Tuesday, November 11, 2008

Correlation verus Causation

Mistaking correlation for causation.

This is probably one of the biggest mistakes we make in nutrition and weight loss research.

If you added up every single grammar error and spelling mistake I have made on this blog over the past year, it still wouldn't even come close to the amount of times people have mistaken correlation for causation in weight loss research.

There are a lot of fancy definitions for correlation, but I like to think of it this way - If two 'things' are correlated, this means they are related.

This is very different than Causation.

Causation is when you change one thing, and you cause another thing to change.

They are similar, but quite different.

For example, I live in the small (but rapidly growing) town of Waterdown. If I were to conduct a research study on the women in Waterdown between the ages of 16 and 22, looking at the relationship between their body composition and the clothing that they own, I would most likely find a significant correlation that would read something like this:

There is a significant correlation in women who own Lululemon workout pants and Nike running shoes and having a low body fat and lower body weight. That is, the women who owned more pairs of workout pants and Nike running shoes, would tend to be leaner and lighter then women who owned less.

This is a correlation.

Obviously the Lululemon pants and Nike shoes did not CAUSE these women to be lighter and leaner (even though it would be great for sales), rather the women that own these clothes tend to go to the gym more often and live a 'fitness lifestyle' that includes watching what they eat, thinking about their health and working out regularly.

So the correlation is in the lifestyle.

This finding happens a lot in research and is often mistaken for causation.

Take for instance all the research that shows that people who eat more frequently tend to be leaner and weigh less than people who eat less frequently.

This would be a great example of a lifestyle correlation.

Over the last ten or so years the trend in the health and fitness industry has been to promote the idea of eating more frequently. And, as a result, people who are heavily into fitness and health tend to eat more frequently then people who are not into the health and fitness lifestyle.

This creates a lifestyle correlation. This does not mean that eating more frequently CAUSES a person to be lean...it just means that people who are lean live a lifestyle that (currently) includes multiple meals per day, because this is what is popular.

But what happens when we remove the lifestyle variable?

Well that is exactly what Karine Duval and her team examined in the research publication "Physical activity is a confounding factor of the relation between eating frequency and body composition"

When you removed physical activity from the equation, the correlation between meal frequency and body composition disappeared.

Or, as they put it:

"It is interesting that the associations between eating frequency and adiposity disappeared after correction for physical activity energy expenditure and VO2peak"

A perfect example of well conducted research proving that we often mistake a correlation for causation.

Scientific research is very difficult. And trying to control all of the components of a persons life so that you can make definitive statements about cause and effect still eludes even the most skilled researcher.

BP

Stumble Upon Toolbar

2 comments:

Redlefty said...

Yeah, that's why I love looking at metastudies but always have a massive grain of salt I take with them.

Mmm... sodium...

Anonymous said...

You make a great distinction between correlation and causation. Many people assume they're the same. Maybe it's because most people just naturally look for easy answers.

I've also noticed that many nutritional (as well as medical) studies seem to have become more and more narrowly focused over the years. I guess this is a result of the increasingly sensitive tools available that allow scientist to study cellular effects at pretty much the DNA level.

However, these narrowly focused studies seem (at least to me) to lead to results that may be beneficial for the item (ie: cell, organ, muscle, etc) under study, but not necessarily beneficial fot the body as a whole. The multiple studies on red wine are good examples.

The studies on red wine all conclude that drinking it is good for your heart as well as for your digestive system. However, they fail to mention the negative effects on your liver or brain -- probably because the study is so narrowly defined.

In order to achieve good health, you have to look at your body holistically. Otherwise, you risk confusing correlation with causation!

Hiram