Complementarity and Aggregate Implications of Assortative Matching: A Nonparametric Analysis

Bryan S. Graham, Guido W. Imbens, Geert Ridder2009 · DOI

In many policy environments, decision-makers face a "fixed pie" problem. A school district operates with a set number of teachers of varying quality and a fixed set of classrooms with different average student abilities. A firm manages a specific pool of managers and several production units. In these scenarios, the total quantity of inputs is constrained; the central challenge is not how to acquire more, but how to match existing inputs to maximize total output.

This research by Bryan S. Graham, Guido W. Imbens, and Geert Ridder provides a rigorous, nonparametric framework for evaluating the aggregate effects of such reallocations.

The Problem: Beyond "Adding More"

Standard econometric tools often focus on the marginal effect of an input: "What happens if we increase teacher quality by one unit?" However, in the real world, inputs are often indivisible and scarce. If a "star" teacher is moved to a low-performing classroom, they must be moved out of another.

If the production technology is "additive"—meaning a high-quality teacher adds the same value regardless of the students they teach—then reallocating them is a zero-sum game for the aggregate outcome. It merely shifts points from one bucket to another.

However, if there is complementarity (e.g., high-quality teachers are significantly more effective when paired with high-ability students), then the specific matching pattern—assortative matching—can change the "size of the pie." The authors focus on measuring these "Aggregate Redistributional Effects" (AREs) when the underlying production function is unknown and potentially complex.

A Two-Step Nonparametric Approach

The authors propose a flexible estimation strategy that avoids the restrictive assumptions of linear models.

1. Estimating the Production Function

The first step involves estimating the relationship between inputs and outputs using a kernel-based approach. A significant challenge in nonparametric estimation is "boundary bias"—the tendency for estimators to perform poorly at the edges of the data distribution. To address this, the authors implement a Nearest Interior Point (NIP) estimator. This method uses Taylor expansions to correct for bias near the boundaries, ensuring that the estimated production function is reliable across the entire range of observed inputs.

2. Simulating Reallocation

Once the production function is estimated, the researchers "average" this function over a new, counterfactual distribution of inputs. To facilitate this, they introduce a family of Correlated Matching Rules. This mathematical framework allows for the evaluation of a spectrum of policies, including:

  • Status Quo: The current allocation observed in the data.
  • Random Assignment: Breaking any existing correlations between inputs.
  • Perfect Positive Assortative Matching (PAM): Pairing the highest-quality inputs together.
  • Perfect Negative Assortative Matching (NAM): Pairing high-quality inputs with low-quality counterparts.

Local Complementarity: A Practical Tool

Estimating the impact of a total system overhaul (like moving from random assignment to perfect PAM) can be statistically demanding. As a more practical alternative, the authors develop a Local Complementarity Measure. This allows a policymaker to ask: "If I make a small, incremental shift toward matching high-quality inputs together, will total output increase?" This measure is often more robust and easier to estimate than global reallocation effects.

Key Insights and Theoretical Contributions

  • The Engine of Non-separability: The justification for any reallocation policy rests on non-separability. If inputs do not interact, the aggregate outcome remains invariant to how they are shuffled.
  • The "Partial Mean" Challenge: Estimating the effect of perfect matching is technically difficult because it involves evaluating the production function along the "diagonal" of the input distribution. The authors demonstrate that while most of their estimators converge at the standard rate ($\sqrt{N}$), the extreme cases of perfect matching converge more slowly.
  • Invariance to Scale: A vital feature of their complementarity measure is its invariance to monotone transformations of the inputs. Since "teacher quality" or "student ability" are often latent constructs measured on arbitrary scales, this ensures the results are not artifacts of scaling.

Validation via Simulation

The paper uses Monte Carlo experiments to demonstrate the efficacy of these tools. By simulating data with known complementarities, the authors show:

  • Consistency: The estimators are remarkably accurate even in relatively small samples ($N=200$).
  • Robustness: The Local Complementarity Measure performs well across various data-generating processes.
  • Necessity of Correction: Ignoring boundary issues in kernel estimation leads to significant bias, validating the NIP approach.

Conclusion

This framework provides a powerful toolkit for researchers and policymakers looking to optimize the use of fixed resources. By moving beyond simple averages and embracing the complexity of how inputs interact, this research offers a path toward more efficient and effective policy design in education, labor, and beyond.

Summary updated

Abstract

This paper presents methods for evaluating the effects of reallocating an indivisible input across production units, taking into account resource constraints by keeping the marginal distribution of the input fixed. When the production technology is nonseparable, such reallocations, although leaving the marginal distribution of the reallocated input unchanged by construction, may nonetheless alter average output. Examples include reallocations of teachers across classrooms composed of students of varying mean ability. We focus on the effects of reallocating one input, while holding the assignment of another, potentially complementary, input fixed. We introduce a class of such reallocations --correlated matching rules --that includes the status quo allocation, a random allocation, and both the perfect positive and negative assortative matching allocations as special cases. We also characterize the effects of local (relative to the status quo) reallocations. For estimation we use a two-step approach. In the first step we nonparametrically estimate the production function. In the second step we average the estimated production function over the distribution of inputs induced by the new assignment rule. These methods build upon the partial mean literature, but require extensions involving boundary issues. We derive the large sample properties of our proposed estimators and assess their small sample properties via a limited set of Monte Carlo experiments.
BibTeX
@misc{graham2009complementarity,
  author = {Bryan S. Graham and Guido W. Imbens and Geert Ridder},
  title = {{Complementarity and Aggregate Implications of Assortative Matching: A Nonparametric Analysis}},
  year = {2009},
  doi = {10.3386/w14860},
}
Figure