Sexual coercion can be defined as the use by a male of force, or threat of force, which increases the chances that a female will mate with him at a time when she is likely to be fertile, and/or decrease the chances that she will mate with other males, at some cost to the female (Smuts & Smuts 1993). It has been evidenced in a wide range of species and may play an important role in the evolution of sexual conflict and social systems. However, identifying sexual coercion in natural systems can be particularly challenging. Notably, while male behaviour may have immediate consequences on mating success (“harassment”), the mating benefits may be delayed in time (“intimidation”), and in such cases, evidencing coercion requires detailed temporal data at the individual level. Moreover, in some species male aggressive behaviours may be subtle or rare and hence hardly observed, yet still have important effects on female mating probability and fitness. Therefore, investigating the occurrence and consequences of sexual coercion in such species is particularly relevant but studying it in a statistically robust way is likely to require a considerable amount of time spent observing individuals.
In this paper, Smit et al. (2022) test three clear predictions of the sexual coercion hypothesis in a natural population of Mandrills, where severe male aggression towards females is rare: (1) male aggression is more likely on sexually receptive females than on females in other reproductive states, (2) receptive females are more likely to be injured and (3) male aggression directed towards females is positively related to subsequent probability of copulation between those dyads. They also tested an alternative hypothesis, the “aggressive male phenotype” under which the correlation between male aggression towards females and subsequent mating could be statistically explained by male overall aggressivity. In agreement with the three predictions of the sexual coercion hypothesis, (1) male aggression was on average 5 times more likely, and (2) injuries twice as likely, to be observed on sexually receptive females than on females in other reproductive states and (3) copulation between males and sexually receptive females was twice more likely to be observed when aggression by this male was observed on the female before sexual receptivity. There was no support for the aggressive male hypothesis.
The reviewers and I were highly positive about this study, notably regarding the way it is written and how the predictions are carefully and clearly stated, tested, interpreted, and discussed.
This study is a good illustration of a case where some behaviours may not be common or obvious yet have strong effects and likely important consequences and thus be clearly worth studying. More generally, it shows once more the importance of detailed long-term studies at the individual level for our understanding of the ecology and evolution of wild populations.
It is also a good illustration of the challenges faced, when comparing the likelihood of contrasting hypotheses means we need to alter sample sizes and/or the likelihood to observe at all some behaviours. For example, observing copulation within minutes after aggression (and therefore, showing statistical support for “harassment”) is inevitably less likely than observing copulations on the longer-term (and therefore showing statistical support for “intimidation”, when of course effort is put into recording such behavioural data on the long-term). Such challenges might partly explain some apparently intriguing results. For example, why are swollen females more aggressed by males if only aggression before the swollen period seems associated with more chances of mating? Here, the authors systematically provide effect sizes (and confidence intervals) and often describe the effects in an intuitive biological way (e.g., “Swollen females were, on average, about five times more likely to become injured”). This clearly helps the reader to not merely compare statistical significances but also the biological strengths of the estimated effects and the uncertainty around them. They also clearly acknowledge limits due to sample size when testing the harassment hypothesis, yet they provide precious information on the probability of observing mating (a rare behaviour) directly after aggression (already a rare behaviour!), that is, 3 times out of 38 aggressions observed between a male and a swollen female. Once again, this highlights how important it is to be able to pursue the enormous effort put so far into closely and continuously monitoring this wild population.
Finally, this study raises exciting new questions, notably regarding to what extent females exhibit “counter-strategies” in response to sexual coercion, notably whether there is still scope for female mate choice under such conditions, and what are the fitness consequences of these dynamic conflicting sexual interactions. No doubt these questions will sooner than later be addressed by the authors, and I am looking forward to reading their upcoming work.
Smit N, Baniel A, Roura-Torres B, Amblard-Rambert P, Charpentier MJE, Huchard E (2022) Sexual coercion in a natural mandrill population. bioRxiv, 2022.02.07.479393, ver. 5 peer-reviewed and recommended by Peer Community in Ecology. https://doi.org/10.1101/2022.02.07.479393
Smuts BB, Smuts R w. (1993) Male Aggression and Sexual Coercion of Females in Nonhuman Primates and Other Mammals: Evidence and Theoretical Implications. In: Advances in the Study of Behavior (eds Slater PJB, Rosenblatt JS, Snowdon CT, Milinski M), pp. 1–63. Academic Press. https://doi.org/10.1016/S0065-3454(08)60404-0
DOI or URL of the preprint: https://doi.org/10.1101/2022.02.07.479393
Version of the preprint: 3
Many thanks for your submission.
I am deeply sorry for not noticing this earlier but I would just have one final suggestion for improvement:
In Figure 1c the occurence of copulations is shown on the x axis and the aggression rate on the y axis. Similarly in the text lines 303-305, you present the mean aggression rate of dyads that copulated vs dyads that did not. Yet, was was tested (and is indeed correct) is whether aggression rate statistically explained the probability to copulate later on, not the other way round. Therefore I feel that a figure showing copulation rate on the y axis as a functon of agression rate on the x axis would better illustrate the analysis (even if it may not look as "nice" as copulation is a binary variable. Similarly, it would be more relevant and adequate given the test performed to provide in the text was the expected copulation probability was for dyads with no aggression vs dyads with an aggression rate of e.g 1 per hour (or another more biologically relevant rate).
Do you think it could be possible (and relevant) to make such a change? Meanwhile I am already writing the recommendation, so hopefully the recommendation will not be delayed at all.
Apologies again and best wishes,
DOI or URL of the preprint: https://doi.org/10.1101/2022.02.07.479393
Version of the preprint: 2
Your preprint entitled “Sexual coercion in a natural mandrill population” has been reviewed again by one of the previous reviewers who only provided a few minor comments to address on a PDF document.
In addition I also have the two following comments:
Line 87: perhaps use a more "geographical" terminology as the term "Old World" reflects a colonial perspective.
Line 228: please state in the text (as you did in your response to one of my previous comments) that results remained similar when using slightly different thresholds and whithout using any threshold.
Once these comments are addressed, I will be very happy to recommend your preprint.
I wish you a very nice day,
DOI or URL of the preprint: https://www.biorxiv.org/content/10.1101/2022.02.07.479393v1
Version of the preprint: 1
Your preprint entitled “Sexual coercion in a natural mandrill population”has now been reviewed and the reviewers’ comments are appended below. As you will see, both reviewers are highly positive about the study, and I share their views, notably regarding the way it is written and how the predictions are carefully stated and tested. Yet they have several comments that need to be addressed carefully before your preprint can be recommended. Please note that the second reviewer also commented on the PDF and you should be able to download this review.
In addition I also have the following comments:
1) In the script 1rstPredictionStats.R
# aggrBinAM: Did the female received aggression from adult males towards the female this day
# harshBinAM: Did the female received aggression from adult males towards the female this day
# aggrBinYMF: Did the female received aggression from groupates other than adult males towards the female this day
# harshBinYMF: Did the female received aggression from from groupates other than adult males towards the female this day
The descriptions of these variables are identical but their values are not identical. Please clarify their differences (agression vs "severe" agression?).
Line 76: I get an error message "Error in etapred + sim.reff : non-conformable arrays". Can you ensure this function can be run and the fit of the model assessed?
Lines 143 and 144 the name of the model output is incorrect.
2) Script 2ndPredictionStats.R
Line 16: .csv is missing (STATinjCyF <- read.csv(2ndPredictionTable.csv))
Line 29: # month: Month of observations (not orservations)
Line 31 and 32: sex ratio instead of ration
3) Script 3rdPredictionStats.R
Line 18 correct the name "3rdPrediction" instead of "3ndPrediction"
Line 24 arrival? (correct other typos also if possible)
Line 60: having a fixed effect seems safer than an offset in a binomial likelihood with logit link (see comment below).
Line 157: spell out GLMM the first time you use it.
Line 160-161: briefly justify why you need to control for these variables.
Line 162: It seems indeed relevant to try to "offset" the probability to observe the event by the length of the observation period. However, I am not certain this would be the right way to proceed with a Bernouilli (binomial) distribution and logit link. Unfortunately, I am not aware of ways to easily "offset" in such cases. Given the very low (about 2%?) probability of observing at least one event within an observation period, I guess the probability to observe 2 of these events is very close to zero (did it ever occur in the present dataset?)? In such case you could instead use a Poisson distribution and then having log(time) as offset would be fine (and no need to scale it I think). If you do have the information of the number of events occuring during the observations (if it did happen more than once at times), then you could use that information as well.
Line 171: is it the probability that she got injured that day or that she was seen with an injury that day? My question is, can we be sure the injury happened on that day? If so, it can be left but if not it may be best to rephrase for clarity.
Line 176: perhaps change mating success for mating probability (if this is what is meant) for clarity?
Line 176-177: for prediction 3, can it be controlled for the familiarity between the 2 individuals? (i.e. their probability/number of interactions). My question is: could the positive relationship between rate of agression and mating probability be solely due to the fact that these two individuals interact more (any "neutral" interaction rate would also be associated with mating probability)? If it could be a possibility, please state it in the discussion. If not, please clarify why not in the method section.Line 187: briefly say why using OSR instead of SR in this analysis.
Lines 191-193: again the offset may be problematic here, although in this case I understand why it may be more interesting to look at effects on the probability to mate than at the number of matings. Perhaps it is best to just use time as a fixed explanatory variable here? That sounds fine by me but otherwise one could build a more customized statistical model (I could think about it if you decide to go down this road, but I am not a statistican and there are for sure better qualified people to help!).
Line 193: this needs some clarification and if possible references. What would be the biases due to too short observations and why is 30 minutes a reasonable threshold to prevent such bias?
Line 211: I understand why you would expect such result if female choose to mate with aggressive males, but it could be that aggressive male mate more, irrespective of whether females can exerce any choice? I would replace "solely" by "potentially" but if I am misunderstanding you can just clarify.
Line 217: State here (instead of in the appendix) that "whenever a singular fit was observed, we reran the relevant model with the bglmer function of the blme package ". I'd actually recommend having the whole "Statistical Analysis" section of the appendix in the main text. Also briefly justify the use of the "optimizer" (control==glmerControl(optimizer="bobyqa")).
Line 226: were not "significantly/clearly" more targeted, or similar rewording (one should not accept the null hypothesis).
Line 239: avoid causal language (positively influence). It is very nicely avoided elsewhere in the result section.
Line 248: predict instead of predicted.
Line 250: if by "strongly" you refer to the statistical significance I would avoid it (as it should rather refer to effect size) and use "significantly", "clearly" or similar wording instead.
Line 255: it may be personal but perhaps avoid using the word "failed" here. Not finding statistically "significant" effects should not be perceived as a "failure".
Line 278: it is not shown that male agression "improves" male mating success. Either change "we showed" for e.g. "our analysis suggests" or change "improves" for a non-causal statement.
Line 289 and 297: again, perhaps don't use the word "failed".
Line 298: rephrase the causal statement.
Line 303 and 307: "on average" more often, or similar wording, as the difference between the two is not tested and the standard deviations provided suggest overlap of the estimates.
APPENDIX: I agree with the reviewer and that most if not all of the appendix can be in the main text. It is relatively short and there is no page limit for the preprint.
Line 26: how do you estimate error (if it is from ref. 1 cite it at the first sentence already) and what is "a few" days? Be specific.
Lines 33-36: again avoid using "a few" days and "several" says and rather provide a mean and/or a range of number of days for each statement.
Lines 103-104: in addition, what seems particularly interesting to show here (rather than p values) is how the effect of the rate of aggression towards the dyad female get affected by including the agression rate towards all groupmates. Could you show this estimate and confidence intervals before and after inclusion here?
I look forward to reading the revised version of this preprint.