Gender differences in SCRABBLE performance and associated engagement in purposeful practice activities

Being female was significantly negatively correlated with SCRABBLE ratings, r(131) = −0.35, p < 0.001 (Supplementary Online Materials, Table 1, subsequently referred to as Table SOM1), which corresponds to d = −0.74. Gender explained around 12% of the variance in SCRABBLE ratings.

A complete correlation table with all variables is reported in Table SOM1. Cumulative SCRABBLE study alone and cumulative serious play were significantly related to SCRABBLE ratings. Current practice activities were described by several variables and could be best accounted for by three variables: SCRABBLE play, general word study (studying dictionaries and word lists), and SCRABBLE-specific practice alone (anagramming and analyzing SCRABBLE games). Our “Appendix” describes which questions provided information for the three groups of practice activities. For a validation of dividing current activity into three variables see SOM-A1 and Table SOM2. We also conducted a hierarchical multiple regression analysis predicting SCRABBLE ratings (see details reported in SOM-A2 and Table SOM3).

Finally, we conducted a mediational analysis of the relation between gender and SCRABBLE ratings (for detailed results see Table 2). When all predictor variables were tested as mediators of gender effects using the bootstrap method (Preacher & Hayes, 2008) we found that three variables had significant indirect effects. Age had an indirect effect tending to diminish the effect of gender. First tournament age had an indirect effect that tended to increase the effect of gender. Finally, current SCRABBLE-specific practice alone also had a significant indirect effect whereby SCRABBLE-specific practice alone tended to decrease the effect of gender. The set of variables included in the analysis predicted 64% of the variance in SCRABBLE rating. All variables statistically significantly related to rating in the final model except cumulative SCRABBLE study, including gender. It is particularly noteworthy that the amount of current general vocabulary study negatively related to rating. We should note that there was no evidence of a response bias in the estimates of females in comparison to males on their time inventories. An analysis of the two cumulative variables does not show a significant between-subject gender difference F(1,133) ≤0.01, p = 0.99, and F(1,133) = 0.03, p = 0.88, with males being insignificantly higher on both.

Our study replicated and greatly extended the findings reported by Halpern and Wai (2007). We found a significant difference in SCRABBLE ratings between men and women (d = −0.74), replicating Halpern and Wai (2007) findings of effects ranging from d = −0.26 (Study 1) to d = −1.25 (Study 2). Halpern and Wai (2007) collected data on the estimated life-long accumulation of hours of SCRABBLE play and study only in Study 1. This study found a significant correlation (r = 0.22, p < 0.05), which explains only 5.0% of the variance of the SCRABBLE ratings.

One of the reasons was that our study represents practice in a much more complex way. One of the most interesting findings of Study 1 is that the effect on SCRABBLE skill of engagement in different study activities conducted alone differs as a function of the particular type of activity performed alone. In fact, the hours of studying vocabulary alone was a negative predictor of SCRABBLE skill, whereas solitary practice analyzing SCRABBLE games and practicing anagrams were strong positive predictors of SCRABBLE skill.

Asking participants to report a single overall estimate for practice alone for each year would provide a sum of these durations, and thus could not predict attained SCRABBLE ratings as accurately as possible. In fact, although the cumulative amount of practice alone was found to be correlated with SCRABBLE rating in our sample (See Table SOM1), this variable does not predict significant independent variance in SCRABBLE ratings as seen in Table 3. The negative regression weight for the log-transformed amount of general word study should not be interpreted as an active cause of lower SCRABBLE performance. In fact, the zero-order correlation between SCRABBLE skill and log-transformed amount of general word study is not significantly different from zero r(131) = −0.09, p = 0.16. An alternate explanation of the negative prediction of rating is that the amount of estimated study time for general vocabulary serves as a suppressor variable because there is a correlated bias in our estimates of the amount of practice activity (Conger, 1974). In this case, the general word study variable would either be unrelated to skill acquisition, or possibly very slightly harmful. If general word study is unrelated to skill acquisition, it would still share irrelevant information with our other practice variables. By allowing unrelated variance to be statistically controlled, it would reveal a stronger relationship between purposeful practice and skill—closer to their true relationship.

Our study found large gender differences in SCRABBLE skill that are comparable to Halpern and Wai (2007). The primary goal of our study was to examine if females’ SCRABBLE skill would remain systematically lower than the males even after controlling for all differences in experience, practice, and career trajectories. We found that after the statistical control of these other factors the gender differences remained significant.

The mediation analysis shows that there are multiple indirect effects of gender on rating. Two of the demographic variables seemed related to the gender differences in SCRABBLE skill. The first is age, females were on average in this sample older, and that was associated with diminishing the gender difference. Females also began tournament play at a later age and this difference was associated with increased gender differences in ratings. However, the fact that we were unable to estimate all variance attributable to practice raises the possibility that with more detailed data on past practice a larger portion of the gender differences would be accounted for, perhaps all reliable gender differences might be ultimately explained. This is particularly possible since current SCRABBLE-specific practice alone appears to significantly decrease the effect of gender according to the mediation analysis.

We first examined correlations between demographic variables and SCRABBLE rating to assess whether the pattern from Study 1 was replicated. The correlation between being female and rating was −0.33 (p < 0.01) compared to −0.35 (p < 0.01) in Study 1. The mean SCRABBLE rating for males was 1410.15 (SD = 291.98) compared to 1215.21 (SD = 269.44) for females, d = −0.69 thus replicating the earlier observed gender differences, and again being among the largest gender differences studied in cognitive psychology. The zero-order correlations of the participants’ gender, SCRABBLE ratings, first tournament age, and current age are reported in Table SOM2. Table 4 shows the means for all variables for both males and females.

All variables measuring practice activities and other types of domain-related activities in SCRABBLE were submitted to a CFA. The results from the CFA are displayed in Table 5 and Fig. 1. The single-factor model assumes that all practice reflects the same construct and had a good fit to the data by the χ²/df and SRMR, while the TLI, χ², CFI, and RMSEA suggested an inadequate fit. The two-factor model involved splitting practice into either practice alone, on the one hand, or playing SCRABBLE and practicing with someone else, on the other, failed to improve on the one-factor model χ²(2) = 1.40, p > 0.05. The three-factor model involved separating practice alone into two types, namely SCRABBLE-specific practice alone and general vocabulary study, and this model had a better fit than both the two-factor model with χ²(2) = 15.86, p < 0.001 and the one-factor model χ²(4) = 18.35, p < 0.01. For the three-factor model, χ²/df and SRMR suggested a good fit to data, the RMSEA a moderate fit, but the TLI, χ², and CFI suggested inadequate fit. The statistically significant improvement in the model fit supports the hypothesis that all types of practice alone do not represent the same theoretical construct. Given our main interest is to mediate gender differences, to improve the fit of the model, we generated a four-factor model by creating a new variable measuring engagement in games that were not part of actual SCRABBLE competitions, including other word games and games against oneself. This model also improved on the model fit over the three-factor model χ²(2) = 18.91, p < 0.001 given a good fit for all measures except for the TLI and CFI, which suggested some misfit. Finally, we generated a five-factor model where playing in a SCRABBLE tournament was considered a separate factor because this activity differs from other types of SCRABBLE play. This model also significantly improved model fit χ²(3) = 8.1, p < 0.05 over the four-factor model and gave good or better fit to data on all measures of model fit.

When the demographic variables were added to the five-factor model and a structural equation model was tested predicting skill, the resulting final model had a good fit to the data with the χ²/df and SRMR, the RMSEA a moderate fit for the TLI, while χ² and CFI suggested an imperfect fit. Overall, the fit suggests that we have an interpretable model as shown in Table 6 (see Fig. 1). In our final model, current SCRABBLE skill was predicted by starting SCRABBLE at a younger age, starting SCRABBLE tournament play younger, and being older. When practice and experience are accounted for, the duration of involvement in SCRABBLE benefits skill, whereas current age is not associated with a decline of skill. Consistent with the pattern observed in Study 1, SCRABBLE practice alone positively predicted skill and general word study negatively predicted skill; no other study variables were significantly related to skill.

Most interestingly, our final model showed no significant relation between gender and skill. To better understand how we explained the gender effect, we examined the partial co-variances within the model. According to this model females were found to study SCRABBLE alone significantly less than males (p < 0.01), but again this did not reflect an overall gap in domain engagement, as females played non-tournament SCRABBLE games significantly more (p < 0.01) than males. Factor loadings for this model are in Table SOM5.

Finally, the mediation analysis from Study 1 was replicated using the variables in Study 2A. In this analysis, to be consistent with Study 1, sum scores were generated based on the five-factor model identified previously. This analysis found the same pattern as in Study 1, with age having a positive indirect effect of b = 72.43 with 95% CI (15.90, 155.85), and where age also significantly increased the effect of gender. Tournament age had a negative indirect effect of b = −86.12 with 95% CI (−168.62, −29.99), where first tournament age significantly accounted for the lower SCRABBLE ratings of females. Finally, SCRABBLE-specific practice alone had a significant indirect effect b = −25.40 with 95% CI (−73.66, −3.39), where increased SCRABBLE-specific practice alone by males significantly accounted for the lower ratings of females. Consistent with Study 1, no other variables were significantly related to gender differences in performance. The detailed results are reported in Table 6. There was no evidence of a general response bias in the reporting of hours in males compared to females on their time inventories. An analysis of the combined gender effect of the five practice activities shows no overall gender difference F(1,120) = 0.64, p = 0.43 with females reporting an insignificantly larger amount.

In Study 2A we were more able to account for the role of estimated accumulated SCRABBLE-specific practice than in Study 1, where the participants gave a single estimate for practice alone that included both SCRABBLE-specific practice and general vocabulary learning. We are confident that our findings in Study 2A are consistent with the findings from Study 1. When we regressed the accumulated estimates of all types of practice alone in Study 1, the standardized regression weight for cumulative practice alone was 0.12, p > 0.05; in contrast, the corresponding regression weights for cumulative SCRABBLE-specific practice was 0.46, p < 0.001 and for cumulative general word study was −0.45, p < 0.001. As skill increases, players engaging in practice activities directly related to SCRABBLE spend relatively less time building their vocabularies, as demonstrated in both concurrent activities in Study 1 and cumulative estimates in Study 2A. While the factor analysis identified five different independent factors to describe the shared variance in engagement in many SCRABBLE-related activities, only two of these factors were statistically significantly related directly to SCRABBLE skill in our Study 2A model. These were the two types of practice identified in Study 1, with general vocabulary study (an activity that is done alone) being negatively related to skill and SCRABBLE-specific practice alone being positively related to skill.

Our goal with Study 2B was to search for variables that could explain the gender differences in engagement in effective practice activities, namely SCRABBLE-specific practice alone, which we previously established to be related to the acquisition of higher SCRABBLE skill. We followed the logic that variables that predicted gender and/or skill would be the best candidates to mediate the behavioral differences (which we had found in Study 2A) to be most pivotal in accounting for the gender differences in SCRABBLE skill.

We found that personality measures, such as Grit and Harmonious Passion did not differ between the genders and that Obsessive Passion was not a significant mediator of the association between gender and enjoyment in SCRABBLE-specific practice alone. Each of the personality variables significantly correlated with SCRABBLE ratings. Grit was significantly correlated with skill r(120) = 0.22, p = 0.02. Harmonious Passion was negatively related to skill r(120) = −0.20, p = 0.03. Obsessive Passion was positively related to skill r(120) = 0.20, p = 0.03. Finally, the personality variables did not significantly predict either skill or gender when other ratings, particularly the influential ratings of enjoyment, were statistically controlled.

Our analysis of the participants’ ratings of practice-related activities revealed surprising findings. The activity judged by the participants to be most relevant to improving one’s skill was playing SCRABBLE, which players rated much more relevant than SCRABBLE-specific practice alone. These ratings are inconsistent with the correlations in Study 1 and Study 2A, which found that amount of engagement in SCRABBLE-specific practice alone, but not SCRABBLE play, correlated with SCRABBLE skill. Our most important finding was that only the ratings of enjoyment of SCRABBLE practice activities predicted unique variance in gender and provided a full mediation of the observed significant relations between gender and amount of engagement in SCRABBLE activities, particularly SCRABBLE-specific practice alone.

The most important finding of Study 2B was that the rated subjective enjoyment of the different practice activities accounted for significant unique variance in gender and fully mediated the relation between gender and the engagement in SCRABBLE-specific practice alone. In sum, the ratings of enjoyment explain the gender difference in the behavior which best predicted SCRABBLE skill in Study 2A and was found to significantly mediate gender differences in skill in Study 1 and Study 2A.

For domains with a long tradition of training with professional instructors, there is a large body of research showing that a substantial proportion of individual differences in skilled performance is attributable to the differences in accumulated deliberate practice. Given that SCRABBLE does not have this body of developed knowledge about effective training, our study was forced to take an alternative and more inductive approach. Therefore, we collected information on the amount of engagement during their careers in a more diverse set of practice and play activities conducted alone or conducted with others. Our analyses in Study 1 showed that the amount of current engagement in SCRABBLE-specific practice alone was statistically significantly related to SCRABBLE skill, but the amount of time spent playing SCRABBLE was not. Our Study 2A showed that accumulated amount of SCRABBLE-specific practice alone was the best predictor of SCRABBLE ratings. The activities of SCRABBLE-specific practice alone match the practice activities in chess that have been found to be predictive of chess skill (Charness et al., 2005).

Our studies provide strong converging evidence for a basic model of skill acquisition in SCRABBLE, emphasizing SCRABBLE-specific practice activities alone and deemphasizing activities designed to build general vocabulary knowledge. This demonstrates that various practice variables had very different relationships with SCRABBLE skill ranging from positive (β = 0.46), to non-significant, to negative (β = −0.45) after controlling for the other practice variables. There have been recent attempts to propose a much simplified definition of deliberate practice as any structured domain activity (Macnamara, Hambrick, & Oswald, 2014). In contrast, our study suggests that purposeful practice cannot be measured by just any operationalization of practice hours, but that effective practice is much more specific and directly related to training activities that focus on changing aspects of the target performance (Ericsson, 2016; Ericsson et al., 1993). Some activities appear to be acting as suppressor variables with negative regression weights, which is inconsistent with the practice of measuring purposeful and deliberate practice by summing up all experience in the domain as proposed by Macnamara et al. (2014).

Study 2A offered an innovative approach for identifying practice-related activities where the amounts of engagement predicted attained skill. We collected data on many of these activities in SCRABBLE and conducted a factor analysis to identify different factors of co-varying levels of engagement in practice activities across the participants’ career in the domain. Our factor analysis identified five different types of practice activity factors, and our structural equation model showed that only the amount of engagement in SCRABBLE-specific practice alone positively related to SCRABBLE skill. In contrast, general vocabulary study was negatively predictive of attained SCRABBLE skill. Study 2A also identified factors associated with playing word games, non-tournament SCRABBLE play, and playing in SCRABBLE tournaments. Engagement in any of these three playing activities did not significantly correlate with attained SCRABBLE skill. Future research is needed to confirm if these factors are consistently identified and if they have a stable relation or lack of relation with skill across the full range of skill.

A theoretical analysis of appropriate, purposeful practice begins with an analysis of the target performance, namely superior performance during competition and tournaments. By generating actions in SCRABBLE situations for which superior actions are available, it is possible to identify cases where a given player does not find the best actions. Either by replaying published SCRABBLE games by elite players or consulting a SCRABBLE computer program, the player can get feedback on the quality of their move. If the selected move is not optimal, then the player needs to review how they generated their action and determine how they would have needed to proceed to generate the superior actions. This described activity corresponds well with the variable SCRABBLE-specific practice, for which we found that more engagement correlated with attained SCRABBLE skill.

In Study 2A we identified two factors related to practice one of which helped account for gender differences in tournament SCRABBLE. Females more frequently played SCRABBLE outside of tournaments than males and engaged less in SCRABBLE-specific practice alone. These differences in time allocation were sufficient to reduce and mediate the superior performance of males.

There are studies of individual differences in component skills and abilities that might provide insights into the acquisition and structure of superior performance in SCRABBLE. Tuffiash et al. (2007) measured different basic verbal abilities and verbal skills and showed that the only ability that distinguished experts from non-experts was solving anagrams (rearranging letters into words)—a task that experts reported practicing by themselves for a significantly longer duration than non-experts. Recently, Hargreaves, Pexman, Zdrazilova, and Sargious (2011) found that the superior verbal fluency of expert SCRABBLE players was only observed for tasks related to SCRABBLE.

There are several different approaches to account for individual differences in engagement in practice activities. One approach is to collect data on how different participants perceive their practice activities regarding their relevance for improvement, the amount of effort required, and the amount of inherent enjoyment of those activities. An alternative approach is to search for correlations with differences in general personality traits. We collected data on three personality measures: Grit, Harmonious Passion, and Obsessive Passion. Our analysis found that the variables which uniquely explained variance in SCRABBLE skill involved perceptions of enjoyment of two types of practice activities, including the key variable of SCRABBLE-specific practice alone. In fact, enjoyment of SCRABBLE-specific practice alone fully mediated the gender differences in that activity.

Pinker’s (2008) general argument that males put more obsessive focus into their careers and competitive domains might explain males higher involvement than females in the more effective learning activities. This argument is also consistent with significantly higher Obsessive Passion scores of males compared to females. One problem with an account based on competitiveness is that we are not sure how to objectively measure competitiveness. The studies mentioned in the introduction experimentally manipulated competitive vs non-competitive situations, but SCRABBLE rating is based only on competitive situations. The closest objective measure we can think of is how often one plays SCRABBLE, and it is interesting that in Study 1 females engaged in statistically significantly more current SCRABBLE play (Table SOM1) while in Study 2A, there were no differences in the amount of tournament play. The only statistically significant differences we found in terms of reaction to competitive situations were that males reported Tournaments being more effortful while females reported playing SCRABBLE was more enjoyable. This could reflect a different approach to competition/playing SCRABBLE that might warrant further investigation.

Differences between genders are observed in some domains with expert performance, but not others. One domain without consistent gender effects is music. The domain of music is a domain with a very long history of training with developed curricula and professional teachers who start supervising training at a very young age. Ericsson et al.’s (1993) study of violinists found that all three groups of students at a music academy engaged in supervised practice, meeting the criteria for deliberate practice. In that domain, all groups had started practice at young ages—around age seven. All groups rated the relevance of practice alone as the most beneficial activity for improving skill. There is a significant body of knowledge about how to best improve at chess as well, and relatively few professional teachers who start supervising practice at more advanced levels with older players. In studies of the acquisition of expert performance in chess, Charness et al. (2005) found a stronger relation between chess skill and the age at the beginning of serious play than at the age of starting in the domain. Additionally, they found generally non-significant, positive effects of tournament play across different samples and types of samples. Most importantly, Charness et al. (2005) found evidence for an independent effect of engagement in purposeful practice for chess skill, even after controlling for other types of practice activities. Gobet and Campitelli (2007) also found that the amount of group practice (defined as playing chess, working with a tutor, or studying with another person) was a strong predictor of attained skill. When tournament chess players rated the relevance of different practice activities, Charness et al. (1996) found the average rating for analysis of games alone to be high (5.9 on a seven-point scale), whereas relevance of playing chess games outside tournaments and against chess computers was low, 3.6 and 3.2, respectively. Even more interesting, Charness et al. (1996) found that the ratings for the relevance of analyzing games alone were significantly correlated with chess ratings positively, whereas the ratings for the relevance of playing games were significantly negatively related. The relevance ratings show that the more skilled chess players knew which activities were associated with improved performance. In the recently established domain of SCRABBLE, we found hardly any evidence that players were able to accurately judge the relevance of different practice activities for improving performance. The players even rated playing SCRABBLE games outside tournaments as being more relevant than SCRABBLE-specific practice alone.

Gender differences in the amount of deliberate and purposeful practice have also been found in research on runners, where more men are distance runners, and male runners train more than female runners (Deaner, 2013). For instance, Ceci and Williams (2010) argue that the most important cause of the difference in professional outcomes is different preferences by women. Our studies have presented significant evidence that differences in performance between females and males are related to differences in the amount of engagement in particular types of practice activities, and that those differences in engagement in practice activities are related to differences in rated enjoyment of those activities. With our current evidence, we can only speculate about the causes of the difference in rated enjoyment of deliberate and purposeful practice.