Keywords

1 Introduction

The social cues conveyed by animated pedagogical agents (APAs) can have an impact on learners in multimedia learning environments (e.g., Baylor and Kim 2004). In most cultures, gender is the primary social categorization (DeFrancisco and Palczewski 2007), and is strongly involved in the organization of interactions (Ridgeway and Smith-Lovin 1999). Gender stereotypes appear at an early age (e.g., Arthur et al. 2008). It is well known that people regard computers and other media as though they were real people (i.e., media equation theory; Reeves and Nass 1996). Gender stereotypes may also exist in the virtual world. For example, when computers provide gender cues, users produce stereotypical responses based on gender (Nass et al. 1997). Gender stereotypes lead to expectations about men and women. Men are regarded as agentic (i.e., independent, masterful and competent), and women as communal (i.e., friendly, unselfish and concerned with others; Eagly and Wood 1991). According to Schunk (1987), a model’s perceived competence can enhance observational learning. Interactions between learners and APAs can be affected by gender stereotypes. As a consequence, male APAs could benefit from gender stereotypes and therefore be more effective.

Some reports, for instance, indicate that male agents are more effective for learners than female ones, as they are judged more positively (Kim et al. 2007; Experiment 1 & 2), arouse greater interest in the task (Kim et al. 2007; Experiment 1) allow for better self-regulation (Baylor and Kim, 2004; Experiment 1) and give rise to enhanced self-efficacy (Baylor and Kim 2004; Experiment 1) or learning outcomes (Kim et al. 2007; Experiment 1; Moreno et al. 2002).

Others, by contrast, suggest that female agents are more beneficial to learners, contributing to their self-efficacy (Baylor and Kim 2004; Experiment 2). Shiban et al. (2015) found a positive effect on learners’ interest in the subject (statistics) for the female, young and attractive agent, compared to the male, older and less attractive agent. Schroeder and Adesope (2015) didn’t find any effect of agent gender.

It is, however, important to emphasize that most of these studies did not take the learners’ sex into account, even though some research in social psychology on the similarity attraction hypothesis (SAH) suggests that it is important to consider gender similarity between learners and APAs.

According to the similarity-attraction hypothesis (SAH), people tend to be more attracted to others who are similar to themselves (Byrne 1971). Same-sex preferences are observed from an early age (e.g., Martin et al. 1999).

In a computer-based learning context, learners could therefore be more attracted to same-sex APAs. Gender similarity has a positive impact on judgments, with girls and boys considering their own gender more favorably (e.g., Yee and Brown 1994). Moreover, according to Bandura (1997) similar attributes, such as gender, between social model and learner can affect self-efficacy which, in turn, has an impact on persistence and achievement (Schunk 1995). In the light of this knowledge, we would therefore expect gender similarity between APAs and learners to give rise to better outcomes. However, studies have yielded heterogeneous results. For example, some studies have highlighted an effect of agent gender independently of the learner’s sex. Baylor and Kim (2003) for instance, demonstrated a main effect of APA gender, with a male agent being judged more extrovert and agreeable in an instructional plan development task. Participants also expressed greater satisfaction with their performances and reported that the male agent facilitated self-regulation more than the female agent did. Furthermore, Arroyo et al. (2009; Experiment 2) observed that girls work better with the male agent. In another study, male and female agents exposed gender-fair beliefs, and provided a twenty-minute narrative about four female engineers, followed by five benefits of engineering careers (Ashby Plant et al. 2009). They observed a positive effect of a female versus male agent on the career interest dimension and mathematics performance.

Arroyo et al. (2009; Experiment 1) meanwhile, found that students tended to perform better in mathematics with a learning companion of the opposite gender, thus contradicting the SAH, although participants did not express a preference for this agent in the similarity condition. As for Behrend and Thompson (2011), they failed to demonstrate any effect of learner–APA gender similarity on either learning, engagement or perceptions of utility – be it of the agent or of the office software training program. That said, the SAH was corroborated in studies, in which Ozogul et al. (2013; Experiment 1) observed that 12-year-old girls rated a program on electrical circuits more positively after working with a gender-matched agent. In another study, Arroyo et al. (2013) found that female characters had a positive effect on girls. Male students had more negative outcomes when a learning companion was present; they obtained their worst outcomes with female characters. Research on the gender of agents and participants has therefore yielded mixed results so far, with no clear endorsement of the SAH. Students’ judgments, motivation and even learning outcomes are not necessarily more positive when the APA’s gender matches their own.

Furthermore, a task’s gender orientation may influence interactions (Ridgeway and Smith-Lovin 1999). For example, children evaluate a computer-synthesized speech more positively and show better self-confidence when voice gender matches either content gender (i.e., female topics: skin care and makeup, princesses, dance; male topics: dinosaurs, football, knights) (Lee et al. 2007). Moreover Lee (2003) showed that participants exhibited a greater conformity with male character about typical male subject (i.e., sport) and with female character about typical female subject (i.e., fashion).

It should be noted that up to now, APA gender studies have usually featured tasks about science, technology, engineering or mathematics (STEM; e.g., Ozogul et al. 2013; Shiban et al. 2015; van der Meij et al. 2015) (often in order to develop girls’ self efficacy in STEM). This may have interfered with the similarity-attraction relationship and could explain equivocal outcomes, as women are underrepresented in the STEM field (Beede et al. 2011), and people associate science and mathematics with men (e.g., Nosek et al. 2009).

The aim of our study was to examine the effect of gender similarity on learners performing a nonstereotypically gendered task. Based on SAH, we predicted that learners would judge the same-sex APA more positively (Hypothesis 1), exhibit greater engagement (Hypothesis 2), and perform better (Hypothesis 3) with a same-gender APA than with an opposite-gender one.

2 Method

2.1 Pretest

We conducted a pretest in order to select a neutral learning task. Participants included 62 children (32 girls, 30 boys) with a mean age of 9.52 years (SD = .31). Consistent with previous studies (Becky 2000; Miller and Budd 1999). We measured children’s preferences for and perceived competence in different academic disciplines according to gender.

This pretest revealed that a foreign language task could be regarded as nonstereotypical, in contrast to mathematics or participants’ native language.

2.2 Test

Participants and Design.

The sample comprised 47 children (24 girls, 23 boys) attending schools in western France. Participants’ mean age was 10.53 years (SD = .27). All of them had French as their mother tongue. They were randomly assigned to a same-gender (n = 25) or opposite-gender APA (n = 22) (see Fig. 1). (Perception of gender of APAs has previously been pre-tested with students).

Fig. 1.
figure 1

Boy and girl APAs in the learning environment

Full size image

Material and Procedure.

The children had to learn new items of English vocabulary presented by the APA on a 12.1” touchscreen tablet. After the learning phase, they could practice in an engagement phase, before undergoing a learning test. Children’s perception of the APA was assessed by a post-task printed questionnaire. The session lasted 40 min.

Measures.

We evaluated learning, engagement, and APAs’ perception.

Learning: children had to recall some of the words learned in the lesson. They had to recall eight English words cued by the French ones (e.g., How do you say journal in English?). To receive a point, the child must remember the word and its pronunciation must be correct. Scores ranged from 0 to 8 points.

Engagement: we measured the number of exercises completed in a revision task (e.g., Behrend and Thompson, 2012). Between the learning phase and learning test, children could practice saying the new words. They could see and listen again twice to each of the 10 words featured in the lesson. Each time, the participant chose to listen the same word again, they received a point. They could end this phase whenever they liked. This allowed us to score their engagement out of 20.

APAs perception: We examined two dimensions of the APAs: their legitimacy and their attractiveness.

Legitimacy: the three items for the 5-point legitimacy scale were adapted from Bavishi et al. (2010); inspired by Choi and Mai-Dalton (1999). Cronbach’s alpha reliability coefficient was .94 for this scale.

Attractiveness: comprising four items, the 5-point scale for measuring attractiveness was adapted from two existing scales (i.e., Khan and De Angeli 2009; McCroskey and McCain 1972; Pratt, Hauser, Ugray and Patterson 2007). Alpha reliability was .94 for this scale.

3 Results

Regarding participants’ perceptions of the APAs, we observed effects of gender similarity on both legitimacy, F(1, 45) = 9.030, p = . 004, ƞ2 = .167, and attractiveness, F(1, 45) = 6.027, p = .018, ƞ2 = .118. Participants who had been assigned to a same-gender APA rated the latter as more legitimate (same-gender APA: M = 3.83, SD = 1.07; opposite-gender APA: M = 2.77, SD = 1.3) and also more attractive (same-gender APA: M = 3.86, SD = .70; opposite-gender APA: M = 3.32, SD = .82). As Levene’s homoscedasticity test showed that there was significant heterogeneity of variance (p > .10) for engagement, we ran a Mann–Whitney test. Engagement was greater among learners who had worked with an opposite-gender APA (M = 6.73, SD = 6.2) than with a same-gender APA (M = 3, SD = 3.46), (U = 158, p = .012).

We failed to find any difference in learning, F(1, 45) = .130, p = .72, between learners who had been assigned a same- versus opposite-gender APA.

4 Discussion

Consistent with SAH, and with our first hypothesis, learners judged the same-gender APA more positively than the opposite-gender APA.

Concerning engagement, results failed to confirm our second hypothesis, as learners were more engaged with the opposite-gender APA. There are two possible explanations for this result. First, our participants may not have sought to compete with the opposite-gender APA, and were thus more strongly engaged. Second, they may have felt more competent with the same-gender APA, and therefore saw less of a need to persevere in order to succeed. Regarding performance on the learning task, results also failed to confirm our third hypothesis.

Our study shows that results can differ and even be contradictory, depending on which measures are used. This might explain the mixed results of previous studies.

Further research with children of various ages is needed to explore these results.

Moreover, the gender orientation of a particular task may strongly influence the user–system relationship, making it necessary to compare several different types of task. It is therefore crucial to conduct further experiments to understand this effect.