Wednesday, 17 June 2009

Correcting the BERL report part 2: Lost output

In Part One, I provided an overview of our critique of the BERL report on the costs of alcohol. We now hit into a rather large cost item: lost output.

This is going to be a long one, so apologies. There's a lot packed into this section, and a lot that we found needed revision. The condensed version for those of little patience: in this single line item, BERL's estimate of social costs exceeds our measure of net external costs by more than $1,500,000,000.

BERL estimates the number of labour hours lost due to premature mortality, excess unemployment, absenteeism, reduced productivity, injuries, road crashes, crime, and imprisonment. They assume that forgone labour would have earned wages matching the average for the worker's age and gender. BERL calculates forgone earnings under those assumptions, then multiplies the total by 1.87: the ratio of GDP to earnings.

Our first adjustment looks at the multiplier. BERL's multiplier makes sense in a world where a missing worker cannot be replaced by another worker or by adjusting the production process to make it more capital intensive. In that case, all of a missing worker's output arguably disappears with the worker. But is that the world we live in? And, in particular, is it the world most likely to apply to workers most likely to be subject to severe problems with alcohol abuse? Rather than apply a multiplier that requires both full employment and that there is no substitutability between capital and labour, we look instead to empirical estimates of labour "value added" and apply a multiplier of 1.1 in place of BERL's 1.87: we reduce costs of lost labour output to 59% of BERL's initial values. We note that in one category, losses due to injury not elsewhere included, it appears that BERL has not applied its 1.87 multiplier. In that case, we inflate BERL's estimate by ten percent for consistency. Ganesh Nana argues that differences between BERL's findings and ours depend on differences in worldview; we're curious what worldview is consistent with the 1.87 multiplier, though Matt discusses his suspicions at section 5.1.3 of the report.

Second, we believe BERL has overestimated excess unemployment due to harmful alcohol use. BERL cites Rayner et al (1984) as showing excess unemployment due to alcoholism as being 10% and interprets this as indicating that alcoholics have an unemployment risk ten percentage points higher than average. We believe the more accurate interpretation of Rayner is that overall unemployment rates may be ten percent higher than they otherwise would be: in other words, the difference between an unemployment rate of 4% and 4.4% for the population as a whole. BERL argues that excess unemployment due to alcohol use affected more than 30,000 workers; total unemployment in 2004 was 92,000. Is it plausible that a third of the unemployed are hard core alcoholics? We don't think so. If our interpretation of Rayner is correct, then unemployment in 2004 was ten percent higher than it otherwise would have been: 9200 workers count as excess unemployed, not 30,000. We reduce BERL's listed costs on this item proportionately.

Third, we adjust BERL's estimate for losses associated with lost output from prisoners and home detainees. BERL uses a survey of prisoners, to which we do not have access, which asks prisoners to what extent alcohol use contributed to their current incarceration. Possible answers include "not at all", "a little", "some", "a lot", or "all". BERL then argues that, if there were no harmful alcohol use, all crimes listed where alcohol contributed at least "some" would disappear. This seems implausible. Of course, if there were no harmful use of alcohol, crime rates would drop. But surely not each and every crime where offenders were willing to say alcohol contributed at least somewhat to their offending. We here make a somewhat arbitrary adjustment, but one that we think makes more sense than BERL's estimate that one hundred percent of these crimes would stop. We say instead that only two-thirds of non drink-driving offences would be stopped if we could eliminate harmful drinking; for drink-driving, we make no adjustment. Crime costs overall then drop by 20%.

Next, we adjust lost output figures for underlying heterogeneity. All of BERL's wage figures depend on the assumption that, but for alcohol, hard core alcoholics and prisoners would have labour force characteristics identical to the rest of the population. The biggest adjustment here is in the crime category. Australian data suggests that the prison population is, unsurprisingly, fundamentally different from the overall population. Giles and Le (2007) find that, a month prior to incarceration, prisoners had labour force characteristics consistent with output levels less than half of those of the rest of the population: in particular, they suffer from much lower employment rates - 47% of rate for the general population. If those incarcerated have much lower employment rates than the rest of the population, we overestimate their forgone earnings if we assume them to have average labour force characteristics. We scale back BERL's estimate on prisoners' forgone labour accordingly. We believe that our number remains an overestimate of the costs of prisoners' forgone earnings: many prisoners work while in prison, and the value of this labour is nowhere counted. BERL says we do not need to worry about this as prisoners are not paid very much; however, we would not expect what is essentially slave labour to be paid anywhere near its marginal product. The value of prisoners' output could easily change the sign of this cost item.

We also scale back forgone earnings from non-incarcerated harmful drinkers. If it is the case that underlying personal characteristics correlate with both adverse labour market characteristics and with severe alcohol misuse, then we cannot assume that forgone labour hours would have been of value equal to the average for age and gender. There's very good evidence from the psychological literature of comorbidity between alcoholism and other mental disorders, with reasonable evidence that mental disorders are prior to alcoholism. We conservatively look only at the likely effects of comorbid depression because that is where we can find reasonable studies with both alcohol comorbidity and effects on employment status. Why does this matter? If harmful alcohol use disappeared, those currently using alcohol harmfully would not suddenly share average labour market characteristics: for many of them, alcoholism is a symptom of deeper psychological problems that themselves seriously adversely affect employment. We follow Pirkola (2005) and assume that twenty percent of harmful alcohol users have a comorbid depression, anxiety, or combined depression and anxiety disorder, and that those individuals would have median incomes 19% lower than average if employed and unemployment rates six times higher even in the absence of alcohol use. The adjustment makes little difference in the overall figures: about $55 million all told. But we thought it important to give more serious consideration to the counterfactual.

Next, BERL lists an offsetting benefit of premature mortality: those dying early consume fewer resources that then are available for use by others. This is, of course, purely a pecuniary externality: the benefits to others are matched by costs to the person suffering premature mortality. We adjust BERL's figure downward slightly to account for cohort heterogeneity as discussed above, then add an internal cost equal in magnitude but opposite in sign to the external benefit.

Finally, we split the corrected costs between those that are internal to the drinker and those that are truly external. As noted in part one, we have good reason to worry about external costs; policy measures aimed at reducing internal costs are far more contentious. We count as external costs the "value added" component of earnings (our 1.1 multiplier) and all earnings forgone by crime victims. Splitting the costs of forgone earnings by those killed in drink-driving incidents is a bit more difficult. If a drink-driver kills only himself in a single-vehicle accident, his lost future income is a cost to himself. If a drink-driver kills only an innocent pedestrian, that pedestrian's lost future income is an external cost. But we know of no New Zealand data showing the proportion of deaths from drink-driving that accrue to drink drivers and those accruing to parties entirely external to the vehicle. So, we instead turn to US data. The NHSTA says that 17% of those killed in drink-driving accidents are persons outside of the vehicle. We consequently apportion costs of drink-driving fatalities as being 17% external and 83% internal.

On this single line item, correction for incorrect multipliers, cohort heterogeneity, and inaccurate interpretation of excess unemployment reduces BERL's listed costs by a billion dollars: from $1,478.4 million to $416.6 million. If we then add back in consumption forgone as a cost to those not enjoying the consumption, total costs rise to $716.2 million. External costs: the costs that are relevant for policy, are -$126.7 million: in other words, an external benefit. For this line item, if we consider only externalities (as we ought to), we find a net benefit rather than a net cost. In short, the external costs of harmful alcohol use are exceeded by the pecuniary externality of consumption forgone.

If we worry about social costs - the costs imposed on others by harmful drinking - BERL overstates things on this single line item by $1.6 billion: the difference between their $1,478.4 million in costs and the more likely external benefits of $126.7 million.


  1. This may be a silly question, but here goes: You say that 'If a drink-driver kills only himself in a single-vehicle accident, his lost future income is a cost to himself.' Isn't it also a cost to others, such as the dead person's dependents, so therefore not a purely internal cost?


  2. Not a silly question. Actually, it's a pretty tough one. My first answer is that life insurance solves it. For example, my wife won't allow me to have a motorbike unless I substantially increase my life insurance policy: she wants extra compensation if I die for a stupid reason. My second answer is that once we start allowing welfare considerations of intrafamily externalities that second-guess the bargains made within households, things get very murky very quickly. There are all kinds of margins on which couples impose costs on each other; this is all part of a very complicated equilibrium.

    That said, there are all kinds of things government can and should be doing to reduce deaths from drink driving besides raising alcohol taxes (very indirect, not terribly effective, imposes large costs on moderate drinkers) or lowering the drink-driving limit (does nothing to deter the folks who already are driving drunk at current limits and who currently are not deterred). Some off-the-cuff potential starters: mandatory third party insurance, having ACC not cover losses to drink drivers, car confiscation for drink drivers who go on to drive while suspended, more blitzes in areas where there's a high probability of catching drink drivers rather than just random draw, mandatory ignition interlocks (breath testers) for convicted drink-drivers' cars for a reasonable period of time post conviction....

  3. I would imagine that the kind of person who is drinking to the point of serious harm isn't very likely to be the kind of person who takes out a life insurance policy, irrational as that may be.

    And I agree, that there are lots of other policy interventions such as you list that would probably be effective in reducing drink-driving deaths. I just think that maybe only noting strictly external costs perhaps over-simplifies the 'very complicated equilibrium'. Afterall, a net external social cost of $146.3 million isn't a terribly strong case for policy intervention at all in the grand scheme of things it seems to me.