Question-1: results

This section presents the results obtained from the first question of the experiment, were non- expert and expert subjects in the supervised and unsupervised condition (controlled environment and online study) performed a series of A/V detection tasks. Figure56, shows the percentages of the subjects who detected correctly the image (out of three images) used to synthesise the audio stimulus. Pearson’s Chi-square tests of independence were performed to examine the relationships and test the null hypothesis of no association between the factors (i) subjects’ detection success rate and harmonicity of the corpora, and (ii) Subjects’ detection success rate and A/V mappings.

Data analysis of non-expert group in the supervised condition shows that there is a strong relationship between the ability of the subjects to detect the correct image and the harmonicity of the audio corpora X2_{(3, N = 224) = 19.22, p = .0. Overall no significant relationship was revealed}

between detection success rate and the mapping X2_{(1, N = 224) = 1.143, p = .28.}

Data analysis of the expert group in the supervised condition show that there is strong relationship between the ability of the subjects to detect the correct image and the harmonicity of the audio corpora was observed X2_{(3, N = 128) = 9.86, p = .02. Overall no significant relationship was}

110

112

Non-sound practitioner

Sound practitioner

Figure 56. Overall effects of the mappings and the corpus on the subjects' ability to detect the correct stimuli. The dotted line represents the overall average of the subjects who rated the A/V associations as similar. Upper row shows the results for the supervised condition and lower row for the unsupervised (i.e. study conducted as online survey).

Data analysis of the non-expert subjects in the unsupervised condition shows that there is a strong relationship between the ability of the subjects to detect the correct image and the harmonicity of the audio corpora X2_{(3, N = 216) = 11.37, p = .010. Overall no significant}

relationship was revealed between the variables detection success rate and the mappings X2_{(1, N = 216) =.70, p = .40.}

Data analysis of the expert subjects in the unsupervised condition shows that there is a strong relationship between the ability of the subjects to detect the correct image and the harmonicity of the audio corpora X2_{(3, N = 312) = 23.15, p < .001. Overall no significant}

relationship was revealed between the variables detection success rate and the mappings X2_{(1, N = 312) =.3.07, p = .079. Figure 57 show that the String and Wind corpus contribute more}

to the strength of the relationship between corpus and perceived similarity in all of the groups and conditions. Post-hoc testing of each group/condition chi-square results was performed for each level of the factor corpus in order to determine which factor levels contribute to the statistical significance that was observed. The results of the post-hoc test are presented in Table 15.

113

Figure 57. Percent of correct detection for each corpus.

Table 15. Post-hoc testing of each group/condition chi-square results by estimating p values from the chi-square residuals for each level of a factor and comparing these to an adjusted benferonni corrected p value. An asterisk (*) indicates statistical significance at adjusted alpha level of .006.

This section presents the results obtained from the comparison between the expert and the non-expert groups in the supervised and unsupervised condition. Pearson’s chi-square test of independence was performed on three between subject variable to test the null hypothesis of no association between: (i) the subjects’ skills and ability for correct discrimination of the A/V stimuli, and (ii) subjects’ skills and the frequency of correct detection between the two mappings, and (iii) between the four corpora.

The comparison between the expert and the non-expert groups in the supervised condition shows that there is an association between the variables subjects skills and discrimination ability X2_{(1, N = 352) = 4.43, p = .035. Examination of the cell frequencies showed that about 70% (45}

114

subjects who detected the correct image was 54% (61 out of 112). Further testing of the relationship between the variables correct detection and skills was performed by layering the participant responses based on the two levels of the mapping factor. The results of the test show that the association between the subjects’ skills and correct discrimination was stronger when the chromatic mapping was tested X2_{(1, N = 176) =4.27, p = .039 and not strong when the}

achromatic mapping was tested X2_{(1, N = 176) = .88, p = .347, see Figure 58. However, Post-}

hoc testing show that the relationship between skills and conditions was not significant, see Table 16.

The comparison between the expert and the non-expert groups in the unsupervised condition show no association between the variables subjects skills and discrimination ability X2_{(1, N = 528) = 0.10, p = .747. Further testing of the relationship between the variables correct}

detection and skills was performed by layering the participant responses based on the two levels of the mapping factor. The results of the test show that there was no association between the subjects’ skills and correct discrimination for none of the mappings: achromatic X2_{(1, N = 264)} = .011, p = .917; chromatic mapping X2_{(1, N = 264) =.332, p = .564.}

Figure 58. Differences between the expert and non-expert subjects in the supervised condition' discrimination ability as a result of the mapping.

Table 16. Post-hoc testing of each group/condition chi-square results for subjects’ skills factor.

Skills chi-square p N Expert (s) .6 .41 128 Expert (u) 2.1 .14 312 Non-Expert (s) 8.5 .0033 224 Non-Expert (u) .43 .51 216 Total X2_{(3, N = 880) = 8.711, p = .033}

115

Since there were no significant differences between the two subject groups in the two conditions, the results from all groups/conditions were merged in order to perform a chi-square test on the merged dataset, for the corpus factor that was the only factor in which we found statistically significant differences in the participant responses. The data analysis of the merged dataset shows that there was a statistically significant relationship between correct detection and corpus: X2(3, N = 880) = 9.8, p < .001. Post-hoc testing was performed to identify which levels of the factor corpus were contributing to this statistical difference. The results show that the string and the wind corpus were the factor levels that contributing to the association between participants’ discrimination ability and corpus, see Table 17.

Table 17. Post-hoc testing of the merged (group/condition) chi-square results for the corpus factor. An asterisk (*) indicates statistical significance.