• No se han encontrado resultados

EL MARCO DE LA CONTABILIDAD DE GESTIÓN EN LOS CENTROS SANITARIOS

3.9. Conclusiones

On average, across each of the conditions, participants made their emotion judge-ments 6.73 seconds after stimulus onset (corresponding to mean stimulus length + 1.63 seconds). As in Study 3, reaction times were not compared across the different feature-attenuation conditions or emotions, since this measure was confounded by differences in stimulus length. Compared to Study 3, participants responded very slightly faster on the whole.

The average confidence rating associated with participants’ judgements, across all of the experimental trials, was 3.04, which was slightly lower than in Study 3. Rank-based factorial ANOVA revealed a very small but significant effect of Feature at-tenuation condition upon confidence ratings, F (3, 680) = 11.52, p <.001, η2 = .05. Post-hoc, Bonferroni-adjusted Wilcoxon-Mann-Whitney tests showed that only two pairwise comparisons between Feature attenuation conditions were significant:

Frequency and Intensity, and Frequency and Original – Frequency being the less

confidently-recognised condition in both cases (see Table 12). There was no signif-icant effect of Feature attenuation, F (3, 820) = 1.56, p = .194, η2 <.01, nor was there a significant interaction with Emotion, F (12, 820) = 0.90, p = .551, η2= .01.

Table 12: Pairwise, Bonferroni-adjusted Wilcoxon-Mann-Whitney tests, comparing differences in participants’ confidence ratings, according to the emotion judged.

Duration Frequency Intensity

Frequency .068

Intensity 1.000 <.001*

Original 1.000 <.001* 1.000

*= significant at .050 alpha-level.

Participants performed slightly better than in the previous chapter, in terms of the mean percentage of correct responses and, as hypothesised, performance was above chance level (20% correct) for all of the feature-attenuation conditions, even at pre-test (Figure 25). However, performance was quite variable on the whole, as indicated by the relatively large standard deviations.

Figure 25: Mean percentage correct emotion recognition, for each condition, across the different testing phases. Error bars show± 1 standard deviation.

Contrary to expectation, the training paradigm appeared to have relatively little effect – ANOVA revealed that there was no significant effect of test phase upon participants’ emotion identification accuracy, F (2, 38) = 1.18, p = .317, η2 = .06.

Despite this, the graph shows evidence of some modest improvement between the

pre-test and post-test blocks. This effect was noticeably stronger in the Duration condition, compared to the other feature attenuation conditions, indicating that training had the greatest effect when stimuli duration cues were preserved. How-ever, across all of the feature attenuation conditions, learning did not appear to generalise to previously-unheard stimuli: emotion identification accuracy was unan-imously lower during the post-test-transfer block, compared to the post-test block.

In fact, only in the Intensity condition did performance improve at all from pre-test to post-test-transfer.

Three-way loglinear analysis produced a final model that retained only the main effect of Emotion, χ2(16) = 45.51, p = .001, indicating that the emotion being judged sig-nificantly influenced identification accuracy. Conversely, neither Feature attenuation condition, χ2(15) = 21.55, p = .120, nor the interaction between the latter and Emo-tion had a significant impact upon identificaEmo-tion accuracy, χ2(12) = 16.85, p = .160.

To examine the effect of Emotion upon identification accuracy in more detail, the equivalent chi-square test was computed. This test produces standardised residuals associated with each level of an experimental condition, thereby providing an ap-proximate index of the extent to which each emotion contributed to the overall effect of Emotion reported above (Sharpe, 2015). According to established convention, standardised residuals of more than ± 2 indicate that values for that level of the variable are less compatible with the null hypothesis (Agresti, 2007; Sharpe, 2015).

Based on this criterion, Fear (Std. residual = -2.036) and Sadness (Std. residual

= 3.504) appeared to primarily drive the main effect of Emotion upon identification accuracy. Specifically, Fear tended to be recognised less accurately than expected by chance, whilst Sadness was recognised more accurately.

Confusion matrices showed that, contrary to the experimental hypothesis, there was

no appreciable effect of the training upon emotion identification accuracy (Figure 26). Certainly, the effect of training was much less strong than with the speech stimuli in Study 3. As in the previous study, however, acclimatisation to the CI-simulated stimuli appeared to reduce the extent to which Neutral was overestimated, as evidenced by fewer false positive responses for this emotion in the post-test-transfer block. Conversely, the overestimation of Anger did not clearly show a decline. As was the case in Study 3, confusions between Anger and Happiness were the most pervasive, taking into account all of the different conditions.

Between the pre-test and post-test-transfer phases there also appeared to be greater consistency in terms of the specific errors made. For example, Sadness was clearly mistaken for Fear more often than any other emotion during the post-test-transfer, whereas at pre-test there was little evidence of any pattern.

Unfortunately, none of the individual emotions were consistently associated with im-proved identification accuracy as a result of the training paradigm. In fact, substan-tial practice effects were limited to only two cases: Neutral stimuli in the Duration condition, and Sadness stimuli in the Original condition.

Sadness appeared to be the most robust emotion to disruption overall, being rela-tively well-decoded across all feature attenuation conditions. Anger stimuli were also relatively well-identified, although with far more false positive errors.

Lastly, inter-individual differences in empathy (EQ) and musical training and ex-pertise (MUSE) appeared to exert little influence on overall task performance. A multiple linear regression analysis found that neither Empathy Quotient scores nor any of the MUSE sub-components explained a significant proportion of the variance in emotion judgement accuracy, once Feature attenuation condition was controlled

Figure 26: Heat-mapped confusion matrices for music stimuli during the pre-test and post-test-transfer phases, depicting the percentage of responses in each emotion cat-egory, for each stimulus emotion. Columns denote presented emotions, rows denote emotion judgement responses. Red = higher values, green = lower values.

for. This was true for emotion identification accuracy at the pre-test, post-test and post-test transfer phases. Additionally, no significant correlations were found with any of the MUSE components or EQ scores.

5.4 Discussion