Using Akaike's Information Criterion AICC) [18] the importance of the variables in the regression equation was measured by removing each in turn, then comparing the goodness-of-fit of the reduced model with the full model containing all terms.
Minimal adequate models were obtained by backward stepwise elimination (i.e. sequentially dropping the least significant terms from the full model, until all terms in the model were significant).
The influence of each independent term within the minimum adequate model was assessed using a likelihood ratio test (L-ratio) between the minimum adequate model and reduced models (with all terms involving the relevant independent factor removed, including interactions).
The 'DNA-relay' model contains all terms described in Equation 3, which takes into consideration diffusion, binding/unbinding, and elastic forces (i.e., dynamic DNA-bound ParA dimers).
Using the maximum heat index parametric model as the base model, all interaction terms between maximum heat index functions and within-season temporal trends, including day of year and month, were not significant.
In order to assess changes in standard errors when adjusting for both education and wealth, we compared results from three model specifications: Model 3 corresponds to the main model described before, Model 1 includes all covariates from Model 3 except all terms related to the wealth index, and Model 2 includes all covariates from Model 3 except all terms related to education level.
Male age did not influence contest behavior through any interactions with other terms included in the models (all interaction terms removed from the models at P > 0.219).
Eventually, we reached a model in which all terms were statistically significant.
*Estimates are from a model that included all terms listed in the table.
The multivariate model used included all terms significant or trend (P < 0·1) to univariate analysis and then eliminated non-significant terms.
