Privatizing kills! Or does it?

March 4, 2014

By Rebecca Goldin, Ph.D and S. Robert Lichter, Ph.D, January 26, 2011

This article originally appeared on STATS

STATS critiques a new study claiming that private liquor stores are hazardous to our health, while government liquor stores save lives.

A study making waves in Canada has spilled over into Virginia’s debate over privatizing state liquor stores. When researchers charged that British Columbia’s privatization program has increased alcohol-related deaths, Virginia and Washington D.C.-based media were quick to play up the local angle. For example, under the headline “Research adds a new twist to Virginia liquor debate,” the Norfolk-based Virginian Pilot reported on January 20:

“As the General Assembly considers Gov. Bob McDonnell’s proposal to privatize Virginia’s state-owned liquor monopoly, Canadian researchers reported this week that alcohol-related deaths increased after a partial privatization of liquor stores in British Columbia.”

The study, which was conducted by the University of Victoria’s Center for Addictions Research, is being publicized in the United States by a California-based alcohol watchdog group called the Marin Institute. The Institute’s research director did not mince words in a statement quoted by the Pilot: “Powerful, profit-driven industry players only want to increase their own alcohol sales regardless of the proven dangers to public health…. Hopefully, this study will put these fruitless and time-wasting debates to rest.”

But the study’s findings, which originally appeared in the medical journal Addiction, are more complicated than the publicity rollout might suggest. For starters, although alcoholic consumption increased in British Columbia (and the rest of Canada) throughout the study period from 2003 to 2008, the alcohol mortality rate did not. To the contrary, it actually declined.1

The period in question saw a sharp rise in the number of private liquor stores and sales and a slight decline in the number of government stores and sales in British Columbia. (On page 7 of the article, the authors put the increase in private stores at 40 percent, although this doesn’t seem to match their data in Table 4 on page 6 , which shows an increase from 727 to 977 stores, or 34 percent.) The  authors include a figure (reproduced below) showing that the rate of alcohol consumption has risen throughout Canada since 2002/03, but it has risen even faster in British Columbia, from approximately 7.8 liters per capita (age 15 and above) in 2002/03 to nearly 8.9 liters in 2007/08.

Figure #1, from article:

figure 1

The article does not provide a similar figure showing changing mortality rates, but the text does list the annual number of alcohol-related deaths from 2003 to 2008. Matching these figures with population data from Statistics Canada, we found that the rate of alcohol-related deaths in British Columbia per 10,000 people gradually rose from 4.70 in 2003 to 4.92 in 2006. However, it then dropped to 4.81 in 2007 and 4.59 in 2008. Thus, the alcohol mortality rate at the study’s endpoint was 2.4 percent below the 2003 starting point. These data are presented here:

Table 1
Alcohol Related Death Rates

  Alcohol- related deaths by cause and ICD-10 code in British Columbia Population Alcohol-related deaths per 10,000 population

2003

1937

4,122,396

4.70

2004

1983

4,155,170

4.77

2005

2016

4,196,788

4.80

2006

2086

4,243,580

4.92

2007

2074

4,309,632

4.81

2008

2011

4,383,860

4.59

% Change 2008/2003

3.8%

6.3%

-2.4%

Sources: Alcohol-related Deaths: From the study
Population:  From Statistics Canada http://www.bcstats.gov.bc.ca/data/pop/pop/BCPop.asp

 

The consumption and mortality rates are compared in the Figure 2, which illustrates the clear absence of a linear relationship between them:

 

figure 2

In fact the authors allude to the absence of a clear province-wide relationship between alcoholic consumption and mortality, when they mention in passing, “The overall increase in alcohol-related mortality … was not statistically significant after controlling for … exogenous variables.” (p 7) However, this implies that there was a positive relationship before controls were applied. But there was no such relationship between consumption and mortality rates. As table 1 shows, the increase in mortality was a function of population growth. Even though the raw number of deaths rose from 1937 to 2011 during this period, the mortality rate decreased from 4.70 to 4.59.

Thus, the authors’ argument linking alcohol-related deaths to privatization is based not on overall trends but on a comparison of trends among 89 smaller geographical units called Local Health Areas (LHA’s). Their key finding is that changes of density in private liquor stores in various LHA’s was associated with higher rates of alcohol-related mortality, after adjusting for demographic and other potential confounding factors. This finding was based on multi-level regression analyses modeling the relationship between logarithmically-transformed LHA rates of alcohol-related mortality and the numbers of different types of liquor outlets for each of the six years in the study period.

The results of this analysis are summarized in the article in Table 5 on p.7, which is reproduced below as our table 2. According to the authors, they show that in the “fully adjusted model,” which controls for socio-economic variables as well as special and temporal confounding factors, “the rate of alcohol-related deaths increased by 27.5 percent with each extra private liquor store.” (p. 7) This apparently refers to the 0.275 entry in the cell of the table representing “Private, Fully adjusted.” But if the table reflects such a logarithmic transformation, as the authors state, it is not clear how a .275 coefficient equals a 27.5% increase in deaths.

 

Table 2: Relationship between log-transformed alcohol related mortality rate per 10,000 population and annual number of different liquor outlets per 1000 population aged 15 years and over in BC in 2003-2008.

table 2

Assuming that we are correct in understanding the authors’ interpretation of these coefficients, however, a statistical oddity appears elsewhere in the table. By the same reasoning, the -0.568 coefficient representing “GLS, Fully adjusted” would mean that the rate of alcohol-related deaths decreased by 56.8 percent with each additional government liquor store. That is, opening a government liquor store will save twice as many lives as opening a private store will cost. Even if these extreme numbers are a misinterpretation of the authors’ poor description of their calculations, the authors themselves point to an inverse relationship between number of government-run stores pre 1000 population, and the number of deaths per 10,000.

The authors list several factors by which government alcohol monopolies might reduce the adverse health effects of alcohol, principally by making it more difficult to purchase – fewer stores, shorter hours, higher prices, fewer promotions etc.  But it seems difficult to find a plausible explanation for why adding a new government-run liquor store in a local community would dramatically reduce alcohol-related mortality, independently of changes in all other sources of alcoholic consumption. From a public health perspective, a logical policy consequence of these findings would be to open new government liquor stores even faster than you close private stores. It is difficult to imagine that this is what critics of alcohol abuse like the Marin Institute had in mind.2

It is hard to tell from the published article whether some confounder might be inadequately represented in a way that accounts for this puzzling finding. Interestingly, Table 5 in the article shows that the original model, which was unadjusted for potential confounders, found that an additional government-owned store actually produced a greater increase in alcohol-related deaths than did an additional privately-owned store. An intermediate “partially adjusted” model found that increases in both types of stores produced increased deaths, with private stores accounting for a greater increase.

It is possible that we are seeing a spurious effect of uncontrolled confounders.For example, people who drink heavily (without that preference being sufficiently attributed to their demographic category) may, for some unknown reason, be drawn to areas where there are private stores. Conversely, businesspeople seeking to maximize profit might locate their new private stores in areas where there are people who drink heavily, and are therefore more likely to die from alcohol-related causes.

It is also possible that the observed effects result from some peculiarity of the LHA’s as the basis for the analysis. This is called the modifiable areal unit problem (MAUP). It refers to the fact that conclusions in geographically-based studies are highly sensitive to the scale and boundaries of the areal units that are used. That is, basing the analysis on a different set of geographical units might produce quite different results. The authors do try to control for the confounding effects of spatial factors, but the fact that significant effects appear only for the LHA’s and not for the province as a whole points in this direction.

Finally, it is possible that the dependent variable is inadequately specified. In measuring alcohol-related mortality, the study makes no distinction between deaths that are directly related to alcohol, such as alcoholic psychoses and dependence, and deaths that are only indirectly related to alcohol, like some neoplasms and diseases of the circulatory and respiratory systems. Nor does it distinguish between short-term causes of death, such as alcohol poisoning, and long-term causes like cirrhosis of the liver. This is particularly relevant to a study that tracked mortality trends over a relatively brief six year period.

These are all only possibilities, and it may be that further data analysis or methodological would resolve the issues we have raised. Nevertheless, researchers should be cautious in interpreting an effect that is found locally but not globally, and one that predicts that building a new liquor store will dramatically raise or lower alcohol mortality, depending on who runs the store.

The relevance of these findings to the privatization debate in Virginia also needs further review. For example, British Columbia adopted a partial privatization plan permitting both government-owned and privately-owned stores to operate during the study period. Virginia’s governor is proposing to auction off liquor store licenses to completely replace the current government-run stores. In addition, British Columbia’s system covers beer and wine as well as distilled spirits, while beer and wine are already sold in privately-owned stores in Virginia.

Finally, we should be even more cautious about using a newly published scientific study as a political football until such questions are resolved. However much advocates may hope that a single study “will put these fruitless and time-wasting debates to rest,” the purpose of scholarly research is to advance debates, not stifle them.

1. Based on the data presentation in the article, it is not clear whether this decline is statistically significant. But these data clearly fail to support a significant positive association between consumption and mortality rates.
2. Of course, if we were to take this seriously, the 56.8 percent would represent a marginal benefit that might diminish as the number of stores grows.

Leave a Reply

Your email address will not be published. Required fields are marked *