The effect of residual acoustic hearing and adaptation to uncertainty in Cochlear Implant users

ONLINE SUPPLEMENTARY MATERIAL

Bob McMurrayDept. of Psychology, Dept. of Communication Sciences and Disorders, Dept. of Linguistics

University of Iowa

Ashley Farris-TrimbleDept. of Linguistics

Simon Fraser University

Michael SeedorffDept. of BiostatisticsUniversity of Iowa

Hannah RiglerDept. of PsychologyUniversity of Iowa

-1

123456789

101112131415161718192021

S1. Participants

A complete description of the CI users who participated in this study is provided below in Table S1. In addition to these standard audiological variables, we also ran a small set of clinical assessments to assess individual differences in language and cognitive abilities. These are explored as moderators in Section S4.

We assessed receptive vocabulary with the Peabody Picture Vocabulary Test (PPVT-IV; Dunn & Dunn, 2007) and expressive vocabulary with the Expressive Vocabulary Test (EVT-II; Williams, 2007). To assess nonverbal IQ, we administered two subtests (i.e. Block Design and Matrix Reasoning) of the Wechsler Abbreviated Scale of Intelligence (WASI-II; Wechsler, 2011). PPVT scores were not available for two participants (both NH), EVT scores were not available for 7 (4 CI, 3 NH), WASI MR scores were not available for 5 participants (2 CI, 3 NH) and WASI BD scores were not available for 5 participants (2 CI, 3 NH). Finally, CI users’ speech perception accuracy was assessed with the CNC word lists that were conducted in their optimal listening configuration (e.g., bimodal CI users got to use both their CI and their hearing aid; bilateral users used both CIs, etc).

All participants achieved standard scores on the PPVT-IV that were within one standard deviation of the mean. The mean PPVT standard scores were 104.4 (SD=11.7) for CI users and 110.6 (SD=10.2) for NH participants (T(48) =1.91, p=0.062). Mean standard scores for the EVT-2 were similar. CI subjects’ mean was 101.2 (SD=15.3) and NH listeners’ mean was 108.6 (SD=9.4; T(43) = 1.86, p = 0.069). Nonverbal IQ scores also did not differ. CI users had mean T scores of 48.0 (SD=9.3) and 48.3 (SD=10.8) on the block design and matrix reasoning subtests of the WASI of while NH listeners had means of 47.4 (SD=9.1) and 51.2 (SD=8.5) on the same tests (Block design: T(45) =0.23, p=0.82; matrix reasoning: T(45) =.96, p=0.34).

Audiological data in Table S1 represent Pure Tone Averages (PTA) for audiograms at 250, 500, 1000, 2000, 4000 and 8000 hz. In a few cases one or two frequencies was not tested (these are indicated with superscripts). PTAs are reported under five listening conditions.

1) Unaided audiograms were taken pre-operatively.2) Hearing Aid (HA) only audiograms are provided for bimodal and hybrid CI users for the

contralateral ear only. Some of the hybrid users wore a hearing aid on their ipsilateral ear but not reported since full PTA covers many high frequencies for which acoustic hearing would presumably have been lost due to the implant.

3) Hearing Aid (HA) only – Low Frequency (LF) audiograms report PTAs for 250, 500, 750 and 1000 for both ears. For several participants, 750 hz was not tested. For these, missing values were replaced with the average of the 500 and 1000 hz responses.

4) CI only – High Frequency (HF) audiograms report PTAs for 2000, 4000 and 8000 hz using just the cochlear implants.

5) Complete audiograms are PTAs for all frequencies in the participants typical listening conditions (all of their CIs plus all of their hearing aids).

Finally, the hybrid CIs reported do not standard long electrodes. The Nucleus EAS hybrid CIs are both 10 mm / 8 electrode CIs; The Hybrid S12 CI is a 10 mm / 10 electrode CI; the Hybrid L24 is a 16 mm / 22 electrode (the lowest 4 electrodes are not used).

-2

22

2324252627282930313233343536373839404142434445464748495051525354555657585960616263646566

Table S1: Summary of CI users participating in this study (“—”indicates PTAs was unavailable, blank cells indicate PTA would not be valid for that type of participant (e.g., a hearing aid PTA for a unilateral participant).

ID GenderAge (yrs) Etiology

Onset of deafness

Age at implant. Ear

Implant type Implant model (manufacturer)

PTA

Unaided HA only,HA only,

LFCI only,

HF CompleteL R L R L R L R L R

18 F 59 unknown 52 57 R unilateral Nucleus CI512 (Cochlear) 93 118 22 2031 M 62 hereditary 33 38 L unilateral Clarion HiRes90K (AB) 119 104+ 38º 33

57 F 48 unknown 1 36 L unilateral Clarion HiFocus 1.2/I-CII (AB) 104 130 25 2468 F 57 unknown 40 44 R unilateral Clarion HiFocus 1.2/I-CII (AB) 130 108 -- --126 M 60 Meniere’s 38 40 R unilateral Clarion Radial Bipolar/Standard 1.0 (AB) 85 97 35 34

34 F 52 unknown 40 42 B bilateral L-Clarion HiFocus II-CII;R-Clarion HiRes90K (AB) 106 108 22 25 23 24

56 F 57 unknown 47 52 B bilateral L-Nucleus CI 24RE;R-Nucleus CI512 (Cochlear) -- -- 17 18 16 18

66 F 51 unknown 31 40 B bilateral Clarion HiFocus II-CII (AB) 118 119 27 30 25 25

67 M 56 unknown 9 50 B bilateral L-Nucleus CI 24RE;R-Nucleus CI422 (Cochlear) 76 107 15 32 15 24

127 M 56 unknown 50 54 B bilateral Nucleus CI 24RE (Cochlear) 130 130 17 23 18 2041 M 58 infection -- 47 L bimodal Clarion HiFocus II-CII (AB) 91 103 87 38 27 3749 F 41 unknown -- 40 R bimodal Nucleus Hybrid L24 (Cochlear) 103 130 47º 23 18º 17º

61 M 34 unknown 31 31 L bimodal Nucleus CI512 (Cochlear) 122 94 -- -- 20 2071 F 27 unknown 18 26 R bimodal Nucleus CI 24RE (Cochlear) 90 108 61 27 23 1979 M 65 unknown 42 61 R bimodal* Nucleus Hybrid S12 (Cochlear) 91 130 38º 18 18 2583 F 60 unknown -- 57 R bimodal Nucleus CI422 (Cochlear) 94 93 57 19 20 18125 M 60 unknown 54 58 L bimodal Clarion HiRes 90K (AB) 79 15 -- -- 33º 24º

128 M 20 unknown 8 18 R bimodal Clarion HiRes 90K (AB) 952 98 -- -- 43 3514 F 57 ASL -- 49 R hybrid Nucleus EAS3(Cochlear) 83 93 61 29 31 25 2315 M 57 hereditary -- 48 R hybrid Nucleus EAS2 (Cochlear) 104 114 59º -- 39 23 2519 M 62 unknown -- 51 L hybrid Nucleus EAS2 (Cochlear) 101 97 762 38 43 27 3135 M 63 unknown 60 62 L hybrid Nucleus Hybrid L24 (Cochlear) 102 94 362 40 20 18º 22º

36 F 47 unknown 32 45 R hybrid Nucleus Hybrid S12 (Cochlear) 88 98 43º 34 22 20 1847 F 57 hereditary 1 51 L hybrid Nucleus EAS3 (Cochlear) 107 107 37 69 14 28º 2870 F 69 unknown -- 56 R hybrid Nucleus EAS2 (Cochlear) 106 101 54º 69 33 22 3273 F 62 unknown 55 55 R hybrid Nucleus EAS3 (Cochlear) 87 98 53º 36 35 18 1874 F 43 unknown 25 41 L hybrid Nucleus Hybrid S12 (Cochlear) 81 77 -- 9 -- 22 2381 F 44 unknown 40 41 R hybrid Nucleus Hybrid L24 (Cochlear) 84 101 -- -- 24 18 17124 M 60 unknown 57 59 R hybrid Nucleus Hybrid L24 (Cochlear) 82 100 44º -- 36 35 18º 22º

129 F 60 unknown -- 53 L hybrid Nucleus EAS3 (Cochlear) 98 98 -- -- -- -- --* While Subject 79 wears a hybrid (S12) implant, he had no residual hearing in the implanted ear (and aided hearing contra-laterally). Thus he was classified as bimodal.º Missing one frequency from audiogram; Missing two frequencies from audiogram

-3

67

69

7071

S2. Complete Descriptive Analysis of Fixation Data.

In the main text, we presented limited descriptive analyses of the fixation with the goal largely of orienting the reader to the kind of data obtained in the VWP to help understand our more targeted analyses.

Figure S1 (which is reprinted from Figure 6 in the main text) shows the likelihood of looking at the target (correct item) and competitor as a function of time and CI group for the unambiguous step 1 in each continuum. During the first few hundred milliseconds, participants are equally likely to fixate the target and competitor (in all cases), but by about 500 msec., these begin to diverge. This 500 msec. does not take into account the 200 msec. it takes to plan and launch an eye-movement or the 100 msec after trial onset, so adjusting for these factors, this divergence can be pinpointed at about 200 msec. after stimulus onset. Both groups of listeners fixate the competitor more than the unrelated item, though it is clear that CI users tend to maintain

-4

A. NH: /b/

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 500 1000 1500 2000 2500

Time (msec)

Prop

ortio

n Fi

xatio

ns

/b//p/unrelated

B. CI: /b/

0

0.9

0 2500

/b//p/unrelated

C. NH: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

D. CI: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

A. NH: /b/

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 500 1000 1500 2000 2500

Time (msec)

Prop

ortio

n Fi

xatio

ns

/b//p/unrelated

A. NH: /b/

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 500 1000 1500 2000 2500

Time (msec)

Prop

ortio

n Fi

xatio

ns

/b//p/unrelated

B. CI: /b/

0

0.9

0 2500

/b//p/unrelated

B. CI: /b/

0

0.9

0 2500

/b//p/unrelated

C. NH: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

C. NH: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

D. CI: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

D. CI: /ʃ/

0

0.9

0 2500

/ʃ//s/unrelated

Figure S2: Timecourse of fixating the target, competitor and unrelated objects. A) For NH listeners after hearing a good /b/ (step 1). B) For all CI users after hearing a good /b/. C) For NH listeners after hearing a good /ʃ/ (step 1). D) For all CI users after hearing a good /ʃ/.

72

737475

7677787980818283

fixations on this item longer. This is consistent with Farris-Trimble, McMurray, Cigrand, and Tomblin (2014).

Figure S2 depicts the data as a function of rStep for the /b/ and /p/ sides of the continuum for both NH and CI listeners. These display fixations to the competitor (e.g., /p/ when the VOT indicated a /b/) as a function of time and the distance of the stimulus to the participant’s boundary, rStep (c.f., McMurray, Tanenhaus, & Aslin, 2002). Here for tokens near the boundary (e.g., the green curves in Figure S3A) participants make many fixations to the competitor, particularly early in the timecourse of processing, and these are reduced at more extreme rSteps. The pattern is roughly the same for the CI listeners (Figure S2B, D) although competitor fixations tend to peak at a higher level, and persist for more time.

-5

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0 500 1000 1500 2000 2500

Time (msec)

Com

petit

or F

ixat

ions -3

-2

-1

0

0.18

0 2500

-4

-3

-2

-1

rStep

0

0.18

0 2500

rStep

+1

+2

+3

+4

+5

rStep

0

0.18

0 2500

+1

+2

+3

+4

+5

rStep

A. NH, /b/

C. NH, /p/

B. CI, /b/

D. CI, /p/

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0 500 1000 1500 2000 2500

Time (msec)

Com

petit

or F

ixat

ions -3-3

-2-2

-1-1

0

0.18

0 2500

-4-4

-3-3

-2-2

-1-1

rStep

0

0.18

0 2500

rStep

+1

+2

+3

+4

+5

rStep+1+1

+2+2

+3+3

+4+4

+5+5

rStep

0

0.18

0 2500

+1

+2

+3

+4

+5

rStep+1+1

+2+2

+3+3

+4+4

+5+5

rStep

A. NH, /b/

C. NH, /p/

B. CI, /b/

D. CI, /p/

8485

8687888990919293

Figure S3 shows the same data for the fricative continua. Here the pattern is roughly the same, with a strongly gradient effect of rStep for both groups, and more competitor fixations overall for the CI users (Figure S3B,D) than the NH liste ners (Figure S3A,C).

One pattern that does emerge from these figures is that the /b/ and /ʃ/ sides of the continua appear to show (at least in these figures) a more gradient pattern of responding (relative to the rStep) than the /p/ and /s/ side of the continua. Bear in mind that these were the short sides of the continua (the boundaries were such that fewer tokens were perceived as /b/ and /ʃ/), so these responses were lower frequency. Thus, these differences may simply reflect a response to the fact that subjects were more uncertain about this lower frequency response.

-6

A. NH, /ʃ/

0

0.05

0.1

0.15

0.2

0.25

0 500 1000 1500 2000 2500

Time (msec)

Com

petit

or F

ixat

ions

-5-4-3-2-1

rStep

B. CI, /ʃ/

0

0.25

0 2500

-4-3-2-1

rStep

C. NH, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep

D. CI, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep

A. NH, /ʃ/

0

0.05

0.1

0.15

0.2

0.25

0 500 1000 1500 2000 2500

Time (msec)

Com

petit

or F

ixat

ions

-5-4-3-2-1

rStep

A. NH, /ʃ/

0

0.05

0.1

0.15

0.2

0.25

0 500 1000 1500 2000 2500

Time (msec)

Com

petit

or F

ixat

ions

-5-5-4-4-3-3-2-2-1-1

rStep

B. CI, /ʃ/

0

0.25

0 2500

-4-3-2-1

rStep

B. CI, /ʃ/

0

0.25

0 2500

-4-3-2-1

rStep

0

0.25

0 2500

-4-4-3-3-2-2-1-1

rStep

C. NH, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep

C. NH, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep+1+1+2+2+3+3+4+4+5+5+6+6

rStep

D. CI, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep

D. CI, /s/

0

0.25

0 2500

+1+2+3+4+5+6

rStep+1+1+2+2+3+3+4+4+5+5+6+6

rStep

Figure S3: Fixations to the competitor as a function of time and rStep (distance from the boundary). A) NH listeners for tokens on the /ʃ/ side of the continuum; B) CI users for tokens on /ʃ/ side of the continuum; C) NH listeners for tokens on /s/ side of the continuum; D) CI users for tokens on /s/ side of the continuum.

949596

979899

100101102

103

S3. Within CI group comparisons of identification data.

Identification: Voicing.

In the main text we reported analyses showing that for the identification of the voicing continua, CI users as a whole showed a shallower identification slope but that there were no differences as a function of acoustic+electric stimulation. Here we report exploratory analyses that we conducted to determine if there were further differences within each listener group (Unilateral vs. Bilateral and Bimodal vs. Hybrid). The first analysis used only individuals in the CIE group and compared unilateral and bilateral CI users. We used the model with maximal random effects, with a single contrast code for listener (bilateral = +.5, unilateral = -.5, centered). The effect of CI type was not significant (B=-0.19, SE=1.20, Z=-0.16, p=.88), nor did it interact with step (B=0.03, SE=0.66, Z=0.04, p=.97). The second analysis used only individuals in the CIA+E group. We again used the maximal model with a single contrast code for listener (hybrid = +.5, bimodal = -.5, centered). The main effect of listener was not significant (B=-0.002, SE=0.45, Z=-0.0, p=.998), nor was the Step CIHybrid vs. CIBimodal interaction (B=0.22, SE=0.45, Z=0.49, p=.62). Thus, there is no evidence of differences within our listener groups, mirroring the lack of difference between electric only and acoustic+electric CI users.

Identification: Fricative Place of Articulation

We also examined differences within the CI groups on fricative identification using similar models to those used above. The first analysis examined the electric-only group, comparing unilateral and bilateral listeners. In this model, the effect of listener was not significant (B=-0.14, SE=0.82, Z=-0.17, p=.87), nor was its interaction with step (B=0.43, SE=0.32, Z=1.33, p=.19).

The second analysis used only the A+E group comparing hybrid and bimodal users (+.5/-5, centered). Here, the effect of listener was significant (B=1.33, SE=0.47, Z=2.80, p=.005), as the hybrid group was more likely to choose /s/ in general (left shifted boundary). However, the interaction with step was not significant (B=0.34, SE=0.25, Z=1.36, p=.17).

While there were no significant differences within the A+E group, the smallish sample size (in each group) coupled with the highly unexpected overall differences in A+E CI users led us to a more descriptive analysis to see if this pattern was robust across the different types of A+E users, or if one or more type of CI configurations was driving the effect. Figure 5A compares the bimodal and hybrid users and suggests there may be small differences that would be detectable with more power. Here, the hybrid users still exhibiting a shallower slope than electric-only listeners, but may not show as large of a difference as the bimodal listeners. We were also worried that this effect may be driven in part by variation among the hybrid implants. These implants are generally shorter than standard implants and have fewer electrodes, potentially reducing sensitivity in the high frequency ranges that are necessary for fricative identification. While this cannot be the whole cause, since the effect was also observed in the bimodal listeners (who had a standard electrode), there is variation in these devices that was worth exploring. In particular, 7 of our hybrid users used variants of the Nucleus EAS system (10 mm / with 8 electrodes); 3 used the Nucleus L24 (16 mm / 22 electrodes of which 4 are not used); and 2 used the S12 (10 mm / 10 electrodes). All are typically mapped to the 700 – 7000 hz range. While there were too few participants in each device type to conduct a proper statistical analysis, Figure S5B shows the same data broken down by hybrid device type. It’s clear that both the EAS and

-7

104

105

106107108109110111112113114115116117118119

120

121122123124125126127128129130131132133134135136137138139140141142143144145146147

S12 (shorter electrode) devices show highly similar patterns of performance with reduced slopes (relative to electric only configurations). In contrast the L24 may show more “normal” performance, though with only three subjects we are hesitant to make any strong claims about device types. More broadly, however, this analysis suggests that two of the three hybrid configurations as well as the bimodal configuration show the shallow slope for fricative perception, suggesting this may be a product of electric+acoustic stimulation more generally, not an isolated effect of one device type.

Fixations: Voicing

In the analyses of fixations reported in the main text we found a main effect of rStep with more looks to the competitor as rStep approached the boundary for both the /b/ and /p/ sides of the continuum. We also found that CI users made significantly more competitor looks than NH listeners, although there was no overall difference between CIE and CIAE users. We did not find strong evidence of any interactions between rStep and CI status with one exception: CIAE users appeared to have a steeper effect of rStep but only on the /b/ side of the continuum.

In this section, we describe additional analyses within the electric-only and A+E groups to determine if there were any differences among electric-only or acoustic+electric configurations. These analyses took the same form as the within-group analyses of the identification data, examining only a single group of CI users (electric-only or acoustic+electric) with a single contrast code for the type of configuration.

For the /b/ analysis comparing A+E groups, the main effect CIHybrid vs. CIBimodal was not significant (B=-0.014, SE=0.015, t(16)= 1.0, p=.34), nor were the interactions with the linear or quadratic effect of rStep (linear: B<0.001, SE=0.013, t(20.6)<0.01 p=.98; quadratic: B=0.008,

-8

0

0.2

0.4

0.6

0.8

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

% /s

/ res

pons

e

CI: bimodalCI: hybrid

NH controlCI: electric only

0

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

CI: electric onlyCI: bimodal

Hybrid: EASHybrid: L24

Hybrid: S12

A. B.

0

0.2

0.4

0.6

0.8

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

% /s

/ res

pons

e

CI: bimodalCI: hybrid

NH controlCI: electric only

0

0.2

0.4

0.6

0.8

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

% /s

/ res

pons

e

CI: bimodalCI: hybrid

NH controlCI: electric only

0

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

CI: electric onlyCI: bimodal

Hybrid: EASHybrid: L24

Hybrid: S12

0

1

-6 -4 -2 0 2 4 6Fricative rStep (ʃ s)

CI: electric onlyCI: bimodal

Hybrid: EASHybrid: L24

Hybrid: S12

A. B.

Figure S5B: Proportion of /s/ responses on the fricative continua as a function of relative frication step. A) A comparison between the two acoustic-electric groups (with electric only and NH listeners for reference); B) A comparison of the three types of hybrids (with electric only and bimodal performance for reference.

148149150151152153154155156

157

158159160161162163

164165166167168

169170171

SE=0.013, t(13.3)=0.6, p=.58). Similarly, within the electric-only groups, the main effect of listener group was not significant (B=0.020, SE=0.018, t(12.4)=1.1 p=.29), nor was the interaction with rStep (linear: B=-0.001, SE=0.014, t(21.9)=-0.1 p=.94; quadratic: B=-0.030, SE=0.017, t(32)=1.8 p=.079). For the /p/ analysis comparing the A+E groups, the difference between hybrid and bimodal listeners was not significant (B=-0.010, SE=0.010, t(18.4)=1.0, p=.31), nor was its interaction with rStep (B<0.001, SE=0.004, t(17.8)< 0.1, p=.99). Within the electric-only groups, there was a significant difference between bilateral and unilateral (B=-0.033, SE=0.014, t(9.2)=2.4, p=.042) with slightly more fixations for bilateral users. However, this did not interact with rStep (B>-0.001, SE=0.004, T(51.8)<.1, p=.98). Thus, there was not strong evidence for group differences within the electric-only or A+E groups.

Fixations: Fricative Place

In the primary analysis of fixations as a function of fricative place, we found main effects of rStep and CI status. As with the analysis of VOT, listeners tended to look to competitors more when the rStep was close to the boundary and CI users made more competitor fixations than NH listeners. Unlike the VOT analysis, however, we also found that acoustic-electric users made even more fixations than electric-only users. Neither CI contrast interacted with rStep suggesting the CI users adapted to the implant by just heightening consideration of the competitor uniformly, without regard to the continuum step.

Once again, we conducted analyses within each group. For the /ʃ/ analysis comparing the A+E groups, there was no significant difference between bimodal and hybrid listener (B=0.005, SE=0.017, t(17.6)=0.3 p=.78), and this did not interact with rStep (B=0.005, SE=0.014, t(22.8)=0.4 p=.70). Within the electric-only groups, there was no difference between unilateral and bilateral CI users (B=0.009, SE=0.016, t(10.2)=0.6 p=.57) and this did not interact with rStep (B=0.009, SE=0.012, t(127)=0.7 p=.47). For the /s/ analysis within electric-only groups, we found no difference between unilateral and bilateral CI users (B=-0.019, SE=0.026, t(9.4)=-0.7, p=.49), and no interaction with rStep (linear: B=0.009, SE=0.005, t(28.6)=1.9, p=.074; quadratic: B=-0.003, SE=0.003, T=-0.9, df=15.2, p=.39). Within the A+E groups, we found a nearly significant difference between hybrid and bimodal CI users (B=-0.037, SE=0.019, t(20.2)=2.0, p=.059) with hybrids looking less to the competitor than bimodals. This did not interact with rStep (linear: B=-0.008, SE=0.006, t(23.2)=1.3, p=.19; quadratic B=0.001, SE=0.003, t(50.4)=0.2 p=.87). Thus, there was not strong evidence for group differences within the electric-only or A+E groups.

-9

172173174175176177178179180181

182

183184185186187188189190191192193194195196197198199200201202203204

205206

207

S4. Moderators of Performance among CI users

Having established the overall pattern of results, we next turned to the individual difference measures to see if any of these measures moderated the effect of the CI on performance. One goal of this was clinical. Are there any subsets of listeners who show steeper identification slopes (despite their degraded input)? The other goal was theoretical. Our analyses of the eye-movements showed clearly that CI users did not moderate their use of fine-grained detail--they simply increased activation for competitor words. Here we asked if there were any subset of users with altered sensitivity to fine-grained detail.

We restricted our analyses only to the CI users and looked at three broad factors: vocabulary skills, non-verbal IQ and speech perception. The former two factors should be seen as potentially mitigating–these abilities were present before the onset of deafness and cochlear implantation, and thus they may impact listeners’ abilities to take advantage of the CI. In contrast, the speech perception measures reflect how well the implant works for speech perception; it can be seen as an outcome variable, allowing us to determine what aspects of performance in these tasks are important for good speech perception.

For language and non-verbal IQ, we had four standardized assessments: the Peabody Picture Vocabulary Test and the Expressive Vocabulary Test were measures of vocabulary; the WASI Matrix Reasoning and Block Design tasks were used to assess non-verbal IQ. We expected these measures to be highly correlated with each other; thus we performed a principle component analysis to reduce them to a smaller number of orthogonal factors. Four participants (three hybrid, one unilateral electric) were missing scores for the EVT; one participant (a hybrid) was missing the Block Design, and two (both hybrid) were missing the matrix reasoning. These were replaced with mean values. We next converted all four scores to z-scores and ran a principle component analysis with varimax rotation. This discovered two factors that together accounted for 79.5% of the variance in the data set. The first captured composite language ability (43.5% of variance after rotation) with the following weights: PPVT: .942; EVT: .901; WASI block design: .098; WASI matrix reasoning: .183); the second captured composite non-verbal IQ (36.0% of the variance after rotation) with the following weights: PPVT: .067; EVT: .264; WASI block design: .833; WASI matrix reasoning: .819). For the speech perception measures, we averaged the CNC words correct and CNC phonemes correct scores into a single score. These scores were missing for three participants (one hybrid, one bimodal and one unilateral) and were replaced with the mean values.

While our primary concern was whether any of the experimental measures are influenced by these measures, prior to conducting these analyses, we also examined them to determine if there were any differences among abilities as a function of CI type (acoustic+electric vs. electric-only). No differences were found (see Table S2). We also observed no significant correlations between speech perception and either vocabulary (R=.169, p=.37) or non-verbal IQ (R=-.041, p=.83).

We did not have sufficient data to attempt a full analysis with all three moderators and their interactions (along with the experimental variables like VOT or

-10

Table S2: Comparison of groups on the three individual difference measures.

Mean (Z score)CIElectric

Language -0.4210IQ -0.1490

208

209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253

fricative step). Thus, for each analysis, we started with a slightly simplified version of the statistical models used above (we dropped the CI vs. NH contrast, since all of the participants were CI users; everything else was otherwise similar), and then added a single covariate along with its interactions with rStep and listener type (A+E vs. E). We then repeated this with the next moderator. In the interest of brevity we will not report the full results of each model, but rather focus on the most theoretically relevant findings.

Identification. Our first set of analyses examined the identification data for the voicing and/or fricative continua. As before, these predicted /p/ (or /s/) responding from step (1-8, centered), and CI type (acoustic-electric vs. electric, +/-.5, centered). While our prior analyses used the maximal random slope structure, when the covariates were added some of these models did not converge, thus, we used a simpler model with a random slope of step on subject and continuum. To this model, we added a single covariate (language, IQ or speech perception) along with its interactions with step, CI type and the three-way interaction. These models found no main effects of the covariates nor interactions between the covariates and step for both the b/p and s/ʃ continua.

Fixations. Our analysis of potential moderators of the fixation data took the same form. We again restricted our analysis to just the CI users. We started with the four primary analysis of the fixation data (on the /b/, /p/, /s/ and /ʃ/ sides of the continuum) that examined rStep and CI type as fixed effects. Each of these four models was repeated three times, adding to each model one of our three continuous covariates and their interactions with the primary factors. We maintained the same random effects structure as the prior models (random slopes of rStep on participant and random intercepts for continuum).

To summarize our results, these analyses found no effects or interactions with either non-verbal

-11

A. Voicing

0

0.02

0.04

0.06

0.08

0.1

0.12

-4 -3 -2 -1 0 1 2 3 4 5 6

Voicing rStep

Com

petit

or F

ixat

ions

Control

CI: High languageCI: Low language

A. Voicing

0

0.02

0.04

0.06

0.08

0.1

0.12

-4 -3 -2 -1 0 1 2 3 4 5 6

Voicing rStep

Com

petit

or F

ixat

ions

Control

CI: High languageCI: Low language

B. Fricatives

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

-4 -3 -2 -1 0 1 2 3 4 5 6

Fricative rStep

Com

petit

or F

ixat

ions

Control

CI: Low languageCI: High language

B. Fricatives

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

-4 -3 -2 -1 0 1 2 3 4 5 6

Fricative rStep

Com

petit

or F

ixat

ions

Control

CI: Low languageCI: High language

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

-4 -3 -2 -1 0 1 2 3 4 5 6

Fricative rStep

Com

petit

or F

ixat

ions

Control

CI: Low languageCI: High language

Figure S6: Fixations to the competitor as a function of rStep and language ability (median split). A) For the b/p continua; B) for the s/ʃ continuum.

254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299

IQ or with speech perception (CNC) scores. However, a consistent effect of language was observed (Figure S6); for the /p/ and /s/ analyses higher language scores were associated with fewer fixations to the competitor (/p/: B=-.010, SE=.004, t(28.9)=-2.2, p=.035; /s/: B=-.027, SE=.008, t(30.8)=-3.6, p=.001). This was not significant for the /b/ (B=-.001, SE=.006, t(28.8)=-0.2, p=.83) or /ʃ/ analyses (B=-.007, SE=.007, t(27.7)=-1.1, p=.29), although as Figure S6 suggests the effects are numerically in a similar direction. This finding is quite consistent with earlier findings that poor language (e.g., SLI) are associated with increased competitor fixations (McMurray, Munson, & Tomblin, in press; McMurray, Samelson, Lee, & Tomblin, 2010). There were no interactions of language with rStep or CI type in any of these analyses.

-12

300301302303304305306307308309

310

References

Farris-Trimble, A., McMurray, B., Cigrand, N., & Tomblin, J. B. (2014). The process of spoken word recognition in the face of signal degradation: Cochlear implant users and normal-hearing listeners. Journal of Experimental Psychology: Human Perception and Performance, 40(1), 308-327.

McMurray, B., Munson, C., & Tomblin, J. B. (in press). Individual differences in language ability are related to variation in word recognition, not speech perception: Evidence from eye-movements. Journal of Speech Language and Hearing Research.

McMurray, B., Samelson, V. S., Lee, S. H., & Tomblin, J. B. (2010). Individual differences in online spoken word recognition: Implications for SLI. Cognitive Psychology, 60(1), 1-39.

McMurray, B., Tanenhaus, M. K., & Aslin, R. N. (2002). Gradient effects of within-category phonetic variation on lexical access. Cognition, 86(2), B33-B42.

-13

311312313314315316317318319320321322323

324