Nintendo 3DS Forums
Go Back   Nintendo 3DS Forums > General > The Lounge

The Lounge Just feel like having a chat? Talk about anything in here that doesn't fit in other sections of the forums.

Young adult sequelae of adolescent cannabis use: an integrative analysis
Old 09-10-2014, 11:47 PM
nasic870's Avatar
Senior Member
 
Join Date: Jul 2011
Posts: 2,171
Default Young adult sequelae of adolescent cannabis use: an integrative analysis

This study (named in the thread title) was funded by the Australian gov't to explore the effects of pot use on teens. The study was a meta-analysis using compiled datasets from three other studies which is atypical of a meta-analysis - they usually use estimates from other studies. They were looking at conditional probabilities, finding out how likely it was to complete high school, college, and attempt suicide based on use and frequency of smoking pot.

Obviously the first thing you hear out of people's mouths is correlation doesn't imply causation... because PhD's and MD's are too stupid to know that. But seriously, probabilities don't measure causation on their own: they just show a lack of statistical independence meaning there's a correlation. What most people don't know is that there has been a lot of work done, a lot of it by economists since they're the ones most interested in causal inference (they also know the math to derive the models unlike other social scientists), which has contributed monumentally to the literature on and understanding of regression analysis (e.g. the creation of instrumental variables to address endogeneity, measurement errors, generally whenever the treatment variable correlates with the error term).

But it's easy to find a statistically significant correlation which is why stats/econometrics (applied regression analysis that's a crux in empirical work in the social sciences) tries to address these issues. For example, you can find a statistically significant relationship between any two variables that move together in time. Say that the average class attendance of a college course drops throughout the semester and that in a remote part of the world, a business's profits are decreasing over the same period of time. If you simply regressed the attendance on the business's profits, you would find a significant relationship which doesn't exist in reality. It makes no logical sense for that relationship to exist. That's an easy issue to address though, you just throw in a time trend where t=1, 2, 3, ..., n where n is the number of samples into your regression.

And the PhD's are interested in causal estimates because they throw in controls, 53 of them into their estimation. The reason they do this is because they want to tease out the effect of smoking pot from the effect of other things which correlates with smoking pot.

This is taken from the actual paper:

Quote:
Several aspects of the study findings support the possibility of a causal relation. First, we recorded strong associations between adolescent cannabis use and all young adult outcomes investigated. Second, the associations had dose-response characteristics with increasing frequency of adolescent use. Third, most associations were resilient to control for potential confounding factors present before and during adolescence.
However, I'm not someone who believes there's a causal relationship and my reason is because I think there's self-selection. Where those who are more likely to commit suicide and drop out of school tend to use pot because there's something specific, some characteristics about those people that makes them more likely to do drugs. And that could be brought on for reasons like their home environment.

They probably thought so too because that's what they try to control for. I don't think they did an adequate job, though. The controls/covariates don't exist for every sample in the dataset which can throw off how relevant they are. The behavioral indicators specifically, which I think are the best candidate for explaining the relationship that's being argued over, are not the same for each dataset. And more problematic than that, the indicators themselves might not even represent actual characteristics of the individuals.

Do school problems (however they define that) necessarily indicate behavioral characteristics? Think about the things that schools punish students for. Think about all the things students can and do getaway with during and outside of school without ever being caught. Or conduct disorder reported by parents. How many parents honestly spend enough time with their children to really know about their conduct? Are there differences between what those parents believe to be appropriate conduct and if so, how does that affect the results?

Also, I think the legality of the issue further attracts people with those certain characteristics and also dissuades others who would smoke pot if it were legal but don't out of fear of being reprimanded (people who probably graduate and don't commit suicide). If true, that should muddle the probability found in the study. There might be sociological and cultural issues specific to Australia and New Zealand which could partially explain the results as well if there are differences found between the same demographics in various countries.

Because of these reasons, I don't think they give me a good excuse to believe there's a causal relationship. If they do psychological case studies and interviews of the samples, they could probably explain the dose usage through behavioral characteristics, home environment, etc. Still, I don't know much about probability theory so I can't talk a lot about their approach but they still need statistical data because of logistics (not feasible to interview all samples in a large sample size because a large sample size is needed to make generalities).

So what are your thoughts on the paper or on smoking weed's effects? What do you think of the reactions people are having?
__________________
READ THE BIBORAN!!! Ansha Abdul Brothers!!! BIBORAN!!!!
Old 09-11-2014, 12:10 AM
logitech's Avatar
Super Moderator
 
Join Date: Jan 2011
Location: North Carolina
Posts: 4,614
Default

causation=/correlation

Spoiler!
__________________

To view links or images in signatures your post count must be 50 or greater. You currently have 0 posts.
Old 09-11-2014, 12:21 AM
nasic870's Avatar
Senior Member
 
Join Date: Jul 2011
Posts: 2,171
Default

I lurk on the gaming side kind sometimes but I don't have an account. I saw it on reddit earlier today but the reason I posted this is because I saw you guys were talking about the GAF thread in the chat. Thought it deserved better discourse than what that thread and reddit had.
__________________
READ THE BIBORAN!!! Ansha Abdul Brothers!!! BIBORAN!!!!
Old 09-11-2014, 12:30 AM
logitech's Avatar
Super Moderator
 
Join Date: Jan 2011
Location: North Carolina
Posts: 4,614
Default

Oh I see. I was pretty involved with the thread on Gaf, that's why I asked and sarcastically said "causation=/correlation."

Regarding the actual article and the OP, I appreciate your efforts to stop any comments like the above and encourage people to think about the study critically and scientifically.

Causation is a difficult thing to show, especially in observational studies such as the ones used in the meta-analysis. To be completely honest, I don't have enough knowledge of statistics to determine how adequately they controlled for their confounders, though I think the list of confounders in pretty exhaustive and I can't think of anything else I would have controlled for. The sample is adequate and I think their choices for end-outcomes were well chosen, so I really can't find much at fault with the study or its design.

That being said this studies meets several criteria that are generally recognized as being needed for establishing causation: there is a strong association, the results are consistent between groups, and there is a temporal and dose-dependent relationship. Further study is obviously needed, but I feel like the results of this article are strong.
__________________

To view links or images in signatures your post count must be 50 or greater. You currently have 0 posts.
Old 09-11-2014, 01:35 AM
nasic870's Avatar
Senior Member
 
Join Date: Jul 2011
Posts: 2,171
Default

I thought it would be interesting to see what economists have to say since they focus on causality:

Marijuana Use and High School Dropout: The Influence of Unobservables finds that selectivity bias accounts for half of the difference in the odds-ratio on graduating with marijuana use. They found that by using an instrumental variable because of endogeneity. And then the rest mostly went away when they controlled for cigarette smoking, probably because of an omitted variable. Meaning their study didn't find any causal link to educational attainment and marijuana use.

Another economics article, funded by the Australian gov't like the psychologist's paper, also found that selection influenced the results while advising that if it's not all selection driven, then trying to intervene for those who have started smoking by age 14 could have drastic improvements on educational attainment.
__________________
READ THE BIBORAN!!! Ansha Abdul Brothers!!! BIBORAN!!!!