Overview

This chapter reports what the 2018 Annual Business Survey, with data reference year 2017, shows for 395,000 metro and 71,000 nonmetro weighted firms in the selected single-unit, nonfarm sample classified by RUCC 1–3 and 4–9. Metro firms reported higher binary rates for new-to-market, process, and marketing innovation: 0.090 versus 0.080, 0.166 versus 0.145, and 0.227 versus 0.193. Separate logistic models and an Oaxaca–Blinder decomposition associate much of the differences with observed endowments; the explained component exceeds the net gap for the first two outcomes and is 87% of the marketing difference. These rounded relationships are not causal effects, and the working-paper status, self-reports, reference period, and single-unit sample limit generalization.

Learning outcomes

Metro rates exceeded nonmetro rates by 1.0, 2.1, and 3.4 percentage points for new-to-market, process, and marketing innovation, respectively.
The model’s explained component exceeded the net gap for new-to-market and process innovation and reached 87% for marketing innovation.
The rates are binary self-reports from a selected single-unit sample, and the decomposition does not prove a policy effect.

Keywords

Annual Business SurveyBinary outcomesMetro firmsNonmetro firmsLogistic regressionOaxaca–Blinder decomposition

The Firm-Level Gap: What the ABS Data Show

The U.S. rural–urban innovation gap in the Annual Business Survey is a difference in reported firm activity, not a score assigned to places once and for all. This chapter examines the study’s sample, the three outcome rates, and the Oaxaca–Blinder decomposition used to investigate them. The reader’s task is to hold two judgments together: the data show higher average innovation measures for metro firms, but they also suggest that unequal observed endowments account for much of that difference without proving that a particular rural condition or policy caused it.

The distinction is important because empirical precision can create a false sense of certainty. A decimal in a table looks exact even when it describes a binary self-report from a selected sample. A label such as “explained” sounds causal even when it means that measured characteristics are distributed differently between two groups. Reading the results well requires keeping the denominator, comparison, model, and boundary attached to every number.

What the ABS comparison actually contains

The study uses the 2018 Annual Business Survey, with a data reference year of 2017. The ABS covers nonfarm employer businesses that file specified federal tax forms and includes information about business and owner characteristics. For this analysis, the authors restrict the sample to single-unit firms. They classify the county of each firm as metro when its Rural–Urban Continuum Code is 1–3 and nonmetro when it is 4–9.

The restriction is not a minor technical footnote. A multi-unit company may report one headquarters or establishment while carrying out innovation elsewhere. Its local context might include several labor markets, supplier networks, and facilities. The survey does not identify which establishment performed a particular innovation. If the study assigned the whole company’s activity to one location, it could attribute a firm-wide outcome to a county that did not supply the relevant conditions. Single-unit firms provide a cleaner connection between the reported business and the county around it, but they also mean the results should not be generalized automatically to every kind of rural enterprise.

The weighted summary table reports 395,000 firms in metro counties and 71,000 in nonmetro counties. Those figures describe the analyzed sample as presented in the table, with ABS survey weights applied. They do not mean that each rural county contains a comparable number of firms or that the displayed county characteristics describe a typical county. The authors specifically caution that county measures calculated from firm-level observations may not represent the true county-level distribution.

The outcomes are the three forms of activity defined earlier: new-to-market, process, and marketing innovation. Each is represented by a binary indicator. A mean of 0.090 for metro firms means that the indicator is one for about 9.0% of the weighted metro observations; it does not mean that those firms produced 0.090 units of innovation. The same interpretation applies to the nonmetro value. This is why the rates provide an empirical anchor but not a measure of innovation intensity or quality.

Published by DeepMox

FAQ

What were the metro and nonmetro innovation rates in the ABS?

In the weighted sample, metro versus nonmetro means were 0.090 versus 0.080 for new-to-market, 0.166 versus 0.145 for process, and 0.227 versus 0.193 for marketing innovation.

What does it mean that the model explained 87% of the marketing gap?

It means the selected model’s explained component accounted for 87% of the observed marketing difference in its decomposition. It does not mean a policy would produce that reduction.

Why can an explained component exceed the total gap?

Because factors can contribute in opposite directions, so their summed associated contributions can exceed the net mean difference. The result is a decomposition, not a set of independent effects.

Previous11 minU.S. Rural–Urban Innovation Gap: From Geography to Endowments

Next13 minU.S. Rural–Urban Innovation Gap: Digital Access and R&D

U.S. Rural–Urban Innovation Gap: What the ABS Shows

The Firm-Level Gap: What the ABS Data Show

What the ABS comparison actually contains

FAQ