| People vary The most extreme variations are allergies. If I eat nuts, I enjoy them. If my brother eats nuts, he could go to the hospital. Then there are things like lactose intolerance. And there are more subtle differences. In one study, researchers fed Thai people and Norwegian people (IIRC) Thai food. then they measured the uptake of certain minerals, and found Thai people got more of them. Then they fed both groups hamburger, and found that the Norwegians got more nutrients.
People metabolize food differently.
Very few studies account for this, and when you simply take the average, you lose the variation. For example, perhaps the average 150 pound woman who is mostly sedentary needs 2000 calories a day (I dunno exactly, but it's about that). But some will need more and some less.
I think it likely that some people can metabolize saturated fat (or any other substance) better than others.
Experiments are impossible In statistics, we distinguish between experiments and observational studies. In an experiment, the researcher can randomly assign people to different treatments. In an observational study, such randomization is impossible (that's a simplification, but this is going to be a long diary anyway). The problem with observational studies is that there can be confounding factors. People who, e.g., give up red meat may make other changes as well - in fact, they probably do. This makes it very hard to disentangle what is going on. Did you lose weight because you gave up heavy cream, or because you started walking to work? Maybe both?
Nearly all nutritional research on humans is observational. To do a true experiment, you'd have to control everything that people ate; that would probably require them living in a lab, for a prolonged period. That's expensive and the people who volunteer to do it may be atypical.
Effect sizes are small Suppose we find that some food is linked to some rare illness. Let's say, just to keep it simple, that the illness affects 1 in 1000 people, but that people who eat thingies have double the risk. To get a good estimate of such a change would require a huge sample. If you sampled 10,000 people, that wouldn't be enough; the 95% confidence interval for the difference would be
-0.27% to 0.07%
(this gets into the whole area of research design - the above is for a prospective design).
Interactions abound In statistics, an interaction means that the effect of one variable is different at different levels of the other variable. For instance, certain minerals and vitamins have to be ingested together for full effect - that is, the effect of vitamin 1 is higher if you also take vitamin 2. Other times, you should NOT take one with another. This makes research hard.
People lie or forget Given that it is very hard to control what people eat, we often rely on retrospective reports. These are going to have inaccuracies in them.
The cult of significance When you see a statement like "Taking Glorp reduces cholesterol" it almost always means that someone did a study and found a statistically significant difference in cholesterol for people who ate glorp vs. those who did not. Then someone else does a similar study, and finds no difference. But statistical significance, although it is by far the most common criterion, is not the best. We usually ought to prefer measures of effect size: How much lower was cholesterol in the glorp eaters? |
