Results are presented in chapter three but the method for deriving the measures used in the results section as well as the statistical procedures used to analyze them are presented below.
2.5.1
Simulated Cochlear Implant (CI) Users
Sound localization behaviour was evaluated using five different dependent variables; these variables are described below:
i) Mean percent of trials with front/back errors: mean percent of responses
located in the wrong front/back field where both the target and response were
Target azimuth is presented on the x-axis and response azimuth is presented on the y- axis. Responses are represented by the black circles. Correct responses lay near the positive diagonal line. Responses that were categorized as front/back errors are found in the yellow boxes.
ii) Mean percent of trials with left/right errors: mean percent of responses in the wrong left/right hemifield and both target and response were
greater than or equal to ±22.5o from the midline (Fig 12);
iii) Slope of the lateral angle regression line: describes the relationship
between target lateral angle and response lateral angle such that if the slope of the lateral angle regression line were equal to 0, mean
responses tended to be focused in the same area regardless of the target location (figs. 13 & 14). This measure assesses the left/right
component of localization in azimuth while disregarding the front/back component of localization in azimuth;
iv) Intercept of the lateral angle regression line: the value of the intercept
(+ or -) indicates the left/right hemifield where lateral bias was observed (figs. 13 & 14);
v) Scatter about the lateral angle regression line: was computed by
determining the average RMS error between the responses and the lateral angle regression line and gives an indication of the consistency of the participants’ responses (figs. 13 & 14). This measures
variability in listeners’ responses independent of the mean response to any particular target.
For each of these variables, 2-way repeated measures ANOVAs were run. Head
movement condition was one factor, and consisted of three levels (static, controlled and free); simulation condition was the other factor, and it consisted of four levels (CI+HA, BiCI, EAS+HA and BiEAS). All statistical tests in the results section are reported using the Greenhouse-Geisser (GG) correction to protect against violations of sphericity (Max & Onghena, 1999). Where interactions were significant, post-hoc analyses evaluated comparisons between: i) head movement conditions (i.e. static vs. controlled and controlled vs. free) within a simulation condition; and ii) between simulation conditions within a given head movement condition. Post-hoc analyses were performed using paired t-tests. Where interactions were not significant but a main effect of simulation was found, the variable was collapsed across head movement conditions and paired t-
Figure 12. Sample data illustrating the analysis of left/right errors in azimuth plots. Scatter plots for response and target azimuth is presented above. The general features of
the figure are described in Fig. 11. Responses that were categorized as left/right errors
Figure 13. Sample lateral angle data illustrating good localization performance. Target lateral angle is presented on the x-axis and response lateral angle is presented on the y-axis. Responses are represented by the black circles. The plot illustrates good performance in lateral angle because the slope (m) of the lateral angle regression line
(blue line) is close to 1, the intercept (yint) is close to 0 and there is very little scatter
Figure 14. Sample lateral angle data illustrating poor performance.
General characteristics of the plot are described in Fig. 13. The plot illustrates poor
performance in lateral angle because the slope (m) of the lateral angle regression line is
relatively flat, the intercept (yint) is a larger integer and there is lots of scatter about the
lateral angle regression line (red arrows represent examples of scatter). Scatter about the lateral angle regression line was quantified as the average RMS of all the possible deviations from the slope of the lateral angle regression line that could be represented by the red arrows.
tests were run between simulation conditions. Given that as many as 26 post-hoc t-tests were run for each dependent variable, False Discovery Rate Control (FDR) was used to determine which comparisons were significant. False Discovery Rate Control (FDR) has been shown to protect against Type I errors by assuming that the significance of all post- hoc tests run within a given family follows a normal distribution (Matsunga, 2007). Therefore, all significant statistical values reported in post-hoc analyses have variable αcritical within a family of tests.
In addition, assessment of sound localization performance in normally hearing
participants was performed for front/back error rates in the static condition using a 1-way repeated measures ANOVA. Post-hoc analyses were performed using paired t-tests such that each of the simulated conditions (i.e. BiCI, CI+HA, BiEAS and EAS+HA) was compared to the normally hearing virtual condition (i.e. presentation of a wideband signal over headphones using in-ear HRTFs). The FDR was also used to determine which comparisons were significant.
Performance on left/right discrimination tests was described by the mean ITD or ILD- JND (i.e. mean value calculated from 3 runs of the left/right discrimination task).
Pearson’s correlation tests were run to determine whether a significant correlation existed between the following variables:
i) Mean ITD-JND and front/back error rates in the controlled movement
condition;
ii) Mean ILD-JND and front/back error rates in the controlled movement
condition;
iii) Mean ITD-JND and slope of the regression line in the static condition;
iv) Mean ILD-JND and slope of the regression line in the static condition.
All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS version 19; IBM Corporation).
2.5.2
Real Cochlear Implant (CI) Users
The sample size for real CI users was not large enough to provide sufficient power for
statistical analyses. Therefore, descriptive statistics are presented in section 3.4 for real
localization behaviour (see Section 2.5.1). Real CI users’ values as well as simulated CI
users’ means and 95% confidence intervals for each of the dependent variables in the simulated CI conditions are depicted to illustrate whether or not the simulations were representative of real CI users. In addition, a qualitative comparison of the head movement behaviours of both real and simulated CI users in the free condition was performed to illustrate whether or not the simulations run were representative of real CI users’ head movement behaviours.
Chapter 3
3
Results
The raw data collected in this study consisted of localization responses, binaural
sensitivity thresholds and head movement tracks. In order to address the hypotheses and objectives, localization performance was quantified in terms of:
1. Front/back error rates, to assess the frequency of front/back reversals;
2. Left/right error rates, to assess the frequency of left/right confusions;
3. Slope of the lateral angle regression line, to assess the listeners’ ability to
differentiate between sound source positions;
4. Intercept of the lateral angle regression line, to assess the listeners’ lateral bias;
5. And scatter about the lateral angle regression line, to assess the consistency of
the listeners’ response.
In addition, analysis of the correlation between binaural sensitivity thresholds and either front/back error rates under controlled head movement condition or slope of the lateral angle regression line in the static condition was performed for simulated CI users. Lastly, descriptive analysis of the similarities and differences between real and simulated CI users was performed.