A model-based evaluation of the contested Massachusetts State Representative races
The competitive landscape for the contested State Representative races to be decided on November 8th range from a very likely GOP hold by incumbent Nicholas Boldyga (R-Southwick) in the Western Massachusetts Republican stronghold of the 3rd Hampden District, to a sure hold in the D+42 5th Suffolk District where Chris Worrell (D-Boston) should replace Rep. Liz Miranda who is running for State Senate.
In order to understand where the other State Rep elections fall between these extremes we have used a simple regression model trained with historical data to provide an estimated range of possible outcomes.
The Bayesian regression model has three variables: a measure of partisanship (the Partisan Voter Index), an incumbency factor (Democratic incumbent, no incumbent, or GOP incumbent), and an adjustment for presidential election years. The model is trained with 424 state legislative elections from 2008 through 2021 and explains a little over 70 percent of variance in the Democratic percentage of the vote. The other 30 percent of the variance is due to important factors like candidate and campaign quality, which will determine the results in competitive races. The general, fundamentals-based nature of the model is also why these ratings should only be used to understand the competitive landscape rather than as a prediction of race results.
The model does not adjust for third-party, independent, or unenrolled candidates, as a simple treatment of the third candidate as an additional model parameter does not provide additional explanatory value. An improved model might treat different types of third-party, independent, and unenrolled candidates differently, depending on their role with respect to the other candidate(s) (challenge from the left for Democrat, libertarian challenge to establishment Republican, etc.). This would require investigation and classification of each individual historical race, and each race being put through the model for inference purposes. We are considering adding additional features to the model that do not require this type of manual coding, including length of incumbency tenure, and an adjustment for a candidate home town effect.
The Bayesian regression model allows us to interpret the results in a probabilistic context and also provides easily interpretable confidence intervals. If you are interested in the technical details you can check out the source code and data for this post.
As an example, let’s take the 1st Middlesex race between Margaret Scarsdale (D-Pepperell), Andrew Shephard (R-Townsend), and independent candidate Catherine Lundeen of Pepperell. The Partisan Voter Index (PVI) of the redistricted 1st Middlesex District is D+2 (or 2.27 with a bit more precision) meaning that over the last two presidential elections the district is 2.27 points more Democratic than the country as a whole. The incumbency variable is set to no incumbent as former 1st Middlesex Rep. Sheila Harrington (R-Groton) left the state house to become Clerk Magistrate of the Gardner District Court. The presidential election variable is false, as it is for all the 2022, off-year races.
If we use the trained model to probabilistically simulate the election several thousand times (sampling from the posterior distribution in technical parlance) we see the Democrat winning 53% of the time, and the Republican 47% of the time. If we look at the simulated Democratic percent of the vote, rather than the win/loss percentage, we see an average Democratic percent of 51% of the vote—making this the most competitive State Rep seat with respect to this model. We can also generate a 90% onfidence interval for the voting result with the Democratic percentage varying from 40% to 62%—a wide enough interval to allow for a comfortable GOP win, or a large Democratic victory. The most likely outcome will be a reasonably narrow victory for Scarsdale or Shephard.
For each race we show the candidates, district Partisvan Voter Index (PVI), and the output of the model: a Democratic win probability, and a projected Democratic one-way margin (points above 50/50 tie) with a 90% confidence interval. It is important to remember that this fundamentals-based model does not take into account particulars of each race including important factors like candidate quality and campaign efforts and outreach, which will actually determine the results in competitive races.
The one race that needs to be interpreted differently is the 2nd Franklin race between Susannah Whipps (I-Athol) and Jeffrey Raymond (R-Athol). We are treating Whipps as an incumbent Democrat, with respect to the model, given that she has a Republican opponent. It would have made sense to treat her as a Republican incumbent if she had a Democratic opponent.
The highlighted races in the table have both candidates with at least a 10% chance of winning and non-trivial margins on both sides of the break-even margin line.
| Contested State Rep race ratings | |||||||
| district | PVI | Incumbent | D | R | other | dem_win_prob | margin |
|---|---|---|---|---|---|---|---|
| 3rd Hampden | R+7 | GOP | Russo (D-Agawam) | Boldyga (R-Southwick) | 0% | ||
| 8th Plymouth | R+1 | GOP | Haikola (D-Raynham) | D'Emilia (R-Bridgewater) | 1% | ||
| 7th Plymouth | R+1 | GOP | Sullivan (R-Abington) | Griffin (W-Whitman) | 1% | ||
| 7th Worcester | R+1 | GOP | Frost (R-Auburn) | Dotson (U-Millbury) | 1% | ||
| 5th Plymouth | EVEN | GOP | Dockter (D-Hanover) | Decoste (R-Norwell) | 2% | ||
| 22nd Middlesex | D+1 | GOP | English (D-Billerica) | Lombardo (R-Billerica) | 2% | ||
| 1st Plymouth | D+3 | GOP | Palmer (D-Plymouth) | Muratore (R-Plymouth) | 3% | ||
| 1st Bristol | D+6 | GOP | Roche (D-Mansfield) | Barrows (R-Mansfield) | 5% | ||
| 3rd Barnstable | D+6 | GOP | Alfano (D-Bourne) | Vieira (R-Falmouth) | 5% | ||
| 2nd Essex | D+9 | GOP | Kassner (D-Hamilton) | Mirra (R-Georgetown) | 8% | ||
| 11th Worcester | D+12 | GOP | Fishman (D-Shrewsbury) | Kane (R-Shrewsbury) | 14% | ||
| 1st Middlesex | D+2 | Scarsdale (D-Pepperell) | Shepherd (R-Townsend) | Lundeen (I-Pepperell) | 53% | ||
| 8th Hampden | D+4 | Arriaga (D-Chicopee) | Goonan (I-Chicopee) | 61% | |||
| 36th Middlesex | R+3 | Dem | Garry (D-Dracut) | Boag (R-Dracut) | 72% | ||
| 9th Norfolk | D+6 | Kalkut (D-Norfolk) | Vaughn (R-Wrentham) | 68% | |||
| 7th Hampden | D+6 | Saunders (D-Belchertown) | Harrington (R-Ludlow) | 71% | |||
| 8th Bristol | R+2 | Dem | Schmid (D-Westport) | Gendreau (R-Westport) | 76% | ||
| 14th Essex | D+8 | Ramos (D-North Andover) | Finn (R-North Andover) | 76% | |||
| 5th Bristol | R+1 | Dem | Haddad (D-Somerset) | Thurber (R-Somerset) | 80% | ||
| 2nd Worcester | R+1 | Dem | Zlotnik (D-Gardner) | Chester (R-Gardner) | 81% | ||
| 19th Middlesex | EVEN | Dem | Robertson (D-Wilmington) | Sarnowski (R-Wilmington) | 82% | ||
| 1st Barnstable | D+9 | Flanagan (D-Dennis) | Post (R-Yarmouth) | Kasparian (WTP-Yarmouth) | 79% | ||
| 12th Plymouth | D+1 | Dem | Lanatra (D-Kingston) | Meschino (R-Plymouth) | Mccoy (NP-Kingston) | 84% | |
| 16th Middlesex | D+11 | Elliott (D-Lowell) | Miller (R-Lowell) | 84% | |||
| 10th Bristol | D+2 | Dem | Straus (D-Mattapoisett) | Swift (R-Mattapoisett) | 88% | ||
| 6th Plymouth | D+3 | Dem | Cutler (D-Duxbury) | Sweezey (R-Hanson) | 88% | ||
| 1st Essex | D+12 | Shand (D-Newburyport) | Fitzwater (R-Salisbury) | 88% | |||
| 4th Worcester | D+5 | Dem | Higgins (D-Leominster) | Dombrowski (U-Leominster) | 93% | ||
| 3rd Bristol | D+5 | Dem | Doherty (D-Taunton) | Coute (R-Taunton) | 93% | ||
| 2nd Bristol | D+6 | Dem | Hawkins (D-Attleboro) | Escobar (R-Attleboro) | 93% | ||
| 2nd Barnstable | D+6 | Dem | Diggs (D-Barnstable) | Peters (R-Barnstable) | 93% | ||
| 2nd Franklin | D+6 | Dem1 | Whipps (I-Athol) | Raymond (R-Athol) | Mckeown (U-Gill) | 93% | |
| 12th Worcester | D+6 | Dem | Kilcoyne (D-Clinton) | Vulcano (R-Northborough) | 93% | ||
| 4th Norfolk | D+7 | Dem | Murphy (D-Weymouth) | Rotondo (R-Weymouth) | 95% | ||
| 9th Plymouth | D+8 | Dem | Cassidy (D-Brockton) | Novak (R-Brockton) | 95% | ||
| 8th Middlesex | D+17 | Arena-Derosa (D-Holliston) | Barnes (R-Millis) | 94% | |||
| 10th Norfolk | D+9 | Dem | Roy (D-Franklin) | Bailey (R-Franklin) | 96% | ||
| 3rd Worcester | D+9 | Dem | Kushmerek (D-Fitchburg) | Packard (R-Fitchburg) | 96% | ||
| 19th Worcester | D+18 | Donaghue (D-Westborough) | Hostage (R-Southborough) | 95% | |||
| 18th Essex | D+10 | Dem | Nguyen (D-Andover) | Dufour (R-Tewksbury) | 97% | ||
| 14th Middlesex | D+22 | Cataldo (D-Concord) | Cleaves (R-Chelmsford) | 97% | |||
| 31st Middlesex | D+13 | Dem | Day (D-Stoneham) | Menounos (I-Winchester) | 98% | ||
| 2nd Middlesex | D+14 | Dem | Arciero (D-Westford) | Xie (R-Westford) | 98% | ||
| 17th Worcester | D+15 | Dem | Leboeuf (D-Worcester) | Fullen (R-Worcester) | 99% | ||
| 2nd Norfolk | D+15 | Dem | Chan (D-Quincy) | Cintolo (R-Quincy) | 99% | ||
| 5th Essex | D+16 | Dem | Ferrante (D-Gloucester) | Sullivan (R-Gloucester) | 99% | ||
| 8th Norfolk | D+17 | Dem | Philips (D-Sharon) | Terban (R-Stoughton) | 99% | ||
| 11th Suffolk | D+28 | Garcia (D-Chelsea) | Taylor (R-Chelsea) | 99% | |||
| 28th Middlesex | D+20 | Dem | Mcgonagle (D-Everett) | Marchese (U-Everett) | 99% | ||
| 3rd Berkshire | D+23 | Dem | Pignatelli (D-Lenox) | Lavery (GR-Becket) | 100% | ||
| 14th Norfolk | D+25 | Dem | Peisch (D-Wellesley) | Rolde (GR-Weston) | 100% | ||
| 5th Suffolk | D+42 | Worrell (D-Boston) | Owens (I-Boston) | 100% | |||
| 1 Whipps is independent, but is treated as an incumbent Democrat for modeling this race. | |||||||