Hiring here instead of placing there

What the data shows

Three series, all from budget columns or from the town's own annual reports.

Placements, 1 March each year, from the Special Services report in each annual town report, sourced to SIMS Report 7 (sources/data/placement-counts.csv):

In-district special education, budget to budget: $4,865,038 (FY25) → $4,964,329 (FY26) → $5,442,383 (FY27 level service) → $5,466,201 (FY27 Balanced).

Out-of-district tuition, budget to budget: $1,164,824 (FY25) → $1,291,293 (FY26) → $700,142 (FY27), a fall of $591,151 in one year.

So placements fell by two thirds over a decade while in-district spending rose. Both statements are facts about published numbers. Neither is a fact about cause.

What the data does not show

It cannot establish that in-district hiring caused placements to fall. At least four explanations fit the same series:

1. In-district capacity absorbed children who would otherwise have been placed — the district's argument.
2. Fewer children needed placement. Cohorts differ. A count measured on 1 March is a snapshot of who is placed, not a measure of who needed placing.
3. Children aged out. A residential placement ending because a student turned 22 looks identical in this series to one ending because the district built capacity.
4. The classification or the reporting changed. The categories are not stable across the run: FY2011 and FY2012 report collaborative placements as a subset of day placements, and every year from FY2014 reports them as a parallel category. FY2021 has a total and no parts.

And the tuition line cannot be read as the cost of placements, for two reasons that are separate and both fatal.

A budget line is net — what the town must raise after everything else paying for the thing is subtracted. Out-of-district tuition is offset by circuit breaker reimbursement from the state, so the line can fall because reimbursement rose, with no change in what any placement costs or in how many there are. build_circuit_breaker.py carries that series; the two have never been reconciled against each other.

And the line has no direction to measure. Eleven budgets from FY17 to FY27 range from $489,918 to $1,291,293, six years up and four down, with a straight-line fit of R² = 0.10. The model holds it flat for exactly this reason — a compound rate off it runs from -45.78% to +11.78% depending only on which year you start. So the FY27 fall of $591,151 is a level change, and a level change cannot be evidence about a trend.

Dollars are not children. The count series and the dollar series measure different things, and dividing one by the other would produce an average cost per placement that no document states and that would be wrong in a specific way: it would mix collaborative, day and residential placements, which differ in cost by a large multiple, and it would divide a net figure by a headcount.

What would be needed to model it

The model would have to answer: if the district employs one more person of a given kind, how many placements of a given kind does it avoid, and what does that save? That decomposes into five quantities, of which this project currently holds none.

1. Cost per placement, by type, gross. Collaborative, day and residential differ by a large multiple and the district budgets them as one line. Gross rather than net, so circuit breaker reimbursement is visible separately rather than silently inside it.
2. Which placements were avoidable, and at what staffing. This is the load-bearing one and it is the hardest: it is a judgment made child by child under an individual plan. No aggregate published anywhere implies it.
3. The marginal relationship between staff and capacity. How many additional in-district staff, of which kinds, create room for one additional child of a given need profile. The district may hold a working view of this; nothing published states one.
4. The time profile of each side. A placement avoided saves for as long as that child would have been placed — potentially many years. A position hired is a recurring cost forever, and grows at the salary rate. Those are different shapes and a single-year comparison of the two would favour whichever is measured over the shorter window.
5. Which fund pays. In-district staff may be paid by grants, circuit breaker, or the general fund; placements draw on circuit breaker. A saving that moves cost from one fund to another is not a saving to the town, and the budget cannot see the difference. This is the standing question in CLAUDE.md and it constrains this analysis as it constrains the special education escalator.

What would settle it. The two documents most likely to move this are DESE's End of Year Financial Report, which separates spending by fund, and the district's own placement-level cost detail — what each placement is billed at, by type, gross of reimbursement. Neither is published. Both are askable.

Absent those, the honest sentence is that the town cannot currently tell whether hiring in district saves money on placements, and that nobody in the public record has shown that it does or that it does not.

Registered as gaps

Per rule 7c, the limits above are registered in sources/data/money-gaps.csv rather than living only here:

• What does an out-of-district placement cost, by type? — closes with the district's placement-level billing detail, gross of circuit breaker.
• How much in-district staffing avoids one placement? — closes with the district's own basis for the ROI claim, if one exists in writing.

Sources