The coefficient estimated by the logit regression technique, similar to other limited dependent variable estimation models, only show the direction of the relationship between the covariates and the probability of a full immunization.
According to this logic, Model M1 contains only the EBM adoption variable, Model M2 includes only control variables, and Model M3 is the full model that incorporates all explanatory variables.
In the final multivariate models only variables with a P level less than 0.05 were retained; non-significant variables were removed by means of a backward selection procedure.
Nevertheless, of the variables retained in both models only spider size (deviance = 7.511, df = 1, P = 0.006) and the interaction between spider size and %UV reflectance (deviance = 8.61, df = 1, P = 0.003) significantly affected honeybee behaviour.
In the final multivariate models only variables with p-value less than 0.05 were retained.
In all models only variables with significant coefficients (p ≤ 0.05) were included.
Of the variables used in the models, only the level of perceived self-efficacy differed significantly between the two professions, being higher for the assistant nurses.
To maintain robust models, only one variable per five events was allowed in the multivariable Cox analyses of each endpoint.
Analyses for each of the response variables included species in the models only if information was available for each of the predictor variables.
This was further simplified into the Best 3 variable model, by using only the top 3 variables (H3K4me1, DNase, H3K27me3), with the highest predictive power.
