Disclaimer Disclaimer Disclaimer Disclaimer Home           Home Home           Home About  site   About  site About  site   About  site About  me    About  me About  me    About  me Current reading   Current reading Current reading   Current reading CONTACT   CONTACT CONTACT   CONTACT Links                  Links Links                  Links Book shelves      Book shelves Book shelves      Book shelves
Blog Archive    Blog Archive Blog Archive    Blog Archive Recent posts  Recent posts Recent posts  Recent posts
Alcohol and “fact”checking in Ireland About a week ago thejournal.ie decided it would be a good idea to ‘FactCheck’ whether  minimum unit alcohol pricing was proven to work, because the Irish Health Promotion  Minister (Marcella Corcoran Kennedy…apparently) made a claim about this with respect to  the Irish government’s alcohol policy. They decided that the best way to go about this was  to put a journalist on this task who does not know too much about the topic or is very good  at statistical methods. You can find the article here (link).  There is one reason to do this, and that is that one has already decided the conclusion and  someone just needs to provide the narrative for this. And indeed, the claim that ‘Minimum  unit alcohol pricing has been proven to reduce health harms, elsewhere in the world’ was  judged to be “Mostly False”… So what is this verdict based on then? This has been nicely spelled out, and gives us the  opportunity to…well, just sigh. Anyway, here they are (rephrased a bit):  - In one study which used data from British Columbia, even though there was a claim  of effectiveness of this policy, in fact the number of deaths rose and fell and  hospitalisations rose every year. We’ll get to this study in a minute… - Researchers used complex statistical models to estimate correlations. Oh dear…. - They found some statistically significant negative associations between price levels  and deaths/hospitalisations, in certain categories and for certain time periods.  This is correct…  - However, the studies cited by the Minister do not reasonably allow for the  conclusion that minimum unit pricing has been “proven to work”. “This is what  my boss told me the conclusion had to be, but I am not really sure why  either….” - Statistical models and certain principles of consumer behaviour would and do  suggest that minimum pricing should lower death and disease rates, but the  evidence available now does not conclusively prove that it has. “I did not  understand, or have not read, the paper….” You can read the rest of the newspaper article yourself if you are interested, but the  bottom line is that because the researchers of a particular study (discussed below) used  statistical models to estimate the effect of the price change, and these estimates do not  correspond to measured alcohol-related mortality rates, the whole study is rubbish. There  is further some mentioning of linear correlations vs. measurements, which is not too far of  the mark, but I don’t think the journalist really understood this bit; so we’ll leave that.  Nonetheless, at face value the article makes sense. This is a bit unfortunate, given that  these kind of complex societal issues are generally speaking, well a bit more complex; and  do not really allow for eye-balling of data to reach conclusions. But unfortunately, that is  exactly what this journalist did. He seems to have an aversion to complex statistical  models, or well, statistical models, and seems to be of the school of thought that indicates  that if you require a statistical model with several parameters in it, it must be wrong.   Now where have we seen such an approach before?...it does ring a bell………  Ah yes! It is the general method to justify the argument against “nanny-stating” and the  basis of the conclusion that any government public health policy has to be ineffective and  meant to punish people by virtue of it being a government policy. Indeed, the Institute of  Economic Affair’s ‘Lifestyle Economics’ section. In fact, the ‘Fact’check article is a near carbon copy of the IEA’s arguments on this. And  indeed (lucky for us), their director has discussed exactly the same study when it first  appeared (in 2013) – objectively called ‘lying with statistics’ (link) – and again following  the Irish newspaper article (link). I guess the similarities between both authors’ work is  what is known as a correlation….. The three articles are fascinating reads in that they clearly indicate a complete lack of  understanding of ‘confounding’, or even the awareness of such a word (I am pretty sure  that in case of the IEA this is incorrect, and this “lack of understanding” is a tool more than an actual lack of knowledge; but I do not know this for sure).   Anyway, you should by now have skimmed the three posts and are aware of how bad this  particular study is (or, preferably, you have scratched your head at using eye-balling data  as a scientific methodology). So let’s have a look at the paper…. It was entitled “The relationship between minimum alcohol prices, outlet densities and  alcohol-attributable deaths in British Columbia, 2002-2009” and was published in 2013 in  the journal Addiction by Zhao and colleagues (link). If you have read the IEA blog posts you  will have noticed that (for unknown reasons to me) Tim Stockwell, the 2nd author, is not in  the IEA ‘Lifestyle Economics’ good books…. In summary, the authors used longitudinal data on alcohol-attributable death rates, rates of  minimum prices for specific beverage types, alcohol outlet density, provincial population  data and other demographic and socio-economic area-level data. In a way, their work was  somewhat similar to recent work I have been involved in looking at effects of alcohol  policies, including cumulative impact zones, in England (paper 1, paper 2). They used  linear (boom! There you go Irish reporter guy; good spot!) multi-level mixed effects  statistical models as well as specific time-series models to analyse the data, which enabled  them to adjust for confounding factors (or, as they are called by some people ‘complex  models’). Now the latter is quite important!   It is always important in epidemiology and public health to do this, but in this particular  case it is even more important because at the same time (2002 to 2009) of the minimum  unit pricing policies, another change occurred in those areas. New alcohol policies let to a  partial privatization of alcohol retail sales, which resulted in a substantial expansion of  private liquor stores in those areas. You can imagine that it is not unlikely that the number  of outlets is correlated to the amount of alcohol sold…  This issue will then counteract any observable effect of minimum unit prices and is a true,  and probably important, confounding factor; in fact, given the likely effects both of these  policies will have (at population level) it is not unlikely that both are effective, but that  the impact of the latter is bigger. This would then show itself in increased rates of alcohol-  related mortality and other health harms despite the effectiveness of minimum unit  pricing. So have we seen this before….like in the three articles above…..? It is possible to separate these effects, and this is not done by eye-balling the data and  shouting out some ad hoc conclusions, but it does require that horrible ‘complex statistical  modelling’. The article has a number of tables showing all the estimates of the model  parameters which are quite helpful, but only if you know what you are looking at.   However, they do indicate exactly what is communicated by the authors with respect to  effects of minimum prices and of private liquor stores (both also backed up by other  studies, I hasten to say), but also show some other little facts that are both interesting and  in agreement with what you would expect (thus suggesting the models are probably  correct-ish). For example, alcohol-related harm rates are lower when there are more  restaurant in an area (compared to liquor stores and bars); that makes sense (well to me it  does; I don’t know how you behave in a restaurant…). Indeed, so what one can do is calculate the correlations between all the variables in the  model and then estimate what would have happened – i.e. what the alcohol-attributable  death rates would have been after minimum unit prices were introduced – if everything  else had stayed the same (!). The latter bit is important since this refers to for example changes in the age distribution  of the population, changes in economic purchasing power, other factors, and especially the  highlighted other alcohol policy of privatization of alcohol retail; all of which may have  changed over time and affected people buying alcohol (and thereby affecting the incidence  of related health effects; in some people), and thus we’d need to keep these at the same  level to isolate and estimate the effect of the minimum unit price policy.   So by doing this it is possible to estimate what should have happened (had privatization and other things not happened) as a result of the minimum unit price policy; and thus how  many people did not die of alcohol-related disease during that period. You can imagine that  all these individual effects are way too complicated to interpret by just eye-balling the  data.   There are some limitations to this (of course):    The first one is that it is not possible to say whose life specifically has been saved,  because they are still alive and it is not possible to determine at an individual level what  impact all the different factors had. We can just say this at an average, population level.     And the second one is that nothing important should have happened in that same period  that was not taken into account in the model. If something had happened, this may have  changed the correlations between the different factors and the estimate of the true effect  could be too large or too small (this depends on the missed factor). However, the good  news is that this factor should be quite big and important and had to be introduced in  exactly the same time period; it is quite likely that someone would have mentioned this  (but again, we don’t know).  So in conclusion, what can we conclude from all this? I would say there are a couple of  important lessons: 1) Confounding is important 2) Don’t just eyeball some data and draw important conclusions  3) It is likely that minimum unit pricing has some positive effect on population    health  4) Some government policies can be a good idea… …and of course….    5)      The ‘FactCheck’ item of thejournal.ie could do with a bit of an upgrade… 
Back   Back Back