The Psychological Record , 1998, 48, 325-332 INDIVIDUAL DIFFERENCES IN COGNITIVE ADDITION MOHAMED BERNOUSSI Universite de Nantes, France This study addresses the problem of individual differences in solving simple addition problems by comparing the result of group to individual data. We found that if the group data are better explained by the currently dominant model (retrieval of arithmetic facts from memory), individual data are explained by different models at the level of the process used. The study of cognitive processes involved in solving simple arithmetic problems has aroused much interest in the past twenty years. Several models have attempted to account for these processes (for review see Ashcraft, 1995; Lemaire & Bernoussi, 1991; McCloskey, Harley, & Sokol, 1991). Most research in this area uses a chronometric method. The measurement of latencies (RTs) can be applied to the formulation of a judgment in a verification task: The subject is asked to verify the accuracy of an already solved problem (e.g., 3 + 4 = 7 True or False ?) or to produce the response in a production task: The subject is asked to solve arithmetic problems and produce the result (e.g., 3 + 4 = ?). The first study in this area was conducted by Groen and Parkman (1972) with children. These authors assumed that the process involved in solving simple addition problems is based on an internal mental counter. Using this counter, an addition problem (like 3 + 4) can be solved in the following way. First, the counter is set to an initial value corresponding to one of the two digits of the problem. Subsequently, the second digit is added to the counter by increments of one. Both the time to set the counter and the time to increase it by one are assumed to be constant. So, the RT will vary only as a function of the number of increments. The results obtained by Groen and Parkman indicated that RT was I am grateful to Dr. Angela Notari-Syverson (Washington Research Institute) for her helpful editing work. I thank the reviewers for their helpful comments on an earlier version of this article. Reprint requests may be sent to Mohamed Bernoussi, Laboratoire de Psychologie, Universite de Nantes. B.P. 81227, Chemin de la Censive du Tertre. 44312 Nantes Cedex3. France. E-mail: [email protected] 326 BERNOUSSI best predicted by the value of the smaller digit of the problem named the minimum (min). This suggests that children assign the biggest value to . the counter and add the smallest value to it. For the adults, however, two variables provided a best fit of RT: the min variable, but also the sum of the two digits. The authors proposed a mixed-model interpretation: Results of simple arithmetic problems are stored in long-term memory, and adults use a direct-access process to retrieve these facts. However, if the direct-access fails, adults may sometimes use a counting process like children. Ashcraft and Battaglia (1978) empirically tested the model proposed by Groen and Parkman (1972) using a simple addition verification task. The results of this study indicated that RTs of adults are best predicted by the square of the correct sum (sum 2). This finding was inconsistent with the Groen and Parkman counting and direct-access model described above. Ashcraft and Battaglia (1978) assumed then that the used process is the retrieval of the sum from a memory network. However, Miller, Perlmutter, and Keating (1984) showed that the structural variable that predicted RTs of adults was the product of the two addends. These authors hypothesized that arithmetical facts are stored in long-term memory in the form of a tabular network (an addition table). On each side of this network are the entry nodes for the integers 0 through 9. The correct sum is stored at the intersection of the entry nodes values corresponding to the two addends. The structural variable "product" represents the surface of the network to browse for retrieving a result. So, to retrieve the sum of 2 + 3, it is necessary to activate the associate node of the first addend (2), then the one partner in the second (3), and browse the surface (2 x 3 nodes) to access to the intersection. Since the study by Miller et al. (1984), several authors share this conception of memory network (see Geary, Widaman, & Little, 1986; Timmers & Claeys, 1990; Widaman, Geary, Cormier, & Little, 1989). These studies, however, have been criticized from both conceptual and methological standpoints. The conceptual criticisms stem from authors who developed alternative models of memory network (see Campbell, 1987, 1995). The methodological criticisms concern primarily two aspects: (1) The choice of structural variables. The structural variables are in fact the mathematical formulation of cognitive processes (Geary, 1987) and used to differentiate models proposed. These variables, however, are strongly correlated with each other, and it is difficult to evaluate statistically the unique contribution of each individual variable. (2) The level of the analyses. Most studies used as a dependent variable the mean RT of the group. But the utilization of the mean masks the different strategies used by subjects (see Siegler, 1987); moreover, it reduces the observed variance, which permits a best fit of data (see Pellegrino & Goldman, 1989). The present study addressed primarily this second aspect, specifically, the problem of individual differences in these tasks. INDIVIDUAL DIFFERENCES IN COGNITIVE ADDITION 327 Although, individual differences in solving simple arithmetic problems in children was extensively examined (see Ashcraft, 1990; Siegler, 1988), few studies have investigated this topic in adults (Geary & Widaman, 1987; Widaman et aI., 1989). In these studies, the authors presented a number of psychometric and chronometric tests to subjects and then performed confirmatory Lisrel factorial analysis (Joreskog & Sorbom, 1993). This analysis presents the advantage of representing at the same time latent variables and the relationships between them. However, the problem of individual differences may be approached in another way, by comparing group data to individual data (Svenson, 1985). The latter approach is adopted here. Our assumption is that adults, like children, use different procedures to solve simple addition problems (see Baroody, 1994; LeFevre, Sadesky, & Bisanz, 1996). Therefore, individual RTs will be fit by different structural variables, even if the group RTs are fit by only one variable. This approach confirms results obtained by other authors (Baroody, 1994; Geary & Wiley, 1991; LeFevre et aI., 1996) by using a different method. In this sense, our assumption is similar to theirs. To test this hypothesis, we will use a comparison between individual and group data. Method Subjects Fifty first year students in psychology participated to this experiment. Materials The problem set was composed of the 100 possible combinations of addends 0-9 and augends 0-9. These combinations were presented both with the correct sum (e.g., 3 + 4 = 7), and an incorrect sum. Fifty problems had an incorrect sum that differed from the correct one by ±1 (small difference), and 50 differed from the correct sum by ±5 (large difference). In all, 200 additions were presented. The frequency and placement of all integers and problems were counterbalanced, and no addend, augend, or sum was presented on consecutive trials. Procedures A verification task was presented using an Atari 1040 ST computer equipped with a monochrome monitor. Latencies (RTs) were measured by an internal clock of the computer with a precision of ±1 ms. Subjects were tested individually and were told that their task was to respond "True" or "False" to the problem presented by pressing the appropriate key. A practice set of 20 problems was presented at the beginning of the experiment. A short break was allowed after 50 problems. The total time to complete this experiment was approximatively 15 minutes. 328 BERNOUSSI Results The principal objective of this study was to examine the individual differences in the area of cognitive addition by comparing group data (general analyses) to individual data (differential analyses); we will present results with respect to this. Accuracy of response. The percentage of error in this experiment was only 2%. This rate is similar for those reported by other studies in this area (e.g., LeFevre et aL, 1996; Miller et aL, 1984). Only RTs of correct responses were taken into account. General analysis. In this analysis we looked for the best structural variable that provides the best fit of RT in the addition chronometric test. For this, we performed a set of regression analyses using as dependent variable the group mean RT for each problem presented and as independent variable the structural variables described above (min, sum, sum 2, and product). These analyses were carried out with SAS (Statistical Analysis System) software using the R2 procedure. This procedure makes it possible to compare simple regressions with independent variables and to select the variable which offers the best fit of RTs. The results indicated that the best predictive variable of RT means was the product structural variable (see Table 1). However, even if the product accounts for more variance, there are no significant differences between R2s of the different structural variables. This result is similar to those reported by other authors (see for example Geary et aL, 1986; Miller et aL, 1984). Table 1 Prediction of Latencies - Group Data Slope Structural Variable Min Sum Sum 2 Product .33 .31 .42 .50 42.5 25.6 3.2 .50 For the product variable, the regression equation is: RT (ms) = 988.42 + 9.28 (product). R2 = .50. F(1, 51) = 52.68, P < .0001. This finding confirmed those obtained in the cognitive arithmetic area (Geary et aL, 1986; Miller et aL, 1984). Moreover, a tie effect was observed in this study. It is well known that ties (a + a) are resolved faster than other problems. The tie problems were resolved 250 ms faster than the other problems in this study. When we suppressed the RT from ties of analyses, we noted an important increase of the variance explained by the product variable. Indeed, the percentage of explained variance rose from 50% to 78%: INDIVIDUAL DIFFERENCES IN COGNITIVE ADDITION 329 RT (ms) = 960.98 + 13.13 (product). R2 = .78. F(1, 33) = 116.61, P < .0001. In conclusion, the results obtained in this experiment confirmed the results of other studies mentioned above. It seems indeed, that the process used by subjects in our study was a retrieval process insofar as the best predictive variable was the product of the two addends. It remains to be seen whether or not the differential analysis confirms this result. Differential analysis. To conduct this analysis, we used as dependent variable each individual's RT, and we looked for the best predictive variable using the regression analysis as for the group data. For these analysis, we select a model if there is only one significant predictive variable (a significant Fwith an alpha level at least equal to .05). This set of analyses yielded the following results (Table 2). Table 2 Structural Prediction of Latencies - Individual Data ------- Model No fit Multiple regression Sum Min Sum 2 Product Number of Subjects 5 2 15 11 6 11 Median R2 Range of R2 ------------.21 .24 .22 .28 .08-.43 .07-.40 .09-.33 .09-.46 Five subjects were not adjusted to any model. No variable made a significant contribution to the explanation of the variance. For two subjects, only one multiple equation of regression predicted their RT, insofar as the structural variables were considered as exclusive and associated to a different type of particular process. In fact, these subjects seemed to have used some strategies, or combinations of strategies not described by the classical structural variables. The RTs of 15 subjects adjusted to the model described by the structural variable "sum." This variable corresponds to a direct-access/ counting process. The RTs of 11 subjects adjusted perfectly to the min model. This answering pattern corresponds, theoretically, to the utilization of a counting strategy. However, we did not assume that they really used this strategy, but rather a procedure similar to counting. The RTs of 11 subjects adjusted to the structural variable "product," and those of six others to the "sum 2" variable. These two structural variables are associated to a retrieval model. These results showed that with exception of the subjects who did not adjust to any model, the group was constituted of three subgroups, each identified by the structural variable which provided the best fit of RTs. 330 BERNOUSSI To study the difference between these groups, we used a one-way ANOVA on mean groups RTs. Results showed a significant difference between groups, F(3, 39) = 2.85, P < .05. 1Z00 .---------------------------------------R 1000 T 800 600 m 400 s ) ZOO o Sum Sum.Z Min Prod Figure 1. Mean RT for each group. As presented in Figure 1, the sum 2 group seems to be the faster one. Partial comparison with Fischer's PLSD revealed only one significant difference between sum 2 and min groups on the one hand, and the product group on the other hand. All the other differences were not significant. Discussion The purpose of this study was to address individual differences in cognitive addition by comparing group data to individual data. In the area of cognitive arithmetic, one important factor is the size effect of the problem (Ashcraft, 1995; Geary, 1996): RTs are slower for larger problems (8+9) than for smaller ones (2+3). The most important studies in this domain demonstrated that adults' RTs are best fit by the product of the two addends. They suggest that adults use a retrieval process (Ashcraft, 1982; Geary et aL, 1986; Miller et aL, 1984). The results obtained in this study indicated that when we use the group data (averaging data), the best structural variable is the product. This result is consistent with other findings in this area (e.g., Miller et aL, 1984; Widaman et aL, 1989). But when we consider individual data, we find a large number of individual differences. Subjects seem to use different strategies, and RTs are fitted by different structural variables. Our findings are also consistent with other studies. Geary and Wiley (1991) using a self-report procedure with students who solved simple addition problems noted that their subjects used nonretrieval procedures in 12% of trials. Recently, Geary (1996) found that American students used a combination of counting and decomposition to solve 27% of the problems presented. All these procedures are non retrieval ones. LeFevre et aL (1996) reported that adults use a variety of nonretrieval procedures to solve simple addition problems. INDIVIDUAL DIFFERENCES IN COGNITIVE ADDITION 331 How can we explain these individual differences? According to Geary (1994) , individual differences in cognitive arithmetic can be approached from three perspectives: psychometric, cognitive , and behavioral genetic. In this experiment, we can make the assumption that at least two categories of variables can explain these differences : psychometric factors (variation in mathematical abil ity or in IQ) and cognitive factors (speed of processing). To test these assumptions, other experiments are necessary. Despite differences in methods, our results together with those of these studies confirm that individual differences in cognitive addition are important to explain adult performance as well as that of children (Siegler, 1987; Siegler & Jenkins, 1989). Taking into account these individual differences permits not only the validation of general cognitive models (Underwood , 1975) but the construction of models relying on both the individual and the universal and achieves complementarity between differential cognitive psychology and general psychology. References ASHCRAFT, M. H. (1982). The development of mental arithmetic : A chronometric approach. Developmental Review, 4, 157-170. ASHCRAFT, M. H. (1990). Strategic processing in children 's mental arithmetic: A review and proposal. In D. F. Bjorklund (Ed.) , Children 's strategies: Contemporary views of cognitive development. Hillsdale, NJ: Erlbaum. ASHCRAFT, M. H. (1995). Cognitive psychology and simple arithmetic: A review and summary of new directions. Mathematical Cognition, 1, 3-34. ASHCRAFT, M. H., & BATTAGLIA, J. (1978). Cognitive arithmetic: Evidence for retrieval and decision processes in mental addition. Journal of Experimental Psychology: Human Learning and Memory, 4 (5), 527-538. BAROODY, A. J. (1994). A reexamination of mental arithmetic models and data: A reply to Ashcraft. Developmental Review, 4, 148-156. CAMPBELL, J. I. D. (1987). The role of associative inference in learning and retrieving arithmetic facts. In J. Siobodab & Rogers (Eds.) , Cognitive processes in mathematics. Oxford: Oxford University Press. CAMPBELL, J. I. D. (1995). Mechanisms of simple addition and multiplication: A modified network interference theory and simulation . Mathematical Cognition, 1, 2,121-164. GEARY, D. C. (1987). Cognitive addition: On the convergence of statistical and conceptual models. Bulletin of the Psychonomic Society, 25(6) , 477-480. GEARY, D. C. (1994). Children 's mathematical development. Washington: American Psychological Association . GEARY, D. C. (1996) . The problem-size effect in mental addition: Development and cross-notational trends. Mathematical Cognition, 2, 63-94 GEARY" D. C. & WIDAMAN , K. F. (1987). Individual differences in cognitive arithmetic. Journal of Experimental Psychology: General, 116(2}, 154-171 . 332 BERNOUSSI GEARY, D. C., WIDAMAN, K. F., & UTILE, T. D. (1986). Cognitive addition and multiplication: Evidence for a single memory network. Memory & Cognition, 14(6), 478-487. GEARY, D. C., & WILEY, J. G. (1991). Cognitive addition: Strategy choice and speed-of-processing differences in young and elderly adults. Psychology and Aging, 6, 474-483. GROEN, G. J., & PARKMAN, J. M. (1972). A chronometric analysis of simple addition. Psychological Review, 79(4), 329-343. JORESKOG, K., & SORBOM, D. (1993). Lisrel8: Structural equation modeling. Scientific Software International. LEFEVRE, J. A., SAD ESKY, G. S., & BISANZ, J. (1996). Selection of procedures in mental addition: Reassessing the problem size effect in adults. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22,1, 216-230. LEMAIRE, P., & BERNOUSSI, M. (1991). Arithmetique cognitive: Processus, developpement et differences individuelles. L'Annee Psychologique, 91, 419-438. MCCLOSKEY, M., HARLEY, W., & SOKOL, S. M. (1991). Models of arithmetic retrieval: An evaluation in light of findings from normal and brain-damaged subjects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17(3), 377-397. MILLER, K., PERLMUTIER, M., & KEATING, D. (1984). Cognitive arithmetic: Comparison of operations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(1), 46-60. PELLEGRINO, J. W., & GOLDMAN, S. R. (1989). Mental chronometry and individual differences in cognitive processes: Common pitfalls and their solutions. Learning and Individual Differences, 1(2),203-225. SIEGLER, R. S. (1987). The perils of averaging data over strategies: An example from children's addition. Journal of Experimental Psychology: General, 106(3), 250-264. SIEGLER, R. S. (1988). Individual differences in strategy choice: Good students, not-so-good students and perfectionists. Child Development, 59, 833-851. SIEGLER, R. S., & JENKINS, E. A. (1989). How children discover new strategies. Hillsdale, NJ: Erlbaum. SVENSON, O. (1985). Memory retrieval of answers of simple addition as reflected in response latencies. Acta Psychologica, 59, 285-304. TIMMERS, L., & CLAEYS, W. (1990). The generality of mental addition models: Simple and complex addition in a decision-production task. Memory & Cognition, 18, 310-320. UNDERWOOD, B. J. (1975). Individual differences as a crucible in theory construction. American Psychologist, 30, 128-134. WIDAMAN, K. F., GEARY, D. C., CORMIER, P., & LITTLE, T. D. (1989). A componential model of mental addition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15(5),898-919.
© Copyright 2026 Paperzz