The News Doesn’t Always Tell the Truth
This post is an addendum to what was written yesterday about problems interpreting medical research results.
Here’s an abstract from a paper I read today about how difficult it is to determine the causes of observed relationships in long-term studies. (Bold part added by me)
While some epidemiologists embraced probabilistic concepts of cause and effect, others maintained that causal mechanisms must ultimately be deterministic. The tension between probabilistic risk factors and deterministic causal mechanisms continues to haunt epidemiology today.
In the video yesterday the doctor spoke about the problems with epidemiology studies. You might read an article saying “New 20 year study finds strong correlation between drinking red wine and lower incidence of pancreatic cancer.”
Often those studies are surveys where people answer questionnaires once a year about what they eat on a weekly basis, and report their observable health factors.
Researchers then dice the data to find relationships, which then get reported to the news.
Causality in Health Research is Difficult
However, for a study like this, the validity of the results might be plagued by these issues:
1). How accurately are people reporting their food intakes? Their health markers? There are a lot of reasons why people might fudge both of those, consciously or not.
2). How do wine drinkers compare to the population at large? They’re probably higher-income and better educated than most, and both of those have strong predictive powers for your health outcomes, regardless of whether you drink wine or not.
3). How do you account for people who drop out of the study as time goes on? What sorts of people actually stick around long enough to complete these studies? Probably nerdy health freaks.
You can put the results through a statistical chop-suey to account for these types of things, but it’s still a blunt tool to arrive at the truth. After you’re done accounting for the above influences, such a study might tell you…..nothing at all.