To answer the question if a robot-initiated touch induces more positive evaluations of the robot, there are several sub-scales and questionnaires that say something about how the participant evaluated the robot after watching the movie.
Attitude toward the robot
To investigate the participants’ attitude toward the robot I carried out a two-way MANOVA on the differences between the post- and pre-measures on the three subscales of the NARS questionnaire as dependent variables. A positive score means an increase in the negative attitude towards the robot, the higher the score the stronger the increase. A negative score means a decrease, vice versa. There was a main-effect for the bond condition, Wilks Λ = 0.83, F (3,58) = 3.96, p = .012, partial η2 = .170, so the
difference scores on the NARS differ by the bond condition. For the main-effect bond there is a significant difference for two subscales of the NARS. Subscale Interaction with Robots in the bond condition decreased (M = -0.07, SD = 0.5) and for the no bond condition increased (M = 0.19, SD = 0.48), this difference is significant (F (1,60) = 4.41, p = .04, partial η2 = .068). Also for the subscale Emotional Interactions with Robots in the bond condition decreased (M = -0.32, SD =0.57) and for the
no bond condition increased (M = 0.08, SD = 0.46), this difference is significant (F (1,60) = 9.88, p = .003, partial η2 = .141). The two subscales (Interaction with Robots and Emotional Interaction with
Robots) of the NARS differ significantly between the bond and no bond condition. In the bond condition, the post-score is lower (participant is more positive attitude toward the robot) than the pre- score and in the no-bond condition the post-score is higher (a negative attitude toward the robot) than the pre-score (see figure 4 and 5). There was no main-effect for the touch condition, Wilks Λ = 0.90, F (3,58) = 2.23, p = .094, partial η2 = .104, and no interaction effect, Wilks Λ = 0.99, F (3,58) = 0.19, p = .906 partial η2 = .009.
Figure 4.
Figure 5.
The mean score of the two subscales Emotional Interaction with the Robot of the NARS (pre- and post- measures).
Physical appearance
How the participants evaluated the physical appearance of the robot was measured by several scales. I carried out a two-way MANOVA on the scores of the Immediacy and Credibility scales as dependent variables and the bond and touch conditions as independent variables. There were no main effects for bond (Wilks Λ = 0.97, F (2,59) = 0.84, p = .437, partial η2 = .028) and touch (Wilks Λ = 0.99, F (2,59)
= 0.46, p = .635, partial η2 = .015), also no interaction effect was found (Wilks Λ = 1.00, F (2,59) = 0.09, p = .918 partial η2 = .003.
I carried out a mixed design MANOVA. The reason that I did a mixed design MANOVA is that I already expect a difference in these measures in advance, which makes it difficult to interpret. I used the pre- and post-measures of the Perceived Human Likeness Scale, Anthropomorphism Scale and the Likeability Scale as dependent variables and the bond and touch condition as independent variables. There was no main effect for the bond condition (Wilks Λ = 0.95, F (3,57) = 1.04, p =.380, partial η2 = .052), no main effect for the touch condition (Wilks Λ = 0.94, F (3,57) = 1.12, p = .348, partial η2 =
.056) and no interaction effect (Wilks Λ = 0.98, F (3,57) = 0.46, p =.711, partial η2 = .024). But within subjects there was a significant difference in the aspect measuring moment (Wilks Λ = 0.79, F (3,57) =
5.14, p = .003, partial η2 = 0.213). Participants score significantly lower post-scores (M = 2.21, SD = 0.74) than the pre-scores (M = 2.43, SD = 0.72) for Anthropomorphism regardless of the experimental condition, this was significant (Greenhouse-Geisser = 1.63, F (1,59) = 8.74, p < .01, partial η2 = .129).
Hence, the participants found the robot less human at the end of the experiment (see figure 6). Also, there was a significant interaction effect between the measuring moment and the bond condition (Wilks Λ = 0.85, F (3,56) = 3.48, p =.022, partial η2 = .155). For the subscale Likeability, in the bond condition decreased (pre-measure: M = 3.78, SD = 0.67 to post-measure: M = 3.64, SD = 0.84) and for the no bond condition increased (pre-measure: M = 3.56, SD = 0.68 to post-measure: M = 3.75, SD = 0.66), this is significant, Greenhouse-Geisser = 0.87, (F (1,59) = 5.54, p <.01, partial η2 = .086 (see figure 7). A
higher score means the participant thinks the robot is more human-like. Participants who bonded with the robot, thought the robot was less human-like at the end of the experiment. And participants who did not bond with the robot, thought the robot was more human-like at the end.
Figure 6.
Pre- and post-measures (pre-, after interaction and after movie) of the Anthropomorphism scale for the four conditions.
Figure 7.
Pre- and post-measurements (pre-, after interaction and after movie) of the Likeability scale for the four conditions.
Trust in the robot
To investigate trust in the robot, measured by the Affective Trust Scale and the Perceived Trust Scale, I carried out a two-way MANOVA on the scores of these scales. The results gave no significant main effects (bond: Wilks Λ = 0.956, F (2,59) = 1.352, p = 0.267, partial η2 = .044 and touch: Wilks Λ =
0.964, F (2,59) = 1.107, p = 0.337, partial η2 = .036) nor interaction effects (Wilks Λ = 0.982, F (2,59) = 0.551, p = 0.579, partial η2 = .018) on the subscales. Participants did not experience more trust in the
robot after watching the movie.
Relation with the robot
To investigate how bonded the participants felt toward the robot after watching the movie I carried out a two-way ANOVA with the Perceived Friendship Scale. The bond and touch condition were the independent variables and the score of the Perceived Friendship Scale was the dependent variable. There were no significant results, main effect bond: F (1,60) = 0.835, p = .365 partial η2 = .014, main effect
touch: F (1,60) = 1.46, p = .231 partial η2 = .024, and no interaction effect: F (1,60) = 1.65, p = .203
I also carried out a mixed design MANOVA for the subscales IOS and Attachment with bond and touch condition as independent variables and the pre- and post-measurements of the IOS and Attachment scale as dependent variables. A higher score means the participant felt more connected toward the robot. There was no main effect for the bond condition (Wilks Λ = 0.99, F (2,56) = 0.26, p =.771, partial η2 = .009), no main effect for the touch condition (Wilks Λ = 0.97, F (2,56) = 0.76, p = .473, partial η2 =
.026) and no interaction effect (Wilks Λ = 0.99, F (2,56) = 0.16, p =.852, partial η2 = .006). Within subjects there was a significant interaction effect between the measuring moment and the bond condition (Wilks Λ = 0.61, F (2,56) = 17.69, p <.001, partial η2 = .387). For both scales, The Attachment Scale in
the bond condition decreased (pre-measure: M = 3.53, SD = 0.89 to post-measure: M = 3.37, SD = 0.92) and for the no bond condition increased (pre-measure: M = 3.24, SD = 0.78 to post-measure: M = 3.59, SD =0.91), which is significant, Greenhouse-Geisser = 2.59, (F (1,57) = 22.5, p < .001, partial η2 = .283
(see figure 8). For the IOS, in the bond condition decreased (pre-measure: M = 3.10, SD =1.63 to post- measure: M = 2.61, SD = 1.5) and for the no bond condition increased (pre-measure: M = 2.34, SD = 1.45 to post-measure: M = 2.97, SD = 1.74), which is significant, Greenhouse-Geisser = 10.48, (F (1,57) = 27.1, p < .001, partial η2 = .322 (see figure 9). Participants, who bonded with the robot before felt less
bonded after the movie part and participants who did not bond with the robot before felt more bonded with the robot after the movie part.
Figure 8.
Pre- and post-measurements (after interaction and after movie) of the Attachment scale for the four conditions.
Figure 9.
Pre- and post-measurements (after interaction and after movie) of the IOS scale for the four conditions. Robot touch
To investigate how the participants in the touch condition experienced the touch of the robot, I carried out an independent t-test. I used the bond condition as independent variable and the subscale-touch (average score consisting of questions about how the participant experienced the touch of the robot) as dependent variable. The results were not affected when the covariate Touch Receptivity was included. The participants in the bond condition (n = 17, M= 4.11, SD = 1.48) scored lower than the control condition (n = 15, M = 5.06, SD = 0.88), which was significant, t = -2.4; df = 25.44; p = .014 (1-sided), d = -0.78. Participants in the bond condition found the touch from the robot less pleasant than the participants in the no bond condition (see figure 10).
Figure 10.
Experience robot-touch for the bond and no bond condition.
Overall conclusion, there were no interaction effects found between bond and touch. In the bond condition the participant had a more positive attitude toward the robot and the no-bond condition had a more negative attitude toward the robot at the end of the experiment. But after the movie part the participant was less positive toward the robot. Participants who bonded with the robot, thought the robot was less human-like at the end of the experiment. And participants who did not bond with the robot, thought the robot was more human-like. Even, the no bond condition is more liking the robot after the movie, and the bond condition was less liking the robot. Also, participants, who bonded with the robot felt less bonded after a touch of the robot. And participants who did not bond with the robot felt more bonded with the robot after being touched. Finally, participants in the bond condition found the touch from the robot less pleasant than the participants in the no bond condition. These findings do not support H2.