A total of 44 students (22 male and 22 female) from the School of Psychology at the University of Kent participated in this study in return for a small payment or course credits. Participants completed the Kinsey scale for the assessment of sexual orientation as part of a pre-screen on our online recruitment system. This is a seven-point scale in which a score of “0” represents complete heterosexuality and “6” complete homosexuality. Only participants who reported to be completely heterosexual (i.e., reporting “0” on the Kinsey scale) were invited to take part (Kinsey, Pomeroy, & Martin, 1948; Kinsey, Pomeroy, Martin, & Gebhard, 1953). The mean age of participants was 21.8 years (SD = 4.2; range 18–35 years). All reported normal or corrected-to-normal vision.
The stimuli consisted of natural beach scenes portraying men, women, and children (5 scenes for each of these four categories). To determine the approximate age of these categories, ten observers (5 males, 5 females) estimated the age of the people in the scenes in a pilot study. This revealed a mean age of 26.4 years (SD = 2.1) for men, 22.8 years (SD = 2.6) for women, 5.7 years (SD = 1.1) for boys, and 4.7 years (SD = 1.4) for girls. The age of the children therefore corresponds to stage 1 (prepubescent) of the Tanner stages of sexual development (see Tanner, 1978). Additionally, a set of control beach scenes without any person content (5 scenes) was included, resulting in a total of 25 scenes. People were portrayed in swim or leisure wear. All stimuli were purchased from an internet photograph database (www.mostphotos.com) and were selected to be of similar composition and size, and to depict the persons in similar poses and with a comparable level of clothing (see Fig. 1). To confirm that these targets were of similar size, their percentage occupancy area in the scenes was calculated. This showed that all person categories occupied a similar amount of space in our scenes (mean = 7.1 %, SD = 3.4, range across person categories = 6.6–7.7 %; one-factor ANOVA: F(3, 19) = 0.14, p = 0.94).
In addition, three versions were created of each scene that were identical in all aspects except for image quality. This resulted in a total of 75 scene images. In the original quality condition, the image quality of the downloaded photographs was retained. In the high-quality version, the images were processed by applying the “Auto Levels,” “Auto Contrast,” and “Auto Color” functions in Adobe Photoshop CS3 to artificially enhance the original photographs. Finally, to create a luminance-controlled version of the stimuli, the photographs were divided into groups of five (one of each category) based on similar luminance values and standard deviation. A mean luminance value and standard deviation were calculated for each of the five groups. Each photo within a group was then re-adjusted to obtain the mean luminance and standard deviation that matched the group value. Therefore, at least one image from each category (men, women, boys, girls, no-person landscapes) had precisely matched luminance values. This particular group-based approach was adopted to avoid the extreme deviation from the natural luminance values of individual scenes. This can occur when a single mean luminance value is derived for large stimulus sets, which can result in some highly distorted and unnatural looking images. Table 1 shows the overall mean luminance values and standard deviation for the different image categories for all scenes. Example stimuli are shown in Fig. 2.
Two questionnaires were also included in the experiment. The first was a general information scale relating to sexual interest and instructed participants to select one or more of five applicable statements (“no sexual interest in adults,” “strong sexual interest in female adults,” “some sexual interest in female adults,” “some sexual interest in male adults,” “strong sexual interest in male adults”). This was included to confirm the sexual interests that participants reported in the pre-screen. In addition, all participants completed the Interest in Child Molestation Scale to ensure that they were solely sexually interested in adults (Gannon & O’Connor, 2011). This scale consists of five short scenarios that describe incidents of child molestation. In response to these scenarios, participants have to rate their arousal, enjoyment, and behavioral propensity to child sex abuse on 7-point Likert scales. This scale has high test–retest reliability (r = .94) and its sexual arousal subscale correlates with the Implicit Association Test, which provides an indirect measure of child sexualization associations (see Gannon & O’Connor, 2011).
The stimuli were displayed using SR-Research Experiment Builder software (version 1.1.0) on a 21″ color monitor, with a screen resolution of 1024 × 768 pixels. Eye movements were tracked using an SR-Research Eyelink II head-mounted eye-tracking system. The Eyelink II was running at a 500 Hz sampling rate, a spatial resolution of <0.01° of visual angle, a gaze position accuracy of <0.5°, and a pupil size resolution of 0.1 % of diameter. The Eyelink II system works by measuring corneal reflection and dark pupil with a video-based infrared-camera eye tracker, which computes the number of camera pixels that are occluded by participants’ pupils. In this system, the diameter of the pupil is recorded as an integer that ranges from 400 to 16,000 units. The device incorporates eye and head tracking that automatically compensates for minor head movements. During the recording of eye movements, participants are instructed to remain seated still but further immobilization (e.g., a chinrest) is not required. This eye-tracking system is compatible with most glasses and contact lenses.
Participants were invited to take part in an experiment on sexual interest and informed that they would be viewing images of males and females of different ages while their eye movements were being recorded. Participants were kept naïve to the full purpose of the experiment until the end. To fully understand observers’ natural interests in these scenes, a free-viewing paradigm was used so as not to constrain spontaneous eye movement patterns. Thus, participants were instructed simply to “view the scenes as you naturally would” (for similar approaches, see Bindemann, Scheepers, & Burton, 2009; Fromberger et al., 2012a, b, 2013; Hall et al., 2011; Hewig et al., 2008; Lykins et al., 2008; Nummenmaa et al., 2012).
Participants were seated in a quiet and windowless room with consistent artificial lighting and positioned approximately 60 cm from the display monitor. The participants’ left eye was tracked and calibrated using the standard Eyelink procedure. To calibrate the eye tracker, observers fixated an initial series of nine target points on the display monitor. Their accuracy was then validated against a second series of nine fixation targets. Calibration was repeated if poor measurement accuracy was indicated. In the experiment, each trial began with a central fixation dot, which allowed for drift correction. This was followed by a gray screen display for 1000 ms, and then the stimulus display for 5000 ms, followed by another gray screen for 1000 ms. This display duration is similar to other studies with static images (e.g., Fromberger et al., 2012a, b, 2013; Hewig et al., 2008; Nummenmaa et al., 2012) and allows for approximately 15 fixations (based on an average fixation duration lasting 200–300 ms, see Rayner, 1998), which is sufficient time to scan the entire scene.
Each participant viewed all 75 stimuli. These were presented in a randomized order that was uniquely generated for each participant by the EyeLink software. Short breaks were inserted every 25 trials, after which the calibration procedure was repeated. On completion of the eye-tracking task, participants answered the general information scale relating to their sexual interests and the Interest in Child Molestation Proclivity scale (see Gannon & O’Connor, 2011).
Confirmation of Sexual Interests
To ensure that participants were not sexually interested in children, responses on the Interest in Child Molestation Scale were analyzed first. An overall interest score was calculated for each participant by combining responses across all subscales (i.e., arousal, enjoyment, behavioral propensity) (for similar analysis, see Gannon & O’Connor, 2011). This produced a total score where a minimum of 15 (low sexual interest in children) and a maximum score of 105 (high sexual interest in children) are possible. The results here converge with those obtained in previous studies with a sample of non-offending community males (Gannon & O’Connor, 2011), such that male observers scored a mean of 18.1 (mode = 15, SD = 5.6, min = 15, max = 30) and 16.8 for female observers (mode = 15, SD = 5.6, min = 15, max = 41). However, an established cut-off point for this scale does not exist. We adopted a simple metric by considering only individuals with scores on the lowest third of the scale (i.e., with scores between 15 and 45). All participants fell within this range.
Sexual orientation was confirmed with the general information scale that was administered following the eye-tracking task (see “Materials” section). In the 22 male observers, 19 reported “strong sexual interest in women” and three selected “some sexual interest in women.” Among the 22 females, 12 selected “strong sexual interest in males” and 10 selected “some sexual interest in males.” Participants reported no other sexual interests in this questionnaire.
For the analysis of the eye-tracking data, all eye movements were pre-processed by merging fixations of less than 80 ms with the preceding or following fixation if it fell within half a degree of visual angle (for similar approaches, see e.g., Attard & Bindemann, 2013; Bindemann et al., 2009; Bindemann, Scheepers, Ferguson, & Burton, 2010). In addition, any fixations that fell outside the dimensions of the display monitor or that were obscured by blinking were excluded. To analyze attention to specific areas within the visual scenes, each image was then coded to define three regions of interest (ROIs), which comprised the head and body of the persons and the scene background. The mean percentage of fixations that fell on these ROIs was then calculated across observer groups (males, females) and stimulus categories (men, women, boys, girls).
For the measure of main interest, observers’ pupillary responses were computed by taking the mean pupil diameter at each fixation, averaged across the duration of a stimulus display. These values were then used to compute an overall mean, across all stimuli, for each participant. The percentage difference (i.e., an increase or decrease) in pupil diameter for each stimulus category (men, women, boys, girls, no-person scenes) from the overall mean was then computed, using the formula: (mean pupil diameter for category × 100)/overall pupil mean. Accordingly, a score of 100 % indicates that the pupillary response to a stimulus category does not differ from the overall mean. Scores higher or lower than this value indicate comparatively larger or smaller pupil sizes (for similar approaches, see Dabbs, 1997; Laeng & Falkenberg, 2007). To simplify the expression of these patterns, these scores were then deducted from 100 so that no change in pupil size is indicated by zero and positive or negative scores reflect relatively larger (dilation) or smaller (constriction) pupil sizes in response to a stimulus category.
We first examined the viewing patterns that the persons in the scenes elicited in male and female observers. To examine this, the percentage fixations to the ROIs were calculated for all stimulus categories (see Fig. 3). Overall, 63 % of fixations fell on the figures in the scenes (range 58 to 71 % across conditions), which indicates that the person content of the scenes was of most interest. A 4 (category: men, women, boys, girls) × 3 (ROI: head, body, background) × 2 (observer sex: male, female) mixed-factor ANOVA revealed a three-way interaction, F(6, 252) = 8.01, p < 0.001, partial η
2 = 0.16. To explore this interaction, two separate 4 (category: men, women, boys, girls) × 3 (ROI: head, body, background) within-subjects ANOVAs were performed for male and female observers.
For male observers, this analysis showed no main effect of category, F(3, 63) = 0.32, p = 0.81, partial η
2 = 0.02, but revealed a main effect of ROI, F(2, 42) = 4.54, p < 0.05, partial η
2 = 0.18, and an interaction between both factors, F(6, 126) = 34.22, p < 0.001, partial η
2 = 0.62. To explore this interaction, Bonferroni-adjusted pairwise comparisons of the stimulus categories were conducted for each ROI. These comparisons show that more fixations were directed at the background of scenes containing boys, girls, and men (39 to 42 %) than scenes depicting women (30 %), all ps < 0.01. In addition, boys (31 %) and girls (32 %) received more fixations to the head than men (27 %) and women (22 %), all ps < 0.01, and men’s heads were also fixated more frequently than those of women, p < 0.01. By contrast, male observers directed more fixations to the bodies (48 %) of women and men (34 %) than those of boys (27 %) and girls (26 %), all ps < 0.001, and more at women’s bodies than those of men, p < 0.001. None of the other comparisons reached significance, all ps ≥ 0.10.
The equivalent analysis for female observers showed no main effect of category, F(3, 63) = 0.16, p = 0.92, partial η
2 = 0.008, but a main effect of ROI, F(2, 42) = 2.58, p < 0.001, partial η
2 = 0.11, and an interaction between factors, F(6, 126) = 8.45, p < 0.001, partial η
2 = 0.29. Bonferroni-adjusted pairwise comparisons of the stimulus categories show that more fixations landed on the head region of boys and girls (both 34 %) than women (22 %) and men (29 %), all ps < 0.001, and on the heads of men than women, p < 0.001. By contrast, more fixations landed on women’s bodies (40 %) compared to boys (29 %) and girls (31 %), both ps < 0.01. No other comparisons reached significance, all ps ≥ 0.08.
Overall, this pattern suggests a clear interest, whereby heterosexual males and females fixate men and women more frequently than children, but are particular biased towards the bodies of adult female targets.
The measure of main interest is pupillary response, which was analyzed in two ways. In the first analysis, pupillary responses were compared for male and female observers across the stimulus categories and image conditions. These data are illustrated in Fig. 4. A 3 (image quality: original, high, luminance-controlled) × 5 (category: men, women, boys, girls, no-person) × 2 (observer sex: male, female) mixed-factor ANOVA revealed a main effect of category, F(4, 168) = 20.35, p < 0.001, partial η
2 = 0.33, but not of quality, F(2, 84) = 1.75, p = 0.18, partial η
2 = 0.04, or observer sex, F(1, 42) = 1.00, p = 0.32, partial η
2 = 0.02. However, an interaction between image quality and observer sex was found, F(2, 84) = 3.36, p < 0.05, partial η
2 = 0.07. Bonferroni-adjusted pairwise comparisons revealed only that female observers exhibited larger pupils than male observers during the viewing of luminance-controlled scenes, p < 0.05. No other differences were significant, all ps ≥ 0.09. An interaction between image quality and category was also found, F(8, 336) = 2.17, p < 0.05, partial η
2 = 0.05, as the no-person beach scenes elicited smaller pupils in the luminance-controlled than the high quality, p < 0.01, and original quality conditions, p < 0.05. No other differences between any of the person content scenes were found, all ps ≥ 0.16. Therefore, image quality was not analyzed further.
An interaction between category and observer sex was also present, F(4, 168) = 2.73, p < 0.05, partial η
2 = 0.06. Bonferroni-adjusted pairwise comparisons revealed smaller pupils in male than female observers during the viewing of men, p < 0.01. Furthermore, in male observers, women elicited larger pupil sizes than men, boys, girls, and no-person scenes, all ps ≤ 0.001. For female observers, women elicited larger pupil sizes than boys, girls, and no-person scenes, all ps ≤ 0.05, but not men, p = 0.26. In addition, pupil responses were larger for scenes depicting boys than girls, p < 0.05. No other differences were observed, all ps ≥ 0.06, and an interaction between the three factors was not found, F(8, 336) = 1.10, p = 0.36, partial η
2 = 0.03. Overall, these results therefore reveal a dilation response in male observers that appears to be consistent with self-reported sex and age preferences. Female observers’ responses are also consistent with their age preferences, but do not correspond with their reported sexual interest in adult men.
In the second analysis, this pattern is confirmed when pupillary responses are compared via one-sample t-tests (with alpha corrected at p < 0.01 for multiple comparisons) with a baseline that reflects the mean pupil diameter across all stimuli (see “Data Analysis” section). This analysis shows that the pupils of male observers were larger than baseline during the viewing of women, t(21) = 5.43, p < 0.001, d = 2.37, and smaller during the viewing of men, t(21) = −3.02, p = 0.006, d = 1.32, and girls, t(21) = −3.1, p = 0.005, d = 1.35. In addition, pupil size was unchanged from baseline in response to boys and no-person scenes, both ts ≤ −1.59, ps ≥ 0.126, ds ≤ 0.69. In female observers, pictures of men, t(21) = 1.49, p = 0.15, d = 0.65, boys, t(21) = −0.12, p = 0.91, d = 0.05, and landscape beach scenes (−1.53 %), t(21) = −2.19, p = 0.04, d = 0.96 did not elicit a change in pupil size from baseline. The pupils were enlarged to scenes with women, t(21) = 4.71, p < 0.001, d = 2.06, and smaller than baseline during the viewing of girls, t(21) = −4.33, p < 0.001, d = 1.89.
Individual Differences in Pupillary Responses
We also sought to explore whether pupillary responses can be informative about the sexual interests of individual observers. For this purpose, the difference in raw pupil size for specific image comparisons (e.g., scenes with men vs. women) was calculated separately for each participant. These data show, for example, that all of the male observers (22/22) recorded larger pupil sizes during the viewing of women than men, and 91 % (20/22) of male observers displayed larger pupils in response to women than girls. In addition, only 22 % (5/22) of these participants showed a greater pupillary response to men than boys. With regard to female observers, 73 % (16/22) showed more pupil dilation during the viewing of women than men. However, 86 % (19/22) of this participant group also exhibited larger pupils in response to women than girls, and 59 % (13/22) recorded larger pupils to men than boys.
The purpose of this experiment was to explore whether pupillary responses to the visual presentation of men and women can provide an indication of a person’s sexual interests. More specifically, we sought to determine whether this approach can be extended to reveal age-specific sexual interests. We first looked at fixation patterns on the person content in scenes. Male observers showed a viewing preference for women over men and children, which was characterized by a high number of fixations on women’s bodies. These results are consistent with other studies, which have shown that heterosexual male observers attend more to images of the opposite sex (Lykins, Meana, & Strauss, 2006; Lykins et al., 2008; Rupp & Wallen, 2007; Suschinsky et al., 2007) and that such preferential viewing behavior is also age-specific (Fromberger et al., 2012a, b, 2013; Hall et al., 2011). Female observers also recorded fewer fixations on the faces of women than men and children, but more on women’s bodies than those of children. Consistent with previous research, heterosexual females therefore showed age-specific viewing patterns but did not exhibit the same strong visual preferences to opposite-sex figures as men (Hall et al., 2011; Israel & Strassberg, 2009; Lykins et al., 2008; Rupp & Wallen, 2007).
The data of main interest were the pupillary responses. In heterosexual male observers, these responses were consistent with their reported sexual interests. Thus, pictures of women elicited a clear pupillary dilation that was not present during the viewing of men and children. In female observers, pupil dilation was also greatest when pictures of women were viewed. In these participants, pupillary recordings therefore do not correspond to their self-reported sexual orientation. However, these responses still appeared to be age-specific as the pupils remained unchanged or constricted during the viewing of children.
These results converge with a recent study that has shown a similar pattern of pupillary responses for heterosexual adult males and females (Rieger & Savin-Williams, 2012). Experiment 1 extends these findings by demonstrating that such pupillary responses are also age-specific. A question that arises, however, is whether these dilation effects could be attributed to a low-level factor such as luminance. To explore this possibility, we also compared scene photographs in which contrast and color were enhanced with a set in which luminance and contrast were equated. The results for these stimulus categories were highly comparable, which suggests that pupillary responses for the different person categories cannot be explained by general variation in luminance.
There is, however, a problem with the luminance adjustment that was employed in Experiment 1. While this manipulation was used to equate luminance across scenes, it does not control other low-level image aspects, such as color, which might also affect pupillary responses (Kohn & Clynes, 1996; Lobato-Rincón et al., 2014). Such information was not matched across stimulus categories in Experiment 1. Consequently, the possibility remains that the results might reflect such image artifacts.
A second explanation is also possible for the observed pupillary responses. While we adjusted the mean luminance of the scenes, we did not measure the sexual attractiveness of the target figures. As a result, this might have been mismatched across categories. Considering that photographs of women elicited more pupil dilation in both male and female observers, it is conceivable, for example, that these pictures were generally more sexually arousing than those of men. To investigate these possibilities, a second experiment was conducted.