Hard to review via Dropbox as it opens in the forum browser. Some quick feedback
You make repeated statements about correlation but you need to preface that by saying what data sets you are correlating and what your methodology is, up front. And you need to deploy better statistical measures than R^2 as that really is basic.
Fat
Don't agree that your first bullet point is proved by your argument. Eg what if eating fat didn't in any way diminish our eating of carbs and protein? It does, but you need to make the point about satiety. See Taubes.
I just noticed the value for Beer and it's insanely low. That value is just plain wrong. Possibly they are seeing the suppressing effect of alcohol on GNG. What time period do they sample FII at?
I think calling protein a Black Swan is hyperbole. Black Swans are something that almost never happen. Protein is mundane. You'll need a better simile for your title I'm afraid.
I think you are missing a point about financial modelling and back fits. While it is not a sufficient condition for a model to back fit the historical data, it is still one of the necessary conditions.
You need to incorporate the prompt response to protein in your discussion which only mentions the slower GNG response.
Based on the discussions here I was expecting you to make the point more that certain refined and purified foods are highly glycemic whether they are protein or carbs. Maybe that's a more important variable to consider than carbs vs protein.
Your correlation data is across a very arbitrary data set. How do you weight the component foods in the data set? It doesn't make sense to weight them equally. Not if you are using the equation that you are testing the correlation for, to apply to real world decisions on injecting insulin. It's a good start to look at the correlation on aggregate, but you need to look at specific cases. What is the worst case deviation of your formula from FII? What's the worst quartile, the SD, etc? How many predictions would be off, and how far off? What's the area under that error curve? And crucially, is it larger or smaller than ignoring protein?
I think it is overstating the case to say that the FII data "proves" fat never requires insulin. It's evidence I guess, but there is also evidence to the contrary.
I would suggest you not say multiple times that people who don't get dosing for protein "have a problem" etc. Straw man, ad hominem, etc. Sure some are not aware, and many find it's not the case for them, but there isn't a cabal of GNG-deniers out there. The truth is more prosaic than that. Just people disagreeing about the relative importance of protein, either personally or in general. It does not strengthen your argument to make ad hominem statements about your opponents, real or imagined
> You make repeated statements about correlation but you need to preface that by saying what data sets you are correlating and what your methodology is, up front.
Noted. I have fleshed out the background a little better.
> And you need to deploy better statistical measures than R^2 as that really is basic.
I have added Person’s r and p values
Fat
> Don't agree that your first bullet point is proved by your argument. Eg what if eating fat didn't in any way diminish our eating of carbs and protein? It does, but you need to make the point about satiety. See Taubes.
Noted. Have elaborated. Not the main point of this article, but good to clarify.
> I think calling protein a Black Swan is hyperbole. Black Swans are something that almost never happen. Protein is mundane. You'll need a better simile for your title I'm afraid.
Hyperbole noted. Have pulled back the terminology to ‘outlier’.
> I think you are missing a point about financial modelling and back fits. While it is not a sufficient condition for a model to back fit the historical data, it is still one of the necessary conditions.
Yes, but a simpler system is often more robust compared to one that uses the ‘kitchen sink’.
You need to incorporate the prompt response to protein in your discussion which only mentions the slower GNG response.
> Based on the discussions here I was expecting you to make the point more that certain refined and purified foods are highly glycemic whether they are protein or carbs. Maybe that's a more important variable to consider than carbs vs protein.
This is somewhat addressed by considering fibre, however I’ve highlighted it as an issue.
> Your correlation data is across a very arbitrary data set. How do you weight the component foods in the data set?
I’ve just tried to look at weighting based on fibre (1x), carbs (1x) and protein (0.54x). I have looked separately at dairy and meat and there doesn’t appear to be anything special about these products that can’t be explained by these variables, at least with the data that we have.
> It doesn't make sense to weight them equally. Not if you are using the equation that you are testing the correlation for, to apply to real world decisions on injecting insulin. It's a good start to look at the correlation on aggregate, but you need to look at specific cases.
This is where we come back to designing a system that is too complex to be robust or viable to be applied by people in real life. Going beyond carbs is already a stretch for most.
> What is the worst case deviation of your formula from FII? What's the worst quartile, the SD, etc? How many predictions would be off, and how far off? What's the area under that error curve? And crucially, is it larger or smaller than ignoring protein?
The reality is that calculating dosages for insulin will always be inaccurate to some degree. Insert Bernstein’s ‘law of small numbers’ argument. The way to manage this conundrum is to keep the insulin load low to try to stay off the roller coaster.
> I think it is overstating the case to say that the FII data "proves" fat never requires insulin. It's evidence I guess, but there is also evidence to the contrary.
I have intentionally used “indicates” rather than “proves”. What I’m doing in this article is eliminating variables that have limited use in application to design a robust but still simple system.
> I would suggest you not say multiple times that people who don't get dosing for protein "have a problem" etc. Straw man, ad hominem, etc. Sure some are not aware, and many find it's not the case for them, but there isn't a cabal of GNG-deniers out there. The truth is more prosaic than that. Just people disagreeing about the relative importance of protein, either personally or in general. It does not strengthen your argument to make ad hominem statements about your opponents, real or imagined.
I’m on the ketogains Facebook page where many seem to think that you just can’t overdo protein. One person said “You’re wrong! If there was a book Gluconeogenis for Dummies I’d buy it for you.” I’m over it now… I’ve taken out the strawman argument of the article. Hopefully the data stands on its own without me having to over emphasis the point.
Thanks again Spiker for the feedback.
https://www.dropbox.com/s/cpvh2cmrmezmuko/Black swans, outliers and reliable trends.docx?dl=0