2. Capítulo 2: Terrenos y entorno de la investigación
2.8 El análisis de los datos arrojados por los cuestionarios
A comparative analysis between the experiments described so far was also conducted, in or- der to better assess the observed sound specificity effect. More specifically, it was compared against the voice specificity effect found in the first experiment (Chapter 1), as well as against the insignif- icant effects in Experiment 2A and 2B. If the sound specificity effect is a robust effect, the compar- ative analysis should reveal that: 1) it is not statistically different from the voice specificity effect, and 2) it is statistically stronger than the insignificant effects in Experiment 2A and 2B. Further, the voice specificity effect was also compared against the insignificant effects in Experiment 2A and 2B, with the expectation that it would be statistically stronger than the insignificant effects.
- Sound specificity vs. voice specificity
The two specificity effects in Experiments 1 and 3 were compared. The data sets of both ex- periments were collapsed into one joint set, which was then used in the analysis. Similar to the analyses for each individual experiment, a mixed-effects regression analysis was conducted, but
with an extra fixed factor added, Experiment (2 levels), coded as: 1 (Exp. 1, previous chapter) and 2 (Exp. 3). The main effects and interaction of the factors Sameness (Voice/Sound) and Experiment were assessed. Since the voice and sound specificity effects were found only for the Accuracy 20 variable, it was the only one included in the analysis.
As expected, there was a main effect of Sameness on Accuracy: β = .15, SE = .04, χ2(1) = 12.09, p < .001. No main effect of Experiment was found, β = -0.12, SE = .14, χ2(1) = .79, p =.38. 21
There was also no interaction between Sameness and Experiment, β = -0.07, SE = .08, χ2(1) = .76 , p =.38, indicating that the voice and sound specificity effects were not statistically different, hence corroborating the first prediction. 22
-
Sound specificity vs. no sound specificityThe sound specificity effect found in Experiment 3 was compared against the insignificant effects found in Experiment 2A and 2B. A mixed-effects regression analysis was conducted, with the extra fixed factor, Experiment (3 levels), coded as: 1 (Exp. 2A), 2 (Exp. 2B), and 3 (Exp. 3). The main effects and interaction of Sound Sameness and Experiment were assessed. Since the sound specificity effect was found only for recognition accuracy (Accuracy), the analysis was per- formed only on this variable.
There was no main effect of Sound Sameness on Accuracy when all three experiments were included in the comparison: β = .04, SE = .03, χ2(1) = 1.45, p = .23. No main effect of Exper- iment was found either, β = -0.12 , SE = .04, χ2(1) = .04, p =.85. Separate comparisons between the sound specificity effect and each individual insignificant effect also revealed no main effect of Sound Sameness on Accuracy: 1) β = .06, SE = .04, χ2(1) = 2.58, p = .11 (Exp.2A-Exp.3); 2) β = . 06, SE = .04, χ2(1) = 2.16, p = .14 (Exp.2B-Exp.3). There was no main effect of Experiment ei23 - ther in each of these comparisons: 1) β = .01, SE = .06, χ2(1) = .03, p = .9 (Exp.2A-Exp.3); 2) β = - 0.25, SE = .14, χ2(1) = 3.27, p = .07 (Exp.2B-Exp.3). 24
There was no interaction between Sound Sameness and Experiment when all three experi- ments were included in the comparison, β = .06, SE = .04, χ2(1) = 2.31, p =.13. Separate compar25 - isons between the sound specificity effect and each individual insignificant effect also revealed no interactions: 1) β = .06, SE = .04, χ2(1) = 2.27, p = .13 (Exp.2A-Exp.3) ; 2) β = .13, SE = .08, 26
In all the models involved in the comparative analysis, slopes for the Experiment factor were added only on a by-item
20
basis.
Only the random slope of Sound Sameness (added by-subjects and by-items) and the random slope of Experiment (added
21
by-items) were included in the respective models.
Only the random slope of Sound Sameness (added by-subjects) and the random slope of Experiment (added by-items) were
22
included in the models.
Only the random slope of Sound Sameness (added by-subjects and by-items) and the random slope of Semantics (added by-
23
subjects) were included in the respective models.
Only the random slope of Sound Sameness (added by-subjects) and the random slope of Experiment (added by-items) were
24
included in the models.
Only the random slope of Sound Sameness (added by-subjects) and the random slope of Experiment (added by-items) were
25
included in the models.
Only the random slope of Sound Sameness (added by-subjects) was included in the respective models.
χ2(1) = 2.69, p = .1 (Exp.2B-Exp.3) , suggesting that the sound specificity effect observed in Ex27 - periment 3 was not statistically different from its insignificant counterparts in Experiment 2A and 2B. Therefore, the second prediction was not satisfied.
-
Voice specificity vs. no sound specificityThe voice specificity effect found in Experiment 1 was compared against the insignificant effects found in Experiment 2A and 2B. A mixed-effects regression analysis was performed, with the extra fixed factor, Experiment (3 levels), coded as: 1 (Exp. 1, previous chapter), 2 (Exp. 2A), and 3 (Exp. 2B). The main effects and interaction of Sameness (Voice/Sound) and Experiment were as- sessed. Since the voice specificity effect was found only for recognition accuracy (Accuracy), the analysis was performed only on this variable.
There was no main effect of Sameness on Accuracy when all three experiments were includ- ed in the comparison: β = .06, SE = .03, χ2(1) = 2.77, p = .1. No main effect of Experiment was found either, β = .07 , SE = .07, χ2(1) = .85, p = .36. Separate comparisons between the voice specificity effect and each insignificant effect revealed a main effect of Sameness on Accuracy: 1) β = .08, SE = .04, χ2(1) = 4.60, p = .03 (Exp.1-Exp.2A); 2) β = .09, SE = .04, χ2(1) = 4.41, p = .04 (Exp.1-Exp.2B). There was no main effect of Experiment in each of these comparisons: 1) β = 28 -0.14, SE = .14, χ2(1) = 1.00, p = .32 (Exp.1-Exp.2A); 2) β = .07, SE = .07, χ2(1) = .82, p = .36 (Exp.1-Exp.2B). 29
There was an interaction between Sameness and Experiment when all three experiments were included in the comparison, β = -0.10, SE = .04, χ2(1) = 5.90, p =.02. Separate comparisons between the voice specificity effect and each insignificant effect also revealed significant interac- tions: 1) β = -0.18, SE = .08, χ2(1) = 5.11, p = .02 (Exp.1-Exp.2A) ; 2) β = -0.10, SE = .04, χ2(1) = 30 5.85, p = .02 (Exp.1-Exp.2B). These results indicate that in line with the prediction, the voice 31 specificity effect was statistically stronger than the insignificant sound specificity effects. The pat- tern of specificity effects across the experiments discussed so far (this chapter and the previous one) is graphically depicted in Figure 3.5.
Only the random intercepts by-subjects and by-items were included in the respective models, since adding any of the ran
27 -
dom slopes led to the models’ convergence failure.
Only the random slope of Sound Sameness (added by-subjects and by-items) and the random slope of Experiment (added
28
by-items) were included in the respective models.
Only the random slope of Sound Sameness (added by-subjects and by-items) and the random slope of Experiment (added
29
by-items) were included in the respective models.
Only the random slope of Sound Sameness (added by-subjects and by-items) and the random slope of Experiment (added
30
by-items) were included in the respective models.
Only random intercepts for subjects and items were included in the models, since adding any random slope led to the mod
31 -