Announcing the new Mass. Numbers

 The blog will now include code and data, along with analysis
by BRENT BENSON

Please check out the announcement post for the re-platforming of the Mass. Numbers blog:

https://www.massnumbers.us/posts/welcome/

We have the goal of eventually bringing over these blogspot-based posts, but first the plan is to publish some detailed posts on the upcoming September primaries and November general elections.

A look at Biden's 2020 performance in Massachusetts compared to Clinton in 2016

Biden did better overall with some evidence of underperformance in municipalities with large numbers of hispanic voters

by BRENT BENSON

While we have yet to get the final results of the 2020 presidential election in Massachusetts, unofficial results show that President-elect Joe Biden outperformed Secretary Hillary Clinton in the Commonwealth, and that this better performance was exhibited in most cities and towns. Four of the municipalities where Biden did significantly worse than Clinton are distinctive in terms of having the highest percentage of hispanic residents, giving some evidence to what seems to be a nationwide phenomenon of Biden doing worse than Clinton with hispanic voters.

This graphic shows the actual relationship between Biden'20 and Clinton'16 two-party margin as a solid blue linear regression line. Compare this to the dashed black line which represents a hypothetical identical margin in 2016 and 2020 and you can see the overall improvement in Biden's margins. 

Municipalities below the dashed black line are cities and towns where Biden underperformed Secretary Clinton and the vertical distance below the line is the magnitude of the margin difference.

The four cities of Lawrence, Chelsea, Holyoke, and Springfield have significantly worse numbers for Biden, with Lawrence giving Clinton a margin of 70 points and Biden a margin of 49—a pretty amazing 21 point swing. It is also notable that these four cities have the highest percentage of hispanic residents in Massachusetts according to 2017 American Community Survey of the U.S. Census Bureau.

We need data at a finer granularity to see how significant the relationship is between Biden/Clinton margin change and percentage of hispanic voters. In particular, it is important to compare areas of similar population (as opposed to municipalities that go from towns with 100s of voters to cities with 100s of thousands of voters) and to control for other demographic differences.

It is also worth noting that while we are focusing on changes in the margin, all of these mentioned margins remain incredibly blue/Democratic. 

Here are the Clinton to Biden margin changes for all 351 municipalities in the Commonwealth.

Comparing the latest Markey/Kennedy US Senate polls

Markey leads by an average of 10 points with few positives for Kennedy
by BRENT BENSON

Three recent polls of the Massachusetts race for U.S. Senate between Senator Ed Markey and Congressman Joe Kennedy III all have positive news for the incumbent, Markey. While Kennedy leads with voters with no college degree, non-white, low-income, moderate, and conservative voters, Markey leads in the larger groups of likely Democratic voters—college graduates, whites, mid- and high-income, and liberal voters—giving him a relatively large lead of 10 points, considering all three polls.

The polls were conducted by Suffolk University Political Research Center, UMass Lowell Center for Public Opinion Polls, and Data for Progress between August 13 and August 25.  The Suffolk poll was conducted with live phone interviews. The UMass Lowell poll was conducted with an online panel matched to sample demographics, and the Data for Progress poll used text-to-web and online modes.

The sample sizes varied from 500 to 800 respondents and all numbers in this article are the leaned voter preference, meaning that voters who were unsure were asked again to make a choice for whom they would likely vote.

Averaging all of the interviews together puts the Markey margin at 10 points.

Comparing the available cross-tabs for the polls gives some useful information and also raises some questions.

The polls are not in agreement on male/female voter preference. The Suffolk poll shows males preferring Markey at a higher margin than female voters, 21 to 1, while the UMass Lowell poll shows an almost opposite finding with male/female Markey margins of 5 to 18. Data for Progress shows similar voter preferences for men and women with margins of 10 and 6 for Markey.

The polls were also not in agreement on age-related candidate preference with the Suffolk poll showing relatively little difference between the preferences of respondents over and under 45 years old (8/10 for Markey), while UMass Lowell and Data for Progress polls showed Kennedy with much higher support with those under 45 (with 16 and 15 point margins for Kennedy) and Markey with much higher support among those 46 and up with margins of 27 and 37 points.

All of the polls agreed that Kennedy is more popular with respondents without a college degree, while Markey is more popular with college degree-holding voters.

Data for Progress did not provide cross-tab breakdowns for our other areas of consideration, but the Suffolk and UMass Lowell polls agree that Markey is more popular with self-identified liberals, while Kennedy does better with self-identified moderates and conservatives. Both polls also point toward a preference for voters with incomes of less than $50,000 being more favorable to Kennedy, while mid- and high-income voters are better for Markey.

Suffolk also included information on region and urbanicity, which show Markey doing better in rural areas and the western and northeastern parts of Massachusetts, while Kennedy keeps it about even in the Boston/Suffolk region, and out-performs Markey in the Southeast and Cape/Islands region. Most of Kennedy's congressional district is in the Southeast region.

Early Voting 

Suffolk provided a breakdown of how many respondents had already voted (21%) and the preference of those voters—62% for Markey and 32% for Kennedy—with margin of +20 for Markey. 

While polling primary elections is very difficult, given problems with identifying likely voters exacerbated by the uncertainty introduced by the COVID-19 pandemic, these polls are good news for Markey and bad news for Kennedy. Even if there is a large polling error and/or an ongoing shift and movement toward Kennedy underway, there is already some locked in early vote which looks like it was positive for Markey. That being said, the race has already shifted multiple times and a Kennedy win is not out of the question.

Partisanship, education, and income were significant factors in MA presidential primary vote share

by BRENT BENSON

The big story out of the Super Tuesday Presidential primaries was the unprecedented Biden surge that put the former Vice President in the strongest position to win the 2020 Democratic nomination. The surge was so strong that it pushed Joe Biden past both Vermont Senator Bernie Sanders and Massachusetts Senator Elizabeth Warren in Massachusetts. Warren and Sanders had been vying for first place in the pre-South Carolina polls, with Biden running a distant third.

I was interested in looking a little deeper at the relative performance of the three major candidates in Massachusetts among different demographic groups to see which types of voters voted for each candidate.

I combined the unofficial city and town election results as reported by the AP and the Boston Globe and joined it together with American Community Survey Census data and results from previous presidential elections to look for relationships.

The city/town comparison is not ideal, in that we lose a lot of interesting variation in the large cities which are aggregated into a single number for each variable, but there was still enough variation to support interesting findings.

The three most significant factors turned out to be the extent of Democratic/Republican partisanship in each municipality as measured by Partisan Voter Index (PVI), the percentage of people with a college degree, and the percentage of people below the poverty line.

Partisanship

Elizabeth Warren had a much higher percentage of votes in more liberal municipalities and Joe Biden had a higher percentage of votes in more conservative cities and towns. Bernie Sanders did marginally better in more liberal areas, but the partisanship score did not explain a significant difference in his performance.

The partisanship score I used was the Partisan Voter Index (PVI) which is an average of the Democratic vs. Republican vote share in the last two presidential elections as compared to the United States as a whole.

Education

Bernie Sanders performed significantly better in cities and towns with a lower percentage of college educated voters, while Elizabeth Warren performed much better in municipalities with a higher percentage of college voters. Education did not make a significant difference in Joe Biden's vote percentage.

Income

The income level in a city or town made a measurable difference to Sanders's and Biden's electoral performance, while not factoring a great deal in Warren's vote share. Bernie Sanders performed better in areas with a larger percentage of people below the poverty line, while Biden performed better in more affluent municipalities.

Which Massachusetts cities and towns are getting more Republican or Democratic?

When comparing presidential margins, it comes down to Trumpiness
by BRENT BENSON

We often think of the partisan lean of a district, town, or city as a fairly static characteristic. One of the common metrics to use for measuring partisan lean of a region is the Partisan Voter Index or PVI which looks at the Democratic/Republican margin averaged over the last two presidential cycles, compared to that of the country as a whole.

But partisan lean is not a static characteristic. Voters in a particular area sometimes vote for a Democratic candidate in one presidential election and a Republican in the next. Voters also move in and out of districts, cities, and towns changing the voting behaviors.

This article looks at the partisan shifts in Massachusetts cities and towns by comparing the Democratic margin in the 2008 Obama vs. McCain election to the margin in the 2016 Clinton vs. Trump election as a way of judging which cities and towns have gotten more Democratic, and which have gotten more Republican. I also look at what demographics might explain these shifts.

Here is a map of Massachusetts cities and towns with colors indicating the partisan voting shifts between 2008 and 2016 presidential elections.

MA presidential vote shifts from 2008 to 2016 (click for interactive map)


You can see some geographic clustering, but are there other demographic factors that help to explain the shifts?

A regression analysis points to two demographic variables that were responsible for 73% of the variation in the partisan voter shift—they were the percentage of residents that did not attend college, and the percent of non-hispanic white residents. (The demographic data was from the 2016 American Community Survey from the US Census Bureau.)



The Massachusetts municipalities that shifted the most towards the Democratic side between 2008 and 2016 are primarily in the affluent Metrowest suburbs of Boston, but also include North Shore towns like Hamilton an Manchester-by-the-Sea and the South Shore towns of Cohasset and Hingham. The two towns with the biggest Democratic shifts were Sherborn and Dover.



The cities and towns with the biggest Republican shifts are primarily in West Central and Western Massachusetts although there is also a cluster on the South Coast including Somerset, Fall River, Westport, Acushnet, and Fairhaven. The town with the largest shift from Obama towards trump is the small town of Monroe on the Vermont border.



The importance of the race variable in the regression is shown by this table sorted by college educational attainment. While most theses cities and towns shifted Republican, the 8 that shifted the Democratic direction show high percentages of minorities. A regression model with only the education component only predicts 60% of variation in the partisan shift, rather than the 73% of the two variable model.

The education component identified by the regression model is consistent with other studies that have been done on shifts from Obama to Trump, notably this Nate Silver Fivethirtyeight piece: Education, not income, predicted who would vote for Trump, although the Massachusetts model benefits more by using the race variable.

The research behind this post was used by Adam Reilly for a WGBH All Things Considered piece.

Here is the full list table of partisan shift for each of the 351 cities and towns of Massachusetts. You can click on the table to open a searchable Google Sheet with the data.

An overview of Massachusetts 3rd Congressional district and primary candidates

Results of the September 4 primary will likely determine the race
by BRENT BENSON

On August 9, 2017 Niki Tsongas shocked the Massachusetts political world by announcing she would not seek another term in the U.S. House for the Massachusetts 3rd District. Tsongas had represented the district after winning a special election to replace Marty Meehan in 2007.

The retirement announcement set of a flurry of exploratory activity and as many as 19 candidates have been in and out of the race. According to the Secretary of the Commonwealth, there are 10 Democratic candidates and one Republican candidate who will be on the primary ballot on September 4, 2018. Based on the partisan lean of the district and the current political climate, it is very likely that the winner of the Democratic primary will go on to succeed Tsongas in the House.

The District

The Massachusetts 3rd Congressional District extends from Haverhill, Methuen, Lawrence, Andover, and Lawrence in the Merrimack Valley, across to Ashburnham, Gardner, and Winchendon in North-Central Mass., and down to Concord, Subury, and Marlborough in Boston's Metrowest region.

The 3rd District is the third-most conservative congressional district in Massachusetts with a PVI of D+9, meaning that it is 9 points more Democratic than the rest of the country, measured by an average of the last two presidential elections.

MA 3rd City/Town PVI (click for interactive map)


There is a wide variation in the partisan lean of the cities and towns in the 3rd District.

The city of Lawrence is the most Democratic in the district at D+33 putting it just above the 97th percentile for Massachusetts cities and towns. Townsend and Ashby are two of the most Republican towns in the Commonwealth at R+7, around 95th percentile in the other direction. The municipality with the district's median PVI is Westford at D+5.

The Republican-leaning towns are primarily distributed along the New Hampshire border and then creeping down into North Central Mass. in Lunenburg, Lancaster, and Winchendon.



There is also a wide variation in population and number of active voters in the cities and towns of the district. I am using the number of voters from each of the municipalities in the 2014 mid-term election as an estimate of the number of 2018 voters.

MA 3rd City/Town 2014 votes (click for interactive map)





Of the top 9 vote-producing municipalities, only Marlborough is not in the northeastern region of the district.

In the following map, the 10 municipalities in the northeast area colored in orange accounted for 51% of the district votes in 2014, while the remaining 27 municipalities in yellow accounted for 49%.

MA 3rd City/Town vote distribution



MA 3rd City/Town median family income (click for interactive map)


In terms of demographics, the Massachusetts 3rd district is home to Sudbury, one the Commonwealth's richest towns with a median family income of $186,507. It is also home to Massachusetts's poorest city in Lawrence, with a median family income of $36,073. The town of Pepperell has the median median-family-income for the district at $94,625.

The richest towns are clustered in the Metrowest area of the district, but extend up through Groton. Andover is the only upper-income municipality in Essex County. The smaller western towns and Fitchburg are on the lower end of the median family income spectrum.





The Candidates

There are 10 Democrats and one Republican on the September 4 primary ballot.



That table is sorted by the amount of contributions to the campaign, a simple but not always accurate judge of candidate viability. I have separated out candidate campaign loans which can inflate total receipt numbers for candidates like Beej Das who have loaned their campaigns much more than they have raised.

It is unlikely that a candidate that raises zero dollars in a very competitive primary like this one is putting in the necessary effort to earn more than a token amount of votes, especially given the resources of those who are raising over a million dollars to reach out through a paid field staff and advertising.

There have been two public polls of the race. The first showed Rufus Gifford in the lead, but with whopping 59% not having a preference. In the second poll, the no preference was down to 29%—still quite high. It showed Dan Koh on top with 19%, followed by Gifford and Barbara L'Italien, both with 13%.



The front-running candidates have been hammering the airwaves with TV ads, inundating mail boxes with mailers, and knocking doors. It is pretty easy to identify the top five finishers, but very difficult to predict their finishing order.

Over-performance in Fitchburg pushed Tran over the top in Worcester and Middlesex special election

Zephir slightly over-performed compared to a modeled result in most other towns
by BRENT BENSON

Republican Fitchburg City Councilor Dean Tran defeated three other opponents, including Democratic Leominster City Councilor Sue Chalifoux Zephir, to replace State Senator Jen Flanagan in the Worcester and Middlesex senate district by a margin of 607 votes (unofficial results).

Tran's victory was mainly due to a staggering 19 point victory in his home city of Fitchburg, which has a strong lean towards Democrats in most elections. Fitchburg has a Partisan Voter Index (PVI) of D+9. Our state legislative regression model predicted a Democratic vote share of 56% for an open seat in a non-presidential year.



Sue Chalifoux Zephir out-performed the model in most of the other municipalities, including an 11 point over-performance in the much-smaller town of Sterling, and kept it reasonably close in the others.

Zephir would have stood a better chance if she could have built a home field advantage in Leominster, similar to Tran's in Fitchburg, but fellow City Councilor Claire Freda came within range of Tran and Zephir with over a thousand votes, by far Freda's strongest showing in any municipality. It seems that Freda's success in Leominster prevented Zephir from running up the numbers in the largest vote-producing city in the district.

Senator-elect Tran will serve out the rest of the current term and will be up for re-election next November, 2018. He will have the advantage of incumbency, but will face a much larger Democratically motivated electorate in a highly-charged mid-term election.