Introduction:

This document contains the models and results for the statistical analyses performed for Experiments 1 and 2 in the article submitted to Brain Sciences by Shae Morgan, Sarah Ferugson, Skyler Jennings, and Ashton Crain. This article is published with Open Access.

Experiment I examines the number of times clear and conversational speech was rated as sounding angry by younger adults with normal hearing, younger adults with a simulated hearing loss, and older adults with hearing loss. We also compared performance with a control group of younger adults with normal hearing who heard stimuli that were processed, but without attenuation (i.e., without the simulated loss). This resulted in a control for the other factors of the processing.

Experiment II examines the number of times clear and conversational speech was rated as sounding angry by older adults with essentially normal hearing (hearing loss was allowed at frequencies >= 4000 Hz)

Sanity check

First, we confirm that the processed control stimuli without hearing loss simulation are similar to the younger adults with normal hearing (YNH) (i.e., that there is no effect of listener group between these two groups - no effect of processing unrelated to the hearing loss simulation)

## Warning: package 'gridExtra' was built under R version 4.0.5

Here's our model comparing Listener Group (LG; varies by analysis), Talker Group (TG; Talker Group A are "good" clear speakers - large intelligibility benefit, Talker Group B are "poor" clear speakers - small intelligibility benefit), and Speaking Style (Style; clear vs conversational) for YNH and control stimuli

YprocData <- subset(dat, LG == 0 | LG==1)
ProcModel <- lmer(Anger ~ Style*LG*TG + (1|TalkerNo), data = YprocData)
anova(ProcModel)

These results show a significant effect of speaking style on perceived anger (expected) and a significant interaction between speaking style and talker group, but no listener group effect, or any interactions involving listener group (suggesting no effect of processing).

Experiment I (EMO3):

The next step is to evaluate the comparison of data from the simulated HL group vs younger adults wtih normal hearing (for a hearing loss comparison) and simulated hearing loss vs older adults with hearing loss (for an aging comparison).

#test  YNH (LG 0), YSIM (LG 2), and OHL (LG 4)
YSimData <- subset(dat, LG == 0 | LG==2 | LG == 4)
YSimModel <- lmer(Anger ~ Style*LG*TG + (1|TalkerNo), data = YSimData)
anova(YSimModel)

Results show a significant effect of style, style x talker group, and 3-way interaction (style x TG is smaller for OHL than for YNH and YSIM). The 3-way indicates that the OHL anger is significantly less than both the YSIM and YNH ratings of anger.

Experiment II (EMO4):

Next, we need to compare older adults with normal hearing with YNH and OHL data from Morgan and Ferguson (2017). The model used to assess this is below -

#test  YNH (LG 0) ONH (LG 3) vs OHL (LG 4)
ONHdata <- subset(dat, LG == 0 | LG==3 | LG == 4)
ONHmodel <- lmer(Anger ~ Style*LG*TG + (1|TalkerNo), data = ONHdata)
anova(ONHmodel)

This analysis yields a similar result - a significant style effect, significant style x talker group interaction, but a 3-way talker group x listener group x style interaction - suggesting more anger selection by YNH than ONH and OHL listeners.

Confirming group differences

Morgan and Ferguson (2017) showed sig. differences between YNH and OHL - so if YSIM is more like YNH and ONH is more like OHL, then we should see differences between YSIM and ONH that are similar to Morgan & Ferguson (2017).

Final analysis to confirm all of this is the comparison of YSIM and ONH - showing a significant listener group difference, style x talker group interaction and 3-way (that the style x talker group magnitude differs by LG)

Statistical model to confirm:

#test  YNH ONH vs OHL
ONHvYsimData <- subset(dat, LG == 2 | LG==3)
ONHvYsimModel <- lmer(Anger ~ Style*LG*TG + (1|TalkerNo), data = ONHvYsimData)
anova(ONHvYsimModel)

Plot for visualizing the effects

Summary Plot:

Final 4-panel plot to summarize all the data: You can see the effect of hearing loss by comparing columns, and the effect of aging by comparing rows. You can see that there is no difference between the columns of data (YNH vs YSIM and ONH vs OHL) and a significant difference between the rows of data (YNH vs ONH and YSIM vs OHL).