The Border Test: How Much More Persuasive Is the Estimate?

This chapter examines the Census study's preferred comparison: single-unit manufacturing firms in adjacent counties on opposite sides of a state border. The question is what this design contributes beyond the broad state-level pattern, how its controls and fixed effects work in plain language, and how to read the reported 8.4-percent relative estimate without turning it into a job effect or a universal law.

What a state border changes in the comparison

The aggregate analysis compares states that can differ in industry, demand, technology, and institutions. The border design tries to make the comparison more local. It takes firms in counties that touch across a state line and asks whether the firms on the two sides show different adoption patterns when their states set different minimum-wage rules.

The logic is not that a border makes two counties identical. It is that neighboring places may share parts of a local economic environment: transport links, regional demand, labor pools, supplier networks, climate, and broad business cycles. At the same time, a state boundary can change the wage floor that applies to a firm. This creates a more focused contrast between policy exposure and the surrounding conditions in which the firm operates.

Suppose two single-unit manufacturers make similar products in counties separated by a state line. One state's minimum wage rises more than the other's. If the firm on the higher-wage side later records an industrial-robot import, the observation fits the mechanism developed earlier: a different labor cost may change the relative attractiveness of a capital-intensive process. The nearby firm supplies a more relevant comparison than a randomly chosen firm in a distant state because many regional conditions may be shared.

But proximity is an assumption, not a guarantee. The firms may have different managers, customers, machinery, financing, or plans for expansion. One county may receive a new supplier or a large order. A state policy may change other than the minimum wage. The border design reduces some sources of difference; it does not make the policy change random or reveal every reason a firm invested.

The study also restricts the analysis to single-unit firms. That helps connect the firm's location to the state wage floor and makes the local comparison easier to interpret. A multi-plant firm could allocate investment across locations, pool labor costs, or make a corporate decision that is not tied to the minimum wage in one county. The restriction improves geographic clarity while narrowing the population to which the result applies.

flowchart LR A[County in State 1] --> B[Single-unit manufacturing firm] C[County in State 2] --> D[Single-unit manufacturing firm] A --- C E[Different state minimum wages] --> B E --> D B --> F[Firm-year robot adoption proxy] D --> G[Firm-year robot adoption proxy] H[Time, industry, border-pair, and firm controls] --> F H --> G I[Unobserved differences and bundled policies may remain] -.-> F I -.-> G