Estimations of biological motion walking direction are affected by observer and walker genders Authors: Qi Sun, Liang-Liang Dong, Fan-Huan You, Dorita H. F. Chang Published: 2024 ABSTRACT A facing-towards bias is commonly reported when observers are asked to judge the motion directions of others. However, it remains unclear just how accurately observers are able to estimate the motion direction of others and whether the gender of the observer and the walker affects the direction estimation. Here, we asked male and female participants to estimate the direction of a point-light walker (PLW) in three experiments. The gender of PLWs was neutral (Experiment 1, 96 participants), clearly male or female (Experiment 2, 72 participants), or more subtlety male or female (Experiment 3, 98 participants). We found that female PLWs showed a stronger reference repulsion bias (RRB) than male PLWs. That is, for female PLWs, the estimates of facing directions were biased away from the boundaries of facing-towards, facing-away, and lateral (left/right)-motion directions. Interestingly, RRBs differed depending on whether the observer was male or female. When the PLW gender difference was clear, the RRB was stronger for female participants than male participants; when the PLW gender difference was reduced, the trend disappeared or was reversed. Finally, the perceived PLW direction was biased towards the previously seen PLW direction, showing serial dependence that was not affected by the PLW and observer genders. In conclusion, the current study shows that observers can accurately estimate PLW directions, but that judgments are curiously affected by both the observers’ and PLWs’ genders. Introduction When navigating the world, it is crucial for observers to accurately perceive both their own self-motion direction and the walking direction of others to avoid head-on collisions. The former issue has been extensively investigated and discussed [1–7], and there is also a comprehensive understanding of the behavioral and neural mechanisms underlying self-motion direction perception [8–13]. In contrast, little work has been conducted to examine whether and how observers accurately perceive the direction of a single walker [14]. Johansson (1973) first found that a few light points attached to the head and major joints of a walker (PLW) could produce a vivid moving form of a human figure (Fig. 1a) [15]. Following this study, extensive studies have shown that observers can extract multidimensional attributes from a PLW, such as gender [16–19], age [20], emotional state [21], social intention [22, 23], and speed [24, 25]. Moreover, we also have a deep understanding of the cortical and subcortical basis of processing different information of the PLW [26–28]. Fig. 1Stimulus and estimation bias illustrations. (a) Each point-light walker (PLW) consisted of 16 white dots. The white lines were invisible in the experiment. The simulated PLW direction was randomly selected from the range of [-90°, 270°), illustrated by the dashed-line circle. Forward (Backward) means that the PLW is moving toward (away from) the observer. Leftward (Rightward) means that the PLW is moving laterally leftward (rightward) that is perpendicular to the observer’s facing direction. (b) illustration of estimation bias. The estimation bias was fit as a sinusoidal function of the actual PLW direction (Eq. 1). The solid line illustrates a sinusoidal curve with a positive amplitude parameter; in contrast, the dashed line illustrates a sinusoidal curve with a negative amplitude parameter Stimulus and estimation bias illustrations. (a) Each point-light walker (PLW) consisted of 16 white dots. The white lines were invisible in the experiment. The simulated PLW direction was randomly selected from the range of [-90°, 270°), illustrated by the dashed-line circle. Forward (Backward) means that the PLW is moving toward (away from) the observer. Leftward (Rightward) means that the PLW is moving laterally leftward (rightward) that is perpendicular to the observer’s facing direction. (b) illustration of estimation bias. The estimation bias was fit as a sinusoidal function of the actual PLW direction (Eq. 1). The solid line illustrates a sinusoidal curve with a positive amplitude parameter; in contrast, the dashed line illustrates a sinusoidal curve with a negative amplitude parameter Despite extensive research, little work has been conducted to reveal just how accurately observers can perceive different motion directions of the PLW. In most studies, participants are presented with a series of PLWs and asked to report whether the PLW is facing them using a two-alternative forced choice paradigm (2-AFC). These studies consistently show a facing-towards bias, such that when given an ambiguous-facing walker, such as one presented in frontal view, observers will tend to report the walker to be facing towards rather than away from them [18, 19, 29–32]. To our knowledge, Sweeny et al. (2012) was the first study to report an estimation bias of observers when judging the directions of PLWs. Participants were shown PLWs with walking directions randomly selected from the range of -63° and 63°. Negative values indicated leftward movement and positive values indicated rightward movement. Their results showed that the perceived motion direction of a single PLW was systematically biased away from the category boundary of leftward/rightward motion in a phenomenon referred to as the reference repulsion bias (see also [33]). They argued that the reference repulsion bias was due to observers’ high sensitivity to the straight-approaching direction, which is a hallmark of categorical perception [34]. When the presentation time of the PLW was reduced, there was a corresponding reduction in the reference repulsion bias. In other work, Yang et al. (2014) found that the discrimination of the PLW motion direction (left vs. right) was more difficult when the PLW was depicted as male (right panel in Fig. 4) than when it was depicted as female (left panel in Fig. 4) [35]. Furthermore, male participants exhibited greater sensitivity than female participants when discriminating the female PLW’s motion direction. These results suggest that both the gender of the PLW and of the observer can affect the discrimination of PLW motion directions. However, other work [14] has shown that observers can estimate PLW motion directions while seemingly ignoring the genders of the PLWs (and of themselves). Therefore, the roles of both the observer gender and the gender of the PLW in direction estimation remain to be clarified. Sweeny et al. (2012) proposed that one might expect to observe a stronger reference repulsion bias in the direction estimation of female PLWs as compared to male PLWs and that male observers may show a stronger reference repulsion bias than female observers [14]. The PLW, of course, contains social attributes (e.g., emotion and intention) and biological attributes (weight and strength), aside from the physical features (e.g., speed and directions). According to [36], statistical analysis indicates that males exhibit more physical aggression than females. Therefore, it might be expected that observers are more sensitive to detecting male PLWs than female PLWs and be more likely to misperceive male PLWs as approaching them in order to avoid danger in a timely manner. This may explain why the facing-towards bias is stronger for male PLWs than for female PLWs [17, 18, 29, 32]. It follows that we might also expect the perceived direction of male PLWs to be more biased towards the reference directions (0°), showing a weaker reference repulsive bias for male PLWs than for female PLWs. Additionally, males are generally stronger and more powerful than females, based on statistics [37]. Compared to male observers, female observers exhibit a stronger facing-towards bias, allowing them to avoid danger more effectively. Consequently, female observers may show a weaker reference-repulsion bias than male observers. Aside from examining the overall estimation bias in PLW direction estimation [14, 33], recent studies have also reported on the spatial-temporal continuity of physical feature perception [38–41]. Fischer and Whitney (2014) showed that the perceived orientations of current trials were biased towards a previously presented orientation, indicating a serial dependence of perception [42]. This serial dependence was robust as long as the feature difference between walkers in successively presented trials was within a certain range and the interval between two trials was less than ~ 20 s [40]. Sun, Gong, et al. (2023) also found a serial dependence in the PLW direction perception [33]. However, it is unclear whether and how the genders of the PLW and the observer might interact with serial dependence in the PLW direction perception. Here, we examined the effects of the gender of the PLW and the observer on estimation bias and serial dependence in the perception of a walker’s motion direction. Experiment 1 investigated how the observer’s gender affected the estimation bias and serial dependence by showing participants PLWs with the neutral gender and extending the PLW direction range (i.e., from − 63° to 63°) in [14] to the entire circle domain (i.e., from − 180° to 180°). In Experiment 2, we presented PLWs that were female or male, and PLWs tested how this affected the motion discrimination of observers of two genders (male/female). Troje (2002b) found that the shoulder of the male PLW tends to be wider than that of the female PLW, and that the male PLW tends to occupy a greater spatial extent than the female PLW. In addition, the shoulder of the male PLW tends to be wider than the hip, but this is not the case for the female PLW. Troje (2002b) also revealed that the male PLW displayed considerable lateral body sway than the female PLW [15]. These features are all closely related with the PLW’s weight and moving speed. To further examine the findings of Experiments 1 and 2, Experiment 3 further blurred the gender of PLWs by changing their appearances. Specifically, we reduced the degree of femininity of the female PLWs by increasing their weights and moving speeds, and correspondingly, we reduced the degree of masculinity of the male PLWs by reducing their weight and moving speeds. Our results show that the PLW and observer genders interactively affect the estimation bias of the PLW direction but do not affect the serial dependence. Experiment 1: the observer’s gender affects PLW direction estimation Methods Participants The current experiment included two conditions. In one condition, all PLWs move towards the observers – the approaching condition; in another condition, all PLWs move away from the observers – the walking-away condition. The two conditions were completed by different groups of participants (Group 1 assigned to the approaching condition: 20 males, 28 females; Group 2 assigned to the walking-away condition: 21 males, 27 females). All were naïve to the purpose of the experiment and had a normal or corrected-to-normal vision. We obtained written informed consent from all participants before the commencement of the experiment. All procedures were approved by the Institutional Review Board at Zhejiang Normal University. Visual stimuli and apparatus In the current experiment, participants were presented with a series of PLWs (Fig. 1a) presented on the center of the monitor. The PLWs were generated from BML kit (https://www.biomotionlab.ca/), a toolkit developed by [15, 43]. Each walker was about 3.07° V × 8.58° H and consisted of 16 white dots that indicated the walker’s head, clavicles, bell button, left and right angles, knees, hips, hands, elbows, and shoulder joints. We changed several parameters of the online PLW generator, others were kept default. Specifically, the parameter Camera.Perspective was changed into Enabled; Walker.Speed was set to 1; Camera.Azimuth that controlled the walking direction was randomly selected from the range of [-90°, 90°] with a step of 9° (Approaching condition) or the range of [90°, 270°] with a step of 9° (Walking-away condition). 0° indicated that the PLW was facing the observer and moving towards the observer; in contrast, 180° indicated that the PLW was facing away from the observer and moving away from the observer. The experiment was programmed in MATLAB using the Psychophysics Toolbox 3. The monitor was ASUS ROG PG279Q (resolution: 2560 H× 1440 V pixels; refresh rate: 60 Hz), and the graphics card was NIVIDA P620. Experimental design and procedure The experiment was conducted in a room with no light and participants’ heads supported by a chinrest. The viewing distance was 56 cm. Participants’ viewing direction (i.e., straight-ahead direction) was aligned with the display center. On each trial, a PLW was presented at the display center for 1 s, followed by the response display in which a white circle was presented with a purple probe. Participants were asked to move the mouse to change the probe’s position on the circle to indicate their perceived PLW direction. Note that, before conducting the experiment, participants were told that the bottom, top, left, and right mid-points of the response circle corresponded to the 0°, 180°, -90° (270°), and 90° PLW directions, respectively, and they could freely position the probe on the circle. After the participants’ response, the next trial started immediately, meaning that there was no inter-trial interval. In the current experiment, two conditions were included. In one condition, all PLW directions are moving towards the participants (i.e., [-90°, 90°], the approaching condition; the white semicircle in Fig. 1a); in another condition, all PLW directions are moving away from participants (i.e., [90°, 270°], the walking-away condition; the blue semicircle in Fig. 1a). Note that, the range of [90°, 270°] corresponds to the union of two ranges: [90°, 180°] and [-90°, 180°). To facilitate the following statistical analysis, we projected these two ranges to the range of [90°, 270°] (Fig. 1b). Each condition corresponded to one session of trials. Each session consisted of 420 trials (21 orientations × 20 trials). All trials were randomly presented and divided into 4 blocks. Before conducting the experiment, each participant was given 10 practice trials randomly selected from the above 420 trials to familiarize the participants with the experiment. The whole experiment lasted for about 30 min. Estimation bias analysis We first examined whether the perceived PLW directions were systematically biased. To do so, we calculated the estimation bias of each PLW direction corresponding to the difference between the perceived PLW direction and the actual PLW direction. Then, for each PLW direction in each condition (approaching and walking-away condition), the mean estimation bias was averaged across the 48 participants. Given that the two conditions were finished by two groups of different participants, to capture the changing trend of estimation biases against the PLW direction, we grouped the mean estimation biases of the two groups of participants together and fitted the mean estimation bias (EB) as a sinusoid function of the actual PLW direction (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\theta\:}_{A}$$\end{document}), given as: 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:EB={a}_{EB}\text{sin}\left(2*{\theta\:}_{A}\right)$$\end{document} in which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} indicates the amplitude of estimation bias. Figure 1b illustrates the curves of Eq. 1 with different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}.\:$$\end{document}This pattern is similar to the reference repulsion bias revealed in [14]. In contrast, a negative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} (dashed line in Fig. 1b) showed the opposite trend. Specifically, when the actual PLW directions were in the range of (-90°, 90°), the perceived PLW directions were systematically biased toward the 0° motion direction; whereas when the actual PLW directions were in the range of (90°, 270°), the perceived PLW directions were systematically biased toward the 180° motion direction. This kind of bias was named as a reference toward bias. Again, the two conditions were completed by two groups of participants. If we fit Eq. 1 to each participant’s data and then evaluate the significance of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}, the findings would be confusing or redundant. Hence, to clarify the data and determine the statistical significance of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} we used a bootstrapping approach. Specifically, we sampled the participants’ data with replacement for 100,000 iterations. For each iteration, we calculated the mean estimation error and fitted the data with Eq. 1. The 100,000 iterations generated a distribution of sampled \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}, which gave the 95% confidence interval (CI) of the sampled \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}. If the lower bound of the 95% CI was larger than 0, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} was significantly larger than 0, showing a lateral-motion bias. In contrast, if the upper bound of the 95% CI was smaller than 0, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} was significantly smaller than 0, showing a central tendency. Next, we repeated the above procedures to examine the size of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} for the female and male participants, respectively. If the 95% CIs of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} for the female and male groups were not overlapped, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}s were significantly different between the two groups. Serial dependence analysis Like the previous studies (e.g [42, 45])., we analyzed the serial dependence on the group level. Specifically, we first calculated the relative PLW direction (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\theta\:}_{R}$$\end{document}) between the previous 1st (n-1st) trial and the current trial (nth, n = 2, 3, etc.) and fitted the estimation error (EE) of the current trial as a first-derivative of Gaussian function (DoG) of the relative PLW direction, given as: 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:EE=\sqrt{2}{a}_{SD}w{\theta\:}_{R}{e}^{-{\left(w{\theta\:}_{R}\right)}^{2}+0.5}$$\end{document} in which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} is the amplitude of the DoG curve. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} was positive, then the perceived PLW direction of the current trial was biased toward the previous 1st trial, showing an attractive serial dependence; if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} was negative, then the perceived PLW direction of the current trial was biased away from the previous 1st trial, showing a repulsive serial dependence. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:w$$\end{document} is the width of the DoG curve. To determine the statistical significance of the DoG curve, a bootstrapping procedure was used. Like the above analysis, we bootstrapped the data in 100,000 iterations. In each iteration, we sampled the participants’ data with replacement and fitted the sampled data with Eq. 2, generating a distribution of sampled \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}. We could calculate the 95% CI of the sampled \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}. If the lower bound of the 95% CI was larger than 0, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} would be significantly larger than 0; vice versa. Like the analysis of the estimation bias, the above procedures were repeated for the serial dependences of male and female participants. If the 95% CIs of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} for the female and male groups were not overlapped, then the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were significantly different between the two groups. Results Estimation bias Figure 2 plots the estimation bias against the actual PLW direction. The data were well captured by the sinusoidal function (Eq. 1) (R2 = 0.56). Importantly, the amplitude parameter was positive (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}=2.68, 95% CI: [2.19, 3.17]), suggesting a reference repulsion bias was in the estimation of the PLW direction. Additionally, the amplitude of female participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}=2.04, 95% CI: [1.34, 2.75]) was significantly smaller than that of male participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}=3.53, 95% CI: [2.91, 4.15]), suggesting that the observer gender affects the reference repulsion bias and male observers show stronger reference repulsion bias than female participants. Fig. 2Results of estimation bias. Estimation bias is against the PLW direction. (a) illustrates the mean estimation bias averaged across all participants. (b) and (c) illustrates the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers represent the results of the approaching and walking-away conditions. The red line is the best-fitting result of a sinusoid function, showing a reference repulsion bias Results of estimation bias. Estimation bias is against the PLW direction. (a) illustrates the mean estimation bias averaged across all participants. (b) and (c) illustrates the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers represent the results of the approaching and walking-away conditions. The red line is the best-fitting result of a sinusoid function, showing a reference repulsion bias Serial dependence Figure 3 plots the estimation error against the relative PLW direction. The DoG function well captured the data (solid lines in Fig. 3, R2s > 0.13). In particular, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were all significantly larger than 0 in the approaching condition (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s > 3.08, 95% CI of each condition was shown in the left panels of Fig. 3) and walking-away condition (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s > 3.69, 95% CI of each condition was shown in the right panels of Fig. 3), showing attractive serial dependences in the perception of PLW directions. Additionally, the 95% CIs of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were well overlapped between the approaching and walking-away conditions, and between the female and male groups, suggesting that the sizes of serial dependence were not significantly different between the approaching and walking-away conditions, and the observer gender did not affect the serial dependence. Fig. 3Results of serial dependence. The estimation error is against the relative PLW direction which was the PLW direction difference between the previous 1st trial and the current trial. (a), (b) and (c) show the serial dependence for all participants, female and male participants, respectively. The left and right panels correspond to the approaching and walking-away conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Results of serial dependence. The estimation error is against the relative PLW direction which was the PLW direction difference between the previous 1st trial and the current trial. (a), (b) and (c) show the serial dependence for all participants, female and male participants, respectively. The left and right panels correspond to the approaching and walking-away conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Summary In summary, the current experiment found that the estimates of PLW directions were systematically biased away from the reference PLW directions (e.g., 0°, 90°, 180°, and 270°), showing a reference-repulsion effect [14]. Especially, the male observers showed a stronger reference-repulsion effect than female observer when showing neutral PLW stimuli. Moreover, we found that the estimates of PLW directions were systematically biased towards the previously seen PLW directions, showing a serial dependence effect [33]. Experiment 2: both the PLW’s gender and the observer’s gender interactively affect PLW direction estimation Experiment 1 revealed a reference repulsion bias and serial dependence in the PLW direction estimation. Notably, the reference repulsion bias was modulated by the observer’s gender. However, the PLW gender was neutral in Experiment 1 (the Walker. Gender parameter is 0 in the BML kit), leaving it unclear whether the gender of the PLW might also affect PLW direction estimation. Troje (2002b) [44] found that, as compared with the female PLW, the shoulder of the male PLW tends to be wider, and that the male PLW tends to occupy a greater spatial extent. In addition, the shoulder of the male PLW tends to be wider than the hip, but this is not true for the female PLW. Accordingly, in the BML kit (https://www.biomotionlab.ca/) developed by Troje group [43, 44], the parameter Walker. Gender can be set to different values to change the appearance of the PLW to align with different genders. Here in Experiment 2, we replaced the neutral PLW with female (Walker.Gender = -6) and male PLWs (Walker.Gender = 6) (left and right panels in Fig. 4a). Fig. 4Stimulus illustration and results of Experiment 2. (a) Experiment 2 stimuli were either female (left) or male (right). The middle panel shows an illustration of a gender-neutral PLW used in Experiment 1. (b) Gender evaluation results. Each diamond maker indicates the mean gender value averaged across 12 participants. Error bars indicate the standard error across 12 participants. (c) Results of estimation biases. Left and right panels illustrate the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers correspond to the male and female PLW conditions. The solid curves are the best-fitting results of a sinusoid function. (d) Results of serial dependence. Left and right panels show the serial dependence for female and male participants, respectively. Black and blue markers correspond to the male and female PLW conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Stimulus illustration and results of Experiment 2. (a) Experiment 2 stimuli were either female (left) or male (right). The middle panel shows an illustration of a gender-neutral PLW used in Experiment 1. (b) Gender evaluation results. Each diamond maker indicates the mean gender value averaged across 12 participants. Error bars indicate the standard error across 12 participants. (c) Results of estimation biases. Left and right panels illustrate the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers correspond to the male and female PLW conditions. The solid curves are the best-fitting results of a sinusoid function. (d) Results of serial dependence. Left and right panels show the serial dependence for female and male participants, respectively. Black and blue markers correspond to the male and female PLW conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Methods Participants Two groups of participants were tested in this experiment. The first Group included 24 participants (12 females, 12 males; 18–25 yrds). Their task was to estimate the gender of the PLWs. The second group had 48 participants (24 females, 24 females; 18–25 yrds). They were instead asked to estimate the direction of the same PLWs. All were naïve to the purpose of the experiment and had a normal or corrected-to-normal vision. We obtained written informed consent from all participants before the commencement of the experiment. All procedures were approved by the Institutional Review Board at Zhejiang Normal University. Visual stimuli, apparatus, and procedure The stimuli were again generated from the BML kit (https://www.biomotionlab.ca/) developed by Troje group [43, 44], as for Experiment 1. The stimulus and apparatus parameters were similar to those of Experiment 1, with the exception of the parameter Walker.gender: Walker.gender was set to -6 to generate the female PLW (left panel in Fig. 4a), and it was set to 6 to generate the male PLW (right panel in Fig. 4a). All other parameters were the same as in Experiment 1. The first group of participants were asked to evaluate the gender of the PLW. Specifically, on each trial, a PLW was presented at the display center for 800 ms. Then, participants were asked to report the gender of PLW. If the PLW was male, they pressed the number 1-key; if the PLW was female, they pressed the number 2-key. The second group of participants were asked to estimate the PLW direction. The trial procedure was similar to Experiment 1 except that the PLW was presented for 800 ms here. Additionally, the PLW directions were randomly selected from the range of [-90°, 270°] with steps of 9°. Each direction was repeated 30 times, leading to 1200 trials in total. Data analysis The data analysis methods were similar to that of Experiment 1. We first analyzed observers’ ability to classify PLW gender. Specifically, we calculated the proportions (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:p$$\end{document}) of pressing the 2-key for the male and female PLWs, and then used the proportions to minus 0.5. Finally, the differences were divided by 0.5. This can be given by: 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:PLW=\frac{p-0.5}{0.5}$$\end{document} If PLW is -1, then the participants accurately recognized the male PLW; if PLW is 1, then the participants accurately recognized the female PLW. Additionally, for the estimation bias of the male PLW, we found that the trends were different in the range of [-90°, 90°] and [90°, 270°] (Fig. 4c) when analyzing the data, so we fitted the data in the two ranges with Eq. 1, respectively. Results PLW evaluation Figure 4b plots the PLW gender classification separately for female and male participants. One sample t test showed that regardless of the female or male participants, one sample t test showed that the gender degree of female PLWs was significantly larger than 0 (ts(11) > 9.53, ps < 0.001, Cohen’s ds > 2.75), while the gender degree of male PLWs was significantly smaller than 0 (ts(11) < -12.23, ps < 0.001, Cohen’s ds > 3.52). A 2 (observer gender: male vs. female) × 2 (PLW gender: male vs. female) ANOVA showed that only the main effect of the PLW gender was significant (Greenhouse-Geisser corrected: F(1.00, 22.00) = 481.61, p < 0.001, partial η2 = 0.96), suggesting that participants can clearly distinguish male and female PLWs. That is, when a male PLW was present, participants tended to take it as male; and vice versa. Importantly, there is no significant difference in distinguishing the two types of stimuli between male and female participants. Estimation bias Left and right panels in Fig. 4c show the results of the direction estimates for female and male participants, respectively. The data show that regardless of whether the participant was male or female, there is a significant reference repulsion bias when judging the direction of female PLWs (blue markers). However, the bias size of female participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} = 9.39, 95% CI: [8.67, 10.11]) was significantly larger than that of male participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} = 7.37, 95% CI: [6.63, 8.11]). This trend is particularly different from that in Experiment 1 in which the PLW gender was neutral. For these PLWs, we found that the bias size of male participants was larger than that of female participants. Similar trend was also observed for the male PLW (black markers in Fig. 4c) when the PLW direction was in the range of [90°, 270°] (female participants: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} = 5.87, 95% CI: [4.94, 6.80]; male participants: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} = 3.55, 95% CI: [2.64, 4.46]). Specifically, a reference repulsion bias was observed, and the bias was significantly larger for female participants than male participants. Additionally, for the male PLW (black markers in Fig. 4c), when the PLW direction was in the range of [-90°, 90°], the 95% CI of the bias for female participants was [-1.98, 0.39] with a mean − 0.79 that includes 0. 0 is within the range of [-1.98, 0.39], so the bias was not significantly different from 0. This indicated the disappearance of bias. However, the 95% CI of the bias for male participants was [-2.60, -0.36] with a mean of -1.48. 0 was not in the range of [-2.60, -0.36], so the bias was significantly smaller than 0. This suggested that that the estimates of PLW directions were systematically compressed towards the straight-approaching direction (0°), a reversed trend to the reference repulsion bias. Experiment 2 found that the reference repulsion bias was stronger for the female participants than for male participants when viewing non gender-neutral PLWs. To clarify this trend, we collapsed the data of female and male PLWs and fitted the estimation bias with Eq. 1 for the female and male participants, respectively. Figure 5 plots the bias size (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}) against the different experiments, clearly showing an inverse trend. Fig. 5Comparison of reference repulsive bias between Experiments 1 and 2. Blue and black markers indicate the results of female and male participants. Error bars are the 95% CI of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}.$$\end{document} Here, we named the neural PLW as low gender difference in Experiment 1 but clear male and female PLWs as high gender difference in Experiment 2 Comparison of reference repulsive bias between Experiments 1 and 2. Blue and black markers indicate the results of female and male participants. Error bars are the 95% CI of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}.$$\end{document} Here, we named the neural PLW as low gender difference in Experiment 1 but clear male and female PLWs as high gender difference in Experiment 2 Moreover, regardless of whether the participant was male or female, the reference repulsion bias was stronger for the female PLW than the male PLW (Fig. 4c). Serial dependence Figure 4d plots the estimation error against the relative PLW direction. The DoG function well captured the data (solid lines in Fig. 4d, R2s > 0.54). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were all significantly larger than 0 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s > 3.29, 95% CI of each condition was shown in Fig. 4d), showing robust attractive serial dependences in the perception of PLW directions. Importantly, the 95% CIs of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were well overlapped between female and male participants, and between female and male PLWs, suggesting that the sizes of serial dependence were not affected by the PLW gender and participant gender. Summary In summary, the current experiment replicated the reference repulsion bias in the range of [-90°, 270°] for the female PLWs and in the range of [90°, 270°] for male PLWs, and serial dependence observed in Experiment 1. Interestingly, however, in Experiment 2, we found that both the PLW’s gender and the participant’s gender may interactively affect the reference repulsion bias. Specifically, when the female and male PLWs were clearly differentiated (Fig. 4b), the reference repulsion bias may be stronger for female participants than for male participants, and the reference repulsion bias may be stronger for female PLWs than for male PLWs. Finally, neither the PLW gender nor the participant gender may affect the serial dependence for estimating PLW direction. Experiment 3: effects of the PLW gender are further confirmed The results of Experiments 1 and 2 indicate that neither the gender of the PLW nor the gender of the observer affected serial dependence. However, there was an interactive effect between the PLW and observer genders on the reference repulsion bias (Fig. 5). Specifically, when the PLW gender was neutral (Experiment 1), the reference repulsion bias was stronger for male participants than for female participants. In contrast, when the PLW genders were clearly differentiated (Experiment 2), the trend was reversed. To examine these findings further, we conducted Experiment 3, in which we altered the appearances of male and female PLWs to blur their gender differences. Specifically, we reduced the degree of femininity of the female PLWs and correspondingly reduced the degree of masculinity of male PLWs. Based on the findings of the earlier experiments, we expected that the difference in reference repulsion bias between female and male participants would disappear under these new manipulations. Methods Participants Two groups of participants were tested in this experiment. The first Group included 24 participants (12 females, 12 males; 19–27 yrds). They were invited to evaluate the gender of the PLWs. The second group had 74 participants (40 females, 34 females; 18–24 yrds)1. They were invited to estimate the PLW direction. All were naïve to the purpose of the experiment and had a normal or corrected-to-normal vision. We obtained written informed consent from all participants before the commencement of the experiment. All procedures were approved by the Institutional Review Board at Zhejiang Normal University. Visual stimuli, apparatus, and procedure The stimuli were again generated from the BML kit (https://www.biomotionlab.ca/) developed by the Troje group [43, 44]. As mentioned in Experiment 2, Troje (2002b) found that the shoulder of the male PLW tends to be wider than that of the female PLW, and that the male PLW tends to occupy a greater spatial extent than the female PLW. In addition, the shoulder of the male PLW tends to be wider than the hip, but this is not the case for the female PLW. Troje (2002b) also revealed that the male PLW displayed considerable lateral body sway than the female PLW [44]. Accordingly, we can blur these differences between the male and female PLWs by increasing the weight of the female PLW and decreasing the weight of the male PLW, and by increasing the speed of the female PLW and decreasing the speed of the male PLW could blur these differences. Therefore, in the BML kit, (1) for the female PLW (left panel in Fig. 6a), the Walker. Speed was set to 1.209 and the Walker. Weight was set to 6; (2) for the male PLW (right panel in Fig. 6a), the Walker. Speed was set to 1 and the Walker. Weight was set to -1.7. Note that, we invited another group of participants to evaluate the effects of these manipulations on the PLW genders (see Fig. 6b for the results). The other parameters for the stimulus, apparatus, and procedures are the same as that of Experiment 2. Please read the corresponding parts in Experiments 1 and 2 for more details. Fig. 6Stimulus illustration and results of Experiment 3. (a) Experiment 2 shows female (left) and male (right) PLWs to participants. The middle panel shows an illustration of a neutral PLW used in Experiment 1. (b) Gender evaluation results. Each diamond maker indicates the mean gender value averaged across 12 participants. Error bars indicate the standard error across 12 participants. The dark and light markers correspond to the results of Experiments 3 and 2, respectively. (c) Results of estimation biases. Left and right panels illustrate the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers correspond to the male and female PLW conditions. The solid curves are the best-fitting results of a sinusoid function. (d) Results of serial dependence. Left and right panels show the serial dependence for female and male participants, respectively. Black and blue markers correspond to the male and female PLW conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Stimulus illustration and results of Experiment 3. (a) Experiment 2 shows female (left) and male (right) PLWs to participants. The middle panel shows an illustration of a neutral PLW used in Experiment 1. (b) Gender evaluation results. Each diamond maker indicates the mean gender value averaged across 12 participants. Error bars indicate the standard error across 12 participants. The dark and light markers correspond to the results of Experiments 3 and 2, respectively. (c) Results of estimation biases. Left and right panels illustrate the mean estimation bias averaged across all female and male participants, respectively. Shaded areas represent the standard error across all participants. Black and blue markers correspond to the male and female PLW conditions. The solid curves are the best-fitting results of a sinusoid function. (d) Results of serial dependence. Left and right panels show the serial dependence for female and male participants, respectively. Black and blue markers correspond to the male and female PLW conditions. Dots indicate the mean estimation error averaged across all participants; shaded areas indicate the standard error across all participants Data analysis The data analysis methods were the same as in Experiment 2. Results PLW evaluation Figure 6b plots the PLW evaluation against the participants with different genders. The light and dark markers correspond to the results of Experiments 2 and 3, respectively. It clearly shows that regardless of female and male participants, one sample t test showed that the genders of PLWs can be clearly distinguished (female PLW: ts(11) > 3.83, ps < 0.0028, Cohen’s ds > 1.11; male PLW: ts(11) < -6.35, ps < 0.001, Cohen’s ds > 1.83). A 2 (observer gender: male vs. female) × 2 (PLW gender: male vs. female) ANOVA showed that only the main effect of the PLW gender was significant (Greenhouse-Geisser corrected: F(1.00, 22.00) = 129.98, p < 0.001, partial η2 = 0.86), suggesting that participants can clearly distinguish male and female PLWs, and there is no significant difference in distinguishing the two types of stimuli between male and female participants. Additionally, the independent-sample t test showed that compared with the PLW evaluation of Experiment 2, the feminine degree of female PLWs was significantly reduced (ts (22) > 2.80, ps < 0.010, Cohen’s ds > 1.14). The masculine degree of male PLWs was also decreased for the male participants (male participants: t (22) = -2.43, p = 0.024, Cohen’s d = 0.99) but not for the female participants (female participants: t (22) = -0.97, p = 0.34, Cohen’s d = 0.40). Overall, these results suggested that the gender difference between the female and male PLW was effectively reduced compared to Experiment 2. Estimation bias Figure 6c shows the results of reference repulsion bias for the female (left panel) and male (right panel) participants. Black (blue) makers and lines show the results of male (female) PLWs. It clearly shows that regardless of male and female participants, there is a significant reference repulsion bias for the female PLW (blue markers). However, the bias size of female participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 5.50, 95% CI: [4.93. 6.07]) was not significantly different from that of male participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 5.12, 95% CI: [4.49, 5.75]). A similar trend was also observed for the male PLW (black makers in Fig. 6c) when the PLW direction was in the range of [90°, 270°] (Female participants: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 5.48, 95% CI: [4.40, 5.99]; male participants: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 5.00, 95% CI: [3.58, 5.64]). When the male PLW direction was in the range of [-90°, 90°], the bias of female participants disappeared (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 2.21, 95% CI: [-0.14, 2.39]), but a significant reference repulsion bias was found in the male participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document} = 1.10, 95% CI: [0.68, 2.93]). As shown in the results of the PLW evaluation, the masculine degree of male PLWs for the female participants was not significantly different between Experiments 2 and 3, so the disappearance of the reference repulsion bias was consistent with Experiment 2. In contrast, the masculine degree of male PLWs for the male participants was significantly reduced in Experiment 2 than in Experiment 3 (Fig. 6b), which indicates that the PLWs become more neutral (Experiment 1) or more feminine (Experiment 2). According to the patterns of the reference repulsion bias for the female PLW revealed in Experiments 1 and 2, it makes sense that the reference repulsion bias was significant for male participants when the masculine degree of male PLWs was decreased, suggesting that the reference repulsion bias for the male PLW was increased compared to Experiment 2. Therefore, Experiment 3 found that the difference in the reference repulsion bias between the female and male participants was reduced as the PLW gender was blurred. To further clarify this trend, we collapsed the data of female and male PLWs and fitted the estimation bias with Eq. 1 for the female and male participants, respectively. Figure 7 plots the bias size (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}) against the different experiments, clearly showing that the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document} of Experiment 3 was between the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}s of Experiment 1 and 2, consistent with our expectations. Fig. 7Comparison of reference repulsive bias between Experiments 1, 2, and 3. Blue and black markers indicate the results of female and male participants. Error bars are the 95% CI of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}. Here, we named the neural PLW as low gender difference in Experiment 1 but clear male and female PLWs as high gender difference in Experiment 2. Because the gender difference was blurred in Experiment 3, so it was named as the middle gender difference Comparison of reference repulsive bias between Experiments 1, 2, and 3. Blue and black markers indicate the results of female and male participants. Error bars are the 95% CI of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{EB}$$\end{document}. Here, we named the neural PLW as low gender difference in Experiment 1 but clear male and female PLWs as high gender difference in Experiment 2. Because the gender difference was blurred in Experiment 3, so it was named as the middle gender difference Moreover, consistent with Experiment 2, the results also showed that the reference repulsion bias was stronger for the female PLW than the male PLW regardless of male and female participants (Fig. 6c). Serial dependence Figure 6d plots the estimation error against the relative PLW direction. The DoG function well captured the data trends (solid lines in Fig. 6d, R2s > 0.67). The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were all significantly larger than 0 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s > 3.11, 95% CI of each condition was shown in Fig. 4d), showing robust attractive serial dependences in the perception of PLW directions. Importantly, the 95% CIs of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{a}_{SD}$$\end{document}s were well overlapped between female and male participants, and between female and male PLWs, further suggesting that the sizes of serial dependence were not affected by the PLW genders and participant genders. General discussion The current study systematically examined the estimation biases of point-light walker (PLW) directions through three experiments. We found that the estimates of PLW directions were repelled away from the reference directions, indicating a reference repulsion bias [14, 33]. Additionally, the estimates of PLW directions were biased towards the previously seen directions, showing serial dependence [33]. We showed that, curiously, the PLW and observer genders interactively affected the reference repulsion bias but did not affect the serial dependence. We review these findings in turn. Firstly, our study reproduced the reference repulsion bias found by Sweeny et al. (2012) [14]. Sweeny and his colleagues selected PLW directions from the range of [-63°, 63°], where the straight-approaching direction (0°) was proposed to be the category hallmark of left and right directions. Due to the heightened sensitivity to the category hallmark, observers bias their direction estimates away from it, resulting in the reference repulsion bias. This has been demonstrated in various features, such as orientation [46], motion direction [47], and heading direction [2]. For example, observers are most sensitive to the vertical and horizontal orientations (0°, 90°, -90°, 180°) [48], and orientation estimates are systematically biased away from these directions [46]. Our PLW direction estimates showed the same pattern (Figs. 2 and 4, and 6). Therefore, the category hallmarks in the PLW estimation included the straight-approaching (0°), facing-away (180°), and lateral motion directions (± 90°). Sweeny et al. (2012) only reported on biases along the straight-approaching direction [14]. Meanwhile, we also showed that the bias patterns were different between the approaching directions (i.e., (-90°, 90°)) and the walking-away directions (i.e., ((90°, 270°)), especially for the male PLWs (Figs. 4c and 6c and c). Generally, the results showed that the estimation accuracies of the approaching directions were higher than those of the walking-away directions, which may because that the approaching directions are more closely related to observers’ safety. Approaching directions can lead to a collision if we cannot accurately estimate and avoid timely. Another interesting finding is that the reference repulsion bias was lower for male PLWs than female PLWs. On one hand, the results indicate that the directions in the range of [-90°, 90°] were more biased towards the facing-toward direction (0°) for male PLWs than female PLWs, which is consistent with previous studies that found male PLWs showed stronger facing-towards bias than female PLWs [18, 35, 49]. On the other hand, the finding indicates that the direction estimates of male PLWs are more accurate than those of female PLWs. This may be related to the common sense that male walkers are more threatening or aggressive than female walkers [17, 36]. To escape potential dangers, we need to accurately estimate male PLW directions. A similar trend is also observed for both male and female observers, with the reference repulsion bias being larger for female observers than for male observers as the genders of PLWs are evidently discriminated (Experiment 2). It can be proposed that this finding is due to the presence of threatening or aggressive information in PLW, which male observers may be more sensitive to than female observers, based on the logic presented in the previous paragraph. However, several studies have shown that female observers are more sensitive to threatening features than male observers [50–53]. Hence, this explanation does not make sense. Another explanation can be that male observers are more sensitive to motion stimuli than female observers [35, 54–56], resulting in higher estimation accuracy for males than females. Therefore, the reference repulsion bias was smaller for males than females. Moreover, Figs. 5 and 7 show that the reference repulsion bias sharply increases with gender differences for female observers but not for male observers. Figure 6b shows that the changes in the gender evaluation were more significant for male participants than for female participants when the weights and motion speeds of PLWs were changed, indicating that male observers are more sensitive to the gender of PLWs than female observers. Accordingly, it would be expected that the reference repulsion bias evidently changed for male observers rather than female observers. Therefore, the above finding could not be attributed to the different sensitivities of the gender discrimination between male and female observers. In addition, previous studies have found that observers can accurately evaluate the gender of PLW using physical form and motion features [43, 57]. The current study generated PLWs with different genders using the BML kit (https://www.biomotionlab.ca/) developed by [43, 44], which alters two physical features simultaneously. Previous research has shown that males exhibit superior motion feature processing abilities [56], while there is little difference between men and women in processing static form features [58, 59]. Therefore, it is unlikely that the modulation of observer genders in the reference repulsion bias is due to changes in physical features. Another possibility is that PLWs also contain sexual information, which becomes more apparent with greater gender differences. However, it should be noted that males tend to exhibit higher sexual arousal than females when exposed to sexual visual stimuli [60]. Therefore, our findings remain unaccounted for, leaving room for future studies. Apart from examining the reference repulsion bias, we also investigated the serial dependence in the PLW direction estimation, as first reported in [33]. Moreover, we discovered that neither the gender of the PLW nor the gender of the observer affected the serial dependence. Yang et al. (2014) found that the discrimination sensitivity of female PLWs was lower than that of male PLWs [35]. Hence, it can be proposed that female PLWs may be more dynamic than male PLWs. Moreover, it has been found that male observers are more sensitive to discriminating PLW direction than female observers [35], indicating that the PLWs are more dynamic for male observers than for female observers. Taubert et al. (2016) have argued that serial dependence occurs for stable attributes, such as gender, while the opposite effect occurs for dynamic attributes, such as facial expressions [61]. Accordingly, it can be expected that the serial dependence in the female PLW may be stronger than that in the male PLW, and that can be stronger for female observers than for male observers. However, our findings were inconsistent with these expectations. What differs from Taubert’s argument [61] is that Alais et al. (2017) found an attractive serial dependence occurred for the dynamic features (e.g., motion directions), but an opposite effect was observed for the static features (e.g., form orientations) [62]. Our findings are consistent with their findings. Taubert et al. (2016) [61] employed static facial stimuli; in contrast, the stimuli of Alais et al. (2017) [62] were the integration of form and motion features (i.e., both dynamic and static features were presented simultaneously). Our PLW stimuli also showed the dynamic (motion direction) and static (gender) features simultaneously. Hence, our serial dependence patterns were consistent with Alais et al. (2017) [62], suggesting that the stimulus types affect the serial dependence. Although the serial dependence in gender perception has been extensively discussed (see [38] for a review), previous studies have focused on morphing stimuli along a male/female continuum, treating gender as an independent physical feature. In contrast, our current study took it as a modulator in the serial dependence. Since the main function of serial dependence is to maintain the spatial and temporal continuity of the physical world [38–41], our results indicate that there is no significant difference in this function between males and females. Therefore, future studies can disregard the effects of gender on serial dependence. In summary, the current study is the first to report biases in the PLW direction estimation (reference repulsion bias and serial dependence) as they interact with observer and walker genders. These findings would be interesting to consider from biological evolutionary perspectives. Why have these gender effects evolved? And what are the neural mechanisms that lead male and female observers to differential estimate biases?