More sweet silliness

A new paper suggests strong public health benefits from junk food taxes in Australia. Blogging has been light as I've too much on deck in the day-job currently, but since a few people have emailed me asking about this one, I might as well put here what I've told them.

The effects found in the paper depend on how consumers respond to price increases. Those elasticities are drawn from a 2013 paper estimating elasticities in New Zealand using an Almost Ideal Demand System setup across a couple dozen food categories in household expenditure survey data. But responsiveness of spending to changes in average prices doesn't tell us about consumption if people can shift to cheaper brands within the same category, as John Gibson's Marsden-funded work shows.

And I expect the problem will be worse due to the apparent coarseness of the categories where they’ve aggregated up to 24 total categories: unobserved shifts within broader categories have more potential for problems than within very fine categories. That isn't their fault as likely due to the level of aggregation at which they have average price data, but it would make a mess of things.

If you’re looking at salt intake and the like, and there’s huge variability within the broad categories, that’s going to be a mess. I remember once when Jennie Connor claimed there’d be huge health benefits from alcohol minimum pricing based on some within-alcohol-category elasticity estimates out of BC (own-price elasticity of wine, beer, and spirits, ignoring cross-price elasticities), but increases in wine prices have people shift to beer or spirits; you need to look at aggregate alcohol consumption in response to things like excise changes to get a handle on it (where excise is a de facto minimum price barring implausible loss-leader stuff). So while the paper accounts for cross price elasticities across broad categories, we don't know what's going on to composition mixes within categories.

So they start from overestimated effects of price on quantity consumed. They then go from quantities to dietary intake, although across the broad categories in some cases I’m not sure how they do that (milk, yogurt and eggs is a single category in the 2013 paper; sauces, sugar and condiments is a category; “restaurant food” is a category; “ready to eat” food is a category, though fruit is a single category). Not sure how they estimate change in salt consumption across the categories where the categories are broad and there’s potential for compositional shift within categories.

I also wonder whether the better approach to this stuff might be to look to direct effects rather than chains of dubious link quality? Look at Gelbach, Klick & Stratmann for example. They went directly from prices to obesity and ignored the consumption chains. Andrew Leigh (Top Australian economist, now also Labor MP) blogged on it when he was at ANU.

Finally, I'm not sure that the paper's cost-of-implementation estimate is anywhere close to right. It comes out of estimates on the costs of changes to alcohol excise. But there's an established regime for alcohol excise and the incremental administrative costs of changes within that won't be large. Plus that’s just administrative cost – zero consideration I can see of forgone consumer surplus. Like, even if you’re balancing it out with healthy food subsidies, there’s a reason people are choosing to eat the tastier things currently.

Caveat: this is just a quick take on the paper. If others have caught problems that I've missed, or if I've misread them, please let me know in comments.