Press "Enter" to skip to content

The Mathematics of Gerrymandering | Wesley Pegden | TEDxCMU


so I’m going to be telling you about
using mathematics to detect
gerrymandering so to start off I just
want to briefly review a basic aspect of
our political system in the United
States so we have this body called the
House of Representatives it consists of
438 members and the way these members
are chosen is that they’re elected in
individual districts 438 individual
districts across the country so each
state gets a number of districts roughly
proportional to population of the state
and there’s single winner-take-all
elections in these districts determine
the outcome so the districts because
they have to because the number of
districts depends on the population of
the state and the district has to be
roughly equal in population a redrawn
every 10 years so this is one of the
reasons why we have the US Census and
because this is how representatives are
chosen there’s no mathematical guarantee
that the number of representatives from
each party in the house will reflect the
proportions of votes cast for each party
overall and congressional elections
right so in particular the way that this
works is not that we look and see if the
Democrats won 52 percent of the vote
then they have to get 52 percent of the
seats that’s not how it works right
instead we have these individual
elections and in practice these two
things can wildly divert diverge so to
give you some examples in 2012 in
Pennsylvania Democrats want a slight
majority of the vote but one only 5 out
of the 18 congressional seats in
Pennsylvania so it was 5 for the
Democrats 13 for the Republicans in a
state where most of the votes were cast
for Democrats okay it’s not just in
Pennsylvania the same thing happened in
North Carolina Democrats won a slight
majority of votes but got only 4 out of
the 13 congressional seats in North
Carolina okay
that same year if you look in Wisconsin
now this is not at the congressional
districts in Wisconsin but the state
legislative districts so this is the the
state’s equivalent of the House of
Representatives
so there’s 99 in these seats in their
state House Democrats won a majority of
votes quite strongly actually 53 percent
but captured a minority of the seats
okay so there’s there can be this huge
divergence between the fractions of
votes cast for each party and the number
of seats they win and just to see that
this really has a nationwide impact in
2012 overall Democrats won more
congressional votes than Republicans but
did not even
they come that close to capturing
majority of the house okay now these
examples may give the impression that
it’s only Republicans gerrymandering
state so that’s not quite true there are
states so Maryland is one where people
think the Democrats are gerrymandering
districts district things but right now
for various reasons the general tilt is
that there seem to be more auditioning
bias to Republicans as you can see in
this last statistics okay now when you
see these results it’s very surprising
and so the question is so somehow what’s
causing this where is this coming from
and we said that the way these elections
are held is that there are these
individual winner-take-all elections in
districts okay so somehow what’s causing
this has to be the districts okay and so
this is a map of the congressional
districts of Pennsylvania that we’re
used for all congressional elections
since the 2010 census okay so there’s 18
districts in this picture by far the
most famous is district 7 which is this
red district here so this has been
called the goofy kicking Donald Duck
district so I’ll be goofy I’m like this
my ears are flying backwards and I’m
very happy to be kicking Donald Duck
okay and right so it seems kind of
ridiculous like how could you possibly
end up with a district safe like this
well one way you could imagine getting a
district a district shape like this is
that well I have this part of the state
close to Philadelphia that’s generally
Democrat leaning suppose that I wanted
to construct a district which would
elect a Republican representative well I
could find little patches of this area
that were barely Republican leaning and
pasted them together to make one
district and then of course I might get
something shaped very strangely and you
get something like this and this is
exactly what gerrymandering is so
gerrymandering is carefully drawing
district lines to serve some political
goal okay and so and we can see in some
cases it seems like people can do this
quite successfully now it turns out that
Pennsylvania has a new congressional map
so this is I said this map has been used
for every congressional action since
2010 but not anymore so now going
forward this is the map that will be
used for congressional elections and
this map is in place because of a
lawsuit brought by the League of Women
Voters against the Commonwealth of
Pennsylvania so there was a trial heard
on an expedited schedule in December and
then the Pennsylvania Supreme Court
ruled the map unconstitutional in
January
the Republican legislature appealed for
emergency stage to the US Supreme Court
and Justice Alito denied their emergency
stay in February and then the
legislature didn’t meet the deadline set
by the Pennsylvania Supreme Court to
draw a new map and so the Penn the
Pennsylvania Supreme Court hired an
expert to draw this map and then finally
so one more appeal went to the Supreme
Court the Republican legislature said
please please let us use this old map at
least one more time and they said no you
have to use a new map and that just
happened March 19 so this is really
late-breaking news okay so we just very
recently had this map so I was an expert
in this case and so one of the things I
want to tell you about is what kind of
evidence you can bring to a court how
you can use mathematics to show that you
can know rigorously that a state is
gerrymandered right so you can really
meet some some legal test of knowing
that there is intent to gerrymander of
state okay now what kind of evidence
could you imagine bringing to such a
case well we saw these statistics that
were quite striking that with more than
50% of the vote Democrats got just 5% of
us just got 5 out of 18 seats and
already I think when you first hear that
that seems like case closed
obviously it’s jerem a nerd what other
explanation could there be right so here
was this old map it had this crazy
property you could have a 513 split with
just a 2% with a 50/50 vote split okay
so at first this seems convincing in
itself and maybe that should be the
whole story but there’s a slight problem
with this okay so I want to carry out a
thought experiment suppose that I
decided to start a mathematical
collaboration with the zoo okay so what
I’m going to do is I’m going to
collaborate with the zoo they’re going
to lend me a troop of monkeys and we’re
going to work hard to train these
monkeys to draw a political district
it’s okay we this is hard work actually
because they have to draw the correct
number of districts and the districts
have to be connected they have to be
roughly equal in population that would
be a really hard one to teach them how
to do okay but suppose I managed to do
that and at the end I had these monkeys
they can draw districts okay and suppose
that they drew this map so there’s just
some sort of random map okay so I didn’t
actually use monkeys to draw this map no
monkeys were harmed and the producing of
this presentation this is so a computer
drew this random like math but it turns
out that this map okay it’s not as bad
as the map we saw before but with the
50/50 vote split this is like a set of
with a 50/50 vote split this is like a
7-eleven seat split so still the
Republicans have a surprising advantage
with this map even though it was not
drawn with intent so what this shows is
that just looking at election results is
not enough to to be confident that
somebody tried to make the map unfair
okay now what’s another kind of evidence
that you could try to bring you could
say well look at goofy kicking Donald
Duck I mean the shapes are ridiculous
what other explanation could there be
and that is a reasonable kind of thing
to look at and the shapes of the
districts were discussed in this League
of Women Voters case but you have to be
a little bit careful about using this as
your sole standard so I’m going to show
you three different maps so I’m gonna
call these so this is the old map and
now here’s map a here’s Matt B and
here’s map C so each of these maps I
mean I’m not sure really what you should
take away from them none of them look
quite as crazy in terms of their shapes
as the old map okay it turns out that
one of these three maps was drawn as a
very biased proposed replacement for the
old map it was almost as gerrymandered
as the old map even though the shapes
are much nicer and the other two were
drawn by an algorithm that ignored
partisanship and was just trying to do
things like minimize County splits but I
would submit that when you just look at
these maps and eyeball them it’s hard to
tell which is which which is the fair
amount which are the fair maps here
which is the unfair map so it turns out
that this first map is the gerrymander
and these two are drawn by the
simulation but I’m not sure that you
should be able to tell just by looking
so this this suggests that shape is also
alone not really enough to evaluate the
strings for bias okay so how will we
how will we test for gerrymandering the
districting okay so our gerrymandering
test is going to be very cautious okay
so we’re not going to infer
gerrymandering just from election
results we’re not going to infer
gerrymandering just from the shapes of
districts instead we’ll only call
something gerrymandered if it if we can
show that it’s carefully crafted for
partisan bias if we so we’re going to
demonstrate somehow that the map had to
be carefully drawn to achieve the
parsing bias that it sees and we’re
going to do this using two things
randomness and mathematics okay and
we’ll see how these two things come into
play okay so first how do we use
randomness to evaluate district things
so to use randomness we’re going to be
making small random changes to
districting so here on the left you see
the legislative district of Wisconsin so
this is this Division of Wisconsin 299
districts okay and we’re going to make
small random changes to these to this
districting over and over and over again
to produce a sequence of districting so
what does a small random change look
like it looks like this does anybody see
what’s changed
so normally nobody can see was changed
but maybe we have so many people
somebody can see what changed so that
changed right there right so I picked a
I picked a little precinct on the
boundary of two districts I checked that
switching this precinct over wouldn’t
disconnect a district wouldn’t change
the population equality too much things
like that and then the change is made it
switches from one district to the next
okay to employ our test we make random
changes like this over and over again to
a districting okay so here something
like hundreds of changes are being made
every second to the districting
okay when we actually do our analysis
for example in Pennsylvania case or we
also submit an amicus brief in the
Wisconsin case we’ve run our analysis
for trillions of steps so to show you a
video running at this speed for the
analysis that we actually do for these
cases it would take you a thousand years
to watch the video okay so this is much
slower than it actually runs okay and
what you can see is that we have this
districting and it’s just sort of being
fuzzed up a little bit right it’s just
taking this little you know random walk
and things are changing a little bit and
you’re getting districts that are not
that different from what you started
with but they’re a little different okay
and what happens when we do this is we
see that the districts get
overwhelmingly fairer than what we
started with okay so in particular I can
say when we run this test we see
trillions of districts of what’s
constant okay how many of them are as
parson as the one we started with okay
so I think that’s to ten millions of the
districts are as bad as the one we
started with that tiny fraction so I
start with this district Wisconsin I
make these small changes to it I get new
district new district new dishing each
just one small step away and then when I
look at this whole sequence almost all
of these are fair than what I started
with okay when you do the same analysis
to Pennsylvania okay so with
Pennsylvania what fraction of maps did I
encounter or as bad as what I start with
and it’s a similar answer here there’s
actually even more zeros this answer is
for ten billions of the maps for as bad
as what you start with so you you store
this map you start making little random
changes and you find that
almost everything you encounter really
almost everything you encounter is as
bad as what you started with so to put
this in the context of the video I said
that to watch the whole video would take
you a thousand years after the first
minute of the video though you’re
already you’ve already left this space
where you see maps as bad as what you
started with so for the rest of those
thousand years
everything is fair than the initial math
okay so it’s quite striking
now when you hear this again it’s like
case closed this is like a very nice
intuitive test for gerrymandering we
just make random changes to it if the
parson ship goes away this demonstrates
the map is gerrymandered but hold on we
have to be a little bit careful because
we had an intuitive test for
gerrymandering before well let’s just
look at the election results if there’s
a big divergence between the the votes
cast and the seats one that seemed like
a reasonable test for gerrymandering and
we saw that that had a problem that on
the monkey map we decided that that
monkey map might be gerrymandered
according to that test right because
there was already a divergence between
seats and votes so do we have the same
problem for this test so far we’ve only
used randomness we have this test it
seems like a reasonable test so is it
possible that the monkey map would also
fail this test because then that’s a
problem then you can’t go to the court
and say look it’s failed our test so we
know that they tried to gerrymander the
state because for all we know this is
just some property of maps in general of
Pennsylvania okay so what we’re asking
is is it possible that if I just take a
typical district of the state just for
example a random districting the monkeys
drew district is it possible that that
districting could just ten naturally to
have the property that when I make small
random changes to it it becomes less
partisan like is it possible if the
Democrats and Republicans are arranged
in the state in some strange way that
gives that our state that property is
that a property their estate could have
and it turns out the answer is no it
can’t have that property and that’s a
consequence of a mathematical theorem
that we proved and okay so the precise
statement of the mathematical theorem is
it’s a slightly technical but not so
much so if you’re interested in math I
encourage you to look it up but roughly
speaking what it says is that if I start
with any typical object that random
changes won’t consistently change the
tip the typical object so in particular
if I really start with a typical
district in Pennsylvania it would not
happen than when I made random changes
to it it would become consistently less
partisan
and this is true not just for Jeremy
nuran but in a very general setting so
there’s there’s a very general setting
which this phenomenon holds and we can
prove that this is the case so what does
this mean
quantitatively it means that when I
discuss results like this for
Pennsylvania
so before we said for Pennsylvania at
most for ten billions of the maps that
we encounter we’re as bad as what we
started with okay what our theorem does
is it lets us quantify how likely that
could be to happen just for a typical
district in the state let’s say you know
somebody just pulls the district out of
the bag they’re not trying to do
anything evil now they’re not trying to
gerrymander the state how likely could
it be that you would fail our test as
badly and there’s a formula that our
theorem gives you some rigor for these
numbers it tells you that the chance is
almost 300 thousands then you would be
this bad okay so then you can go to the
judge you can say look it failed our
test the probability of failing this
test if nothing bad was going on is just
300 thousands and so I mean this is a
crazy level of certainty that you
usually can’t bring to a courtroom you
even have a mathematical proof asserting
that you’re bound on the probability of
being misled is correct okay and so this
is how we combine mathematics and
randomness to detect gerrymandering and
interstates okay thank you very much [Applause]
Please follow and like us: