Saturday, April 14, 2012

Bad Science: Correlation and Regression


Analyzing the quality product of a road surfaced based on some quality measure in either the material itself or compaction process makes sense and is proven by statistics to have a positive correlation.  Typically these correlations are very precise where the line of best fit can be used to estimate the quality of the final product.  I think these are very useful in conveying statistical findings to the field personal in the form of heuristics or “rules of thumbs”.  One of these useful ones is that for every 1% deviation in compaction spec equates to 10% off the life the road.  This rule of thumb can be used to send home a point to workers that their attention to quality is magnified in the service life of the final product and this is why quality in their work means so much. To use statistics in this manner is a powerful management tool to foster anotomony in employees.

Observational studies do not establish cause and effect it is a logical fallacy. It can be used to generate hypothesis that then can be tested to establish cause and effect.  This explains that some statistics can fit these lines of best fit but it is not the major contributing factor to the final outcome.  In other words for example we all know that “people who eat breakfast are less likely to be obese” and this can be proven statistically but is that the mechanism that is factoring to obesity? Perhaps people who eat breakfast go to sleep on time and aren’t snacking and playing video games to the wee hours of the morning?

No comments:

Post a Comment