Manufacturing Economy
The State Evidence Does Not Establish That a Broader Manufacturing Mix Guarantees Growth
Across 49 states with complete 2019 industry portfolios, the correlation between initial manufacturing concentration and 2019–2024 real manufacturing growth is -0.23. Removing Texas and Washington changes it to -0.06.
Editorial evidence cutoff: September 9, 2026. Published September 29, 2026. Observation periods are stated throughout; older figures are retrospective evidence.
Diversification can sound like a self-evident growth strategy in September 2026 industrial discussions. The historical BEA panel available by September 9 offers a limited test: compare the 2019 industry mix with real manufacturing growth through 2024. This is a retrospective association with explicit sensitivity checks, not a new causal estimate or a forecast for the next five years.
Does a broader manufacturing portfolio predict stronger state manufacturing growth? The idea is plausible enough to appear in economic-development arguments, but a straightforward test produces a much less decisive result. Across forty-nine states with complete 2019 industry portfolios, the correlation between initial concentration and subsequent real manufacturing growth is negative 0.23. Remove Washington alone, and it falls to negative 0.08.
The relationship looks weak when states are compared by ranks as well: the rank correlation is negative 0.09. These results do not prove that diversification is unimportant. They show that this particular five-year growth comparison cannot bear a strong claim about its benefits. The analysis links each state’s 2019 manufacturing mix with its 2019–2024 real manufacturing value-added change, using BEA state GDP records.
THE TEST STARTS BEFORE THE GROWTH WINDOW
Concentration is measured from the shares of nineteen detailed manufacturing industries in each state’s 2019 current-dollar manufacturing value added. The shares are squared and added to produce an index: a high value means more activity is concentrated in fewer of the nineteen groups. This initial portfolio is then compared with the subsequent change in the state’s published real manufacturing value added.
Using the starting portfolio avoids describing an end-of-period mix as though it had existed before growth occurred. It does not solve the larger problem of causality. States begin with different industries, resources, markets, histories and institutions. Their starting concentration is one characteristic among many. The test asks whether there is a simple association across the observed states. It does not isolate what would happen if a state deliberately changed its industry mix.
WASHINGTON HAS A LARGE INFLUENCE ON THE LINE
The full-sample Pearson correlation is negative 0.23: higher initial concentration is associated with weaker subsequent growth in this panel. Washington’s real manufacturing value added declined 21.88% over the period, making it an influential observation. Excluding Washington alone moves the correlation to negative 0.08. Excluding Texas alone leaves it at negative 0.22. The two exclusions do not have symmetrical effects.
Removing both Texas and Washington gives negative 0.06, but that result is a post hoc sensitivity check, not a cleaner preferred estimate. Those states are real members of the population, and excluding inconvenient observations would be poor analysis. The point of the exercise is to show influence. A relationship that changes materially when one state is removed should be described with that dependence visible, rather than presented as a general law governing the other states.
CHANGING THE MEASURE DOES NOT PRODUCE ONE STABLE ANSWER
The rank correlation is negative 0.09. It asks whether states with higher concentration tend to occupy lower positions in the growth ranking, reducing the influence of the precise distances between observations. Its small magnitude contrasts with the somewhat more negative ordinary correlation. The difference suggests that unusually positioned observations matter to the fitted linear relationship.
A separate size filter moves the other way. Restricting the sample to the forty-one states with at least $5 billion of manufacturing value added in 2019 produces a correlation of negative 0.34. That is a stronger negative association, but it describes a different population. The correct response is not to select whichever coefficient best supports a preferred story. It is to disclose the alternatives and recognize that the result depends on the population and summary measure chosen.
A MISSING INDUSTRY IS NOT A ZERO-SIZED INDUSTRY
The main panel includes forty-nine states because Wyoming lacks one of the nineteen detailed sector observations in 2019. Replacing that missing value with zero would change both the denominator and the distribution of shares, potentially making its manufacturing portfolio look more concentrated than the available evidence supports. Wyoming is excluded from this matched calculation instead.
The District of Columbia is excluded throughout because the analysis is defined as a state panel. These choices leave nineteen-industry portfolios with consistent coverage for the included states. They also establish a limit: the result is about those forty-nine observed portfolios. A complete table is not an administrative detail in a concentration study. Missing values can alter the measure that the article is trying to relate to growth, creating an apparent pattern from data handling rather than economics.
GROWTH AND RESILIENCE ARE DIFFERENT OUTCOMES
The outcome measured here is cumulative real manufacturing value-added growth over five years. It does not measure volatility, the depth of a downturn, recovery speed after a common shock, or dependence on a single employer. A diversified portfolio could conceivably grow more slowly while varying less from year to year. A concentrated portfolio could grow quickly because its leading sector had favorable conditions during this window.
The nineteen-sector classification introduces another limitation. Different industries may depend on the same customers, financing conditions or upstream inputs, while one broad category may contain several distinct product markets. A low concentration score therefore does not necessarily describe independent sources of economic risk. To test resilience, the outcome and the definition of diversification would need to be designed for that question, with annual paths, shared shocks and economic links examined directly.
THE UNSTABLE RESULT IS THE FINDING
The useful conclusion is narrower than either “diversification works” or “diversification does not matter.” In this matched state panel, the simple relationship between initial industry concentration and subsequent real manufacturing growth is weak under a rank comparison and highly sensitive to Washington. A size restriction strengthens it, which adds another reason to resist a universal interpretation.
For a September policy or location argument, the result raises the evidentiary threshold. A claim about resilience needs a measure of resilience; a claim about faster growth needs a credible growth comparison. A broad historical portfolio can still be useful to a business, but this test cannot establish that it caused better outcomes or guarantees them in the future.
For a public claim that a broader manufacturing mix guarantees faster growth, that is a meaningful challenge. The claim needs a defined outcome, a defensible comparison and evidence that survives reasonable changes in the sample. The next investigation could examine different starting years, annual volatility, industry-specific shocks and the composition of the broad categories. Until that work is done, the transparent result is the appropriate one: the intuitive story is not reliably demonstrated by this straightforward test, and the observations that make it appear persuasive must remain visible to the reader.
Sources and evidence
Evidence period: 2019 industry composition; 2019–2024 real manufacturing growth. The frozen evidence record lists the source files and verified hashes available September 9, 2026. Source revision: 5e4fb7726c3d40060c0151c086c903baae856cab. Later live-data updates do not alter the historical evidence in this article.
Published 2026-09-29.