On Econometric and Statistical Research in Accounting

Statistical methods are not inherently faulty. But they can be, and  far too frequently are, misused. So, to turn your metaphor on its head, much accountics econometrics work is more like spraying manure in a perfumed room, or more like a skunk spraying in a perfumed room.

Statistical methods are used for classifying, associating, predicting, infering (causally as well as associatively), organising, and  learning.  It is important to always keep in mind in which context you are using  statistics. 

1. 

In the accountics stuff I am familiar with, determining association is the avowed objective, but the language subtly takes a predictive turn in discussions. The reason usually is the positivist dogma having to do with absence of causation in a naive positivist's lexicon.

I have been stunned by well known accounticians professing that we do not study causes because there are no statistical methods for causal inference. And to the last person, these folks have not heard of modern statistical tools for the study of causation in statistics. Ignorance is bliss in this wonderland. Social scientists, however, have used them for a long time. Theological commitments are dangerous for  ANY "science".

2.

Classification is the first step in learning. It is only VERY recently that accounting folks have started talking about the use of classification by use of clustering, support vector machines, neural nets, etc., but most of these discussions take place in non-mainstream contexts.

3.

Many of the techniques in 2 are nowadays considered part of the  field of machine learning, a hybrid between statistics and computing. I am sure one of these days, when they have become stale elsewhere, they'll be used in accounting. Mainstream accountics academics  are far too conservative to accept any statistical method unless they have been certified stale.

4.

Often, in conversations, accountics folks revert to counterfactual statements. That is natural in the sciences. Underlying such statements are usually causal inferences. It is in this context that I had made observation 1 above.

Building a better mousetrap is a legitimate objective   of sciences, and therefore predictive models are essential component of any science. Accountics' theological commitment to positivist dogma makes them schizophrenic in that they can not admit causality without jeopardising their philosophical suppositions and yet can not ignore it if they are to  maintain their credibility as scientists.