(a) Correlations between dimensions
Odour pleasantness was strongly correlated with ratings of partner desirability (rs=0.854, n=659, p<0.001). Both pleasantness and desirability were equally and negatively related to perceived odour intensity (session 1, rs=−0.325, n=659 and 658, p<0.001).
Within-session correlations between pleasantness ratings were highly skewed and more positive than expected by chance (women not using the pill: t57=4.82, p<0.001; pill users in session 2: t27=3.62, p=0.001; see the electronic supplementary material, figure S1). Between-session ratings were also correlated in the control group (t59=4.86, p<0.001), but correlations were no higher than expected by chance for pill users (t36=0.66, p=0.515). Between-session repeatability in pleasantness ratings was significantly higher in the control group than the pill group (Wilcoxon test, z=2.26, p=0.024).
Similar patterns were found for odour intensity and desirability (electronic supplementary material, figure S2). However, between-session ratings of odour intensity by the pill group were highly correlated (t36=3.02, p=0.005), indicating that low repeatability was specific to ratings that indicate odour preference (pleasantness, desirability) rather than to changes in women's olfactory sensitivity.
(c) Preferences in normally cycling and pill-using women
To our surprise, we found no significant effect of MHC dissimilarity on odour pleasantness or desirability scores in session 1, where no women were using the pill (p>0.68, ). Intensity ratings of all 110 women showed a tendency for odours of dissimilar men to be rated as stronger, but this non-significant effect was weakened when analysis was restricted to the core sample of British women and shirts not perceived as smelling of tobacco or perfumed products. Across all ratings, there was no correlation between allele sharing and either odour pleasantness (rs=−0.002, n=660, p=0.95), partner desirability (rs=0.013, n=659, p=0.73) or intensity (rs=−0.046, n=659, p=0.24).
Table 1 Mean scores given to three MHC-similar and three MHC-dissimilar male odours by 110 normally cycling women tested during the late follicular phase (session 1). The core sample excludes non-UK women and shirts worn by non-UK men or those that were noted (more ...)
In session 2, where some women were using the pill, we again found no significant differences in any comparison (electronic supplementary material, table S3). Our results therefore suggest that, at least in our sample, there was neither a significant general preference for MHC dissimilarity across normally cycling women, nor a significant preference for MHC similarity associated with pill use.
We checked whether the non-significant effect described above might be owing to the inclusion of a proportion of men who, across the sample, were assessed only under one condition (i.e. only as a MHC-similar/dissimilar man). This might be a problem if the odours of such men were unusual or especially (un)attractive. Of all shirts rated in this experiment, 47 were from men assessed only as MHC similar, and 48 from men assessed only as MHC dissimilar (7% each; the other 86% of shirts were from men assessed by at least one woman in both MHC-similar and MHC-dissimilar conditions). There were no differences in odour pleasantness, desirability or intensity between these men (independent-samples t-tests, all p>0.17). Furthermore, recalculating mean MHC-similar and MHC-dissimilar ratings for each woman, with these men excluded, had little effect on the results (compared with : pleasantness, t109=0.32, p=0.75; desirability, t109=0.24, p=0.81; intensity, t109=1.98, p=0.051; core sample: all p>0.4).
Following previous studies (Wedekind et al. 1995
; Wedekind & Füri 1997
), we next compared scores assigned to male shirts when presented to MHC-similar and MHC-dissimilar women (i.e. men as unit of analysis). In this analysis, we used z
scores (i.e. with a mean of zero and standard deviation of 1) to control for variability in the use of the rating scale by individual women (cf. Roberts et al. 2005b
; full details, also using raw scores, are given in the electronic supplementary material, table S4). In session 1, we found no difference in ratings when individual men's odours were assessed as MHC similar or MHC dissimilar, neither in the entire sample (paired t
=79 men, t
=0.54, 0.35 and 0.93 for pleasantness, intensity and desirability, respectively, all n.s.) nor the core sample (n
=1.06, 0.38 and 0.27, all n.s.; electronic supplementary material, table S4). We found no effects of male heterozygosity on odour pleasantness, intensity or partnership desirability; mean pleasantness/desirability scores were higher for heterozygotes in non-users, and lower in pill users, but these differences did not approach significance (electronic supplementary material, table S5). Following Wedekind et al. (1995)
, we tested for an association between MHC dissimilarity and the number of times women indicated that shirt odours reminded them of either partners or relatives. However, we found no significant effects in either session (electronic supplementary material, table S6).
(d) Changes in relation to pill use
Although we detected no general MHC-associated preferences, we next looked for potential shifts in preferences across sessions. We first calculated a within-session difference score between mean ratings of MHC-similar and MHC-dissimilar odours for each rater, subtracting similar scores from those for dissimilar odours (positive scores indicate preference for MHC-dissimilar odours). We then used doubly multivariate repeated measures ANOVA to test for changes in relative preference for MHC dissimilarity.
We found no significant main or interaction effects when using the whole sample. However, with the core sample, we found a significant session–group interaction (F3,71=3.05, p=0.034), driven mainly by desirability ratings (F1,73=3.63, p=0.061; pleasantness F1,73=0.22, p=0.64; intensity F1,73=0.01, p=0.92). Excluding the one woman who used a progesterone-only pill did not affect the main interaction (F3,70=3.07, p=0.034) but increased the effect of desirability ratings (F1,72=4.19, p=0.044). This interaction () is indicative of a decreasing preference for dissimilarity across the two sessions among the pill-using group and, to a lesser extent, an increasing preference for dissimilarity in the control group.
Figure 2 Mean difference in odour ratings for MHC-similar and MHC-dissimilar men by pill-using and non-pill-using women in two rating sessions (open bar, first test; filled bar, second test). Positive scores indicate preference for MHC-dissimilar odours. (a) odour (more ...)
Finally, we considered the possibility that differential use of the rating scale between sessions might obscure any relevant effects (e.g. women's familiarity or distaste for the odours may have changed as a result of experience in session 1, and might have differed between the pill and control groups). We therefore repeated the analysis using z scores. This made little qualitative difference to the analysis, again showing a significant session–group interaction (F3,64=2.82, p=0.046), driven by desirability ratings (F1,66=4.07, p=0.030; pleasantness F1,66=0.55, p=0.46; intensity F1,66=0.37, p=0.54).
(e) Differences between women
We detected a difference in the use of rating scales between treatment groups, which was evident even in session 1, before the pill group began pill use: mean scores given to all six shirts were higher for the pill group, for both odour pleasantness (t108=3.28, p=0.001) and partner desirability (t108=3.21, p=0.002). However, there was no difference in the ratings of odour intensity (t108=1.19, p=0.238), indicating that the differences for pleasantness and desirability were unrelated to differences in the ability to smell the odours.
We also noted a significant difference in responses from women who were grouped according to whether they reported being in a current relationship. In session 1 (none using the pill), we found a significant relationship status–MHC interaction (F1,83=4.72, p=0.033), such that paired women gave higher partnership desirability scores to MHC-dissimilar men, and single women preferred MHC-similar men (a). The same interaction for odour pleasantness ratings bordered on significance (F1,83=3.92, p=0.051), but there was no effect for odour intensity (F1,83=1.06, p=0.307). Relationship length was unrelated to MHC-odour preference, but among women in relationships, we found a near-significant association between MHC-odour desirability scores (but not pleasantness or intensity) and the frequency with which women reported fantasizing about sexual relationships with other men (F2,45=2.68, p=0.080), such that women who did so more frequently gave higher desirability scores to MHC-dissimilar odours (b).
Figure 3 Effects of relationship status on MHC-correlated odour preferences. (a) Differences in partnership desirability ratings of MHC-similar and MHC-dissimilar male body odours by 42 single and 43 paired women. The interaction is significant (p=0.033). (b) (more ...)
There was no significant association between intention to initiate pill use and current relationship status: 22/41 pill users and 40/72 non-users reported being in a relationship (X2=0.45, p=0.85). Despite this, in view of the effect of relationship status, we repeated the main repeated-measures analysis of pill use on ratings, this time adding relationship status as a between-subjects factor. The results remained qualitatively unchanged: there was a significant session–group interaction (F3,69=2.95, p=0.039), driven by odour desirability (F1,71=3.53, p=0.064), but no significant interactions for session–relationship status (F3,69=0.46, p=0.712), relationship status–group (F3,69=0.45, p=0.717) or session–relationship status–group (F3,69=0.04, p=0.990).
There was no relationship between self-rated attractiveness of women raters and pill use (Mann–Whitney tests; facial attractiveness: U=1254, p=0.82; physical attractiveness: U=1174, p=0.44; both n=39pill and 66non-pill). Self-rated facial attractiveness did not vary among single or paired women (U=1349.5, p=0.90, n=57single and 48paired), but self-assessed physical attractiveness was higher among paired women (U=1056.5, p=0.038). However, tests of MHC preference in the first test revealed no effect of physical attractiveness (entered as a covariate) on preference either in a model without relationship status (main effect, p=0.37; interaction p=0.70) or with it (main effect, p=0.44; interaction p=0.42; the MHC–relationship status interaction remained significant, F1,77=4.17, p=0.045). Including self-rated attractiveness (either facial or physical) as a covariate in the main repeated measures ANOVA across tests only increased the significance of the session–group interaction reported above (F3,66=4.18, p=0.009 and F3,66=3.45, p=0.021, for facial and physical attractiveness, respectively).