How Well Does the Shangri-La Diet Work? (part 1: breakdown by starting weight)
The most basic question about the Shangri-La Diet is how well it works. As the host of The Amazing Race might say, there are many ways to answer this question, each with their own pros and cons. Thanks to Rey Arbolay, we can answer this question in a new way: by looking at the data in the Post Your Tracking Data Here section of the Shangri-La Diet forums, where more than 80 people have posted.
The goal of SLD, of course, is weight loss. The problem with just plotting pounds lost as a function of time on SLD is that a loss of 10 pounds is quite different to someone who starts at 120 pounds and someone who starts at 300 pounds. To deal with this I have divided the results by starting weight. In the graphs below, the ranges of weights were chosen so that there would be roughly the same number of people in each graph.
Starting weight makes a big difference — at least, between the two extremes. Many of the people who aren’t losing very fast are in the lowest quartile. There are still a few outliers — people at the higher starting weights who are losing slowly if at all — but not many.
The next step is to do an analysis that estimates the effect of starting weight and somehow removes it. It might be better to divide people by BMI rather than starting weight, and maybe the y axis can be improved.
August 28th, 2006 at 6:11 pm
I’d try a model of the form,
log(y_jt) = a_j + b_j*t + error_it,
where I’m using the notation j for persons and t for time. I’d then model the intercept a_j as a regression, including height and sex as predictors, and then b_j as a regression, including a_j, height, and sex as predictors.
Ideally of course you’d also have a matched set of people who _aren’t_ on the diet, but perhaps that’s too much to hope for…
P.S. Excellent use of multiple graphs per page. My only suggestions here are: (i) make the lines a little thinner (lwd=.5), (ii) keep the color but get rid of the guide lines at -1/wk and -2/wk; these overwhelm the pictures, I think. I’d also arrange in a 4 x 2 grid: the 4 weight classes you have, but 2 columns, one for the men and one for the women.
September 20th, 2006 at 2:15 pm
Before we take out the arsenal of regressions
Are the people who lost a lot of weight more likely to post about it?
September 20th, 2006 at 2:22 pm
Are the people who lost a lot of weight more likely to post about it? Well, no, that is not a problem because it is a prospective study — people post their weights as they go along. The better question is: are people who are going to lose a lot of weight more likely to post about it?