which is utterly useless since people will typically under-estimate their calorie intake by 20-60%
Wouldn't this be corrected by randomization? Or is it only overweight people who under-report intake and over-report expenditure?
In other words, suppose I had a hypothesis that redheads were 10-15 pounds lighter than the average person. I could do a survey, but I know that everybody under-reports their weight. However, since redheads under-report just as much as everybody else, it all comes out in the wash -- unless *only* redheads *or* everybody else under-reports their weight.
So if everybody is under-estimating intake and over-estimate expenditure, it wouldn't make a different because both groups. In other words, there's error but not necessarily bias.