Would it happen anyway?
A jobs grant only creates value for the jobs that would not have come without it. This screen asks how often the grant has to be the deciding reason for the wages it causes to equal the grant, and compares that with what research finds.
The deal
What it needs
On this timetable the bar is 14.0%; with every job from year one it would be 10.3%. At the 12.7% median the wages the grant causes fall a little short of it on this timetable, and would clear it only if every job came at once. The gap is small: a wage above the stated floor, real wage growth or a lower discount rate could close it, and the research median itself is for typical incentives, not one this size.
- Needed for this deal: 14.0%, or 10.3% if every job started in year one (shaded between)
- Studies with no obvious selection bias, median: 3.4%
- All studies, median: 12.7%
- Studies likely to overstate it, median: 23.5%
- Research range: 2% to 25% of incentivized firms
| If the grant decides | Wages caused, minus the grant | Grant per job it caused |
|---|---|---|
| 3.4% (no obvious selection bias) | −$18.9M | $1.5M |
| 12.7% (all studies) | −$2.3M | $394K |
| 23.5% (likely to overstate) | +$17.0M | $213K |
Award #338
The staff's $19.20, reconciled
From the stated return to what the grant plausibly caused, per grant dollar
- Staff score sheet$19.20About $480.0M of household income over 10 years (the stated ratio times the grant). The sheet says the figure is discounted ("discounted to present money value"); it names no rate and no model.
- The promised wages, counted once and discounted$7.15–$9.73$243.3M: every promised job at the promised wage for 10 years, at a 4% real discount rate, as if the grant caused all of them. So the stated figure is about 2.0 times the discounted wages, which suggests it adds the spending those wages set off; the sheet does not say.
- Only the deals the grant decides, 3.4%$0.24–$0.33Median of studies with no obvious selection bias (Bartik 2018).
- Only the deals the grant decides, 12.7%$0.91–$1.24Median across all 34 estimates.
- Only the deals the grant decides, 23.5%$1.68–$2.29Median of studies likely to overstate it.
The thin line marks $1: past it, the wages the grant causes exceed the grant. Ranges run from one reading of the agreement’s deadlines (solid) to every job from year one (light); a row crosses the line only if its range does. Wages are not a net gain: most workers would earn something in their next-best job. Even the last two rows overstate what the region gains.
How the answer moves
The same deal under other assumptions a reader might defend
| Assumptions | Must decide at least | Against the research medians |
|---|---|---|
| Direct-wage screen: every wage a gain, 10 years, 4% real (the rate on Triumph's score sheet #205), on the timing set above | 14.0% | Above both |
| Jobs counted for 20 years at a 2% rate | 5.9% | Between them |
| Plus a 1.5 regional spending multiplier | 4.0% | Between them |
| Jobs start three years after the grant is paid | 11.6% | Between them |
| Jobs from year 4, building up evenly to the full number by year 7 (the #338 agreement's timetable) | 14.0% | Above both |
| 70% of the jobs go to people who already live in the region | 20.0% | Above both |
| Wages grow 2% a year faster than prices (the growth score sheet #205 pairs with its 4% rate) | 12.7% | Between them |
| A 2% real discount rate (OMB A-94 Appendix C, 30-year real Treasury; OMB says these rates are for cost-effectiveness analysis, not benefit-cost analysis of public investment) | 11.7% | Between them |
| A 7% real discount rate (OMB Circular A-94's base case for public investment, reinstated April 2025) | 18.0% | Above both |
| Only 30% of each wage is a real gain over the next-best job | 46.6% | Above both |
| 30% real gain, 70% of jobs to current residents, jobs from year 4 | 55.0% | Above both |
Each row changes only the assumptions named and keeps the deal’s grant, jobs and wage, starting from the timing set above: first jobs after 3 years, all of them 4 years later. Rows that name a timing replace it. A multiplier counts spending the wages set off; whether it belongs in a benefit to the region, rather than a transfer from elsewhere in it, is contested. The research medians are 3.4% (no obvious selection bias) and 12.7% (all studies).
What counts
Change what the screen counts
How this works
The benefit counted is the yearly pay of the jobs, times the three shares above, discounted over the years the jobs count, from the standing of residents of the eight Triumph counties. It is a direct-wage screen, not the most the deal could be worth: spending by the new workers, tax revenue and spillovers would add to it, and wages below workers’ next-best jobs would take from it. The scales it down to the jobs the grant actually caused. The break-even is the grant divided by the wages counted if the grant caused every job.
The research figures come from Bartik, “‘But For’ Percentages for Economic Development Incentives,” Upjohn Institute Working Paper 18-289 (2018), which reviewed 34 estimates from 30 studies. It concluded that typical incentives tip between 2% and 25% of the firms that receive them. Studies with no obvious selection bias had a median of 3.4%; across all studies the median was 12.7%; studies likely to overstate the share had 23.5%. The medians are not scaled to the size of a given incentive, and a larger incentive relative to the firm’s costs should decide more deals. Page reference pending: the paper’s PDF could not be retrieved to confirm it.
Not counted: other incentives on the same deal (state programmes, local abatements, conduit financing), which would shrink the grant’s own share of the decision; tax revenue, spending by the new workers, and spillovers to other firms, which would raise the benefit; public services for new residents, which would lower it; and the next-best use of the grant money, which is assumed to be worth a dollar of regional income per dollar. Treat the result as a screen: a deal that fails here needs a strong, specific reason to believe the grant is decisive.