A recent paper by Roberto Foa (University of Melbourne) and Yascha Mounk (Harvard) suggesting that support for democratic governance is declining among younger generations has generated a fair bit of buzz recently. The New York Times ran a relatively long story on it, highlighting a graph purporting to show declines in support for democracy among younger generations that had been making the rounds on Twitter (see for example here, retweeted nearly 6,000 times). Below is the NYT version of the graph (which, as far as I can tell does not appear in the actual paper, so I am a little unclear about where exactly these data came from):
These "declines" are dramatic, but this is essentially a snapshot of two pooled cross-sectional datasets (Waves 5 and 6 of the World Values Survey data, corresponding roughly to surveys conducted around 2006 and 2012). By "snapshot" what I mean is, we survey individuals at a single point in time (e.g., in 2012), and then we plot the percentage of individuals responding that it is "essential" to live in a democracy according to the year they were born. Fine. Nothing wrong with that. And I don't doubt that these estimates are correct in showing that individuals born more recently are less likely to say that it is essential to live in a democracy.
However, it's important to note that these estimates, on their own, say nothing about trends in beliefs about democracy, unless one is willing to assume that individual perceptions of how "essential" it is to live in a democracy are fixed at birth. I'm not a political scientist, but I certainly know that many of my political opinions have changed over the years (there must be data on this, but I'm too lazy to search for it right now), and I'd gather that is true for many others.
Nevertheless, the message the New York Times was pushing clearly did suggest that faith in democracy was declining:
Saying the percentage has "plummeted" strongly suggests that is has declined in a substantial way over time, yet the graph shown above cannot tell us whether or not that is true.
The authors of the original paper made it pretty straightforward to see how they constructed their numbers, and the WVS has a brilliant website where you can easily download the data, so I grabbed the data and wanted to see whether a closer look at more than one survey would provide a similar picture.
With a little tweaking (code is at the bottom of the post) I think I got pretty close to reproducing the NYT figure:
Okay, so now it seems like we are working with basically the same outcome data as did the authors. What I'm interested in here is the change in the fraction of people reporting that living in a democracy is "essential" (measured as 10 out of 10 on a scale of importance). Because the WVS asked this question in both Wave 5 (around 2006) and Wave 6 (around 2012), we can look at a couple of things. First, we can see whether or not the proportion reporting that it's essential to live in a democracy has "plummeted" between the two surveys, as the NYT claimed above.
Second, and more relevant to the broader point I believe the authors were making about younger generations not valuing democracy in the same way as their elders, we can look and see, for individuals born in the same decade, whether their views are actually changing. In the table below I have calculated, just for the United States, the percentage reporting that it is essential to live in a democracy for each birth cohort, by survey year for waves 5 and 6 of the WVS:
|Average within cohort change||-1.2%|
A few of things are worth noting here. First, as the authors of the paper and the NYT story emphasized, the decrease in the belief that it is essential to live in a democracy across birth cohorts is quite large (>70% for those born in the 1930s vs. ~35% for those born in the 1980s), and indeed this is troubling, particularly if younger generations do not change their thinking as they get older. (And as I noted before, I'm not a political scientist--there may be reams of very good evidence out there that opinions like this don't change much as people age, in which case this is all the more distressing. But an age effect is not the same thing as a secular trend, which is the point I am making here.)
Second, the claim that the NYT made about the "plummeting" of the percentage reporting that it essential to live in a democracy is a hard to see in these data. The overall percentage of decreased from 53% to 51% in the roughly 6 years between survey waves. I do think it is probably a bad sign for this kind of indicator to be declining over time, but I find it very difficult to characterize this as a "plummeting" change.
Third, and perhaps most interesting, if you look across the rows you can see something interesting, which is the change within the same birth cohorts. For example, the first row shows that 75% of people born in the 1930s said it was "essential" to live in a democracy in 2006, but 6 years later only 71% of people in that same birth cohort said it was essential. If anything, this suggest that it is older, rather than younger individuals, that seem to be "losing" their belief that living in a democracy is essential. In fact, the percentage regarding it as essential to live in a democracy is actually increasing over time in the "younger" cohorts of the 60s and 70s. (In fairness, the 1980s cohort does show a decline). The average within-cohort change (weighted by cohort size) is a decline of about 1 percentage point, but you can see this is due to very different patterns for older and younger cohorts.
I did this same exercise for 3 of the other countries shown in the figures above (data for England and New Zealand for this outcome were not available in both survey waves) and plotted the results below. Each graph shows the change in opinion within each birth cohort between the two surveys, and a 95% confidence interval for each estimate).
The estimates I showed in the table above for the United States are reflected here, and one can also see that it is hard to paint a picture of anything like robust declines in believing that it is essential to live in a democracy. In Australia, for example, the fraction of individuals who consider it essential to live in a democracy is actually increasing for every single birth cohort, despite the fact that it is lower in general for younger cohorts (the overall percentage in Australia also increased from 63% to 68% between the two surveys). Of course, there is measurement error here so nothing is really definitive for any country, but this pattern of results is considerably more nuanced than the NYT reported. I have not read the entire academic paper in detail, but I think it is safe to say that the authors do not use only this information to form their conclusions about the prospects of democracy as a form of government in rich countries. But I do think that many people (including Amanda Taub at the NYT) have stretched the interpretation of these data in ways that go well beyond what they are capable of revealing.
For anyone that is interested I have provided the Stata code I used to generate these numbers at the bottom of this post:
capture log close log using declining-democracy.txt, replace text // program: declining-democracy.do // task: replicate NYT (How Stable Are Democracies? ‘Warning Signs // Are Flashing Red’ http://nyti.ms/2gdv7I6) // author: sam harper \ 30nov2016 // #0 // program setup version 14 set linesize 80 clear all macro drop _all // #1 // load World Values Survey data for Wave 5 * Wave 5 data use V1 V2 V162 V236 using /// "WV5_Data_stata_v_2015_04_18.dta", clear * restrict to AUS, SWE, NED, USA, NZL, GBR keep if inlist(V2,752,36,528,840,554,826) * rename variables rename (V1 V2 V162 V236) (wave country impdemo byear) * generate survey year midpoint gen syear = 2006 label var syear "Survey year (midpoint)" * save temporary dataset tempname wvs5 save `wvs5', replace // #2 // load World Values Survey data for Wave 6 * now for Wave 6 data use V1 V2 V140 V241 using /// "WV6_Stata_v_2016_01_01.dta", clear * restrict to AUS, SWE, NED, USA, NZL, GBR keep if inlist(V2,752,36,528,840,554,826) * rename rename (V1 V2 V140 V241) (wave country impdemo byear) * generate survey year midpoint gen syear = 2012 label var syear "Survey year (midpoint)" * append Wave 5 data append using `wvs5' // #3 // generate outcomes and birth cohort categories * country order for graphing recode country (752 = 1 "Sweden") (36 = 2 "Australia") /// (528 = 3 "Netherlands") (840 = 4 "United States") /// (554 = 5 "New Zealand") (826 = 6 "Britain"), gen(cname) * generate outcome variable: How impt to live in a democracy? * 1 = not at all, 10 = absolutely recode impdemo (10=1 "Yes") (1/9=0 "No") (-5 -4 -2 -1 = .), gen(demo) label var demo "Democracy 'essential'?" * generate birth cohort categories and labels egen bcohort = cut(byear), at(1930(10)1990) icodes label define bcohort 0 "1930-39" 1 "1940-49" 2 "1950-59" /// 3 "1960-69" 4 "1970-79" 5 "1980-89" label values bcohort bcohort label var bcohort "Birth cohort" // #4 // plot outcomes by birth cohort * reproduce NYT figure qui logit demo cname##bcohort, nolog qui margins cname#bcohort marginsplot, by(cname) recastci(rarea) /// byopts(rows(1) imargin(medsmall) /// title("Percentage of people who say it is 'essential' to live in a democracy", /// size(medium)) /// note("Source: Author's calculations of the World Values Surveys, Waves 5 and 6", /// color(gray) size(vsmall))) /// ylabel(0 "" .25 "25%" .5 "50%" .75 "75%" 1 "100%", angle(horizontal)) /// xlabel(0 "1930s" 5 "1980s") ciopts(color(ltblue)) /// plotopts(lcolor(edkblue) mcolor(edkblue) msymbol(O) msize(small)) /// ytitle("") scheme(tufte) name(nyt, replace) xsize(6.5) graph export "nyt-replication.png", replace * within cohort change in proportion reporting "essential" qui logit demo cname##bcohort##wave if cname<5, nolog qui margins r.wave, over(cname bcohort) marginsplot, by(cname) xdim(bcohort) recast(scatter) recastci(rspike) /// byopts(rows(1) imargin(medium) /// title("Change in % saying it is 'essential' to live in a democracy, 2006 to 2012", /// size(medium)) /// note("Source: Author's calculations of the World Values Surveys, Waves 5 and 6", /// color(gray) size(vsmall))) /// ylabel(-.2 "-20%" -.1 "-10%" 0 "" .1 "10%" .2 "20%", angle(horizontal)) /// xlabel(0 "1930s" 5 "1980s") ciopts(color(ltblue)) /// plotopts(lcolor(edkblue) mcolor(edkblue) msymbol(O) msize(small)) /// ytitle("") scheme(tufte) yline(0, lcolor(red)) name(change, replace) graph export "change.png", replace log close exit