Yesterday, we looked at the relationship between drug use and economic patterns. We saw that drug use was associated with both higher levels of state economic output as well as higher levels of unemployment.
Today, I turn to the relationships between drug use and economic class. My colleague Charlotta Mellander charted the relationships between drug use and the percentage of a state's economy that is made up of two classes: the creative class--that is, people who work in knowledge-based, artistic, and professional occupations; and the working class--those who work in production, transportation, and construction jobs.
While the associations between drug use overall are weak, the patterns for marijuana and cocaine are significant. Take the creative class: both marijuana and cocaine use are positively and significantly related to states with higher concentrations of the creative class.
Correlation coefficient: 39**
Correlation coefficient: 36**
Now look at the results for the working class, where the pattern is reversed. Both marijuana and cocaine are negatively and significantly related to the concentration of working class jobs in state.
Correlation coefficient: -.35**
Correlation coefficient: -.36**
Note: * indicates statistical significance at the .05 level; ** indicates significance at the .01 level.
Richard Florida
Come on, this is just lazy pop statistics. The way you are presenting the data here suggest that you believe there is a causal relationship between pot/cocaine use and increased number of "creative" occupations. If that is really what you believe then show a model that actually controls for alternative hypotheses (e.g., percentage of a population that is enrolled in 4 year colleges). There might be a lot of pot smoking in Vermont because UVM is traditionally known as a hippie/party school. There might also might be a larger creative class in Vermont because of the presence of UVM. That does not mean that the higher percentage of pot smokers is what makes UVM generate a creative class (or that the presence of a creative class in Vermont is responsible for the higher levels of pot smoking).
Note the use of the word "relationship". He does not imply causality, rather it is just correlation, which is interesting. He doesn't say "drugs cause people to be creative", but that this class tends to use more drugs. Why that is has been left open.
note to self: read all comments before posting.
sorry sparky
Quick comment about the youth effect on our drug variables. We've only included the age group +26, to exclude any such effects. The idea is to capture individuals who to a higher extent would vote, have a job, an income etc (well, maybe not personality :-). However, there is also data on illegal drug use available from the age of 12.
i still don't agree with your statistical methods, but at least you're making a statement that i can agree with in this one!
In the three postings on drug use vs demographics, you look at factors according to states only. Can you break them down some other way? For example, cities are centers of both the greatest concentrations of creative class, the most wealth, but also a lot of grinding poverty. For example, Baltimore is both the drug addicted poverty trap of THE WIRE and also world leading Universities. The two populations do not mix, in general.
Thus, any correlation between drug use and wealth, creative class and employment is due not to any particular causality, only that they are more commonly found where there is a third factor, a city.
Pretty weak all around, esp. conjuring up an image of creative hipsters sitting around smoking pot and listening to Miles Davis. The thing that stood out more to me in terms of association is not "creativity" (however defined and measured by BLS), but geography. New England and other culturally liberal states stand out in the chart for their high use, cum acceptance, of drug use. The geography of drug use seems to coincide with the geography of religious identification. According to Gallup, the U.S. States With highest percentage of the Non-Religious are, in order, Oregon, Idaho, Washington, Colorado, Maine, California, New Hampshire, Nevada, and Arizona, (note that pot-loving Vermont is near the top of the unaffiliated list in the ARIS/American Religious Identification Survey on religion) which matches up pretty neatly with the states at the top of these charts (with a higher R2, I suspect). Likewise at the other end. Seems more plausible than Florida's premise. It seems sometimes you can crosstab any two sets of numbers and come up with a hypothesis.
MR FLORIDA!
Please see me after class!!! Amazingly enough you have managed to foreshadow nearly every statistical misstep POSSIBLE - Prior to drawing ANY conclusion. You need to speak with me so that I can validate your understanding, BEFORE the close of drop/add ....
Of course these are CORRELATIONS -- nothing else. And NO, we cannot break them down on a finer level since we use data from another source (state level) --but that would have been GREAT material to work with in a statistically nicer way. And YES--it would be great to test in a multivariate context, but with 50 observations any regression with at least some explanatory variables would leave us with a VERY LOW degree of freedom (a regression with 50 observations is never great either).
So, we basically do what we can do-- we use the aggregates, we don't claim there is any causality involved, we just say what we see in those bivariate relations (which very well may be proxies of something else), and that's it. We also run partial correlations for most of this, controlling for other factors, e.g. human capital levels, but the relations stayed the same.
All - Thanks for your comments. I was always the geeky type who liked to stay after class to learn more, anyway:-)
Charlotta, Thanks for weighing in on the analytics. Just to clarify, "Charlotta" here is Charlotta Mellander, a regional economist and statistician, and my frequent collaborator. She puts the matter in the proper perspective.
Part of the issue may be that I put the qualifiers in the first post, from Monday, but then didn't repeat them in subsequent posts. Basically what I said there is that we came across a treasure trove of sstate level data on illegal drug use from the National Survey of Drug Use and Health. So, Mellander and Cambridge University personality psychologist, Jason Rentfrow, decided to take a look at the relationship between drug use and various political, economic, and psychological characteristics of states. We know that there has been abundant, terrific research on the effects of factors like income, poverty, and race on the propensity to use drugs. But we wanted to extend our own work which has focussed on role of the psycho-social, as well as economic factors, on state and regional outcomes. We noted there a pioneering study by Rentfrow, Sam Gosling, and Jeff Porter that identified the effects of personality factors on state-level economic and social outcomes. And I state that we basically wanted to extend this line of research to see if and how these various economic, demographic, and personality factors might be related to drug use.
I also pointed out that as provisional findings we were sharing early results and correlations or associations, we were not putting forth a causal model, though we hope that over time our research could lead toward a better, deeper understanding. And I said, that while we are knee-deep in this more extensive research project, our preliminary results looked so interesting we thought we would report them and encourage feedback.
We are really interested in the kinds of things that cluster at the sub-national level. This relates SJE's important comment about cities. We wish we had the city or metro-level data and some day perhaps we will. But our goal here is to extend out work which has been looking at economic outcomes like income, wages, jobs, housing prices and innovation to include social outcomes as well. This is out first try really, and the state level data allows us to begin to do some empirical work in that space so to speak.
Airish, do send a link to that data on religious affliation if you can locate it.That is something we'd like to look at.
Thanks again, all for your helpful comments. Hope this clears some stuff up.
RF
Re: Airish, do send a link to that data on religious affliation if you can locate it.That is something we'd like to look at.
I sent you that information yesterday -- don't see it posted. Did you recieve it?
Thanks for the clarification/explanation. While I understand data limitations, Mellander's note about sample size of 50 and "explanatory variables would leave us with a VERY LOW degree of freedom" doesn't hold water. You have a theory...to say otherwise would be disengenuous. Given that you have a theory (and even allude, with the mention of partial correlations to the existence of intervening variables), any presentation of bivariate correlations is almost deceptive. Is it more important that you meet artificial standards of acceptability (i.e., p
Interesting that California, which has the wacko liberal rep, is pretty close to the center. And what's up in Vermont & RI??
Maybe some of the differences lie in that the "working class" is generally far more exposed to random drug tests potentially leading to unemployment than the "creative class".
This seems to be hard to post as a reply, so here it is as a comment.
Note that the issue of religious affiliation and participation has been extensively surveyed by a number of organizations, including Gallup. There are a couple of surveys that are quite illuminating regarding the point I was making in my post. Gallup polled self-reported church attendance in 2006 (http://www.gallup.com/poll/22579/church-attendance-lowest-new-england-highest-south.aspx) and the results correlated very well with the drug use data (in the expected direction), with VT and NH at the low end and AL and LA at the other end of the spectrum. A more recent survey Gallup designed to measure the “relative religiosity of population segments” asked: "Is religion an important part of your daily life?" http://www.gallup.com/poll/114022/State-States-YES%20Importance-Religion.aspx. The results appear to be very highly correlated to the drug use data. I don’t have a calculated R2, but it appears that it would be much higher than some of the attributes addressed in these articles. Another survey I noted is the American Religious Identification Survey (http://www.americanreligionsurvey-aris.org/), which measures formal affiliation. Of course, religiosity is likely a surrogate for a set of social and cultural attitudes that also manifest themselves in behaviors such as drug use, but it seems more plausible that religious beliefs might arguably influence one’s propensity to use drugs, which is legally proscribed for the most part. I think that hypothesis is more plausible than the others advanced in these articles, such as working in a “creative” profession.