• Keine Ergebnisse gefunden

Perception with the Switch Procedure

6.4 Additional Analyses for Experiment 1

According to the considerations above, the data collected in Experiment 1 were inspected in a re-analysis (HTfin/‘same’ vs. ‘switch’) and a control analysis (HTfin vs. T1). It was hypothesised that if these two kinds of ‘new’

analyses were appropriate for the evaluation of the data collected with the Switch Procedure, the findings of the main analysis should be confirmed. To be more precise, it was expected that the effect found in the main analysis would be fortified in the other two analyses. Infants’ ability to discriminate the highly contrastive non-words pata and medo should become even more clear-cut. The strongest effect was expected for the re-analysis. In the control analysis the effect was expected to be a little bit weaker compared to the re-analysis since the amount of observations usable for the discrimination test was bisected in this case.

6.4.1 Re-Analysis

As explained in section 6.3.1, the re-analysis took all infants’ reactions to the first change in the sound signal into account (HTfin/‘same’ vs. ‘switch’),

11Note that each infant’s attention during the habituation phase decreases to the same preset criterion. The number of habituation trials each subject takes to reach this criterion varies individually.

6.4 Additional Analyses for Experiment 1 159 ignoring the second auditory change in the group of infants tested in the sw/sa order (see figure 6.3b). With a comparison of infants’ pre-shift versus post-shift attention the re-analysis follows the logic of classical habituation paradigms. It was predicted that the effect found by comparing ‘same’ and

‘switch’ trials in the main analysis would be fortified in the re-analysis.

6.4.1.1 Results

For the re-analysis of the infant control test, no outliers had to be excluded, since all looking times remained within the range of ±2 standard devia-tions of the mean looking time. Looking time was the dependent variable and the factors subject (as random factor), condition (pata | medo), trial(HTfin/same |switch) andtest order(sa/sw |sw/sa) were added as independent variables in a full-factorial ANOVA that included all pos-sible interactions. There was a main effect of trial [F(1,9) = 13.87, p <

.001, MHT f in/same = 6.64, SD= 3.78, Mswitch = 10.61, SD = 3.95]. Infants looked longer to the ‘switch’ than to the final habituation/‘same’ trial and the difference was highly significant, indicating that the 6- to 8-month-old German infants were able to discriminate the contrast. None of the other factors and none of the interactions reached the level of significance. The results are illustrated in figure 6.4.

In addition to the crucial comparison of pre- versus post-shift attention (HTfin/‘same’ vs. ‘switch’), another ANOVA was run in order to find out whether the looking times to the final habituation trial generally differed from the ‘same’ trial. 1 out of 32 observations (3.13%) was excluded since it diverged more than two standard deviations from the group mean looking time. The statistical design was identical to the preceding ANOVA with the exception that the factor trialnow consisted of the opposition ‘HTfin

| same’ (instead of ‘HTfin/same | switch’). The analysis yielded no main effect and no interactions, indicating that the final habituation trial does not differ significantly from the ‘same’ trial – irrespective of condition and test order.

Figure 6.4: German 6- to 8-month-olds’ looking behaviour in Experiment 1 (pata vs.medo; re-analysis).

6.4.1.2 Discussion

Conform to the predictions, the results of the re-analysis provide more ev-idence for the insights yielded by the main analysis. 6- to 8-month-old German infants clearly discriminate the two pseudo-words pata and medo.

This finding supports the assumption that the re-analysis is a reliable means to evaluate the data collected in the Switch Procedure. As expected, the effect became stronger in the present analysis (p < .001) than in the main analysis (p < .05). Therefore, it can be claimed that the re-analysis is effi-cient in revealing effects which may be weak or (as will be shown in chapter 7) even remain invisible in the traditional way of analysing the data.

6.4.2 Control Analysis

The data collected in Experiment 1 were further inspected in a control ana-lysis. As described in section 6.3.1, this kind of analysis compares infants’

attention during the final habituation trial to their attention during the first test trial, irrespective of the order in which infants heard the two test tri-als. Consequently, only one half of the subjects – the ‘experimental group’, that is, those infants who were tested in the sw/sa order – were useful

in-6.4 Additional Analyses for Experiment 1 161 dicators concerning contrast discrimination in this analysis. For this group, the analysis corresponds to a comparison of pre- and post-shift attention as in classical habituation paradigms and in the re-analysis of the Switch Procedure described before. The other half of the infants, those tested in the sa/sw order, acted as a control group that was presented two identical trials (see also figure 6.3c). It was predicted that the control analysis would confirm the results of the main analysis and the re-analysis, revealing suc-cessful discrimination ofpata and medo. Accordingly, it was expected that looking times during the final habituation trial and the first test trial would differ significantly for the experimental group of infants (tested in the sw/sa order) but no effect was expected for the control group (sa/sw order).

6.4.2.1 Results

None of the observations had to be excluded in the control analysis due to too much deviating looking times. The dependent variable in the ANOVA was looking time, and the factorssubject (as random factor), condition (pata | medo), trial (HTfin |T1) and test order (sa/sw | sw/sa) were independent variables in a full-factorial design including all possible inter-actions. The analysis yielded a main effect of trial [F(1,9) = 9.15, p <

.05, MHT f in = 6.18, SD = 2.85, MT1 = 9.53, SD = 4.79] with infants look-ing longer to the first test trial than to the final habituation trial. The interaction trial × test order nearly reached the level of significance [F(1,9) = 4.84, p = .0553]. There were no other effects or interactions.

As mentioned above, in order to make sense this kind of analysis requires an interaction between the factors trial and test order. The fact that this interaction just missed the level of significance in the statistical analy-sis must probably be attributed to a relatively low amount of observations as against a comparatively high amount of factors. Since the factor con-dition was not involved in any of the effects, an additional ANOVA was performed, identical to the first one with the exception that now condi-tion was not included as a factor. This second ANOVA yielded a main effect of trial [F(1,13) = 10.13, p < .01] as well as the expected inter-action of trial × test order [F(1,13) = 5.36, p < .05]. Subsequent planned comparisons revealed the predicted pattern. In the control group (sa/sw test order), looking times to the final habituation trial did not differ significantly from those to the first test trial (which is a ‘same’ trial in this case) [MHT f in = 7.01, SD = 2.48, MT1 = 7.93, SD = 4.15, p > .05]. In the experimental group (sw/sa test order), in contrast, looking times to the first test trial (a ‘switch’ in this case) were significantly longer than to the

final habituation trial [t= 3.89, p < .01, MHT f in= 5.35, SD= 3.11, MT1 = 11.14, SD= 5.10]. The diagram in figure 6.5 illustrates these results.12

Figure 6.5: German 6- to 8-month-olds’ looking behaviour in Experiment 1 (pata vs. medo; control analysis).

6.4.2.2 Discussion

The results obtained by the control analysis are a further confirmation of the insights gained by the previous analyses. First of all, they provide more evidence for the finding that German infants aged 6 to 8 months had no difficulties in discriminating the phonetically highly contrastive pseudo-word pairpata–medo. This replication of results supports the assumption that the control analysis is a proper way of investigating infants’ discrimination skills.

Moreover, it can be observed that the effect in the present analysis (p < .01) was stronger than in the main analysis (p < .05) but weaker than in the

re-12It was pointed out by Paula Fikkert (p.c.) that, although infants were exposed to identical stimuli up to that point, the looking times for the final habituation trials differ between the control group (sa/sw-order; MHT f in = 7.01) and the experimental group (sw/sa-order; MHT f in = 5.35). Given standard deviations which exceed the difference between these two values (∆ = 1.66), it seems that the discrepancy has to be considered as random variation. The same is true for the control analysis conducted for Experiment 2 (see section 7.2.4).