the weed, period, cohort problem

| Gabriel |

Over at EconLog Bryan Caplan asks about why marijuana decriminalization has gained such traction in the last decade. I ran a few xtabs on the SDA GSS data and said that it appeared to be mostly cohort replacement. Caplan and his commenter “agnostic” both asked if there wasn’t something special about recent periods. To check this out, I did the analysis slightly more formally and got this graph by plotting % favoring legal weed as a function of birth cohort, broken out by period.

So it looks like Caplan and agnostic were on to something — cohort is a big part of it but so is period. Since the period effect is non-monotonic it’s probably a true period effect rather than an age thing. I think the simplest way to describe the pattern I see is that more recent cohorts are more favorable to legal weed than older cohorts, but both the slope and the intercept rise in the last decade. My only guess as to why this is that it’s a policy diffusion process seeded by the 1996 California plebiscite that legalized “medical” marijuana — but I would chalk it up to diffusion, wouldn’t I?

To replicate this go to SDA and get the variables GRASS, AGE, and YEAR. I used the cleaning script and dictionary SDA provides and added this to the end.

gen cohort=YEAR-AGE
recode GRASS 2=0
keep if cohort> 1900
twoway (lowess GRASS cohort if YEAR<1980) /*
    */ (lowess GRASS cohort if YEAR>=1980 & YEAR<1990) /*
    */ (lowess GRASS cohort if YEAR>=1990 & YEAR<2000) /*
    */ (lowess GRASS cohort if YEAR>=2000) /*
    */, legend(title(period) order(1 "70s" 2 "80s" 3 "90s" 4 "00s")) xtitle(Birth Cohort) ytitle(Pro-Legalization)
graph export grass.png, replace