Transit Equity: Scale and Correlation

Yesterday I noted that the ACS 5-year/Census block group demographic correlations were significantly lower than the 2000 Census tract correlations.

So let’s look at the differences:

  1. Different (newer) data
  2. Different areas – we included all Census tracts that touched any part of the service district, while we only used block groups that were substantially inside the service district
  3. Finer-grained data (about 3 block groups for every tract)

So to start peeling it back, let’s isolate some of those differences. First let’s recall the 2000 Census tract correlations.

  • Transit Score/Density: 0.67 (1.00 would be perfect correlation)
  • Transit Score/Percent non-white: 0.40
  • Transit Score/Median 1999 Household Income: -0.52

So for apples/apples, let’s take ACS 2009 data for the same census tracts and correlate (spreadsheet here):

  • Transit Score/Density: 0.67
  • Transit Score/Percent non-white: 0.28
  • Transit Score/Median Household Income: -0.50

That’s a lot closer! What I think we see is that over the last decade, we’ve become more racially diverse and the population groups of color have dispersed through more of the region.

So next let’s deal with the fringes of the region. If we drop the tracts largely outside the service district here’s what we get (spreadsheet):

  • Transit Score/Density: 0.62
  • Transit Score/Percent non-white: 0.28
  • Transit Score/Median Household Income: -0.50

So the density correlation drops, because we just dropped out some tracts with low density and effectively no transit service.

But the big jump comes when we move to block groups (spreadsheet):

  • Transit Score/Density: 0.53
  • Transit Score/Percent non-white: 0.23
  • Transit Score/Median Household Income: -0.39

So the big learning for me is that scale matters – looking at a census tract level averages things out too much, and we’re likely to have more insights analyzing block groups.

[As always, better statisticians are welcome to correct me.]

What’s next – we’ll accumulate some more metrics for possible correlation, then we’ll start to look at outliers (things that don’t have service levels the correlations would predict) and hopefully we’ll recruit a real statistician who can help us do some multivariate correlations.

2 responses to “Transit Equity: Scale and Correlation”

  1. Good stuff.

    Two opinions:

    1) I’d convert from square meters to square miles for better density understanding. It’s hard to picture meters and it’s such a small unit when discussing a regional scale.

    2) This needs to be mapped! GIS all the way once you get the data side figured out.

    That’s all I’ve got.

  2. It’s interesting to me that the correlation between transit score and household income is strongly negative. It seems likely to represent a real phenomenon—on average, people are richer in suburban-style neighborhoods, which have worse transit access.

    However, I wonder if median household income truly captures how rich people are. I would be interested in seeing how changing it to median HH income per person changes things. I would expect that it would make the inner city (with on average smaller household sizes) look richer.

    Personally, I think the truth is somewhere in between HH income and per capita HH income. For example, consider a family of four with two parents who each earn $50,000; and a single person who earns $25,000. Is the family FOUR TIMES better off than the single person? I don’t think so (so HH income alone isn’t quite right). But I think it’s also easy to argue that the single person is in more financial difficulty (and so HH income per capita isn’t quite right either, as both HHs have a per capita income of $25,000).

    I’m not a statistician, but I’m a data analyst with statistics experience. I could run some multivariate regressions on your data if you’d like! Feel free to contact me about it at 240-888-1349 or .

Leave a Reply

Your email address will not be published. Required fields are marked *