An agent-based model of the role of resource dynamics and the environment in human female life cycles

Dear Claudio Tennie, Cecilia Padilla-Iglesias, and anonymous reviewer,

Thank you very much for the helpful comments on our preregistration! Based on your feedback, we have now (1) expanded on the explanation for how we summarize and investigate the model output, (2) clarified the feasibility of the simulation across the parameter space, and (3) checked the wording and presentation throughout. Please find our detailed replies to the comments below (in bold).

All our best,

Pablo José Varas Enríquez (on behalf of all co-authors)

Review by anonymous reviewer 1, 21 Dec 2023 11:10
C1.0: This is a very interesting project aimed at better understanding how variation in (human) female life history emerges. This agent-based model expands on our understanding of (human) female life history in that it does not attempt to find an optimal life history, but is explicitly designed to generate and quantify variation. The introduction is extremely well written and does an excellent job pointing out the findings and limitations of previous life history evolution models and how this study expands on them. As a behavioral ecologist and someone who has been closely following recent debates in human life history research I welcome the emphasis on understanding variation and the factors that generate it, such as variation in resource acquisition, or redistribution through transfers. I have to caveat my review though by saying that I am not a modeler myself, so while I understand the general logic of what the authors propose I have too little experience with modeling to really comment on the appropriateness of the methods. 

R1.0: Thank you for your positive feedback! 

C1.1: One general thing that struck me was the vast parameter space that the authors propose to explore (e.g. 3362 combinations of parameter values for one question), which to me immediately raised the questions of (1) computational feasibility, as well as (2) clarity of communication. I would have liked to see some preliminary exploration of computation times, and some kind of estimate of how long these proposed parameter sweeps might take; I'm saying this because we routinely run statistical models that take days or even weeks, which can severly constrain the amount of robustness checks, sensitivity analyses, etc. that we can do. 

R1.1 Thank you for your concern regarding the computational feasibility and clarity of communication.

Regarding the first point of computational feasibility, we developed the code on a standard desktop computer, where a full run of a single model took less than a minute. For the exploration of the parameter space, we will use the servers available at the Department of Human Behaviour, Ecology and Culture at the Max Planck Institute for Evolutionary Anthropology, which speeds up the run of a single model to less than a few seconds. Hence, there should not be a problem regarding the computational feasibility of covering the parameter space. Furthermore, we decide to decrease the granularity of the parameter values that we will analyse from values every two 0.02 to 0.05 (e.g. from 0.1-0.13 to 0.1-0.15), considering that such differences would not imply major unexplored areas of the parameter space and, therefore, no major differences in the analysis (lines 180-194 and lines 705-728). This way, the parameter space is reduced, increasing the computational feasibility while using the computational infrastructure available in our research institution. 

Regarding the second point of communication, we made changes in section 3 “Model analysis” (lines 698-764) to clarify how the different scenarios build up from each other, allowing us to keep track of the causes behind the differences in the outputs, and we also added Figure 4 to have a visual representation of the process (for details see responses below). Furthermore, we decided that we will show the whole distribution of values of each life history trait of interest (e.g. longevity, lifetime reproductive output) instead of only the coefficient of variation (lines 729-735). We believe that analysing the distribution of life history traits of the whole population instead of using a summary statistic can give a more complete understanding of the differences observed in our analyses.

C1.2: With regard to communication, it would have been nice to see how these vast parameter sweeps will be presented and interpreted; generally the authors are quite clear about their predictions and the interpretation of possible outcomes (which is great!), I just had a bit of a hard time visualizing this for such vast parameter spaces. 

R1.2 Thanks for making more explicit the need for a more visual representation of the outcomes of our model. It is more complicated to visualise when there are multiple parameters occurring simultaneously. However, we expect that Figure 4, mentioned earlier, helps to understand better the possible differences that we expect in the outcome variables of our model. In this figure, we now show what the predictions mean in terms of shifts in the distributions of expected values in the different scenarios.

C1.3: Another broader comment refers to transfers. I really like that transfers are included in this model, but I had some concerns/questions. First, it really wasn't clear to me until quite late in the manuscript that transfers are distinct from maternal investment and that they follow a different logic. I think this should be flagged earlier, because for most of the text I had intuitively thought of transfers as directed towards kin. 

R1.3 Thanks for highlighting that it was not clear enough that maternal investment follows a different dynamic than resources transfers among random individuals in the population. We made the clarifications in the text between lines 226-253 to clarify our assumptions, and that resource transfers do not follow a bias towards kin nor need, nor other mechanisms. Additionally, we clarify that maternal investment is a separate dynamic from resource transfers between lines 254-261.

C1.4 Second, while maternal investment follows a clear logic (patterned by kinship and need), other transfers are seemingly random. For example, individuals have the same probability of transfering to any other individual (line 393) and "transfers ... are therefore not guided by relatedness" (line 398); instead, they are only structured by life history stage, such that they mostly flow from older to younger stages (line 449). This is said to follow previous work showing downward flows of resources across generations; however, such previous work (e.g. the cited Hooper et al 2015 paper) also emphasize kinship, need, reproductive value, etc. in patterning such downward flows. In short, I don't really follow the logic of why transfers couldn't be patterned by relatedness and need, just like maternal investment, and this is not justified well. I'm not saying this absolutely needs to be changed, I would have just liked the authors to be more upfront about why it is set up this way. For example, there are conditions in which investing in *any* other group member can be beneficial, as shown by the literature on group augmentation or cooperative breeding in harsh environments, and it can easily be argued that transfers from older to younger individuals make pretty much universally sense as older individuals have more experience and are more likely to overproduce while younger individuals have higher residual reproductive value.

R1.4 Thanks for highlighting the lack of clarity regarding the assumptions behind the resource transfer dynamics in our model. We clarified in the text now (lines 226-241) that the aim of this model is to remain as agnostic as possible, considering previous theoretical developments and empirical work to characterise the biological constraints of the female human life cycle. 

Minor points:

C1.5: Line 352-353: so mothers will let the most needy offspring die if they cannot cover their needs?

R1.5 Exactly. We made some changes in lines 391-390 to make it more clear.

C1.6: Line 666: Does "probability of transferring resources" here refer to maternal investment and other transfers or only the latter?

R1.6 In line 666 it referred to the resource transfers between individuals, not the maternal dynamics. We added some clarifications in lines 699-703 in order to avoid confusions.

Everything else was very clearly written and presented, thank you!

Review by Cecilia Padilla-Iglesias, 01 Apr 2024 09:16
C2.0 I congratulate the authors for the clear and interesting framing of the study. It has a well-defined purpose and question, very relevant for understanding human evolutionary dynamics and I’d definitely recommend it to move to the following stage.

R2.0: Thank you for your positive feedback!

C2.1 However, I believe some of the model assumptions could be revisited to ensure outcome-neutral conditions as well as to make sure that the insights of the model match the patterns observed in human populations, and therefore that the model serves to explain the questions it is set to explain. In addition, when turning this report into a publication, the general theoretical framing of the study should be revised for clarity and to make sure the necessary background is provided to justify the research questions. I provide some additional (more specific) suggestions below:

R2.1 Thank you for your insightful general opinion of the model assumptions and the theoretical framing we use. We clarify in the following replies how we address them considering your specific suggestions.

C2.2 The introductory paragraph reads a bit confusing. It is unclear which statements refer to humans in general, to women or to women from particular societies - and therefore what are the parameters that the study wishes to explain. I’d maybe start with a more general (1-2 line) description of the human life cycle and then talk about the female life cycle (especially because then you go on to talk about the human life cycle, so it is not clear what is female-specific). 

R2.2 Thanks for your suggestions. We made changes in lines 19-44 in order to clarify that we are constantly talking about the female human life cycle, and female populations. Furthermore, we also specify in the same lines which are the parameters we want to analyse in our study when we state the aims of our computational model.

C2.3 Similarly, before the sentence in line 33 where you introduce the potential explanations for this variation - I’d introduce which dimensions of variation you will talk about.

R2.3 Thanks for highlighting the sentence of line 33. We added information in lines 34-40 in order to clarify what we mean with variation, and that we are focusing on the individual stochasticity of resource dynamics.

C2.4 In the last paragraph of the introduction, I’d separate research questions from predictions (or eliminate questions and just phrase it in terms of hypotheses + predictions given that the nature of this research is to test those predictions).

R2.4 Thanks for sharing your approach on how to present our research questions, and the patterns we may expect in our results. It has been a long-standing discussion among the co-authors and other colleagues that work on theoretical models on how to think, and present, hypotheses and predictions when developing computational models. The main issue is how to avoid possible biases in the model because of our own expectations and predictions. To tackle this issue we follow a generative inference approach for the development of our model. Hence, more than stating specific hypotheses that will be tested by our model, what we aim for is to develop a model with explicit causal mechanisms that would help us to understand how individual-level resource dynamics informs population-level life history traits distributions. Therefore, we would only be able to infer possible explanations from the distributions of life history traits that emerge from the model based on the parameter values we set for the individual-level dynamics. This way, we would avoid building a model that would just give as an output the results we expect from more formal predictions. We made clarifications in lines 143-1467, and in Table 2 to align better with the presentation of the research questions and aims of the model, but we did not separate the research questions from the hypotheses, as you suggested, because of the arguments described above This study is at the step to generate predictions, rather than to test specific predictions.

C2.5 I like Figure 1, but it is unclear to me why the individual in the lower part of the figure is sharing 50% of their harvest in the high and low sharing conditions. Maybe I have misunderstood something but in Table 1 you make transfers probabilistic, so in that case the image should represent an increasing probability of sharing?

R2.5 Thanks for highlighting the lack of clarity in Figure 1. We changed it to a new version to better represent what you see in Table 1, which better follows the probabilistic approach used in general throughout the model. We hope that the new version makes it more clear about the dynamics of the model. 

C2.6 In lines 177-189, when you specify the parameter values you will use, please specify the outcomes the model will produce and the measurements you will take and analyse to answer the questions. The same in Table 2, in the analysis plan authors need to mention the outcome.

R2.6 Thank you for pointing out the need to clarify the outcome. In lines 175-178 we clarified that the outcomes of the models are the distributions of the different life history traits we use to characterise the female human life cycle. We also added a clarification in the caption of Table 2 to specify that “life cycle variability” within the table refers to the outcomes of the model.

C2.7 Similarly, in lines 177-189, where you mention that some parameters lead to populations going extinct, I believe exploring what combinations of resource dynamics and behaviours lead to non-viable populations would also be an interesting model insight to explore as it would point out prohibitive dynamics in human populations.

R2.7 Even though we agree that the scenarios under which populations go extinct could show some prohibitive dynamics in human populations, we decided to not account for them in our analyses. The aim of our model is to observe how the distributions of the female human life cycles change under different resource dynamics, within the biological constraints of the female human life cycle. Therefore, the question of extinction is outside of what we are aiming for with this study. We therefore decided to decrease the number of combinations we will present in our results to focus on the particular patterns we want to investigate.

C2.8 Lines 728-744: When you specify why you remove the possibility of individuals transferring resources prior to ensuring their reproduction because that results in a simulated population that is closer to those observed in human populations - it is unclear why precisely fixing that parameter to be necessary for the model to produce realistic populations. In other words, fixing other parameters could lead also to realistic populations (even when including reproductive costs within the definition of surpluses). This is because, for example, among hunter-gatherer societies, resource transfers are done at the time of resource acquisition and prior to ensuring one’s needs are covered. So in my eyes, to explore the influence of resource transfers on female life cycles, this would be an important parameter to explore.

R2.8 Thanks for your comments regarding the assumptions we use in the model for resource transfers, and how we define resource surplus. We clarified the definition of resource surplus available for resource transfers, which is the upper limit of resource sharing in the model, and the logic behind this definition in lines 786-816. We argue that, according to the literature and the model exploration we performed, there is no evidence showing a relationship between the sharing dynamics and the reproductive cost for an individual. Therefore, including the reproductive costs in the definition of resource surplus is a valid assumption in our model. Hence, even though it might be interesting to compare different definitions of resource surplus and their impact on the female human life cycle, it goes beyond the scope of our research questions, which already requires us to make multiple assumptions to simplify the model and have a clear mechanistic path to explain the expected differences in the output of our model.

C2.9 In general, the model is explained in a lot of detail but the analyses of its outcomes not really. The authors mention they will calculate the coefficient of variation across individuals in a population. How will these values be aggregated across simulation runs, how will they be compared across populations? Please provide details of the statistical approach that will be used. 

R2.9 Thanks for pointing out the lack of explanation regarding the analyses of the outcomes. It is true that compared to the definition of the model there are less details on how are we planning to analyse the outcomes. For this, we made changes in Section 3 “Model Analysis” (lines 698-764, and Figure 4) in order to make it more clear which are the different steps that will be used to analyse the influence of resource dynamics on the female human life cycle. Specifically, we will no longer calculate the coefficient of variation for each life history trait but rather analyse, and show, the distribution of the values of the whole population. We believe that such a difference will make things easier to understand and visualise. Furthermore, we will not perform statistical analyses considering that our model is a mechanistic one. This means that the differences we observe will only occur because we have a clear understanding on how the changes in a specific parameter for an individual translates into a difference in their life cycle, as it is detailed described in earlier parts of the manuscript. Additionally, since we will be analysing the whole distribution of the life cycle and not a summary statistic, we will have the information of the whole population and not a sample from it. 

C2.10 Similarly, the authors specify that they will run sensitivity analyses. How will these be performed? There’s extensive literature on different methods used to performed sensitivity analyses, so please specify which approach will be used, which parameter values will be explored, and which outcomes (if not all) will be used to compare across those parameter values.

R2.10 Thanks for highlighting the need for further clarifications regarding sensitivity analyses. We agree that there is an extensive literature dedicated to sensitivity analysis. What we aim to see is how much our results would change considering differences in population size, as well as the survival and reproductive costs. We understand now that this might not be a sensitivity analysis in its more traditional understanding. Therefore, we decided to change the wording from a sensitivity analysis to a robustness check. We believe that robustness check might be a better fit in order to explain what we aim to do in this section. You can see the changes in lines 748-764.