Which of the following sounds more like the real world?
In world A, you walk down the grocery aisle. You want to buy some soda. You see that the price of your favourite brand of soda has gone up. You decide not to buy any soda and you consequently drink less soda.
In world B, you walk down the same aisle with the same intention. On seeing that the price of your favourite brand of soda has gone up, you look around to see if it’s available at a lower price point in cases of cans, or if a different comparable brand is available, or you give the discount brand a second look. Or, you decide to hold off that week because you still have some stored in the cupboard from last time it was on special and you can buy a few cases next time it’s on special. Either you spend a bit less to drink the same amount of a discount brand, or you don’t spend anything this week but still drink about as much as you otherwise would have because you’re drawing down your stocks.
Turns out that most of the studies that try to estimate the effects of soda taxes assume we live in world A. I live in world B, and suspect you might too.
Waikato Professor of Economics John Gibson provided a superb presentation on the problem at the Ministry of Health at lunchtime today.
How effective a fat or soda tax is at curbing consumption depends on how responsive people are to changes in prices. If people are not very responsive to changes in prices, taxes will not do very much to change consumption.
Most measures of price responsiveness depends on very poor data. For example, in New Zealand, studies will use the Household Economic Survey’s measure. The HES provides an income share measure: what proportion of your weekly family expenditure went to fizzy drinks, or to meat, or to any of a wide variety of other categories. That, combined with a price series on the average price of sodas, is used to get an assumed measure of quantities consumed. So when average prices go up, total family expenditures on the fizzy drink category get divided by a higher average price. The change in quantity consumed then comes out of that.
But while that can work well if the product is very specifically defined, like “6 pack of 330mL cans of Coke Zero”, it stops working if the category is “fizzy drinks”. Why? People respond in two ways to price hikes. They can reduce the total quantity consumed, but they can also change what they buy: shifting from more expensive products to packaging that costs less per unit (say, an 18-pack of cans instead of a single can for a lower unit price), or shifting from a more expensive brand to a store brand. All of the HES-derived figures assume zero change in the quality of what is purchased and consequently assume all of the movement is on the quantity side.
John points out that this is hardly a problem unique to New Zealand. Hundreds of studies do this and consequently overestimate the actual responsiveness of quantity to prices. Economists at least are typically well aware of the problem: it’s been pointed out at least as early as 1955.
Deaton proposed a potential solution – which was about as good as you could get without direct observations on all of unit values, prices and quantities. Under a few restrictive assumptions, you can estimate quality changes by looking at the unit prices paid by households of different incomes.
Gibson uses data from Vietnam where they have unit values, budget shares, and prices for 45 food and beverage groups to look at how biased price elasticities are when they don’t control for changes in quality. Using the standard method, Gibson found price elasticities around -0.8. Correcting properly for quality changes, which his data lets him do, measured elasticities dropped to about -0.20. In other words, the true price elasticity of demand is a quarter of what typically gets reported. And while the Deaton solution helps somewhat, it remains pretty far off.
Here’s the key slide. The “Unrestricted elasticity of quantity with respect to own-price” shows Gibson’s elasticity estimates that recognise quality shifts – again, this is only possible where you have all the data, and we typically don’t. The second line shows the price elasticity you’d get if you used the standard method which assumes that nobody makes changes along a quality dimension when prices change. The third line gives the elasticity measure you’d get using Deaton’s method.
Your bottom line: sugar taxes will probably do about a quarter as much as you might have previously thought in changing consumption. Sure, this is data out of Vietnam, but note that opportunities for quality shifting are larger in richer and more developed economies, because there will be more brands at more price points. And where opportunities for downshifting stop if you’re already on the lowest possible price point because you’re very poor, that constraint will bind less in richer places.
And worry too that all of this still overestimates the price responsiveness of consumption as compared to purchases where people can store durable goods like soda.
Jenesa pointed to these problems in her report, citing Gibson’s then-in-progress work. There’s a lot of wishful thinking in the public health sector. One prominent public health researcher/activist (Chatham House) argued at the MoH presentation that even if taxes didn’t change behaviour that much, they still could be worth trying as they can’t really do harm. Where do you start….