Some models are used for classification, others for prediction and in the case of drivers analysis, models can be utilised for explanation.
Depending on how the data has been collected and the sample size, different techniques can be used, ranging from MaxDiff to regression variants.
Different types of regression are suitable depending on the outcome variable of interest e.g. Linear regression (continuous outcomes), Logistic (binary outcomes), Ordered logit (ordered categories), Multinomial (unordered categories), Bayesian (observations by ‘group’), Poisson (counts), Negative Binomial (counts).
State of the art for ‘Drivers Analysis’ is Shapley/Johnsons Relative Weights analysis, although straightforward correlations and principal components analysis can add and inform in this space.
Typical outputs include MaxDiff rankings, Regression importance scores, Correlation heatmaps, Relative Importance results or classification models.