## Monday, 6 July 2009

### Errata

When Matt and I drafted our report on the BERL study, we sent it on to BERL for comment. Reverse engineering anything like this can't be done perfectly, and we didn't have access to their underlying workings, so we were looking for any bits where we misunderstood what BERL was doing. The only reply we received was that they didn't have time to look at it. So, we released it. Wednesday night, I had the chance to chat with Adrian Slack, the study's main author. He pointed out lots of areas of disagreement. On some of these, like that someone going to a company-sponsored Christmas party imposes an externality on his employer if he shows up at work hung over the next day, we're not going to agree. But, on two points, I think he was right. And so I sent him an email Sunday (or, rather, Matt did as I haven't his email address) telling him so and showing the effect on the final figures. Here are the two corrections.

First, we misinterpreted Rayner on excess unemployment. But, correcting it adds almost nothing to our overall costs. Rayner argues that about 5.15% (he uses measures for prevalence rates of 4.3% and 6%) of the working age population suffers reduced employment rates due to alcohol use. The 2004 working age population was 3,094,400. The employment rate in 2004 was 63%. So, the employment rate among 159,362 individuals was ten percentage points lower, for a workforce reduction of 15,936 individuals. BERL argued for reduced employment of 31,100 workers; we apply Rayner and get about 16,000 (our first estimate was 9,200). BERL gave a cost measure of \$877.5 million. Of that, \$469.25 was forgone earnings and the rest was multiplier. So, to adjust BERL's figure, we just apply the fraction [15,936/31100] to their measure of forgone earnings, [update: and make a minor adjustment for cohort heterogeneity] then inflate by our preferred 1.1 multiplier rather than their preferred 1.87 multiplier. Our costs on this measure then are \$192 million in private costs and \$19 million in external costs (where we'd previously said \$11.1 million in external costs). So, we increase our policy-relevant costs measure by about \$8 million.

There's one other bit worth noting in Rayner, that BERL chose to ignore: Rayner says that applying the full wage costs is the upper bound, not the lower bound. Here's Rayner:
At one extreme, the economy is assumed to be at full employment, in the sense of output being subjected to a labour constraint. In this case, the loss of a person from the workforce means that the total output is reduced by the value of his/her marginal productivity, which will equal his/her gross wage if employers are profit maximizers.

The opposite extreme sees unemployment in the economy caused by demand constraint. The loss of a worker now has little output implications, since he/she can be replaced by another from those unemployed, at the cost of only a small initial retraining.
Rayner goes on to remind the reader that his use of the first assumption is an upper-bound estimate. So adding a 1.87 multiplier would be inappropriate, and so too is our 1.1 multiplier.

Second, we misread BERL on insurance overheads. It makes sense to see those as sitting alongside paid out claims on the right hand side of an equation where total paid premia are on the left hand side. But working to correct this gives us a number at substantial variance to BERL's. BERL lists insurance administration costs of \$132.2 million. At page 123, BERL argues that alcohol-attributed car crashes cost \$45 million in 2005/6; since car crashes make up a majority of all insurance claims, they then attribute half of all overhead costs to alcohol. Let's look to their claimed source, the Insurance Council of New Zealand. In 2006, total claims incurred were \$1,609,672,399; business costs (overhead) were \$762,889,565. Total "motor commercial and private" claims were \$739,416,181. BERL says alcohol-attributed car crashes cost \$45 million. This is 2.8% of all paid claims. 2.8% of total overhead costs is \$21.3 million. If we add in property damage, we get a proportionate share of insurance administration of \$28.3 million. I'm not sure how BERL arrived at its estimate of \$132.2 million. We adjust upwards our cost measure from \$0 to \$28.3 million. I'll apportion all of this as being external assuming that insurance cannot adequately price for risk type on this margin; this is an upper bound assumption. If drunks have to pay more for car insurance, attenuate downwards our measure accordingly (or rather split some off into private costs).

So, total upwards adjustment: \$36 million.

There's one other adjustment we need to make as well though. This one has been bothering me for some time, but I've only been able to check the figure today. I could earlier only find figures on excise taxes of \$516 million. But I've seen them cited elsewhere as being much higher. I knew there was a portion collected by Customs as excise-equivalent duties, but I couldn't find the figure. Sam Direen points me to it (thanks!). I left out excise taxes collected of \$197 million. We increase private costs of tax and external benefits of tax accordingly.

Net errata then: external costs increase \$36 million, external benefits increase by \$197 million. We previously found net external costs of \$146.3 million. The careful reader will note that we now have a positive figure: net external benefits totaling \$37.8 million. We apologize for any inconvenience our overestimate of the net external costs of alcohol may have caused any of our loyal readers. I wouldn't go out on a limb and claim that there are strong net external benefits from alcohol; rather, I'd say that both figures were effectively zero given the margins of error in this kind of work.