...curiously, if I did look, would I find that this had been addressed?
I’m not saying you should look. I’m just saying, time spent is not so often rewarded, whereas anyone that has kept with the discussion would readily know.
Furthermore, it helps to keep track of and repeat one’s assumptions. To write, “doesn’t work at all” doesn’t sound right without some qualifications.
That’s very funny. I meant that I hadn’t read the previous posts:
Previously: Mainstream Nutrition Science on Obesity, Atkins Redux, Did the US Government Give Us Absurd Advice About Sugar?, What Causes Obesity?
At some point, it was probably discussed in which context it was appropriate to look at averaged data...I was hoping for a recap. I have composed a comment about whether I should have read the previous posts in the open thread.
[...] the answer to the [...] question is clearly “no.” In the sense that the vast majority of people end up regaining weight regardless of the diet they are on.
If the diet should work for some people, but not most people, why should you say it doesn’t work at all?
There appears to be some substance in the disagreement.
Suppose that there are many different ‘metabolic’ types of people, and different diets work with different efficacies for these different types. Perhaps any diet works well with a small subset of people (we already know some people don’t need to diet). Perhaps particular diets work well with certain subsets of people. Maybe a particular diet works extremely well for one subset, in that they lose weight and keep it off, whereas it only works moderately well for others (they gain the weight back or must combine different diets). This model—which seems quite likely anyway—would imply two things:
the effectiveness of a diet for small subsets of people would be completely obfuscated by averaging the data—for example, in all groups in Figure 2, dieters kept off some of the weight on average. This might mean that some people gained all the weight back, whereas some kept it off, or it might mean that everyone gained some back. (Note that showing standard error, rather than standard deviation, especially hides the role of individual variation.)
it would not be reasonable to expect a diet to work by working for everyone; instead, each person must find which diet, if any, works for them, and then we would say that a diet “works” if it works for any subset of people that would otherwise be overweight
I’m pretty certain that this paradigm is mainstream. Increasingly, we’re recognizing that medical solutions (for example, blood pressure medications or cancer therapies) need to be tailored to individual people.
Here’s what the authors of the 2008 paper (Figure 2) have to say about individual differences:
We enrolled few women; however, we observed a significant interaction between the effects of diet group and sex on weight loss (women tended to lose more weight on the Mediterranean diet), and this difference between men and women was also reflected in the changes in leptin levels. This possible sex-specific difference should be explored in further studies. The data from the few participants with diabetes are of interest, but we recognize that measurement of HOMA-IR is not an optimal method to assess insulin resistance among persons with diabetes.
Gender and diabetes are individual differences that are easy to recognize. Who knows (yet) how many characteristics are relevant for diet. If someone is sensitive to sugar like me (can’t go long periods without eating, tends to put on weight on their belly), the low-carb diet might work for them. Another complication is that changes in gut-bacteria populations over longer time-scales might result in long-term changes in your diet type. For example, over-weight people have different gut bacteria, diets change gut bacteria and gut bacteria might change your equilibrium weight. (The media is dominated by these studies recently.) I find a positive effect—once I’ve been on the low-carb diet for a few months, I’m not as sugar-sensitive and can maintain my ideal weight with a higher level of carbs than before I began the diet. Over time I lose the effect (too many holiday parties?) and need to return to a stricter low-carb diet.
I don’t fault the discussion for not accounting for gut bacteria yet (the diet science literature is huge, and you have to begin somewhere)...but would you agree that it is complicated enough that we can not infer much of anything from average responses to a diet, other than that a particular diet doesn’t work for everyone?
There appears to be some substance in the disagreement.
Suppose that there are many different ‘metabolic’ types of people, and different diets work with different efficacies for these different types. Perhaps any diet works well with a small subset of people (we already know some people don’t need to diet). Perhaps particular diets work well with certain subsets of people. Maybe a particular diet works extremely well for one subset, in that they lose weight and keep it off, whereas it only works moderately well for others (they gain the weight back or must combine different diets).
Assuming this is true, I don’t see how it contradicts anything I have said.
Do you dispute any of the following:
The vast majority of people end up regaining regardless of the diet they are on;
In general, the central and essential problem of successful dieting is dealing with the problem of difficult-to-resist urges;
Taubes’ approach is not a general solution to this problem.
Yes, I dispute with the first point. In any case, you can’t infer this from figure 2.
But that wasn’t my initial point. My initial point, asked in a different way, is what fraction of people would a diet have to be effective for, for you to say that it works “at all”? This is a substantive question, not just semantics, because I would like to convince that an expectation of 100% is not reasonable—rather 20 or even 10% would be quite good.
But that wasn’t my initial point. My initial point, asked in a different way, is what fraction of people would a diet have to be effective for, for you to say that it works “at all”?
I’m not sure, since I generally don’t use “at all” in positive sentences. Besides, it’s just a question of semantics.
This is a substantive question, not just semantics, because I would like to convince that an expectation of 100% is not reasonable—rather 20 or even 10% would be quite good.
Well it depends on how you define “reasonable” or “good.” In terms of deciding whether a diet is worth trying, a 10% success rate might make it worthwhile. But that’s not what we’re talking about—we’re talking about whether Taubes’ underlying model is correct or not. If his model is correct, one would expect a substantially higher success rate.
Yes, I dispute with the first point.
Well let’s quantify things. I assert that weight loss recidivism rates are well over 80%. What’s your estimate of the percentage?
I assert that weight loss recidivism rates are well over 80%. What’s your estimate of the percentage?
I expect that it is very high also. I think the situation could be better if we understood more.
-- we’re talking about whether Taubes’ underlying model is correct or not. If his model is correct, one would expect a substantially higher success rate.
From skimming through the posts, I’m not sure what Taubes’ underlying model is. Would you summarize it in a couple sentences?
In general, the central and essential problem of successful dieting is dealing with the problem of difficult-to-resist urges;
In your opinion, are these urges physical or cultural/psychological? (I think that a diet that does “work” must handle the ones that are physically based.)
I’m not sure, since I generally don’t use “at all” in positive sentences. Besides, it’s just a question of semantics.
You’re right, it works better to consider the negative context. How many people would the diet need to be effective for before you would not say that the diet does not work at all?
I expect that it is very high also. I think the situation could be better if we understood more.
Then why do you dispute my assertion that the vast majority of people end up regaining regardless of the diet they are on?
From skimming through the posts, I’m not sure what Taubes’ underlying model is. Would you summarize it in a couple sentences?
According to Jack, Taubes’ position is that generally speaking, if people eliminate refined carbohydrates (whatever that means) from their diet, they can eat other foods ad libitum and get and stay thin. Apparently Taubes believes that there is some important and dominant interplay between refined carbohydrates, insulin, and body fat. I don’t know the details of his model.
In your opinion, are these urges physical or cultural/psychological?
Physical.
How many people would the diet need to be effective for before you would not say that the diet does not work at all?
Now that I know what Taube’s model is, roughly, I think that it is probably correct, for some small subset of people (but not vanishingly small). I think a softer version, the effectiveness of a low-carb diet, is true for a larger, significant, but still minority subset of the population. It doesn’t matter what I think though – my point is the qualifier: that a model can be true (or a diet can be effective) for a population subset.
This is why I agree with you that the majority of people will regain their weight—if they are trying a diet at random (like in the study) rather than systematically, and without a working model of how a diet is not a one size fits all type of thing.
It occurs to me that you consider the regaining of weight as evidence that his model doesn’t work, because he also said you could eat as much as you want and stay thin. But there may be some people for whom this is true, even within the study sample. It would be useful to see the individual data.
Now that I know what Taube’s model is, roughly, I think that it is probably correct, for some small subset of people (but not vanishingly small).
Assuming that’s true, it doesn’t contradict what I have said. In fact, now that you concede that the vast majority of people end up regaining regardless of the diet they are on, it seems you do not dispute any of my basic points.
I rather suspect that what’s been presented to you has been made out of straw.
Take that up with Jack.
Me:
But anyway, you seem to be saying that, according to Taubes, if you simply avoid eating refined carbohydrates, you can eat other foods ad libitum and avoid obesity. Is that pretty much it?
If the criteria is that over half won’t benefit, almost every drug on the planet. I wasn’t exaggerating.
Here’s a paper that explains NNT, and on page four there are some examples. NNT under 10 is pretty good no matter what time span or endpoint of interest we’re looking at. Then there’s also NNH.
I see, I guess when I think of ‘most drugs’, I think of common drugs like aspirin and Tylenol that work pretty much for everyone. (But ‘most’ drugs (and treatments) doesn’t mean most ubiquitously used.) I’ve updated, thanks.
I think of common drugs like aspirin and Tylenol that work pretty much for everyone.
That’s not true either. I know a person who on occasion has headaches. They respond well to Tylenol and don’t respond to Ibuprofen or Advil. The latter two are common drugs which do not work for her.
Maybe you can hire someone to read them for you and prepare an executive summary :)
Sure, that’s fair. :)
...curiously, if I did look, would I find that this had been addressed?
I’m not saying you should look. I’m just saying, time spent is not so often rewarded, whereas anyone that has kept with the discussion would readily know.
Furthermore, it helps to keep track of and repeat one’s assumptions. To write, “doesn’t work at all” doesn’t sound right without some qualifications.
Dude, the thread has 39 comments at the moment, two of which are yours. If you are too busy to read 37 comments, then I doubt it is worth engaging.
That’s very funny. I meant that I hadn’t read the previous posts:
At some point, it was probably discussed in which context it was appropriate to look at averaged data...I was hoping for a recap. I have composed a comment about whether I should have read the previous posts in the open thread.
Whatever, if there’s anything I said which you disagree with or would otherwise like to discuss with me, feel free to quote it and respond.
Ok, I wanted to ask about this bit:
If the diet should work for some people, but not most people, why should you say it doesn’t work at all?
To emphasize the fact that even though it might work better than some other diet, it still (generally speaking) does not work.
Anyway, your question is simply about my word choice, right? You do not dispute my substantive point?
There appears to be some substance in the disagreement.
Suppose that there are many different ‘metabolic’ types of people, and different diets work with different efficacies for these different types. Perhaps any diet works well with a small subset of people (we already know some people don’t need to diet). Perhaps particular diets work well with certain subsets of people. Maybe a particular diet works extremely well for one subset, in that they lose weight and keep it off, whereas it only works moderately well for others (they gain the weight back or must combine different diets). This model—which seems quite likely anyway—would imply two things:
the effectiveness of a diet for small subsets of people would be completely obfuscated by averaging the data—for example, in all groups in Figure 2, dieters kept off some of the weight on average. This might mean that some people gained all the weight back, whereas some kept it off, or it might mean that everyone gained some back. (Note that showing standard error, rather than standard deviation, especially hides the role of individual variation.)
it would not be reasonable to expect a diet to work by working for everyone; instead, each person must find which diet, if any, works for them, and then we would say that a diet “works” if it works for any subset of people that would otherwise be overweight
I’m pretty certain that this paradigm is mainstream. Increasingly, we’re recognizing that medical solutions (for example, blood pressure medications or cancer therapies) need to be tailored to individual people.
Here’s what the authors of the 2008 paper (Figure 2) have to say about individual differences:
Gender and diabetes are individual differences that are easy to recognize. Who knows (yet) how many characteristics are relevant for diet. If someone is sensitive to sugar like me (can’t go long periods without eating, tends to put on weight on their belly), the low-carb diet might work for them. Another complication is that changes in gut-bacteria populations over longer time-scales might result in long-term changes in your diet type. For example, over-weight people have different gut bacteria, diets change gut bacteria and gut bacteria might change your equilibrium weight. (The media is dominated by these studies recently.) I find a positive effect—once I’ve been on the low-carb diet for a few months, I’m not as sugar-sensitive and can maintain my ideal weight with a higher level of carbs than before I began the diet. Over time I lose the effect (too many holiday parties?) and need to return to a stricter low-carb diet.
I don’t fault the discussion for not accounting for gut bacteria yet (the diet science literature is huge, and you have to begin somewhere)...but would you agree that it is complicated enough that we can not infer much of anything from average responses to a diet, other than that a particular diet doesn’t work for everyone?
Assuming this is true, I don’t see how it contradicts anything I have said.
Do you dispute any of the following:
The vast majority of people end up regaining regardless of the diet they are on;
In general, the central and essential problem of successful dieting is dealing with the problem of difficult-to-resist urges;
Taubes’ approach is not a general solution to this problem.
Yes, I dispute with the first point. In any case, you can’t infer this from figure 2.
But that wasn’t my initial point. My initial point, asked in a different way, is what fraction of people would a diet have to be effective for, for you to say that it works “at all”? This is a substantive question, not just semantics, because I would like to convince that an expectation of 100% is not reasonable—rather 20 or even 10% would be quite good.
I’m not sure, since I generally don’t use “at all” in positive sentences. Besides, it’s just a question of semantics.
Well it depends on how you define “reasonable” or “good.” In terms of deciding whether a diet is worth trying, a 10% success rate might make it worthwhile. But that’s not what we’re talking about—we’re talking about whether Taubes’ underlying model is correct or not. If his model is correct, one would expect a substantially higher success rate.
Well let’s quantify things. I assert that weight loss recidivism rates are well over 80%. What’s your estimate of the percentage?
I expect that it is very high also. I think the situation could be better if we understood more.
From skimming through the posts, I’m not sure what Taubes’ underlying model is. Would you summarize it in a couple sentences?
In your opinion, are these urges physical or cultural/psychological? (I think that a diet that does “work” must handle the ones that are physically based.)
You’re right, it works better to consider the negative context. How many people would the diet need to be effective for before you would not say that the diet does not work at all?
Then why do you dispute my assertion that the vast majority of people end up regaining regardless of the diet they are on?
According to Jack, Taubes’ position is that generally speaking, if people eliminate refined carbohydrates (whatever that means) from their diet, they can eat other foods ad libitum and get and stay thin. Apparently Taubes believes that there is some important and dominant interplay between refined carbohydrates, insulin, and body fat. I don’t know the details of his model.
Physical.
In what context?
Now that I know what Taube’s model is, roughly, I think that it is probably correct, for some small subset of people (but not vanishingly small). I think a softer version, the effectiveness of a low-carb diet, is true for a larger, significant, but still minority subset of the population. It doesn’t matter what I think though – my point is the qualifier: that a model can be true (or a diet can be effective) for a population subset.
This is why I agree with you that the majority of people will regain their weight—if they are trying a diet at random (like in the study) rather than systematically, and without a working model of how a diet is not a one size fits all type of thing.
It occurs to me that you consider the regaining of weight as evidence that his model doesn’t work, because he also said you could eat as much as you want and stay thin. But there may be some people for whom this is true, even within the study sample. It would be useful to see the individual data.
Assuming that’s true, it doesn’t contradict what I have said. In fact, now that you concede that the vast majority of people end up regaining regardless of the diet they are on, it seems you do not dispute any of my basic points.
I rather suspect that what’s been presented to you has been made out of straw.
Take that up with Jack.
Me:
Jack:
Link:
http://lesswrong.com/lw/je4/critiquing_gary_taubes_part_2_atkins_redux/abjs?context=1#comments
Good point. If we applied the same principle to drugs, very few of them would be used at all.
What examples were you thinking of?
If the criteria is that over half won’t benefit, almost every drug on the planet. I wasn’t exaggerating.
Here’s a paper that explains NNT, and on page four there are some examples. NNT under 10 is pretty good no matter what time span or endpoint of interest we’re looking at. Then there’s also NNH.
I see, I guess when I think of ‘most drugs’, I think of common drugs like aspirin and Tylenol that work pretty much for everyone. (But ‘most’ drugs (and treatments) doesn’t mean most ubiquitously used.) I’ve updated, thanks.
That’s not true either. I know a person who on occasion has headaches. They respond well to Tylenol and don’t respond to Ibuprofen or Advil. The latter two are common drugs which do not work for her.
Advil == ibuprofen. The latter is the active ingredient in the former.
You’re right—I’ve meant Aleve (naproxen), not Advil.
I haven’t looked at their NNTs, but if you compare them to placebo and also take into account the harm, they might seem less amazing.
Interesting, yes, I found in one place that aspirin had an NNT of about 5 for headache.