Interesting read. Thanks for sharing. While I see your general point, I believe your assertion is untrue. Within the regression world, technicians are well aware of problems with direct interpretation of betas(weights, coefficients, etc..) as those are swayed by scale ( think dependent continuous variables measured in Kilograms vs. Grams etc…). In most regression problems, Betas for categorical variables represent the relative relationship of each level in the category to a baseline level so yes they are in interpretable in that respect. I agree that there is no causal relationship ( please don’t tell the VP to go sending thank you notes lol), but the essence of a model is to describe correlated relationships between inputs and outputs, and in that respect weights, especially standardized ones are a formidable tool for understanding the push-pull dynamics of dependent-to-outcome relationships. Do you know that ANOVA, one of the premiere approaches in Experimental design is essentially equivalent to regression?
