Thursday, July 29, 2010

Variable definitions are tricky

Tyler Cowen calls this the blog post of the year and I am inclined to agree. It gets at a classic problem of labeling of variables that measure a lot of different things. In this case, the scale that is being evaluated measures both the size of government as well as the quality of government. It's hardly a shocking discovery that low levels of corruption are good for economic growth. But it leads to very different policy conclusions than a discovery that large government is simply bad.

[to be fair these factors may be even more correlated than the analysis suggests as it might be easier to police corruption in a small government]

Once again, this is a lesson that we should take to heart when considering complex diseases in epidemiology. Low adherence to medication could be driven by factors like access to care or being health conscious. But it could also be a proxy for which populations show a higher burden of uncomfortable side effects. Disentangling these complex activities (like decisions to remain adherent) isn't trivial and can lead us down very misleading pathways if we aren't critical.

Professor Cowen's pointer is well taken.

No comments:

Post a Comment