Jay Parsons, VP and Deputy Chief Economist at RealPage, outlines why the mass eviction narrative may be over-exaggerated.
Nearly all studies predicting an evictions tsunami are based on the Census Household Pulse Survey. The name implies prominence, but the Census itself cautions that HPS is far less robust than other datasets. There are two big fundamental flaws.
First, it’s an incredibly small survey. In the most recent periods, the Census captured around 70,000 respondents per surveyed period. That’s not just renters. That’s all households. Reports indicate that renters comprised anywhere from 16% to 25% of respondents. That implies somewhere between 10,000 to 20,000 renters participated in each of the recent Pulse surveys. How significant is that? It represents between 0.02% and 0.04% of the nation’s 47.2 million rentals. Not good.
Second, the Census itself disclosed a big methodology challenge related to “non-response bias.” (The Census employs many brilliant researchers, but they’ve long been hamstrung by underfunding.) This gets a little wonky, but it’s critical to understanding the facts behind the evictions narrative – and why HPS may be under-representing renters paying their rent.
The Census labels HPS data as “experimental,” urges users to “take caution” when analyzing the data, and recommends researchers include warnings on non-response bias in their work (which very few do). Census researchers usually work around small-sample issues with sophisticated weighting methodologies. But the Census admits unusual challenges with non-response bias in HPD surveys—noting that “weighting adjustments are limited by a lack of information about nonresponding units.” Because HPD covers so many new topics (including rent payments), there’s no benchmark on what subpopulations are missed or underrepresented. In simple terms? It means HPS has a huge blind spot because it doesn’t know what it doesn’t know.
Despite all these caveats from the creators of the survey, eviction forecasters have largely ignored the warnings and mispositioned the data to be rock solid.
Read more here.
