Intentional discrimination versus disparate impact
I’m paying lots of attention to the Supreme Court’s coming decision on The Fair Housing Act. A New York Times editorial of this morning does a good job explaining the issues, including the concern that Chief Justice Roberts seems to think we’ve moved past racial discrimination in this country.
The burning question is whether housing developments and the like are responsible only for intentionally discriminating against individuals, or whether they are responsible in a more general, statistical sense, of having disparate impact on groups of people. The New York Times, like me, hopes for a broader reading, consistent with the 11 courts of appeals decisions over the last 40 years. From the Times:
The ability to show discriminatory effect has only become more important as intentional discrimination has become harder to prove. To take one prominent example, the Justice Department relied on it to negotiate the largest-ever fair-lending settlement — $335 million — with Bank of America in 2011. The bank’s mortgage unit, Countrywide Financial, had charged higher average fees and interest rates to black and Latino borrowers than to whites with the same credit risk, a practice that former assistant attorney general Thomas Perez called “discrimination with a smile.”
This case is focused on housing, but of course it could generalize to all sorts of other systems, including job applications and credit applications among others.
If we stick to the “intentional discrimination” only, we are opening up a door to (even more) widespread use of algorithmic decision-making that produces unfair and discriminatory results. And as it turns out, it’s easy to produce a model that effectively discriminates.
And if you are not in charge of your own system, then who is?
mathbabe 
” And as it turns out, it’s easy to produce a model that effectively discriminates.” What do you mean by this? Are you referring to a model that exists?
LikeLike
There definitely are known models of loan default risk that, although they don’t formally include race, manage to tag black people as worse risks based on other commonalities in the data. People who were having to fight to overcome discrimination in the past, have histories that look worse in the present.
LikeLike
One of the problems with disparate impact in housing is the desire of people to have contact with others who are like themselves. Would I feel uncomfortable being the first white guy in an all-black neighborhood? Yes. If I knew there was another white family on the block would it be easier? Yes.
It’s easy to run simulations that demonstrate that the desire to have as few as 25% of your neighbors like you leads to completely voluntary segregation.
If you are used to living in dense urban environments where there are lots of diverse groups, this may be hard to understand, but the desire to have others like yourself around is almost universal.
LikeLike
My favourite simulation showing this result is here Parable of Polygons.
LikeLike
When our promised apartment in Married Student Housing in Berkeley failed to materialize, we were the second and third white people to move into the all Black projects in Richmond, CA. Long story short, our neighbors made us feel very unwelcome, and we were able to break the lease at the end of the first month and move into Married Student Housing in Albany, CA, which was integrated ethnically, racially, and internationally.
LikeLike
Disparate impact (discriminatory effects) and disparate treatment (discriminatory intent)
were principally types of data to be used by minority (members of legally recognized
suspect class) to establishing standing in the complaint process. Disparate impact
involved a complainant showing that s/he was a) claiming harm, b) that the harm was
due to being a member of a suspect class, and c) others in the suspect class were
similarly treated in a discriminatory fashion.
Once disparate impact analyses were presented, the defense was to address why
this data was not compelling. If no defense was or could be mounted, the complainant
prevailed. Or the defense might have been judged to be inadequate. Again, the
complainant would prevaiil.
Those battling against efforts to equalize opportunity or provide a legal venue for
redress over time were instrumental in reducing the primacy disparate impact based
complaints. This was done in a variety of ways. In this era the Supreme Court’s
majority has shifted to disparate treatment: unambiguous, witnessed acts of
racial, gender, etc. prejudice and discrimination. In effect, the bar for legal evidence
has been set much much higher. Instead of beginning with the possibility that
“where there’s smoke, there may be fire,” the courts are instructed to look for
“the smoking gun in the hands of the accused.”
While I know of no mathematical model that entails using proxies for race and
gender, thereby sidestepping the matter of disparate treatment), we know from
social practices (housing development and its sidekick, education) that there’s
disparate impact and treatment, but nothing explicit in these that suggests covert’
or overt racial, etc. discrimination.
As to housing segregation, the matter isn’t personal preference, but whether a
person in a suspect category seeking to live in a particular area is treated
equally in rental or purchasing conditions. Personalizing is one of the diversions
that takes the eye off trends and legal conflicts which are often messy affairs.
LikeLike
Ultimately disparate impact leads to some troubling problems. Cathy, how many Blacks do you have in your building? How many Chinese-Americans? How many Korean-Americans? Irish, Italian, Greek, … Can you really have a microcosm of the ethnic makeup of Manhattan, of NYC in your building? You are certainly in violation if disparate impact is used to measure discrimination. Same in my building. How about postal workers? In my local post office they are all Black. Are they in violation? No middle aged short white guys playing in the NBA. Must be discrimination. Disparate impact.
LikeLike