observation in autism spectrum conditions
around 30 years with similar appearance to the real human B The virtual robot agent was created by replacing the limb segments of the human agent with grey cylinders.
6.3.2 Interference Effect generated by biological motion
The mixed model 2 (Group: ASC, control) x 3 (Actor form: virtual human agent, virtual robot agent, real human) x 2 (Congruency: congruent, incongruent) ANOVA (excluding CV conditions) revealed a trend toward an interaction between Actor form, Group and Congruency (F(2,40) = 2.80, p = 0.07, ηp2 = 0.12). However, this interaction was not significant, there were no other main effects or
interactions (Figure 6.6). The interaction between Actor form, Group and Congruency reached significance when age and full-scale IQ were included as covariates (F(2,36) = 3.21, p = 0.05, ηp2 =
0.15). Since incongruent movements were expected to generate greater error plane variance than congruent movements, we conducted 1-tailed simple effects analyses with age and IQ as covariates. These demonstrated that the control group produced significantly more error plane variance when observing incongruent (adjusted mean(SEM) =415.22(96.16)) compared to congruent (350.03(77.57); F(1,18) = 3.27, p < 0.05) movements conducted by the virtual human agent, and more variance when observing incongruent (491.04(122.41)) compared to congruent (416.25(122.41)) real human
movements, although this trend did not reach significance (F(1,18) = 2.53, p = 0.06). In contrast, the
5 Variance scores significantly deviated from the normal distribution (Shapiro-Wilk W < 0.05). To ensure the effects
reported above were robust against violations of normality, data was log transformed and the 2x2x2x2 ANOVA rerun (as above with factors Group (ASC, control); Actor Form (virtual human agent, virtual robot agent), Actor Motion (BM, CV) and movement Congruency (congruent, incongruent) and age and IQ as covariates). This reanalysis also showed a significant interaction between Group x Actor Form x Congruency (F(1,18) = 4.92, p < 0.05). In addition there was a marginally significant Group x Actor form x Actor Motion interaction (F(1,18) = 4.38, p = 0.051). Differing from the above results, there was no significant Actor Motion x Group interaction (F(1,18) = 2.35, p = 0.14). Simple effects analyses showed that the Group x Actor form x Actor Motion interaction, which was not found in the above analysis, was driven by greater variability in movements when observing CV versus MJ movements conducted by the robot agent (adjusted log transformed mean BM (SEM) = 2.41 (0.11), CV = 2.47 (0.10); F(1,18) = 6.02, p < 0.05) but not human agent (BM = 2.47(0.10), CV = 2.47 (0.10); F(1,18) = 0.01, p = 0.93) for the ASC group but not for the control group (Robot agent BM (2.46 (0.10)) versus CV (2.43 (0.09) F(1,18) = 2.17, p = 0.16); Human agent BM (2.43(0.09)) versus CV (2.43(0.09); F(1,18) = 0.05, p = 0.82).
140 difference between the incongruent (411.73(105.84)) and congruent (456.90(123.54)) conditions for virtual robot agent movements was non-significant (F(1,18) = 0.68, p > 0.05).
For the ASC group the difference between incongruent and congruent conditions was non-significant for virtual human agent movements (incongruent = 331.13(105.35)), congruent = 358.36(84.99); F(1,18) = 0.47 p > 0.05) and virtual robot agent movements (incongruent = 339.13(115.95), congruent =
329.62(135.34); F(1,18) = 0.25, p > 0.05). Individuals with ASC demonstrated a trend in the opposite direction (variance for congruent movements greater than variance for incongruent movements) for real human observation (incongruent = 292.25(140.80), congruent = 355.01(134.11); F(1,18) = 1.48, p = 0.06)6.
To further investigate the Group x Actor form x Congruency interaction the Interference Effect was calculated for virtual human agent BM, virtual robot agent BM and real human. The Interference Effect figures were entered into a 3x2 ANOVA with factors actor form and group and with age and IQ as covariates. The ANOVA demonstrated a significant interaction between Group x Actor form (F(2,36) = 3.22, p = 0.05, ηp2 = 0.15). Simple effects analyses demonstrated that, compared to the control group,
the ASC group exhibited a reduced Interference Effect of real human actions and this difference was marginally significant (F(1.18) = 3.88, p = 0.065). No other simple effects analyses were significant (all p > 0.06).
There were no significant correlations between Interference Effect scores and ADOS total score or scores on the ADOS reciprocal social interaction or communication subscales (all p > 0.05).
6 To ensure these effects were robust against violations of normality data was log transformed and the 2x3x2
ANOVA analysis rerun. This reanalysis also showed a significant interaction between Group x Actor form (F(2,34) = 3.18, p = 0.05, ηp2 = 0.15).
141
Figure 6.6. Adjusted mean (+/-SEM) Interference Effect (incongruent minus congruent variance) is
displayed. The control group exhibit a significant Interference Effect when observing human agent biological motion (BM) movements and a marginally significant Effect for real human movements but no Interference Effect for robot agent BM movements. The ASC group did not exhibit a significant Interference Effect for human agent, robot agent or real human movements.
6.4 DISCUSSION
Individuals with ASC and control participants executed horizontal sinusoidal arm movements whilst observing congruent (horizontal) and incongruent (vertical) movements conducted by a 3D virtual reality agent with either human or robot form, which moved with either biological motion (BM) or at a constant velocity (CV). Participants also executed the same arm movements whilst observing a real human making congruent or incongruent arm movements. Finger-tip position was recorded and average variability in the error (vertical) plane was the dependent variable. Results from control participants replicated the previously reported effect of actor form (Kilner et al., 2003): control participants exhibited a significant Interference Effect for observation of the virtual human agent and the real human but not for the virtual robot agent. In contrast, individuals with ASC did not exhibit a significant Interference Effect for either the virtual human agent, real human or virtual robot agent. There was no effect of actor motion (BM vs CV) on the Interference Effect for either group. This is the first demonstration that, whereas control adults exhibit a greater Interference Effect in response to human form compared to robot form actors, individuals with ASC do not show this Interference Effect.
142 6.4.1 Interference effect in healthy controls
Previous Interference Effect studies have not investigated separable effects of form and motion using 3D stimuli. To enable the manipulation of actor form (human versus robot) and actor motion (BM versus CV) whilst keeping all other factors constant, the current study used 3D virtual agents. As expected, we found that control participants exhibit a greater Interference Effect in response to human form compared to robot form actors. The lack of a difference in the Interference Effect between BM and CV conditions for the control group was unexpected as it goes against previous findings of greater Interference Effects for BM compared to CV movements (Kilner et al., 2007a). Furthermore, this finding contrasts with the results reported in Chapter 3 which showed that movements with BM but not CV kinematics are automatically motorically simulated. However, as Kilner and colleagues (2007a) suggest, this effect may depend on experience and prior expectations of how a stimulus should move. Although Kilner and colleagues (2007a) found a significant difference in the magnitude of the Interference Effect mediated by a human video stimulus that moved with MJ BM compared to CV, they found no difference between MJ and CV ball stimuli. The authors argue that, whereas participants were unlikely to have previously seen videos of humans moving with CV, through exposure to computer animations, participants may be equally familiar with CV and MJ ball movements. It might therefore be equally possible to simulate the movement of the MJ and CV ball and both may create an Interference Effect. This explanation is of relevance to the current study. Both Chapter 3 and the study by Kilner and colleagues (2007a) reported differences between BM and CV movements using video stimuli. In contrast the current study employed computer-animated virtual agents. Our participants may have had previous sensorimotor experience with (for example) computer game virtual agents moving with biological motion or CV, such previously acquired sensorimotor experience could result in an Interference Effect for CV movements conducted by virtual agents.