I think the story goes like this: there are correlations between weight and health. There are disputes, but let us skip that and assume for the sake of argument that thin people live longer than fat people. The next question is whether this is causal. If a fat person makes a big effort and becomes thin, will he have a long life, just like a person who was thin to begin with?

We can fill in the details of what the experiment looks like. We start with 200 fatties who want to get thin, and a random assignment of 100 to the control group, who get the usual crap advice about diet and exercise. Since this doesn't work they remain fat. The others get special good advice about diet and exercise. They diet and exercise sucessfully and become thin. Then we wait.

Later (10 years? 20 years?) we follow up. Did the newly thin live longer? It is hard to be sure of the outcome in advance, the observed correlations do not guarantee it.

This is where Paul Campos jumps in a says "hey wait a minute, we cannot do that experiment because we do not have special good advice to hand out."

That is a controversial position, and the key issue seems to be the duration of follow up. Qutoing from How effective are traditional dietary and exercise interventions for weight loss?

Data from the scientific community indicate that a 15-wk diet or diet plus exercise program produces a weight loss of about 11 kg with a 60-80% maintenance after 1 yr. Although long-term follow-up data are meager, the data that do exist suggest almost complete relapse after 3-5 yr.

Think about planning the experiment that I sketched earlier. The key to getting a crisp, clear result is a ten or twenty year follow up measuring all-cause morbidity and mortality. The designer of the experiment needs to chose an intervention for the trial group, but the weight loss interventions that follow up five years later are reporting almost complete relapse for all interventions. So design of the experiment shudders to a halt.