Measuring Sexual Preferences Using Eye Movement Data

A Proposal for Research

Michael Wogan, Ph.D., J.D.
Rutgers University
Camden College of Arts and Science
September 30, 2002

The purpose of the proposed project is to establish a method of measuring eye movements (sequence of movements, fixation time, and pupil diameter) as a way of assessing sexual preferences among normal males and a comparison group of convicted sex offenders. The results will contribute to understanding the cognitive processing of personally salient material. Once perfected, the method may be used as a non-invasive screening device in a variety of situations. It will also provide a method for measuring any changes in sexual preferences among offenders that might occur as a result of treatment, thus supplementing measures presently used to predict the likelihood of re-offense.

Eye Movements as an Indicator of Mental Activity

    The usefulness of eye tracking as a way of predicting recall for complex events was first demonstrated by Loftus,
Loftus and Messo (1987). Loftus et al. found that a record of eye movements provided evidence of a weapon focus, a tendency for observers of a crime scene involving a weapon to cluster their eye fixations on the weapon to the relative neglect of other elements in the scene. When a weapon was present, observers failed to remember other details of the scene with normal accuracy.

    Interest or novelty attracts eye fixations. Viewers fixate more frequently, earlier, and with longer duration on objects that are unfamiliar or out of place in a scene (e.g., an octopus in a farm scene; Loftus and Mackworth, 1978; see Henderson,
Weeks, and Hollingworth, 1999). In a facial recognition task, observers viewing famous faces make fewer fixations, sample fewer regions, show less constraint (lower predictability) in their transitions to successive locations, and make more symmetric fixations as compared to scanning of nonfamous faces (Althoff and Cohen, 1999).

    Eye movements and gaze fixation are particularly sensitive indices of interest and preference (Ohta, 1999). Images to which the person is particularly sensitive attract more and longer fixations. Threatening images of human faces attract a higher proportion of eye fixations than low-threat faces (Moog, McNamara, Powys, Rawlinson, Seiffer, and Bradley, 2000). In a study comparing individuals with Generalized Anxiety Disorder (GAD), depressive disorder, and normals, observers were presented with 4 types of faces: threatening, sad, happy, and neutral.  Observers with GAD looked first or sooner, and longer, toward the threatening faces than toward the neutral faces (Mogg, Millar, and Bradley, 2000). In another study, depressives showed reduced attentiveness to happy faces (Bradley, Brendan, Mogg, and Millar, 2000).

    These findings provide evidence of a link between individual preferences and the timing and length of eye fixations.

    The findings are consistent with studies which link eye movements and activity in the brain (Carpenter, Just, Keller,
Eddy, and Thulborn, 1999; Irwin and Brockmole, 2000; Spivey and Spirn, 2000; Stuyven, Van der Goten, Vandierendonck, Claeys, and Crevits, 2000). The brain directs eye movements, but the eyes provide information to the brain which is then used to guide further activity. The direction of the gaze anticipates motor movements (Land and Tatler, 2001). In guiding motor movements, the eyes fixate on landmarks that will anchor the next movement.  Once the motor movement is well underway, the eye fixation point moves to the next landmark (Eberhard, Spivey-Knowlton, Sedivy,
and Tanenhaus, 1995; Johansson, Westling, Baeckstrom, and Flanagan, 2001).

    Eye movements are closely linked to the processing of spoken language, whether on the input or the output side.
Resolving ambiguity in the recognition of spoken words is faster when there is also visual information (Allopenna, Magnuson, and Tanenhaus, 1998; Tanenhaus, Spivey-Knowlton, Eberhard, and Sedivy, 1995). Speakers begin fixating on pictured elements less than a second before naming them; the eye movements anticipate the order in which the elements are named (Griffin and Bock, 2000).

    The direction of the gaze correlates with the type of material being retrieved during a mental task. Richardson and
Spivey (2000) asked their subjects to scan a 3 x 3 grid in which information was presented in some of the squares. When the information was removed and only blank squares remained, the subjects were asked questions that could be answered using information from one of the squares. Anticipating the answer, before beginning to speak the subject's gaze shifted to the empty square where the information had been displayed.

    There is an increase in pupil diameter as the cognitive complexity of a task increases, whether the task is a non-verbal
one (Just, Carpenter, and Hemphill, 1996) or a linguistic one (Carpenter, Just, Keller, Eddy, and Thulborn, 1999; Just and
Carpenter, 1995). Pupil dilation increases in response to increasingly painful stimulation. The magnitude of pupil dilation is correlated with subjective pain reports, and with the latency and magnitude of brain evoked potentials (Chapman, Oka,
Bradshaw, Jacobson, and Donaldson, 1999; Oka, Chapman, and Jacobson, 2000)

    It seems reasonable to conclude that the direction of the gaze is a useful index of attention, interest or arousal, and the
extent or depth of cognitive processing, particularly if information regarding eye movements is combined with a measure of
pupil diameter.

Current measures of sexual preference: Phallometric assessment

    Self report measures of sexual preference may be adequate for use with normal males in a research setting, but because of the possibility of faking, self report measures are unsuitable for use with offender populations, even when anonymity is
guaranteed. Among convicted offenders, self-report measures of sexual preference do not correlate very well with
phallometrically measured arousal (Firestone, Bradford, McCoy, Greenberg, Curry and Larose, 1998; Harris, Rice, Quinsey, and Chaplin, 1996; Quinsey, Steinman, Bergersen, and Holmes, 1975).  Yet in an offender population, an assessment of sexual preferences is an important part of assessing an individual's potential for re-offense. (Sex offenders are almost exclusively male and the discussion which follows is focused exclusively on males.)

    One of the best predictors of re-offense among sex offenders is phallometric assessment of sexual arousal in response to deviant sexual stimuli. Using a penile plethysmograph, the person is shown slides, videotaped interactions, or listens to a recorded passage depicting sexual activity between consenting or non-consenting adults, sexual activity between adults and children, or non-sexual violence between adults. From this data a deviance index is constructed which predicts the likelihood of re-offense (Barabee and Marshall, 1988; Firestone, Bradford, Greenberg, and Larose, 1998;
Harris and Rice, 1996; Harris, Rice, Quinsey, Chaplin, and Earls, 1992; Lalumiere and Quinsey, 1993; Quinsey, Rice and Harris, 1995; Rice, Harris and Quinsey, 1990; Zuckerman, 1971).

Arousal patterns in sub-groups of offenders

    Sub-groups within the population of sex offenders can be distinguished using phallometric measures. The clearest finding
to emerge from the research is the distinction between rapists and normals. In some studies rapists show reduced, but still
greater-than-normal response to coercive sex (Baxter, Barbaree, and Marshall, 1986). In other studies they show as great a response to depictions of coercive sex, and to non-sexual, violent episodes as to normal sex (Earls, and Proulx, 1986;
Lalumiere and Quinsey, 1994; Proulx, Aubut, McKibben, and Cote, 1994; Quinsey, and Chaplin, 1984; Quinsey, Chaplin, and Upfold, 1984; Rice, Chaplin, Harris, and Coutts, 1994). In these studies, normals always show less arousal to depictions of violence and/or coercive sex than to consensual sex.

    The arousal pattern of child molesters can be distinguished from that of rapists (Miner, West, and Day, 1995) and normals (Barsetti, Earls, Lalumiere, and Belanger, 1998; Quinsey and Chaplin, 1988a; Quinsey, Chaplin, and Carrigan, 1979; Seto, Lalumiere, and Blanchard, 2000). Within the group of child molesters, those with female victims are distinguishable from those with male victims (Castonguay, Proulx, Aubut, McKibben, and Campbell, 1993; Laws, Hanson, Osborn, and Greenbaum, 2000; Miner, West and Day, 1995; Seto, Lalumiere, and Blanchard, 2000). Child
molesters with female victims have interest patterns that are closer to normals than they are to those of homosexual pedophiles (Castonguay, et al., 1993).

    Intrafamilial incest offenders are usually distinguishable from extrafamilial offenders (Castonguay, Proulx, Aubut,
McKibben, and Campbell, 1993; Frenzel and Lang, 1989; Miner and Dwyer, 1997; Rice, Quinsey, and Harris, 1991), with the intrafamilial offenders having arousal patterns that are closer to those of normals. Some researchers have failed to find any differences between intra- and extra-familial offenders (Quinsey, Chaplin, and Carrigan, 1979), especially when offenders with exclusively female victims are considered (Barsetti, Earls, Lalumiere, and Belanger, 1998).

    Age of the victim is sometimes a distinguishing factor among child molesters, but the effect has not been well studied.
Many studies lump together prepubescent children across a broad range of ages (Quinsey, Chaplin, and Carrigan, 1979; Rice, Quinsey, and Harris, 1991). When stimulus materials are sorted by apparent age, differences in age preferences appear that are consistent with the offender's choice of victim. Arousal patterns for offenders with preferences for young children (ages 5 to 10) are distinguishable from those with preferences for older children (pre-pubescent adolescents ages 11 to 13; Castonguay, Proulx, Arbut, McKibben and Campbell, 1993; see also Quinsey, Steinman, Bergersen, & Holmes, 1975). As Castonguay, et al. and others point out, the age of consent for young females is culturally dependent. A preference for young, but definitely post-pubertal girls is biologically "normal," whereas a sexual preference for pre-pubertal girls is not.

Disadvantages of phallometric assessment

    Phallometric assessment has several disadvantages. The method is invasive, technically complex, and requires special
equipment and specially trained personnel to administer. There is a risk of false positive identifications among normal males.  The measure is ineffective with intrafamilial sex offenders, especially those with female victims, who often show normal patterns of sexual arousal (Barbaree and Marshall, 1989; Barsetti, Earls, Lalumiere and Belanger, 1998).

    No one is sure of the exact percentage, but there is a fairly high level of false negatives among convicted sex
offenders who appear "normal" even though they are not. For example, in the study by Castonguay, et al. (1993) presentence offenders showed less arousal than those who had been sentenced.  In some studies, 30% of the offenders tested did not show sufficient arousal to derive a usable score (Barbaree, Seto, Serin, Amos, and Preston, 1994; Freund and Blanchard, 1989).  (But see Castonguay, et al., 1993, "[None of our tests] have yielded a phallometric record with absolutely no changes," p. 506.) Finally, there is concern among researchers that the measure may be susceptible to faking (Harris, Rice, Chaplin and Quinsey, 1999; Lalumiere and Earls, 1992; Proulx, Cote and Achille, 1993; Quinsey and Chaplin, 1988b; Rice, Harris and Quinsey, 1990).

    The plethysmographic method is also politically unpalatable. Several jurisdictions have discontinued using it following public concern that prison psychologists were "showing pornography to prisoners." Even though the charge is not true--
specially developed, non-pornographic slides, verbal descriptions, and videotapes are used--these charges are politically explosive and hard for public officials to resist, making the plethysmographic method unsuitable for use with a broad population of offenders.

Current measures of sexual preference: The Abel Screen

    Another instrument, the Abel Screen (Abel, 1994; Abel, Lawry, Karlstrom, Osborn, and Gillespie, 1994;
http://www.abelscreen.com) combines viewing time with data from a self-report questionnaire. The person being tested rates the sexual attractiveness of slides of individual persons using a computer keyboard. The slides include males or females, children, adolescents, or adults, presented clothed or nude (Abel, Lawry, Karlstrom, Osborn and Gillespie, 1994). The person advances to the next slide by pressing a key, giving an approximate measure of viewing time or Visual Reaction Time (VRT). The VRT is longer for photographs which the viewer finds more attractive (Lang, Searles, Lauerman, & Adesso, 1980; Quinsey, Rice, Harris, & Reid, 1993; Zamansky, 1956). The measure has been validated with child molesters by comparing the VRT score with plethysmographic measures of arousal (Abel, Huffman, Warberg, and Holland, 1998).

Disadvantages of the Abel screen

    Although the Abel screen performs well with admitted offenders, it's effectiveness in predicting recidivism in an
offender population has not been well tested. Most testing of the Abel Screen has been done with child molesters and it is not clear how well the measure performs with rapists. With child molesters, the discrimination between offenders and normals is better when the offender's preference is for young boys rather than young girls (Abel, Lawry, Karlstrom, Osborn and Gillespie, 1994).

    Abel's Relapse Prediction Score is derived by combining VRT data with static variables, such as present age, age at which offending began, number of prior convictions, and type and diversity of victims. The incremental validity gained by adding VRT data to the other measures is uncertain, and no data have been reported. Finally, the formula for combining scores into a Relapse Prediction Score is proprietary and cannot be used without the payment of licensing fees. Although some states have adopted the Abel screen, it should probably be viewed as an experimental, rather than a well-established measure.

Research Design and Methods

    The research involves presenting the viewer with photos or slides showing two persons in an ordinary social interaction on a computer screen. The figures in the slides are paired using two dimensions, male-female, and adult-child. The viewer is asked to rate the attractiveness of each person in the photo. As part of obtaining informed consent, it will be explained to the participant that the study is designed to learn how a person determines the attractiveness of people in photographs. Eye movements and pupil diameter are recorded while the viewer scans the figures. The participant presses the spacebar on a computer keyboard to advance to the next slide.

    The figures shown are adult males and females, and children of different ages, presented in various pairings. Each photograph shows one type of pairing, for example, adult male with adult female; adult male with young male child; adult female with latency-age male child, and so forth. There will be 60 photographs in all, three for each type of pairing, resulting in three blocks of 20 photographs each. The order of presentation is randomized within each block.

    Money (1999) points out that preferred forms of sexual interaction and sexual partner tend to be somewhat unique to the individual, with a critical developmental period occurring around age eight. Although the research evidence is sparce (Frenzel and Lang, 1989; Greenberg, 1999; Murphy, Haynes, Stalgaitis and Flanagan, 1986; Quinsey, Chaplin and Carrigan, 1979), clinicians believe that pedophiles develop preferences for children within a narrow age range. For example, Laws, Gulayets, and Frenzel (1995) present individual profiles of phallometrically measured preferences of heterosexual pedophiles (their Fig. 5) and homosexual pedophiles (their Fig. 6). Both groups are small, consisting of only four cases in each group. As expected, the groups differ in the amount of arousal shown in response to male vs. female photos. Within each group, however, there are marked individual differences in the age of the preferred target.

    To capture these preferences, photographs of children will be pre-screened for apparent age by judges, using developmental criteria (Tanner, 1962; 1971). Photos which are ambiguous as to perceived age will be omitted.

    Data will be collected from two groups of research participants, with 75 cases in each group. Normal males with no
history of conviction for sex offenses will be recruited.  Similar procedures, using identical photographs, will be
followed using convicted child molesters at the New Jersey Adult Diagnostic and Treatment Center (ADTC), which is a treatment center for sex offenders run by the N.J. Department of Corrections (DoC).