Thursday, 4 July 2013

Limit fast food outlets?

Should we blame fast food outlets for obesity?

Canterbury student Alice Robertson thinks so. The University's press release on her internship project has been picked up in a few places (Herald, Press). Alice's paper isn't yet available, but I've been promised a copy when it is. It sounds like it's mainly a literature review; she notes Day and Pearce, 2011 as particularly relevant. That paper found clustering of fast food outlets around schools. Such clustering doesn't prove an obesity link, just that the kinds of places that wind up being decent venues for primary and intermediate schools are also the kinds of places where fast food outlets wind up locating.

Robertson suggests limiting the number of fast food outlets near schools.

Rachel Webb is a doctoral student in our Economics department. She's presenting some of her thesis work at this year's NZAE meetings. She hasn't sought any press releases on her work because she likes making sure everything's nailed down before talking to the press office. And so her paper still says "don't cite without permission". But she's said it's ok for me to post on what she's been up to.

She's investigating links between obesity and high birth weights. In her quest for instruments that might correlate with obesity risk but that should not have any independent effect on high birth weight risk, she thought about fast food venue concentration. There's some evidence that such venue concentration affects obesity. If if doesn't independently affect high birth rate risk, then it can be an instrument (subject to the usual validity tests).
The density of different categories of dining establishments with a particular focus on fast food restaurants within the Territorial Local Authority (TLA) area that the woman resides in comprises my next set of instruments. A significant relationship between fast food restaurant density and obesity has been a prominent finding by health researchers over recent times. Rosenheck’s (2008) systematic review of 16 studies concludes there exists a significant relationship between fast food restaurant density and obesity[21]. It is generally agreed that fast food proximity lowers the notional cost of eating high caloric food and can therefore lead to higher obesity risk though the causality of the relationship is disputed [22]. It should not have any direct effect on high birth weight risk. However, like with rurality, there are plausible factors which could correlate with both food venue type and concentration and high birth weight risk. For instance, if unhealthy food venue options tend to concentrate in areas where people tend to be less health conscious for reasons that transcend deprivation level, ethnicity, age, rurality, or wider region then food venue type and concentration may have an avenue of correlation with high birth weight risk outside of the effect on obesity that I am unable to control for and could invalidate its use as an instrument. It is also possible that food venue type and concentration may be correlated with high birth weight risk through the effect of weight gain during pregnancy. Validity tests are required to check the soundness of this instrument.
So Rachel wasn't interested in the effects of fast food restaurant density on obesity per se: she was just looking for plausible instruments. And she's found something rather interesting.
A curious finding from the first stage results was that the fast food restaurants density in a TLA did not have the expected effect on obesity measures. The majority of the fast food chains showed consistently negative coefficients in the first stage and particular chains such as Hell’s Pizza, Burger Wisconsin, and McDonalds frequently showed a significant negative relationship with the propensity to be overweight, obese, and morbidly obese. KFC and Pizza Hut were the only chains to have a generally consistent positive relationship with obesity risk. It is not clear what is driving these findings as both median income of the TLA and the deprivation level of the meshblock have been controlled for suggesting it is unlikely to be socio-economic status, nor could it be the effect of living in urban areas as rurality variables are also included. The overall number of fast food establishments per person in a TLA was generally insignificant so it doesn’t appear to be driven by substitution away from less healthy options such as fish and chips either. More research into the effect of fast food on obesity is warranted. 
So, fast food restaurant density, in her regressions on New Zealand data, tends to reduce the prevalence of obesity. When she'd first presented this to the Department, I'd wondered whether what she was picking up was that folks hitting McDonald's would otherwise have been going to a fish'n'chip shop and eating even worse food; she's checked that, as noted in the blockquote, and that wasn't driving things.

Rachel wouldn't draw policy conclusions from her thesis work. But I'll draw one: we shouldn't be too hasty to ban fast food outlets near schools. I'll draw a second one: had Rachel sought press releases about her work, there would have been less uptake. There's reasonable media demand for panics about fast food restaurants, and about alcohol, and about "the kids these days".


  1. The story I read in the herald about the interns study just had this recommendation about limiting the number of fast food outlets. It didnt even explain what the study was or what the findings were etc. I found it very bizarre.

  2. Idle speculation..

    Kids who graze at such places to and from school and maybe during the lunch hour will usually be *walking* with quite a few kilograms in a backpack. The exercise might be burning off at least some of the extra calories.


  3. Regarding the paper, I would check whether maybe the funny results for the fast food restaurants are due to collinearity. What happens when one uses a summary measure of fast food restaurant density?

    More generally, I have a problem with the interpretation of the results that I have with most IV papers. IV regressions (if they work) measure the effect of the endogenous regressor on the dependent variable *as influenced by the instrument*. I'm no expert, but it seems obvious that some of the correlation between mother's and child's weight is due to genes. The analysis in question, in contrast, measures the effects of maternal overweight to the extent that maternal overweight is influenced by the specific environmental channels that the instruments measure. This makes the paper more interesting from a policy perspective. For example, if we knew that the causal influence of maternal on babies' weight were through genetic channels only, this should lead us to not worry (in this respect) about the harmful effects of these environmental factors.

  4. She notes that there's no particular difference when using the summary measure, but I'll check whether she used that without the individual restaurant controls. There are just so many more fish and chip shops and other restaurants relative to McD's that I have a hard time seeing substantial collinearity between the McD numbers and the overall figure...

  5. Must have overlooked the bit about the summary measure.