 6. The statistics of the trend test results for these five RT functions are presented in Table 2.
Mean reaction time (RT) of hits of studied critical words, correct rejection (CR) of related distractors, hits of studied list words, and false alarm (FA) and CR of the critical lure as a function of memory-set size (MSS) of Experiment 2. To view this figure in colour, please visit the online version of this Journal.
As expected, the bowedness of the RT functions of the studied critical words and the studied list words disappeared (see Table 2). As indicated in Table 2, the studied critical word, the related distractor, and the studied list word all showed a significant linearly increasing trend, with no indication of a quadratic trend. This change supported the hypothesis that when one of the two distractors was an unrelated word (with the other being the critical lure), subjects relied progressively more on the semantic categorical information for a yes/no decision as the list length exceeded 10 items.
However, the change of the unrelated distractor to a related distractor did not seem to have affected the shapes of the FA and CR RT functions for the critical lure. That is, the patterns of the FA and CR functions for the critical lure were not affected by the relatedness of the distractor. Especially worth noting was that the positions of the cross-over point of the critical-lure FA and CR RT functions were almost identical. To confirm that the cross-over of the FA and CR RT functions of the critical lure in Figure 6 indicated a significant interaction, an ANOVA was conducted on the data of these two functions. The mean RT of the CR response (3768 ms) was significantly longer than that of the FA (3298 ms), F(1, 591) = 7.36, MSE = 5,164,059, p < .007. The MSS effect was also significant, F(5, 1362) = 2.89, MSE = 7,134,454, p = .013. Importantly, the response by MSS interaction was significant, F(5, 1362) = 8.22, MSE = 7,134,454, p < .0001. Thus, the crucial findings in Experiment 1 concerning the FA and CR RT function patterns were replicated and were shown to be independent of the semantic relatedness of the distractor and the shape of the RT functions of the studied words.
In the subspan section of the functions, the mean RT for the FA to the critical lure (3567 ms) was significantly longer than the corresponding CR mean RT (3043 ms), F(1, 432) = 5.27, MSE = 4,993,913, p = .022. In contrast, in the supraspan section, the mean RT of FA to the critical lure (3201 ms) was significantly shorter than that of the CR to the critical lure (4087 ms), F(1, 578) = 19.08, MSE = 13,457,794, p < .0001. Again, the reversed roles of the automatic and controlled processes in producing and checking false memory across the sub- and supraspan lists were replicated when the studied words showed a normal pattern of the RT/MSS functions.
General Discussion
Many studies reported that subjects could consciously re-experience the illusory contextual, physical, and temporal details associated with the presentation and studying of the critical lure. For example, they gave high confidence ratings and “remember” judgment, and even attributed list serial positions to the critical lure (Lampinen, Meier, Arnal, & Leding, 2005; Lampinen et al., 1999; Payne et al., 1996; Roediger & McDermott, 1995; Tulving, 1985). It has been suggested that the critical lure might be generated during encoding and rehearsed with the other list words (Cramer, 1970; Goodwin, 2007; Goodwin et al., 2001; Rhodes & Anastasi, 2000; Roediger, Balota, et al., 2001; Underwood, 1965). In addition, there were conceptual and perceptual priming effects of the semantic associates on the critical lure, suggesting that subjects might be conscious of that word (McDermott, 1997; see also Seamon, Luo, & Gallo, 1998). However, the present data failed to show any upward trend in the RT/MSS function of the false positive recognition of the critical lure, either in the subspan or in the supraspan lists, suggesting that the “memorized” critical lure was probably not being searched for in the same way as was a memory-set word.
On the other hand, when the critical word was externally encoded, its RT profile showed a linearly increasing trend with the increase in MSS in both experiments. We think that this test provided evidence showing that the memory of the critical lure is not the same as the memory of the words in the memory set. Also, regarding the issue of whether the false memory arises in the encoding or retrieving or both phases (Roediger, Balota, et al., 2001), the FA RT functions of the critical lure seemed to be more consistent with the idea of the false memory being the product of the retrieval stage than the product of the encoding stage. A related question is whether subjects responded false positively faster to the critical lure in larger MSSs because they were unwilling to spend the time to search for “the word in their memory” or because the identities of the memory-set items were not available due to a storage limitation. An increasing RT function with the increasing MSS is consistent with the search interpretation whereas a decreasing RT function is not. The CR to the critical lure showed an increasing RT function, suggesting that when subjects returned a correct “no” response to the critical lure, they probably performed a search of the memory set. In contrast, unlike the list-word hit function, the FA function of the critical lure showed a decreasing function, suggesting that the response was probably based on a sense of strong familiarity rather than a search. Thus, the faster critical-lure FA for the larger MSSs was probably not due to a storage limitation, but more likely due to a time-efficiency consideration. The increasing RT functions for the hits of the studied words also suggested that subjects probably performed the time-consuming search even for the very large MSSs of 10 or more items (except when the distractor was an unrelated word as in Experiment 1, and the MSS exceeded 10, at which point subjects appeared to increasingly rely on a semantic categorization strategy for a decision; see Figure 4).
Although the FA rate of the critical lure was relatively low, the subset of the responses on which the FA RT function was based met the behavioural criterion of false memory—that is, a positive recognition of a word that subjects did not study. Moreover, even though the FA rate of the critical lure obtained in this study was relatively lower than the typical DRM false recognition rates due to the averaged list length in this study being shorter than 15 items, its overall mean FA rate (around .34) was much higher than the FA rate of the related distractors (around .09; see Table 3). Hence, one cannot simply dismiss these FA responses as regular recognition errors. Further, a comparison of the critical lure CR rate function with the related distractor's CR rate function revealed that the slopes of these two functions showed a strikingly large difference (see Figure 5). In other words, the MSS effect was distinctively larger for the critical lure FA response (reciprocal of its CR) than for the same response to the related distractors. Therefore, the response rate profile of the false memory displayed the earmark features that separated it from those of the other probes. The same can be said of the false memory's RT profiles. That is, overall, responses to the critical lures were slower than those to the other probes, consistent with reports in other studies (e.g., Cabeza, Rao, Wagner, Mayer, & Schacter, 2001; Jou, 2011; Jou et al., 2004), and also the slopes of the critical lure RT functions were much steeper than those of the other probes (see Figures 4 and 6). These distinctive features of false memory revealed by this test paradigm indicated that there are objective differences between true and false memories despite a lot of subjective similarities between these two types of memories (D. A. Gallo, 2006; Jou & Flores, 2013).
Previous studies used an increased FA or false recall rate of the critical lure resulting from an increased number of semantic associates in the study list as evidence of increased spread activation. On the other hand, a decreased FA rate in recognition or a decreased false recall rate due to a warning against the nonpresented thematic word is taken as evidence of monitoring (Robinson & Roediger, 1997; Roediger, Balota, et al., 2001). In the present study, by examining RT as a function of MSS, we captured these two processes in action. We showed that as the spread activation accrued from additional associates, the FA response to the critical lure speeded up, but at the same time, the successful monitoring process as reflected by the CR responses took a progressively longer time. The RT pattern of these two responses in the supraspan section was consistent with findings in some other studies in which rejection of a familiar distractor was found to take a longer time than acceptance of it because the rejection of a familiar distractor requires the retrieval of contextual or episodic details (Gronlund & Ratcliff, 1989; Hintzman & Curran, 1994; Rotello & Heit, 2000).
The RT pattern of these two responses in the subspan section, however, has not been reported in the literature (at least to our knowledge). The importance of this finding is that the RT profiles of the FA and CR of the critical lure in the subspan section of the MSS showed a pattern opposite to what the dual-process theories of recognition (Jacoby, 1991; Yonelinas, 2002) and the activation/monitoring theory of false memory (D. G. Gallo & Roediger, 2002; Roediger, Balota, et al., 2001) will predict. For the supraspan section, the FA response was faster than the CR response, suggesting that it derived from a familiarity-based, faster, activation spreading process, which is what the dual-process theories and the activation/monitoring theory predict. However, the pattern of the RT results in the subspan section showed that, relative to a CR of the critical lure, the acceptance of the critical lure appeared to be a slower, more strategic process. This pattern of results suggests that the false recognition in the subspan condition may not be driven by an automatic spreading of activation, but probably by a slow, deliberate, constructive process, which typically takes place at a later stage in retrieval (Brainerd et al., 2002; also see Jou, 2008, for evidence of a dual-stage critical-lure output in recall).
The slower FA (than the CR) in the subspan section might be similar to a misguided second thought that opposed an otherwise faster, accurate CR. If subjects had followed their faster, more spontaneous initial reaction to the probe, they would have made the correct response of rejecting the critical lure. Thus, just as the FA to a critical lure can be driven by two processes of opposite natures, so can the correct response (i.e., a CR) to the critical lure across the sub- and the supraspan MSSs. In the subspan section, the CR to the critical lure appeared to be a relatively quicker, more intuitive, direct process, whereas in the supraspan section, it appeared to be a slow, controlled process. Again, the dual-process theories of memory and the activation/monitoring theory of false memory typically characterize the two processes with the properties as revealed in the supraspan section of the DRM list, with a quick, familiarity-based process generating false memory, and a slow, recollection-based process curbing false memory (Jacoby, 1991; Yonelinas, 2002). To our knowledge, not much research has been done on the reverse version of the dual process found in this study in which a quicker, automatic-like response can actually be correct, whereas a slower, strategic-like judgment can be error-prone. By using the DRM semantic associate lists, we demonstrated reliably in two experiments under what condition the standard dual process was in operation and under what condition the reverse dual process could occur.
As well, we demonstrated that the true and false memories behave in distinctively separable ways in this paradigm. The hit RT of a true memory item including the studied critical item (in Experiments 1 and 2) increased with the increase in MSS whereas the FA RT of the critical lure decreased with the increase in MSS. Since RT reflects the activation level, this indicates that the activation level of a true memory item decreases with MSS whereas that of a false-memory item increases. The important implication of this fact is that anyone who endorses the idea that the longer the DRM list, the stronger the activation of the critical lure (e.g., Robinson & Roediger, 1997) should also recognize that this very phenomenon is the earmark of false memory. Again, an externally encoded piece of information will decrease in activation as the list of items becomes longer. Regardless of what other subjective, phenomenological characteristics are ascribed to false memory, the positive correlation of the critical-lure activation level with the list length is the objective evidence of false memory.
Before considering the theoretical implications of our findings to conclude the discussion, we ask the question of why the critical-lure FA RT trend that we observed was the opposite of what Coane et al. (2007) did. As noted earlier, they found both a significant increase in the critical-lure FA rate from MSS 5 (.08) to MSS 7 (.22) and a significant RT increase from MSS 5 to MSS 7 in the same experiment. They interpreted the increase in the critical-lure FA RT as a result of increased monitoring as the MSS increased. Regardless of the interpretations, we thought that the different results were probably the results of differences in methods and procedures between the two studies. We therefore carefully compared the details in the methods of the two studies. We found that there were many design, material, and procedural differences between their experiments and ours. Among them were (a) they compared only two MSSs, 5 and 7, whereas we compared 6 MSSs, from 4 through 14; (b) their subjects studied a memory set 4 times in Experiment 1, and 3 times in Experiment 2, whereas our subjects studied it only once; (c) their “no” response trials made up 75% (Experiment 1) and 66% (Experiment 2) of the trials whereas ours made up only 50%; (d) we used a 20-s counting filled task in the retention interval whereas they did not. Any or a combination of these methodological variations could possibly have produced the different RT results of the FA response to the critical lure across the two studies. With that many design, material, and procedural differences between Coane et al.'s study and our study, it is not possible to answer why the results were different.
Finally, can the two major theories of false memory, the activation/monitoring (D. G. Gallo & Roediger, 2002; Roediger, Balota, et al., 2001) and the fuzzy-trace theories (Brainerd & Reyna, 1998; Brainerd, Reyna, & Brandse, 1996; Reyna & Brainerd, 1995), explain the main findings from this study—that is, the RT cross-over interaction of response with MSS factor for the critical lure? As noted earlier, the activation/monitoring theory can account for the critical-lure FA and CR RT function pattern in the supraspan portion of the RT functions in that the activation of the critical lure increases as the list length of semantic associates increases, and that the spread of activation from the semantic associates to the critical lure is automatic and fast whereas the monitoring process is controlled and slow. But can it handle the reversed RT function pattern of these two responses in the subspan portion of the MSS scale? We realize that this theory can explain the RT data of the full MSS range by having the definition of monitoring tweaked slightly. The monitoring process is thought to be a controlled, slower, and strategic operation to oppose a faster automatic process (Jacoby, 1991). When it wins over the faster, automatic initial process, it produces a slower response and one that is the opposite of the initially triggered, faster, automatic tendency (which is “killed” internally before it is actualized in an explicit response). And the stronger this initial tendency is, the slower and more effortful the overriding monitoring process will be. Accordingly, if the initial, more intuitive response tendency is correct, the slower, more strategic, overriding response derived from monitoring will be wrong since it is assumed to oppose the initial response tendency. Thus, if it is assumed that when people re-consider an initial spontaneous decision, not only is the second-guess decision slower but it is also the opposite of the initial decision, then the activation/monitoring theory can explain both the sub- and supraspan sections of the FA and CR RT functions. In other words, in the subspan section of the function, the initially triggered spontaneous tendency is a no response (and this tendency is stronger the shorter the list), but the strategic, slow monitoring process generates the opposite yes response (unfortunately a wrong one) and overrides the initial no response.
The concept of an opponent and trade-off relationship between the gist and verbatim memories implied in the fuzzy-trace theory can also handle the cross-over interaction and the whole spectrum of results displayed by the two critical lure's RT functions. The idea is that when the number of items falls within the span limit, and the individual items can be tracked verbatim, the verbatim memory predominates. Because verbatim memory drives the CR, and gist memory drives the FA response, a stronger verbatim memory generates a faster CR and a slower FA response. When the number of items exceeds the span limit, not every individual item can be easily tracked. As a result, the gist memory predominates. A stronger gist memory coupled with a weaker verbatim memory generates a faster FA accompanied by a slower CR response. The gradual shift in the relative potency of these two processes resulted in a dynamic reversal of the dominance relationship between gist and verbatim memories at about MSS 7, which therefore marked the boundary between the sub- and supraspans for false memory. The ordinal reversal of the CR and FA RT functions across the memory span boundary also suggests that false memory is primarily generated by an automatic process in the supraspan memory sets (which accords with the common notion about the source of false memory), but probably by a strategic mechanism in the subspan sizes of memory sets.
The dual false-memory retrieval process mirrors the dual process in veridical-memory retrieval from short and long lists in the Sternberg task (Burrows & Okada, 1975; Corballis et al., 1980; Okada & Burrows, 1978) and in other memory tasks (Jou & Aldridge, 1999). Thus, there seems to be a dual-process of retrieval associated with sub- and supramemory spans for false memory as well. An important implication of the present findings for recognition memory research in general is that memory span or limitation (Cowan, 2001; Cowan et al., 2008) should be taken into consideration as a relevant factor whether one studies true memory or false memory.
Footnotes
1
The standard materials in the Sternberg paradigm are digits and letters, the so called high frequency items of a small vocabulary (Wickens et al., 1985), whereas in this study, the materials were semantic associates. The list length in the standard Sternberg paradigm ranges from 1 to 6 items, whereas in our study the shortest list was 4 items long and the longest list 14 items long. In the standard paradigm, no interpolated task is used, whereas in this study, a 20-s interpolated counting task was used to boost the false memory rate, which presumably made the item retrieval to be from the secondary (long-term) rather than from the immediate memory regardless of the length of the list. For these reasons, the term Sternberg paradigm is used in a very broad sense in this paper with it mainly referring to its basic assumption of a positive correlation between the RT and memory-set size (MSS). Because of these changes, we do not expect that every aspect of our results will correspond to what is observed in the standard Sternberg paradigm. In addition to other differences, we expect that all these changes will lengthen the absolute recognition RT (Roediger et al., 1977) compared with the typically obtained RTs in a standard Sternberg task.
2
Jou and Aldridge (1999) found a similar bilinear RT function in an alphabet serial position and interval distance estimation task (e.g., “What is the alphabetic serial position of E?” “What is the alphabetic step distance between E and J?”), with the alphabetic quantitative information within the subspan range producing a steeper linear slope and that of the supraspan range a flatter slope. Based on this finding, they suggested that the process of retrieving alphabetic position and interval information from subspan ranges may be serial whereas the retrieving process from supraspan ranges is predominantly parallel. Therefore, this dual-mode retrieval pattern seems to hold for both episodic and semantic memories.
3
We found out the word frequency counts for the related words in the DRM lists from Davies (2008) The corpus of contemporary American English. The average frequency was 101 per million words, with a standard deviation of 312, and a range of 0.06 to 3792. Thus, the frequency range of the semantically related words was much wider than that of the unrelated words although the average frequencies for these two types of words were close. Each list of unrelated words was selected by the first author and was reviewed by one of his associates. When a disagreement arose, a different word was selected on which a consensus could be reached.
4
Subjects in this experiment were constantly switching between counting backward and memory recognition test. The switching between the tasks incurs a task-switching cost. Because we tested only one probe for each list, that probe was the sole bearer of the task-switching cost (see Jou, 2014, for how it affects RT in the Sternberg paradigm; and see Mayr & Kliegl, 2000; Monsell, 2003; Monsell et al., 2003; Rogers & Monsell, 1995, for its effects in other types of experiments). Therefore, we expect to find an overall elevation of RTs across the board, but especially for the FA and CR RTs to the critical lure to be even longer than was found in previous studies (e.g., Jou et al., 2004). Also, previous studies found RT of false alarm (FA) to the critical lure to be longer than hit RT to the studied words (Cabeza et al., 2001; Jou, 2011; Jou et al., 2004), and CR RT to the critical lure to be even longer than its FA RT (Jou et al., 2004). We expect to observe the same RT magnitude order for these three types of responses in the present study.
5
In Jou et al. (2004), also using DRM semantic associate lists but with a fixed-set testing procedure in which multiple probes were presented for test, a typical RT for a hit to the studied word was about 1,600 ms, that of a FA to the critical lure was about 2,000 ms, and that of a CR to the critical lure was 3,000 ms. As Jou (2014) demonstrated, the RT of the first trial in a recognition task of this type can be up to 80% slower than the trials past the task-switching-cost phase (e.g., past the 3rd trial). When 80% of 2,000 ms and 3,000 ms (critical lure FA, and CR RT, respectively, obtained in Jou et al., 2004) is added to each RT due to the task-switching cost, it is about 3,600 ms for the FA, and over 4,000 ms for the CR of the critical lure, which were roughly what we obtained for these two responses in this experiment. In a typical DRM recognition experiment, MSS and the number of probes at test are not varied; therefore, using a fixed-set testing procedure is not a problem. However, as indicated by Jou (2014), when MSS is varied as in this study, using a fixed-set testing procedure introduces confounds (RTs of shorter lists with fewer probes at test are impacted by the task-switching cost to a greater extent than longer lists). Hence, the single-probe test is the only confound-free way to measure the recognition time with the result of obtaining an overall slower response. In addition, a 20-s interpolated counting task could also slow down the recognition response (see Roediger et al., 1977).
6
Starting from the semester when Experiment 2 was conducted, research participation became a requirement for the completion of the introductory psychology course. There were many more introductory psychology students than available research projects could accommodate in that semester. Therefore, one reason that we tested more subjects in Experiment 2 than in Experiment 1 was to lower the number of incomplete grades for the introductory psychology courses. However, this measure also helped towards meeting the needs of the experiment for a large sample. Given that only one RT was collected for a list of words, the RT variability was very high, and the frequencies of FAs for the critical lure was low for the small MSSs. Also, due to a random selection of items from the 15 words of DRM lists for different MSSs, the association strengths between the selected words (Roediger, Balota, et al., 2001) for different MSSs (especially the small MSSs) might not be well equated. However, a large sample was expected to minimize the discrepancies in backward association strength (BAS) for small and large MSSs.
Acknowledgements
We thank Sarah Silva and Clarissa Garcia for help with collecting part of Experiment 1 data, and Liann Pena for collecting part of Experiment 2 data. We also thank Mark Huff, Charles Brainerd, and an anonymous reviewer for helpful comments on a draft of this article.