Coming across a Psychology Today article on “The Link Between Cannabis and Psychosis” is a little like watching this clip from The Onion. It’s a well written article but someone in a different mental state might read it as “everyone knows you’re high and it’ll probably make you go insane.” I have taken enough econometrics to suspect that there might be some selection effects in the observational studies that this article mentions. I’ve also taken enough econometrics to wonder whether I could answer this question for myself. This post chronicles my attempt to do just that.
Is the problem of finding this evidence that you’ve searched for associations with schizophrenia rather than psychosis more broadly defined? That’s all well and good but if you follow up people who experience cannabis induced psychosis, four years later roughly 1/4 to 1/2 transition to schizophrenia https://academic.oup.com/schizophreniabulletin/article/46/3/505/5588638#
Yeah I don't think my lit review was comprehensive but I was writing this for a senior thesis and I discovered the bad data before I did a deep dive on previous research. This post was meant to just follow my research process rather than be an air tight defense of a claim.
Hey I'm trying to start a DM with my favorite lesser-known sub stackers and commenters on their articles (that is, people like Sam Atis). Would you care to be a part of it? I would be grateful for your insights! If you're interested email generalpurposeemail1@gmail.com your discord username and user ID # and I'll add you.
1. Schizophrenia is diagnosed between the ages of 15-35, so the derivative of incidence might just capture the millennial cohort distinct from Gen X and Z. If net MJ consumption goes up but adolescent consumption goes down then macro scale analysis will be too heavily confounded.
2. The medical literature suggests an effect of regular heavy/daily use but not intermittent use; the risk is more like pain pill overdoses than tobacco carcinogens. If everyone smokes 1 extra joint per week because of legalization, that might affect marginal incidence in the 5+ joints-a-week population, but everyone else will be fine.
3. Cannabis can cause psychotic episodes, which shows it impacts the same systems that drive schizophrenia. But, it’s possible that someone who has multiple cannabis-induced psychotic episodes across years satisfies the DSM-V criteria but isn’t a ‘true’ schizophrenic. Does that make our priors go up or down?
I’ve done a bit of research on this topic earlier, which left me mildly concerned about weed and very concerned about the quality of illicit drug studies. Your post is a great attempt but GIGO. Thanks for the attempt tho.
1.) This is right. I've checked the data and adolescent use is increasing more than overall use. I can also break down the IHME Schizophrenia data by age group but it doesn't change the results. The concerns with the IHME data are the same though.
2.) This is very true, the NSDUH data does have some data on intensity of use but to you need special privileges to break it down at the state level for privacy reasons so I couldn't get it. But I think this is actually really important and other literature emphasizes the intensive vs extensive growth. I think it is plausible that legalization also increases intensity but it isn't measured.
3.) For the analysis I'm doing it's fine if there is misdiagnosis as long as the misdiagnosis isn't significantly different in legal states vs illegal ones. The diagnosis bias would be washed out in the comparison between states as long as they diagnose in the same way .
Overall I agree that this isn't super convincing. I was working on this topic for a senior thesis but because of these problems I've moved on. I'm glad you found it interesting though!
Hey I'm trying to start a DM with my favorite lesser-known sub stackers and commenters on their articles (that is, people like Sam Atis). Would you care to be a part of it? I would be grateful for your insights! If you're interested email generalpurposeemail1@gmail.com your discord username and user ID # and I'll add you.
Whoa, very cool. Where did you learn the methodology for synthetic (weighted average) difference-in-difference? I think you described it to me in the past with regards to a different research domain.
Second, can you clarify this: "we can do placebo tests on the control group states that never legalized and see how Colorado’s deviation from its control group compares to these placebos that didn’t actually change their cannabis policies"
Are you comparing [Colorado vs. synthetic Colorado] to [non-legal states vs. their synthetic counterparts]?
Lastly, it's sad that the schizophrenia data is scarce; the "Schizophrenia by State" graph is a very honest inclusion, but I think it reveals that your overall approach doesn't work because the state trends are endogenous.
2. That's basically right. Colorado deviates a lot from its synthetic control after 2012, but we would expect some deviation from the control group even if nothing changed just due to random variation since we only match the control to the real data for 2002-2011. The placebo graph compares the deviation of states that didn't change their laws to Colorado's deviation.
3. Yeah it is sad. This is why I didn't end up doing this topic for my thesis.
I think you have overlooked a lot of evidence that cannabis (particularly THC) causes psychosis - summarised in this thread https://twitter.com/psychunseen/status/1504138625713246210?s=46&t=h_aFsZlD-I_rxx3iml2TfA
- and also see this meta https://pubmed.ncbi.nlm.nih.gov/26884547/
Is the problem of finding this evidence that you’ve searched for associations with schizophrenia rather than psychosis more broadly defined? That’s all well and good but if you follow up people who experience cannabis induced psychosis, four years later roughly 1/4 to 1/2 transition to schizophrenia https://academic.oup.com/schizophreniabulletin/article/46/3/505/5588638#
Let me know if this is what happened!
Yeah I don't think my lit review was comprehensive but I was writing this for a senior thesis and I discovered the bad data before I did a deep dive on previous research. This post was meant to just follow my research process rather than be an air tight defense of a claim.
Hey I'm trying to start a DM with my favorite lesser-known sub stackers and commenters on their articles (that is, people like Sam Atis). Would you care to be a part of it? I would be grateful for your insights! If you're interested email generalpurposeemail1@gmail.com your discord username and user ID # and I'll add you.
1. Schizophrenia is diagnosed between the ages of 15-35, so the derivative of incidence might just capture the millennial cohort distinct from Gen X and Z. If net MJ consumption goes up but adolescent consumption goes down then macro scale analysis will be too heavily confounded.
2. The medical literature suggests an effect of regular heavy/daily use but not intermittent use; the risk is more like pain pill overdoses than tobacco carcinogens. If everyone smokes 1 extra joint per week because of legalization, that might affect marginal incidence in the 5+ joints-a-week population, but everyone else will be fine.
3. Cannabis can cause psychotic episodes, which shows it impacts the same systems that drive schizophrenia. But, it’s possible that someone who has multiple cannabis-induced psychotic episodes across years satisfies the DSM-V criteria but isn’t a ‘true’ schizophrenic. Does that make our priors go up or down?
I’ve done a bit of research on this topic earlier, which left me mildly concerned about weed and very concerned about the quality of illicit drug studies. Your post is a great attempt but GIGO. Thanks for the attempt tho.
1.) This is right. I've checked the data and adolescent use is increasing more than overall use. I can also break down the IHME Schizophrenia data by age group but it doesn't change the results. The concerns with the IHME data are the same though.
2.) This is very true, the NSDUH data does have some data on intensity of use but to you need special privileges to break it down at the state level for privacy reasons so I couldn't get it. But I think this is actually really important and other literature emphasizes the intensive vs extensive growth. I think it is plausible that legalization also increases intensity but it isn't measured.
3.) For the analysis I'm doing it's fine if there is misdiagnosis as long as the misdiagnosis isn't significantly different in legal states vs illegal ones. The diagnosis bias would be washed out in the comparison between states as long as they diagnose in the same way .
Overall I agree that this isn't super convincing. I was working on this topic for a senior thesis but because of these problems I've moved on. I'm glad you found it interesting though!
Hey I'm trying to start a DM with my favorite lesser-known sub stackers and commenters on their articles (that is, people like Sam Atis). Would you care to be a part of it? I would be grateful for your insights! If you're interested email generalpurposeemail1@gmail.com your discord username and user ID # and I'll add you.
Whoa, very cool. Where did you learn the methodology for synthetic (weighted average) difference-in-difference? I think you described it to me in the past with regards to a different research domain.
Second, can you clarify this: "we can do placebo tests on the control group states that never legalized and see how Colorado’s deviation from its control group compares to these placebos that didn’t actually change their cannabis policies"
Are you comparing [Colorado vs. synthetic Colorado] to [non-legal states vs. their synthetic counterparts]?
Lastly, it's sad that the schizophrenia data is scarce; the "Schizophrenia by State" graph is a very honest inclusion, but I think it reveals that your overall approach doesn't work because the state trends are endogenous.
1. Basically everything I know about synthetic controls come from this useful doc: https://www.jstatsoft.org/article/view/v042i13
2. That's basically right. Colorado deviates a lot from its synthetic control after 2012, but we would expect some deviation from the control group even if nothing changed just due to random variation since we only match the control to the real data for 2002-2011. The placebo graph compares the deviation of states that didn't change their laws to Colorado's deviation.
3. Yeah it is sad. This is why I didn't end up doing this topic for my thesis.
This is awesome, thanks for doing this and writing it up!
Thank you for reading <3