0 évaluation0% ont trouvé ce document utile (0 vote)
13 vues19 pages
Researchers investigate whether a single dimension (e.g., strength) is sufficient to account for recall and judgments of learning (JOLs) they find repeated support for the sufficiency of a single dimensions for both recall and JOLs. The answer will facilitate development of a parsimonious and consistent theory of JOLs, they say.
Researchers investigate whether a single dimension (e.g., strength) is sufficient to account for recall and judgments of learning (JOLs) they find repeated support for the sufficiency of a single dimensions for both recall and JOLs. The answer will facilitate development of a parsimonious and consistent theory of JOLs, they say.
Researchers investigate whether a single dimension (e.g., strength) is sufficient to account for recall and judgments of learning (JOLs) they find repeated support for the sufficiency of a single dimensions for both recall and JOLs. The answer will facilitate development of a parsimonious and consistent theory of JOLs, they say.
How Many Dimensions Underlie Judgments of Learning and Recall?
Evidence From State-Trace Methodology
Yoonhee Jang and Thomas O. Nelson University of Maryland The authors used state-trace methodology to investigate whether a single dimension (e.g., strength) is sufficient to account for recall and judgments of learning (JOLs) or whether multiple dimensions (e.g., intrinsic and extrinsic factors) are needed. The authors separately manipulated the independent variables of intrinsic and extrinsic cues, determining their state traces for recall and JOLs. In contrast to the supposition that intrinsic cues have similar effects on both recall and JOLs whereas extrinsic cues affect JOLs less strongly than recall (i.e., 2 dimensions underlying recall and JOLs), the authors found repeated support for the sufficiency of a single dimension for both recall and JOLs (not only immediate JOLs but also delayed JOLs) across a variety of intrinsic and extrinsic cues. Keywords: metacognition, judgments of learning (JOLs), state-trace analysis, intrinsic and extrinsic cues, single-dimensional versus multidimensional theories of JOLs This research concerns a kind of metacognitive monitoring known as judgments of learning (JOLs), which are judgments that occur during or after acquisition and are predictions about future test performance on recently studied items (Nelson & Narens, 1994, p. 16). JOLs are one of the most frequently investigated self-monitoring judgments and have been investigated across diverse areas of psychology (reviewed in Schwartz, 1994). In spite of the large amount of research on JOLs, there is little consensus about what kind of theoretical structure underlies them. In this research, we take an initial step toward under- standing this structure by asking whether a single dimension provides a sufficient account of JOLs or whether multiple dimensions are needed. The answer to this question will facil- itate development of a parsimonious and consistent theory of JOLs and, more generally, of metacognitive monitoring. Al- though this particular question has not been addressed previ- ously, several theories have been proposed that bear on this issue. Single-Dimensional Theories of JOLs Direct-access (i.e., trace-access) theory proposes that people monitor memory content, assessing the magnitude along some underlying single dimension (e.g., memory strength in Busey, Tunnicliff, Loftus, & Loftus, 2000; Loftus, Oberg, & Dillon, 2004). According to direct-access theory, a strong correspon- dence is expected between recall and JOLs because recall and JOLs are affected by the same underlying factor. The observed correlation between JOLs and recall performance may be less than unity because of noise, limited degrees of fineness in the JOLs, and other stochastic issues that can be accounted for by probability theory. Other ideas for a single dimension underlying JOLs include ease of processing (e.g., Begg, Duft, Lalonde, Melnick, & Sanvito, 1989) or retrieval fluency (e.g., Benjamin, Bjork, & Schwartz, 1998), although it has not been claimed specifically that these mechanisms are solely responsible for JOLs. Instead, the claim is that one of these mechanisms may underlie both recall and JOLs, but this will happen only to the extent that recall relies on this mechanism (e.g., retrieval fluency may sometimes mislead people into giving inappropriately high JOLs when it has a greater effect on JOL magnitude than on recall; Benjamin et al., 1998). Multidimensional Theories of JOLs: Intrinsic and Extrinsic Cues From the Cue-Utilization Framework Koriats (1997) cue-utilization framework proposes that peo- ple evaluate several different cues that are differentially pre- dictive of subsequent recall and that JOLs are based on heuris- tics that attempt to forecast the likelihood of recall. For instance, research by Begg et al. (1989; replicated and extended by Wixted, 1992) found that although JOL magnitude was greater for high-frequency words than for low-frequency words, Yoonhee Jang and Thomas O. Nelson, Department of Psychology, University of Maryland. This research was partially supported by Cognition and Student Learning (CASL) Research Program Grant R305H030283 from the Institute of Education Sciences of the U.S. Department of Education. We thank Donald Bamber, Morris Goldsmith, David Huber, Thomas Wallsten, and Michael Dougherty for their many valuable comments and Geoffrey Loftus for useful information about his recent studies. Yoonhee Jang is deeply indebted to Thomas O. Nelson, who passed away January 14, 2005, for his careful guidance throughout her grad- uate career. Correspondence concerning this article should be addressed to Yoonhee Jang, Department of Psychology, University of Maryland, College Park, MD 20742. E-mail: yjang@psyc.umd.edu Journal of Experimental Psychology: General Copyright 2005 by the American Psychological Association 2005, Vol. 134, No. 3, 308326 0096-3445/05/$12.00 DOI: 10.1037/0096-3445.134.3.308 308 subsequent recognition performance was greater for low- frequency words than for high-frequency words. This crossover interaction 1 was interpreted as establishing that more than one factor is required to explain both JOLs and memory perfor- mance. However, recent research by Benjamin (2003) revealed that JOLs correctly predict both recognition (better perfor- mance for low-frequency words) and recall (better performance for high-frequency words), although that did not occur in the initial study/test sequence, suggesting that the single- dimensional account is contingent on metacognitive knowledge acquisition. The cue-utilization framework assumes that intrinsic cues in- volve characteristics of the study items that are perceived to disclose the items a priori ease or difficulty of learning (Koriat, 1997, p. 350), whereas extrinsic cues are factors that pertain either to the conditions of learning or to the encoding operations applied by the learner (p. 350). For instance, Koriat (1997) proposed item difficulty and item relatedness as prototypical ex- amples of intrinsic cues and number of study presentations and study duration as prototypical examples of extrinsic cues. On the basis of scale-dependent interactions (which are potentially prob- lematic, as discussed below), Koriat speculated that intrinsic cues have approximately the same effects on recall as they have on JOLs, whereas extrinsic cues have greater effects on recall than on JOLs, which he interpreted as requiring a multidimensional struc- ture underlying recall and JOLs. The Present Approach and Formulation of the Problem By separately manipulating intrinsic and extrinsic cues, we tested a single-dimensional account of JOLs and recall versus a multidimensional account. Because it is often assumed that recog- nition involves more than one process (e.g., memorability vs. discrimination), we considered only recall in the present study, focusing our question on whether a single factor underlies both JOLs and memorability. We conceptualize the relation between JOLs and recall in the following way. If JOLs arise from only a single dimension, then independent variables can affect only that one underlying dimen- sion (perhaps with noise and/or bias). However, if JOLs arise from multiple underlying dimensions, then the independent variables can differentially affect those underlying dimensions. For instance, Figure 1 illustrates the single-dimensional model in which the two independent variables (e.g., item difficulty and number of study presentations) are assumed to affect a single dimension of the memory representation (D1), which determines both JOLs and recall: JOLs f(D1), and recall g(D1), where f and g are positive monotonic functions, with f not necessarily identical to g. Thus, both JOL magnitude and recall are assumed to be monotonic functions of the same single dimension of the memory representation, D1. In contrast to this single-dimensional account, one possible multidimensional model is illustrated in Figure 2, which is based on the hypothesis that intrinsic cues have the same effects on both recall and JOLs whereas extrinsic cues affect recall more strongly 1 Crossover interactions are important because they allow conclusions of an underlying multidimensional structure, in contrast to converging/diverg- ing interactions, which do not require an assumption beyond a single- dimensional structure (e.g., Dunn & Kirsner, 1988). Converging/diverging interactions (i.e., scale-dependent interactions) also have problems of meaningfulness (Krantz & Tversky, 1971; Loftus, 1978; Townsend & Ashby, 1984), as elaborated in the General Discussion section. Figure 1. A single-dimensional model for the relation between two independent variables (i.e., item difficulty as an intrinsic cue and number of study presentations as an extrinsic cue) and two dependent variables (i.e., judgments of learning [JOLs] and recall). The single underlying dimension is referred to as D1, and both f and g are monotonic functions with f not necessarily identical to g. 309 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL than JOLs. This multidimensional model (which is only one of many possible multidimensional models) contains two underlying dimensions (D1 and D2), where D1 is a monotonic function of both item difficulty and number of study presentations, and D2 is a monotonic function of only number of study presentations (i.e., a two-dimensional model). JOL magnitude is primarily determined by D1 as in the single-dimensional model. Although JOL magni- tude may also be affected by D2, the effect is small or nonexistent, as indicated by the dotted arrow between D2 and JOLs. By contrast, recall is equally affected by both D1 and D2: JOLs f(D1), and recall g(D1, D2), where f and g are positive mono- tonic functions, as described above. Hence, item difficulty exerts similar effects both on JOLs and on recall, whereas number of study presentations affects JOLs less strongly than recall. This particular two-dimensional model is in accord with the cue- utilization framework. State-Trace Analysis We utilized state-trace analysis (originally proposed by Bam- ber, 1979) to test the number of dimensions underlying JOLs and recall. State-trace analysis is related to conjoint measurement theory (e.g., this relationship is elaborated on by Loftus et al., 2004) and the logic of additive and multiplicative effects (see Loftus, 2002). State-trace analysis is achieved by means of a scatter-plot graph showing the covariation of two dependent vari- ables and the manner in which the independent variables affect the dependent variables. A major goal of state-trace analysis is to competitively evaluate conceptualizations of single-dimensional and multidimensional theoretical structures by evaluating curves, referred to as state traces, within such a scatter plot. According to a review by Dunn and James (2003), State-trace analysis comprises a conceptual framework within which models of the relationships between different dependent variables can be represented. It also incorporates a method for identifying and testing these relationships. . . . Thus, if two dependent variables are functions of the same latent variable, the resulting state-trace is a one-dimensional curve in two-dimensional state space. (pp. 404405) The application of state-trace analysis to the present situation is illustrated in Figure 3, which shows predicted outcomes from the Figure 1 single-dimensional model (see Figure 3A, 3B, and 3C) as compared with predicted outcomes from the Figure 2 two- dimensional model (see Figure 3D, 3E, and 3F). The first two panels of each row illustrate the separate effects of the independent variables on the two dependent variables of recall (see Figure 3A and 3D) and JOL magnitude (see Figure 3B and 3E). In isolation, these first two panels do not qualitatively differentiate between the single-dimensional account versus the two-dimensional account of the dependent variables. Instead, this qualitative model compari- son is achieved by combining the two plots into state-trace plots, as shown in the third panel of each row. As illustrated in Figure 3C, the critical prediction of the single- dimensional model is that both the one-presentation and two- presentation curves lie along a single curve. The location of each curve arises from the causal paths shown in Figure 1: (a) from item difficulty through D1 to JOLs, and from item difficulty through D1 to recall; and (b) from number of study presentations through D1 to JOLs, and from number of study presentations through D1 to recall. The critical aspect of the state-trace analysis is highlighted by the two arrows in Figure 3C, which identify the overlapping portion of the one-presentation and two-presentation curves, with the upper arrow identifying easy items that have been presented once and the lower arrow identifying difficult items that have been Figure 2. A hypothetical two-dimensional model for the relation between two independent variables (i.e., item difficulty as an intrinsic cue and number of study presentations as an extrinsic cue) and two dependent variables (i.e., judgments of learning [JOLs] and recall). The two underlying dimensions are referred to as D1 and D2, and f and g are monotonic functions with f not necessarily identical to g. The dotted arrow between D2 and the JOLs indicates that there is little or no effect of D2 on JOLs. 310 JANG AND NELSON presented twice. In the single-dimensional model, different condi- tions giving rise to a particular JOL rating must also give rise to a particular recall performance level (and vice versa), considering that any specific value on a dependent variable is obtained only through dimension D1. Therefore, the critical test of the single- dimensional model is the extent to which the two curves fall atop each other. This can be formally stated as follows. Assume that C(i, j) represents the joint condition of the ith level of one independent variable (where i can take on the values of, say, a or b) and the jth level of another independent variable (where j can take on the values of, say, q or r). Then, according to the single-dimensional model in Figure 1, the following must be the case: (a) If C(a, r) produces the same JOL as does C(b, q), then C(a, r) must produce the same value of D1 as did C(b, q); and (b) because C(a, r) and C(b, q) have the same value of D1, they must produce the same recall. In short, the strong prediction of the single-dimensional model is that whenever JOL magnitude is the same for C(a, r) as for C(b, q), then recall must be the same for C(a, r) as for C(b, q). Hence, in Figure 3C, the two curves must be atop each other throughout the range in which the two curves have the same values of recall (or the same magnitudes of JOLs), as indicated by the portion of the two curves between the two arrows. By contrast, the data pattern that definitively specifies a multi- dimensional interpretation is one in which the two curves are separated, such as in Figure 3F. The two-presentation curve could fall either to the left or to the right of the one-presentation curve, with each pattern indicating the particular connection strengths of the multidimensional structure. For instance, the prediction from the cue-utilization framework (from Figure 2) is illustrated in Figure 3. Predictions of the two models from Figures 1 and 2 (the top three panels show illustrative outcomes predicted by the single-dimensional model, and the bottom three panels show illustrative outcomes predicted by the two-dimensional model from the cue-utilization framework). Panels A, B, D, and E show traditional data in which the two dependent variables (i.e., recall and judgments of learning [JOLs]) are plotted as functions of the two independent variables (i.e., item difficulty and one vs. two presentations). Panels C and F show state traces in which JOL magnitude is plotted against the percentage of correct recall. The data shown between the two arrows in Panel C are the overlapping portion of the one-presentation curve and the two-presentation curve. 311 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL Figure 3F, wherein the two-presentation curve falls to the right of the one-presentation curve. The location of these two curves arises from the causal paths shown in Figure 2. For this example, JOLs are determined by only D1, and so both of the presentation conditions provide the same value of JOLs (viz., 55%). However, because recall is determined by both D1 and D2, the percentage of correct recall is higher in the two-presentation condition (e.g., 80%) than in the one-presentation condition (e.g., 60%). Hence, the two-presentation curve in Figure 3F falls to the right of the one-presentation curve (instead of the two curves falling atop each other as in Figure 3C). It is important to note that a multidimensional model is capable of producing a single curve, such as would be the case for equally weighted inputs and outputs to each of the dimensions. However, if such a single-curve pattern is observed repeatedly, across dif- ferent situations, this indicates at a minimum that the multiple dimensions are entirely redundant and function as a single dimen- sion. For this reason, a single experiment finding a single curve cannot definitely rule out the possibility of multiple dimensions that happen to have conspired to produce a single curve for that particular situation. It is for this reason that claims of a single dimension are most effectively made by examining multiple experiments. It is also important to note that traditional tests of interactions between independent variables appear to test something similar to that tested using state-trace analysis, but there are critical differ- ences. Specifically, interactions in an analysis of variance (ANOVA) assume a linear combination of variables, whereas state-trace analysis allows for nonlinearities in the relationship between the underlying dimensions and the dependent measures. Indeed, it is possible to observe a significant interaction even though the state traces lie along a single curve. This fact serves to highlight the inherent limitations of the general linear model. Specific Goals of the Present Research The primary goal of the present research was to assess the number of dimensions required for JOLs and recall. Specifically, we wanted to empirically test the predictions of the single- dimensional model to determine whether it is sufficient to account for JOLs and recall or, alternatively, to determine whether a multidimensional model is necessary (e.g., the cue-utilization frameworks version of a two-dimensional model). A secondary aim of our experiments was to separately examine this relationship for delayed JOLs as well as for immediate JOLs (which were the only kind of JOLs investigated by Dunlosky & Matvey, 2001, and by Koriat, 1997). Delayed JOLs typically predict subsequent recall better than do immediate JOLs (Nelson & Dunlosky, 1991). There is not yet a consensus explanation for this delayed-JOL effect, which can be explained through a variety of mechanisms (e.g., Kimball & Metcalfe, 2003; Nelson & Dun- losky, 1992; Spellman & Bjork, 1992; Weaver & Kelemen, 1997). Although cursory consideration of that issue suggests different mechanisms for immediate and delayed JOLs, some recent studies have proposed that the delayed-JOL effect may instead correspond to different settings within a single underlying mechanism (e.g., Nelson, Narens, & Dunlosky, 2004). Nelson et al. (2004) reported that JOL accuracy differences can be explained by the differential breakdown in the number of dyads comprising immediate and delayed JOL accuracy. Most of the dyads for immediate JOLs consist primarily of items that can be both recalled at the time of the JOL, whereas most of the dyads for delayed JOLs consist primarily of one item that can be recalled and one item that cannot be recalled at the time of the JOL. In other words, for immediate JOLs, discrimination between the two items is relatively difficult (i.e., discrimination between recalled items), whereas for delayed JOLs, discrimination between the two items is relatively easy (i.e., discrimination between a recalled item vs. a nonrecalled item). It is important to note that this explanation appeals only to a natural increase in variability of memorability as a function of delay and is consistent with a single-dimensional interpretation of JOLs and recall. Thus, we wanted to evaluate the state-trace plots for both immediate and delayed JOLs. This procedure allowed additional opportunitiesnot only from immediate JOLs but also from de- layed JOLsfor the single-dimensional model to fail and for the necessity of a multidimensional model to be confirmed through the observation of separate state traces. Overview of the Experiments All experiments used paired study and cued recall testing, with JOLs given in response to the cue word. Separate experiments manipulated the intrinsic cue of item difficulty or item relatedness and the extrinsic cue of number of study presentations or study duration. One intrinsic cue and one extrinsic cue were manipulated in each experiment, and all experiments included both delayed and immediate JOLs. We used a 2 2 2 repeated measures design for each experiment in which the three independent variables were an intrinsic cue (viz., easy vs. difficult items for Experiments 1A and 1C; related vs. unrelated items for Experiments 1B, 1D, and 2), an extrinsic cue (viz., one vs. two presentations for Experiments 1A, 1B, and 2; short vs. long presentation for Experiments 1C and 1D), and timing of JOLs (viz., immediate vs. delayed JOLs). The two dependent variables were the percentage of correct recall and JOL magnitude. For each experiment, we investigated the assumption that the independent variables of intrinsic and extrinsic cues had significant effects on both recall and JOL magnitude. Significance of these effects is a critical prerequisite for state-trace analysis because, otherwise, an observation of a single trace could result from a null effect for one of the independent variables. In all, the experiments yielded 10 state-trace plots that were used to compet- itively evaluate the predictions from the single-dimensional model (see Figure 3C) versus a multidimensional model (e.g., the two- dimensional model from the cue-utilization framework; see Figure 3F). Experiments 1A1D For Experiments 1A and 1C, the manipulated intrinsic cue was item difficulty, whereas for Experiments 1B and 1D, it was item relatedness. For Experiments 1A and 1B, the manipulated extrinsic cue was number of study presentations, whereas for Experiments 1C and 1D, it was study duration. During the study phase, partic- ipants were instructed to learn each pair and to make a JOL about the likelihood that they would subsequently be able to recall the target word when the cue word was presented. During the test 312 JANG AND NELSON phase, they were instructed to recall the target word in response to each cue word. The main hypotheses under investigation were (a) in accord with the single-dimensional model, the state-trace curves should fall atop each other, versus (b) if the two curves do not fall atop each other, then a multidimensional model is needed (e.g., for the two-dimensional model from the cue-utilization framework, the two-presentation state-trace curve should fall to the right of the one-presentation state-trace curve in Experiments 1A and 1B, and the long-presentation state-trace curve should fall to the right of the short-presentation state-trace curve in Experiments 1C and 1D). Method Participants. Forty-five volunteers from undergraduate psychology courses at the University of Maryland received course credit in return for their participation in each of Experiments 1A1D. Participants in all experiments of this study were treated in accord with the Ethical Princi- ples of Psychologists and Code of Conduct (American Psychological Association, 1992). Materials. For Experiments 1A and 1C, 64 SwahiliEnglish transla- tion equivalents were drawn from the norms of Nelson and Dunlosky (1994) according to the normative likelihood of the English word being recalled when the Swahili word was presented. Thirty-two pairs were normatively easy pairs (e.g., yaiegg) with a mean normative probability of recall of .25 (range .18.55), and 32 were normatively difficult pairs (e.g., nafakacorn) with a mean normative probability of recall of .05 (range .02.08). For Experiments 1B and 1D, 64 nounnoun pairs were constructed on the basis of the pairs used by Dunlosky and Matvey (2001). They were divided into two lists according to the degree of associative relatedness between the two nouns comprising each pair. Thirty-two pairs consisted of nouns that were moderately related (e.g., stovekitchen), and the remaining 32 pairs consisted of nouns that were not obviously related (e.g., bottlecalendar). In all experiments, the first 8 pairs constituted practice, and the last 8 pairs were excluded from recall so as to prevent recency effects. The remaining 48 pairs comprised two blocks of 24 pairs per block and were the only pairs that were analyzed. Procedure. Participants in Experiment 1A studied SwahiliEnglish word pairs and then indicated their JOL for each pair when the Swahili word appeared alone as the cue for the English word. During study, each SwahiliEnglish pair was presented in the center of the screen for 9 s. All pairs were randomly ordered anew for each participant, with the restriction that at least four pairs separated the two presentations of a given two- presentation pair. A self-paced JOL occurred for each pair and was prompted with only the Swahili word and the question How confident are you that in about 10 minutes from now you will be able to recall the second word of the pair when prompted with the first? The participants reported their estimate on a scale ranging from 0 (definitely will not recall) to 100 (definitely will recall) (e.g., 20 20% sure, 40 40% sure, 60 60% sure, and 80 80% sure). Immediate JOLs versus delayed JOLs were randomly assigned to the pairs, with the restriction that one half of the pairs in each condition received immediate JOLs and the other half received delayed JOLs. Each immediate JOL occurred immediately after the offset of the pair. After the final immediate JOL or the final study trial of a given block of 24 pairs, JOLs occurred for the first third of the pairs that were slated to receive delayed JOLs within that block; then, JOLs occurred for the second third of the pairs slated to receive delayed JOLs within that block, followed by the JOLs for the last third of the pairs slated to receive delayed JOLs within that block. The order of presentation of the pairs within each third of the pairs was randomly determined anew from study to delayed JOLs. During the self-paced test phase, the participants were instructed to attempt to recall the English translation equivalent when cued by a Swahili word. If they had no guess, then they typed NEXT to proceed to the next test trial. The procedure of Experiment 1B 2 was identical to that of Experiment 1A except that, during the study phase, each pair was presented for 5 s and the first noun of the pair was the cue word during the JOL and recall phases. The procedure of Experiment 1C was identical to that of Experiment 1A except that, during the study phase, one half of the pairs were presented for 5 s per pair and the other half were presented for 15 s per pair. For Experiments 1C and 1D, no pair was presented more than once, and the order of presentation of each pair was randomized anew for each participant. The procedure of Experiment 1D was identical to that of Experiment 1B except that, during the study phase, one half of the pairs were presented for 2 s per pair and the other half were presented for 8 s per pair. Results and Discussion For each experiment, we first report the outcome for the pre- requisite of significant effects of intrinsic and extrinsic cues on both recall and JOL magnitude. The descriptive statistics are shown in Figures 4A, 4B, 4D, and 4E8A, 8B, 8D, and 8E, and the main effects of intrinsic and extrinsic cues (and the two-way interaction between them, along with the three-way interaction of them and immediate vs. delayed JOLs) from the 2 2 2 (i.e., timing of JOLs, extrinsic cue, and intrinsic cue) ANOVAs of all experiments are referred to in the text; the complete results from the ANOVAs are reported in Appendix A. Next, the results of the state-trace analysis are reported, with the means and standard error of the means shown in Figures 4C and 4F8C and 8F. Analyses of item-by-item JOL accuracy are not relevant to the hypotheses under investigation but, for completeness, are reported in Appen- dix C. Throughout, all differences reported as statistically signif- icant have p .05, and estimates of effect size (ES) are reported as partial eta squared for statistically significant effects. Prerequisite: Effects of intrinsic and extrinsic cues on recall and JOL magnitude. The mean percentage of correct recall for each condition for items receiving immediate JOLs is shown in Figures 2 Before Experiment 1B was conducted, we ran an experiment that was identical to Experiment 1B. In the earlier experiment, the pattern of the first 30 participants data collected at the beginning of the semester was incon- sistent with that of the final 15 participants data collected toward the end of the semester. Because we were concerned that the two samples might not be homogeneous, we reran the experiment, which we report here as Experiment 1B. The results of Experiment 1B are most similar to those of the first 30 participants data from the earlier version of this experiment. We mention that the data from the first 30 participants were more stable data than those from the final 15 participants (e.g., 23 out of the 24 results from the 2 2 2 ANOVAs for each of the three dependent variables the percentage of correct recall, JOL magnitude, and gammayielded smaller standard deviations for the first 30 participants than for the final 15 participants). Also, when three extreme outliers were removed from the final 15 participants, the resulting outcomes were similar to those of the first 30 participants data. We have no explanation for the relatively unusual performance of those three outliers, but we mention this for completeness. 313 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL 4A7A, and the corresponding mean for items receiving delayed JOLs is shown in Figures 4D7D. The mean magnitude of JOLs for each condition for items receiving immediate JOLs is shown in Figures 4B7B, and the corresponding mean for items receiving delayed JOLs is shown in Figures 4E7E. As shown in Appendix A and in Figures 4A, 4B, 4D, and 4E7A, 7B, 7D, and 7E, the pattern of results was quite consistent across Experiments 1A1D. Thus, for both recall and JOL mag- nitude, the prerequisite was met that both the effect of intrinsic cues and the effect of extrinsic cues were significant. For both recall and JOL magnitude, neither the two-way interaction of intrinsic and extrinsic cues nor the three-way interaction involving intrinsic and extrinsic cues was statistically significant. The pre- requisite allows for analyses of state traces, as described next. State traces of JOLs and recall. The outcome of the state-trace analysis of recall and JOL magnitude for items receiving imme- diate JOLs is shown in Figures 4C7C, and the corresponding outcome for items receiving delayed JOLs is shown in Figures 4F7F. Of primary importance, each of those panels of Figures 4 and 5 shows that the two-presentation curve falls atop the one- presentation curve, and each of those of Figures 6 and 7 shows that the long-presentation curve falls atop the short-presentation curve (also note that the bidirectional standard errors are quite small). The consistency of this outcome across Experiments 1A1D sug- gests that the extra flexibility of a multidimensional model, such as the two-dimensional model from the cue-utilization framework, is not needed, with the results most parsimoniously explained through a single underlying dimension for both recall and JOLs. Figure 4. Results of Experiment 1A. Panels A and B show the mean percentage of correct recall and the mean magnitude of judgments of learning (JOLs) for items having immediate JOLs, whereas Panels D and E show the corresponding data for items having delayed JOLs as a joint function of item difficulty and number of study presentations. Panel C is the state-trace plot of the mean magnitude of immediate JOLs against the mean percentage of correct recall, whereas Panel F is the state-trace plot of the mean magnitude of delayed JOLs against the mean percentage of correct recall. Each vertical and horizontal hash mark depicts the standard error of the mean. 314 JANG AND NELSON Experiment 2 Experiment 2 was similar to Experiment 1B, in which the intrinsic cue was item relatedness and the extrinsic cue was num- ber of study presentations, except that the instructions during the study/JOL phase were changed to encourage participants to inten- tionally use a comparison process when they made JOLs. Suppose that a participant studied a pair comprised of unrelated words (e.g., bottlecalendar) and responded with a JOL of 30%. If the partic- ipant subsequently studied a pair of related words (e.g., stove kitchen), then the rating might increase, say, to 70% because the person could compare and contrast the degree of the relatedness in the second pair with that in the first pair (Koriat, 1997). Because participants might not have used such a comparison process in Experiment 1B (and in Experiments 1A, 1C, and 1D as well), the instructions in Experiment 2 were constructed to encourage such a comparison process, assuming that such a comparison process may constitute a critical aspect of a multidimensional account of JOLs. Method Participants. Forty-five volunteers from undergraduate psychology courses at the University of Maryland received course credit in return for their participation in Experiment 2. Figure 5. Results of Experiment 1B. Panels A and B show the mean percentage of correct recall and the mean magnitude of judgments of learning (JOLs) for items having immediate JOLs, whereas Panels D and E show the corresponding data for items having delayed JOLs as a joint function of item relatedness and number of study presentations. Panel C is the state-trace plot of the mean magnitude of immediate JOLs against the mean percentage of correct recall, whereas Panel F is the state-trace plot of the mean magnitude of delayed JOLs against the mean percentage of correct recall. Each vertical and horizontal hash mark depicts the standard error of the mean. 315 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL Materials and procedure. The materials and procedure of Experiment 2 were identical to those of Experiment 1B except for the instructions during the study/JOL phase. The participants were informed that the members of each pair were related for some of the pairs but not for other pairs, and they were encouraged to make greater JOLs for related items than for unrelated items. Results and Discussion Prerequisite: Effects of intrinsic and extrinsic cues on recall and JOL magnitude. The mean percentage of correct recall for each condition for items receiving immediate JOLs is shown in Figure 8A, and the corresponding mean for items receiving delayed JOLs is shown in Figure 8D. The mean magnitude of JOLs for each condition for items receiving immediate JOLs is shown in Figure 8B, and the corresponding mean for items receiving delayed JOLs is shown in Figure 8E. As shown in Appendix A and in Figure 8A, 8B, 8D, and 8E, there were significant main effects of the intrinsic and extrinsic cues on both recall and JOL magnitude. The two-way interac- tion of intrinsic and extrinsic cues on JOL magnitude was not significant (as in Experiments 1A1D), whereas the two-way interaction of intrinsic and extrinsic cues on recall was signif- icant. This interaction was revealed as a greater effect of number of study presentations for the case of unrelated items (where recall was intermediate) compared with related items (where recall was closer to ceiling). Presumably, this interac- tion resulted from a ceiling effect, but it is important to note that, as described next, the state-trace analysis placed these Figure 6. Results of Experiment 1C. Panels A and B show the mean percentage of correct recall and the mean magnitude of judgments of learning (JOLs) for items having immediate JOLs, whereas Panels D and E show the corresponding data for items having delayed JOLs as a joint function of item difficulty and study duration. Panel C is the state-trace plot of the mean magnitude of immediate JOLs against the mean percentage of correct recall, whereas Panel F is the state-trace plot of the mean magnitude of delayed JOLs against the mean percentage of correct recall. Each vertical and horizontal hash mark depicts the standard error of the mean. 316 JANG AND NELSON conditions along a single curve. There were no three-way interactions for either recall or JOL magnitude. With the pre- requisite met, analyses of state traces are described next. State traces of JOLs and recall. The outcome of the state-trace analysis of recall and JOL magnitude for items receiving imme- diate JOLs is shown in Figure 8C, and the corresponding outcome for items receiving delayed JOLs is shown in Figure 8F. As in Figure 5 of Experiment 1B, each of those panels shows that the two-presentation curve falls atop the one-presentation curve. As in Experiments 1A1D, the consistency of the result suggests that a multidimensional account is not needed and that a single- dimensional account is sufficient. General Discussion The primary goal of this research was to investigate the structure underlying recall and JOLs by applying state-trace methodology to determine whether one underlying dimension is sufficient or whether multiple underlying dimensions are needed (e.g., the two dimensions proposed in the cue-utilization framework). Across all experiments investigating immediate and delayed JOLs, all 10 state-trace plots of recall and JOL magnitude consistently yielded state-trace curves that fell atop each other, as predicted by the assumption that only one dimension underlies both JOL magnitude and recall. The failure to disconfirm the single-dimensional model Figure 7. Results of Experiment 1D. Panels A and B show the mean percentage of correct recall and the mean magnitude of judgments of learning (JOLs) for items having immediate JOLs, whereas Panels D and E show the corresponding data for items having delayed JOLs as a joint function of item relatedness and study duration. Panel C is the state-trace plot of the mean magnitude of immediate JOLs against the mean percentage of correct recall, whereas Panel F is the state-trace plot of the mean magnitude of delayed JOLs against the mean percentage of correct recall. Each vertical and horizontal hash mark depicts the standard error of the mean. 317 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL occurred even when participants in Experiment 2 were instructed to intentionally use a comparison process that should have in- creased differential effects of intrinsic and extrinsic cues on JOLs (per the cue-utilization framework). At first glance, the finding that the pattern of state traces for immediate JOLs did not differ from that of state traces for delayed JOLs is surprising because previous researchers speculated that the relation between recall and JOLs might change over time. How- ever, the results of this study are in accord with a formulation that ascribes most of the greater accuracy of delayed JOLs to different ratios of easier versus more difficult discriminations between items without invoking different psychological processes for immediate versus delayed JOLs (Nelson et al., 2004). A multidimensional model will imitate the single-dimensional model, yielding state-trace curves that fall atop each other, if the relations from the independent variables through each of the di- mensions to the dependent variables are weighted to the same degree. Manipulating various combinations of the intrinsic and extrinsic cues, our experiments afforded multiple opportunities to rule out the single-dimensional model in at least a particular case. Across all experiments, however, the results that consistently yielded evidence for the single-dimensional model suggest that the multiple dimensions are unnecessary. The question of what is the single-dimensional structure that underlies JOL magnitude and recall is a topic for future research. Whether the single-dimensional structure is strength, ease of pro- Figure 8. Results of Experiment 2. Panels A and B show the mean percentage of correct recall and the mean magnitude of judgments of learning (JOLs) for items having immediate JOLs, whereas Panels D and E show the corresponding data for items having delayed JOLs as a joint function of item relatedness and number of study presentations. Panel C is the state-trace plot of the mean magnitude of immediate JOLs against the mean percentage of correct recall, whereas Panel F is the state-trace plot of the mean magnitude of delayed JOLs against the mean percentage of correct recall. Each vertical and horizontal hash mark depicts the standard error of the mean. 318 JANG AND NELSON cessing, retrieval fluency, or something else is an open question. However, research designed to answer that question should go beyond postulating metaphorical structures and instead operation- alize the various possibilities to distinguish empirically between them. Although the present research attempted to address the issue of dimensionality from a theoretically agnostic standpoint, we none- theless make a few remarks about the relation between this re- search and Koriats (1997) cue-utilization framework. We used the same set of independent variables that Koriat dichotomized into intrinsic versus extrinsic cues. Thus, our conclusion that JOLs are based on a single-dimensional construct is limited to the set of independent variables manipulated in our experiments according to Koriats dichotomy. Although we cannot generalize to other nonexamined independent variables, we can generalize our con- clusions to the independent variables that Koriat viewed as being prototypical for his intrinsic versus extrinsic distinction. Presum- ably, other variables might suggest a multidimensional structure. In fact, a candidate set of dimensions of JOLs is under consider- ation. Koriat, Bjork, Sheffer, and Bar (2004) recently reported some evidence for two underlying dimensions of JOLs. They showed that JOLs are insensitive to retention interval relative to recall, suggesting a distinction between experience-based and theory-based JOLs. They attributed the indifference of JOLs to retention interval to the predominant dependence on subjective experience (i.e., experience-based JOLs). Although further empir- ical research is needed to fully understand how the theory-based knowledge functions and can be combined with the experience- based knowledge as Koriat et al. suggested, it should be empha- sized that this dual-basis view serves as one potential multidimen- sional model. To explore the difference between our conclusions and those of Koriat (1997), we conducted another ANOVA for each experi- ment, treating the contrast of recall and JOLs (labeled measure by Koriat) as a repeated variable. The complete 2 2 2 (i.e., measure, extrinsic cue, and intrinsic cue) ANOVAs of all experi- ments are reported in Appendix B. According to the cue-utilization framework, there should be an interaction of measure and extrinsic cue, whereas there should be little or no interaction of measure and intrinsic cue. Specifically, the interaction of measure and extrinsic cue should yield the pattern of results indicating that recall is much higher than JOL magnitude in the strong level of extrinsic cues (i.e., two presentations in Experiments 1A, 1B, and 2; long pre- sentation in Experiments 1C and 1D). Note that because Koriats experiments examined only immediate JOLs, the comparison be- tween our results and those of Koriat should be limited to only our immediate JOL conditions (although the patterns of results in the immediate and delayed JOL conditions of this study are similar). The ANOVAs showed that, first, the interaction of measure and extrinsic cue was significant in all experiments except for Exper- iment 1C. The discounted effect of extrinsic cues on JOLs for items of two presentations (i.e., underconfidence, as shown in Table B2) was found in Experiments 1B and 2, which is consistent with the hypothesis of the cue-utilization framework. This pattern of interaction, however, was not found in any of the other exper- iments. Indeed, Experiment 1A yielded the opposite pattern; JOLs were overestimated for items of one presentation (i.e., overconfi- dence, as shown in Table B2). Second, the interaction of measure and intrinsic cue was significant in all experiments; Experiments 1A and 1C yielded overestimated JOLs for difficult items (and for easy items of Experiment 1C), whereas Experiments 1B, 1D, and 2 yielded underestimated JOLs for related items. Neither of the results was consistent with the hypothesis of the cue-utilization framework. In the present research, on the whole, the conceptual distinction of the intrinsic versus extrinsic cues failed functionally to confirm the predictions from the cue-utilization framework that whereas intrinsic cues have similar effects on both recall and JOLs, extrinsic cues affect recall more strongly than JOLs. Indeed, Koriat (1997) found inconsistent effects of intrinsic cues in his experiments; for instance, whereas his Experiments 2 and 3 yielded equivalent effects of intrinsic cues on both recall and JOLs, his Experiment 1 yielded greater effects on JOLs than on recall (see Figure 2, top panel, p. 354), and his Experiment 4 yielded weaker effects on JOLs than on recall. Likewise, other findings inconsistent with conclusions derived from the cue-utilization framework have been reported; for instance, Dunlosky and Matvey (2001) concluded that both outcomes are inconsistent with pre- dictions from the cue-utilization framework [which] provides more of an empirical generalization (i.e., a taxonomy of effects) and not a theoretical explanation for why various factors differentially influence JOLs (p. 1186). In addition, Busey et al. (2000) re- ported that exposure duration had a similar effect on JOL magni- tude as on memory performance, whereas the amount of rehearsal had a greater effect on JOL magnitude than on memory perfor- mance (although both exposure duration and the amount of re- hearsal are extrinsic cues). The assumption of Koriats intrinsic versus extrinsic distinction was based on converging/diverging interactions that were scale dependent in the sense that the conclusion of an interaction de- pends critically on the particular scaling both of JOL magnitude and of recall. That is, although Koriats interactions were signifi- cant in the statistical sense (of rejecting the null hypothesis of parallel curves), they were not meaningful in the measurement sense (i.e., conclusions drawn from them will carry over only to the particular measures he reported or to some linear transforma- tion of them, but there is no evidence that the relationship between those values and the underlying structure is necessarily linear). By contrast, a positive monotonic nonlinear transformation could transform Koriats interactions to parallel (cf. Krantz & Tversky, 1971, and the admonishment about drawing conclusions from scale-dependent interactions in the tutorial by Loftus, 1978). Such converging/diverging interactions, in contrast to crossover inter- actions (or interactions in which two curves have opposite- direction slopes), are known in the literature as being problematic as a basis for inferring underlying multidimensional structures (e.g., Dunn & Kirsner, 1988; Krantz, Luce, Suppes, & Tversky, 1971; Loftus, 1978). An advantage of the state-trace methodology used in the present research over standard parametric ANOVAs is that it is not beset with the aforementioned problem of meaningfulness of conclu- sions that occur when conclusions are based on scale-dependent interactions. As readers can prove to themselves, any monotonic (linear or nonlinear) transformation can be applied to the values we reported for JOL magnitude and recall without eliminating the overlap of the curves in the state-trace plots shown in Figures 4C and 4F8C and 8F. Our conclusions about an underlying single- dimensional structure being sufficient to account for the perfor- mance in our experiments are meaningful across all monotonic 319 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL transformations of JOL magnitude and recall that are shown in our 10 state-trace plots. By contrast, conclusions about converging/ diverging interactions can change if monotonic-but-nonlinear transformations are applied to the figures shown in Koriat (1997). The state-trace methodology used in this research not only circum- vents the problems of interpretation of scale-dependent interac- tions (the limitations of the general linear model) but also can explore the same issues as dissociation techniques, but in a stron- ger manner (see Busey et al., 2000; Loftus, 2002; Loftus & Irwin, 1998; Loftus et al., 2004; for a tutorial on state-trace analysis, see the Appendix in Harley, Dillon, & Loftus, 2004). References American Psychological Association. (1992). Ethical principles of psy- chologists and code of conduct. American Psychologist, 47, 15971611. Bamber, D. (1979). State-trace analysis: A method of testing simple theories of causation. Journal of Mathematical Psychology, 19, 137 181. Begg, I., Duft, S., Lalonde, P., Melnick, R., & Sanvito, J. (1989). Memory predictions are based on ease of processing. Journal of Memory & Language, 28, 610632. Benjamin, A. S. (2003). Predicting and postdicting the effects of word frequency on memory. Memory & Cognition, 31, 297305. Benjamin, A. S., Bjork, R. A., & Schwartz, B. L. (1998). The mismeasure of memory: When retrieval fluency is misleading as a metamnemonic index. Journal of Experimental Psychology: General, 127, 5568. Busey, T. A., Tunnicliff, J., Loftus, G. R., & Loftus, E. F. (2000). Accounts of the confidence-accuracy relation in recognition memory. Psy- chonomic Bulletin & Review, 7, 2648. Dunlosky, J., & Matvey, G. (2001). Empirical analysis of the intrinsic extrinsic distinction of judgments of learning (JOLs): Effects of relat- edness and serial position on JOLs. Journal of Experimental Psychol- ogy: Learning, Memory, and Cognition, 27, 11801191. Dunn, J. C., & James, R. N. (2003). Signed difference analysis: Theory and application. Journal of Mathematical Psychology, 47, 389416. Dunn, J. C., & Kirsner, K. (1988). Discovering functionally independent mental processes: The principle of reversed association. Psychological Review, 95, 91101. Harley, E. M., Dillon, A. M., & Loftus, G. R. (2004). Why is it difficult to see in the fog? How stimulus contrast affects visual perception and visual memory. Psychonomic Bulletin & Review, 11, 197231. Kimball, D. R., & Metcalfe, J. (2003). Delaying judgments of learning affects memory, not metamemory. Memory & Cognition, 31, 918929. Koriat, A. (1997). Monitoring ones knowledge during study: A cue- utilization framework to judgments of learning. Journal of Experimental Psychology: General, 126, 349370. Koriat, A., Bjork, R. A., Sheffer, L., & Bar, S. K. (2004). Predicting ones own forgetting: The role of experience-based and theory-based pro- cesses. Journal of Experimental Psychology: General, 133, 643656. Krantz, D. H., Luce, R. D., Suppes, P., & Tversky, A. (1971). Foundations of measurement: Vol. 1. Additive and polynomial representations. New York: Academic Press. Krantz, D. H., & Tversky, A. (1971). Conjoint-measurement analysis of composition rules in psychology. Psychological Review, 78, 151169. Loftus, G. R. (1978). On interpretation of interactions. Memory & Cogni- tion, 6, 312319. Loftus, G. R. (2002). Analysis, interpretation, and visual presentation of data. In H. E. Pashler (Series Ed.) & J. T. Wixted (Vol. Ed.), Stevens handbook of experimental psychology: Vol. 4. Methodology in experi- mental psychology (3rd ed., pp. 339390). New York: Wiley. Loftus, G. R., & Irwin, D. E. (1998). On the relations among different measures of visible and informational persistence. Cognitive Psychol- ogy, 35, 135199. Loftus, G. R., Oberg, M. A., & Dillon, A. M. (2004). Linear theory, dimensional theory, and the face-inversion effect. Psychological Review, 111, 835863. Nelson, T. O., & Dunlosky, J. (1991). When peoples judgments of learning (JOLs) are extremely accurate at predicting subsequent recall: The delayed-JOL effect. Psychological Science, 2, 267270. Nelson, T. O., & Dunlosky, J. (1992). How shall we explain the delayed- judgment-of-learning effect? Psychological Science, 3, 317318. Nelson, T. O., & Dunlosky, J. (1994). Norms of paired-associate recall during multitrial learning of Swahili-English translation equivalents. Memory, 2, 325335. Nelson, T. O., & Narens, L. (1994). Why investigate metacognition? In J. Metcalfe & A. P. Shimamura (Eds.), Metacognition: Knowing about knowing (pp. 125). Cambridge, MA: MIT Press. Nelson, T. O., Narens, L., & Dunlosky, J. (2004). A revised methodology for research on metamemory: Pre-judgment recall and monitoring (PRAM). Psychological Methods, 9, 5369. Schwartz, B. L. (1994). Sources of information in metamemory: Judgments of learning and feeling of knowing. Psychonomic Bulletin & Review, 1, 357375. Spellman, B. A., & Bjork, R. A. (1992). When predictions create reality: Judgments of learning may alter what they are intended to assess. Psychological Science, 3, 315316. Townsend, J. T., & Ashby, F. G. (1984). Measurement scales and statistics: The misconception misconceived. Psychological Bulletin, 96, 394401. Weaver, C. A., III, & Kelemen, W. L. (1997). Judgments of learning at delays: Shifts in response patterns or increased metamemory accuracy? Psychological Science, 8, 318321. Wixted, J. T. (1992). Subjective memorability and the mirror effect. Journal of Experimental Psychology: Learning, Memory, and Cogni- tion, 18, 681690. 320 JANG AND NELSON Appendix A Complete 2 2 2 (Timing of JOLs, Extrinsic Cue, and Intrinsic Cue) Analyses of Variance Experiment Mean percentage of correct recall Mean magnitude of JOLs F(1, 44) MSE p ES F(1, 44) MSE p ES Experiment 1A T 1 2.13 423.21 .15 E 99.03 506.63 .001 .69 91.31 230.05 .001 .68 I 218.45 363.95 .001 .83 170.98 207.82 .001 .80 T E 1 21.03 114.24 .001 .32 T I 1 15.97 191.79 .001 .27 E I 1 1 T E I 1.38 247.16 .25 1.27 122.88 .27 Experiment 1B T 1 1 E 74.76 368.69 .001 .63 88.77 140.63 .001 .67 I 159.52 609.57 .001 .78 121.40 355.20 .001 .73 T E 1.89 254.98 .18 2.94 499.72 .09 T I 7.59 264.52 .05 .15 1 E I 1 1 T E I 1 1 Experiment 1C T 2.01 498.11 .16 36.53 614.77 .001 .45 E 18.68 334.60 .001 .30 24.37 114.76 .001 .36 I 190.40 308.71 .001 .81 160.19 171.34 .001 .78 T E 1 1.53 206.39 .22 T I 1 2.89 131.54 .10 E I 1 1 T E I 4.04 274.55 .05 ns 4.04 80.97 .05 ns Experiment 1D T 1.08 516.47 .30 2.30 603.08 .14 E 55.65 324.82 .001 .56 88.19 120.24 .001 .67 I 232.90 475.74 .001 .84 179.88 293.47 .001 .80 T E 11.80 183.33 .01 .21 44.70 111.49 .001 .50 T I 14.67 311.38 .001 .25 4.22 119.59 .05 .09 E I 1 1 T E I 1 1 Experiment 2 T 1 1.29 690.27 .26 E 105.60 316.11 .001 .71 95.41 165.84 .001 .68 I 106.39 528.72 .001 .71 89.35 345.45 .001 .67 T E 1.50 404.50 .23 6.73 148.55 .05 .13 T I 5.43 355.40 .05 .11 6.20 140.62 .05 .12 E I 13.76 229.62 .01 .24 1 T E I 1.95 228.22 .17 1 Note. Effect size (ES) is reported only when the F value was significant. JOLs judgments of learning; T timing of JOLs (immediate vs. delayed JOLs in all experiments); E extrinsic cue (number of study presentations in Experiments 1A, 1B, and 2: one vs. two presentations; study duration in Experiments 1C and 1D: short vs. long duration); I intrinsic cue (item difficulty in Experiments 1A and 1C: easy vs. difficult items; item relatedness in Experiments 1B, 1D, and 2: related vs. unrelated items). (Appendixes continue) 321 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL Appendix B The complete results from 2 2 2 (i.e., measure, extrinsic cue, and intrinsic cue) analyses of variance are reported in Table B1. The results from follow-up simple effect tests of the interactions between measure and extrinsicintrinsic cue in the condition of immediate- judgments of learning (JOLs) are reported in Table B2, and the corresponding results in the condition of delayed JOLs are reported in Table B3. Table B1 Complete 2 2 2 (Measure, Extrinsic Cue, and Intrinsic Cue) Analyses of Variance Experiment Immediate JOLs Delayed JOLs F(1, 44) MSE p ES F(1, 44) MSE p ES Experiment 1A M 1.13 728.11 .29 1 E 73.05 322.00 .001 .62 76.58 606.78 .001 .64 I 142.74 285.45 .001 .76 161.81 445.93 .001 .79 M E 19.58 168.59 .001 .31 9.59 48.70 .01 .18 M I 27.87 169.31 .001 .39 5.66 108.50 .05 .11 E I 1 1 M E I 1.79 86.77 .19 4.99 68.24 .05 .10 Experiment 1B M 3.08 655.92 .09 8.50 422.25 .01 .16 E 70.98 276.92 .001 .62 53.01 365.97 .001 .55 I 198.78 428.38 .001 .82 81.52 656.72 .001 .65 M E 13.45 169.82 .001 .23 1 M I 20.26 231.69 .001 .32 15.78 80.60 .001 .26 E I 1 1 M E I 1 1 Experiment 1C M 54.64 650.31 .001 .55 1 E 13.80 181.63 .001 .24 11.58 578.99 .01 .21 I 151.47 247.01 .001 .78 103.79 444.06 .001 .70 M E 1.47 222.99 .23 2.09 60.51 .16 M I 14.69 151.99 .001 .25 16.58 52.55 .001 .27 E I 2.36 161.15 .13 2.82 361.99 .10 M E I 1.32 145.84 .26 1 Experiment 1D M 1.07 616.28 .31 5.37 285.60 .05 .11 E 24.83 145.75 .001 .36 87.34 359.75 .001 .66 I 245.53 433.75 .001 .85 108.06 516.68 .001 .71 M E 5.28 146.28 .05 .11 1 M I 33.16 165.67 .001 .43 9.99 840.28 .01 .19 E I 1 1 M E I 1 1 Experiment 2 M 7.75 1443.01 .05 .15 7.24 613.86 .05 .14 E 75.92 209.59 .001 .63 56.50 588.60 .001 .56 I 148.95 396.92 .001 .77 50.92 565.57 .001 .54 M E 6.50 157.19 .05 .13 7.82 79.61 .01 .15 M I 4.44 324.60 .05 .09 6.67 83.18 .05 .13 E I 8.12 218.94 .01 .16 1.55 240.69 .22 M E I 10.55 117.02 .01 .19 3.71 67.36 .06 Note. Effect size (ES) is reported only when the F value was significant. JOLs judgments of learning; Mmeasure (recall vs. JOLs in all experiments); E extrinsic cue (number of study presentations in Experiments 1A, 1B, and 2: one vs. two presentations; study duration in Experiments 1C and 1D: short vs. long duration); I intrinsic cue (item difficulty in Experiments 1A and 1C: easy vs. difficult items; item relatedness in Experiments 1B, 1D, and 2: related vs. unrelated items). 322 JANG AND NELSON Table B2 Simple Effect Tests Following the Interactions Between Measure and ExtrinsicIntrinsic Cue: Immediate JOLs Experiment and interaction t(44) p Over/underconfidence Measure Extrinsic Cue 1A Recall vs. JOLs of one presentation 3.19 .01 Overconfidence Recall vs. JOLs of two presentations .88 .38 1B Recall vs. JOLs of one presentation .10 .92 Recall vs. JOLs of two presentations 3.02 .01 Underconfidence 1D Recall vs. JOLs of short presentation .08 .94 Recall vs. JOLs of long presentation 1.94 .06 2 Recall vs. JOLs of one presentation 1.86 .07 Recall vs. JOLs of two presentations 3.41 .01 Underconfidence Measure Intrinsic Cue 1A Recall vs. JOLs of difficult items 3.36 .01 Overconfidence Recall vs. JOLs of easy items 1.30 .20 1B Recall vs. JOLs of unrelated items .70 .48 Recall vs. JOLs of related items 4.42 .001 Underconfidence 1C Recall vs. JOLs of difficult items 9.29 .001 Overconfidence Recall vs. JOLs of easy items 4.56 .001 Overconfidence 1D Recall vs. JOLs of unrelated items 1.94 .06 Recall vs. JOLs of related items 3.26 .01 Underconfidence 2 Recall vs. JOLs of unrelated items 1.56 .13 Recall vs. JOLs of related items 3.55 .01 Underconfidence Note. JOLs judgments of learning. (Appendixes continue) 323 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL Table B3 Simple Effect Tests Following the Interactions Between Measure and ExtrinsicIntrinsic Cue: Delayed JOLs Experiment and interaction t(44) p Over/underconfidence Measure Extrinsic Cue 1A Recall vs. JOLs of one presentation 1.40 .17 Recall vs. JOLs of two presentations 1.13 .26 2 Recall vs. JOLs of one presentation 1.64 .11 Recall vs. JOLs of two presentations 3.41 .01 Underconfidence Measure Intrinsic Cue 1A Recall vs. JOLs of difficult items 1.43 .16 Recall vs. JOLs of easy items 1.21 .24 1B Recall vs. JOLs of unrelated items 1.02 .31 Recall vs. JOLs of related items 4.54 .001 Underconfidence 1C Recall vs. JOLs of difficult items 2.57 .05 Overconfidence Recall vs. JOLs of easy items 1.42 .16 1D Recall vs. JOLs of unrelated items .58 .56 Recall vs. JOLs of related items 3.28 .01 Underconfidence 2 Recall vs. JOLs of unrelated items 1.59 .12 Recall vs. JOLs of related items 3.57 .01 Underconfidence Note. JOLs judgments of learning. Appendix C Mean Goodman-Kruskal gamma correlations between recall and judgments of learning are reported in Tables C1C5, and the complete results from analyses of variance of the gammas are reported in Table C6. Table C1 Mean Goodman-Kruskal Gamma Correlations Between Recall and JOLs of Experiment 1A as a Function of Timing of JOLs, Item Difficulty, and Number of Study Presentations Item difficulty Timing of JOLs Immediate (M .41, SEM .09) Delayed (M .76, SEM .04) Overall One study presentation Two study presentations One study presentation Two study presentations M SEM M SEM M SEM M SEM M SEM Easy .28 .23 .64 .13 .95 .05 .66 .17 .63 .08 Difficult .28 .20 .43 .17 .54 .18 .89 .06 .53 .07 Overall .28 .18 .53 .14 .74 .09 .78 .09 Note. JOLs judgments of learning. 324 JANG AND NELSON Table C2 Mean Goodman-Kruskal Gamma Correlations Between Recall and JOLs of Experiment 1B as a Function of Timing of JOLs, Item Relatedness, and Number of Study Presentations Item relatedness Timing of JOLs Immediate (M .28, SEM .07) Delayed (M .77, SEM .05) Overall One study presentation Two study presentations One study presentation Two study presentations M SEM M SEM M SEM M SEM M SEM Related .13 .17 .04 .17 .80 .07 .73 .09 .43 .06 Unrelated .55 .13 .41 .15 .73 .09 .80 .08 .62 .06 Overall .34 .11 .22 .13 .76 .07 .77 .06 Note. JOLs judgments of learning. Table C3 Mean Goodman-Kruskal Gamma Correlations Between Recall and JOLs of Experiment 1C as a Function of Timing of JOLs, Item Difficulty, and Study Duration Item difficulty Timing of JOLs Immediate (M .29, SEM .10) Delayed (M .77, SEM .05) Overall Study duration: 5 s Study duration: 15 s Study duration: 5 s Study duration: 15 s M SEM M SEM M SEM M SEM M SEM Easy .58 .16 .08 .22 .83 .09 .80 .13 .57 .09 Difficult .13 .20 .38 .26 .53 .20 .90 .10 .49 .09 Overall .35 .14 .23 .18 .68 .11 .85 .08 Note. JOLs judgments of learning. Table C4 Mean Goodman-Kruskal Gamma Correlations Between Recall and JOLs of Experiment 1D as a Function of Timing of JOLs, Item Relatedness, and Study Duration Item relatedness Timing of JOLs Immediate (M .22, SEM .08) Delayed (M .76, SEM .04) Overall Study duration: 2 s Study duration: 8 s Study duration: 2 s Study duration: 8 s M SEM M SEM M SEM M SEM M SEM Related .01 .18 .29 .17 .66 .15 .70 .10 .41 .06 Unrelated .29 .14 .28 .16 .83 .09 .86 .06 .56 .06 Overall .15 .10 .28 .10 .74 .09 .78 .05 Note. JOLs judgments of learning. (Appendix continue) 325 STATE TRACES FOR JUDGMENTS OF LEARNING AND RECALL Received December 10, 2003 Revision received April 19, 2005 Accepted April 26, 2005 Table C5 Mean Goodman-Kruskal Gamma Correlations Between Recall and JOLs of Experiment 2 as a Function of Timing of JOLs, Item Relatedness, and Number of Study Presentations Item relatedness Timing of JOLs Immediate (M .36, SEM .08) Delayed (M .83, SEM .04) Overall One study presentation Two study presentations One study presentation Two study presentations M SEM M SEM M SEM M SEM M SEM Related .34 .21 .03 .20 .79 .14 .89 .07 .51 .07 Unrelated .44 .15 .61 .11 .73 .12 .89 .08 .67 .07 Overall .39 .13 .32 .13 .76 .10 .89 .05 Note. JOLs judgments of learning. Table C6 Complete 2 2 2 (i.e., Timing of JOLs, Extrinsic Cue, and Intrinsic Cue) ANOVAs of Goodman-Kruskal Gamma Correlations Between Recall and JOLs Independent variable Experiment 1A Experiment 1B F(1, 12) MSE p ES F(1, 19) MSE p ES T 14.70 .22 .01 .55 44.06 .21 .001 .70 E 1 1 I 1 9.40 .16 .01 .33 T E 1 1 T I 1 4.53 .34 .05 .19 E I 1.05 .28 .33 1 T E I 5.30 .22 .05 .31 1 Experiment 1C Experiment 1D F(1, 9) MSE p ES F(1, 18) MSE p ES T 18.95 .24 .01 .68 27.73 .41 .001 .61 E 1 1.01 .28 .33 I 1 2.30 .37 .15 T E 1.54 .27 .25 1 T I 1 1 E I 6.42 .25 .05 .42 1 T E I 1 1 Experiment 2 F(1, 14) MSE p ES T 28.14 .24 .001 .67 E 1 I 3.38 .21 .09 T E 1 T I 2.85 .36 .11 E I 2.52 .21 .14 T E I 1 Note. Effect size (ES) is reported only when the F value was significant. JOLs judgments of learning; ANOVAs analyses of variance; T timing of JOLs (immediate vs. delayed JOLs in all experiments); E extrinsic cue (number of study presentations in Experiments 1A, 1B, and 2: one vs. two presentations; study duration in Experiments 1C and 1D: short vs. long duration); I intrinsic cue (item difficulty in Experiments 1A and 1C: easy vs. difficult items; item relatedness in Experiments 1B, 1D, and 2: related vs. unrelated items). 326 JANG AND NELSON