Although we randomized the order in which the images were shown, we still tested for a potential order effect. A series of post hoc ANOVAs revealed that regardless of which image participants saw first, the first image shown was rated as more androgynous (M = 4.35, SD = 1.98) than subsequent images (M = 4.02, SD = 2.37), less anthropomorphic (M = 3.12, SD = 1.30) than the subsequent images (M = 3.59, SD = 1.88), and less homophily was felt towards the first image (M = 2.50, SD = 1.27) than towards subsequent images (M = 2.89, SD = 1.59). Finally, participants were less likely to choose the first image to represent themselves (M = 1.91, SD = 2.37) than subsequent images (M = 2.37, SD = 1.65). The same tests conducted after removing the first image shown did not indicate order effects. Therefore, unless otherwise noted, the remaining analyses were performed excluding the first image shown. These order effects results are discussed in more detail later.
Individual Differences and Avatar Image Type Influence on Image Perception
RQ2: What was the influence of participant’s biological sex and avatar image type on perceptions of androgyny?
An ANOVA revealed main effects for both participant’s gender, F (1, 1612) = 27.16, p < .001, and image type (human male, human female, animals, and objects), F (3, 1613) = 418.37, p < .001, on ratings of femininity. Male participants (M = 4.44, SD = .06) rated the avatars as more feminine than female participants did (M = 4.03, SD = .06). Post-hoc Scheffe tests indicated that all image types were different from each other (p < .01 for all comparisons). Human males (M = 2.29, SD = 1.49) were rated as least feminine and human females (M = 5.76, SD = 1.52) the most feminine. Animal avatars (M = 4.22, SD = 1.61) and objects (M = 4.68, SD = 1.72) were rated in between, with objects being perceived as more feminine than animal avatars. The effect size for participant gender was very small (ηp2= .02), however, especially when compared to the effect size for image type (ηp2= .44).
For masculinity, a main effect was found for image type, F (3, 1613) = 523.46, p < .001, ηp2= .49. Again, similar to femininity, human males (M = 6.02, SD = 1.42) were rated more masculine than human females (M = 2.34, SD = 1.45). Animals (M = 4.92, SD = 1.46) and objects (M = 4.78, SD = 1.73) were again in the middle of the scale. Although animals and objects were rated between human males and human females in both the masculinity and femininity scales, in both cases they were closer to the highest group than to the lowest. That is, they were closer to the human males in masculinity and closer to human females in the femininity scale.
The perceptions of androgyny also showed main effects for gender, F (1, 1608) = 13.59, p < .001, ηp2= .01, as well as image type, F (3, 1608) = 254.92, p < .001, ηp2= .32). Men (M = 4.50, SD = .07) rated the images as more androgynous than women (M = 4.13, SD = .07) but the effect size is very small. Image type, on the other hand, had a good-sized effect on androgyny. Post-hoc Scheffe tests indicated that human male (M = 2.80, SD = 1.97) and human female (M = 3.12, SD = 2.08) images did not significantly differ in their ratings of androgyny (p = .08). There were also no significant differences between animal (M = 5.85, SD = 1.80) and object images (M = 5.50, SD = 1.92) in their ratings of androgyny (p= .16). Nevertheless, human male and female images differed from animal and object images in their ratings of androgyny (p < .001 for all comparisons).
A similar pattern was found for androgyny-multiplicative. The ANOVA showed a small main effect for gender, F (1, 1608) = 23.18, p < .001, ηp2= .01 and a moderate effect for image type, F (3, 1608) = 156.07, p < .001, ηp2= .23. Again, men (M = 4.05, SD = .04) rated the images as more androgynous than women (M = 3.80, SD = .04) did. Post-hoc Scheffe tests indicated that there were no significant differences between the human male (M = 3.39, SD = .79) and human female (M = 3.35, SD = .75) image (p = .94), nor were there significant differences between the animal (M = 4.42, SD = 1.26) and object (M = 4.54, SD = 1.36) images (p = .52). In addition, human male and female images differed from animal and object images in their ratings of androgyny (p < .001 for all comparisons).
RQ3: What was the influence of participant’s biological sex and avatar image type on anthropomorphism?
An ANOVA with participant’s biological sex and avatar image type as factors revealed only a main effect for image type (human male, human female, animal, and object), F (3, 1623) = 617.24, p < .001, ηp2= .53) on ratings of anthropomorphism. Post-hoc Scheffe tests indicated that there were no significant differences between the human male and human female images in their anthropomorphism ratings (p = .05), but that these groups were significantly different from the animals and objects (p < .001 on all comparisons). Human males (M = 4.79, SD = 1.32) and human females (M = 4.56, SD = 1.43) were the most anthropomorphic groups, followed by animals (M = 2.39, SD = 1.40) and objects (M = 1.45, SD = .84).
RQ4: What is the influence of participant’s biological sex on the type of image they would choose to represent them?
Male participants overwhelmingly preferred choosing a human male avatar while women preferred the choice of a human female avatar. An ANOVA with participant’s gender and image type as factors indicated a strong interaction between these terms, F (3,1614) = 53.07, p < .001. See Figure 3 for means. Interestingly, an ANOVA with the same factors on attraction produced significant results only for image type, F (3,1619) = 83.96, p < .001, but not for gender or any interaction. Both male and female participants had the same attraction ratings for the avatar images and post-hoc tests indicated that females were the most attractive (M = 3.82, SD = .08), followed by men (M = 3.26, SD = .06), and then by nonhumans (M = 2.51, SD = .08) and objects (M=2.51, SD=.08). Nonhumans and objects did not differ. People were more likely to choose avatars that were human-like and of the same gender (males choosing male avatars and females choosing female avatars).
RQ5: What was the influence of participant’s computer usage and efficacy on the dependent variables?
Regression analysis on the dependent variables using computer usage factors, computer efficacy, age, and gender showed some significant effects. Computer usage math/science was a significant but very small predictor of androgyny (β=−.047, p < .01), androgyny-multiplicative (β=−.064, p = .02), anthropomorphism (β= .07, p < .01), attraction (β= .08, p < .01), homophily (β= .065, p = .02), and likelihood of choosing an image (β= .083, p < .01). “Author/researcher” computer usage was also a significant but very small predictor of homophily (β=−.073, p < .01) and likelihood of choosing an image (β=−.06, p = .02). The small effect sizes, however, make it questionable to conclude that there is any meaningful influence of computer usage or computer efficacy in the dependent variables.
Features of the Avatar Image that Influenced Perception
A series of linear regressions were run with effects coded values (detailed below) to examine what design features influence the perceptions of the images. The features analyzed, effects coding, and the images that fall into each set are the following: designed image male gender (−1, female; 0, undetermined, 1 male/male: m1–m5 and m1h–m5h; female: f1–f5 and f1h–f5h; undetermined: all other images), image is of an animal (0, not an animal; 1, animal/a1–a5), image is of an object (0, not an object; 1, object/o1-o5), image has head and torso or just a head (0, head only; 1, head and torso/m1–m5 and f1–f5; head only, m1h–m5h and f1h–f5h), image is of a child (0, not a child; 1, child/m4, m4h, f4, f4h), and the image rendering quality (0, low quality; 1, high quality/m1, m1h, f1, f1h). We also included the participant’s gender (−1, female; 1, male), computer usage, and computer efficacy to investigate how individual differences affected the perceptions of the images along with the images characteristics.
RQ6: What features of the avatar image influenced the perception of gender?
Looking first at the characteristics that predict the perception of the image gender, the analyses revealed that only two characteristics were significant predictors. The strongest predictor was the image designed gender (β= .81, p < .01). Participants could clearly identify the gender of the images as designed. Being a child character’s image negatively predicted perceived gender, however, (β=−.15, p < .01); that is, images of children were more prone to be rated as female or undetermined.
Moving now to the characteristics that predict the perceptions of masculinity and femininity, the regression coefficients show that image designed male gender was the stronger predictor of femininity (β=−0.63, p < .01) and masculinity (β= .67, p < .01). That is, male images (coded as 1) were perceived as less feminine and more masculine, and vice versa for female images (coded as −1), with undetermined in the middle. Another predictor of femininity (β= .23, p < .01) and masculinity (β=−.23, p < .01) was whether or not the image was that of a child character. The images of children were perceived as more feminine and less masculine. Lastly, being an object was also a predictor of femininity (β= .14, p < .01), but not of masculinity, so if the image portrayed an object, this avatar was perceived as more feminine.
An animal image was the strongest predictor of androgyny (β= .47, p < .01), followed by being an object (β= .40, p < .01). Interestingly, the quality of the image reduced the androgyny perception (β=−.13, p < .01). For androgyny-multiplicative, object images (β= .33, p < .01) and animal images (β= .37, p < .01) had an effect, but not the image quality. Participant gender, however, appeared as a negative predictor of androgyny-multiplicative (β= .13, p < .01). Male participants perceived the images as more androgynous-multiplicative.
RQ7: What features of the avatar image influenced the perceptions of anthropomorphism?
An ANOVA revealed a main effect for image type (human male, human female, animal, and object), F (3, 1623) = 617.24, p < .001, ηp2= .53, on ratings of anthropomorphism. Post-hoc Scheffe tests indicated that there were no significant differences between the human male and human female images in their anthropomorphism ratings (p = .05), but that these groups were significantly different from the animals and objects (p < .001 on all comparisons). Human males (M = 4.79, SD = 1.32) and human females (M = 4.56, SD = 1.43) were the most anthropomorphic groups, followed by animals (M = 2.39, SD = 1.40) and objects (M = 1.45, SD = .84).
Anthropomorphism behaved similarly to androgyny, and being an animal or object image was the only meaningful predictor. Being an object character was a very strong negative predictor of anthropomorphism (β=−.66, p < .01), while being an animal was also a strong negative predictor (β=−.47, p < .01), though less strong than object. These results indicate that images of objects are perceived as less anthropomorphic, followed by animal images, with human images being perceived as the most anthropomorphic.
RQ8: What features of the avatar image influenced the perceptions of attraction, credibility and homophily?
Attraction was negatively predicted by the image designed male gender (β=−.16, p < .01), whether or not the image was that of an animal (β=−.20, p < .01) or an object (β=−.19, p < .01) and positively predicted by whether or not the image was of a child character (β= .14, p < .01). These results indicate, for example, that the most attractive avatar would be one based on a human child image. However, although four features influenced attraction perceptions, it is important to note that this model only accounts for a small fraction of the variance in this variable (R2= .16).
Credibility was negatively predicted only by whether or not the image was an object (β=−.33, p < .01) or animal (β=−.28, p < .01). The effect sizes show that, compared to humans, animal images reduce the perception of credibility and object images reduce it even more. Homophily was also predicted only by whether the image was an object (β=−.26, p < .01) or an animal (β=−.26, p < .01). Similar to the results for attraction, the variance accounted for by the above regression models was small; the R2 for credibility was .13 and for homophily it was .11.
RQ9: What features influenced the choice of an avatar?
The features analyzed did not have much influence on whether participants would choose that image to represent them. As reported above, participants were most likely to choose an avatar that represented the same gender, and being an object image reduced slightly how much that image would be chosen (β=−.13, p < .01). We must note that the variance accounted for in the model is extremely small, however (R2= .03).
Although one must be cautious when interpreting null results, it is interesting to note that one of the features analyzed had no effect on any of the dependent variables. Unexpectedly, the presence or absence of torso (vs. floating head) had no effect on the participants’ perceptions. Even when the regression was run using only the human images, the presence or absence of torso had no significant effect on any of the variables. The effect sizes for head compared to head and torso in this case were β=−.03, p = .11 for femininity, β=−.01, p = .95 for masculinity, β= .00, p = .99 for perceived male gender, β= .03, p = .34 for androgyny, β=−.02, p = .49 for androgyny-multiplicative, β=−.04, p = .22 for anthropomorphism, β= .06, p = .07 for attraction, β= .00, p = .93 for credibility, β= .00, p = .90 for homophily, and β=−.01, p = .87 for likelihood of choosing the avatar.
Relationships Among the Perception Variables
The previous section examined how a variety of features influenced perceptions of anthropomorphism, androgyny, credibility, homophily and attraction. In this section, we ask how these variables relate to each other. To answer this question we computed Pearson correlations between the variables (shown in Table 3).
As expected, femininity, masculinity, and image gender were highly correlated with each other, as were the two measures of androgyny. Anthropomorphism was negatively correlated with androgyny (r =−.51, p < .01), androgyny-multiplicative (r =−.44, p < .01), and positively correlated with attraction (r = .45, p < .01), credibility (r = .46, p < .01), homophily (r = .45, p < .01), and the likelihood to choose the image (r = .31, p < .01). Attraction was negatively correlated with masculinity (r =−.31, p < .01), androgyny (r =−30, p < .01), and androgyny-multiplicative t (r =−.33, p < .01) but positively correlated with femininity (r = .10, p < .01) to a lesser degree. It was also positively correlated with credibility (r =−.49, p < .01), homophily (r = .49, p < .01), and the likelihood of choosing the image (r = .45, p < .01). Credibility was negatively correlated with masculinity (r =−.15, p < .01), androgyny (r =−.30, p < .01), and androgyny-multiplicative (r =−.31, p < .01). Credibility also correlated with anthropomorphism and attraction, as indicated before. Finally, credibility had a good correlation with homophily (r = .51, p < .01) and a moderately low correlation with the likelihood of choosing the image as an avatar (r = .37, p < .01). Homophily was highly correlated with likelihood of choosing the image (r = .3, p < .01). In fact, the image that received the highest homophily ratings also received the highest credibility ratings, was most likely to be chosen, and it was rated the second most attractive. Similarly, the image that received the lowest homophily ratings received the lowest credibility ratings, was least likely to be chosen, and also least attractive.
Overall, the gender-related variables were intercorrelated and the remaining variables formed another intercorrelated group. Anthropomorphism was unique in the sense that it correlated with all variables.