How the model gets its numbers

Why this document exists

Transparency is the point. Every figure the model reports can be traced back to a published, authoritative data source and a documented method.

This page is written for the people who rely on the estimates — economic developers, public officials, site selectors, and business leaders — rather than for specialists. Where our results differ from other widely used models, we can tell you exactly why in a single sentence. That openness lets you defend the numbers to a skeptical audience, because you can see how they were made. Nothing here requires a background in economics; technical terms are defined as they appear.

What the model does, in plain terms

When a new employer opens, an existing one expands, an event draws visitors, or households earn and spend money, the effect does not stop at that first transaction. Suppliers receive new orders, those suppliers buy from their own suppliers, and the workers involved spend their wages at local businesses. The model traces these ripples through the regional economy and reports the total effect, broken into three layers:

• Direct effects — the activity you start with: the jobs, wages, and sales at the business or event itself.
• Indirect effects — the supply chain: additional activity at the firms that supply the direct activity, and their suppliers in turn.
• Induced effects — the spending of wages: activity generated when workers in both layers spend their household income at local businesses.

For each layer, the model reports four standard measures: employment (jobs), labor income (wages, salaries, and benefits), value added (the region's contribution to GDP), and output (total sales or production) — for the state as a whole or for any individual Georgia county.

The data foundation

The model rests on the official statistical accounts of the United States — the same primary sources that other professional impact models draw upon, but used directly and visibly rather than repackaged.

Because the foundation is public, any analyst — including one checking our work for a client — can locate the underlying numbers and verify that we started from the official record.

From national patterns to Georgia

The national accounts describe the United States as a whole. A dollar spent in Georgia behaves differently from a dollar spent across the entire country, because no single region makes everything it needs. When a Georgia business buys a specialized input not produced in the state, that purchase — and its ripple — leaves the local economy. Capturing this “leakage” correctly is the central task of regionalization.

To do this, the model uses a published, peer-reviewed technique known as the Flegg Location Quotient (FLQ) method. In plain terms, it compares how concentrated each industry is in a region against its concentration nationally, adjusts for the size of the region, and estimates what share of each input a local economy can realistically supply from within. Inputs the region cannot supply locally are treated as imports and correctly excluded from the local ripple.

The FLQ method was introduced by Flegg, Webber, and Elliott (1995) and refined by Flegg and Webber (1997, 2000). It is one of the most thoroughly tested non-survey regionalization methods in the regional-science literature and consistently outperforms simpler location-quotient approaches. We apply it to every county using that county's own employment and wage structure, producing a distinct profile for each of the 159 counties. Before release, the county-level results were tuned and then locked against an independent professional benchmark so the estimates land in a defensible range rather than drifting on untested assumptions.

Analyzing an industry the region doesn't have yet

This construction has a consequence that matters enormously for economic development: the model can analyze an industry even where it does not currently exist. Models built from observed local accounts cannot — if an industry has no local presence, there is nothing in their tables to shock. But the central act of economic development is evaluating a prospect that isn't there yet.

Because this model carries the complete national technology recipe for all 402 industries, a hypothetical new entrant's input requirements are always known. The county's actual industrial mix then determines what share of those purchases can be supplied locally versus imported — the machine shops that are present capture their share of the ripple; the specialty inputs that aren't are correctly treated as imports. For the entrant's own local-purchasing behavior, the model assumes it sources the way the state-average version of that industry does, scaled to the county's size — a conservative, documented stand-in.

Three cautions apply when using this for prospecting: the entrant is modeled as behaving like an established state-average operation (a brand-new plant may import more in its first years); the model is static, so it does not capture suppliers relocating because the plant arrived; and the direct jobs-to-output relationship uses the county's own wage structure (see §05), so the prospect is evaluated at local rather than national productivity.

How an impact is calculated

Once the model has a regional picture of how industries are linked, it converts a starting event into total impact using economic multipliers. A multiplier answers a simple question: for each dollar (or job) of direct activity, how much total activity results once the supply-chain and wage-spending ripples are counted? The model reports two standard types:

• Type I multipliers count the direct effect plus the supply-chain (indirect) effect.
• Type II multipliers add the induced effect of household spending on top, giving the fuller picture of total regional impact.

The jobs figure is also county-specific. Since the August 2026 calibration, jobs per dollar of output are adjusted county by county using each county's own industry wage structure from the QCEW, on a simple, checkable principle: within an industry, output per worker moves with pay per worker, so a rural operation paying below the state average runs with more workers per dollar of output than its metro counterpart. Each county's self-employment mix — from the Census Bureau's Nonemployer Statistics — then shifts jobs toward the industries where proprietors genuinely concentrate, holding the county total to what its wage data supports. Both adjustments were validated industry by industry against county benchmark runs of a leading commercial model.

The induced effect is geographic: how much of each wage dollar returns to a county's economy depends on how much of the household shopping basket the county can actually supply, and on where its workers live — measured worker by worker from the Census Bureau's LODES commuting data. A self-contained mountain county keeps a larger share of its payroll ripple than a bedroom community whose workers take their paychecks home elsewhere, while a metro core is credited for the near-work spending of its large in-commuting workforce. Both influences are calibrated against county-level benchmark runs of a leading commercial model, and a region built from many counties smoothly recovers the state's own validated behavior as commuting is internalized.

Results can be expressed in the dollar year you need. The model includes a price-index adjustment so impacts can be stated in current-year dollars and compared cleanly against other estimates, which are often reported in a different dollar year.

The kinds of analysis the model supports

• Industry events — a new or expanding business, a construction project, or an operating facility, entered as jobs, output, or income, at a broad sector or a specific detailed industry.
• Household consumer spending — the impact of household purchases, useful for retail, tourism, and population-driven analyses.
• Household income — income that households receive and then spend, with the realistic share spent locally rather than saved applied automatically.
• Household income by tier — the same analysis across nine income brackets, because a dollar of income to a lower-income household is spent very differently from a dollar to a higher-income one — producing more accurate results for programs that affect specific income groups.

Beyond the multipliers: the added result layers

Around the core impact engine, the model adds several result layers that traditional input-output tools do not provide. Each is built from public federal data, and each carries its assumptions on its face — the relevant caveat is printed on the result card itself and in every export.

• Tax impacts — the federal, state, and local tax revenue supported by the total impact. Every rate is an effective rate: actual government collections divided by the actual tax base (BEA national accounts, IRS Statistics of Income, Census state tax collections, Census of Governments local finance). Statutory rates, exemptions, and compliance are therefore baked in rather than assumed. Personal income tax varies by household income bracket — including the negative net federal rates for lower-income households created by refundable credits. Local rates are county-specific: every local government's FY2022 tax collections (county, cities, school districts, special districts — from the Census of Governments individual-unit files) are aggregated to the county and set against the county's own economic base (BEA county GDP and earnings), so a resort county's occupancy taxes and digest show up in its local line while a low-levy county reads correspondingly lower. Small counties are shrunk toward the state average, residential property follows where a county's workers actually live (the same LODES residence data as the "where workers live" layer, with second-home digests assigned to the business side), and the county rates aggregate exactly back to Georgia's published statewide local collections. Multi-county regions blend member counties' rates; statewide analyses use the Georgia average. Tax revenue is one side of the ledger; for a true net fiscal impact — revenues and service costs on a community's actual digest, millage, and budgets — the appropriate tool is LOCI®.
• Jobs by occupation — the job impact translated into occupations with Georgia median wages and the size of the existing statewide workforce, using the U.S. Bureau of Labor Statistics' industry staffing patterns (OEWS). National staffing patterns are applied to Georgia industries — state-level staffing matrices are not published — while all wages and employment bases are Georgia's own.
• Who gets the income — the labor-income impact distributed across nine household income brackets, from each industry's actual wage distribution (OEWS percentiles, fitted industry by industry). Worker wages are mapped to household brackets directly — a single-earner approximation, noted on the card.
  
  What varies by county and what does not. The dollar total is the model's own labor-income impact and is fully county-specific. The percentage split across brackets is not: it is driven by the industry composition of the scenario, and the county model files supply per-industry output, employment, income and value-added multipliers but not per-industry supply-chain composition or wage shares. Those two vectors are statewide. Two counties running an identical scenario will therefore show the same bracket percentages and different dollar amounts. Making the mix genuinely county-specific would require county industry-by-industry transaction tables, which are available for only a handful of Georgia counties.
• Where workers live — for county and regional analyses, the share of the jobs supported that are held by residents of the selected area versus commuters, from the Census Bureau’s LODES origin–destination commuting data (LEHD LODES 8.4, 2023 reference year, JT00 all jobs). Both the main and auxiliary origin–destination files are used, so workers commuting in from Alabama, Tennessee, Florida, South Carolina and elsewhere are counted rather than dropped — material in border counties, where out-of-state residents can hold a sixth or more of local jobs. LODES covers jobs reported through state unemployment-insurance systems, so proprietors, most farm labor and uniformed military fall outside its scope.
  
  Why this layer counts jobs and not dollars. LODES observes headcounts by county of residence, not payroll. Multiplying the labor-income impact by a headcount share would assume every job pays the same regardless of where its holder lives, and that assumption is wrong in a predictable direction: in most Georgia counties, residents hold a disproportionate share of the lower-paying jobs at local worksites and commuters hold more of the better-paid ones. The dollars are therefore allocated in the next layer, where the wage level is observed alongside the geography. The two retained shares will differ — a county can hold 21% of the jobs and 16% of the payroll — and that gap is the finding, not a discrepancy.
• Where low- and high-wage workers live — the residence split above, crossed with wage level. The dollar amounts come from the model’s own industry wage distribution (the OEWS percentiles behind Who gets the income), so they are specific to the industries in the scenario. The geography of each band comes from LODES, which publishes commuting flows separately for three earnings segments: SE01 ($1,250 per month or less), SE02 ($1,251–$3,333) and SE03 (over $3,333). Annualised, those cuts fall at roughly $15,000 and $40,000, coinciding with existing boundaries in the nine-bracket household distribution — bracket 1 maps to SE01, brackets 2–3 to SE02, brackets 4–9 to SE03 — so the two sources are joined without interpolation. SE01 and SE02 are reported as a single under $40,000 column: the lowest segment is a rounding error in dollar terms, typically well under a tenth of a percent of payroll, and giving it its own column implies a precision the thresholds cannot carry.
  
  What this table can and cannot show. The LODES earnings thresholds are fixed in nominal dollars and have not been indexed since the series began in 2002 — $40,000 then bought roughly what $70,000 buys today, so a cut that was set near the median now sits below Georgia’s median wage. By the 2023 reference year the top segment holds roughly 83% of Georgia jobs. The table therefore separates low-wage from higher-wage commuting patterns reliably, but it cannot distinguish a $45,000 job from a $150,000 job by county of residence: every bracket above $40,000 is allocated using the same observed commuting pattern. Differences among places within the $40,000-and-over column are differences in dollars, not in geography. It follows that adding finer bands above $40,000 would add no geographic information at all — each would return the identical residence shares while looking more precise, which is why the table stops at two. A genuine finer split would require American Community Survey microdata (place-of-work and residence PUMAs with continuous earnings), trading observed county detail for sample-based estimates on much coarser geography; it is not used here. Read this layer as low-wage versus everything else. The income distribution proper is Who gets the income, which runs off the full OEWS wage distribution and needs no residence geography.
  
  The high-wage skew column divides a place’s share of the $40,000-and-over payroll by its share of the under-$40,000 payroll. A value of 1.00 means the place captures better-paid and low-wage income in the same proportion; above 1.00 means it captures disproportionately more of the better-paid payroll. Within a band, commuting patterns are assumed uniform across industries — LODES publishes earnings and industry as separate margins and does not cross them.
• Uncertainty ranges — 90% intervals on every headline figure, produced by re-solving the entire model 120 times with its two governing behavioral parameters drawn from their calibration uncertainty. The direct effect a user enters is exact; the modeled ripple carries the uncertainty. To our knowledge no commercial impact model publishes error bars.
• Solve for a target — the analysis run in reverse: enter the total outcome needed (jobs, output, income, value added, or state-and-local tax revenue) and the model back-solves the required direct investment. Because the model is linear, the solution is exact.
• LOCI® multipliers — the scenario's employment and labor-income multipliers (Type I and Type II), formatted for entry into a LOCI® project profile, where they convert a direct fiscal impact into a total fiscal impact on genuine community data.

How our estimates differ from a leading commercial model — and why

Many clients will, reasonably, compare our results against a widely used commercial impact model. As of the August 2026 calibration this comparison has been done systematically: 2,617 industry-level benchmark comparisons across thirteen Georgia geographies, spanning the state, metro cores, suburbs, mid-size cities, and the smallest rural counties. Output multipliers agree within a few percent almost everywhere, and total-jobs estimates land within about ±6% in three-quarters of the validation counties. The estimates are close, but not identical, and the remaining differences are deliberate and explainable. Each reflects a transparent methodological choice, not an error — and in every case the reasoning is one sentence long.

Taken together, these choices make our estimates somewhat more conservative than a commercial alternative in specific, identifiable situations — and we would rather a client be able to explain a number than be surprised by it.

Uncertainty, limits, and good practice

No economic impact estimate is a precise measurement; every one is a modeled approximation, and honesty about that is part of the method. A few principles guide responsible use:

• Estimates are most reliable for larger regions and broader industries, where the underlying data is richest.
• Estimates for a single small county combined with a single narrow industry carry the widest uncertainty, and should be read as an indicative range rather than a point figure.
• The model reflects the economy as captured in the most recent benchmark data; genuinely novel activities with no historical analog are inherently harder to model.
• Impact results describe gross activity generated and do not, by themselves, net out activity that might have occurred anyway — interpreting that is a matter of analytical judgment.

Where one of these limits applies most sharply — a small, specialized county — the tool surfaces a short caveat at the point of analysis, so the reader is reminded to treat the result as a band.

As of the 2026 update, the tool no longer merely advises this humility — it quantifies it. Every headline figure carries a 90% range, computed by re-running the full model over the calibration uncertainty of its two governing parameters (the regionalization exponent and the household spending rate). Typical industry scenarios carry roughly a ±5–6% band on output; household-spending scenarios, which lean harder on consumer behavior, run wider at roughly ±14%.

In summary

Built on the official statistical accounts of the United States, regionalized to Georgia and its counties with a published, peer-reviewed method, and calibrated against an independent professional benchmark.

The results are transparent by design: the data is public, the method is documented, and every difference from other models has a clear, one-sentence explanation. That is what allows the numbers to stand up to scrutiny — not the authority of a black box, but the visibility of the reasoning behind them.

Key sources and references

• U.S. Bureau of Economic Analysis. Benchmark Input-Output Accounts of the United States, 2017. Released 2023 with the comprehensive update of the National Economic Accounts.
• U.S. Bureau of Labor Statistics. Quarterly Census of Employment and Wages (QCEW).
• U.S. Bureau of Economic Analysis. Personal Consumption Expenditures (PCE) and price indexes.
• U.S. Bureau of Labor Statistics. Consumer Expenditure Survey (CEX).
• U.S. Census Bureau. Economic Census and NAICS industry classification.
• Flegg, A. T., Webber, C. D., & Elliott, M. V. (1995). On the appropriate use of location quotients in generating regional input-output tables. Regional Studies, 29(6), 547–561.
• Flegg, A. T., & Webber, C. D. (1997). On the appropriate use of location quotients in generating regional input-output tables: reply. Regional Studies, 31(8), 795–805.
• Flegg, A. T., & Webber, C. D. (2000). Regional size, regional specialization and the FLQ formula. Regional Studies, 34(6), 563–569.
• U.S. Bureau of Economic Analysis. National Income and Product Accounts (NIPA) and Regional Economic Accounts (SAGDP, SAINC).
• Internal Revenue Service, Statistics of Income. Historic Table 2 — individual income by state and size of adjusted gross income.
• U.S. Census Bureau. Annual Survey of State Government Tax Collections; Census of Governments — State & Local Government Finances; American Community Survey.
• U.S. Bureau of Labor Statistics. Occupational Employment and Wage Statistics (OEWS), May 2024 — national industry staffing patterns; Georgia wages and employment.
• U.S. Census Bureau. LEHD Origin-Destination Employment Statistics (LODES), 2023 — county commuting flows.