As I see it, there are a gap between theoretical work in statistics and real-world data analysis; and differences in opinions among applied statisticians with regards to their approaches to data analysis.
For an example of what I mean by a gap between theoretical work and real-world data analysis, it is known that multiple testing inflates the probability of a type I error yet this is often ignored in practice.
For an example of what I mean by differences in opinions between applied statisticians, some statisticians perceive collinearity in (the data of) a regression model as a problem, some don’t and some will say that it depends on the situation at hand (though often a discussion on what these contexts are and what the suggested solution, if any, is not available in introductory texts). Also, there are some differences in opinion with regards to centered and uncentered measures of collinearity.
Further, there is a great deal of literature that discusses methods that offer an improvement over existing methods (in some situations) but these methods are not implemented in most statistical software making them inaccessible for the applied statistician that lacks either of the mathematical ability to understand these methods or lacks the coding skills to implement them (which casts doubt on the practical significance of such methods).
Further, there are differences in opinion among theoretical works as well though I don’t have a specific example in mind.
Some say that data analysis is more art than science; but how do you as statisticians, with far more experience in both research and data analysis than I, deal with this?
I would really appreciate your help and advice.
