Friday, 15 August 2014

The plain packs meta-lie (part one)

In recent weeks the various dodgy claims made by plain packaging campaigners have coalesced into a single meta-lie which, I suspect, will now be repeated again and again.

The meta-lie has two components. Firstly, that Australia has seen the biggest ever decline in smoking prevalence since plain packaging was introduced. Secondly, that tobacco sales fell by 3.4 per cent in Australia in the first year of plain packaging (with the implication that this is an unusually large drop).

These claims have been repeated by Public Health England...

In Australia, official data is already demonstrating the impact. Its latest national triennial survey shows the fastest decline in smoking rates in over 20 years, and customs and excise data shows a fall of 3.4% in tobacco sales by volume in the first year of standardised packaging... According to the latest official national survey of tobacco use, the daily smoking rate fell markedly from 15.1% to 12.8% between 2010 and 2013 – a record 15.2% decline.

They are also being parroted by ASH Ireland today in the Irish Times...

Latest research results from Australia on its standardised packaging of tobacco products are very encouraging. Official data from its latest national survey show the fastest decline in smoking rates in over 20 years with an 11 per cent relative reduction in the prevalence of smoking. Customs and excise data in the same survey show a fall of 3.4 per cent in tobacco sales by volume in the first year of the legislation.

Let's take the claim about smoking prevalence first. This contains a lie within a lie. The figures that campaigners are referring to compare 2010 with 2013. Plain packaging was only introduced in December 2012 so anything that happened in two-thirds of that period cannot possibly be attributed to olive green fag packets. Moreover, there was a whopping 25 per cent tax hike on tobacco in 2010 which the government predicted would reduce the number of smokers "in the order of 2 to 3 per cent, or around 87,000 Australians."

Using two data points three years apart when plain packaging wasn't in place for most of the period in question is obviously a poor method of assessing the impact of the policy. We do have data that shows annual changes in smoking prevalence, but plain pack campaigners ignore it because it clearly doesn't support the claim they want to make.

Leaving all that aside, the claim about smoking prevalence doesn't stand up even on its own terms. The graph below shows smoking prevalence in Australia since 1995 (in three year increments).

Anyone whose job does not depend on not seeing things can see that there has been a steady, longterm decline in the smoking rate. Prevalence fell in every three year period and did so consistently and predictably within a narrow range of 0.9-2.4 percentage points.

The biggest decline in this period was not—as campaigners are claiming—between 2010 and 2013 (2.3%), but between 1998 and 2001 (2.4%). 2010-13 did not see the biggest decline in smoking rates since records began (as Martin Dockrell claimed on the radio recently). It was not even the biggest decline in the last fifteen years.

So why do they say it was? Simon Chapman seems to be the source of this particular bit of spin. He's keen to focus on the percentage differences between the percentages rather than look at the decline amongst the whole population. Therefore, he's looked at the percentage drop from 15.1% to 12.8% (which is 15.5%) and decided that this is the only number worth talking about. His method has the effect of downgrading what was genuinely the biggest decline in the time series that occurred between 1998 and 2001 (21.8% to 19.4% = 11.0%).

To see why this is a statistical trick, consider the graph below which shows a constant decline of 5 units from 50 to 0.

The decline is completely straight and linear, but if you measured it using Chapman's method you would get the impression that the decline speeds up rapidly and exponentially (see the percentage figures on the horizontal axis). In the early stages of the decline, the percentage drop is quite small (10%, 11.1% etc.), but as the numbers are diminished the same drop leads to much bigger declines in relative terms, ultimately ending in declines of 50% and 100%. Every step down represents the biggest ever decline up to that time even though the drop is always exactly the same.

Therefore, if the number of smokers is dwindling year-on-year (as it is in most Western countries) and there is a constant and steady decline in the smoking rate, it is a mathematical inevitably that you will keep seeing a record decline if you measure it as Chapman does even if the rate of decline of smoking in the population does not accelerate. Of course, the smoking rate does not fall by exactly the same amount every year so it is not a certainty that records will always be broken, but so long as the decline is roughly constant the probability of a record being broken rises as the numbers get smaller.

This is not the way the data are normally looked at, nor is it the way they should be looked at. Chapman resorts to this tactic because it helps disguise the simple fact that the smoking rate has been falling a steady but pedestrian pace for many years and has continued to fall at the same pace since 2010.

Plain packaging wasn't introduced in 2010 or 2011. It was introduced in December 2012, so the question of whether there was a record drop in the smoking rate between 2010 and 2013 is academic so far as that policy goes. Nevertheless, for the record, there wasn't.

To be continued...

No comments: