Thursday, 18 June 2015

Racial bias in mortgages?

Simon Collins at the Herald asked me for comment on a paper alleging racial bias in mortgage lending; his story's now up.

The paper is available here. It shows, using ordered logit regression, that people who self-identify as being more easily identified as Maori are less likely to own their own home, correcting for income and a few other variables. The paper's empirics say absolutely nothing about mortgages or banks. But the study nevertheless concludes:
"To sum it up in one sentence: results from a large national probability sample of Māori indicate that the more Māori you look, the less 'mortgage worthy' you are."
Here are a few alternative hypotheses:
  • The empirics correct for current employment and current income but not past employment and past income. If Māori employment histories are more varied than non-Māori, and if this also follows the "is identified as Māori " indicator, Māori will have less accumulated wealth at any given level of income, and this is not controlled in the study.
  • If those who look more Māori are given preference in state housing, then home ownership would also be attenuated.
  • If parental resources are negatively correlated with looking more Māori , then that also affects ability to put together a deposit on a house. Note too the potential influence of holding household wealth under Māori land tenure.
I also think they've an error in how they described the magnitude of the effect. Remember that this is an ordered logit regression. So you can't just take the point estimate and multiply it by the number of interval steps to get an accumulated effect; you have to ask your stats package to give you a predicted value at the different values of the category. At page 11, it really looks like they linearised from the point estimate:
Some readers may be wondering how large this effect is in practical terms. One way to think about it is like this: when statistically adjusting for numerous other demographics, such as differences in income, region of residence, and education, a Māori person with a score of 5.55 on our Perceived Appearance measure of Māori identity would be twice as likely to not own their home relative to someone with a score of 1 in Perceived Appearance. This is a statistically significant association, which in our view represents a large and extremely important difference in the rate of home ownership based solely on merely appearing more Māori.
They have an odds ratio of 0.82, which ought to mean that a step change increase in perceived appearance score from the mean score reduces likelihood of owning a home by 18%. I don't think that means that if you go 5.55 steps in the other direction (1/0.18) from the mean score doubles your likelihood of home ownership, except under some pretty strong assumptions. But it's been a little while since I've played around in ordered logit.

Here's the bit where Collins quoted me - entirely fairly:
However Dr Eric Crampton of the NZ Initiative think-tank said there could be many other explanations for this besides racial bias. For example, people who looked more Maori might have parents who did not have freehold properties to use as collateral for loans, a factor that was not surveyed.
"Banks would be throwing money away if they decided to not lend to somebody simply based on looks," he said.
Mortgage brokers Bruce Patten in Auckland and Karen Essex-Mooney in Blenheim both said they had never seen a mortgage application turned down because the borrowers were Maori. They said many borrowers now applied online and never actually met the lenders.
New Zealand Bankers' Association chief executive Kirk Hope said racial stereotyping was not in the banks' or their customers interests especially within such a competitive part of the banking sector.


  1. Sigh. Tut tut tut. What silliness.

  2. Wasn't there a study a few years' back that found that there was some causality from SES to self-identification as maori? If so, this could be a source of correlation.

  3. Yes, I recall a commenter here pointing to that several years ago. It was something produced within government, I think, and then never talked about because it caused unrest.

  4. Found it:

  5. I haven't read the paper. You say it doesn't mention mortgages. Can I infer from that that they didn't actually ask the participants if they had tried and been declined a mortgage? Wouldn't that be an easier way to determine to test their theory?

  6. They were using others' survey and were constrained by the questions others asked. But the step to blaming the banks is a bit interesting where there are other plausible competing hypotheses and zero evidence around banks in their data.