Once upon a time the Gateway Hypothesis* related to the fact that people who smoked cigarettes might at some point attempt to consume marijuana and/or cocaine. Cigarette consumption has collapsed in many countries, yet people still seem to consume marijuana and cocaine. So now the Gateway Hypothesis is that people who vape (i.e. consume electronic cigarettes) might take up the consumption of combustible cigarettes.
We are treated to shock-horror headlines such as:
That’s from The Australian – one of the more sensible and least hysterical media outlets.
Vaping is a “gateway” to conventional smoking with those using e-cigarettes three times as likely to take up tobacco, a study reveals.
“We found non-smokers who use e-cigarettes have around three times the risk of taking up regular smoking than non-smokers who have avoided e-cigarettes.”
The latest research, a joint project by the Australian National University and University of Melbourne, was commissioned as part of the so-called “Vaping Inquiry” agreed to by Health Minister Greg Hunt in 2018.
Well, it turns out that wasn’t an actual study, but rather a meta-study performed by researchers at the ANU and Melbourne Uni. The paper can be found here.
I have many criticisms of that paper (and other similar studies) but here I want to highlight some statistical sleight of hand that is going on.
This sleight of hand is well illustrated by a paper published in Tobacco Control in 2018. The table below is reproduced from Best et.al (2018) in total. Note that the definition of having smoked a cigarette is having taken ‘just one or two puffs’.
Best et.al (2018) interpret the table as follows: Of 2125 students in the sample, 183 students consumed e-cigarettes (169+8+6 = 183). Of those 183, 74 students also consumed combustible cigarettes. Therefore 40.4% (74/183) of Never Smokers who had consumed e-cigarettes went on to consume combustible cigarettes. By contrast only 12.8% of Never Smokers went onto consuming combustible cigarettes (249/1942).
That is one interpretation of the data. That interpretation, however, is highly misleading. It artificially partitions the students into combustible cigarette consumers and e-cigarette consumers, and then investigates any overlap. It assumes that which is yet to be proven. The Gateway Hypothesis suggests: those individuals who had consumed e-cigarettes are more likely to subsequently become consumers of combustible cigarettes. The important distinction here is: more likely than whom? There are four groups of students in the sample. Those students who never consumed either a combustible cigarette or an e-cigarette. Then we have students who consumed combustible cigarettes who never consumed e-cigarettes. The existence of these two groups of students provide evidence against the Gateway Hypothesis. Then we have a group of students who consumed e-cigarettes who had never consumed (even took a puff or two) of a combustible cigarette. The existence of this group of students is, at least, inconsistent with the Gateway Hypothesis. To the extent that any of these students may well have become combustible cigarette consumers, but did not, this group of students would provide evidence that rejects the Gateway Hypothesis. Unfortunately, due to the casual empiricism that is so common in this literature, that interpretation of the data is not tested by Best et.al (2018) or any other researchers in the field. Finally there is a group of students that consumed e-cigarettes that went on to consuming combustible cigarettes (or, at least, had a puff or two of a combustible cigarette).
To illustrate the inadequacy of the interpretation of the data in the table – the data in the various underlying studies are often presented in a confusing and misleading manner – I have summarised Best et.al’s (2018) table 2 as below:
Of the 2125 students in the sample, 1693 had never consumed either a combustible cigarette or an e-cigarette. That is 79.67% of all students. 249 students had consumed a combustible cigarette, but had not consumed an e-cigarette – that is 11.73% of all students. Then 109 students had consumed e-cigarettes, but had not consumed combustible cigarettes – that is 5.13% of all students. Then 74 students had consumed both combustible cigarettes and e-cigarettes – that is 3.48% of all students.
Yet by carefully partitioning the student sample, Best et.al (2018) are able to inflate a figure where 3.48% of students use e-cigarettes before combustible cigarettes into a result consistent with the Gateway Hypothesis. If we look back to the original Best et.al (2018) table, we see that only four students who had regularly consumed e-cigarettes (weekly) went onto having had a puff (or two) of a combustible cigarette. That is 1.2% of all combustible cigarette consumers and 0.19% of all students. Who can even tell if those four students had gone on to being regular combustible cigarette consumers? The Best et.al (2018) summary statistics are too opaque to reveal that (important) detail
When viewed in this light, the “evidence” for the Gateway Hypothesis is very weak. The analysis that shows otherwise is contrived to show a result that simply cannot be supported by the data. The Best et.al (2108) study, however, is not an outlier. It is not somehow an anomaly. It is typical of the literature.
To demonstrate this point I partially reproduce table 2 from Soneji et.al (2017) below. The first three columns are directly reproduced. The first column shows the study being referred to, the second column shows the probability of an e-cigarette user taking up the consumption of combustible cigarettes, while the third column shows the probability of a non-e-cigarette user taking up the consumption of combustible cigarettes. The results appear to be very similar to those reported in Best et.al (2018).
In order to establish the veracity of those data I investigated the summary statistics of each of the seven papers. In the case of Miech et.al, interpretable summary statistics were not included in the study. Similarly in the case of Wills et.al, I could not sensibly interpret the summary statistics that they did report. In the other five instances, understanding and interpreting the summary statistics was a non-trivial task. I submit that the reporting on summary statistics is deliberately opaque in order to mask to underlying reality that consuming e-cigarettes is not a gateway to the consumption of combustible cigarettes. In each study I attempted to establish the number of combustible cigarette users, the number of e-cigarette users, and the number of non-e-cigarette users. The results of that exercise are shown in the final three columns of the table.
Looking at the totals row, across the five studies where I was able to get meaningful summary data, 712 individual became combustible cigarette users, of those 183 had initially consumed e-cigarettes (183/712 = 25.7%). That means that 74.3% of individuals who became combustible cigarette users did so directly without first consuming e-cigarettes. That result, however, is distorted by Barrington-Trimis et. al. This is the only underlying study that actually provides unambiguous support for the Gateway Hypothesis. This study has an astonishing small sample size of 298. In the absence of that outlier study the percentage of individuals who became smokers after having consumed e-cigarettes falls to 19.5%. That means that over 80% of individuals who become combustible cigarette consumers did not do so after consuming e-cigarettes.
The statistics demonstrating a Gateway Hypothesis are dodgy. The fact is that the evidence supporting such a gateway is very weak. Worse, the hypothesis that vaping is an exit from smoking combustible cigarettes is simply not being tested.
Anyway – my friends at the Consumer Choice Center have just published two papers (one for Australia and the second for the EU) where I spell out this sleight of hand and poor hypothesis development in greater detail.
*Some readers will recognise the Gateway Hypothesis as being a slippery slope argument.