Capital-Labor Substitution and Economic Efficiency

K. J. Arrow, H. B. Chenery, B. S. Minhas, R. M. SolowThe Review of Economics and Statistics 1961 · DOI

This 1961 classic, authored by a "dream team" of economists including Nobel laureates Kenneth Arrow and Robert Solow, fundamentally changed how we model the engine of the economy. It introduced the Constant Elasticity of Substitution (CES) production function, providing a flexible mathematical bridge between the rigid models of the past and the complex reality of industrial production.

TL;DR

The paper introduces the CES production function, a versatile model that allows the ease of substituting capital for labor to vary by industry. By analyzing international data, the authors prove that the traditional assumption—that capital and labor are easily interchangeable—is often wrong, with major implications for global trade and income distribution.

What Problem Does This Paper Address?

In the mid-20th century, economists generally relied on two extremes to describe how inputs (capital and labor) create output:

  1. The Cobb-Douglas Model: Assumed a "unitary" elasticity of substitution ($\sigma = 1$). In plain English: if labor becomes 1% more expensive, firms perfectly offset this by using 1% more capital.
  2. The Leontief (Fixed Proportions) Model: Assumed zero substitution ($\sigma = 0$). Think of a bicycle: you need exactly two wheels and one frame; adding more wheels doesn't help if you don't have more frames.

The problem? Real-world data didn't fit these "all or nothing" buckets. The authors sought a more general mathematical framework that could account for varying degrees of substitutability across different industries and countries.

Methods

The authors began with an empirical observation: in almost every industry, labor productivity (output per worker) is highly correlated with the wage rate.

They performed a regression analysis on cross-country data to find the relationship between the value added per unit of labor ($V/L$) and the wage rate ($w$). They discovered that this relationship is linear in logarithms: $$\log(V/L) = \log(a) + b \log(w)$$ Here, the coefficient $b$ represents the elasticity of substitution. From this empirical "law," they mathematically derived a new production function (the CES function) that allows $b$ to be any constant value, not just 0 or 1.

Datasets

The study utilized a robust (for the time) cross-country dataset consisting of:

  • 19 Countries: Ranging from highly developed economies like the U.S. and Canada to developing nations like India, Iraq, and El Salvador.
  • 24 Manufacturing Industries: Including textiles, iron and steel, and dairy products.
  • Timeframe: Census data primarily from 1949 to 1955.

The list of countries and their corresponding wage data can be seen in the table below:

Table 1: Countries in the Sample

Key Insights and Ideas

  • The "Aha" Moment on Productivity: High labor productivity in wealthy countries isn't just about better "tech"; it's a rational response to high wages. When labor is expensive, firms are forced to substitute it with more capital, which naturally raises the output per remaining worker.
  • $\sigma$ is rarely 1: The authors found that for most manufacturing industries, the elasticity of substitution is significantly less than 1. This means capital and labor are "stickier" than the Cobb-Douglas model suggests.
  • Factor-Intensity Reversals: This is a crucial blow to classical trade theory. If two countries have different elasticities of substitution, a product that is "labor-intensive" in a low-wage country (like Japan in 1953) might become "capital-intensive" in a high-wage country (like the U.S.).
  • Separating Efficiency from Distribution: The CES model introduces distinct parameters for efficiency (how much you get out of your inputs) and distribution (how the income is shared between capital and labor).

Key Results

The authors’ regression analysis (summarized in Figure 3) showed remarkably high correlations ($R^2$ often exceeding 0.9) between wages and productivity.

Table 2: Results of Regression Analysis

Quantitatively, they found:

  • In 14 out of 24 industries, the elasticity of substitution ($b$) was significantly different from 1.
  • The median elasticity for manufacturing was roughly 0.8 to 0.9, suggesting that capital and labor are less substitutable than previously assumed.
  • They successfully used the model to explain why the "Leontief Paradox" (the observation that the U.S. exported labor-intensive goods despite being capital-rich) might exist due to these varying substitution rates.

Limitations and Future Work

  • Data Constraints: The authors acknowledge that 1950s international census data was "fragmentary" and often lacked reliable measures of capital stock.
  • Neutrality Assumption: The model assumes that technological progress improves the efficiency of capital and labor at the same rate ("Hicks-neutral"), which may not be true in practice.
  • Constant Returns to Scale: The derivation assumes that doubling all inputs exactly doubles output, ignoring potential economies of scale.
  • Future Directions: The authors suggest that future research should focus on "disaggregated" data—looking at specific machine types or labor skills—to see if the CES properties hold at a more granular level.

Summary updated

BibTeX
@article{arrow1961capital,
  author = {K. J. Arrow and H. B. Chenery and B. S. Minhas and R. M. Solow},
  title = {{Capital-Labor Substitution and Economic Efficiency}},
  journal = {The Review of Economics and Statistics},
  year = {1961},
  doi = {10.2307/1927286},
}
Figure