The spandrels of Santa Barbara? A new perspective on the peri-ovulation paradigm

Numerous recent studies document peri-ovulatory increases in women’s physical attractiveness and in their preferences for certain male traits, which appear to be linked to cyclical fluctuation in levels of ovarian hormones. Changes in physical traits are subtle, leading to the widespread conclusion that cues of ovulation are adaptively concealed. It has been argued that concealment enables women to pursue a mixed-mating strategy characterized by pair-bonding with opportunistic extrapair copulation with high-quality mates when conception is possible. Men’s preferences for peri-ovulatory cues and women’s cyclical preference shifts are thus also argued to reflect distinct psychological adaptations. Here, we re-interpret the evidence for each of these putative adaptations and argue that cyclic changes need not result from incomplete concealment of ovulatory status. We also emphasize that ovarian hormone levels underpin between-individual differences in both women’s attractiveness and their mate preferences, which influence the sexes’ mate-choice decisions. We contrast the salience and efficiency of these within- and between-individual effects, and suggest that, although both effects have the potential to influence mating behavior, the between-individual effects are stronger and more robust. This introduces the possibility that both men’s increased attraction towards peri-ovulatory women and women’s cyclical shifts in preference for certain male traits may not constitute adaptations per se, but are, instead, inevitable by-products, or "perceptual spandrels," of putative adaptations related to between-individual differences in reproductive potential. Our framework parsimoniously explains several observable phenomena and potentially could transform the way evolutionary researchers interpret changes in behavior associated with the menstrual cycle.


INTRODUCTION
Evolutionary psychologists use a Neo-Darwinian framework to provide answers to questions concerning the origin of specific psychological mechanisms. Central to this endeavor is the concept of adaptation. This term can be used in 2 senses. First, it is employed to characterize the process by which nonrandom or unequal reproduction, driven by specific heritable traits of an organism (i.e., the process of natural selection), results in organisms displaying some kind of fit to the environment. Second, it is used to describe the end-products of this process, that is, traits that have evolved as a direct consequence of the fitness benefits they conferred on their bearers. In this paper, we refer only to the latter sense. Despite the centrality of adaptation in evolutionary thinking, there is a long-standing debate regarding the criteria by which a trait can be convincingly labeled as such, because traits that resemble adaptations may in fact originate via different evolutionary processes. For instance, Gould and Vrba (1982) introduced the concept of exaptation, which characterizes traits that evolved to fulfill 1 function but subsequently become beneficial for a different purpose.
In their (in)famous paper, Gould and Lewontin (1979) criticized researchers for concluding too readily that certain traits and behaviors were adaptations. An apparently adaptive trait is not always an adaptation, they argued, but may sometimes be a spandrel; an inevitable byproduct of the development of another adaptive trait, without itself being a direct product of selection, perhaps due to gene pleiotropy, linkage disequilibrium, or developmental constraints. Note that spandrels may also be subsequently selected for new functions. Gould and Lewontin (1979) employed an analogy from architecture (a spandrel is the triangular area between supporting arches that are produced as an inevitable by-product of arch formation, but which thereby provides a surface that could then be exploited for decoration), specifically from the Basilica di San Marco in Venice (but see Houston 2009 for a critique of this architectural metaphor). They further argued that it is important to differentiate between adaptations, exaptations, and spandrels if one is interested in explaining the origins of a trait and how it may play a role in current behavior. In many cases, this is a difficult task because it involves traits, which are the end-products of past selection; hence, we must infer the nature of previous selection pressures when attempting to reconstruct their origin.
In the case of psychological mechanisms, in particular, we face another disadvantage: these do not fossilize and we therefore cannot trace the temporal order of their origin. However, Andrews et al. (2002) provide a comprehensive analysis of approaches to identifying adaptation in the context of behavioral science, arguing that the main evidential burden is whether the trait displays features of special design: that is, it should be possible to demonstrate that the trait evolved to perform a specific function. Relevant criteria vary according to the nature of the trait under examination (e.g., morphological and cognitive traits may require different criteria), but include features like proficiency, efficiency, and specificity of the trait in question. Surprisingly, however, behavioral scientists have rarely employed this approach as a guideline for analysis of specific traits (for an exception, see Williams and Taylor 2006).
With the rise of evolutionary thinking in psychology, Gould and Lewontin's (1979) distinction between adaptations and spandrels, and the difficulty of identifying adaptation, have once again been brought to the fore. Although most researchers likely understand Gould and Lewontin's distinctions and recognize that, as Williams (1966) noted, adaptation is an onerous concept, it is apparent that empirical work too rarely pits competing hypotheses against each other in order to distinguish adaptationist explanations from by-products/spandrels. This applies to many evolutionary subdisciplines, but arguably it is especially true of the strongly adaptationist research program pursued by evolutionary psychologists within the so-called "Santa Barbara school" (Bolhuis et al. 2011;Schulz 2012); here, in line with others, we use this term as a label for a particular way of reasoning, rather than a reference to a particular academic affiliation. The existence of a psychological adaptation is often inferred on the basis of a putative ancestral fitness benefit, and then tested by investigating whether particular behaviors that are elicited under various kinds of experimental (and less frequently, naturalistic) conditions are consistent with the existence of such an adaptation. Demonstrating the potential benefits of a trait is certainly an important and necessary step; however, as we have outlined, it may be insufficient evidence to conclude that a trait represents an adaptation, at least until it has been tested against alternative explanations for its evolutionary origin. To be clear, we do not wish to claim that evolutionary psychologists have entirely failed to consider nonadaptationist interpretations; rather, we are simply stressing that by-product explanations are not given quite the same level of empirical attention and are often not put to the test in the same systematic fashion (indeed, nonadaptationist explanations are frequently represented by a cursory note in a manuscript's discussion; for a notable exception, see Thornhill and Gangestad 2008).
Here, we apply the approach advocated by Andrews et al. (2002) to re-examine a suite of hypothesised adaptations involved in human mating behavior associated with the menstrual cycle. We first briefly address the question of whether cyclical changes in women's appearance are consistent with the interpretation of an adaptation for concealment of ovulation. Although many researchers conclude that these cyclical changes in appearance are by-products of co-occurring physiological fluctuations, we argue that the subtlety of the resulting changes in appearance need not be explained by selection for concealment.
We then address 2 intensively studied facets of human mating psychology that many researchers also consider to be psychological adaptations: 1) the proclivity for men to discriminate women's fertility status from physical and behavioral cues and 2) an increased preference in women for specific male traits around ovulation. In common with many researchers, we agree there might be selective benefits to both phenomena. However, we propose that, rather than being adaptations, both facets might be inevitable by-products (or perceptual spandrels) of more generalized hormone-dependent mate-choice mechanisms.

CONCEALED OVULATION AS AN ADAPTATION
Based primarily on comparisons between humans and their closest extant relatives, chimpanzees, women historically have been considered to show concealed ovulation (e.g., Symons 1979). This is because, unlike humans, female chimpanzees undergo conspicuous cyclical changes in perineal morphology that index current fertility in a probabilistic manner and are attractive to males (Wallis 1992;Nunn 1999). Earlier hypotheses on this topic often assumed conspicuous cyclical changes to be the ancestral state for humans (e.g., Burt 1992). More recent phylogenetic analyses, however, suggest this may not be the case (Sillen-Tullberg and Moller 1993;Nunn 1999;Duda and Zrzavý 2013). Indeed, the last common ancestor of chimpanzees and humans may have been characterized by very moderate or even inconspicuous cyclical changes in morphology. Nonadvertised ovulation might therefore be due simply to differences in selective pressures acting on different primate lineages, a by-product of bipedality (Pawlowski 1999) or perhaps something not obviously related to sexual behavior, such as the presence of large adrenal glands (reviewed in Thornhill and Gangestad 2008).
It remains the case, however, that many researchers see human ovulation as concealed. Functional hypotheses for concealed ovulation include the promotion of paternal care (Strassmann 1981) or paternity confusion (Hrdy 1979) and argue that concealment enables women to engage in extrapair copulations with males of high genetic quality while retaining paternal investment of a primary bonded partner, and that this is especially likely to occur if the latter is of relatively low genetic quality (e.g., Gangestad and Thornhill 2008;Thornhill and Gangestad 2008;Haselton and Gildersleeve 2011;Gangestad and Haselton 2015). Thornhill and Gangestad (2008) provide an extensive review of the evidence for concealed ovulation as an adaptation. They describe 3 key features that are thought to characterize concealment: 1) females should not alter their behavior in relation to their estrous/menstrual phase; 2) sexual desire should be decoupled from the underlying hormonal changes governing the cycle, and 3) physical and behavioral changes associated with ovulation should be greatly reduced or absent. Although they concede that none of these requirements appear to be strictly met in humans, they conclude that human ovulation is indeed concealed. They base this conclusion primarily on the third of these requirements, arguing that, because physical and behavioral cues are not easily detectable, this likely means that females have been under direct selection to suppress them. This reasoning leads to the concept of "leaky cues," such that the ability to entirely suppress the hormone-dependent physical changes associated with ovulation are constrained by the necessity of maintaining reproductive function (see also Gangestad and Thornhill 2008;Haselton and Gildersleeve 2011;Gangestad and Haselton 2015).
Irrespective of whether there has been recent selection for concealment in the human lineage, or whether inconspicuous ovulation has been retained from the last chimpanzee-human common 1250 Downloaded from https://academic.oup.com/beheco/article-abstract/26/5/1249/242042 by guest on 28 July 2018 ancestor, most scholars agree that cyclical fluctuations in women's physical traits have not been selected for specifically (i.e., they are not "signals"; for an alternative view, see Guéguen 2012), but instead are by-products of the hormonal changes that underpin ovulation (i.e., they are "cues"; Pawlowski 1999; Thornhill and Gangestad 2008). With regard to these cues, the distinction between nonadvertised ovulation (or "undisclosed" ovulation: Thornhill and Gangestad 2008) and concealed ovulation rests critically on whether or not selection has acted to suppress ovulatory cues directly. Thus, if ovulation is simply nonadvertised, then these physical and behavioral changes are just "cues." By contrast, if ovulation is concealed, these changes must be viewed as "leaky cues." Which interpretation is correct? We do not necessarily wish to argue that ovulation is not concealed in humans; there is certainly evidence that could be used in favor of this hypothesis (see review in Thornhill and Gangestad 2008). However, we suggest that the subtlety of the physical and behavioral changes across women's menstrual cycles can plausibly be interpreted as simple by-products of the underlying hormonal changes that occur without resorting to the claim of an adaptation in women to suppress these cues. This view is supported by the fact that the other 2 foundational features of the concealment hypothesis are not met. Furthermore, we think it is interesting to note that there is a growing body of literature showing that women behave in ways that attract more attention around ovulation, for example, by choice of more revealing or attractive clothing (Haselton et al. 2007;Beall and Tracy 2013). Is this a means of attracting the attention of potential extrapair partners, as it has been interpreted, even if it also tends to increase jealousy and proprietary behavior in a partner? Or could it be that women simply choose their clothing in response to increased courtship motivation linked to their current hormonal profile? We do not know the answer, but this form of "advertisement" would appear to be directly at odds with the concept of adaptive concealment.

IS CURRENT FERTILITY DETECTION REALLY AN ADAPTATION?
A separate question from how cyclical changes evolved is whether other people, particularly men, possess a specific adaptation to respond to cycle-related cues. We begin by briefly summarizing the evidence that has amassed over recent years.
Studies have shown that women's physical attractiveness, assessed in several sensory modalities, fluctuates across the menstrual cycle, with a peak in the late follicular phase, thus potentially providing cues to current fertility. For example, facial photographs taken during the follicular phase receive higher attractiveness ratings than those of the same women in the luteal phase (Roberts et al. 2004). Cue discrimination may be based on changes in soft facial tissue (Bobst and Lobmaier 2012) or a combination of shape and skin texture cues . Similarly, recordings of women's voices during the follicular phase are rated as more attractive than during the luteal phase (Fischer et al. 2011;Puts et al. 2013), likely due to elevated voice pitch (Fischer et al. 2011). Attractiveness of axillary (Kuukasjarvi et al. 2004;Havlíček et al. 2006;) and vaginal (Doty et al. 1975) odors follows the same pattern. These changes may be underpinned by fluctuations in levels of ovarian hormones. For example, progesterone levels negatively predict facial attractiveness, and interactions between estradiol and progesterone predict voice attractiveness (Puts et al. 2013).
Cyclical variation in physical cues is mirrored by evidence from behavioral displays (Cantú et al. 2014), gait (Fink et al. 2012), and, as mentioned above, choice of clothing style (Haselton et al. 2007) and color (Beall and Tracy 2013). Together with physical changes, these behavioral cues may also influence men's attractiveness judgments of women (Miller et al. 2007;Cobey et al. 2013), although clothing choices are dynamic, likely to be culturally dependent, and might be consciously deployed according to social context .
Some researchers therefore argue that male attention to such cyclical fluctuations represents a psychological adaptation that 1) increases their chances of conceptive sex and/or 2) serves to intensify mate-guarding behaviors around ovulation and thus avoid the costs of cuckoldry (e.g., Gangestad and Thornhill 2008;Thornhill and Gangestad 2008;Haselton and Gildersleeve 2011).
The idea that men's ability to detect cues of current fertility represents a psychological adaptation has intuitive appeal because the potential benefits are clear. However, we can employ the approach of Andrews et al. (2002) to re-examine this claim more closely. For example, a defining characteristic of an adaptation is its efficiency in achieving the associated fitness-related benefits. When considered from this perspective, it is immediately obvious that this putative adaptation is not particularly efficient. The physical changes in women's appearance around ovulation are almost beyond our capacity to measure, despite the aid of sophisticated technologies. For instance, a study examining changes in women's voices found that even detailed acoustic analysis of vocal parameters precluded unequivocal identification of the follicular phase (Fischer et al. 2011).
These low-amplitude fluctuations in the cycle-associated cues are matched by low proficiency in detection of ovulation by perceivers. In experimental tests using paired stimuli obtained under the same conditions from the same woman during the follicular and luteal phases (e.g., Figure 1), fertile stimuli were selected as most attractive only slightly above chance (~55% for faces: Roberts et al. 2004; 60% for clothing: Haselton et al. 2007). Although it has been argued that cyclical shifts in some modalities (e.g., body odor) are relatively large , effect sizes from any such experiment likely over-estimates the actual effect: simultaneous presentation of paired stimuli from the same woman never actually happens in real life. Actual cycle-related attractiveness judgments are, by definition, temporally distributed and, given low discrimination rates even in laboratory experiments, the extent to which they affect perception in real-world settings is questionable, even for long-term partners (Roberts et al. 2004).This might also explain why cross-sectional (in contrast to within-subject) studies frequently fail to find cyclical changes (e.g., Thornhill and Gangestad 1999).
To illustrate this point, consider the following thought experiment: imagine entering a room full of adult women and attempting to detect those near ovulation. We suspect that few people would consider themselves able to do this reliably, unless equipped with sophisticated devices such as hormone kits or transvaginal ultrasound equipment. If our sensitivity to detect leaking gas were as good as our ability to detect ovulation, the result would be a host of unwanted deaths.
As we have described, proponents of the concealment hypothesis argue that such cyclical changes are expected to be very subtle, because selection has directly acted on women to hide them. An alternative adaptive explanation is that their effects might be restricted to men who interact on a daily basis with particular women, such as their partners (e.g., Roberts et al. 2004). In this sense, the mechanism allows men to assess variation in fertility once they are partnered and does not function as a mechanism that assists in selecting potential mating partners in the first place. Although such effects are plausible (see e.g., Haselton and Gangestad 2006 for cyclic fluctuations in male mate guarding), to our knowledge, there are no studies available that explicitly compare ratings made by women's partners against those of men who are unfamiliar with a particular woman. Instead, studies on menstrual-cycle-related attractiveness almost always involve men unknown to the women and make the assumption that they are testing realistic mate-choice (i.e., decision-making) processes.
This emphasis on decision-making has led to a focus on research that uses closely controlled experiments that test whether or not men can "detect" signs of ovulation (for a review, see Haselton and Gildersleeve 2011). We have nothing against such studies in principle: they are both interesting and insightful, but they do not necessarily reflect real-world partner choice. Indeed, studies in other animals suggest that the limits of detection of variability in traits such as symmetry, assessed in laboratory-measured tests, renders such traits as ineffective for discriminating natural variation (e.g., Swaddle 1999;Swaddle et al. 2008).
Against this background of evidence, we acknowledge that a possible adaptation for concealment of ovulatory state, leading to "leaky cues," might be the reason for relatively low efficiency of male preferences for current fertility cues, and that, as Thornhill and Gangestad (2008) suggest, in an evolutionary arms race between women's adaptive suppression of ovulatory cues and men's adaptations to detect them, women have the upper hand. We argue, however that an antagonistic coevolutionary race between these 2 adaptations is not the only plausible interpretation for the low efficiency of the male current fertility-related preferences. We now turn our attention to providing an alternative explanation for this phenomenon.

CUES OF POTENTIAL FERTILITY
In contrast to our ability to detect cues of ovulation, we consider ourselves quite proficient at judging individual variation in attractiveness. The concept of beauty appears to be a human universal. People readily agree on who is attractive and who is not (Langlois et al. 2000), and judgments are often consistent across people from different cultures and ethnicities (Jones et al. 1995). Facial attractiveness is evident even to infants (Langlois et al. 1991). Put another way, neither men nor women would have much trouble dealing with the task of picking out attractive women within a busy room.
Proficiency in detecting attractiveness also implies efficiency in its communicative function. Factors underlying between-individual variation in attractiveness are well-documented. In the face, for example, attractiveness is mediated particularly by symmetry, sexual dimorphism, averageness, and health (Gangestad and Scheyd 2005;Rhodes 2006). Most researchers agree that these traits are considered attractive because they reveal individual variation in mate quality leading to potential benefits in terms of reproductive success (Jokela 2009;Pflüger et al. 2012, but see Silva et al. 2012), and that preferences for these traits have been favored by selection. In terms of women's attractiveness to men, physical traits such as facial femininity (Smith et al. 2010), gynoid fat distribution, waistto-hip ratio (WHR; Singh 2002), vocal characteristics (Brunings et al. 2013), and breast shape (Jasieńska et al. 2004) are considered especially important, although their relative significance may vary across cultures. These traits develop during puberty under the influence of steroid hormones, particularly estrogen, and in adults, they correlate with estrogen levels (Singh 2002;Law Smith et al. 2006;Grillot et al. 2014). For example, individual variation in facial attractiveness is correlated with late-follicular estradiol levels (Durante and Li 2009), women with larger breasts and smaller WHRs have higher sustained estradiol and progesterone levels across the cycle (Jasieńska et al. 2004), and ratings of women's bodies (Grillot et al. 2014) or faces and bodies (Durante and Li 2009) are positively predicted by differences in their estradiol levels ( Figure 2). Selection on preferences for these traits is likely to be strong, because hormone levels provide cues to a woman's future, Variation in women's attractiveness across the menstrual cycle. (a) Face images are shown of the same woman in the follicular and luteal phase (from Roberts et al. 2004). In (b), the same image of 1 woman has been transformed to display specific shape and skin texture changes that occur across phases (from Bobst and Lobmaier 2012). Follicular images are judged more attractive, but within-cycle differences are subtle compared with between-individual variation (a vs. b). (c) Mean ratings of axillary odor samples of 12 women. Follicular samples were assessed by male raters as smelling most pleasant, but there is greater between-individual variation in ratings (from Havlíček et al. 2006). or potential, fertility, at least over immediately forthcoming cycles (Lipson and Ellison 1996;Venners et al. 2006) and potentially over the lifespan. On this basis, the psychological mechanisms for discriminating attractiveness of potential partners are also considered to be adaptations (e.g., Thornhill and Gangestad 2008). For balance, it should also be noted, however, that our understanding of associations between ovarian hormones and female physical attractiveness is currently far from settled (see e.g., Puts et al. 2013). This may partly be attributed to the methodology used. First, most studies correlate an individual body trait (e.g., WHR) with levels of a particular hormone (e.g., estradiol). However, because potential fertility is influenced by the concerted action of several hormones, it might be more appropriate to assess combined effects (e.g., progesterone to estrogen or estrogen to testosterone ratio; Johnston and Franklin 1993). Second, for convenience, many studies use saliva samples, which provide measures of levels of free hormones. However, levels of free hormones might be obscured by body mass index because fat tissue is negatively associated with sex hormonebinding proteins; studies employing blood samples to assess both free and bound hormone levels likely provide more valid data (for a full discussion, see Grillot et al. 2014). Finally, development of physical attractiveness-related characteristics may depend mainly on levels of hormones during puberty and subsequently become relatively stable. Although it is frequently assumed that pubertal levels of hormones and their adult levels are highly correlated (e.g., Law Smith et al. 2006), this might only be partly true and cross-sectional studies with adults may thus underestimate this effect.

THE PERCEPTUAL SPANDREL HYPOTHESIS
Most scholars tend to view 1) current (i.e., whether a given woman is presently in the fertile phase of her menstrual cycle) and 2) potential (i.e., a given woman's overall likelihood of being able to conceive) fertility-related contributions to attractiveness as 2 independent adaptations (e.g., Thornhill and Gangestad 2008). At first glance, this seems reasonable: one describes variation in attractiveness within individuals, whereas the other relates to between-individual variation. However, both sets of judgments hinge on the same underlying mechanism, namely levels of ovarian hormones, and could thus be expected to be strongly interlinked (see Figure 3a). Understanding the evolution of preferences for such traits thus depends on an objective assessment of the relative selection pressures on the ability to discriminate current and potential fertility.
As described above, the low proficiency of males to detect current fertility indicates that selection on men's preferences for cues of high potential fertility may outweigh selection pressures on sensitivity to the physical changes related to current fertility. Further support for this contention comes from contrasting the levels of within-cycle and between-individual variation in studies of women's faces and axillary odor: between-individual variability in facial appearance is higher than within-cycle changes (Figure 1a,b). The same is true of variability in odor attractiveness (Havlíček et al. 2006; Figure 1c): although, on average, women's follicular phase odors were judged most attractive, the odor of many women during the menstrual and luteal phases was judged as more attractive than the odor of other women during the fertile phase.
Thus, even if ovulation detection is an adaptation, it appears to have a relatively small effect on men's behavior compared with their assessment of between-women variation in attractiveness. This, along with the similarities in underlying hormonal mechanism, raises the possibility that, rather than being an adaptation (again, defined by being a trait that evolved as a direct consequence of the benefits they confer), discrimination of ovulation-linked cues might arise as a by-product of a generalized preference among men for cues of high estrogen levels (or low progesterone to estrogen ratio) that discriminate between women in terms of potential fertility. The arguments presented by Gould and Lewontin (1979) with respect to by-products thus apply here. Although Gould and Lewontin focused on morphological traits, it is important to emphasize that such by-products are not restricted to morphology and their analogy can also be applied to psychological functioning (Gould 1991;Buss et al. 1998;Andrews et al. 2002). Thus, we suggest that perception of cyclical changes in physical attractiveness could be an example of a psychological, or more specifically, a perceptual spandrel. Note that this fulfills the criteria for an evolutionary by-product as recently discussed by Park (2007). In his critique of the loose use of this term, he proposed that, in order to avoid confusion, the term "by-product" should not be used either to describe behavioral outcomes or effects of particular psychological mechanisms that arise as a consequence of novel inputs to the system, but should be applied only to the mechanisms themselves.  Although phrased somewhat differently, Roney (2009) reached a similar view. He argued that apparent preferences for female cues of ovulation are consistent with a general mechanism for cues of chronically high estradiol in women. He further suggests that this preference may be derived, in functional terms, from benefits gained through discrimination of differences in ovarian hormone levels between ovulatory and nonovulatory cycles, rather than from the benefits gained through discriminating within-cycle variation. We think this is a persuasive and underappreciated point.
Indeed, we would go further: the argument can also be extended to between-individual differences in ovarian hormones levels. For instance, because estradiol levels vary between individuals, and these differences underpin between-individual differences in attractiveness (Figure 2), it is plausible that men's attention to estradiol-linked physical traits when making between-woman attractiveness judgments may also explain estradiol-linked within-individual attractiveness differences, regardless of whether these occur within or between cycles. Attention to ovarian hormone-linked markers of high potential fertility (e.g., facial femininity, body shape) could reap selective benefits in addition to targeting imminent ovulatory or conceptive cycles, because estradiol levels may reflect younger age (Mersereau et al. 2008) or relatively good nutritional status/ condition (Venners et al. 2006;Ziomkiewicz et al. 2008), or be linked to traits like nulliparity and earlier onset of menarche (Windham et al. 2002). Indeed, between-individual differences in estradiol are robust and maintained across the cycle (Figure 3b), providing more stable and reliable cues of potential fertility than associations between estradiol concentration and ovulation (Figure 3). In contrast, 1 recent study found that within-individual changes in facial and vocal attractiveness were related to progesterone levels but between-individual differences were not (Puts et al. 2013). As these authors note, however, they collected hormonal samples at variable points of women's cycles, which could have the effect of underestimating any between-individual effects by a considerable margin. We clearly need more studies to disentangle these issues.

CYCLICAL SHIFTS IN WOMEN'S MATE PREFERENCES
The conceptual framework developed above could also conceivably be applied to the accumulating evidence for changes in women's mate preferences across the menstrual cycle. These preference shifts are characterized by an increased peri-ovulatory preference for traits that appear to be markers of male quality. Such shifts in preferences have been reported for a range of sexually dimorphic physical traits, including relatively masculine faces, bodies, and voices (Penton-Voak et al. 1999;Puts 2005;Little et al. 2007;Jones et al. 2008). Like changes in women's physical appearance, these changes are hormonally mediated. For example, preference changes for masculine male faces or vocal traits are mediated by changing levels of estradiol or its precursors such as testosterone Welling et al. 2007;Bobst et al. 2014).
Currently, there are 2 main suggestions for why these shifts occur. The most influential and pervasive of these is the idea that psychological mechanisms are affected by cyclical hormonal fluctuations to increase attraction to markers of high genetic quality when conception is likely, providing a mechanism by which women maximize the quality of mating partners: the "ovulatory shift hypothesis" . Despite its prevalence and widespread acceptance within evolutionary psychology, this idea remains controversial and heavily contested outside this sphere (e.g., Gildersleeve et al. 2013;Harris 2013). Most recently, this has resulted in the publication of 2 meta-analyses aimed at establishing the robustness of the phenomenon. Far from reaching consensus, these papers produced widely contrasting results. Although Gildersleeve et al. (2014a) found evidence for cycle-shifts mainly in short-term mating contexts,  found no support for cycle-shifts influencing female mate preferences. Given that each research group was dealing with a largely overlapping body of literature, these results seem rather puzzling at first. On closer inspection, the differences are not, in fact, surprising.
First, although the sample for both meta-analyses did indeed overlap ( Stricker et al. (2006). Dashed lines represent reference levels for the mid-luteal phase (days +5 to +9 from the luteinizing hormone (LH) peak). In women with the highest estradiol levels, serum concentrations during the mid-luteal phase (red dashed line) are close to those found in women with median levels (black line) even at the estrogen peak, and far exceed those in women with low levels (blue line). (b) Putative estrogen-dependent mechanisms for temporal change in women's attractiveness and expressed mate preference. The schematic depicts typical variation in estradiol levels across 2 menstrual cycles in women with relatively high (H) or low (L) average estradiol. Most previous studies focus on within-cycle (WC) changes. Roney and Simmons (2008) refer to differences in mean estradiol between different cycles (BC; means for each cycle are represented by green dashed lines). Here, we depict cycle 2 in the low-estradiol example (L) as an exposed nonconceptive cycle with reduced estradiol levels (see Grillot et al. 2014). Mean estradiol levels between individuals (BI) are depicted with solid blue lines. data from 10 published and 8 unpublished studies unique to their sample, whereas Gildersleeve et al. (2014a) presented unique data from an additional 4 published and 6 unpublished studies. In addition, the 2 studies used different inclusion criteria for the variables selected and structured their analyses differently. Most substantively, Gildersleeve et al. (2014a) conducted a 2-tiered analysis, in which they first investigated whether there was an overall influence of cycle phase on preferences for a broad array of mate characteristics. This was followed by a more restrictive analysis that excluded those traits not widely accepted as cues of genetic quality, as well as excluding studies that either asked women to state their preferences (rather than directly rate stimuli for attractiveness) or which used stimuli that did not allow for any direct assessment of the trait in question (e.g., when vignettes describing visual characteristics were used, rather than using visual stimuli). By contrast, the analyses of  consisted of testing for cycle-shifts for each mating characteristic separately and did not combine effects across traits. Finally, Gildersleeve et al. (2014a) split their sample by relationship context (short-term, long-term, and unspecified), whereas  did not differentiate their sample in this way.
Given these differences in approach, the contrasting results are more easily explained. Indeed, the debate continues more or less unabated (Gildersleeve et al. 2014b;Harris et al. 2014;Wood 2014;Wood and Carden 2014). This is perhaps to be expected, however: as Ferguson (2014) points out in his commentary on this debate, meta-analyses rarely solve scientific disputes, because differences in theoretical outlook, along with the stance taken on a particular issue, inevitably influence the manner in which data are selected for inclusion in such analyses and how they are used to test a specific hypothesis. We feel this is a key point to make, but on slightly different terms. That is, in our view, part of the problem lies in characterizing within-individual preference shifts as adaptations when they should be considered as by-products of more general differences that exist between women. Theoretical reconceptualization along these lines may therefore pay dividends with respect to the design of future empirical work, rather than continued attempts to interrogate the existing data and establish some kind of consensus on the size and robustness of the effects.
To expand on this just a little, perhaps the most interesting aspect of the current meta-analyses, in our view, is the size of the effects found and the variability in these across different traits; it seems that these do not just represent variability in the rigor of the methods and the definitions employed in such studies, but tell us something about the phenomenon itself. For example, Gildersleeve et al. (2014a) show that, although there was a significant overall effect of cycle phase on women's mate preferences, and a significant effect of short-term relationship context, the effect sizes were small to moderate (broad sample: overall, g = 0.15; short-term relationships, g = 0.21; narrow sample: overall, g = 0.17; short-term relationships, g = 0.26). Small effect sizes also characterize the analyses for individual traits, and there was also considerably more variability in the associations found (e.g., significant or marginally significant effects were found in the short-term context for body masculinity (g = 0.35) and behavioral dominance (g = 0.19), but not for facial symmetry (g = 0.3) or facial masculinity (g = -0.02)). In addition, Gildersleeve et al. (2014a) showed that, after controlling for relationship context, higher effect sizes were associated with a between-participant rather than a within-participant design. These authors point out that there are many potential unmeasured confounds and, as a result, it is difficult to know true effect sizes. In addition, the relationship between effect size and the presence of an adaptation is unlikely to be linear. We sympathize with this view and note that true effect sizes might be heterogeneous and, in different cases, either under-or over-estimated. Given the strong public interest in shifts in mating behavior across the menstrual cycle, it is important for scholars to provide context and convey what, if any, "real-world" differences exist across the cycle. As we see it, however, the variable and elusive nature of the phenomenon is precisely what one would expect from a by-product explanation: that is, variability is not the consequence of experimental/measurement error and noise, but an inherent characteristic of the phenomenon under study. Thus, rather than speaking to a lack of methodological rigor within the current paradigm, as some suggest (Harris et al. 2014;Wood and Carden 2014), this variability and inconsistency might be more productively viewed as a mischaracterization of the nature of the effect: the effects of individual within-cycle shifts should, in fact, be small, statistically weak, and variable if they prove to be the by-products of more robust between-women differences.
The same point can be made for the finding that, in both sets of analyses, stronger effects were found for those studies that specified a broad window of fertility (i.e., of around 9 days as opposed to 6; Harris et al. 2014;Wood and Carden 2014). As Gildersleeve et al. (2014b) point out, this does not necessarily indicate less precision in the measurement of the fertile phase, as argued by Wood and Carden (2014) and Harris et al. (2014). Indeed, Gildersleeve et al. (2014b) argue that using a 9-day window is a superior approach as it correctly captures fertile days of women with cycles longer than 28 days. Although this is correct, it is also true that the number of women having shorter cycles relative to those having longer cycles is relatively equal, which is why 28 days is usually chosen as an "average cycle length" (Münster et al. 1992). Consequently, the use of a 9-day window may misclassify a similar number of women with shorter cycles as being fertile when they are not. In addition to issues of precision, however, stronger effects with a 9-day window might also suggest that shifts in preference are associated with levels of estrogen or other ovarian hormones (or their ratios) which fluctuate across the entire follicular phase of the cycle, rather than with the fertile window specifically; something that, again, would be in line with a by-product explanation, rather than one based on within-cycle shifts themselves being a target of selection.
The other explanation offered for why within-cycle shifts occur, proposed by Roney, is that shifts in preference are underpinned by differences in estrogen levels between different kinds of menstrual cycles, specifically between ovulatory/anovulatory or between conceptive/nonconceptive cycles (Roney and Simmons 2008;Roney 2009). Again, however, following a similar line of argument to the above, we suggest that any such within-woman changes in preference strength for male traits potentially could be explained by an even more general hormonal effect. That is, the attraction to phenotypic markers of male quality may be regulated by between-woman differences in ovarian hormone levels (for a similar argument, see Dixson 2009, ch. 6). To understand our argument more fully with respect to reconceptualizing both the ovulatory shift hypothesis and Roney's variant on this theme, it is necessary to re-examine the underlying hormonal mechanisms.

INDIVIDUAL VARIATION IN WOMEN'S PREFERENCES
A functional relationship between ovarian hormones and expression of preference for high-quality male traits (evidenced by studies of within-woman preferences) would also be expected to shape the 1255 Downloaded from https://academic.oup.com/beheco/article-abstract/26/5/1249/242042 by guest on 28 July 2018 level of demand for attractive characteristics among women who vary in average levels of ovarian hormones. Although direct evidence is currently lacking, there is persuasive indirect support for this prediction. For example, as estradiol levels predict betweenindividual differences in facial and body attractiveness (Figure 2), high estradiol women have high "mate value." Such women may be relatively selective (Parker 1983), choosing and retaining male partners with the most attractive characteristics. For example, assortative mating for attractiveness is one of the most striking and robust findings in human mate choice (Berscheid et al. 1971). In terms of specific attractiveness components, women's self-rated attractiveness (which is strongly correlated with other-rated attractiveness) predicts stronger preference for male facial masculinity and symmetry (Little et al. 2001) and vocal masculinity (Feinberg et al. 2012). Women with more attractive bodies show stronger preferences for facial cues of male quality (Penton-Voak et al. 2003;Jones et al. 2005;Smith et al. 2009). These findings indicate a relationship between women's estrogen level and their preference for the same kind of characteristics that are also chosen when estrogen peaks in the cycle, around ovulation. They are also consistent with studies in animals demonstrating that female condition influences mate choice (Bakker et al. 1999;Cotton et al. 2006). Although we use the example of variation related to estrogen, links to attractiveness may also be dependent on between-individual variation in other hormones such as testosterone and progesterone or even combinations or interactions of several hormones (e.g., estrogen to testosterone ratio). The precise hormonal mechanisms associated with the preferences described remain a challenge for the future studies. Importantly, differences in female condition also moderate the size of cycle-dependent changes in preferences and behavior. For example, although women express greater preference for masculine voices in the late follicular than the luteal phase, this effect is significantly less pronounced in women with relatively high estrogen levels ). Attraction to male muscularity is similarly more phase-dependent among women of lower self-perceived mate value (Millar 2013). Sensitivity to such changes in male muscularity is enhanced by greater self-evaluation of mate value during the fertile phase, thereby reducing costs of inappropriate partner choice (Beaulieu 2007). This may also influence relationship dynamics: women of lower mate value experience greater ovulatory shifts in mate-guarding (Haselton and Gangestad 2006). These effects were previously interpreted as part of a mixed-mating strategy. More specifically, women of lower mate value are expected to experience difficulties in attracting partners of high genetic quality. As a consequence, it is argued that there should be activation of a contingent adaptation in which such women should be more likely to opportunistically pursue extrapair sex with men of high genetic quality or current condition, while attempting to maintain strategic longterm relationships with the males who provide care and provision to the offspring. Such extrapair sex can be functional only when pursued during the fertile window when conception is possible (e.g., Haselton and Gangestad 2006).
In contrast, we think that the observed patterns might be due to a ceiling effect on preferences among high-estrogen women. Such ceiling effects could lead to lower cyclical variation among highestrogen women compared with the change experienced by women with relatively lower average estrogen levels. Future studies should specifically address these 2 competing interpretations. The mixedmating strategy hypothesis predicts that the cyclical changes in women of lower mate value will be more pronounced in coupled compared with single women, whereas these differences are not expected according to the perceptual spandrel hypothesis. Several previous studies touch partly on this issue, though they focus not on preferences but on actual behaviors. For example, Durante et al. (2008) found that single women, compared with coupled women and those scoring high on a relationship satisfaction scale, revealed more skin during fertile periods of their cycle. A second study found no effect of relationship status on changes in sexiness of clothing style across the cycle and only single women showed cyclic increase in interest for going out (Schwarz and Hassebrauck 2008). None of these findings are predicted by the mixed-mating strategy hypothesis, which would posit such effects in partnered women but not single women.
It has been further argued that the mid-cycle peak in preferences for cues of high-quality partners should be pronounced in shortterm in contrast to long-term mating contexts or unspecified contexts. As shown above, the effect sizes of the cyclic fluctuations in the short-term context in the meta-analysis of Gildersleeve et al. (2014a) varied greatly, being sometimes higher compared with studies in which relationship length was unspecified and sometimes not. We would, however, question the extent to which the concept of a "short-versus long-term mating strategy" is primarily theorydriven rather than data-driven and might have been imposed on participants by researchers. It is possible that such ideas only have salience and relevance in some cultures and not others, and this is an empirical issue that requires further exploration.
In summary, then, although one could postulate 2 independent adaptations-one for within-cycle variation in preferences and a second being sensitive to ovarian hormone-dependent individual quality-a more parsimonious explanation is that selection acting on between-individual expression also produces cyclical variation as an epiphenomenon. The evidence showing that these 2 processes are interlinked is in complete agreement with this view.

PHYLOGENETIC CONSIDERATIONS
Our by-product argument for both male preferences for cues of current fertility and cyclic variation in female preferences is inherently based on assumptions related to phylogenetic sequence. Though we have partly addressed these issues above, here we aim to summarize possible phylogenetic scenarios. The most elaborate scenario was proposed by Thornhill and Gangestad (2008). In brief, their argument was based on the notion that most mammals are only sexually active during a discrete fertile period (commonly coined estrus) and that this was perhaps true of the common ancestor of the anthropoid primates. In contrast, anthropoid primates are sexually active throughout the cycle, even when conception probability is extremely low. These authors argue that the functionality of estrus sexuality has nevertheless been retained in humans. Based on this, Thornhill and Gangestad propose the concept of dual sexuality, which argues that mate preferences when fertile and nonfertile are divergent. Specifically, this idea acknowledges that, although women are receptive to sex throughout the cycle (termed "extended sexuality"), preferences and sexual behavior during the fertile phase of the cycle reflect a suite of adaptations designed to sire offspring with males of high genetic quality (for details, see . Moreover, the ovulatory shift hypothesis argues that women's increased fertile phase preference for highquality males should be specific to short-term, but not long-term, mating contexts. In this view, sexuality during the nonfertile phase of the cycle is thought to serve a different purpose, namely promoting long-term relationships and bonding (see e.g., Gangestad and Haselton 2015 for a recent argument of this kind) or paternity confusion. Several moderating factors to this hypothesis have been proposed, including that women coupled with low-quality partners should show relatively greater cyclical shifts in desire for extrapair men or that there should be larger differences in cyclic shifts between single and coupled women (see above).
In contrast, we argue that the loss of a strict association between sexual behavior and estrus, occurring at the root of the anthropoid primate lineage (for discussion, see Dixson 2009), enabled relatively independent evolution of mate preferences and sexual behavior, the latter of which may have been co-opted for other functions than conception. Due to various environmental, social, and other factors, selection then led to variability in mating systems and preferences across anthropoid primate species (for review, see Dixson 2012). Although (in contrast to the majority of anthropoid primates) humans form long-term pair-bonds or small and relatively stable polygynous groupings, mate preferences would reflect relatively stronger selection pressures on men to select among women based on cues of long-term fertility (Roberts and Havlíček 2013;Stewart-Williams and Thomas 2013). Putative adaptations to longterm pair-bonded relationships would involve adult attachment and romantic love (Fletcher et al. 2015), which have indeed been reported across various cultures (Jankowiak and Fischer 1992). Note that this scenario does not call for the relatively complex series of adaptations associated with the concept of dual sexuality, but only invokes variation in selection pressure for males of different species to differentially value cues of current or potential fertility. Indeed, it could further be argued that the reproductive costs associated with extrapair copulations (e.g., risk of being abandoned with a highly altricial offspring) might be sufficiently high so as to preclude adaptations for dual sexuality to evolve in any case (see Dixson 2009 for a similar critique). However, this issue would benefit from formal mathematical models, which would help to assess the circumstances under which we may expect such adaptations to evolve in our human ancestors.

OBJECTIONS AND COUNTER-ARGUMENTS
In the main body of this paper, we have proposed that discrimination of cues of ovulation by men and cyclical variation in women's mate preferences may both be by-products of between-individual variation in cues and preferences, underpinned by similar hormonal mechanisms. Objections to our proposal could include the following: 1) We have misinterpreted the direction of causality. That is, selection has acted directly on perception of hormonelinked within-individual cues of current fertility and that perception for between-individual differences arose as a by-product. Indeed, it was recently argued that primate sexual swellings originally developed as signals of current fertility and were subsequently co-opted as signals of potential fertility signals (Huchard et al. 2009). 2) Even if preferences relating to current fertility initially arose as a byproduct of traits related to potential fertility, they have subsequently been the target of selection and formed new adaptations (secondary adaptations). 3) Low efficiency in perception of current fertility cues might be a consequence of counter-selection in women to conceal ovulation. Finally, 4) the ability to detect cues of current fertility among humans is not a by-product, nor is it adaptive; rather, it is a vestige of the evolved ability to detect current cues of ovulation shown by our primate relatives.
Based on our comparison of features of special design, such as efficiency and proficiency, the first 2 objections appear unlikely. Furthermore, they are undermined by the fact that estradiol, the predominant and most potent estrogen in reproductively active women, is not in itself a reliable predictor of current fertility. Although it increases during the late follicular phase and triggers the luteinizing hormone surge, it occurs at sustained levels during the luteal phase and even across menstruation, and varies considerably between different women and different cycles (Stricker et al. 2006; Figure 3). This would result in a poor match between timing of ovulation and its associated effects on physical cues and women's preferences, if, as has been argued; these are directly regulated by estradiol. On the other hand, between-individual differences in estradiol are sustained across the cycle and average levels would thus provide relatively stable and reliable cues of potential fertility (Figures 2 and 3).
The third objection would necessitate that physical markers revealing a woman's current fertility were more obvious during our evolutionary history than they are today. There are 2 main versions of this argument. The first version often emphasizes the human-chimpanzee comparison and assumes (though often implicitly) chimpanzee-like cyclic fluctuations in appearance as an ancestral state (e.g., Symons 1979). However, as we previously outlined, recent phylogenetic analyses indicate that conspicuous markers of current fertility are an evolutionary novelty in the chimpanzee lineage, which makes this scenario unlikely (Nunn 1999;Duda and Zrzavý 2013; see also Pawlowski 1999). The second version concurs with phylogenetic reconstructions on the evolutionary novelty of the conspicuous cyclic fluctuations in morphology in the chimpanzee and assumes that cyclic physical changes are by-products of the menstrual cycle physiology in an ancestral species. However, it also requires that, somewhere in the evolutionary lineage of modern humans, selection favored females with less obvious cyclic changes because of the potential advantages for paternity confusion (e.g., Thornhill and Gangestad 2008). This might explain the low efficiency of current fertility-related preferences. Although there is robust evidence showing that ancestral females were unlikely to show conspicuous fluctuations, the phylogenetic studies are uncertain as to whether there was selection for concealment in the human lineage, but our by-product hypothesis does not require such an evolutionary step. Furthermore, as argued above, it is at least possible that the altriciality of human offspring and the high costs of abandonment may have precluded selection for dual sexuality, which in turn makes selection for concealment less likely, though this is definitely a potential area of further research.
Finally, if the preferences for cues of current fertility are merely vestigial, but are not adaptive, and have not been selected against in modern humans, they would not be expected to be linked to the preferences for cues of potential fertility, as proposed by the perceptual spandrel hypothesis. However, as reviewed above, this appears not to be the case.

CONCLUSIONS AND IMPLICATIONS
Whether and how men can detect women's current fertility status and how current fertility influences women's mate preferences have each generated enormous research interest. Within-individual change in hormone levels, particularly estrogen, is the putative mechanism underlying both research paradigms. As we have described, most researchers interpret changes in women's attractiveness or preferences in terms of their current fertility within a single menstrual cycle. Roney and co-workers argue that these changes may be byproducts of variation in estrogen levels across different cycles (Roney and Simmons 2008;Roney 2009). In contrast, we suggest that both kinds of effect might be explained by a general hormone-dependent mechanism based on between-individual differences in women's average estrogen levels. Women with relatively high estrogen levels are likely to have high potential fertility and selection will therefore strongly favor men to attend to estrogen-linked physical cues. Women with high average estrogen are also expected to exhibit higher demand for attractive male traits. The available evidence is consistent with this hypothesis. Furthermore, although Roney's hypothesis can account plausibly for most of the data and requires only a single mechanism to explain different observed effects, our idea additionally captures well-documented differences in attractiveness and preference between different individuals. According to our view, withinindividual changes in women's attractiveness and preferences across or between cycles are most likely by-products, or perceptual spandrels, rather than specific and independent adaptations.
In their defense of adaptationism, Buss et al. (1998) state "…we could not find a single example of an empirical discovery made about humans as a result of using the concepts of exaptations or spandrels." In contrast, we argue that, although between-individual cues of potential fertility and preferences are indeed suggestive of adaptations, the corresponding effects associated with current fertility may well be excellent examples of spandrels. We are aware that this suggestion is controversial and we hope it will provoke researchers to develop tests to confirm or refute it. As Houston (2009) points out, following his analysis of architectural adaptations, the lesson of Gould and Lewontin's (1979) paper is that adaptationist explanations can be dismissed even when there is evidence in their favor; an error we would certainly be at pains to avoid. Fruitful approaches might include direct tests of effect sizes for cyclical changes versus between-individual differences; more comparisons of the relative effects of ovarian hormone variability within/between cycles and between individuals; consideration of dose-dependent effects and interactions among different hormones; more attention to criteria of special design such as efficiency and proficiency; and consideration of the relative selection pressures that operate on markers of current and potential fertility in relation to socio-ecological conditions, including differences in individual motivation brought about by, for example, relationship status. Finally, we hope that our analytical approach will stimulate greater attention to alternative interpretations in other psychological domains.