Theory-based categorization under speeded conditions.

Christian C. Luhmann Woo-kyoung Ahn Thomas J. Palmeri Vanderbilt University

Corresponding Author: Christian C. Luhmann Department of Psychology, Vanderbilt University Wilson Hall, 111 21st Avenue South, Nashville, TN 37203 (615) 322-2835 [email protected]

Running head: speeded categorization

Luhmann

Theory-based categorization 1

Abstract

A largely accepted view in the categorization literature is that similarity-based reasoning is faster than theory-based reasoning. In the current study, we explored whether theory-based categorization behavior would continue to be observed when people are forced to make category decisions under time pressure. As a specific test of the theory-based view of category representation, we examined the causal status hypothesis, which states that properties acting as causes are more important than properties acting as effects when categorizing an item (Ahn, Kim, Lassaline, & Dennis, 2000). Subjects learned four categories of items composed of three features and learned causal relations between those features. In two experiments we found that participants gave more weight to cause features than to effect features even under rapid response conditions. categorization.

We discuss implications of these findings for

Luhmann

Theory-based categorization 2

When categorizing in everyday life people recruit information from various sources. Previous work on categorization has generally focused on two types of information. On one view, concept learning and use involves computing the similarity between novel objects and a stored representation of a category (e.g., Kruschke, 1992; Nosofsky, 1986; Smith & Medin, 1981; Rosch & Mervis, 1975). An alternative view assumes that people have theory-like background knowledge that includes relations between properties and influences categorization (Carey, 1985; Keil, 1989; Murphy & Medin, 1985; Rips, 1989). For example, adults categorize animals with the appearance and behavior of a horse but the cow insides and lineage as cows (Keil, 1989). This behavior suggests that lineage has a special status beyond perceptual features, presumably due to our lay theories of biology. While these two views have been traditionally portrayed in opposition (e.g. Sloman, 1996; Murphy & Medin, 1985), recently many advocate the use of both kinds of information (e.g. Smith & Sloman, 1994). However, these proposals typically put the two views on unequal footing. A persistent bias is that similarity-based categorization proceeds more rapidly than theory-based information. This assumption may be motivated by the belief that novices (e.g., children) use similarity-based reasoning and is thus a simpler mode of reasoning (cf. Keil, Smith, Simons, & Levin, 1998). Smith and Sloman (1994) state this explicitly by arguing that theory-based reasoning is, “more analytic and reflective than similarity-based categorization” (pp. 377-378).

Luhmann

Theory-based categorization 3

Smith and Sloman (1994) examined the effect of time constraints on theory-based categorization. Following Rips (1989), subjects received a forced-choice task where each item consisted of a description of an object (e.g., circular object with a 4 inch diameter) and two possible responses: one signifying a theory-based decision (e.g., calling it a pizza rather than a quarter because of the constraints imposed by the minting process) and the other signifying a similarity-based decision (e.g., calling it a quarter because the object is more similar to quarters; but see Nosofsky & Johansen, 2000). While Rips’ (1989) subjects preferred the theory-based response, when Smith and Sloman asked their subjects to respond as quickly as possible, they failed to reproduce these results. Only when instructed to talk aloud while categorizing did subjects prefer the theory-based response. Smith and Sloman concluded that a “…possible constraint…is that the situation encourage people to articulate and explain their reasons for categorization, rather than encourage rapid judgments” (p. 383). This study constitutes the main evidence that theory-based categorization is a slow process. One problem with this Smith and Sloman’s (1994) interpretation is that any judgment resulted in subjects’ accepting bizarre objects as category members. For the pizza/quarter question, one must entertain the idea of both a non-standard quarter (violating ontological beliefs) and an unusual pizza (violating pragmatic beliefs). Thus, the task may become a judgment between a “lesser of two evils, ” which does not represent a naturalistic test of the theory use.1

1

It is important to point out that Smith and Sloman (1994) never obtained a preference for similarity-based choices in their speeded condition. When speeded, subjects simply showed no preference when offered both similarity and theory-based responses.

Luhmann

Theory-based categorization 4

More recent studies have suggested that theory-based categorization may be faster than previously thought. Lin and Murphy (1997) taught subjects about novel objects and their intended functions. Some object parts were central to this function while others played minor functional roles. The results showed that functionally central parts influenced categorization judgments more than functionally irrelevant parts even when the picture of the object to be categorized was presented for only 50ms, demonstrating rapid influence of domain knowledge. Palmeri and Blalock (2000) reported a similar pattern of results. Following Wisnewski and Medin (1994), they had subjects categorize drawings supposedly drawn by ”creative children” or ”non-creative children.” Subjects were able to categorize using this background knowledge (e.g., the amount of emotional expression) even when the drawings were shown for only 200ms. Their results suggest that theory-use does not require lengthy periods of reflection. Although this recent evidence is compelling, the effect of background knowledge demonstrated in these two studies is limited to perceptual categorization. The current studies utilize verbal stimuli as in Smith and Sloman (1994) to examine speeded theorybased effects in conceptual categorization. More importantly, both studies fail to provide a principled account of the theory-based mechanisms at work. Palmeri and Blalock (2000) allowed the subjects to generate categorization rules given their background knowledge (i.e. category labels), but did not attempt to characterize this process. Lin and Murphy (1997) assume that theory-based categorizing of artifacts relies on functional features, but it is not clear why some functional features (e.g., used to hang a tool on the

Luhmann

Theory-based categorization 5

wall) are more central than others (e.g., used to grab an animal’s neck) (but see Ahn, 1998). In the current study, we test a well-articulated theory-based mechanism under speeded conditions. As a specific test of the theory-based view, we have chosen to investigate the causal status hypothesis (Ahn, 1998; Ahn, et al., 2000), which states that features of an object that act as causes in one’s domain theory are more important than features that act as effects. To test this hypothesis, Ahn, et al. (2000) presented subjects with novel categories with a list of features (e.g., X, Y, Z). Subjects were also told that feature X caused feature Y, which in turn caused feature Z. When subjects received descriptions of possible category members, those items missing feature Z were rated as the best category members, items missing feature X were rated as the worst category members, and items missing feature Y fell midway. These findings strongly support the idea that causes are more important than effects when categorizing. The purpose of the current studies is to examine whether the use of theories requires extended deliberation. To do so, we employed a methodology similar to Ahn, et al. (2000); we created artificial categories with causally related features and had subjects judge the category membership of items. In addition, we manipulated the amount of time pressure on making such judgments.

Experiment 1

Method Participants.

Luhmann

Theory-based categorization 6

Twenty-nine Vanderbilt University undergraduates participated in partial fulfillment of course requirements. Materials. The stimuli consisted of four fictional animals each with three features (e.g. X, Y, and Z). The features were described as forming a causal chain such that feature X causes feature Y and feature Y causes feature Z. An included summary diagram contained arrows depicting the presence and direction of a causal relationship. The features were chosen to facilitate intuitive causal connections between adjacent features (e.g., a small heart causes a low body temperature presumably due to weak circulation) in order to approximate the effect of entrenched naïve theories. To ensure that the three features in each category did not vary in importance in the absence of causal information, a separate set of 30 subjects were presented with features without causal information and rated the category membership likelihood of items missing a single feature. The pre-test results showed no significant differences between the ratings of items missing the feature X (M = 3.90), items missing feature Y (M = 4.38), and items missing the feature Z (M = 4.25; all p’s>.4). Thus, we concluded that the features were equated for a priori saliency. Procedure. The experiment was run on Apple iMacs using RSVP (Williams & Tarr, No date). During the learning phase, subjects were first given the opportunity to study the description of each animal one at a time. For each animal they were given the list of features, the causal relationships between them, and the summary diagram. While each description remained on the computer screen, subjects were instructed to “write about

Luhmann

Theory-based categorization 7

how you think each feature causes the next.” This was done to encourage subjects to think causally about the features instead of as a simple ordered list. After viewing and writing about all animals, subjects were presented with 6 blocks of trials during which they were prompted with the name of an animal and required to select (using a mouse-click) that animal’s features from a table containing all 4 animals’ features. On each trial one of four tables was randomly chosen, each with a unique arrangement of the features so that subjects could not simply memorize locations in the table. Subjects were additionally required to select the correct features for each animal in the appropriate causal order (see below). After each selection, subjects were presented with feedback about their selection as well as a summary of that animal’s features and the causal relationships. Successfully responding to the entire set of animals twice consecutively, with one allowed error, permitted subjects to move on to the next block. In the first two blocks responses were unspeeded. In the last four blocks subjects were told to respond as quickly as they could, and any responses that took longer than 5 seconds were counted as incorrect. This speeded element was added to automatize the use of the novel causal background knowledge thereby approximating real-life lay theories. In addition, on half of the blocks subjects were asked for the features in the forward (e.g. X, Y, Z) order, while in the other half in the backward order (e.g. Z, Y, X) to prevent features from being rated as central simply because it was always presented first in the list. The order manipulation alternated across blocks always beginning with a forward block. Once completing the learning phase, subjects proceeded to the transfer task. They were presented with items missing a single feature and asked to rate the likelihood that

Luhmann

Theory-based categorization 8

the item belonged to its target category on an 8-point scale (with 1 labeled as “Definitely Unlikely” and 8 labeled as “Definitely Likely”). Subjects were told that since every item would be missing exactly one feature there would be no perfectly good category members and no perfectly bad category members and were encouraged to use the entire scale. Trials began with the name of one of the animals (the target category) appearing on the screen for one second and then replaced by a triad of features. The three features belonged to the target category with one feature negated by appending the phrase “does not.” The features were presented in a triangular arrangement with the location of each feature in the triangle randomly determined for each trial to discourage reliance on spatial information. Additionally, the negated feature was displayed in red to facilitate reading of stimuli. This should not have qualitatively altered performance since the other two features are logically inferable from the negated feature given that each item would contain the target category’s features with exactly one feature negated. There were 4 blocks of these trials each consisting of each of the 12 items (4 categories with 3 features each) presented 4 times for a total of 48 trials presented in random order. In two of the blocks (Speeded condition), subjects were instructed to answer “as quickly as possible while still remaining accurate” and given an example of how medical professionals often had to make decisions that were both accurate and rapid. In the other two blocks (Unspeeded condition), they were told to “take as much time as needed”. The two conditions alternated across blocks and were counterbalanced across subjects. The RTs in the speeded blocks (M=1560ms) were significantly faster than the

Luhmann

Theory-based categorization 9

RTs in the unspeeded blocks (M=3202ms), t(170)=21.09, p<.05, verifying the instructional speed manipulation. Results and Discussion In the unspeeded condition, we expected to find results similar to those of Ahn, et al. (2000). That is, items missing feature Z should be rated as more likely category members than those missing feature X. The critical question was whether this causal status effect would disappear in the speeded condition, as suggested by Smith and Sloman (1994). The results are summarized in Figure 1. A 2 (speed condition: speeded vs. unspeeded) X 3 (item type: missing feature X vs. missing feature Y vs. missing feature Z) repeated measures ANOVA was performed on the data. We observed a significant main effect of item type, F(2, 56)=22.69, MSE=44.39, p<.0001. No other effects were significant. Insert Figure 1 about here.

Planned comparisons showed that items missing feature Z were rated significantly higher than those missing the features X or Y in both the speeded (t(28)=5.00, p<.05) and unspeeded (t(28)=5.07, p<.05)conditions. The difference between items missing feature X and those missing feature Y was not significant (t(28)=1.47, n.s.), possibly because the feature Y also served as a cause of another feature, making the difference between the X and the Y feature less pronounced (see also Kim & Ahn, 2002). An analysis of subjects’ first block of transfer trials demonstrates similar findings (Luhmann, 2002) ruling out practice effects as an explanation of speeded performance. Overall, these results demonstrate that it is possible to categorize using causal knowledge even when time for extensive reflection is not allowed. It is tempting to

Luhmann

Theory-based categorization 10

contrast our findings with those of Smith and Sloman (1994). In their experiment, subjects needed to talk aloud while making the judgment in order to demonstrate theorybased behavior. It is possible that the paradigm used by Smith and Sloman (1994) created a situation in which theory-use was more difficult to apply than our situation as suggested above. Nevertheless, our findings clearly question the assertion that all theoryuse necessitates lengthy reflection.

Experiments 2A and 2B In Experiment 1, subjects in the Speeded condition were simply asked to respond as quickly as possible. Given this freedom, some subjects responded very quickly but others responded significantly more slowly. In Experiments 2A and 2B, we imposed stricter control over subjects’ response times by enforcing specific deadlines. One methodological complication with establishing appropriate deadlines is that it is difficult to determine whether a particular deadline is short enough to challenge the categorization system but not so short as to eliminate intelligent responses. That is, if speeded conditions attenuate the causal status effect, it may be because theory-based reasoning does not take place during rapid categorization or because the deadline is too short to produce any reasonable response. For this reason, we also tested whether similarity information could be used under short deadlines. In this way, the casual status effects can be compared to similarity-based categorization at each deadline and it can be inferred whether any breakdown of the causal status effect is due to the inability to complete theory-based processes or if reasonable responses are impossible for other

Luhmann

Theory-based categorization 11

reasons (e.g. inability to read stimuli). If the latter is true, then subjects should have difficulty with both kinds of categorization. Similarity is frequently calculated based on how many attributes an item has in common with other members of the category (e.g., Tversky, 1977). Therefore, as a similarity-based determinant for feature weighting, we manipulated the base rates of each feature within a category (i.e., what percentage of category members possess a feature). This measure has been shown to positively correlate with typicality ratings (Gluck & Bower, 1988; Rosch & Mervis, 1975). Experiments 2A and 2B contain similarity conditions that provide base rate information in much the same way causal information was provided in Experiment 1. Comparing these conditions, and with the addition of deadlines, we attempt to provide a more rigorous test of the causal status effect under speeded conditions.

Method Participants. One hundred eight Vanderbilt University undergraduates were assigned to either the Causal condition (n=25 in Experiment 2A, n=30 in Experiment 2B) or the Base-Rate condition (n=23 in Experiment 2A, n=30 in Experiment 2B). Materials. Subjects in the Causal conditions were given the same stimuli and causal information as used in Experiment 1. Subjects in the Base-Rate conditions were given the same stimuli but all features were redistributed across the categories to discourage the spontaneous generation of causal links and the causal information was replaced by the

Luhmann

Theory-based categorization 12

base rates of each feature. Thus, in the Base-Rate condition, each category was described as having three features (e.g. X, Y, and Z) such that 100% of category members possessed feature X, 80% of possessed feature Y, and 60% possessed feature Z. Paralleling the effect of base rate on typicality judgments (Rosch & Mervis, 1975), items in the Base-Rate condition missing feature Z (60%) should be rated as better category members than those missing feature X (100%) with feature Y (80%) falling somewhere in the middle. Procedure. The learning phase for the Causal conditions was identical to that used in Experiment 1. That is, subjects received the description of animals and causal relations, generated explanations for causal relations, and then completed a “selection task” where three features of a category were selected in the requested causal order. The learning phase for the Base-Rate conditions was constructed to be as similar as possible to that for the Causal conditions. Subjects in the Base-Rate conditions first received the description of animals one at a time along with base rate information for each feature. To inflict direct experience with base rate information, participants then received a set of items constructed to mirror the stated base rates2, categorized each into one of the stimuli categories, and received feedback after each trial along with a summary of that animal’s features and base rates. Blocks of 30 such trials alternated with blocks of a “selection task” like that used in the Causal condition. Subjects in the Base-Rate condition were instructed to select features in order (forward or backward) dictated by their base rate rather than their position in the causal chain. The feature matrix, exit When a feature did not appear in an item, a feature from one of the other animals was substituted. 2

Luhmann

Theory-based categorization 13

conditions, speed manipulation, and order of blocks were all identical to the Causal condition. The transfer phase for both conditions was identical to that used in Experiment 1 except for a modified speed manipulation. Instead of an instruction manipulation, Experiment 2 employed a signal-to-respond procedure (Lamberts, 1998). Thus, each feature triad was presented for a certain, set, amount of time ranging from 5000ms to 300ms. Subjects were instructed to respond to the item immediately after the item disappeared from the screen. Responses could not be made before the item disappeared or more than 300ms after the disappearance of the triad. If a response was delivered late, subjects were told to respond more rapidly, and the data point was not included in our analyses. There were again four blocks of 48 trials. Each block used one of four durations. In Experiment 2A the durations were 5000ms (representing an “unspeeded” condition), 2250ms, 1500ms (the mean RT from Experiment 1), and 750ms. In Experiment 2B the durations used were 1500ms, 750ms, 500ms, and 300ms. These blocks were ordered randomly for each subject. Results and Discussion The results can be seen in Figure 2. In Experiment 2A, a 2 (knowledge condition: Causal vs. Base-Rate) X 4 (speed condition: 5000ms vs. 2500ms vs. 1500ms vs. 750ms) X 3 (item type: missing feature X vs. missing feature Y vs. missing feature Z) ANOVA was performed with repeated measures on the latter two factors. The main effect of item type was significant, F(2, 90)=64.77, p<.0001, and this effect significantly interacted with knowledge condition, F(2, 90)=6.73, p<.005. We observed a significant interaction

Luhmann

Theory-based categorization 14

between item type and speed, F(6, 276)=16.73, p<.0001, however, the three-way interaction between item type, speed, and knowledge condition failed to reach significance, F<1. No other effects were significant. Planned comparisons were conducted to determine whether a significant effect was present at each deadline for each knowledge condition. For both knowledge conditions items missing feature Y always fell between the two other item types but were not consistently significant (see Luhmann, 2002), probably due to lack of power. Examining the difference predicted to be largest in both Base-Rate and Causal conditions, the items missing feature Z were rated as significantly better category members than those items missing X at all deadlines (all p’s < .05). Insert Figure 2 about here.

In Experiment 2B, a 2 (knowledge condition: Causal vs. Base-Rate) X 4 (speed condition: 1500ms vs. 750ms vs. 500ms vs. 300ms) X 3 (item type: missing feature X vs. missing feature Y vs. missing feature Z) ANOVA was performed with repeated measures on the latter two factors. We observed a significant main effect of item type, F(2, 116)=10.51, p<.0001, that did not interact with knowledge condition, F<1. No significant main effect of speed was observed, F(3, 171)=1.60, n.s., but this must be interpreted in light of a significant interaction between speed and item type, F(6, 342)=4.04, p<.001, which is examined below. A significant main effect of background condition was observed, F(1, 57)=5.22, p<.05, but this factor did not interact with either of the other two factors. The three-way interaction between background condition, speed, and item type significance was also significant, F(6, 342)=2.14, p<.05.

Luhmann

Theory-based categorization 15

Planned comparisons were carried out to determine at what deadlines the background information had an effect on categorization (items missing feature Y again fell between the other two item types but were not consistently significant, see Luhmann, 2002). For the Base-Rate condition, subjects rated items missing feature X significantly lower than items missing feature Z in the 1500ms condition (M = 3.80, 4.82 respectively), t(29)=3.15, p<.005, and the 750ms condition (M = 4.39, 4.96 respectively), t(29)=2.05, p<.05, but not in the 500ms, t(29)=.18, p=.86, or 300ms, t(29)=.92, p=.37, conditions. In the Causal condition, items missing feature X differed those missing feature Z in the 1500ms condition (M = 3.66, 4.18 respectively), t(29)=2.17, p<.05, the 750ms condition (m = 3.81, 4.37 respectively), t(29)=2.74, p<.05, is marginally significant in the 500ms condition (m = 3.87, 4.31 respectively), t(29)=2.01, p=.053, and non-significant in the 300ms condition, t(29)=.75, p=.46. In light of these further results, it is clear that causal knowledge can be utilized when categorizing stimuli under even faster conditions than those created in Experiment 2A. Subjects in this experiment were able to categorize according to the causal information even when allowed only 750ms to view the exemplar. Furthermore, baserate information, which has been considered key determinant of similarity (Rosch and Mervis, 1975), did not result in differential responses under the 500ms deadline while causal information produced a marginally significant difference. These results provide strong evidence that theory-based categorization need not be slower than similarity-based categorization.

Luhmann

Theory-based categorization 16 General Discussion

Our results suggest that the causal status effect, an example of theory-based categorization, is not resigned to situations where categorizers have an opportunity to deliberate. This was demonstrated with both the more naturalistic instructional manipulation as well as the more controlled signal-to-respond method. In fact, all speeded manipulations failed to eliminate the effect until Experiment 2B when subjects were only allowed 500ms to view the text and an additional 300ms to respond, the same deadline where similarity information (in the form of feature frequencies) first failed to produce an effect. Our results clearly contradict the prevailing notion that the effects of background knowledge are found only in slow, deliberative settings and suggest that future models of categorization should reconsider how to integrate similarity mechanisms with background knowledge. While we have shown that one variant of theory-based categorization may be fast, this might not be true of all other variants. Theory-based categorization that is more analogous to problem solving (e.g., akin to Smith & Sloman, 1994) might not show similar immunity to speed. Another open question concerns whether any causal reasoning taking place during the transfer tasks, or subjects were simply retrieving precompiled notions about feature importance derived during learning. The current findings bolster the idea that use of similarity is not necessarily more primary than theory-use. Instead, it appears that people’s use of theories and similarity are inexorably intertwined. If similarity needs to be constrained by theories (e.g., Keil, 1981; Murphy & Medin, 1985), it makes sense that theories are influential at an early stage of categorization.

Luhmann

Theory-based categorization 17 References

Ahn, W. (1998). Why are different features central for natural kinds and artifacts?: The role of causal status in determining feature centrality. Cognition, 69, 135-178. Ahn, W., Kim, N. S., Lassaline, M. E, & Dennis, M. J. (2000). Causal status as a determinant of feature centrality. Cognitive Psychology, 41, 361-416. Carey, S. (1985). Conceptual change in childhood, MIT Press, Cambridge, MA. Hampton, J. A. (1995). Testing prototype theory of concepts. Journal of Memory and Cognition, 34, 686-708. Keil, F. C. (1981). Constraints on knowledge and cognitive development. Psychological Review, 88, 197-227. Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press. Keil, F. C., Smith, W. C., Simons, D. J., & Levin, D. T. (1998). Two dogmas of conceptual empiricism: Implications for hybrid models of the structure of knowledge. Cognition, 65, 17-49. Kim, N. S., & Ahn, W. (2002). The influence of naïve causal theories on lay concepts of mental illness. American Journal of Psychology, 115, 33-65. Kruschke, J. K. (1992). ALCOVE: An exemplar-based connectionist model of category learning. Psychological Review, 99, 22-44. Lamberts, K. (1998). The time course of categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 695-711.

Luhmann

Theory-based categorization 18

Lin, E. L., & Murphy, G. L. (1997). Effects of background knowledge on object categorization and part detection. Journal of Experimental Psychology: Human Perception and Performance, 23, 1153-1169. Luhmann, C.C. (2002). Theory-based categorization under speeded conditions. Unpublished master’s thesis, Vanderbilt University, Tennessee, Nashville. Medin, D. L., & Shoben, E. J. (1988). Context and structure in conceptual combination. Cognitive Psychology, 20, 158-190. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289-316. Nosofsky, R. M. (1986). Attention, similarity, and the identification-categorization relationship. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 510-520. Nosofsky, R.M., & Johansen, M.K. (2000). Exemplar-based accounts of “multiple-system” phenomena in perceptual categorization. Psychonomic Bulletin & Review, 7, 375-402. Palmeri, T. J., & Blalock, C. (2000). The role of background knowledge in speeded perceptual categorization. Cognition, 77, B45-B57. Quine, W. V. (1977). Natural kinds. In Schwartz, S.P. (Ed.), Naming, Necessity, and Natural Kinds. Cornell University Press, Ithaca, NY. Rips, L. J. (1989). Similarity, typicality, and categorization. In S. Vosniadou & A. Ortony (Eds.), Similarity analogical reasoning (pp. 21-59). New York: Cambridge Univ. Press. Rosch, E., & Mervis, C. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573-605.

Luhmann

Theory-based categorization 19

Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3-22. Sloman, S. A. & Rips, L. J. (1998). Similarity as an explanatory construct. Cognition, 65, 115. Smith, E. E. & Medin, D. L. (1981). Concepts and Categories, Harvard University Press, Cambridge, MA. Smith, E. E., & Sloman, S. A. (1994). Similarity- versus rule-based categorization. Memory and Cognition, 22, 377-386. Tversky, A. (1977). Features of similarity. Psychological Review, 84, 327-352. Vygotsky, L. S., (1962). Thought and language (E. Hanfmann and G. Vakar, Trans.), MIT Press, Cambridge, MA. Williams, P. & Tarr, M. J. (No date). RSVP: Experimental control software for MacOS [Online]. Available: http://www.tarrlab.org/rsvp/ [2002, July 10]. Wisnewski, E. J. & Medin, D. L. (1994). On the interaction of theory and data in concept learning. Cognitive Science, 18, 221-281.

Luhmann

Theory-based categorization 20 Author Note

Christian C. Luhmann, Department of Psychology, Vanderbilt University, Wookyoung, Department of Psychology, Vanderbilt University, Thomas J. Palmeri, Department of Psychology, Vanderbilt University. This project was supported by a National Institute of Mental Health Grant (RO1 MH57737) to Woo-kyoung Ahn and XX. Correspondence concerning this article should be sent to Christian C. Luhmann, Vanderbilt University, Department of Psychology, 301 Wilson Hall, Nashville, TN, 37203; [email protected].

Luhmann

Theory-based categorization 21

Figure Captions

Figure 1. Results from Experiment 1. Error bars indicate standard error. Figure 2. Results from Experiment 2. The top graph contains the results from the BaseRate condition. The bottom graph contains those from the Causal condition. Error bars indicate standard error.

Luhmann

Theory-based categorization 22

Figure 1.

6

Speeded Unspeeded

5.25 (.23) 5.02 (.21)

5

3.46 (.29)

3.51 (.30)

3.85 (.22)

Rating

4

3.77 (.20)

3

2

1 Missing Feature X

Missing Feature Y Item Type

Missing Feature Z

Luhmann

Theory-based categorization 23

Figure 2. 2A-Missing X 2A-Missing Z 2B-Missing X 2B-Missing Z

Base-Rate 6

Rating

5

4

3

2 5000

2500

1500 Presentation Duration (ms)

750

500

300

2A-Missing X 2A-Missing Z 2B-Missing X 2B-Missing Z

Causal 6

Rating

5

4

3

2 5000

2500

1500 Presentation Duration (ms)

750

500

300