Scale versus scope in the diffusion of new technology: evidence from the farm tractor

• DOI: http://doi.org/10.1111/1756-2171.12230
• Published date: 01 June 2018
• Author: Daniel P. Gross
• Date: 01 June 2018

RAND Journal of Economics

Vol.49, No. 2, Summer 2018

pp. 427–452

Scale versus scope in the diffusion of new

technology: evidence from the farm tractor

Daniel P. Gross∗

Although tractors are nowused in nearly every agricultural field operation and in the production

of nearly all crops, they first developed with much more limited application. Early diffusion was

accordingly rapid in these narrower applications but limited in scope until tractor technology

generalized. The sequence of diffusion is consistent with a model of Research and Development

(R&D) in specific- versus general-purpose attributes and with other historical examples, sug-

gesting that the key to understanding technology diffusion lies not only in explaining the number

of different users, but also in explaining the number of different uses.

1. Introduction

Technology diffusion is widely viewed as a leading explanation for productivity growth

and productivity differences across industries, firms, and geographic regions. For example, it is

frequently argued that facilitating the diffusion of modern production technologies to manufac-

turing and agriculture in developing countries is a key to lifting incomes and breaking a cycle

of poverty. More generally, diffusion is typically viewed as the fastest path to the technology

frontier. Research on technology diffusionhas made significant inroads in explaining variation in

its scale, treating as fixed the total potential market. Considerably less attention has been paid to

changes in scope—the set of potential applications, and thus the size of the market itself—despite

that this extensive margin is one of the principal dimensions along which technologies spread.1

∗Harvard University and NBER; dgross@hbs.edu.

I am grateful to Barry Eichengreen for his support in this project. I also thank Dominick Bartelme, Carola Binder, Susan

Carter, Brad DeLong, Alex Field, Joel Mokyr, Petra Moser, Alan Olmstead,Mar tha Olney, Paul Rhode, Daniel Robert,

Richard Sutch, Noam Yuchtman,and the Editor and referees for helpful comments and suggestions, as well as participants

in the BEHL Economic History Lunch, All-UC Hundred Flowers conference, and NBER DAE Summer Institute poster

session. I am grateful to Richard Sutch for sharing data on hybrid corn diffusion from the USDA Agricultural Statistics. I

thank the BerkeleyEconomic Histor y Lab and All-UC Group in Economic History for financial support. This research was

additionally supported by NSF Graduate Research FellowshipGrant no. DGE-1106400 and an EHA Graduate Fellowship.

The appendices referenced in this article are available online at www.dpgross.com/.All errors are my own.

1As Griliches (1957) shows, logistic models of technology diffusion are parametrized by (i) whenit begins, and

(ii) the rate at which it proceeds. These two parameters characterize what I refer to in this article as “scope” and “scale,”

respectively. Research on diffusion has overwhelmingly focused attention on the latter, which has been attributed to

heterogeneous costs and benefits (Duflo, Kremer, and Robinson, 2008; Suri, 2011), fixedcosts of adopting an indivisible

technology (David,1966; Olmstead, 1975), and changes in relative factor prices (Manuelli and Seshadri, 2014), as well as

C2018, The RAND Corporation. 427

428 / THE RAND JOURNAL OF ECONOMICS

This article shows that the historical diffusion of farm tractors—a technology which revo-

lutionized 20th-century crop production and is a fixture in modern agriculture—was the result

of not only an increasing number of users, but also a growing number of uses. The tractor first

developed for narrow applications with existing complementary equipment, exogenously high

demand, and relatively lower R&D costs, and initial diffusion was accordingly rapid for these

applications, but otherwise limited in scope. Only later did tractor technology become sufficiently

general in purpose for its diffusion to be broad based and pervasive. This pattern of expanding

scope is consistent with other historical examples and with economic theory, which suggests

that in this context, R&D will naturally progress from specific- to general-purpose variants of

an innovation, and that these technical advances will (i) drive the development of additional

complementary technologies, and (ii) and directly translate to an increasing scope of diffusion.

Lags in diffusion can therefore be the result of hold-ups and market failures in R&D that stymie

the generalization of existing technology.

The article opens by reviewing the history of the farm tractor. Here, it is useful to first clarify

what the tractor is: farm tractors are vehicles that tow and powerthe ag ricultural implements that

do the day-to-day workof plowing, planting, cultivating, and harvesting crops. Though now used

in nearly every agricultural field operation and in the production of nearly all crops, tractors first

developed for use in tillage and harvesting grain. Early, fixed-tread models could not navigate

row crops without destroyingthe crop, and this generation of tractor technology was therefore not

a candidate to replace draft power on corn-growing farms at any price. By the 1930s, however,

a more versatile, general-purpose design had emerged, making it possible for these farms to

“replace their horses and mules with one general-purpose tractor” (Sanders, 2009).

The era of the tractor in US agriculture begins in the late 1910s, prior to which diffusion was

effectively zero. Using serial numbers and production data from the four major manufacturers of

this period, I first verify that fixed-tread models dominated tractor production up until the early

1930s, accounting for 96% of tractors manufactured from 1917 to 1928, and 91% through 1932.

During the 1930s, the industry made a near-complete transition to general-purpose models, which

comprised over 85% of units produced between 1933 and 1940.

I then use county-level data from the US Census of Agriculture to show that the initial wave

of tractor diffusion in the 1920s was concentrated in the Wheat Belt states of North Dakota, South

Dakota, and Kansas, whereas a second wavefrom 1930 to 1940 was concentrated in the Corn Belt

states of Iowa, Illinois, and Nebraska. This sequence is plainly visible in maps of wheat versus

corn intensity and diffusion (Figure 1). Numerically, I find that county-level diffusion from 1925

to 1930 was 0.4 percentage points greater with every percent of farmland in wheat but did not

vary with farmland in corn. From 1930 to 1940, the pattern is precisely reversed, following the

introduction of the general-purpose tractor and mechanical corn harvestor; in the 1940s, diffusion

rounds out in counties with little of either crop and primarily growing hay. The results are robust

to a wide variety of controls, sample restrictions, and definitions of diffusion—establishing that

they are not due to changes in farm sizes, local factor prices, financial conditions, dealer networks,

New Deal relief, the Dust Bowl, the contemporaneous diffusion of hybrid corn, or other features

of Midwest agriculture that might have affected tractor demand in this period.

The question remains as to why the tractor’s development followed this sequence. To put

structure around this phenomenon, I introduce a model of innovation where a firm develops a

technology with application-specific and general-purpose technological attributes, and where the

value of the innovation depends on the evolving quality of complementary technologies. The

model borrows ideas from the framework developed by Bresnahan and Trajtenberg (1995) to

characterize general-purpose technologies, while endogenizing the path of product development.

Intuitively, this model suggests that product features develop in the order in which they are most

valuable, implyingthat new technologies will often first be invented for narrow applications where

to suboptimal decision-making due to credit constraints (Clarke, 1991), information spillovers (Conley and Udry,2010;

Dupas, 2014; Munshi, 2004), and individual biases (Duflo, Kremer, and Robinson, 2011).

C

The RAND Corporation 2018.