1. Introduction

It has been proposed that certain physical features honestly signal an individual's health and reproductive potential (Buss, 1994, Symons, 1995). If one were sensitive to these features, then one would be able to gauge the health and fertility of a possible partner. Certain specific values of these features could signal optimal health and fertility; thus, an individual whose features correspond to these values would be regarded as optimally attractive. From this, it follows that in environments where the optimal values for these features differ due to differences in environmental pressures, the attractiveness preferences should also be different (Anderson, Crawford, Nadeau, & Lindberg, 1992; Brown & Konner, 1987; Ember, Ember, Korotayev, & de Munck, 2005). Moreover, as observers moved from one environment to another, it would be adaptive for them to alter their attractiveness preferences to those that accurately reflected optimal health and fertility in their new environment.

In women, two potentially critical physical features are shape and weight scaled for height (in kilograms per meters squared). This latter factor is called the body mass index (BMI; Bray, 1998). For shape, research has focused on the ratio of the circumference of the waist to the circumference of the hips [the waist–hip ratio (WHR)] among women. For women in Western Europe and the USA, a low WHR (i.e., a curvaceous body) is suggested to correspond to the optimal fat distribution for high fertility (Wass et al., 1997, Zaadstra et al., 1995); hence, this shape should be highly attractive within these cultures. The optimal fat distribution is proposed to correspond to a ratio of 0.7, and this is suggested to be optimally attractive (Furnham et al., 1997, Henss, 2000, Singh, 1993, Singh, 1994). BMI also appears to be a strong predictor of attractiveness in Western observers (Fan et al., 2004, Puhl & Boland, 2001, Thornhill & Grammar, 1999, Tovée et al., 2002, Tovée et al., 1999, Tovée et al., 1998). There are also advantages to using BMI as a basis for mate selection because BMI provides a reliable cue to female health (Manson et al., 1995, Willet et al., 1995) and reproductive potential (Frisch, 1988, Lake et al., 1997, Reid & Van Vugt, 1987, Wang et al., 2000). For Caucasian women in Western Europe and the USA, Tovée et al. suggest that the balance between the optimal BMI for health and fertility is struck at around a value of 19–20 kg/m2, which, in their studies, is also the preferred BMI for attractiveness. However, they also suggest that the ideal may vary in different racial groups and different environments (Tovée & Cornelissen, 2001), and a number of studies have suggested that resource-poorer societies prefer a heavier female body than more affluent societies (Anderson et al., 1992, Brown & Konner, 1987, Ember et al., 2005).

To test whether attractiveness preferences are adaptive and whether the preferences are different in environments where there are different local optima for survival and reproduction, we have tested the attractiveness preferences of Caucasian observers in the UK and Zulu observers in rural South Africa. Although a lower BMI and a more curvaceous body is associated with higher health and fertility in the UK, such a body is associated with low socioeconomic status (SES), poor nutrition, and disease (e.g., Clark et al., 1999, Mvo et al., 1999) in rural South Africa. One would anticipate that a higher BMI would be preferred in this environment, which is consistent with previous studies in Tanzania using line-drawn figures (Marlowe & Wetsman, 2001, Wetsman & Marlowe, 1999). In a general population, as BMI rises, so does WHR (South African Department of Health, 1998, UK Department of Health, 2003). Hence, as a secondary effect of preferring a heavier BMI, observers may also choose a less curvaceous body. Additionally, we have tested the preferences of a cohort of Zulus who have moved to the UK to see whether their attractiveness preferences are shifting to adapt to their changed environment. We have also tested the attractiveness preferences of British people of African descent, who were born and who grew up in the UK, as a final control group.

2. Method

We recruited four groups of participants from the UK and South Africa. The first group was composed of 100 British Caucasians, equally divided between the sexes. The second group consisted of 35 South Africans (19 males, 16 females) from Mshwati-Mpolweni, in KwaZulu-Natal (Eastern Seaboard of South Africa). This group was made up of ethnic Zulus with little command of any other language and with intermittent education and was employed either as subsistence farmers or as temporary laborers and domestic workers in nearby towns like Howick. The third group was composed of 52 Zulu migrants into Britain (25 males, 27 females), all of whom were born in South Africa and moved to Britain in the 18 months before taking part in the study. The final group was made up of 66 Britons of African descent (34 males, 32 females), all of whom were born and raised in Britain. There were no significant differences in the age ranges of the different groups (see Table 1).

Table 1. Summary of the proportion of variance accounted for by BMI and WHR in the regression analysis of attractiveness judgements, plus the peak BMI for each group and the gradient of the relationship between attractiveness and WHR

	Group	Age, mean (S.D.)	BMI variance	WHR variance	Peak BMI	WHR gradient
	UK Caucasians	24.76 (6.96)	84.1	7.4	20.85	−6.54
	SA Zulus	25.6 (4.47)	82.5	7.5	26.52	+6.18
	Zulu migrants to the UK	26.56 (6.87)	86.2	1.7	23.99	−2.38
	Britons of African descent	24.41 (4.53)	76.5	7.3	20.68	−8.25

Participants in each group were asked to rate black-and-white images of 50 real women in front view. To generate the images, we videotaped consenting women standing in a set pose at a standard distance, wearing tight gray leotards and leggings in front view. Images were then frame-grabbed and stored as 24-bit images (see Tovée et al., 2002, for an example). The heads of the women in the images were obscured so that they could not be identified and so that facial attractiveness would not be a factor in subject's ratings. Although previous studies have manipulated the relative ranges of BMI and WHR to explore the relative contributions of these features to attractiveness judgements (e.g., Tovée et al., 2002, Tovée et al., 1999), in this study, the ranges of BMI and WHR used were not constrained and represented the widest range in our image library. The ranges seem to be not inconsistent with population data reported by epidemiological studies, such as the Health Survey for England 2003 (UK Department of Health, 2003).

For this experiment, the images of women were printed on sheets of A4 paper so that each image covered the entire page. This was done so as to facilitate replication of the study in all locations. Participants were presented with a booklet to record their ratings, where the first page consisted of brief instructions and a worked example of a rating and where the final page requested participants' demographic details (age, gender, ethnicity, weight, and height). Other pages in the booklet provided a nine-point Likert scale, which appeared below the question ‘How beautiful is the person in the photograph?’ and on which participants were asked to record their ratings. All participants were tested individually, with the only difference in procedure between the different settings being the language used. The questionnaire was in English for all groups except for the rural South African (SA) sample, for whom the questionnaire was translated into Zulu. Within the image set, individual images were presented in a randomized order.

3. Results

3.2. Multiple regression results

A multiple polynomial regression was used to model the contributions of BMI and WHR to the attractiveness ratings. Fig. 1 shows plots of the attractiveness ratings as a function of BMI for all four groups, with all sets being significantly explained by BMI (p<.001 in all cases). Fig. 2 shows the corresponding relationship between attractiveness and WHR. None of the observer groups showed a significant correlation between attractiveness ratings and WHR (p>.05). This suggests that for this image set, WHR has only a weak effect on attractiveness ratings for the observer groups tested. As in previous studies (e.g., Tovée & Cornelissen, 2001, Tovée et al., 1999, Tovée et al., 1998), we modeled the data using a multiple regression model (see Altman, 1991) to estimate the variance of attractiveness ratings explained by BMI and WHR. The model, run separately for the different groups, was

where y is the attractiveness rating, a is the intercept, x1 is the WHR, x2 is the BMI, x3 is the BMI2, x4 is the BMI3, and e is random error.

Fig. 1. Comparison plots of the attractiveness ratings by the four observer groups as functions of BMI. Each point represents the average attractiveness rating for a particular body by all the observers in one of the observer groups. Panel A illustrates a comparison of the responses of UK Caucasians (open squares and continuous line) and SA Zulus (filled circles and dashed line); Panel B illustrates that of the SA Zulus (filled circles and dashed line) and the Zulus in the UK (open triangles and continuous line); Panel C illustrates that of the Zulus in the UK (open triangles and continuous line) and Britons of African origin (filled stars and dashed line).

Fig. 2. Comparison plots of the attractiveness ratings by the four observer groups as functions of WHR. Each point represents the average attractiveness rating for a particular body by all the observers in one of the observer groups. Panel A illustrates a comparison of the responses of UK Caucasians (open squares and continuous line) and SA Zulus (filled circles and dashed line); Panel B illustrates that of the SA Zulus (filled circles and dashed line) and the Zulus in the UK (open triangles and continuous line); Panel C illustrates that of the Zulus in the UK (open triangles and continuous line) and Britons of African origin (filled stars and dashed line).

The variance explained by BMI and WHR in the best subsets analysis from this model is shown in Table 1.

4. Discussion

This study found substantial differences in the perception of female attractiveness in the different groups, a result consistent with some cross-cultural studies (e.g., Craig et al., 1996, Furnham & Baguma, 1994, Marlowe & Wetsman, 2001, Wetsman & Marlowe, 1999, Yu & Shepard, 1998). There was no difference between the preferences expressed by the UK Caucasian and UK African groups. There were large differences between these two groups and the SA Zulus (Fig. 1). The peak attractiveness for the SA Zulu observers is shifted to a higher BMI, and most strikingly, the attractiveness ratings of the SA Zulus do not show a dip in attractiveness for bodies with higher BMI values (see Fig. 1A). There is also a difference in the pattern of attractiveness preferences with changing WHR. Both the UK groups show a preference for a more curvaceous figure; the opposite appears true for the SA Zulus (Fig. 2A). This latter change may be, at least, partially a result of the preference of the SA Zulus for heavier bodies, which tend to be less curvy. The preferences of the Zulus who have moved to the UK seem to be intermediate between these two positions (Fig. 1, Fig. 2). The fact that there are differences between the UK Caucasian and UK African groups and the SA Zulus might be explained in terms of adaptations to the different environmental pressures.

In many traditional, non-Western settings, it is argued that body fat is believed to be an indicator of wealth and prosperity (McGarvey, 1991), with obesity as a symbol of economic success, femininity, and sexual capacity (Ghannam, 1997, Nasser, 1988). In less affluent societies, there is often a positive relationship between increased SES and body weight. Only high-status individuals would have been able to put on body weight, which would explain why many of the world's cultures had or have ideals of feminine beauty that include plumpness (Anderson et al., 1992, Brown & Konner, 1987), as it would have been advantageous for women to be able to store excess food as fat in times of food surplus. Conversely, with the food abundance in many industrialized nations, fatty foods are easily available, and it is those with higher SES who are more able to keep their weight down (Furnham & Alibhai, 1983), whereas fatness became associated with poverty, poor diet, and lack of slimming activities and exercise.

People in rural South Africa are living in a low-resource, economically deprived society. Fifty-six percent report going hungry, and most households do not have electricity, running water, or significant amounts of household durable goods (which is taken as a proxy of SES; South African Department of Health, 1998). Among people living in rural South Africa, a higher female body weight is perceived to reflect affluence, high status, and good health (Clark et al., 1999, Mvo et al., 1999). It is therefore not surprising that our results suggest that a higher female BMI is regarded as attractive. These preferences may be reinforced by the current health problems prevalent in South Africa. There are long-standing problems with infectious diseases, including lower respiratory tract infections, meningitis, diarrhea, septicemia and TB, which, when combined with low levels of immunization, make potential infection a serious possibility (South African Department of Health, 1998, South African Department of Health, 2000). The health consequences linked to these serious diseases include weight loss, and this is reflected in the perception that a lower body mass may signal potential parasitic infection or disease (Clark et al., 1999, Mvo et al., 1999).

The most prevalent and the most important of the potential diseases is HIV/AIDS infection (Blacker, 2004, Bradshaw et al., 2003, Caldwell, 1997, Hosegood et al., 2004). AIDS is now the single largest killer of men and women in South Africa, and this mortality rate is set to rise (Bradshaw et al., 2003, Hosegood et al., 2004). It is estimated that in 2005, 16.5% of the population of KwaZulu-Natal were HIV positive, the highest level of any province in South Africa (Shisana et al., 2005). The potential effect of AIDS on attractiveness preferences is magnified when one considers that the other important causes of illness and death (such as pulmonary or respiratory diseases) are associated with older adults, whereas the prevalence of HIV infection and the rate of AIDS deaths are highest in young and middle-aged adults (i.e., those who are most likely to be actively seeking partners; Bradshaw et al., 2003, Shisana et al., 2005). For example, it is estimated that in 2005, a quarter of 20- to 24-year-old women in South Africa were HIV positive, and a third of 25- to 29-year-old women were infected (Shisana et al., 2005). Additionally, in 2004, nearly 41% of women attending antenatal clinics in KwaZulu-Natal were HIV positive (South African Department of Health, 2005). The possibility of HIV/AIDS infection is recognized as a significant risk in partner choice by young people in KwaZulu-Natal (Bernardi, 2002, Varga, 1997), and a potential cue to HIV infection is low body weight. HIV/AIDS infection causes a characteristic loss of body mass, called wasting syndrome (Cohan, 1994, Kotler & Grunfeld, 1995). The timing of the onset of wasting can be quite variable, and if the onset occurs later in the course of the disease, it can be accompanied by other symptoms. However, moderate-to-severe weight loss can also occur before the onset of other symptoms (Polsky, Kotler, & Steinhart, 2004). Thus, a lower body weight may signal HIV infection (either on its own or in conjunction with other symptoms) and could reinforce a mate strategy that favors heavier bodies.

Additionally, a higher preconception maternal BMI is correlated with a higher birth weight for the resulting child (Baker et al., 2004, Bhargava, 2000, Mohanty et al., 2006); this may convey an important advantage given that one of the most common causes of death for newborn children and for children below 1 year old is low body weight (Statistics South Africa, 1998). However, this must be balanced with the fact that a higher maternal BMI is also linked to reduced fertility and increased problems during pregnancy (e.g., Gross et al., 1980, Lake et al., 1997). A higher BMI is also linked to a range of health problems including diabetes, hypertension, cardiovascular disease, and stroke (Manson et al., 1995, Willet et al., 1995). However, in the conditions that prevail in KwaZulu-Natal, the positive features of a higher female BMI may outweigh the potential dangers, which may explain the pattern of attractiveness ratings by the SA Zulus.

In contrast, the situation is very different in the UK. The prevalence of HIV/AIDS is comparatively low (European Centre for Epidemiological Monitoring of AIDS, 2003, UK Collaborative Group for HIV and STI Surveillance, 2005), and the general mortality rate in young women is very low. The primary killer for this group is cancer (particularly breast cancer, as well as cervical and ovarian cancer; Brock & Griffiths, 2003, Office for National Statistics, 2003). A lower BMI is also associated with lower levels of cancer (Calle et al., 2003, Garfinkel, 1985), and a lower BMI value is also associated with general long-term female health (Manson et al., 1995, Willet et al., 1995) and reproductive potential (Frisch, 1988, Lake et al., 1997, Reid & Van Vugt, 1987, Wang et al., 2000). This is true not only for the general population but also for African-Britons (Cappuccio et al., 1997, Chaturvedi, 2003). Additionally, BMI may be an indicator of SES. For a given age group, BMI is inversely proportional to nutrition and SES (i.e., a higher BMI is correlated with poorer nutrition and lower SES; Darmon et al., 2002, James et al., 1997, UK Department of Health, 2003). Thus, there are a number of reasons, based on health, fertility, and SES, which might influence a UK observer to favor a lower body weight.

An alternative explanation may be that the average BMI and pattern of fat distribution may differ between the populations in KwaZulu-Natal and the UK, with consequent differences in body size and shape in the two populations. Under such circumstances, it is possible that the different attractiveness preferences reflect the different ranges of BMI and WHR values found in the two environments. It might be argued that one cannot develop a preference for a particular size or shape if it is not present in the environment. For example, Sugiyama (2004) reported that Shiwiar women from Ecuador have a higher average WHR value than reported for women in the Western world and suggests that the attractiveness preferences of the Shiwiar for a higher WHR must be seen in this context. In the case of the current study, the epidemiological evidence suggests similar average female WHR values in the UK and KwaZulu-Natal populations. For example, the South Africa Demographic and Health Survey reports an average WHR of 0.83 (S.D.=0.13) for adult women from KwaZulu-Natal, and in the UK, our analysis of data from the Health Survey for England for women in the same age group shows an average WHR of 0.82 (S.D.=0.07), and specifically for African-British women, the average WHR is 0.81 (S.D.=0.17; UK Department of Health, 2003). However, this does not preclude the possibility that there are differences in body fat distribution that are not captured by WHR, which may influence attractiveness preferences. The average BMI values from the two populations are also similar [28.5 (S.D.=9.86) for Zulu women, 26.8 (S.D.=5.44) all women in the UK, and 28.8 (S.D.=9.57) for African-British women as a specific group] (South African Department of Health, 1998, South African Department of Health, 2000, UK Department of Health, 2003).

For attractiveness preferences to be adaptive, they also have to be flexible so that when conditions within an environment change or when someone moves between environments, his or her preferences should alter to reflect this change in the local optima for survival and reproduction. Hence, when someone moves from a rural SA environment (where risk factors for disease are highly correlated with low body weight) to the UK (where the opposite is the case), one would expect that his or her preferences would shift to reflect these changed priorities. Our data are consistent with this hypothesis. Eventually, over time, the preferences of the migrant population should change to become similar or identical to those of the indigenous population, which would be consistent with our data on the preferences reported by Britons of African origin.

The fact that Zulu migrants into Britain seem to have modified their preferences and the fact that the Britons of African origin have the same preferences as the UK Caucasians imply that these preferences are part of a flexible behavioral repertoire, acquired through social learning, which allows humans to adapt and respond to changing conditions within an environment or when moving between environments. In this case, the change in preferences is likely to be mediated both by social interactions and the media environment. The Western media emphasize a preference for a slim and shapely body as the female ideal, and the ubiquitous nature of this ideal in television, magazines, and advertisements is likely to accelerate the putative changes to attractiveness preferences (Barber, 1998, Silverstein et al., 1986, Tovée et al., 1997).