Commonwealth Report Findings Land and water For policymakers Elected Officials Updates Weekly Method Search

Commonwealth · Report

Methods and limits

Terms in plain English

The report uses a few statistical terms. Every finding is an association between numbers; none shows that one thing caused another.

  • Median: the middle value. Half the towns are above it and half below. Unlike an average, one very large city cannot pull it up or down.
  • Per resident: a town’s dollars divided by its population, so towns of different sizes can be compared.
  • Smallest fifth, largest fifth: the 351 towns ranked by population and split into five groups of about 70.
  • Times its share: what a town or group received divided by what an even split would give it. 1.0 is exactly its share, 2.0 is twice, 0.5 is half. The burden index is read the same way.
  • The model, and “more than the model expects”: a statistical model predicts each town’s money from facts about it, such as population, road miles, jobs, need and institutions. “More than the model expects” means more than towns with the same facts received.
  • Log dollars: the model compares towns by percentage differences rather than raw dollars, so a very large town does not swamp a very small one.
  • R² (R-squared): the share of the differences between towns that a model accounts for, from 0% to 100%.
  • Out of sample, cross-validated: the model is built on some towns and tested on towns it has not seen, repeated until every town has been tested. It keeps a model from looking better than it is by memorizing the data.
  • Standard deviation: a common yardstick for how spread out a set of numbers is, roughly the typical distance from the average. “Per standard deviation of exposure” compares towns one typical step apart in how much of the decade they were represented by a leader.
  • p, or p-value: how likely a pattern at least this strong would appear by chance alone if there were no real link. p < 0.001 means less than 1 chance in 1,000; p = 0.015 means about 1.5 in 100. By common convention, a value under 0.05, 1 in 20, is treated as unlikely to be chance. A small p does not say the pattern is large, and it does not show cause.
  • 95% interval: the range of values the data are consistent with. Ranges built this way contain the true value 95 times in 100.
  • Odds ratio: compares how likely something is for one group against another. 1.0 means no difference; 1.40 means the odds are 40% higher; 0.34 means about a third as high.
  • Percentage point: the plain difference between two percentages. Going from 20% to 27% is 7 percentage points, a 35% increase.
  • Rank correlation: how closely two rankings of the towns line up, from −1 to 1. 1 means the same order, 0 means no relation, and a negative number means opposite orders.
  • Controls, “at the same size,” region fixed effects: ways of comparing like with like, so that size or region cannot produce the result.
  • After inflation, real dollars: dollars adjusted so that different years buy the same amount, using consumer prices in the Boston area (CPI-U) or the federal price index for what state and local governments buy.

Data sources

Data corrections, 18–21 October 2026

Two errors were found and fixed before this report. DESE numbers 33 towns differently from the Department of Revenue, and one file had read DESE numbers as DOR codes, which put $524.8 million of regional school payments on the wrong town in 31 towns; statewide totals did not change. Separately, five fiscal years had been credited to the General Court seated in January of the fiscal year rather than the one that enacted the budget; every year now carries its enacting court, on that court’s district lines. No headline result reversed. Three within-town power results lost significance and are not reported: a House member moving into a top post, Senate Republican representation, and within-town Senate leadership tiers. The cross-town Senate stipend effect also fell back above p = 0.05 and is not reported.

A third issue surfaced on October 19. CTHRU records cherry-sheet aid as the cash paid after assessments for charter tuition, school choice and regional transit are deducted; ten-year Chapter 70 is $46.35 billion net against $57.10 billion gross. This report uses the gross DESE and cherry-sheet figures for all formula aid, as the Fairness Index does, and the town-level model was rerun with gross formula aid as its control; every power result reported here held. The old figures had also credited $9.01 billion of charter and virtual-school tuition to the town where each school sits, which put Hadley 8th in formula aid per resident; on gross formula aid to the town government and its school districts, the basis the Fairness Index uses, it is 294th.

A fourth correction was made on October 21. The first measure had counted every clause that named a town, whatever the recipient: a local organization named with the town, such as the Gloucester Marine Genomics Institute, a state facility such as the Pocasset Mental Health Center in Bourne, or a state agency program. It now counts only money to the city or town government: the government itself, its departments and schools, and municipal projects with no other recipient. Where the Comptroller’s payment records show who was paid, they decide; otherwise the clause text does, read clause by clause where it is unclear. Organizations now count on the second measure, and state facilities, named state parks and beaches and state agency programs on the third. The first measure fell from $485 million to $215 million, and the Fairness Index, the district-money results and the town-level model were recomputed. The House district-money results keep their direction. The Senate results changed: on the first measure the data would seem to indicate that district money rises with the stipend and is higher for the 12 reform-side senators, where before nothing was significant; on the wider measures neither holds.

The Fairness Index: blend and lenses

Two features of the method bear on how the scores read. First, the blended score tracks the per-resident (Equal) lens closely: across the 351 towns the rank correlation between the blend and the Equal lens is 0.98 on all aid and 0.995 on discretionary money. A town’s blended rank is close to its per-resident rank, which is why section 9 and every town page show the four lens scores separately. Second, the Effort lens divides the residential property tax by residents’ income. In practice that rate rises with median household income (rank correlation 0.24) and with school spending above the required minimum (0.37), and falls with poverty (−0.26), so a town’s Effort score reflects its spending choices and tax base as well as how heavily its residents are taxed. The lens is shown as built.

Named legislators

Legislators are named for facts of record: posts, stipends, sponsored amendments, enacted items, public roll-call votes and the money that reached their districts. Each association reported with a name is an association, not a finding of cause. The web edition carries a page for each legislator, with no composite score.

Known weaknesses

  • House dispositions are inferred by matching amendment text against the consolidated amendments; the House records most amendments only as “Consolidated A–G.”
  • Leadership base-text and conference money cannot be assigned to a member from the public record. Of the 61 enacted clauses of $950,000 or more placed in a town (second measure), 37 match no member’s amendment.
  • Earmark figures are enacted text, not payments. A held or unspent earmark counts in full.
  • Of $5.44 billion in enacted earmark clauses, 15% can be placed in a city or town counting everything placed, 13% placed in the city or town and 4% to the city or town government; the large clauses are statewide program floors.
  • Section 8’s cost figures come from the Division of Local Services: FY2026 spending is incomplete (55 of 351 towns filed), pensions are not separable from other fixed costs, and cherry-sheet detail before FY2010 could not be retrieved.
  • Section 11’s counts of local earmarks are the administration’s and the Massachusetts Taxpayers Foundation’s; they match none of the three measures.
  • Competitive grants appear at two totals. $2.73 billion is the amount awarded in the award catalog (sections 1 and 3). The Comptroller’s payments to town governments and their school districts total $2.52 billion, the figure the Summary and the Fairness Index use; payments to all recipients, including organizations other than town governments, total $3.68 billion.
  • Merit scores rest on 300-character excerpts for 56% of rows and are not used.
  • Capacity is proxied by population band; form-of-government data could not be obtained in a consistent form for all towns.
  • Matching each grant program to its burden measure is judgment; every match is listed, and a second set of plausible measures leaves the rankings close (rank correlation 0.94). Sixty-seven western towns have no digital FEMA flood map, and their resilience grants are left out of the burden index rather than guessed.
  • Aid to towns in regional school districts is an attribution by enrollment; no compiled source gives each member town’s assessment.
  • The breadcrumbs result is association. The House sample of reform voters is small and aging.
  • The Secretary of State’s lobbyist registry blocks automated access, so lobbying is not part of these tests.

Sources

Massanomics Weekly Every Tuesday

Your city or town, in the numbers that matter.

  • This week’s story, with where your city or town stands
  • Economics, housing and the state money that reaches you
  • Land, climate, history and politics
  • Side-by-side findings that show how it all connects
Free
every week
Sign up
One email a week. Unsubscribe from any issue.

Feedback. Every figure here is built from public records, and every claim in the report is checked against its source before it is published. The work has also been reviewed independently. If you believe something should be corrected or clarified, let us know. We will look into it promptly, and any correction is published with its date on the Method page. Provide feedback · Contact us