One psychological issue of research importance is what determines female attractiveness (Fallon & Rozin, 1985; Singh, 1993; Wiggins, Wiggins, & Conger, 1968). One reason for this research interest is the hypothesis that the media use of very thin models may play a role in the general dissatisfaction women have with their bodies and the increasing number of cases of eating disorders among women (Mable, Balance, & Galgan, 1986; Silverstein, Perdue, Peterson & Kelly, 1986).

While these studies have examined many different features related to female attractiveness, the samples of nearly all of the studies are drawn from college age students. These samples differ in many ways from the general population and may limit generalization. For example, the age range of most college samples is limited to individuals in their late teens to twenties. As people age, they may change in what they prefer in female figures.

Providing access to a more diverse sample is one way that the World-Wide Web can enhance the external validity of psychological investigations since the World-Wide Web has access to people around the world. However, there are significant potential problems with data collection over the World-Wide Web, e.g., the lack of control of the experimental environment. In order to examine the usefulness of the Web for data collection, experiments on the determinants of female attractiveness were conducted in both a laboratory situation and over the Web.

Subjects.

In the two laboratory studies, 333 subjects participated (nearly the same number in each study). In the World-Wide Web studies, 543 subjects participated. Prior to beginning any of the experiments, subjects were asked to fill out personal data forms and to sign informed consent forms. The subject's characteristics are shown in Table 1a for the laboratory studies and Table 1b for the World-Wide Web studies.

Stimulus Figures.

One set of studies used forward facing figures adapted from Fallon and Rozin (1985; Figure 1a). The figures were varied by weight and by bust width to hip width ratio. There were nine levels of weight going labeled in steps of ten from 10 to 90. There were three levels of bust to hip width ratios given bust > hips, bust = hips and bust < hips. When unequal, the larger width was 1.2 times the smaller width. There were 27 unique figures.

The other set of studies used side facing figures adapted from Wiggins et al. (1968). These figures were varied by weight, breast size, and buttocks size. There were six levels of weight set to match the above figure levels from 10 to 60. There were three sizes of breast and buttock for each weight value labeled small, medium and large. Thus, there were 54 figures.

(a) (b)

Figure 1. Example figures from the experiments judging (a) frontal view images and (b) side view images of female figures. Both figures are of a weight value of 30 for their experiments. Figure 1a is adapted from Fallon and Rozin (1985) and Figure 1b is adapted from Wiggins et al. (1968).

Procedure.

All of the studies employed a magnitude estimation procedure with a standard. The standard figure was given an arbitrary attractiveness value of 200. The experimental figures were presented one at a time paired with the standard. Subjects gave an attractiveness rating to indicate the attractiveness of the experimental figure relative to the standard, e.g. a figure perceived as half as attractive would be given a value of 100.

In the laboratory studies, the order of the figures was randomized and the same order was used for all subjects. The subjects were run in groups of various sizes using overhead projectors to present the stimulus figures.

After the laboratory studies were completed, the experiments were converted to HTML, the language of the World-Wide Web. Each figure was originally on a computer, file so it was simple to embed these images in Web pages. Each Web page contained the standard figure labeled with a value of 200. Below each test figure was an input box for the subject to enter their attractiveness rating. An example of the Web pages are shown in Figure 2. The figures were presented in a random order with rotation so that the last figure in the list was moved to the beginning of the list. By the nature of the Web, testing is individual.

(a)

(b)

Figure 2. Example Web pages from the two online versions of the experiments. (a) Page from the frontal view experiment. (b) Page from the side view experiment.

Thus, both experiments are as identical as possible given the different technologies used to conduct the experiments.

The side-view study was a 9 (weight) x 3 (bust to hip width ratio) x 2 (gender of subject) mixed factorial design. The frontal-view study was a 6 (weight) x 3 (breast size) x 3 (buttock size) x 2 (gender of subject) mixed factorial design. In both studies all variables, other than the gender of the subject, are within subject variables.

Laboratory Results.

For the laboratory data all of the scores are range-corrected to place all subjects' data on the same scale range of 1 to 100.

A 9x3x2 mixed design Analysis of Variance (ANOVA) was run on the side-view data and a 6x3x3x2 mixed design ANOVA was run on the frontal view data. Tukey's HSD was used to determine significant differences between means as the post-hoc test. While many of the main effects and interactions are significant at an a = 0.05 level, only some of the more important effects will be discussed in this poster.

(a)

(b)

Figure 3. Range corrected magnitude estimates of attractiveness for the Frontal-view laboratory experiment as a function of weight value, bust to hip proportion and gender of the subject. (a) Female subject ratings. (b) Male subject ratings.

Most of the important trends can be seen in the significant three-way interaction of weight, bust to hip width ratio and gender of the subject shown in Figure 3, F(16, 1760) = 2.00, p < .003. First, The effect of weight was significant with the preferred weight value being 30 for both female and male subjects. This finding contrasts sharply with Fallon and Rozin's (1985) finding that males preferred the 30 weight figure and females preferred the 20 weight figure more often. This difference may be due to changes in perception over time or the different methodologies. In general, the bust=hip figures were preferred and the bust>hips figure was least preferred especially around the preferred weight and for females. However, for the thinnest figures (weight 10) this trend is reversed somewhat in that the both males and females significantly prefer the bust>hips figure. At the optimal weight, females apparently do not make any distinction between the bust=hips and the bust<hips figure but males still significantly prefer the bust=hips figure over all other bust to hip ratios.

Figure 4. Range-corrected magnitude estimates of attractiveness for the Side-view laboratory experiment as a function of weight value, breast size, and gender of the subject. (a) Female subject ratings. (b) Male subject ratings.

For the side-view figures, the trends are similar and can be seen in the significant interaction of weight, bust size and gender of the subject shown in Figure 4, F(20, 2380) = 8.36, p < .0001. Buttock size played a very minor role in the results obtained with this sample and will not be discussed. From Figure 4 it is clear that preferred weight is at a weight value of 20. Overall females prefer a moderate breast size and males prefer large breast sizes but these preferences are only found for lower weight values. Not as visible, but significant, is the finding that males tend to rate thinner figures higher than females and that females tend to rate heavier figures higher, F(5, 595) = 3.21, p < .008.

World-Wide Web Results and Comparison of the Two Data Sets.

Since subjects can quit the experiment at any time, many of the subject data sets are incomplete. Therefore, it is not possible to range-correct the data which requires complete data sets. In addition, since data collection is ongoing, no complete statistical analysis has been performed Trends will be identified and pair-wise comparisons between means have been made used a Bonferonni t.

The main trends are very similar to the trends found in the laboratory studies. For example, the overall preferred weight is still a weight of 30 for the frontal-view figures (Table 2, Figure 5) and a weight value of 20 for the side-view figures (Table 3, Figure 6). However, for the frontal-view figures there was a non-significantly higher rating for the weight value of 40 when the bust=hips and at the optimal weight the bust<hips figures were preferred, t(220)=1.75, p>.05. In the laboratory studies, subjects significantly preferred the 30 weight figure over the 40 weight figure. One difference between the laboratory and Web results in the side-view figures is that both males and females significantly prefer the moderate breast size at the optimal weight in the Web studies. Recall that for in the laboratory studies, males preferred the larger breast size at all thinner weights (Table 4, Figure 6). However, at the thinnest weight, both genders prefer the larger breast size (Table 5, Figure 6).

Figure 5. Magnitude estimates of attractiveness for the Frontal-view World-Wide Web experiment as a function of weight value, bust to hip proportion and gender of the subject. (a) Female subject ratings. (b) Male subject ratings.

Figure 6. Magnitude estimates of attractiveness for the Side-view World-Wide Web experiment as a function of weight value, breast size, and gender of the subject. (a) Female subject ratings. (b) Male subject ratings.

The differences between laboratory and Web experiments may be due to the differences in the population being drawn from but there is not sufficient data on the population the test that inference. Breaking the data down by age does not reveal that the greater preference for the bust<hips at the optimal weight or other differences noted in the data can be accounted for by this variable.

One trend that has been observed in the data that seems to be related to the age of the subject is how males and females differently prefer thin versus heavy women. As was noted for the side-viewing data, males tend to give higher ratings than females for thinner figures and the reverse is true for heavier figures. In the World-Wide Web, this trend is observed in all age groups. However, for older subjects the weight value where males start rating the figures lower than females is at a higher weight value than for younger subjects (Figure 7).

Figure 7. Difference between mean magnitude estimates of attractiveness as a function of weight value for different age groups of subjects from the World-Wide Web experiments. (a) Frontal-view experiment. (b) Side-view experiment.

Experimental Conclusions.

1. Both males and females tend to prefer female figures of a moderate weight and not an ultrathin figure. These studies did not find that females prefer figures that are of a lower weight than males. This result was found by both laboratory and Web data sets and is contrary to Fallon and Rozin (1985).
2. There is a tendency to prefer figures that are wider at the hips than at the bust. This might be better stated that there are distinctly less favorable ratings for figures that are wider at the bust than at the hips at optimal weights.
3. Women tend to prefer moderate breast sizes while males are more likely to prefer larger breast sizes when the female figures are at the optimal weight.
4. For thin figures, the preference is for larger breast and bust sizes by both male and female subjects.

Conclusions about the Web as an Experimental Medium.

1. The data obtained by the Web versions of these experiments seem to be tapping the same psychological phenomena as in the laboratory experiments.
2. The sample is more varied than the laboratory sample, but cannot be claimed to be representative. As the Web increases in popularity and more people have access to it, the sample may become more representative. The Web samples are already moving in that direction.
3. Data collection is easy as it proceeds automatically.
4. Data collection is getting faster. Both samples would be nearly twice as large if the data collected in June 1996 were added.

Fallon, A & Rozin, P. (1985). Sex differences in perceptions of desirable body shape. Journal of Abnormal Psychology, 94, 102-105.

Mable, H., Balance, W., & Galgan, R. (1986). Body image distortion and satisfaction in university students. Perceptual and Motor Skills, 63, 907-911.

Silverstein, B., Perdue, L., Peterson, B., & Kelly, E. (1986). Some correlates of the thin standard of bodily attractiveness for women. International Journal of Eating Disorders, 5, 145-155.

Singh, D. (1993). Adaptive significance of female physical attractiveness: Role of waist-to-hip ratio. Journal of Personality and Social Psychology, 65, 293-307.

Wiggins, J., Wiggins, N., & Conger, J. (1968). Correlates of heterosexual preference. Journal of Personality and Social Psychology, 10, 82-90.