Measuring the impact of a binary variable on a quantitative response in a non-parametric framework.

Summary This paper proposes a method for quantifying the effect of a qualitative explanatory variable on a quantitative response that is more flexible to use than the simple coefficient of a multiplicative GLM model. The objective is to have a measure that does not require the proportionality assumption of the GLM and that is completely uncorrelated with the effect of the other explanatory variables included in the model. The approach is illustrated using data on the cost of automobile claims, for which we seek to quantify the impact of the adjuster who assessed the claim on the amount of the assessment.