January 15, 2013 | 1
As marijuana is being legalized in Washington and Colorado states, its proliferation and use raise legitimate issues regarding its dose-dependent and long-term effects. One key question is whether pot leads to cognitive decline and a lowering of IQ, especially if its consumption is started at an early age. Answering this question is important for users, families and policy makers to have a realistic idea of personal and legal policies regarding widespread cannabis use.
Last year, Madeline Meier and her group from Duke University reported results from the so-called Dunedin study which tracked a group of 1037 people from their birth to age 38. These volunteers’ pot smoking histories were monitored at periodic intervals from age 18 onwards. The study found a troubling decline of IQ and cognitive abilities among regular pot smokers, especially those whose habit kicked in during their teens. No explicit causal relationship was assigned between the two facts, but the correlation was positive and significant. The study naturally raised a lot of questions regarding the wisdom of early pot use, especially in light of its current legalization in two states.
Now a study by Ole Rogeberg from the Ragnar Frisch Center for Research in Norway has called this study into question, both for its methodology and its conclusions. The first thing to realize about any such study, even if you don’t know the details, is that there are going to be several confounding socioeconomic factors in assessing any relationship between cannabis use and IQ. Medicine and psychology are not exact sciences, and following a large sample of people for 38 years and assessing correlation – let alone causation – between any two factors is going to be confounded by a large number of other correlated and uncorrelated variables in an inherently uncontrolled experiment. The second thing to realize is that effects of socioeconomic status (SES), including finances and education on IQ are well-documented. The so-called Flynn-Dickens model explores the interplay between inherited variability in IQ and the resulting socioeconomic effects which in turn influence IQ through a feedback loop; for instance, children with lower IQ are likely to be introduced to less cognitively challenging environments, which further impairs improvements in IQ.
Rogeberg’s study takes SES into consideration and details how it can be applied to the original data. For instance, if we consider education as a (admittedly noisy) proxy for SES, then it turns out that there is a difference in IQ drops between users with higher education and those with high-school education, with the former drop being roughly half as much as the latter. This drop is independent of cannabis use, and while the sample size is small, it indicates that even a noise proxy like differences in education can partly account for IQ differences.
To more quantitatively explore the effects of SES rather than cannabis use on IQ, Rogeberg constructs a simulation model that excludes marijuana use. Instead the model relies on two assumptions; first, that SES actually predicts cannabis exposure (with poor SES correlating with more exposure) and second, that kids with low SES receive a boost in IQ from compulsory schooling, a boost that declines over time as this imposed environment withers away. The model finds that these two socioeconomic effects essentially reproduce the original data as shown in the figure below.
Now while the results from the model are persuasive, it’s a model after all; a simplified representation that only captures a little bit of reality. A good study would recommend further measures to distinguish between two conflicting paradigms, and this is what Rogeberg does. For instance the original study looked at a few confounders but only in terms of individual effects which were small. But small individual effects can add up to larger confounding factors and Rogeberg suggests that the previous group should have done this summation. In addition, the data also found higher correlation between marijuana use and IQ declines for teen-onset users, but this finding would also be consistent with a socioeconomic model that predicts lower IQ declines for adult-onset users because of their better SES. It’s also important to establish a baseline; the study did not document what the variation in IQ between the users was before they started using cannabis. Finally, comparing your results against the null model is paramount in any statistical study; in this case the question to ask would be, how much of the variation supposedly caused by cannabis use can be explained compared to the variation known to be caused by purely socioeconomic factors in other established studies.
Ultimately this modeling is not just an academic exercise, since a model correlating socioeconomic status with IQ has very different policy implications from one correlating IQ and cannabis use. At the very least it alerts us to the complex web of confounding interrelations that needs to be taken into consideration when conducting these kinds of studies. Rogeberg is quite emphatic about his findings in the conclusion:
“Meier et al.’s estimated effect of adolescent-onset cannabis use on IQ is likely biased, and the true effect could be zero. It would be too strong to say that the results have been discredited, but fair to say that the methodology is flawed and the causal inference drawn from the results premature. Furthermore, should a direct effect of adolescent-onset cannabis use remain after controlling for confounders, the Flynn–Dickens model suggests an alternative causal path through which this may occur. This model, too, would predict reduced IQ in so far as heavy, persistent, ad- olescent-onset cannabis use involves a culture and norms that raise the risk of dropping out of school, getting entangled with crime, and other such behaviors. Unlike a neurotoxic effect, however, this effect would be nonpermanent and mediated by the cognitive demands of different environments. Because the effect in this case would be a result of culture rather than pharmacology, it would also have different policy implications.”
In an email sent earlier today, Dr. Rogeberg clarified the parts about education-as-proxy and about establishing the change in IQ before starting marijuana consumption. Here’s what he had to say:
“The education-as-noisy-proxy paragraph discusses something that I have found difficult to state in a simple manner, and I became a bit uncertain as to whether I had gotten the point clearly enough across when you wrote “This drop is independent of cannabis use, and while the sample size is small, it indicates that even a noise proxy like differences in education can partly account for IQ differences.” What I do is compare the effects of high-cannabis exposure (to use that terminology) for two educational groups. The “effects” are defined in terms of how large a drop the heaviest user-group had relative to non-users (within their own educational category). I.e., (IQ-drop of heaviest users with low education minus IQ drop of non-users with low education) compared with (IQ-drop of heaviest users with high education minus IQ drop of non-users with high education). It is this latter parentheses, effect-of-cannabis-on-high-educated that is roughly half the size of the first parentheses. Same goes for the effect of the user-group with the second-highest “exposure”. These numbers are pulled from their original research (and thus based on their actual data).
The other thing regards the sentence concerning what the “variation in IQ between the users was before they started using cannabis”. I think you are referring to the test that would show whether the IQ-changes of the different cannabis-groups were similar or different prior to their cannabis use. Put differently, if we look at IQ changes from, say, 9 to 11, were these the same for all the groups (who at that time had yet to begin using cannabis)? If the IQs were already changing at different rates, it would throw doubt on their (central) assumption that cannabis-exposure was the sole source of later differences between the groups.”
Give a 1 year subscription as low as $14.99X