Predictions are Hard, Especially About the Future

DC is having a special election for City Council today, which has once again aggravated the long-running wounds of DC racial politics. I was reading the stories about the candidates, clicking from link to link, when suddenly I stopped at this arresting graphic:

Why is this so interesting? Because as DC residents know, that graphic was way off. Census data indicate that DC lost its black majority sometime in 2011, not 2014. If current trends continue, the city will become majority white sometime in the next decade.

Okay, so he was off by a few years. What's the big deal?

Because that graphic was made in late 2008. You expect long-range demographic forecasts to be off by a few years on either side. You don't expect it to be this hard to project outcomes a couple years in the future.

But this is in no way a criticism of Rob Goodspeed, who made the image. He was working off of the latest data available, and a (correct) belief that demographic change tends to be pretty slow. What this illustrates is three things:

1. The remarkable speed of demographic change in the District of Columbia. DC is creating a lot of high-paying jobs for highly educated professionals, who disproportionately tend to be white. Long-time black residents, who are less likely to have a college degree, aren't participating in the boom. As a result, the new residents are creating bidding wars that DC's semi-skilled workers, and even its old middle class, can't possibly win. (Nor can a lot of educated journalists and non-profit workers, who are increasingly worried that they're going to be priced out of the housing market before they can save a reasonable downpayment). I've never seen anything like the pace of gentrification here--and I lived through successive gentrification waves on New York's Upper West Side.

2. The limitations of the census data. Goodspeed was working with data from the 2006-7 American Community Survey, which became available in 2008. But the pace of gentrification was accelerating all through the aughties (and as far as I can tell, continues to rocket ahead). Using the 2009 ACS, which became available in 2010, would have produced a more accurate prediction--although as best I can tell, the 2009 ACS also underpredicted the actual amount of change recorded in the 2010 census. In part that's because the change kept accelerating, but there may be other reasons: for example, elderly residents may have been more likely to return their forms, and easier to reach by phone if they didn't. And elderly residents in the District are much more likely than the general population to be black.

3. The uncertainty of even "easy" predictions. Demographic prediction is one of the most accurate forms of prognostication; changes are almost always slow, so a straight line extrapolation is more likely than most such forecasts to be roughly accurate. Nonetheless, as this shows, there's always a chance that you'll end up pretty wide of the mark. Goodspeed did an excellent job with the data he had, and he was still too conservative in his estimate. Remember this the next time you see someone making a prediction based on much fuzzier and more volatile macroeconomic data.