Zombies are hard to kill. Since the classical zombie only really occurs in fiction, accounts vary. But it never seems easy. Things that would kill or at least stop any normal living creature barely seem to faze the undead.

Zombie statistics are at least as hard to kill. These statistics, despite being horribly unsound, insinuate themselves into public debate and stay there. Tim Harford’s excellent BBC series More or Less makes a running feature of the statistics that, no matter how often you think you’ve cut them down, pick themselves up and lumber on.

I was very surprised to see one of New Zealand’s worst zombie statistics come back to life a couple of weeks ago. Put to the grave almost a decade ago, it returned at Alcohol Action New Zealand’s conference in August. BERL’s Ganesh Nana was reported to have claimed that alcohol-related harm costs every New Zealander $1,635 per year, for an annual total of $7.8 billion.

The figure seems to be an inflation and population-growth update of BERL’s 2009 estimate of the social costs of harmful alcohol use. The figure was nonsense at the time; an inflation adjustment to nonsense is hardly an improvement.

How it came back
Let’s go through just what went wrong in that prior study, and why I was surprised to see its resurrection.

The Ministry of Health and the Accident Compensation Commission hired BERL to tally the social costs of harmful alcohol use. BERL said those costs amounted to $4.8 billion, including $1.52 billion in costs of loss of life, $1.48 billion in labour costs, $699 million in drug production costs, and $290 million in health care costs.

But the figure was not sound. Infected if you like.

To begin with, and most obviously, BERL counted as a social cost every dollar spent on alcohol by any drinker consuming more than about two pints of beer per day. That provided just under $700 million of BERL’s headline figure. Drinking that much wouldn’t be great for your health, but does it seem reasonable to consider someone else’s spending on beer as some kind of cost to the country? Drinking just over two pints per day hardly makes you an alcoholic; that those drinkers voluntarily part with their own money to buy their drink suggests they get some benefit from it that should be counted.

A social cost?
BERL’s initial work even counted excise taxes paid by heavier drinkers in that tally; they later updated their figure so that drinkers’ excise payments would not be counted as a social cost.

It gets worse.

New Zealand’s Ministry of Transport assesses the value of a statistical life. That figure matters for policy: when the government is trying to decide whether it makes sense to put median barriers on highways, it needs to assess how much it is willing to spend to save a life.

The Ministry of Transport’s estimate assesses all the costs of a fatality, including lost productivity. So when MoT produces a cost-benefit assessment for a roading improvement, it will include the benefits of fewer lives lost using its measure of the value of a statistical life, and the productivity benefits of people being able to continue working instead of recuperating in hospital. The productivity benefits of avoiding fatalities are already included the value of a statistical life and are not then counted twice.

Double counting lives
BERL added together the value of statistical lives lost using the Ministry of Transport’s figure, and lost output due to premature mortality. Including both is double-counting.

Alcohol-related crime is an obvious and real social cost, but BERL’s method for adding it up was rather questionable. The government surveys prisoners and asks whether alcohol contributed to their offending. If a surveyed prisoner said that alcohol was responsible for “some”, “a lot”, or “all” of their offending, the entire cost of that crime was attributed to alcohol.

Similarly, while heavy drinkers incur obvious health care costs, estimating those costs isn’t easy. BERL used Australian work by Collins & Lapsley as model. That Australian work estimated the burden imposed by alcohol across a huge range of medical conditions. But, for some conditions, Collins and Lapsley said that alcohol provided a protective effect: alcohol reduces the burden of cardiovascular disorder and diabetes, for example.

So Collins & Lapsley provided a figure that added up the costs of all of the disorders where alcohol imposes a burden, and netted out the benefits from reduced costs where alcohol use saves the health system money. BERL decided to follow a somewhat different method: they zeroed out every disorder where alcohol reduces costs.

Even leaving aside the perhaps more philosophical issue of whether policy should weigh the costs that we all impose upon ourselves through our own decisions, BERL’s method included substantial methodological flaws that worked to provide them with a rather large figure.

Just a quarter of the cost
Matthew Burgess, Brad Taylor and I attempted to correct some of these errors to see how much of the figure could be considered as policy-relevant using standard analysis. We estimated that less than a quarter of BERL’s figure could count as policy-relevant external costs after adjusting for double-counting and after setting to one side costs that drinkers impose upon themselves. While those latter private costs are certainly real, they cannot reasonably be included in a measure of net social costs without taking some account of the benefits enjoyed by people who drink more than a couple of pints of beer per day.

Even a well-constructed social cost figure would not tell us anything about appropriate policy. For that, we need measures of the cost-effectiveness of any proposed policy. There can be policies that are remarkably cost-effective even if measured net social costs are not particularly high, and policies that are startlingly ineffective even if net social costs are huge.

Suppose someone tallied up the social costs of boy-racers – including the amount they spend on their cars and on petrol. If anyone then proposed hiking the petrol excise tax to reduce those social costs, we would rightly consider it ludicrous – even if making speeding more expensive might reduce racing by a bit. Speeding tickets seem a bit better targeted.

How to react to Zombies
BERL’s number seemed dead. The figure had drawn a fair bit of ridicule in the economics community. But now it’s back.

The best part of any zombie story isn’t the medical mechanics of the undead. It’s how the rest of us react to the zombies, and how society defends itself against them.

Big social cost figures are not designed to inform good policy. They instead whet public demand for action – and especially when costs that people impose upon themselves, like their spending on alcohol, are presented as a cost to the broader community.

These statistics do not emerge out of nowhere. We need to be able to defend ourselves against them when they come back from the dead.