Creativity and Intelligence Revisited : A Latent Variable Analysis of Wallach and Kogan ( 1965 )

Many decades of research have shown that creativity and intelligence are modestly related. Some studies, however, have found that creativity and intelligence are essentially unrelated. The best example may be Wallach and Kogan's (1965) landmark study of 151 children. In that study, 10 measures of creativity didn't correlate with 10 measures of intelligence and academic achievement (average r = .09). The present research reanalyzed these data using latent variable analysis, which can (a) assess the relations between latent creativity and intelligence variables and (b) model method variance shared by the creativity tasks. Consistent with past research, the latent originality and fluency variables significantly predicted intelligence. The relations' magnitude (around r = .20) was consistent with past research, suggesting that Wallach and Kogan's data replicate other studies of creativity and intelligence. Article: The study of creativity and intelligence is one of the oldest areas of creativity research, perhaps because of Guilford's (1950, 1967) landmark work in both fields. Psychometric research generally finds modest relationships between creativity and intelligence—within the range of r = .10 to .30—depending upon the nature of the creativity and intelligence tests and the range of intelligence in the sample (Kim, 2005; Sternberg & O'Hara, 1999). Creativity tests that involve convergent thinking, such as the Remote Associates Test, typically show higher correlations with intelligence (e.g., Kane et al., 2004; Mednick & Andrews, 1967) relative to creativity tests that involve divergent thinking, such as alternate uses tests (e.g., Plucker, 1999; Preckel, Holling, & Wiese, 2006). In an analysis of nearly 400 correlations, Torrance (1975) concluded that creativity and intelligence are only modestly related. Consistent with Torrance's conclusion, a recent meta-analysis found an overall creativity-intelligence relation of r = .174 (Kim, 2005). In light of the consistently modest relations between creativity and intelligence, studies that found essentially no relationship are particularly interesting. Perhaps the best-known study is Wallach and Kogan's (1965) classic research on creativity and intelligence in fifth-grade children. Wallach and Kogan administered 5 measures of creativity, each of which yielded a score for originality and fluency, and 10 measures of intelligence and achievement to 70 boys and 81 girls. The creativity tests were given in a casual, game-like atmosphere, which was expected to facilitate creativity. Wallach and Kogan found that the creativity tests correlated strongly with each other (average r = .41), that the intelligence and achievement tests correlated highly with each other (average r = .51), but that the creativity tests did not correlate with the intelligence tests (average r = .09). This pattern—unrelated sets of coherent creativity and intelligence tests—set Wallach and Kogan's (1965) research apart from similar studies. Getzels and Jackson's (1962) classic study, for instance, was unable to achieve strong internal consistency between the tests. The creativity tests correlated as highly with the intelligence tests as they did with each other, which implies a single latent factor. Moreover, critics of their study (McNemar, 1964) estimated the relation between creativity and intelligence to be r = .40, higher than Getzels and Jackson thought. Wallach and Kogan's study thus represents an unusually good example of an instance in which creativity and intelligence were essentially unrelated. Wallach and Kogan's (1965) study, although undoubtably sophisticated for its time, was limited by the state of statistical knowledge (cf. Cronbach, 1968). In 1965, researchers were unable to do much with multiple measures of a single construct. In modern research, Wallach and Kogan's data would be modeled with latent variable analysis, which uses observed variables to estimate relationships between latent constructs (Kline, 2005; Loehlin, 2004). Apart from its many statistical benefits—the ability to model method variance among them—latent variable analysis fits the view of creativity and intelligence as abstract latent factors responsible for patterns of relations in observed measurements (Plucker & Renzulli, 1999). Although latent variable analysis was unavailable to Wallach and Kogan, modern researchers can reanalyze their data to test their conclusions and to estimate new models that were not considered by Wallach and Kogan. Latent variable analyses can be conducted without access to the full set of raw data. Given sufficient descriptive statistics— means, standard deviations, sample sizes, and a correlation matrix—researchers can reconstruct a covariance matrix for analysis (Kline, 2005). Wallach and Kogan's research is well-suited for latent variable analysis. First, the researchers collected multiple measures of each construct. Each student completed 10 measures of intelligence and achievement and 10 measures of creativity. The 10 creativity measures can be split into 5 measures of originality and 5 measures of fluency. Second, Wallach and Kogan's sample size (n = 151) is sufficient for latent variable analysis. Although a larger sample is always desirable, a sample of 151 suffices if the measured variables relate strongly to the latent variable and if the model has relatively few parameters (see Jackson, 2001, 2003). Moreover, concerns about errors due to small sample sizes can be alleviated by bootstrapping procedures. Third, the study used high-quality measures of creativity and intelligence. The measures of creativity invented by Wallach and Kogan have become classics in the field, and the measures of intelligence and achievement were standard, widely-used measures at that time. Finally, unlike other studies (e.g., Getzels & Jackson, 1962), Wallach and Kogan's measures of creativity and intelligence formed distinct and coherent clusters, which is necessary for modeling them as distinct latent factors. The present research reanalyzed Wallach and Kogan's data using latent variable analysis. This research had two aims. First, it tested if the relation between creativity and intelligence is still negligible when the data are analyzed with more sophisticated methods. Several factors could have deflated the creativity-intelligence relation in Wallach and Kogan's research. For instance, the creativity tests shared method variance that was not shared with the intelligence tests. Conversely, the measures of intelligence and achievement shared method variance that was not shared with the creativity tests. Furthermore, the low average creativity-intelligence correlation may have been deflated by a few outlying values. A couple of the 10 intelligence measures failed to correlate appreciably with any creativity measure; when correlations are simply averaged, these measures will drag down the average creativity-intelligence relationship. Second, the present research sought to estimate models that were not estimated by Wallach and Kogan. In particular, their data offer a chance to explore the relations of originality and fluency to intelligence and to distinguish between intelligence and academic achievement. AN OVERVIEW OF THE ANALYSES Measures of Creativity Five creativity tests were administered. In the instances test, children generated instances for four categories (e.g., things that are round). In the alternate uses test, children generated uses for eight common objects (e.g., newspaper, knife, cork). In the similarities test, children generated ways in which pairs of objects were similar for 10 object pairs (e.g., milk and meat, a train and a tractor). In the pattern meanings test, children viewed 8 abstract visual designs and suggested interpretations of the patterns. In the line meanings test, children viewed 9 pictures of continuous lines and suggested interpretations of the lines. Each test thus involved between 4 to 10