Philly Calling: Who voted for Krasner in the 2017 Democratic Primary?

Longer version of this article for statisticians (Open Office format)

Larry Krasner is the most progressive candidate to run for the District Attorney position (on a major party line) in recent history. In May 2017, he won the Democratic primary with 38.3% of the vote. This beat the two released internal polls (average of 21.4%) and my prediction of 28.1%. I use statistical analysis at the division level to look at where his vote came from.


100.00 % 1686/1686 Precincts Completed.

Overall winner(s) Denoted by Winner vs Runner(s) up

Candidate Name Party Votes % Of Total Votes
KHAN, JOE DEMOCRATIC 31182 20.31 %
Write In, 14 0.01 %


Plurality Results Map - Shows who won each division.

My own map of just the Krasner vote. Dark red represents Krasner support. Click on it to expand. Uses cutoffs of 15, 20, 25, 30, 35, 45, 55, and 65%.
Created using TileMill.

Gym 2015
For a starting point, I use Helen Gym's 2015 Democratic Primary vote as a proxy for the percent of progressive Democrats in a division. Gym was the last person to win a city-wide Democratic primary from the left-wing of the city's progressive movement. I also tested Sherrie Joyce Cohen's 2015 primary result but it was less significant than Gym's. Gym's 2015 vote was the strongest factor in predicting Krasner's vote.

The next most important factor was race. There is a direct impact of race on Krasner's vote and also an indirect effect through modifying the impact of canvassing and ward leader endorsements.

There is some evidence that Krasner filled part of the niche left empty by weakness in the El-Shabazz campaign. For instance, El-Shabazz and Krasner were the only two candidates whose vote percentages were higher in majority Black divisions and lower in majority White divisions. Both Krasner and El-Shabazz had lower vote shares in Hispanic divisions.

Krasner loses 0.2% vote for each 1% White in a division. Ex. in a 50% White division, he'd lose 10% compared to his baseline support.

Krasner loses 0.1% vote for each 1% Hispanic in a division. You can see in the map above that Krasner did poorly in Kensington.

Krasner did better with young people (18-44) than older people (45 plus).

Krasner gains 0.27% for each 1% of people age 18-44 (as percent of the total population of the division). For instance, if there was a university division where everyone was 18-44 - Krasner would gain 27%.

I have canvassing data from Reclaim Philly, 215PA, BLOC Party, Neighborhood Networks, and PA Student Power. Note this didn't include canvassing from other groups including the official Krasner campaign, the ACLU, and other groups. I did a simple comparison between divisions where we canvassed and those we didn't. In divisions where we canvassed Krasner's vote increased by 10%.

Ward Leader Endorsements
Krasner got nine ward leader endorsements and they played a significant role in his vote. Each ward endorsement increased Krasner vote by 10% in the ward.

Race's impact on Canvassing
Reclaim Philadelphia is a predominantly white organization (approximately 80% based on personal observations of meetings). I followed up on research by Nico Pastore and Vanessa Baker who both observed that canvassing was stronger when the canvassers were better able to relate to the voters. I'm not sure what percent of the canvassing was done by Reclaim, but I decided to test this and found that our canvassing had a much stronger impact in areas with more Whites (or fewer Blacks).

In my model, canvassing increased Krasner vote by 20% (100% White division), 10% (50% White division), or 0% (0% White division).

If you look at it from the perspective of percent Black in a division, canvassing would increase Krasner vote by 19% (0% Black division), 8% (50% Black), or actually decrease by 2% in a 100% Black division.

This could also be due to having stronger canvassing efforts in areas that are more White. This would especially make sense for Reclaim Philly's efforts - that our strongest efforts would be in the predominantly White areas where our members live.

Ward Leader Endorsements and Race
This is strong evidence that the ward leader endorsements were more significant in Black districts. For instance, an endorsement would increase Krasner vote by 7.6% in a 50% Black division and 15.2% in a 100% Black division. If a division was 100% White, the impact of a ward leader endorsement was approximately zero. Of course you need to be careful looking as these findings as the sample size of nine ward leader endorsements and 66 wards is small.

How Many Votes did Our Canvassing Get?
I estimate that Krasner gained approximately 2700 votes from canvassing of the groups that were included in my data.

If we had canvassed all of the divisions with a similar level of effectiveness, it would have gained Krasner approximately 14600 votes.

If we had canvassed all of the divisions with canvassers that better matched the race of their division, we'd have increased Krasner support by even more votes - possibly 20,000 to 30,000.

That said, my models are likely to under-estimate the impact of canvassing due to not including the canvassing impact of the official Krasner campaign, the ACLU, and possibly other groups.

How Many Votes Did Endorsements Get?
I estimate that Krasner gained 2800 votes from ward leader endorsements. If he had ward leader endorsements in all wards, he would have gained approximately 10200 votes.

Interestingly, according to this model Krasner would have won the election comfortably without any canvassing (from our member groups) or ward leader endorsements as he won with a a 27700 vote margin. So that margin was approximately five times the size of the impact of our canvassing and endorsements.

Krasner's support can be partially explained by Gym's 2015 vote, race, canvassing, age, and endorsements.

Race probably played a much stronger role than income in the DA race. This appears to be a general fact in US society. However in this analysis there is some uncertainity due to my race data being far more accurate than my income data (due to the limits of Census data).

Not only do candidates receive support based on their race, but the impact of canvassing depends on the race of the canvassers and voters, and the impact of endorsements depends on the race of the voters (and probably endorsers too - but I didn't test that). This canvassing result is perhaps the most interesting finding as it has lots of implications for both canvassing and organizing. Predominantly white organizations, like Reclaim Philly, need to decide whether they should transform themselves (even though it is likely to slow organization growth) so that they are able to do city-wide campaigns/organizing, or to focus on canvassing/organizing their base (and form alliances with multi-racial and people of color organizations).

If ward leader endorsements are more effective in Black wards, then if Philadelphia continues to become more White - both ward leaders, committee people, and the rest of the Democratic party establishment will slowly lose their influence.

If Reclaim Philadelphia and allies want to take over (or reform) the Democratic party, it might be possible to do so with a strategy of getting ward leader endorsements, reducing the power of ward leaders who do not support our values, increasing turnout, and canvassing. However, my analysis shows that these factors by themselves do not explain the Krasner victory.

Unfortunately it is impossible to measure the impact of other factors. Notably Krasner did very well with Google Search Trends (He got 39% of the searches in the 30 days before the primary) and Facebook likes (47% share). These two variables indicate above-average supporter enthusiasm for the candidate. He also had the most money and did well with endorsements from people who weren't ward leaders. It is hard to measure the impact of organizational endorsements without knowing how much resources they put into their endorsement. But as these are city-wide factors, they cannot be included in a ward division model.

If anyone has suggestions or corrections, please email me so I can update this article!