Injecting Data into the SF Rent discussion

2017 May 17
by Daniel Lakeland

There was a massive shitstorm on Andrew Gelman’s blog about SF rents, which are of course a hot button topic. Since I have the complete ACS microdata, I figured I could inject some data into the discussion at least. So here it is, the complete distribution of rents for occupied units each year in all of the SF related Census PUMAs… In total about 8328 household observations across 2010-2015


Click to see full size

Tabular values from each year:

  YEAR   MedRent
1 2010 1.0000000
2 2011 0.9565217
3 2012 1.2173913
4 2013 1.3043478
5 2014 1.3913043
6 2015 1.3913043

Note, the above was the original table I published, but it is affected by the fact that the Census discretizes the rent. Adding a normal perturbation to the normalized rent with sd = 0.01 creates a smoothed distribution, and then you can get better estimates on the median and mean of the real distribution:

  YEAR   MedRent  MeanRent
1 2010 0.9959745 1.0020145
2 2011 0.9621274 0.9797048
3 2012 1.2243687 1.2844332
4 2013 1.3043896 1.3762188
5 2014 1.3854716 1.4446855
6 2015 1.4144859 1.5111725

So clearly the distribution has ratcheted up each year, and more at the right tail than anywhere else (mean more than the median). Also Top-coding is a big problem here, lots of the rents are reported at the maximum allowable top-coded value rather than their real value because of privacy issues.

2 Responses leave one →
  1. Morgan permalink
    May 20, 2017

    The break in distribution shape between 2011and 2012 seems impossibly large. Any idea what happened there?

Leave a Reply

Note: You can use basic XHTML in your comments. Your email address will never be published.

Subscribe to this comment feed via RSS