Tobacco excise incidence

The latest paper by Callison and Kaestner* makes me a bit more worried about the incidence of Tariana Turia's proposed excise increases. Recall that the MoH is modelling things using a fairly high estimate of price elasticity of demand. At paragraph 55 here, MoH says:

the price elasticity of demand (ie. the extent to which each percentage increase in the tobacco price causes consumption to fall) - current estimates are that each 10% rise in tobacco prices will engender about a 4-5% drop in tobacco consumption, but based on recent Treasury revenue returns and behavioural research by Auckland University, there is emerging evidence that price responsiveness may be increasing.

Presumably the -0.4 to -0.5 elasticity estimate is what they had NZIER use in their analysis; MoH seems to be suggesting this is a lower bound.

Callison and Kaestner start by surveying the existing literature.  My clipping from their survey:**

• Lewit and Coate (1982): adult smoking participation elasticity (age 35+) of -0.15; younger smokers more elastic.
• Evans and Farrelly (1998); Farrelly et al (2001): same dataset as above; younger smokers respond to price changes; adults aged 40+ not responsive.
• Wasserman (1991): Adult participation elasticity -0.17 in 1985.
• Ohsfeldt (1998): Participation tax elasticity for teen and young adult males of -0.15 to -0.22, but only -0.07 for males over age 45.
• Tauras (2006): Same dataset as Ohsfeldt but more recent years' coverage: participation elasticity among adults 18+ of -0.12.
• DeCicca and McLeod (2008): Post-2001 data. Participation elasticity of -0.3 for 45-59 year olds and -0.2 for 45-64 year olds. 
• Gallet and List (2003): A survey of existing elasticity estimates giving a price elasticity of demand (I read this as combination of participation and intensity elasticity) of -0.32 for adults aged over 24.

When O'Dea ran his analysis, he used a price elasticity of demand of -0.5 and a participation elasticity of -0.2. 

Callison and Kaestner use a difference-in-difference method comparing smoking prevalence in states with large excise increases with those in states that didn't increase taxes. Among adults, they found participation and price elasticities of demand on the order of -0.02 to -0.05: a 10% tax increase reduces consumption by -0.5% or less, not 2% and not 5%. They write:

Considering all the evidence, we conclude that there is insufficient justification for the widespread belief that raising cigarette taxes will significantly reduce cigarette consumption among adults, even young adults. Our evidence suggests that, at best, increases in cigarette taxes will be associated with a small decrease in cigarette consumption and that it will take very sizeable tax increases, on the order of 100%, to decrease smoking by as much as 5%.

Maybe that's why the Australian Treasury modelled tobacco excise revenues on the assumption that excise revenues will be increasing.

Recall that, on an assumption of a -0.2 participation elasticity, O'Dea reckoned that, for a 20% price increase, 129,200 households in the bottom four deciles that continued to smoke would suffer losses of $396 per household; these were balanced by gains to 5,400 quitting households of $2,988 per household in cash savings (no accounting for reduced consumption benefits for those who enjoyed smoking) plus 2 Quality Adjusted Life-Years in each household. They value a QALY at $50k. So losses to non-quitting households for the 20% tax increase total $51.2 million while gains to quitting households total $16.1 million plus $540 million in QALY benefits. 

Callison and Kaestner say that the participation elasticity isn't -0.2, it's -0.05 to -0.02. What happens if you simply halve O'Dea's participation elasticity estimate to -0.1? Then half as many households quit: we have 2700 quitting households (in the bottom 4 deciles) enjoying aggregate cost savings of $8 million  plus $270m in QALY against 131,900 non-quitting households paying an extra $52.2 million in tax. If we go all the way to the -0.02 estimate, then a tenth as many households quit: 540 quitting households gain a total of $55.6 million (including QALY) and 134,060 non-quitting households suffer aggregate losses of $53.1 million.

The O'Dea numbers are calibrated around a 20% price increase from 2005 prices and prevalence levels that then obtained; MoH has been talking about excise increases well in excess of that. I'd be nervous about applying the Callison and Kaestner estimates to New Zealand as total prices here are well in excess of anything they'd have in the US data. On the other hand, non difference-in-difference estimates from the US aren't crazily out of the ballpark on participation elasticities: O'Dea was using -0.2, which is well in line with the US studies that used somewhat less sophisticated techniques. 

Note as well that if reduced effective income hurts QALY measures, then we perhaps need to add QALY costs on the non-quitting side as the poorest households effectively get poorer. 

Another cool bit of the Callison and Kaestner paper: they use only MSA data and correct for distance to nearest lower-tax jurisdiction to try to correct for leakage from smuggled product - the find little effect on their estimates from cross-border purchases. But if it is the case that smuggling rings are a bit more organized and folks with semi-loads of smokes from lower-tax jurisdictions don't much care whether they're driving 50 miles or 250 miles, then smuggling might still confound their measure: it probably well captures individuals' decreased likelihood to drive out of state to get smokes, but it might not catch organized groups handling the transport. As each state has its own particular tax stamp for cigarette packets, such opportunities may be limited. 

If New Zealand goes ahead with very large tobacco excise increases, the government might consider some offsetting income tax cuts for lower decile groups. Callison and Kaestner didn't split their sample by income decile rather than just by age group to see whether price responsiveness varied by income; price elasticities split by income could be important in assessing overall incidence of the NZ changes. But as smoking is fairly concentrated among lower decile groups, overall estimates likely aren't far from estimates restricted to lower decile groups. 

It could also be fun to have iPredict run markets on aggregate tobacco excise revenues over the next few years. 

* Previously noted here.

** Note that the participation elasticity is the percent change in the proportion of the sample who report smoking given a percent change in the price. So if a 10% price increase is associated with 5% of the sample quitting, the participation elasticity would be -0.5. MoH above is using a total consumption elasticity that combines those quitting with reductions in smoking among those who don't quit entirely. Callison and Kaestner also list some of the smoking intensity elasticities: the percentage change in the amount that you smoke (conditional on your smoking) given a percentage change in price. These values are usually about half of reported participation elasticities.