« How to Consume Big Data (from McGraw-Hill blog) | Main | Causal thinking »


Feed You can follow this conversation by subscribing to the comment feed for this post.


While a hamburger is processed, it is still red meat. The processed meats all have something done to them involving cooking and usually adding chemicals. That is why they are treated differently.

Harcombe doesn't seem to like the idea of multivariable adjustment, but it really does work. One of the problems in this data is that age varies between the meat consumption quintile groups. Once you correct for age there is a trend for risk with total meat consumption. Correcting for the others has some problems but generally seems safer. I'm not going to think about what all the path implications of all the covariates are.

Basically with these types of analyses, it really is a question of what other covariates are associated with the factor of interest and are we successfully correcting for them and are there some that we missed. Harcombe seems to be against doing these corrections. It is also possible to overcorrect by including covariates on the causal pathway. That is if consuming meat causes the effect then we shouldn't correct for it. This probably applies partly to BMI and total energy, so it may be overcorrecting. There are probably other things that are missed so that the model is under corrected.

One interesting result is that low meat eaters have higher cholesterol. A suspicion is that people with high cholesterol are modifying their diet, and it is having little effect on their cholesterol. It is an interesting possibility as it would reduce the apparent effect of eating meat. Harcombe seems to think that cholesterol is good. Try telling that to 40 year olds who have very high cholesterol and a heart attack.


Ken: The reason why she complained about processed and unprocessed is that the research separately estimated the impact of the two types of red meat. I skipped that detail as it's not really pertinent to my point here.

I don't endorse Harcombe's point of view on multivariate adjustments and she's painting with a broad brush. What I do like is the way she interacts with the data analysis. As a consumer, we cannot afford the time nor have the expertise to replicate everything that researchers have done but we also must not give blind trust.

As you pointed out, multivariate adjustment is based on a model and it is only as good as the model assumptions. It's a challenge when the amount of confounding is so much.


Multivariate adjustments are, indeed, very useful but can be tricky. Note Ken's comments about undercorrection / overcorrection.

Looking at the less adjusted data [Death rate (Z)] I see Q1-Q4 with nothing going on, and Q5 showing a big effect. (a threshold model) It's easy to be suspicious, at least, of an adjustment that converts this to a monotonic, nearly linear effect.


You have to use common sense as well -when you look at past large healthy populations like Japan they mostly ate rice and vegetables. Then when newer generations move to the West or as fast food chains creep in over there they start to get heart disease, cancer and other western diseases. So whether it's a bag of chips, bacon or a big mac you're best to avoid any of them I would imagine.

Low meat eaters don't change their cholesterol because they probably still eat dairy or other types of meat like fish which also have cholesterol. Saturated fats from coconut oil can also make the liver produce more cholesterol. Dean Ornish and others have shown you can reverse heart disease with a plant-based diet.

Criticizing the person for making money or being a vegan is called an ad hominem attack and doesn't usually work in rational arguments.

The comments to this entry are closed.

Get new posts by email:
Kaiser Fung. Business analytics and data visualization expert. Author and Speaker.
Visit my website. Follow my Twitter. See my articles at Daily Beast, 538, HBR, Wired.

See my Youtube and Flickr.


  • only in Big Data
Numbers Rule Your World:
Amazon - Barnes&Noble

Amazon - Barnes&Noble

Junk Charts Blog

Link to junkcharts

Graphics design by Amanda Lee

Next Events

Jan: 10 NYPL Data Science Careers Talk, New York, NY

Past Events

Aug: 15 NYPL Analytics Resume Review Workshop, New York, NY

Apr: 2 Data Visualization Seminar, Pasadena, CA

Mar: 30 ASA DataFest, New York, NY

See more here

Principal Analytics Prep

Link to Principal Analytics Prep