Designing cultural multilevel selection research for sustainability science

Humans stand out among animals in that we cooperate in large groups to exploit natural resources, and accumulate resource exploitation techniques across generations via cultural learning. This uniquely human form of adaptability is in large part to blame for the global sustainability crisis. This paper builds on cultural evolutionary theory to conceptualize and study environmental resource use and overexploitation. Human social learning and cooperation, particularly regarding social dilemmas, result in both sustainability crises and solutions. Examples include the collapse of global fisheries, and multilateral agreements to halt ozone depletion. We propose an explicitly evolutionary approach to study how crises and solutions may emerge, persist, or disappear. We first present a brief primer on cultural evolution to define group-level cultural adaptations for resource use. This includes criteria for identifying where group-level cultural adaptations may exist, and if a cultural evolutionary approach can be implemented in studying a given system. We then outline a step-by-step process for designing a study of group-level cultural adaptation, including the major methodological considerations that researchers should address in study design, such as tradeoffs between validity and control, issues of time scale, and the value of both qualitative and quantitative data and analysis. We discuss how to evaluate multiple types of evidence synthetically, including historical accounts, new and existing data sets, case studies, and simulations. The electronic supplement provides a tutorial and simple computer code in the R environment to lead users from theory to data to an illustration of an empirical test for group-level adaptations in sustainability research. Electronic supplementary material The online version of this article (10.1007/s11625-017-0509-2) contains supplementary material, which is available to authorized users.


Generate a group-structured dataset
It is always useful to create a simulated dataset so that you can practice and refine your analysis methods in advance of data collection and study design. We start by creating a simple, random, group structured dataset. You may change these values yourself by editing the code.

groups <-8 indivs <-20
Next, we generate a table of group-wise parameters which will be used to create the actual dataset. These include the presence of the trait under study, and its costs and benefits.

Using Rogers' Inequality
The first method we will demonstrate is applying Rogers (1990) inequality for evaluating whether, in a given case, group selection is strong enough to favor the evolution of an altruistic trait. Rogers' 1990 paper constructs a model of group selection by selective emigration, not on cultural diffusion or cultural selection. However, the assumptions he makes in his model are suitable for our purposes here. Rogers (1990), Page 401, equation 3 reads: where: b represents the fitness benefit of altruism across the entire population, c denotes the average fitness cost of altruim within groups, and F st is the fraction of trait variation which can be attributed to groups. F st is commonly used in population genetics, and is called the 'fixation index for population structure'.

Step 1: Estimate b, fitness benefit of altruism across whole population
To do this we calculate the average fitnesses of altruists and non-altruists and substract them. Step 2: Estimate c, average fitness cost of altruism within groups First, it is useful to make a group-wise data summary. This corresponds to our means dataframe, from above. We use dplyr to calculate group-wise means. library(dplyr) gstats<-as.data.frame( data[,c("group","trait","net")] %>% group_by(group) %>% summarise_all(mean)) Next, we calculate the fitness cost of altruism within each group.

Using the Price Equation
The Price Equation can be used to compare the relative effects of individual and group selection. We use McElreath and Boyd's (2007) formulation of Price equation, as follows: The left hand side of the equation, W ∆Z denotes total change, or "evolution," where W is mean fitness, and ∆Z is average change in trait. We will not compute these separately here, and they cannot be computed separately without temporal data.
On the right hand side we have: W g is the average fitness in group g, Z g is the trait frequency in group g, W ig is the fitness of individual i, in group g, and W ig is the trait of individual i, in group g, so that Cov(W g , z g ) is the covariance between group trait and group fitness, and E(Cov(W ig , z ig ) is average individual-level trait-fitness covariance.
Calculating and comparing these terms is very straightforward. We will use the same groupstructued dataset, data, which we generated in the beginning. For clarity, some code will be redundant with the Rogers example.
1. These data do not include change over time (multiple measurements of trait values), but these computations provide the basic approach which could be extended to time series data.
2. Also, because the dataset is randomly generated it may not produce interesting interactions between individual and group-level selection. We suggest running it a few times to see the various possibilities.
Tim Waring, 2017 University of Maine