In summary, the studies of social status in primates do not appear to provide strong evidence that social status is an important predictor of CAD in monkeys. Rather, they seem to suggest that the relationships between social status and CAD may be sex-specific, and in particular that subordinate status is a more fallible predictor of atherosclerosis than is sometimes assumed. Indeed, in male monkeys it appears that dominant, not subordinate, status may sometimes be pathological. With respect to the studies which have examined the effects of social stress on CAD outcomes, the studies are suggestive, but there are probably too few data at present to come to any firm conclusions.
One study here is of particular interest, as it is widely cited as providing supporting evidence for a pathological relationship between low social rank and poor health 
. In this study monkeys were randomly allocated to 4-member groups for 8 weeks until the social rankings stabilised. Dominant monkeys were then housed together, and the subordinates housed together, forming 4 groups in total: females that were initially dominant, and remained dominant after regrouping (n
11), monkeys that were initially dominant, but became subordinate after regrouping (n
11); those that were initially subordinate, but became dominant (n
8) and those that were initially subordinate, and remained so (n
12). One finding from this study is often cited in support of the psychosocial interpretation of health inequalities, to the effect that among dominants that became subordinate the extent of coronary artery atherosclerosis increased by 500% - which is used as evidence of the harmful effects of low social status (See Table S2
). This figure is however based on a comparison between the extent of atherosclerosis among monkeys who stayed at the same rank (either dominant or subordinate) (mean extent of CAA approximately
), and those who became subordinate (mean
) and these data from only 42 monkeys are adjusted for skewness, heterogeneity of variance, and two potential CAD risk factors (thigh circumference and ratio of TPC-HDL). The unadjusted data (as estimated from the original scatterplot shown in Figure S2
) seem to show no meaningful association between plaque size, and rank. Moreover, monkeys who were previously subordinate, but became dominant, also experienced an increase in CAA, by 44% greater than the subordinates that stayed subordinate. If any message is to be drawn from this small study, then the message is that change in status in either direction, and not status itself, is pathogenic. The researchers who conducted the study themselves point this out: “All animals with altered social positions (dominants that became subordinate, and subordinates that became dominant) had worsened coronary artery atherosclerosis”
(page 725) The straightforward “500% increase” claim on its own is therefore misleading; if anything, subordinate monkeys are better off staying subordinate.
Finally, though we did not formally assess publication bias, the risk of this and other biases in this small sample of small studies should also be borne in mind, including outcome reporting bias.
Use of the primate evidence in public health
Although as we have seen the evidence appears equivocal, these studies are frequently cited in epidemiological and public health literature in support of the “psychosocial hypothesis”, a conceptual model describing the relationships between social stress, human hierarchies, and human health outcomes 
. Robert Sapolsky's baboons (not included in this review, as objective CAD measures have not been collected on this group) are particularly prominent in these discussions:
“The Whitehall and Serengeti studies are in a sense starting from opposite ends of a possible bridge. While the baboons show hierarchically associated variations in physiological responses to stress that are consistent with health effects, the civil servants show hierarchical variations in health outcomes that must emerge from some physiological pathway.”
The experimental studies in macaques are also used to suggest that the low status is harmful in monkeys and by extension, in humans. As illustration, shows the frequency of citation of most of the primate studies whose results are described above. A sizeable percentage of citations derive from articles in epidemiology and public health journals, but more interesting is the popularity of the Shively and Clarkson (1994) study described above – most of its citations are by researchers outside of the field of primatology, and in these papers it is often cited in the context of discussion of the direct health effects of human social status 
. Despite its equivocal findings, this study, more than any other primate study, is used to support the view that low rank in human societies is directly harmful to health. The disconnect between the reality of available non-human primate data and the interpretation given to it by commentators from various disciplines has been commented on in a different context 
A common form of citation is to refer to both the Shively and Clarkson study of female monkeys (which studied coronary atheroma), and Sapolsky's study of male monkeys (which examined blood lipid levels) (See Table S2
for examples). This suggests that the finding relating to CAD is robust and also relates to males, but when studies of CAD in male monkeys are examined the results are, if anything, in an opposite direction to those (possibly) found by Shively for female monkeys. A further popular claim is that a dramatic five-fold increase in atherosclerosis was generated by downward mobility in the social hierarchy, again referring to the Shively and Clarkson study, though the unadjusted data show little effect.
The limits to generalisation
The evidence from these studies does not provide strong support for a psychosocial explanation of health inequalities. The data derive from studies, which are, almost by necessity, small, ranging from 23 to 193 animals. By contrast, the number of statistical tests carried out in these studies is often large, and power calculations to justify the sample sizes are absent. The need for small sample sizes is understandable in primate research, as is the need to include animal participants in more than one study, but similar biases apply to animal studies as apply to other epidemiological studies, including lack of intention to treat analyses, and lack of blinding of outcome assessment in most studies. Drop-outs (e.g. in this case due to animals dying) are however generally low, at least compared to community-based studies in humans. Other unknown observer biases may also be operating; for example, one meta-analysis of field studies of baboon behaviour has reported that observers recording behaviour of troops of baboons had a tendency to favour watching larger groups, which tend to travel less far, and also differ in other behaviours 
Primatologists themselves have warned repeatedly about over-generalising from primate data to human societies 
. Indeed the data may not even be generalisable between similar species of monkey, as comparative research and field studies suggest that there are striking differences in group composition, social spacing, dominance and aggression between species 
. The social and hierarchical behaviour of Macaca fascicularis
, the species used in many of these studies, may not therefore even be representative of all of its own genus, which raises doubt about extrapolation to higher primates.
Robert Sapolsky has raised another problem for researchers seeking to generalise to humans. He describes “the circuitous and often tragic routes by which primates come to find themselves in laboratories…It is not generally known that there is an extremely high mortality rate among primates during transit…survivors may well be those which have passed successfully through what evolutionists refer to a “selective bottleneck” – and either may be physically and psychiatrically robust, or permanently weakened by their vicissitudes…they could well be “supermonkeys”
. Even if this is not the case, he concludes that the primates available for study represent a far-from-random sample.
This in itself represents a limitation of our review, in that it examines evidence from a selected group of animals. Perhaps more importantly in some cases the investigation of social status is not a primary focus of the study (see , column 1), and in all cases the numbers are small and so they may have had limited power to detect real differences in outcomes between subordinate and dominant (or stressed and unstressed) animals. This heterogeneity in study design and study purpose makes drawing firm conclusions about CAD and social position in primates difficult.
Simplistic interpretations of human hierarchies have also been criticised by Rohde (2001), who agrees that human hierarchies are common, with governments, religions, workplaces, and schools often arranged hierarchically 
. The term “hierarchy” is of Greek origin, originally referred to ranks of ecclesiastical rulers, and was later used to describe the pecking order of angels. Hierarchies often appear rapidly and spontaneously, and phrases like “pecking order” itself are well–recognised. However interpretation of human hierarchies is often difficult, and Rohde suggests four reasons for this. Firstly, higher-ranking people often make paradoxically submissive gestures, (such as allowing others to pass through a doorway first, and there is the obligation of the strong to care for the weak in some religions and in other forms of social obligation, and for the well to care for the sick). Secondly, human hierarchical aspirations can be expressed purely cognitively, with no obligation to action, because humans, with symbolic thought, can model likely outcomes of confrontations. Thirdly, there is a thick cultural veneer (including manners) which overlays hierarchical relations in humans. Finally, humans may have many alternative hierarchies, with perfectly satisfactory alternative hierarchical positions 
Primate social rank and the stress response
This overview has concentrated on CAD outcomes, but studies have also examined associations between dominance hierarchies and hormonal responses. These markers of stress response also appear to suggest psychoneuroendocrine pathways underlying the development of human inequalities in cardiovascular disease. For example it has been suggested that there is an established relationship between high basal cortisol levels and subordinate status, and that this is consistent across primate species 
. However a comparative analysis of this issue reached the opposite conclusion. This study involved a systematic review of data on subordinate/dominant cortisol ratios across ten different primate species, and found cortisol ratios associated with dominant/subordinate status to be highly heterogeneous. The authors concluded that there is no consistent relationship between social rank and stress response in primates and argue that there is no generalisable relationship between social status and any aspect of stress physiology across monkey species 
(). A recent large study has also found that the highest rank wild male baboons had an unfavourable profile of stress-related hormonal measure, in contradiction to the anticipated favourable effects of hierarchy on such indicators 
Table 4 Variation in subordinate/dominant cortisol ratios, by primate species .
Finally, this review provides an example of the difficulty in extrapolating from very few or single studies, particularly when those few studies provide conflicting evidence. There are many examples in the literature of where single studies taken out of context may be misleading, which has fostered an awareness of the risks of relying on single studies, and of the need for comprehensive systematic reviews of the evidence. However such reviews are still relatively uncommon outside of the health and social sciences, though the case has been made previously for the need for more systematic reviews of animal research in particular, both to summarise existing literature and to help direct future research 
Two conclusions can be drawn from this review of the monkey evidence on social rank and CAD. The first is that non-human primate studies present limited evidence for an association between rank and CAD in monkeys; the effects of stress, and social status appear to be more inconsistent than is often assumed, and the relationships may be sex-specific. The data presented to support these associations themselves are also limited, deriving from small studies in highly–selected populations. We took the view that the strongest evidence is likely to come from studies which analyse the impact of these psychosocial factors on CAD outcomes, rather than intermediate outcomes. However it may also be informative in future to conduct a systematic review of studies which address factors related to CAD even if they do not directly measure it.
Secondly, generalisation of these data to human societies may not be warranted, and is against the advice of the primatologists conducting such studies. Given the pre-eminence of Robert Sapolsky's Serengeti baboons in the public health literature, it is probably appropriate to conclude with Sapolsky's own views on rank in monkeys, and its application to humans:
“It seems virtually meaningless to think about the physiological correlates of rank outside the context of a number of other modifiers… This dovetails nicely with the de-emphasis of rank in other niches of primatology …It leads to a final, somewhat obvious point – if we are endlessly struck with the complexity of these issues as they apply to non-human primates, the complexity expands exponentially when considering humans”