Heterogeneity in U.S. Farms: A New Clustering by Production Potentials

Authors: Asif Rasool and David Abler (Penn State University)

What Is the Issue?

For over a century, U.S. policymakers have grappled with how to design effective agricultural policy in a system often treated as monolithic. However, U.S. agriculture is anything but uniform. From small family-run operations to large capital-intensive enterprises, the farm sector is deeply heterogeneous in land use, production capacity, labor dependence, and mechanization. Yet federal and state agricultural programs are often formulated with a "one-size-fits-all" lens, overlooking this structural diversity.

This lack of precision in policy design has growing consequences. As the agricultural sector becomes increasingly interwoven with climate, labor, and global supply chain challenges, the ability to differentiate policy interventions by farm type becomes critical. Current regional delineations, such as USDA's Farm Resource Regions, were developed using data and methodologies that no longer capture the evolving production landscape. These outdated frameworks risk masking policy-relevant differences between regions and may limit the effectiveness of economic support, conservation incentives, and rural development initiatives.

Figure 1. Counties (red dots) have a preponderance of small, labor-intensive, high-value farms. Source: Authors' analysis of USDA Census of Agriculture data (2002–2017).

What Did We Find and Why Does It Matter?

Our analysis revealed six distinct clusters of agricultural regions in the United States, each characterized by a unique combination of farm size, mechanization, labor dependence, capital investment, and production output: (1) small, low-resource farms; (2) medium-sized farms; (3) highly mechanized farms; (4) small, labor-intensive, high-value farms; (5) large farms; and (6) high-productivity farms. These clusters reflect underlying structural and resource-based differences in U.S. agriculture that are invisible in traditional regional classifications. Most of the farms in the Chesapeake Bay Watershed fall into clusters (1) and (4).

Variation among clusters is multidimensional. For instance, some counties exhibit large-scale operations in terms of land area but employ relatively few workers or machines. Others are densely packed with small, high-value, labor-intensive farms that generate disproportionately high asset values and local economic activity. This complexity challenges common assumptions that agricultural productivity can be neatly categorized as "small," "medium," or "large."

Geography alone is a poor predictor of agricultural similarity. Clusters are not defined by state boundaries or USDA regions. For example, counties in New York and California may share more similar production characteristics than neighboring counties within a single state. This spatial mismatch has serious implications for policy targeting. Programs based on outdated or overly broad regional divisions risk allocating resources inefficiently or ineffectively.

Figure 2. Counties (red dots) have a preponderance of small, low-resource farms. Source: Authors' analysis of USDA Census of Agriculture data (2002–2017)

Our findings show that meaningful agricultural policy requires a cluster-aware lens. Programs such as conservation incentives, disaster relief, rural infrastructure investment, or climate adaptation strategies will be more impactful if designed with the underlying production profile of each region in mind.

What Did We Do?

We developed a nationwide clustering of U.S. counties based on agricultural production potential. Rather than relying on state lines or legacy USDA regions, our analysis used county-level data from four recent Censuses of Agriculture (2002, 2007, 2012, 2017) to identify structurally similar farming areas using an unsupervised machine learning technique known as agglomerative hierarchical clustering, a type of AI.

We selected 15 variables that reflect five key dimensions of farm structure: land area and value, labor use, capital investment in machinery and infrastructure, farm output, and government program participation. These variables were carefully chosen to ensure broad representation of production factors, and we verified that their correlations were low, enabling us to capture distinct and independent signals from each input.

To group counties into meaningful clusters, we applied Ward's minimum variance method, which minimizes variation within clusters while maximizing differences across them. Six distinct clusters emerged from this method. Each cluster reflects a unique production strategy, shaped by natural conditions, economic development, and historical investment patterns.

Publication completed for this work

Rasool, A., & Abler, D. (2023). Heterogeneity in US farms: A new clustering by production potentials. Agriculture, 13(2), 258. https://doi.org/10.3390/agriculture13020258