Because this indicator variable showed a very low probability (<0.1) in all classes in all LCA models that were investigated, it was decided to exclude this indicator variable from the analysis and to re-run the LCA on 12 indicator variables.
This is because these variables showed an influence on the outcome variable and there is a need to identify whether each has been confounded by another variable or not.
Genomic control parameters were also computed for the loge(SF) and body tissue iron outcomes because the variables showed bimodal sample distributions which would result in a violation of the assumption of normally distributed residual errors in the linear regression analysis and the potential for an incorrect false positive rate.
The model's covariates were week of injection, lot number, and interaction between week and lot variables, selected because these variables showed evidence of modifying results.
We used summer temperature because this variable has shown to be linked to species richness [ 31, 34].
Variables were expressed as percentages, mean values with standard deviation (SD) and because some variables showed a skewed distribution, also as median plus interquartile range (IQR).
Because some variables showed abnormal distributions, we elected to systematically transform all variables [ 23] to normalize their distributions, as verified with the Wilk-Shapiro test, thus allowing the use of parametric statistical analyses.
Because the investigated variables showed a nonnormal distribution, nonparametric statistical analyses were applied (Spearmann's correlation test, Wilcoxon signed rank Test).
Because many background variables showed different patterns in Eastern and Western European countries we also present results separately for these two regions.
This indicated that the model constructed using contemporaneously both temperature and atmospheric pressure suffered from collinearity problems, because the two variables showed strong negative correlation (r = −0.86, P < 0.01).
