5.8.2.1 Response Latencies
A pseudo-R2 calculated for linear mixed models showed that the random effects and fixed effects together in this model described 55.89% of the variance in RTs for both word and nonword targets; random effects described 37.50% of the variance in RTs while fixed effects described 18.39% of the variance in RTs. Table 5-7 presents the estimated standardised coefficients for the fixed effects in the model. Visual inspection of residual plots for the model also did not reveal any obvious deviations from homoscedasticity or normality, thus the model was kept as the full model in which p-values were obtained by likelihood ratio tests of the full model with the effect in question against the model without the effect in question. Results from the model are described in the following sub-sections.
Covariates. As seen in Table 5-7, the following covariates were significant: trial order number, refresh rate, sex, and age. Participants were more likely to be faster as they progressed through the lexical decision. They were also more likely to be slower when using a computer with a slower display refresh rate. Females were more likely to be faster than males. Last, older participants were more likely to be slower.
Amount of Qur’an memorisation. There was a significant main effect of amount of memorisation on RTs, β = .035, SE = .016, χ2(1) = 4.916, p < .05. Participants who had memorised more of the Qur’an were more likely to be slower in lexical decision than participants who had memorised less.
Qur’an vocabulary knowledge. There was a significant main effect of Qur’an vocabulary knowledge on RTs, β = -.109, SE = .017, χ2(1) = 38.374, p < .001. Participants with more Qur’an vocabulary knowledge were more likely to be faster in lexical decision than participants with less Qur’an vocabulary knowledge. It may be worth noting that there is also a significant two-way interaction between memorisation and vocabulary knowledge on RTs, β = -.033, SE = .015, χ2(1) = 6.637, p < .05, but the interpretation of this interaction should be contextualised by the significant three-way interaction between lexicality, memorisation, and
Lexicality. Overall, participants were more likely to be faster when responding to real words than when responding to nonwords, as shown by the pairwise comparison in the model, β = -.173, SE = .060, χ2(1) = 331.780, p < .001.
A significant three-way interaction between lexicality, memorisation, and vocabulary knowledge indicated that this lexicality effect on RTs (difference in RTs of real words versus nonwords) was moderated by both amount of memorisation and vocabulary knowledge, β = -.013, SE = .006, χ2(1) = 4.841, p < .05. As can be seen in Figure 5-22, as amount of vocabulary knowledge increases, the increase in lexicality effect on RTs as amount of memorisation increases gets smaller; participants who had very high vocabulary knowledge thus were much faster in identifying real words and in rejecting nonwords if they had memorised more of the Qur’an than if they had memorised less of the Qur’an.
Figure 5-22. Lexicality × Vocabulary Knowledge × Memorisation interaction:
Predicted RTs for word and nonword targets based on final linear mixed effects model. Results are presented as a function of memorisation and Qur’anic vocabulary knowledge (QVT) z-scores. Bands are based on 95% confidence intervals.
5.8.2.2 Accuracy
A pseudo-R2 calculated for generalised linear mixed models showed that the random effects and fixed effects together in this model described 27.21% of the variance in accuracy for both word and nonword targets; random effects described 14.64% of the variance in accuracy while fixed effects described 12.57% of the variance in accuracy. Table 5-8 presents the estimated standardised coefficients and for the fixed effects in the model. Visual inspection of residual plots for the model also did not reveal any obvious deviations from homoscedasticity or normality, thus the model was kept as the full model in which p-values were obtained by likelihood ratio tests of the full model with the effect in question against the model without the effect in question. Results from the model are described in the following sub-sections.
Covariates. As seen in Table 5-8, the following covariates were significant: trial order number, refresh rate, sex. Participants were more likely to be accurate as they progressed through the lexical decision. They were also less likely to be accurate when using a computer with a slower display refresh rate. Females were more likely to be accurate than males.
Amount of memorisation. There was a significant main effect of amount of memorisation on accuracy, β = .211, SE = .059, χ2(1) = 12.366, p < .001. Participants who had memorised more of the Qur’an were more likely to be accurate than participants who had memorised less; for each standardised unit change in amount of memorisation, the log odds of accuracy increase by .211.
Qur’an vocabulary knowledge. There was a significant main effect of Qur’an vocabulary knowledge on accuracy, β = .639, SE = .064, χ2(1) = 83.669, p < .001. Participants with more Qur’an vocabulary knowledge were more likely to be accurate in lexical decision than participants with less Qur’an vocabulary knowledge; for every one standardised unit change in Qur’an vocabulary knowledge, the log odds of accuracy increase by .639.
Lexicality. Overall, participants were more likely to be accurate when responding to real words than when responding to nonwords, as shown by the pairwise comparison in the model, β = .865, SE = .058, χ2(1) = 151.980, p < .001.
Significant two-way interactions indicated that this lexicality effect was moderated by both amount of memorisation, β = -.159, SE = .060, χ2(1) = 6.800, p < .05, and Qur’an vocabulary knowledge, β = -.172, SE = .065, χ2(1) = 6.837, p < .05, respectively. As amount of memorisation increases, lexicality effect decreases, i.e., the difference in the accuracy of real words versus that of nonwords decreases (see Figure 5-23). Figure 5-23 also shows that participants who had memorised more of the Qur’an were predicted to have a higher probability of accurately rejecting nonwords (almost as high as accurately identifying real words) than participants who had memorised less of the Qur’an. Similarly, as Qur’an vocabulary knowledge increases, lexicality effect decreases (see Figure 5-24); participants who had more Qur’an vocabulary knowledge were predicted to have a higher probability of accurately rejecting nonwords (almost as high as accurately identifying real words) than participants who had less Qur’an vocabulary knowledge.
Figure 5-23. Lexicality × Memorisation interaction: Predicted probability of
accuracy for word and nonword targets based on final linear mixed effects model. Results are presented as a function of amount of memorisation z-scores. Bands are based on 95% confidence intervals.
Figure 5-24. Lexicality × Vocabulary Knowledge interaction: Predicted probability
of accuracy for word and nonword targets based on final linear mixed effects model. Results are presented as a function of Qur’an vocabulary knowledge (QVT) z-scores. Bands are based on 95% confidence intervals.
Although the three-way Lexicality × Vocabulary Knowledge × Memorisation interaction was not significant, β = .076, SE = .063, χ2(1) = 1.428, ns., plotting it showed an interesting trend in that the lexicality effect on accuracy can be attenuated by both memorisation and vocabulary knowledge. As can be seen in Figure 5-25, as amount of memorisation increases, the lexicality effect on accuracy tends to decrease as amount of vocabulary knowledge increases; memorising more Qur’an may thus help participants with lower Qur’an vocabulary knowledge in rejecting nonwords more accurately. Looking at the three-way interaction in another way in Figure 5-26, as vocabulary knowledge increases, the lexicality effect on accuracy also tends to decrease as amount of memorisation increases; knowing more Qur’an vocabulary not only helps to improve the probability of accurately identifying real words but also in improving the probability of accurately rejecting nonwords. The lexicality effect does not exist only for participants with very high vocabulary knowledge across all levels of memorisation or for participants with more memorisation across all levels of vocabulary knowledge. In contrast, participants with both less memorisation and smaller vocabulary knowledge were not only predicted to perform the poorest in identifying real words but were also predicted to perform even much worse in rejecting nonwords.
Figure 5-25. Lexicality × Vocabulary Knowledge × Memorisation interaction:
Predicted probability of accuracy for word and nonword targets based on the full generalised linear mixed effects model. Results are presented as a function of Qur’an vocabulary knowledge (QVT) and memorisation z-scores. Bands are based on 95% confidence intervals.
Figure 5-26. Lexicality × Vocabulary Knowledge × Memorisation interaction:
Predicted probability of accuracy for word and nonword targets based on the full generalised linear mixed effects model. Results are presented as a function of memorisation and Qur’anic vocabulary knowledge (QVT) z-scores. Bands are based on 95% confidence intervals.
5.8.2.3 Lexicality (All Data)
Like the RT analyses with word targets, there was also a rather unusually high exclusion of RT data during the cleaning of the data to ensure that the lexicality RT analyses provided reliable and interpretable results. To examine the sensitivity of the results to the exclusion of observations, a supplementary analysis was done with all word and nonword RT data that was more than 200ms as faster latencies typically indicate either a technical or participant error. Table 5-7 presents the estimates for the full linear mixed effects model.
Overall, there was a similar pattern of findings for both the analyses with cleaned data and all data. The only difference was that unlike in the analysis with cleaned data, the analysis with all data showed a significant two-way interaction between Qur’an vocabulary knowledge and lexicality.