“Wisdom” and Statistical Modelling

Kind of a strange topic, perhaps.

But for some reason I was re-reading “Bayesian Forecasting and Dynamic Models” by Mike West and Jeff Harrison on a quiet Sunday afternoon, and found myself struck by the evident “wisdom” in their introductory comments (some similarities perhaps with the tone and content of Frank & Witten’s “Data Mining”, but I will leave that for another day).

I don’t much like statistics as technique or as over explicated assumptions of convenience : I love statistics as thought and learning.

So here are some quotes from the West and Harrison book, for your interest

A time series model is essentially a confession of ignorance, generally describing situations statistically without relating them to explanatory variables.

The key message for the modeller is “THINK, and do not sacrifice yourself to mathematical magic”.

..data analysis should not replace contextual thinking, nor should it promote mathematical formulae that have no defensible applications

All analytic methods have some contribution to offer, but they must be seen as servants of explanatory thought and not its usurper

.. modellers may, without question, accept sales statistics as their routine observations. The real objective is to forecast customer requirements. Clearly, sales statistics represent what is sold. As such they reflect the ability of the system to meet requirements, and not necessarily the actual requirements themselves.

A major problem for many mathematically based learning systems has been that of accommodating subjective information. This is particularly important at times of major change. .. The vital role of a model structure facilitating expert intervention .. Usually, at times of intervention, there is additional uncertainty with an accompanying sharp change in beliefs about the future.

factors that dominate a system at one level of aggregation may be insignificant at other levels, and vice-versa

a model should focus on a limited range of questions and not try to answer both macro and micro questions ..

Well, there is more along these lines in there.

Leave a Comment