Alabama Democratic Primary Proves New York Times’ Nate Cohn Wrong about Exit Polling

When I wrote the piece about exit polling in my series on Election Fraud Allegations, I specifically addressed the arguments Nate Cohn expanded yesterday in the New York Times into a post decrying conspiracy theorists who think the Democratic primary was stolen from Bernie Sanders. As is the custom for writers of his class, Cohn simply ignored my counter arguments. I’ll rehearse, in brief, Cohn’s expanded argument and how it fails to address the key points I made previously. More importantly, there is now much more powerful, visually-stunning statistical evidence to buttress the claim that old, provably hackable machines in particular counties helped Hillary Clinton outperform expectations. In short, the bigger the precinct size in terms of total votes for Clinton and Sanders, the better Clinton did, even when controlling for such factors as racialization and age.

Cohn’s basic argument, like so many others, is that exit polls in the United States are not designed to catch fraud. Tax evasion laws were not intended to catch gangsters, but Al Capone landed in jail anyway. Cohn is more specific than previous commenters, however, going into some detail on why early voting and age demographics could have skewed exit polls toward Sanders. The problem, as I noted previously, is that these theoretical arguments do not work when applied to specific places where exit polling failed. I used the case of Alabama for precisely these reasons. The exit poll released by the networks when the voting booths closed in Alabama was off by fourteen percentage points, far more than for any of the previous elections Cohn mentions. Cohn had made his arguments about early voting and race in brief on Twitter previously, and I addressed them specifically accordingly. Alabama did not have early voting (beyond the very basics required by federal law for military and infirm voters). Likewise, the exit poll for Alabama included so few young voters by percentage that there is no mathematical way to make that explanation work.

But what about my explanation? Three of the top four Alabama counties by population have very old voting machines that were badly hacked by a “red team” of university security experts more than eight years ago. Clinton did far better in those counties than in demographically similar counties in Alabama and elsewhere in the South where exit polls did not fail (notably in North Carolina). I am now working on a much larger project analyzing voting share down to the precinct level in large counties in every state that held primaries. Nicholas Bauer pointed this phenomenon out to me based on his precinct level analysis in New York City.  The following is a teaser; the rest will be released in a larger report in about two weeks:

The one county of the four largest by population in Alabama that does not use provably hackable voting machines shows a vote share per candidate by precinct size that matches what statisticians would expect in areas that are similar in terms of race, economics, wealth, and the like. Madison County’s data trend line is basically flat or horizontal.

Madison County Flat Line Chart

You may notice that the chart eliminates the last three precincts in Madison County. Why? Well, one of the arguments against doing this kind of analysis at the state level is that perhaps larger precincts or counties by vote total are in urban areas with a stronger concentration of people of color. The three largest precincts in Madison County skew the data a bit in precisely this manner, as do the largest precincts in big counties in states like Oklahoma and Connecticut where exit polls did not miss. The largest three precincts in Madison voted much more heavily for Clinton, and the polling locations were two large African American churches and an African American seniors center.  Here’s a more detailed graphic analysis with those three precincts included:

Madison Alabama CHART 2

Note that the data trend lines are still close to horizontal and parallel, but now favor Clinton a bit more as precinct size increases – a bit more meaning that overall the spread between Clinton and Sanders grows by about ten to eleven percent from smallest to largest precinct size grouping. This is a substantial increase, but it is entirely predictable given particular age and ethnic demographics. As I argued previously, Madison’s voting machines are not provably hackable, and the vote by precinct size model looks clean.

Now, let’s look at Jefferson County, Alabama’s largest by population: Jefferson County AL CHARTThe data trend lines (the black dotted lines) are nowhere near parallel. Clinton does massively better on average as precinct size increases. Why, it’s almost as if someone planned such a thing.

Jefferson County’s non-white population has been increasing each year over the past decade or more and stood at 49.4% last July 1 according to U.S. Census figures. Republicans sometimes win political races in Jefferson County, however. As you might imagine, political districting in Alabama is heavily gerrymandered, nearly all non-white voters are registered or lean Democratic, and nearly all white voters are registered or lean Republican. In other words, racial differences from small to large precinct size do not explain the more than 50% increase in Clinton’s win margin between the smallest precincts and the largest precincts because perhaps as many as 90% of all Democratic voters in Jefferson County on March 1 were people of color.

Cohn responded to a similar chart I posted on Twitter yesterday for East Baton Rouge, Louisiana with a combination of mocking (imagine that, Clinton doing well with black voters! … which ignores the necessary differences among black voters given this data) and more serious arguments (“racial polarization”). It’s the more serious argument that matters for the larger study. Eventually, the back and forth with Cohn led to him sneering at me to “do the math” assuming white voters were 80% Republican and 20% Democrat. That would mean less than 10% of voters in the Democratic primary on March 1 were white since white people make up just 45% of East Baton Rouge (EBR). We can be more generous: 13.7% of EBR Democrats in 2013 were white, and, assuming a non-existent surge of white independents enthusiastically registering Democrat a month ahead of Louisiana’s closed primary, let’s spot Cohn 20% white Democratic voters.

The math doesn’t work.

There simply aren’t nearly enough white voters to make that scatter plot graph make sense based on “racial polarization.”

Furthermore, Mecklenburg County (Charlotte), North Carolina is highly racially polarized with the majority of its schools 80% one race. Clinton won healthily in North Carolina by 14% and by a landslide in Mecklenburg (a 22% spread). But the exit polling was accurate in North Carolina and Mecklenburg County’s data trend lines are basically horizontal and parallel. So are Wayne County (Detroit) data lines. As well, there are key counties that are more than 80% or even 90% white where the same steady increase by precinct size shows up for Clinton. But all of that is getting ahead of the game. More in a fortnight.

Sometimes “Jesus” isn’t the answer in Sunday School; sometimes “race!” or “math!” can’t solve for irregularities in the Democratic primaries.

These statistical arguments are completely independent of, but reinforce the exit polling argument. It is quite the hat trick, actually – hackable voting machines, wildly wrong exit polls, and a Clinton vote share that smoothly increases in keeping with total votes by precinct size. It’s a hat trick now demonstrable in a dozen states or more.  At some point, the onus will shift from so-called “conspiracy theorists” to those who think the Couple Clinton to be so morally pure and upright that they’d never pay a team of hackers to laugh their way through the United States’ horribly insecure voting landscape.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s