Professional Documents
Culture Documents
DOI: 10.1111/ajae.12273
ARTICLE
1
Department of Economics, The Orfalea College
of Business, California Polytechnic State Abstract
University, San Luis Obispo, California, USA There is a chronic shortage of agricultural labor in the
2
Morrison School of Agribusiness and Resource US. Although growers increasingly turn to guest-worker pro-
Management, Arizona State University, Tempe,
grams to meet their labor needs, few regard immigrant
Arizona, USA
3
workers as a viable long-term solution. Many producers of
International Research Institute for Climate and
Society, Columbia University, Palisades, New
labor-intensive agricultural commodities regard mechaniza-
York, USA tion as a clear long-term solution, making the slow rate of
adoption of mechanized harvesting equipment in the US an
Correspondence empirical puzzle. In this article, we demonstrate that wage-
Timothy Richards, Morrison School of
setting farmers have an incentive to “overmechanize,” or
Agribusiness and Resource Management, Arizona
State University, Tempe, AZ, USA. employ more than the cost-minimizing level of capital when
Email: trichards@asu.edu capital and labor are substitutes, but “undermechanize”
when labor and capital are technical complements. This out-
Funding information come can cause agricultural labor problems to persist under
Social Science Registry, Grant/Award Number:
AEARCTR-0006532; USDA NIFA, Grant/Award complementarity. To assess the potential role of farm under
Number: 2019-67023-29415 investment in labor augmenting capital equipment, we
examine labor market outcomes following the adoption of
non-autonomous harvesting aids on a large strawberry farm
in Central California. We develop an econometric model of
peer-affected productivity that controls for the group perfor-
mance of farm workers operating in crews and find that
mechanical aids complement labor in strawberry production,
a finding that helps explain not only the relative lack of
mechanized harvesting in strawberry production but, more
generally, the persistent productivity gap in agricultural
industries. We examine the broader implications of our the-
ory for the slow rate of adoption of mechanical harvesting
technologies in US agriculture by comparing general wage
trends across several labor-intensive and non-labor-intensive
industries in California.
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction
in any medium, provided the original work is properly cited.
© 2021 The Authors. American Journal of Agricultural Economics published by Wiley Periodicals LLC on behalf of Agricultural & Applied Eco-
nomics Association.
KEYWORDS
farmworkers, imperfect labor markets, labor shortage, mechanization,
peer effects, productivity
JEL CLASSIFICATION
J22, Q12, Q18
1 | INTRODUCTION
There is a chronic shortage of agricultural labor in the US (Mercier, 2014). Given the increasing dif-
ficulty of relying on foreign guest workers for harvest labor, and the inherent distaste for agricultural
jobs among domestic workers, a potential solution to alleviating the domestic farm labor shortage is
to increase the productivity of existing workers. One way to increase labor productivity is to raise
piece-rate compensation; however, recent research suggests that raising piece rates contributes only
minimal productivity effects in manual-labor settings (Hill, 2018; Richards, 2020; Stevens, 2018). Yet
managers may be reluctant to pay higher piece rates than neighboring farms as a practical matter,
because paying higher wages to a fixed pool of farm workers has the potential to stimulate aggressive
wage competition from rivals. Given that most work crews tend to perform close to their maximum
productivity when harvest conditions are favorable, paying higher piece rates to farm workers may
not be a viable solution. In this article, we consider the potential for grower investment in mecha-
nized harvesting aids to raise agricultural labor productivity.
In principle, agricultural mechanization can help resolve labor-shortages in one of two ways:
(i) capital can substitute for labor, as in the dairy and processing tomato industries, thereby obviating
the need to hire workers; or (ii) capital can complement labor, raising worker productivity and all-
owing each worker to produce more output. Although these two cases may appear to be observation-
ally equivalent in terms of addressing farm labor problems, we demonstrate that the implications for
capital investment rates materially differ when labor markets are oligopsonistic. Specifically, we con-
struct an oligopsony model of farm labor procurement to show that farms have an incentive to
“overmechanize,” that is, to employ more than the cost-minimizing level of capital when capital and
labor are substitutes but “undermechanize” in cases where labor and capital are technical comple-
ments. The reason that technical complementarity can stifle capital investment in agriculture is that
employing mechanized harvesting solutions raises the marginal value product of labor for the
adopting farm, stimulating aggressive wage competition among rival agricultural producers that bids
up equilibrium wages in the rural community. Higher equilibrium farm wages, in turn, disincentivize
capital investment in the long run, disrupting the virtuous cycle envisioned in “induced-innovation”
stories of aggregate productivity growth.
It is well known that growers can increase the productivity of their workforce, or even replace
workers altogether, by adopting some form of mechanized-labor solution. Although the past decade
has seen a dramatic increase in the pace of innovation in agricultural mechanization (Gallardo &
Sauer, 2018), most harvesting technologies in their current form augment existing manual labor
rather than completely eliminating it. Given the growing importance of mechanized solutions in
labor-intensive agricultural industries, the lack of empirical estimates on the impact of mechaniza-
tion on productivity, and the associated impact of mechanized harvest aids on wages and worker
productivity, is surprising.
We frame our analysis of the marginal value of mechanical harvesting aids on labor productivity
using data from a large strawberry farm in Central California. Like most strawberry growers in
California, the farm we examine combines manual harvesting techniques with labor using a
non-autonomous trailer that travels through the field with workers placing picked produce on a con-
veyor behind the trailer.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1437
Our analysis contributes to a growing literature on the effect of peers on own-worker productiv-
ity among workers in teams. Because field workers are typically organized into teams to perform har-
vest operations, controlling for peer effects is essential to avoid misspecification bias. Peer effects
among workers in teams is a subject that has been studied both in agricultural (Bandiera et al., 2005,
2010; Benyishay & Mobarak, 2019; Conley & Christopher, 2001; Foster & Rosenzweig, 1995; Hill &
Burkhardt, 2021; Kondylis et al., 2017; Stevens, 2018; Vasilaky & Leonard, 2018) and non-
agricultural settings (Bramoullé et al., 2009; Bursztyn et al., 2014; Cohn et al., 2014; Cornelissen
et al., 2017; De Giorgi et al., 2010; Feigenberg et al., 2013; Glenn Dutcher et al., 2019; Guryan
et al., 2009; Marmaros & Sacerdote, 2002; Mas & Moretti, 2009; Murphy, 2019; Park, 2019; Tan &
Netessine, 2019). Common among these applications is a careful consideration of the “reflection
problem” inherent in estimating peer effects (Manski, 1993), where the estimation of peer effects on
own productivity is confounded by the fact that the individual is also part of the group. Indeed,
Angrist (2014) argues that many empirical models of endogenous peer effects are biased by the fact
that there is a simple, mechanical–statistical relationship between own productivity and group pro-
ductivity. To address this potential problem, we follow the approach developed by Mas and
Moretti (2009) to measure individual productivity in groups, while taking care to ensure that the
general criticisms of Angrist (2014) are addressed.
An advantage of using farm-level data to estimate the effect of harvest machinery on labor pro-
ductivity is that farm capital investment may complement labor in some agricultural settings and
substitute for it in others, confounding inferences obtained from highly aggregated production data.
Typical sources of agricultural production data lack sufficient firm-level detail to allow researchers to
disentangle the marginal product of capital from the marginal product of labor in a traditional
production-function context.
In this article, we utilize firm-level data from a large, California strawberry farm to estimate the
marginal value of augmenting labor with mechanical aids using an econometric model of peer-
affected productivity. Our farm-level data allow us to observe the amount produced per worker per
unit of time, the crew composition of workers, and whether or not an individual worker was aided
by mechanical means.
Our model predicts that mechanization results in lower farm wages in the case of substitutes, but
higher farm wages in the case of complements. This outcome generates clear empirical implications
for how farm wages, employment, and farm-level investments in mechanization differ across indus-
tries according to both industry- and product-specific technical relationships between labor and cap-
ital in agricultural production.1
After carefully controlling for the potential impact of peer influence on workers’ output in our
firm-level strawberry data, we find that machine-aided workers are able to produce approximately
5%–6% more output per unit of time than those who do not receive mechanical assistance. More
importantly, our empirical findings also suggest that the type of mechanization we describe here,
which involves mechanical harvest aids commonly used for horticultural crops, serves to comple-
ment labor rather than substitute for it. Therefore, we expect wages in industries relying on such
labor-augmenting mechanical aids to rise faster than in comparable industries that do not rely on
such technology.
Our empirical analysis of general wage trends show that this is indeed the case. Wages in the
strawberry industry have risen some 35.2% faster than in comparable industries that do not rely on
machines to augment harvesting operations. To the extent that farm labor is procured under oligop-
sony conditions, this trend toward higher wages in industries that provide mechanized harvesting
assistance to workers is likely to be associated with persistent under-capitalization in the farm input
mix and reduced agricultural productivity over time relative to other sectors of the economy.
1
We describe these relationships as product specific in cases where the relationship between labor and capital differs based on the intended end
use of the product, as in the case of fresh tomatoes versus processing tomatoes.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1438 FARM LABOR PRODUCTIVITY
There is considerable evidence that firms exert oligopsony power in labor markets. As Bhaskar
et al. (2002) observe, any labor market with an active search-and-matching process, search frictions,
bargaining, or heterogeneous preferences for employer attributes is best interpreted as imperfectly com-
petitive (Bhaskar et al., 2002; Staiger et al., 2010). In providing evidence of inelastic labor supply from the
US nursing market, Staiger et al. (2010) explain that “monopsonistic behavior may be pervasive, with
individual firms facing upward-sloping labor supply curves because of the presence of oligopoly, differen-
tiation between firms, moving costs, costly job search, or efficiency wages” (p. 211). Depending on the
commodity, firms in agriculture may face one, or all, of those conditions, a feature of farm labor markets
documented by Dickens et al. (1995) and Richards (2018). The notion of oligopsony labor markets in
agricultural settings is consistent with broader evidence of oligopsony power in agriculture due to sunk
investment in specialized assets (Just & Chern, 1980; Rogers & Sexton, 1994).
Our analysis contributes to the substantive literature on agricultural labor employment, the methodo-
logical literature on estimating labor productivity with peer effects, and to the broader literature on pro-
ductivity growth and mechanization in agriculture in several ways. First, our theoretical finding that
firms have an incentive to underinvest in capital when labor and capital are complements contributes to
a long-standing debate between “push” and “pull” theories for the declining number of workers in agri-
culture. Although the pull theory maintains that the relative concentration of capital in urban centers,
and the rise of relative wages, has attracted labor out of agriculture and into the manufacturing-and-ser-
vice-based labor market in cities, the push theory maintains that labor is pushed off farms by the rapid
rate of technological change in agriculture (Gallardo & Sauer, 2018; Kislev & Peterson, 1981).2 To the
extent that the underlying mechanism is a manifestation of induced innovation (Hayami &
Ruttan, 1970), we would expect to see relatively large capital flows into agriculture and high rates of pro-
ductivity; however, Gollin et al. (2014) and Herrendorf and Schoellman (2015) document exactly the
opposite, citing evidence of a persistent productivity gap between agriculture and other industries that
suggests an excessive allocation of workers to the agricultural sector. Our findings underscore this out-
come using microeconomic data from a single, representative farm rather than relying on the use of indi-
rect, macroeconomic evidence typical of this literature.
Second, we contribute to the empirical literature on the rate of mechanization and productivity
growth in agriculture. Although researchers in the macroeconomics literature carefully document a
persistent productivity gap between agriculture and other industries (Gollin et al., 2014;
Herrendorf & Schoellman, 2015), the mechanisms they cite remain somewhat speculative. In con-
trast, we provide theoretical and empirical evidence on the precise causes of undercapitalization, and
lower productivity, in an important industry in the agricultural sector. Our empirical evidence on
wage trends in mechanized and non-mechanized industries supports our theoretical explanation,
providing a microfoundation for the effect of mechanization on labor productivity that is consistent
with macroeconomic observations.
Third, our evidence of labor-capital complementarity also contributes to the empirical literature
on the peer-effects on productivity. Although others estimate the peer effect on productivity in simi-
lar, agricultural contexts (Bandiera et al., 2005, 2010; Hill & Burkhardt, 2021; Stevens, 2018), we are
the first to use a peer-rich environment to isolate the effect of mechanization on observed levels of
productivity and characterize the nature of agricultural production in an economically important
industry. Although our data are specific to the strawberry industry, there are many other examples
of the type of labor-augmenting technology that we describe here.
In the next section, we provide some background on labor-productivity estimation in agricultural set-
tings and how it frames our own analysis. In the third section, we derive a theoretical model that gener-
ates testable hypotheses regarding the implications for wage and capitalization trends in agriculture,
conditional on whether mechanization complements, or substitutes for, labor. We describe our empirical
model of peer-induced productivity estimation in the fourth section, and our empirical approach to
2
Alvarez-Cuadrado and Van Long (2017) maintain that both of these dynamics have dominated during different eras, and both are responsible
for the movement of labor out of agriculture.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1439
testing for industry-level wage trends. The fifth section presents our data and identification strategies,
and provides some model free evidence derived from both the productivity and wage-trend data sets. We
present and discuss our findings in the sixth section, and the final section concludes.
where p denotes the price of the farm commodity. In the continuation game, Ki is fixed. Accordingly,
farm i selects the effective wage rate in Stage 2 to satisfy the first-order necessary condition
π ii ¼ pf L wi Lii Li ¼ 0, ð1Þ
where f L ¼ ∂f ð∂L K ,L Þ
i i
i
is the marginal product of labor. Notice that pfL wi > 0 in equilibrium, which
captures the oligopsony outcome that wages are depressed below the marginal value product of
labor.
Let wi ðK Þ denote the simultaneous solution to Equations (1), where K = (K i , K j ) is the vec-
tor of first-stage harvest capital choices of the farms. For the analysis to follow, we assume a
wage increase by the rival farm raises marginal profit for farm i, π ij i > 0, which ensures that
the reaction functions slope upwards in farm wages. We also assume the usual stability condi-
tion for wage-setting oligopsony models that an increase in the wage of farm i decreases mar-
ginal profit for farm i more than an increase in the wage offered by farm j increases farm
profit, jπ iii j > jπ iij j.
In the first-stage mechanization game, profit for farm i is
π i K i ,wi ðK Þ ¼ pf K i , Li wi ðK Þ, wj ðK Þ wi ðK ÞLi wi ðK Þ, wj ðK Þ rK i :
∂π i i ∂wj
i ¼ pf K r þ pf L wi Lj ¼ 0,
∂K ∂K i
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1441
Proposition 1: If farm labor and farm capital are technical complements (substitutes),
fLK > 0 (< 0), then firms underinvest (overinvest) in farm capital relative to the cost-
minimizing level, pfK = r.
Proof: The choice of Ki by farm i affects wages for farm workers in the continuation game
according to
2 32 3 2 3 2 3
6 7 6 7
6 i i 76 7 6 7
7 ¼ 6 pf Li 7dK i þ 6 7 j
6 π ii π ij 76 dw 6 0 7dK ,
54 5 4 LK i 5
i
4 4 5
j
j j
π ji π jj dwj 0 pf LK Lj
∂π i
where ∂Kii ¼ pf LK Lii is the effect of farm mechanization on the marginal productivity of labor. This
term is positive when farm labor and capital are technical complements (fLK > 0) and negative in the
j j
case of technical substitutes (fLK < 0). Letting Δ = π iii π jj π iij π ji > 0 denote the determinant of the
coefficient matrix, the mechanization choice of farm i affects farm wages in Stage 2 of the game as
j
dwi π
i jj s
i ¼ pf LK Li ¼ f LK , ð2Þ
dK Δ
j
dwj π
i ji s
i ¼ pf LK Li ¼ f LK , ð3Þ
dK Δ
s
where “ ¼ ” denotes “equal in sign”.
The intuition for this result is that firms select compensation rates for workers that depend on the
first stage choice of farm mechanization. In cases where farm labor and harvesting aids are technical
complements (fLK > 0), increasing the stock of mechanical harvest aids in stage 1 shifts labor demand
function outward for farm i, which causes the rival farm j to increase its farm wage in the continuation
game. Generating such a response is undesirable from the perspective of firm i, which provides an incen-
tive for farm i to choose a lower level of mechanization than the cost-minimizing level.3 As a result, both
farms undermechanize in the market equilibrium, leading to an inefficient use of resources in which the
marginal value product of labor exceeds the wage rate and the marginal value product of capital is greater
than the rental rate. The opposite is true for substitutes, as farms overinvest in capital, with the marginal
value product of capital being less than the rental rate. In either case, the marginal rate of technical sub-
stitution of labor for capital diverges from the ratio of factor prices.
Proposition 2: If farm labor and farm capital are technical complements (substitutes),
fLK > 0 (< 0), then in the symmetric oligopsony market equilibrium K* = Ki,* = Kj,*.
3
We have anecdotal, and logical, support for this outcome. First, the manager of the farm that provides the data described below would not let
us conduct a field experiment that involved increasing piece rates for some workers, and not for others, because he was worried his workers
would talk to workers at other farms, thereby manifesting the wage spiral we describe in our manuscript. Second, Dey and Flinn (2005) develop
a model of oligopsonistic wage bargaining in which employer-provided healthcare insurance serves as a complementary input that increases
worker productivity. Firms avoid destructive wage competition by providing healthcare insurance, increasing employee retention without
inciting labor-market competition. Third, Billikopf (2008) provides extension advice to growers in California that includes an explicit warning
to not increase piece rates unilaterally, lest other growers respond proportionately.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1442 FARM LABOR PRODUCTIVITY
Proof: In the symmetric market equilibrium K* = Ki,* = Kj,*, the implied compensation rate is
the same for workers at both farms w ¼ wi ¼ wj . An increase in mechanization affects wages as
dw dwi pf LK Lii j j
s
¼ ¼ π ji π jj ¼ f LK ,
dK i dK i wi ¼wj Δ
FIGURE 1 Production for crews that work with mechanized aid or work on the ground
In this section, we derive an empirical framework appropriate to estimate the marginal value of
mechanization and its broader impact on growers’ incentives to mechanize. For this purpose, we fol-
low a two-step empirical procedure. In the first stage, we test for whether the marginal productivity
of labor rises or falls in the level of mechanization, which is the core parametric relationship required
by our theoretical model, using a highly granular, firm-level data set on productivity and mechaniza-
tion. Based on the results of this stage, we then use data on aggregate wage measures (from the Quar-
terly Census of Employment and Wages [QCEW]) to test the implications of our microlevel findings
on more macrolevel, industry-wide data (QCEW, 2021). We combine our findings from these two
stages to draw more general implications regarding the effect of mechanization on wages, and labor
employment. We describe each stage in turn.
productivity by using the leave-out mean and extract a true signal of group productivity using the
Mas and Moretti (2009) approach.
Following Mas and Moretti (2009), we first estimate the permanent productivity of each
individual and then construct peer means for each harvesting crew. With this approach, individual
productivity in the second-stage of the model is written as:
yidm ¼ θi þ αd þ Pidm ðθ1 ,…,θi1 ,θiþ1 ,…,θn Þ þ πN idm þ τM idm þ eidm , ð4Þ
where yidm is individual productivity (defined as boxes picked per hour), i indexes the individual,
d the date, and m the type of work (m = 1 for machine, m = 0 for ground). The unit of observation
in our estimation is a worker-day type. On a single day, a given worker may work for several hours
in a machine-assisted setting (i.e., “machine”), with a given set of co-workers, and then they may
work for additional hours without machine assistance (i.e., “ground”), and with a different set of
coworkers. Their productivity will also differ between these two parts of the day. We divide each
worker day into two observations for the cases where a worker participates in both types of work on
a given day.4 Our estimation will treat these two parts of the day as distinct observations so that we
can estimate the effect on productivity of machine assistance, after controlling for peer effects.
Midm is a dummy for whether observation (i, d, m) is machine-assisted work (i.e., Mid0 = 0 and
Mid1 = 1 for all i, d). The coefficient τ is the primary coefficient in which we are interested, estimat-
ing the effect on productivity of the use of mechanized aids. Nidm refers to the size of the crew (the
number of employees plausibly influencing the worker i each day), and eidm is an error term.
The function Pidm is a “peer pressure function” that measures the impact of peers on own pro-
ductivity; it is a function of θi, the permanent productivity of each of the other members of the
crew for that date and type of work. In this model, the values of θi are interpreted as permanent-
productivity measures, or inherent abilities, of each worker in the data set. The model intends to
measure the impact of peers’ pure ability on productivity, so these permanent-productivity measures
must be purged of any peer-on-peer effects before entering back into the peer-effect model (the
“reflection problem” of Manski 1993).
Defining the peer-pressure function (Pidm) as the simple leave-out mean of all other co-workers
on the same crew as individual i on a particular date d and work-type m, the model then becomes:
where θidm is the leave-out-mean of the other crew members’ permanent productivity.5 In this
expression, β > 0 is interpreted as positive spillovers due to a worker perhaps keeping up with faster
workers, or modeling the behavior of faster workers, whereas we interpret β < 0 as empirical evidence
of free-riding, or a worker who responds to observing more productive workers by reducing his or
her own effort through sheer demoralization.
In order to estimate this second-stage peer effect, however, it is first necessary to estimate the
permanent productivity of all workers, in a manner consistent with Equation (5). Therefore, our
first-stage model involves estimating θi from the fully-saturated expression:
X
n
yidm ¼ θi þ αd þ γ k T kidm þ πN idm þ τM idm þ eidm , ð6Þ
k¼1
4
On a given day, an individual may do only ground work, only machine work, or both. There are a large number of workers who never do
machine work (about 57% of workers). Among the workers who sometimes do machine work, it is most common for a day to involve both
types of work (about 43% of worker days), and it is less common to spend the entire day on machine-assisted work (about 24% of worker
days). When a crew splits the day between both types of work, the composition of the crew often (about 40% of the time) changes between the
two parts of the day.
5
Mas and Moretti (2009), however, estimate the model in (5) in first differences, obviating the need to include the location of the individual, or
the hour, date, and store interaction terms.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1445
where T kidm ¼ 1 when worker i and worker k ≠ i are working on the same crew on date d and work-
type m, and T kidm ¼ 0 otherwise. We consider alternatives to this definition, explained in more detail
in the online appendix.
Our data provide detailed information on individual productivity and crew membership for each
date and worker. Perhaps most importantly, we also know whether or not the worker used the Har-
vest Pro machine or not, which provides a binary indicator of the presence or absence of mechanized
aid (our Midm) variable. Because our data include measures of both individual productivity, and crew
composition, the Mas and Moretti (2009) approach is well-suited to testing our core hypotheses
regarding the relationship between mechanization and productivity.
In estimating Equation (6), we control for a complete set of individual (i) and date (d) fixed
effects in estimating the permanent productivity (θi) of worker i. In Equation (6), the vector T kidm
consists of indicators for all the relevant combinations of workers in the data for each particular
crew. That is, we include indicators for each case where i works with k such that γ k measures the
spillover effect that worker k has on each of his or her peers. This parameter measures whether spe-
cific combinations of individuals have a measurable effect on the productivity of individual
i regardless of the productivity of the group.
In terms of econometric identification, allowing for this level of crew heterogeneity speaks to the
individual-covariates requirement of Angrist (2014). By removing these effects, we ensure that there is no
mechanistic relationship between individual and group productivity—one that is assured by the nature of
the data.
This approach imposes some structure on the worker externalities when estimating each individual’s
permanent productivity. The productivity effect of working with worker k is always γ k and the effect of
working with workers k, j, and l is γ k + γ j + γ l. In the Mas and Moretti (2009) framework, in contrast, it
is possible that the effect of working with k, j, and l is different than the effect of working with k and
j plus the effect of working with l only, but in this setup they are assumed to be the same. To allow for
this more general form of peer effects, we consider another specification where, in addition to the Tk
dummies used in Equation (6), we also include specific crew-composition dummies in the first-stage
model for every unique crew composition that occurred at least δ times in the data, and then test for the
value of δ that provides the best fit to the data using the procedure described in the online Appendix.
For any given value of δ, the first-stage model with crew composition dummies is:
X
n X
C
yidm ¼ θi þ αd þ γ k T kidm þ μc Dcidm þ πN idm þ τM idm þ eidm , ð7Þ
k¼1 c¼1
where c = {1, 2, …, C} represents all combinations of workers that make up a full crew minus one
person that occurred on at least δ days. Dcidm is a dummy that is equal to 1 if the other crew members
that worked with individual i on date d for work-type m exactly corresponds to the crew composi-
tion c, and Dcidm ¼ 0 otherwise.
In our case, there is considerable variability from 1 day to the next in terms of crew composition,
which we document below. Once in the crew, workers typically remain with the crew for the entire
season, but there is substantial rotating out, days with missing crew members, or individuals doing
different jobs within the same crew. For example, we selected one of the largest crews in the farm
and examined how the composition of the crew changed over the course of an entire work year, or
harvest season. We found that the composition of the crew differed from 1 day to the next on 95%
of the cases. More importantly for the Mas and Moretti (2009) approach, there is considerable vari-
ability in the identity of the workers to each focal worker is exposed from day to day, so, as a practi-
cal matter, any direct interpersonal effects could come from any one of a number of different
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1446 FARM LABOR PRODUCTIVITY
individuals in the crew. Therefore, these direct interpersonal effects that form part of the first stage
of Mas and Moretti (2009) is readily estimable in our data.
6
Conversations with industry officials at the California Table Grape Association (CTGA) suggest that wine grapes in the Central Valley of
California can be mechanically harvested, but not table grapes. Further, Gray (2020) argues that 90% of all wine grapes are mechanically
harvested, using technology more akin to the processing-tomato harvester than the mechanical aids used in the strawberry industry, so we are
confident that our harvesting-wage data from the QCEW refer to manually harvested table grapes and not mechanically harvested wine grapes.
Including county-level fixed effects controls for county-level variation in harvesting technologies. With respect to apples, Zhang et al. (2016)
argue that, despite their theoretical-availability “[S]emi-automatic harvesters for fresh market apples have not been commercialized due to
unacceptably high incidence of apple bruising” (p. 1165), and that “[A]pples are still primarily hand-harvested by seasonal farm employees”
(p. 1165).
7
We examine the robustness of our findings with more flexible polynomial time trends, but our qualitative conclusions remain the same.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1447
quarter—to address any concerns with omitted variable bias that may affect our parameter of inter-
est, the estimated wage trend.
Our empirical model, therefore, is written as:
log wrjq,t ¼ ϕ1r þ ϕ2j þ ϕ3q þ ϕ4 log wrjq,t1 þ ϕ5 log Erjq,t þ ϕ6 log Lrjq,t þ ϕr7 T þ μrjq,t , ð8Þ
where wrjq,t is the output-price deflated weekly wage for workers in crop r, county j, quarter q, and
year t; Erjq,t is the number of establishments reporting employees for each crop, county, quarter, and
year; Lrjq,t is the number of workers employed; T is a linear time trend, ϕ1r, ϕ2j, and ϕ3q are crop,
county, and quarter fixed-effects, respectively; and μrjq,t is an i.i.d. error term.8
Our parameter of primary interest is ϕr7, which is the trend rate of wage growth in each industry.
If the estimated value of ϕr7 for strawberries, for example, is significantly greater than the equivalent
value for the other, more manual-labor-intensive crops, then our data provide empirical support for
our core hypothesis. Namely, when mechanization complements labor inputs, we expect wage
growth to be higher than otherwise. In the next section, we describe the data we use to estimate both
parts of our model in more detail.
8
Focusing on weekly wages does not lead to a tautological relationship between mechanization and wages. Farm managers set piece rates
consistent with the market, which involves paying workers on the machine a lower piece rate than those who work “on the ground” so as to
adjust for any mechanization benefit. This is common industry practice, so even in aggregate, there is no mathematical relationship between the
extent of mechanization and the weekly wage. Intuitively, there are simply not enough machine positions to accommodate all of the employees,
so if the manager sets a common piece rate across both machine and ground jobs, there would clearly be competition for positions on the
machine, and ground work would come to be seen as something of a punishment. Because the overriding concern of farm management is to
attract and retain workers, these types of problems have to be avoided. Billikopf (2008) describes these issues in some detail, so we assume his
recommendations as to the optimal, or preferred, management of piece-rate programs from a human-resource management perspective are
taken seriously by most managers.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1448 FARM LABOR PRODUCTIVITY
composition of the other crews, of course, differs from this one, the pattern and level of variation is
similar to this crew. We interpret this level of variation as sufficient to identify any peer effects in
the data.
Because some workers do not do machine-assisted work, there is a potential concern that our
results are biased by selectivity—that those workers who do machine work have larger productivity
boosts from machine work than other workers. However, because work assignment is done at the
crew level, not the individual level, the ability of workers to choose their preferred type of work is
limited and the ability of the farm to assign individual workers to the type of work based on their
ability is also limited. Although workers are free to opt out of machine work, in most cases workers
follow their crew assignment. To illustrate this, we divide our sample into those who do only ground
work (“ground only”) and those who do some machine work (“machine”). Of the 528 ground-only
workers, 99 (19%) of them work on crews that only do ground work. An additional 334 (63%) of
them only work on days when their crew does exclusively ground work. Thus, machine work is not
an option for 82% of ground-only workers, based on their crew and dates worked.
Summary statistics are reported in Table 1 for both ground-only and machine workers. When
comparing the ground-work productivity for ground-only workers to machine workers, we note that
average ground-work productivity is higher for the machine group than the ground-only group,
suggesting that our machine-effect estimates are based on a sample with above average productivity.
However, the ground-only group includes over 90 workers who work only 1 day in the entire season
and over 150 additional workers who work more than 1 day but less than a week (whereas the
machine group has very few of these short-term workers). The short-term workers tend to have
lower productivity and are a big reason for the productivity gap observed between ground-only
workers and machine workers. After controlling for number of days worked, the difference in pro-
ductivity between the two groups is much smaller. In the online Appendix, we provide additional
evidence that our results are not driven by the selection of high ability workers into machine work.
Ground-only Machine
Total workers workers
Number of observations 53,391 19,060 34,331
Number of observations, ground work 37,184 19,060 18,124
Number of observations, machine work 16,207 0 16,207
Length of season (days) 199 199 199
Days with ground work 199 199 199
Days with machine work 170 0 170
Number of workers 929 528 401
Median age 28 30 27
% Female 25% 28% 21%
Productivity, mean (SD) 10.71 (4.78) 9.19 (3.95) 11.55 (5.00)
Productivity ground work, mean (SD) 10.15 (4.62) 9.19 (3.95) 11.16 (5.05)
Productivity machine work, mean (SD) 11.99 (4.90) 11.99 (4.90)
Number of crews 13
Crew size, mean (SD) 21.9 (7.0)
Crew size ground work, mean (SD) 21.7 (6.4)
Crew size machine work, mean (SD) 22.3 (8.1)
Note: Unit of observation is a worker/date/type of work where type of work is machine or ground. Productivity is defined as the number of
boxes per hour. “Ground-only workers” never do machine-assisted work; “machine workers” sometimes do machine-assisted work.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1449
F I G U R E 2 Variation in crew-composition: crew # 7 in 2017. % Full crew measures the percentage of the maximum
number of workers who ever worked with this crew who are working each day, for Crew 7, 2017 harvest year
Prior to estimating our structural model of worker productivity, we examine the production data
using a model-free or reduced-form approach in order to determine whether there are any patterns
apparent in our data that may either support or contradict our core hypothesis. Because of the com-
plexity of the peer-effect estimation method, our reduced-form estimates show whether there is any
impact of mechanization on individual productivity while only controlling for individual and date
fixed effects, without controlling for the peer effect as in Mas and Moretti (2009). We find the mar-
ginal effect of working with the machine to be 0.659 (standard error = 0.029) with a sample size of
N = 53, 640 daily productivity observations. We interpret this summary evidence as clear support
for at least a preliminary expectation that productivity should be higher when working with the
machine, as opposed to working on the ground. In terms of our theoretical model, this summary evi-
dence is interpreted as support for a complementary impact of mechanization on labor productivity.
Conclusive evidence of this effect, however, requires that we estimate the full two-stage model of
productivity, in which we control for the full vector of peer, date, crew, and machine effects. These
results are reported in the next section.
In the second-stage of our empirical approach, that is, testing the implications of our findings for
aggregate wage trends, we use data from the Quarterly Census of Employment and Wages (QCEW,
Bureau of Labor Statistics) QCEW (2021). Our QCEW data describe average wages paid to har-
vesters for each commodity, on a quarterly basis, over a 1990–2019 sample period. Importantly, the
QCEW is a census, and not a survey, so every employer who reports hiring employees for
unemployment-coverage purposes is captured in our sample. Because it is a census, the QCEW is
likely to be highly representative of actual wage trends and is less likely to be biased by either non-
reporting problems or self-selection into the data sample.
Our aim with this data is to test for differences in wage trends that may be due to the adoption
of mechanized harvesting aids in berry harvesting. Our empirical design accounts for the fact that
market wages are locally determined, so we gather wages from the QCEW at the county level, and
our sample counties vary by commodity. For each commodity, we use average wages for the top-five
counties, by production volume, except for where confidentiality regulations prevent the disclosure
of wages and employment.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1450 FARM LABOR PRODUCTIVITY
For each commodity, the sample counties are: Ventura, Monterey, Santa Barbara, Santa Cruz,
and Orange (strawberries); Kern, Fresno, Napa, Tulare, and Sonoma (grapes); Santa Cruz, El
Dorado, San Joaquin, and Sonoma (apples); and Ventura, Santa Cruz, and Santa Barbara (other
berries). In the empirical model described above, we control for any unobserved variation in mecha-
nization rates across counties by including a set of county-level fixed effects. Our QCEW data con-
tains measures of the average weekly wage paid to harvesters, the number of employees in the
county, and the number of establishments reporting, so contains a number of potentially important
covariates.
From first principles, the demand for labor, whether in competition or in oligopsonistic markets,
is a function of the marginal product of labor, and output prices. Therefore, wages are determined
not only by productivity but the interaction of productivity and output prices. Therefore, we account
for the output-price effect in different specifications by either including output prices as explanatory
variables or by deflating observed wages by an output-price measure. For this purpose, we use
producer-price index (PPI) values from the Bureau of Labor Statistics (commodity price indices) for
US fresh fruit markets for each commodity on a monthly basis for the sample period (1990–2019).
We then aggregate these monthly prices into quarterly average price-index values that are compara-
ble to the quarterly wage values from the QCEW data.
In the empirical model, we consider various specifications for the wage-trend model, including
output prices as explanatory variables, and estimating the trend in real wages, calculated by deflating
nominal wages with the producer price index for each commodity. Our wage normalization proce-
dure is straightforward, as we simply divide the observed wage each month by the contemporaneous
PPI value for the same commodity and month. We summarize the wage-trend data in Table 2 below.
Examining this summary data shows that the largest percentage increase in wages are actually for
workers in the crops we define as more likely to be manually harvested: grapes and apples. However,
more conclusive tests for differences in wage trends requires a formal econometric model that con-
trols for county, commodity, and quarterly fixed effects, as well as for trends in output prices. We
present our findings from this econometric model in the next section.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1452 FARM LABOR PRODUCTIVITY
we allow for heterogeneous peer effects between the two types of job. In each specification, we esti-
mate the model across all crews, dates, and field locations for the 2017 harvest year. After determin-
ing whether machines complement or substitute for labor from this specification, we then examine
industry-level wage data for the broader implications of our findings.
Our first set of estimates are in Table 3 below, where the coefficients are interpreted in terms of
boxes per hour. In the first two columns of Table 3, we present the estimates from our core models
(based on Equation (6), including both the direct-machine effect (Model 1) and the machine effect
interacted with the peer effect (Model 2). These estimates show that there is indeed a strong, positive
peer effect among strawberry harvesters. Our finding in this regard is consistent with the empirical
literature on peer effects in agriculture (Bandiera et al., 2007; Chang & Gross, 2014; Hill &
Burkhardt, 2021), so is not, in itself, particularly surprising. However, when we interact the peer
effect with an indicator as to whether worker i worked on the machine or not, we find an additional
peer effect among machine workers, but the direct effect of the machine is no longer statistically dif-
ferent from zero. We interpret this result as suggesting that most of the productivity effect from
using the harvest machine is due to the fact that workers must interact in order to make the machine
work to maximum effectiveness. Regardless of the specific mechanism, the marginal effect of using
the machine is strongly positive, whether considered on its own or in combination with the team
nature of strawberry harvesting.
In addition to the Tk dummies used in Models 1 and 2, we also examine the effect of allowing for
specific crew-composition dummies as in Equation (7). In Table 3, the estimates for the “optimal”
value of δ are shown for Models 3 and 4, which allow for the direct and peer-interacted machine
effects as in Models 1 and 2. Models 5 and 6 provide estimates of a model in which δ is fixed at two
for comparison purposes. These estimates show that the core estimates are robust to changing how
we define the peers to which each worker is likely exposed. That is, the peer-effect estimate in Model
3 is very similar to the equivalent estimate from Model 1. The most dramatic difference between
Model 4 and Model 2 concerns the interaction between machine and -peer, where the direct
machine effect term is now negative and statistically significant. However, the interaction-term esti-
mate is larger in Model 4 than in Model 2, so the net effect of mechanization on productivity is
about the same in both models. The net machine effect shown in Table 3 is calculated by adding the
direct machine effect and the peer-interacted machine effect, using the mean value of peer ability
(i.e., θi , the average θi across all other workers). This net effect is the expected effect of mechaniza-
tion on productivity when working with peers of average ability. In Models 5 and 6, we see that
changing from the optimal value of δ to a value of δ = 2 again has little impact on the estimates, as
the peer, machine, and interaction effects differ little between Models 5 and 3, and Models 6 and 4.
The various specifications in Table 3 vary in how we allow for peer effects in the first-stage,
permanent-productivity estimation model. We also examined the robustness of our results to varia-
tions in how we define the machine effect in the first-stage. Although the initial specification of the
first-stage model is simple and intuitive, it assumes that the machine effect is the same for all
workers in the data. We also consider an alternative specification in which we interact the machine
effect with individual fixed effects in the first-stage model and re-estimate the second-stage model
with the revised permanent-productivity estimates. The details of this procedure are provided in the
online Appendix, but the results show that our core findings, namely that worker productivity rises
when using the machine, remain robust to how we define the first-stage machine effect. Conse-
quently, we estimate the wage-trend model with the null hypothesis that wages should have been ris-
ing in mechanically assisted industries (strawberries and other berries) relative to wage trends in
more manually harvested industries (apples and grapes).
HAMILTON ET AL.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1454 FARM LABOR PRODUCTIVITY
The estimates of our wage-trend model are shown in Table 4 below. We estimate several specifica-
tions in order to examine the robustness of our findings to different specifications of the wage vari-
able and the dynamic structure of the data. In the first two models (Models 1 and 2), we estimate the
model with the dependent variable defined as the log of average weekly wages. In Model 2, we
account for the fact that output prices should influence wages by including average commodity
prices as a regressor in the model. However, because this specification likely suffers from endo-
geneity bias, we deflate the observed wage by the average commodity price for the remaining three
specifications (Models 3–5). Our most comprehensive model (Model 5) also accounts for the likely
autoregressive nature of the wage data.
The results in this table show little difference in the estimated wage trends (the first four rows)
between the four commodities. However, the deflated-wage specifications provide a substantially bet-
ter fit than the first two models and are less likely to reflect bias from the clear endogeneity of the
output price. In Model 3, we see that the real-wage trend among strawberry workers was substan-
tially higher than for either grape or apple harvesters, an insight that does not change after we add
the number-of-workers and number-of-establishment variables in Model 4. When we add the auto-
regressive term in Model 5; however, the results in Table 4 show a considerable improvement in the
quality of fit and a moderation of wage trends for all commodities. After controlling for dynamics
inherent in the wages in each industry, we find that real wages in the strawberry industry trended
upward over the 1990–2019 sample period at an average rate 31.5% faster than among grape har-
vesters, and 38.9% faster than apple harvesters. Comparing wages for other-berry harvesters yields a
similar pattern.
Further, the difference in wage trends between manual and mechanically aided industries are
each statistically significant. There are four comparisons that are relevant in our model: the trend in
strawberries versus grapes (t = 55.16), strawberries versus apples (t = 65.60), other berries versus
grapes (t = 71.41), and other berries versus apples (t = 79.71). Therefore, each pairwise comparison
provides empirical support for our wage-trend hypothesis.9
Other factors may help explain the sharp differences in wage trends in our four sample indus-
tries: For example, the competitive structure among growers, plant genetics that make fruit easier to
harvest over time, or worker preferences for each industry are three of the more likely. It may also
be due to a relative shortage in commodity-specific labor or workers who simply do not want to do
the comparatively undesirable work of picking strawberries. However, by controlling for fixed county
and crop effects, we control for as many of these unobservable factors as possible.
More generally, there may be other explanations for the observed underinvestment in capital that
leads to our observed wage trends. First, wages are notoriously volatile, so the returns to investing in
mechanization may imply a substantial real option. However, it is well understood that the hysteretic
effect associated with real option values lead to a delay in adoption and not the absence of adoption
(Dixit, 1992; Richards & Green, 2003). Fortunately, the real option explanation lends itself relatively
easily to empirical testing, as real option values, and hence the extent of hysteresis, rise in the volatil-
ity of the underlying stream of returns. In the absence of data to test this hypothesis directly on
adoption data, we test for this possibility by including the volatility of commodity returns in our
wage-trend model. Defining volatility as the three-quarter moving standard deviation of wages, we
find that it does not have a statistically significant effect on wages (see empirical results in the online
Appendix).
9
Our findings may be due to differing wage trends across counties in our cross-county sample. To examine this possibility, we re-estimated our
preferred specification (Model 5) for only Santa Cruz County, focusing only on crops grown in this county: strawberries, apples, and other
berries. Our new estimates find that the qualitative difference between the commodity-wage trends is the same as in Table 4, namely, wages
trended higher over our sample period for strawberries relative to apples (t = 8.213) and other berries relative to apples (t = 2.734), as in the
full sample.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1455
Second, economies of scale may be another valid explanation as berry farms may be relatively
small operations in which technology investments may simply be scale infeasible. However, the aver-
age farm is approximately 54.0 acres and hires roughly thirty-seven workers (Census of
Agriculture, 2017), but there are several farms that are as large as the farm we worked with in this
article (i.e., in the order of 1000 acres and 600–1000 workers per year, on average).10 At the same
time, the Harvest Pro machine is relatively scale-neutral as only a few workers (15) can operate the
machine at the same time, and the farm represented here operates about a dozen of them. More
importantly, however, farms in this industry have a rental option for their Harvest Pro machines
(our sample farm rents its machines). The fact that there is an active rental market suggests that nei-
ther lumpiness nor the irreversibility of technology investments matter. If a new machine were
invented, for example, the adoption decision would consist of simply terminating a lease agreement
and signing a new one.11
In general, therefore, our findings support our theoretical expectations that mechanical aids that
complement labor, rather than substitute for labor, are likely to be associated with higher wages than
otherwise. Under competitive labor markets, higher implied wages for harvest workers in industries
where innovations in harvesting technology complement labor, in turn, should help alleviate farm
supply shortages. Under labor market oligopsony, however, higher wage trends are associated with
undercapitalization when mechanized harvesting aids complement harvest labor, disrupting this vir-
tuous cycle. Although studying this latter effect is a matter for future research, and a different data
source, the fact that time-series data on adoption of labor-augmenting machines simply does not
exist provides indirect evidence that the effect is present. That is, strawberry-harvesting technology
has been present for many years, but adoption is still incomplete and sporadic.12
Our findings are important to both the policy discussion on labor shortages and the potential for
mechanization to provide a viable solution. Although on the surface it would seem that mechaniza-
tion of the sort we describe here should help alleviate labor shortages and would be the expected
response if induced innovation were the rule of the day (Hayami & Ruttan, 1970), our findings tem-
per this conclusion. Rather, if mechanization allows wages to rise and leads to undercapitalization,
then the problem of over relying on labor inputs is likely to persist. More generally, our findings sup-
port the findings from the macroeconomics literature on why there are persistent productivity gaps
in agriculture (Gollin et al., 2014; Herrendorf & Schoellman, 2015). If more industries mechanize in
a manner similar to the strawberry industry we describe here, and undercapitalization follows, then
lower productivity is likely to persist.
10
At a land-valuation of roughly $50,000 per acre, the average farm represents a substantial capital investment (UC Davis, 2016).
11
We thank a reviewer for suggesting these other potential explanations.
12
Clemens et al. (2018) point out that if farmers anticipate labor shortages, such as with the termination of the Bracero program in 1964, their
substitution of capital for labor can actually lead to zero wage effects, despite a sharp reduction in labor supply. However, their theoretical
model is based on a perfectly competitive labor-market assumption, and assumes that technology is available for all industries. On this point,
they note that “[F]or crops that lack a feasible advanced technology…output falls and wages rise (p. 1474),” which is precisely our case. In fact,
if the technology is not available, then capital literally cannot substitute for labor, and incremental “process” improvements like the one we
describe manifest as complementary with labor. That is, if capital cannot do away with labor, then it has to settle for making each unit of labor
more productive.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1456 FARM LABOR PRODUCTIVITY
mechanical-investments on a micro (farm-level) scale and then empirically examine the implications
for labor-market outcomes using more aggregate data.
We examine the question of whether mechanization complements or substitutes for labor using
a highly detailed data set of daily-production records from a large strawberry farm in the Central
Coast of California. Controlling carefully for the effect of peer-group influence on productivity, we
find that mechanization complements labor inputs, raising the productivity of each employee. Our
test for relative wage trends in mechanized versus non-mechanized industries shows that wages have
trended significantly higher in mechanized industries in which the nature of the machines that are
used complement, rather than substitute for, labor. Our findings support the implications of our the-
oretical model and help explain findings from the macroeconomics literature that document a per-
sistent productivity gap between agriculture and other, non-agricultural industries.
Our findings are relevant to the literature on agricultural labor shortages (Charlton &
Taylor, 2016; Richards, 2020), as authors often suggest that investments in mechanization form at
least part of a solution to the “access to labor” problem. Although adopting mechanized aids of the
sort we describe here and increasing the productivity of existing employees allows farmers to make
better use of a limited pool of labor, our findings suggest that higher wages are likely to depress
capital investment on farms. If farmers are less willing to invest in capital that ultimately leads to
higher wages, then long term productivity growth is likely to be lower, and the problem will persist.
Our findings suggest at least one important policy implication. Namely, we emphasize the
general point in Fuglie et al. (2019) that publicly funded R & D may provide a way out of the under-
investment trap we identify. That is, if private-sector firms do not perceive a demand for
productivity-enhancing harvesting innovations, then they will not be forthcoming from industry.
However, if the dynamic we describe truly does arise from a market imperfection (labor-market
buying power), then there is a valid role for government research agencies to step in and seek a tech-
nology solution that is not only feasible (a difficult matter in fruit-harvesting), but sufficiently cost
effective to be adopted by growers once it is perfected.
Our research suggests other opportunities for future research. For example, the mobility of
workers between industries is likely to affect the elasticity of labor-supply facing each farm and the
degree of oligopsony power exercised by representative growers in the strawberry industry. Although
the skills of the workers in our data are portable between different crops, if workers could not move
from crop to crop, then investments in harvest technology would be more likely to be labor reducing
than wage increasing as in our case.
Our study is not without weaknesses. As we explain above, our findings are conditional on our
strawberry farm being representative of others in the industry. Although we have no reason to
believe that this is not the case, other growers may manage their workers, and their technology, dif-
ferently from ours. Second, our aggregate wage analysis relies on the comparability of wages in the
strawberry industry and related berry industries with those in the apple and grape industries. To the
extent that our counterfactuals are not accurate, our aggregate wage-trend comparisons will not be
valid. That said, our conversations with officials from all industries involved suggest that our
assumptions are at least approximately accurate. Third, there are always empirical issues that
confound the estimation of peer effects on productivity in team-based environments (Angrist, 2014).
Although we carefully control for the sources of bias that are typically addressed in the literature,
other forms of bias nevertheless may exist that confound our estimates of the marginal productivity
of mechanization. Fourth, we do not have data on mechanized-harvesting adoption rates for the
strawberry, or any other, industry. Detailed data on the extent of mechanization would permit a
direct test of our model and would allow for much more, and more insightful, research in this area.
ACKNOWLEDGMENTS
We gratefully acknowledge funding from the USDA NIFA Grant 2019-67023-29415. This research
was pre-registered at Social Science Registry, Trial AEARCTR-0006532
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1457
R EF E R EN C E S
Alvarez-Cuadrado, Francisco, and Ngo Van Long. 2017. “Capital–Labor Substitution, Structural Change, and Growth.”
Theoretical Economics 12: 1229–66.
Angrist, Joshua D. 2014. “The Perils of Peer Effects.” Labour Economics 30: 98–108.
Bandiera, Oriana, Iwan Barankay, and Imran Rasul. 2005. “Social Preferences and the Response to Incentives: Evidence from
Personnel Data.” Quarterly Journal of Economics 120: 917–62.
Bandiera, Oriana, Iwan Barankay, and Imran Rasul. 2007. “Incentives for Managers and Inequality among Workers: Evidence
from a Firm-Level Experiment.” Quarterly Journal of Economics 122: 729–73.
Bandiera, Oriana, Iwan Barankay, and Imran Rasul. 2010. “Social Incentives in the Workplace.” Review of Economic Studies
77: 417–58.
Barattieri, Alessandro, Susanto Basu, and Peter Gottschalk. 2014. “Some Evidence on the Importance of Sticky Wages.”
American Economic Journal: Macroeconomics 6: 70–101.
Benyishay, Ariel, and A. Mushfiq Mobarak. 2019. “Social Learning and Incentives for Experimentation and Communication.”
Review of Economic Studies 86: 976–1009.
Bhaskar, Venkataraman, Alan Manning, and Ted To. 2002. “Oligopsony and monopsonistic competition in labor markets.”
Journal of Economic Perspectives 16(2): 155–74.
Billikopf, Gregorio 2008. “Piece Rate Pay Design.” https://nature.berkeley.edu/ucce50/ag-labor/7research/7calag06.htm.
Bramoullé, Yann, Habiba Djebbari, and Bernard Fortin. 2009. “Identification of Peer Effects through Social Networks.”
Journal of Econometrics 150: 41–55.
Bursztyn, Leonardo, Florian Ederer, Bruno Ferman, and Noam Yuchtman. 2014. “Understanding Mechanisms Underlying
Peer Effects: Evidence from a Field Experiment on Financial Decisions.” Econometrica 82: 1273–301.
Census of Agriculture. 2017. “The U.S. Department of Agriculture’s National Agricultural Statistics Service (NASS).”
Volume 1, Geographic Area Series, Part 51. https://www.nass.usda.gov/Publications/AgCensus/2017/Full_Report/
Volume_1,_Chapter_1_US/usv1.pdf
California Farm Bureau Federation 2017. “Survey: California Farms Face Continuing Employee Shortages.” https://www.cfbf.
com/news/survey-california-farms-face-continuing-employee-shortages/. Accessed September 24, 2020.
Chang, Tom, and Tal Gross. 2014. “How Many Pears Would a Pear Packer Pack if a Pear Packer Could Pack Pears at Quasi-
Exogenously Varying Piece Rates?” Journal of Economic Behavior and Organization 99: 1–17.
Chang, Tom, Joshua Graff Zivin, Tal Gross, and Matthew Neidell. 2016. “Particulate Pollution and the Productivity of Pear
Packers.” American Economic Journal: Economic Policy 8: 141–69.
Charlton, Diane, and J. Edward Taylor. 2016. “A Declining Farm Workforce: Analysis of Panel Data from Rural Mexico.”
American Journal of Agricultural Economics 98: 1158–80.
Clemens, Michael A., Ethan G. Lewis, and Hannah M. Postel. 2018. “Immigration Restrictions as Active Labor Market Policy:
Evidence from the Mexican Bracero Exclusion.” American Economic Review 108: 1468–87.
Cohn, Alain, Ernst Fehr, and Lorenz Goette. 2014. “Fair Wages and Effort Provision: Combining Evidence from a Choice
Experiment and a Field Experiment.” Management Science 61: 1777–94.
Conley, Timothy, and Udry Christopher. 2001. “Social Learning through Networks: The Adoption of New Agricultural
Technologies in Ghana.” American Journal of Agricultural Economics 83: 668–73.
Cornelissen, Thomas, Christian Dustmann, and Uta Schönberg. 2017. “Peer Effects in the Workplace.” American Economic
Review 107: 425–56.
De Giorgi, Giacomo, Michele Pellizzari, and Silvia Redaelli. 2010. “Identification of Social Interactions through Partially
Overlapping Peer Groups.” American Economic Journal: Applied Economics 2: 241–75.
Dey, M. S., and Flinn, C. J. (2005). An equilibrium model of health insurance provision and wage determination. Econometrica
73(2): 571–627.
Dickens, Richard, Stephen Machin, Alan Manning, David Metcalf, Jonathan Wadsworth, and Stephen Woodland. 1995.
“The Effect of Minimum Wages on UK Agriculture.” Journal of Agricultural Economics 46: 1–19.
Dixit, Avinash. 1992. “Investment and Hysteresis.” Journal of Economic Perspectives 6(1): 107–32.
Feigenberg, Benjamin, Erica Field, and Rohini Pande. 2013. “The Economic Returns to Social Interaction: Experimental
Evidence from Microfinance.” Review of Economic Studies 80: 1459–83.
Foster, Andrew D., and Mark R. Rosenzweig. 1995. “Learning by Doing and Learning from Others: Human Capital and
Technical Change in Agriculture.” Journal of Political Economy 103: 1176–209.
Fuglie, Keith, Madhur Gautam, Aparajita Goyal, and William F. Maloney, eds. 2019. Harvesting Prosperity: Technology and
Productivity Growth in Agriculture. Washington, DC: World Bank Group, Chap. Sources of Growth in Agriculture.
Gallardo, R. Karina, and Johannes Sauer. 2018. “Adoption of Labor-Saving Technologies in Agriculture.” Annual Review of
Resource Economics 10: 185–206.
Glenn Dutcher, E., Regine Oexl, Dmitry Ryvkin, and Tim Salmon. 2019. Do Competitive Incentives in Team Production Lead
to More Effort or More Sabotage?. 1–60. Athens, OH: Department of Economics, Ohio University.
Gollin, Douglas, David Lagakos, and Michael E. Waugh. 2014. “Agricultural Productivity Differences across Countries.”
American Economics Review 104: 165–70.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1458 FARM LABOR PRODUCTIVITY
Graff Zivin, Joshua, and Matthew Neidell. 2012. “The Impact of Pollution on Worker Productivity.” American Economic
Review 102: 3652–73.
Gray, Blake 2020. “Vineyard Mechanization: Quality at a Distance.” https://www.winebusiness.com/news/?go=getArticle&
dataId=230585.
Guryan, Jonathan, Kory Kroft, and Matthew J. Notowidigdo. 2009. “Peer Effects in the Workplace: Evidence from Random
Groupings in Professional Golf Tournaments.” American Economic Journal: Applied Economics 1: 34–68.
Hayami, Yujiro, and Vernon W. Ruttan. 1970. “Factor Prices and Technical Change in Agricultural Development: The
United States and Japan, 1880–1960.” Journal of Political Economy 78: 1115–41.
Herrendorf, Berthold, and Todd Schoellman. 2015. “Why Is Measured Productivity So Low in Agriculture?” Review of
Economic Dynamics 18: 1003–22.
Hill, Alexandra E. 2018. “The Minimum Wage and Productivity: A Case Study of California Strawberry Pickers.”
Unpublished.
Hill, Alexandra E., and Jesse Burkhardt. 2021. “Peers in the Field: The Role of Ability and Gender in Peer Effects among
Agricultural Workers.” American Journal of Agricultural Economics 103: 790–811.
Just, Richard E., and Wen S. Chern. 1980. “Tomatoes, Technology, and Oligopsony.” Bell Journal of Economics 11: 584–602.
Kislev, Yoav, and Willis Peterson. 1981. “Induced Innovations and Farm Mechanization.” American Journal of Agricultural
Economics 63: 562–5.
Kondylis, Florence, Valerie Mueller, and Jessica Zhu. 2017. “Seeing Is Believing? Evidence from an Extension Network
Experiment.” Journal of Development Economics 125: 1–20.
Manski, Charles F. 1993. “Identification of Endogenous Social Effects: The Reflection Problem.” The Review of Economic
Studies 60: 531–42.
Marmaros, David, and Bruce Sacerdote. 2002. “Peer and Social Networks in Job Search.” European Economic Review 46:
870–9.
Mas, Alexandre, and Enrico Moretti. 2009. “Peers at Work.” American Economic Review 99: 112–45.
Mercier, Stephanie. 2014. “Employing Agriculture: How the Midwest Farm and Food Sector Relies On Immigrant Labor.”
Chicago Council on Global Affairs, December.
Murphy, Francis X. 2019. “Does Increased Exposure to Peers with Adverse Characteristics Reduce Workplace Performance?
Evidence from a Natural Experiment in the US Army.” Journal of Labor Economics 37: 435–66.
Paarsch, Harry J., and Bruce Shearer. 2000. “Piece Rates, Fixed Wages, and Incentive Effects: Statistical Evidence from Payroll
Records.” International Economic Review 41: 59–92.
Park, Sangyoon. 2019. “Socializing at Work: Evidence from a Field Experiment with Manufacturing Workers.” American Eco-
nomic Journal: Applied Economics 11: 424–55.
QCEW. 2021. “Quarterly Census of Employment and Wages, Bureau of Labor Statistics.” Bureau of Labor Statistics. https://
www.bls.gov/cew/
Richards, Timothy J. 2018. “Immigration Reform and Farm Labor Markets.” American Journal of Agricultural Economics 100:
1050–71.
Richards, Timothy J. 2020. “Income Targeting and Farm Labor Supply.” American Journal of Agricultural Economics 102:
419–38.
Richards, Timothy J., and Gareth P. Green. 2003. “Economic Hysteresis in Variety Selection.” Journal of Agricultural &
Applied Economics 35: 1–14. https://doi.org/10.1017/S1074070800005897
Rogers, Richard T., and Richard J. Sexton. 1994. “Oligopsony and Monopsonistic Competition in Labor Markets.” American
Journal of Agricultural Economics 76: 1143–50.
Shearer, Bruce. 2004. “Piece Rates, Fixed Wages and Incentives: Evidence from a Field Experiment.” Review of Economic
Studies 71: 513–34.
Shi, Lan. 2010. “Incentive Effect of Piece Rate Contracts: Evidence from Two Small Field Experiments.” BE Journal of
Economic Analysis & Policy 10(1).https://www.degruyter.com/document/doi/10.2202/1935-1682.2539/html
Staiger, Douglas O., Joanne Spetz, and Ciaran S. Phibbs. 2010. “Is There Monopsony in the Labor Market? Evidence from a
Natural Experiment.” Journal of Labor Economics 28: 11–236.
Stevens, Andrew. 2018. “Temperature, Wages, and Agricultural Labor Productivity.” Working Paper, Department of Agricul-
tural and Resource Economics, University of California, Davis, Berkeley.
Tan, Tom Fangyun, and Serguei Netessine. 2019. “When You Work with a Superman, Will You Also Fly? An Empirical Study
of the Impact of Coworkers on Performance.” Management Science 65: 3495–517.
UC Davis. 2016. “Sample Costs to Produce and Harvest Strawberries.” Working Paper. Davis, CA: University of California at
Davis.
Vasilaky, Kathryn N., and Kenneth L. Leonard. 2018. “As Good as the Networks they Keep? Improving Outcomes through
Weak Ties in Rural Uganda.” Economic Development and Cultural Change 66: 755–92.
Zhang, Zhao, Paul H. Heinemann, Jude Liu, Tara A. Baugher, and James R. Schupp. 2016. “The Development of Mechanical
Apple Harvesting Technology: A Review.” Transactions of the American Society of Agricultural and Biological Engineers
59: 1165–80.
14678276, 2022, 4, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ajae.12273 by INASP/HINARI - PAKISTAN, Wiley Online Library on [20/09/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
HAMILTON ET AL. 1459
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the pub-
lisher’s website.
How to cite this article: Hamilton, Stephen F., Timothy J. Richards, Aric P. Shafran, Kathryn
N. Vasilaky. 2022. “Farm labor productivity and the impact of mechanization.” American
Journal of Agricultural Economics 104(4): 1435–1459. https://doi.org/10.1111/ajae.12273