Synergy and Learning Effects of Informal Labor-Sharing Arrangements

341KB Sizes 0 Downloads 49 Views

Report

PDF Reader
Full Text

World Development Vol. 99, pp. 1–14, 2017 0305-750X/Ó 2017 Elsevier Ltd. All rights reserved. www.elsevier.com/locate/worlddev

http://dx.doi.org/10.1016/j.worlddev.2017.06.004

Synergy and Learning Eﬀects of Informal Labor-Sharing Arrangements DAWIT K. MEKONNEN a and JEFFREY H. DORFMAN b,* a International Food Policy Research Institute (IFPRI), Washington, DC, USA b University of Georgia, Athens, USA Summary. — We study the eﬀects of informal labor-sharing arrangements and other social interactions on farmers’ productivity in a developing country context, testing whether these types of social and work interactions lead to productivity gains through learning, synergy, or both. Using a rich panel data set of Ethiopian subsistence farmers, we estimate a distance function of grains production and ﬁnd large productivity gains (approximately 33% and 29% in 1999 and 2004) from labor sharing due to synergy eﬀects that boost labor productivity. However, labor sharing does not lead to learning as the productivity gains observed in years with labor sharing disappear in following years if the farmers do not continue to engage in labor sharing. Labor-sharing partners are either neighbors, relatives, members of the same funeral and religious associations, or have plots next to each other, which together reduce labor sharing as a single venue for learning. However, the synergy eﬀect is strong enough to warrant the design of extension and outreach policies that recognize and utilize farmers’ informal social networks such as labor-sharing arrangements. Ó 2017 Elsevier Ltd. All rights reserved. Key words — distance function, eﬃciency, labor exchange, labor sharing, learning, social networks, synergy

1. INTRODUCTION

In this study, we focus on an oft-neglected aspect of social networks that can have direct implications on farmers’ productivity—the synergy eﬀect from informal labor-sharing arrangements. The synergy eﬀect refers to productivity gains that come from working together such as speed gains and being less bored by tedious agricultural activities or working harder while observed by the labor-sharing partners. In these arrangements, a household head invites members of other households in his network to help him with speciﬁc agricultural activities. Labor sharing is used in a wide range of agricultural activities including land preparation and plowing, weeding, harvesting, and threshing, providing opportunity for labor-sharing partners to inﬂuence productivity in all stages of production. Labor-sharing use for land preparation and plowing can make required labor available for on-time sowing of crops, while its use for weeding means pest and weeds that can aﬀect crop productivity are checked early on. Labor-sharing use during harvest means quick completion of harvest, which could save a lot of harvest loss, particularly in seasons where untimely rain during harvest periods can destroy crops on the ﬁeld. Consistent with this synergy eﬀects

The inﬂuence of social networks on individuals’ behavior and success has long been of interest to sociologists, and this interest has recently been picked up by economists. The social network studies in agriculture have mainly focused on the impact of networks on technology adoption and diffusion (Genius, Koundouri, Nauges, & Tzouvelekas, 2014; Krishnan & Patnam, 2014; Liverpool-Tasie & WinterNelson, 2012; Maertens & Barrett, 2012; Falco & Bulte, 2013; Conley & Udry, 2010; Bandiera & Rasul, 2006; Munshi, 2004; Foster & Rosenzweig, 1995) and risk sharing (Fafchamps & Lund, 2003; Dercon & Krishnan, 2003). In addition, Santos and Barrett (2010) applied social network theories on the role of identity in farmers’ search for information while Krishnan and Sciubba (2009) analyzed the role of the number of links and network architecture in determining the impact of social networks on outcomes. However, what type of networks facilitate learning and the context in which they do that is still an active area of enquiry. Krishnan and Patnam (2014), for instance, found that neighbors are important sources of learning for adoption of agricultural activities in Ethiopia, more so than extension agents whose eﬀects fade away over time. Songsermsawas, Baylis, Chhatre, and Michelson (2016), on the other hand, found that 60% of farmers’ revenue is explained by peers, but the peer eﬀects are signiﬁcant among farmers’ self-reported peers, especially among those peers who are farmers’ main advisors for agricultural matters, rather than geographically deﬁned neighbors. The mechanisms through which social learning aﬀect technology adoption can have direct implications for the design of agricultural extension and training programs. In many developing countries contact or progressive farmers, who serve as points of contact between extension agents and other farmers, are ubiquitously used as messengers of information (Kondylis, Mueller, & Zhu, 2017). An extension system based on contact farmers presumes that these contact farmers will inﬂuence other farmers in their networks to follow their lead and adopt new production practices.

* We would like to thank Drs. Berna Karali, Cesar Escalante, Esendugue Greg Fonsah, Nicholas Magnan, Yared Seid, and Beliyou Haile for their reviews and comments at diﬀferent stages of the paper. We are grateful to Yisehac Yohannes for generously responding to our data-related queries. We also would like to thank the three anonymous referees and the editor of the journal for their helpful and constructive comments. However, any and all errors are our responsibility. The data for this study have been made available by the Economics Department at Addis Ababa University, the Center for the Study of African Economies at University of Oxford, and the International Food Policy Research Institute. Funding for data collection was provided by the Economic and Social Research Council (ESRC), the Swedish International Development Agency (SIDA), and the United States Agency for International Development (USAID); the preparation of the public release version of these data was supported, in part, by the World Bank. AAU, CSAE, IFPRI, ESRC, SIDA, USAID, and the World Bank are not responsible for any errors in these data or for their use or interpretation. Final revision accepted: June 25, 2017. 1

2

WORLD DEVELOPMENT

narrative, households in the data that we use for the study, report that they use labor-sharing parties for quick completion of tasks, due to unavailability or expensiveness of hired labor, and for completion of tedious agricultural activities in a group. Other households respond to labor-sharing requests by a farmer not based on wages but in expectation that the household will reciprocate the labor supply when they make a similar request later. In addition to the synergy eﬀect, the identiﬁcation strategy described below can also pick up learning eﬀects from previous season labor-sharing arrangements, to the extent that skills and technologies are varied among labor-sharing partners. However, any ﬁnding on lack of learning eﬀects may not necessarily imply that social networks are not important for learning. Rather, a ﬁnding of no learning eﬀects may be because labor sharing is often done with ‘‘like-minded” people on ﬁelds with crops that everyone grows and is used to grow, with techniques that do not change much, which does not provide the necessary context for learning to occur. The fact that labor-sharing partners are also relatives, neighbors, and belong to other social and religious associations can also reduce the role it can play as a single venue for learning. The objective of this article is to analyze the impact of these diﬀerent interactions and learning opportunities on agricultural productivity using a rich panel data set of Ethiopian subsistence grain farmers. We investigate to what extent involvement in informal labor-sharing arrangements aﬀects productivity above and beyond the direct impact of the additional labor to production. In other words, we want to know whether labor sharing means more than an increase in labor supply. This question will be answered aﬃrmatively if there are increasing returns to working together (synergy) or if labor-sharing facilitates mutual learning. We also examine whether social interactions such as funeral association membership and educational opportunities like extension programs and oﬀ-farm work lead to learning that increases agricultural productivity. We believe our article contributes to the growing literature on rural household networks in at least two ways. First, it expands the role of social networks beyond learning and diﬀusion of agricultural technologies by exploring the synergy eﬀects of labor-sharing arrangements that can aﬀect productivity and eﬃciency in a rural agricultural setting. Second, it investigates to what extent ordinary interactions with other farmers can boost productivity through the inﬂuence and leadership of some farmers, with implications for the design of production-increasing policies. If observation and interaction with an average farmer is not enough, but rather training and educational opportunities of new technologies and skills are necessary, then it clearly deﬁnes what role labor-sharing arrangements, as common as they are, should be expected to play in advancing productivity and eﬃciency in rural areas. This article should shed light on such policy questions and could point the way toward important agricultural production improvements. 2. SYNERGY VERSUS LEARNING EFFECTS After accounting for the direct impact of labor sharing in production in terms of increased labor supply, we hypothesize that informal labor-sharing arrangements aﬀect agricultural productivity and eﬃciency in two ways: the synergy eﬀect and the learning eﬀect. The synergy eﬀect is the result of the physical presence of the labor-sharing partners on the farmer’s plot and it refers to productivity gains that come from work-

ing together such as speed gains and being less bored by tedious agricultural activities or working harder while observed by the labor-sharing partners. The learning eﬀect is the skills learned and information obtained from the laborsharing partners that the household can put into use to improve its productivity and eﬃciency even on plots and at times when a labor party is not present. Labor sharing is expected to have an impact on farmers’ current level of technical eﬃciency based on the current and previous period labor-sharing status. Farmers are, therefore, grouped into four types as shown in Table 1. Type_I farmers are those who do not use labor sharing this year, and do not have prior labor-sharing experience. Type_II farmers refer to those who used labor sharing in the current year, and also have prior labor-sharing experience in at least one of the previous survey rounds. Type_III farmers are those who do not use labor sharing this year but have prior labor-sharing experience. Type_IV farmers are those who use labor sharing in the current year but do not have prior labor-sharing experience. The empirical application in this article uses the Ethiopian Rural Household Survey (ERHS) and though the econometric estimation is undertaken using the 1999 and 2004 survey rounds, we have used the 1994, 1995, 1997, 1999, and 2004 survey rounds to classify farmers based on their history of labor-sharing participation. Initially, we include labor-sharing participation in the season to see the total eﬀect of labor-sharing participation on farmers’ eﬃciency. The eﬀort to decompose this total eﬀect of labor sharing into synergy and learning eﬀects relies on the comparison of production eﬃciency among the four labor-sharing types in a subsequent estimation. The key to such decomposition is that synergy, as deﬁned above, requires labor-sharing participation in the current season, while learning from labor-sharing partners can happen in previous laborsharing uses even if the farmer does not participate in one in the current season. For instance, the learning eﬀect of labor sharing can be discerned if the technical eﬃciency of type_III farmers is greater than that of type_I farmers. This is because neither type_III nor type_I farmers use labor sharing in the current season. Thus, there will be no synergy eﬀect to consider and the only diﬀerence in the eﬃciency of type_I and type_III farmers should come from the previous periods’ participation of type_III farmers, which we called the learning eﬀect. Likewise, the synergy eﬀect of labor sharing can be discerned by comparing the production eﬃciency of type_II and type_III farmers. Both types of farmers have labor-sharing experiences prior to the current production season, and hence the opportunity to learn from their labor-sharing partners in the past, but type_II farmers use labor sharing in the current season as well with the potential to gain from the synergy eﬀect of labor sharing. Thus, if the eﬃciency of type_II farmers is greater than the eﬃciency of type_III farmers, that shows the presence of the synergy eﬀect. If the technical eﬃciency of type_II farmers is greater than that of type_III farmers, and the technical eﬃciency of type_III farmers, in turn, is greater than that of type_I farmers, then labor sharing has both learning and synergy eﬀects. This is because if there were only learning eﬀects, the technical eﬃciency of households who have used labor sharing both in the current season and in the past (type_II farmers) would be the same as those who did not use labor sharing that year but have used it in a previous season (type_III farmers). If there were only a synergy eﬀect, there would be no diﬀerence in the eﬃciency of type_III and type_I farmers because neither used labor sharing for that season. Using the same arguments to compare eﬃciency diﬀerentials between type_I, type_II,

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS

3

Table 1. Farmers classiﬁcation based on labor-sharing participation Survey year

Farmer type

Previous seasons (1994, 1995, 1997)

Current season (1999)

1999 1999 1999 1999

I II III IV

No Yes Yes No

No Yes No Yes

Survey year

Farmer type

Previous seasons (1994, 1995, 1997, 1999)

Current season (2004)

2004 2004 2004 2004

I II III IV

No Yes Yes No

No Yes No Yes

type_III, and type_IV farmers shows whether there is neither, either, or both learning and synergy eﬀects as summarized in Table 2. There are certain concerns in this identiﬁcation strategy. We have carefully addressed these concerns as follows. (a) Classiﬁcation of farmers based on their labor-sharing history The ﬁrst concern is on the classiﬁcation of the farmers based on their labor-sharing history with incomplete information on their labor-sharing status in between survey years (because the surveys we use are done in 1994, 1995, 1997, 1999, and 2004). However, such incomplete information of labor-sharing status in between the survey years does not signiﬁcantly change the assignment of farmers to the above labor-sharing types. For instance, the non-zero probability of labor-sharing use by all farmers in the years the survey was not undertaken does not change the deﬁnition of type_II and type_III farmers, which together constitute 60% and 70% of the study sample in 1999 and 2004. This is because the main characteristics of type_II farmers is that they have used labor sharing in the current season (which we know for a fact because every survey round solicits the information on whether the farmer used labor sharing in that season) and have used labor sharing in some of the previous rounds. If they use labor sharing in between survey rounds, that only strengthens their assignment to this group. In addition, the main characteristics of type_III farmers is that they have not used labor sharing in the current season but have prior labor-sharing experience, thus this grouping is also not aﬀected by farmers using labor sharing in between survey periods. This means that we can identify the synergy eﬀect fairly well by comparing the performance of type_II and type_III farmers. The possibility of labor-sharing participation by some type_I farmers in between survey periods is concerning because it may mean that some type_I farmers could actually belong to type_III. Given that we have observed the laborTable 2. Identiﬁcation of learning and synergy eﬀects Technical eﬃciency

Eﬀect of labor sharing

III > I II > III > I II > III = I II = III > I II = III = I

Learning Both learning and synergy Synergy but not learning Learning but not synergy Neither Learning nor synergy

I, II, III, and IV refer to the type of farmers as deﬁned in Table 1.

sharing participation of these farmers a number of times, we do not expect the number of farmers in type_I who could actually be type_III farmers to be signiﬁcant, if any. For the 1999 observations, for instance, previous labor-sharing experience refers to whether the farmer used labor sharing in 1994, 1995, and 1997, giving us the chance to observe them in three out of the ﬁve previous production seasons. In the 2004 survey around, we have a chance to observe their prior labor-sharing history for an additional one year. The same argument also goes to type_IV farmers. It is possible that some type_IV farmers may have used labor sharing in between survey years (making them more of type_III). However, as of 1999, for instance, they have not used labor sharing in four of the six production seasons that we have data for during 1994–99. In 2004, we have additional data point where they did not use labor sharing. Thus, we are not concerned that signiﬁcant number of farmers could be type_I or type_IV farmers while they should have been type_III. However, we cannot completely rule out the potential for error for some farmers in type_I or type_IV that may aﬀect the precision of our estimates. (b) Endogeneity of labor-sharing participation The second issue has to do with the potential endogeneity of labor-sharing participation as farmers who (continuously) use labor sharing could be well-connected farmers with high social capital to tap into, who can be reasonably expected to perform better even in the absence of labor-sharing participation. The potential endogeneity of labor-sharing participation is addressed using generalized method of moments estimation where labor sharing is instrumented with the ratio of male adults to total number of adults (male-ratio), whether the household head was born in the village, and whether the spouse of the household head was born in the village. The ERHS data set we use in the study show that 90% of laborsharing participants are male and 99% are adults between the ages of 10 and 70 (with an average age of a participant around 34 years old). Hence households with higher share of male adults are more likely to ﬁnd it easier to form or join a labor-sharing party. At the same time, the male ratio in the household (or sex ratio in general) is unlikely to depend on differences in sexual behavior as it does not have a one-to-one correspondence with family size. Whether the household head and spouse of the household head were born in the village, are likely to inﬂuence the household’s ability to reliably attract labor-sharing partners over time. Given that 55% of labor-sharing partners are relatives or know each other in many other ways, the fact that either

4

WORLD DEVELOPMENT

or both household heads were born in the village provides a bigger pool of people to make labor-sharing arrangements with (making these instruments relevant to the labor-sharing variable), while birth place is obviously not in one’s control (satisfying the exogeneity criteria for a good instrument). (c) Potential reverse-causality problem The third concern deals with potential reverse causality problem since farmers might have a certain expectation about yields, based on their assessment of the crop stand, and subsequently decide about labor sharing, posing a challenge on the direction of causality. This is particularly a problem if farmers are deciding to use labor sharing for activities that occur later in the season, when farmers have a good sense of expected productivity. The data provide an easy way out because we know who used labor sharing for what (land preparation, weeding, harvesting, and threshing). As a way of robustness check, we have used a new deﬁnition of labor-sharing participation where a household is not considered to participate in labor-sharing arrangements if he used it only for harvesting and threshing. This means that some farmers who were assigned as type_II will be re-assigned as type_III if they use labor sharing only for harvesting and threshing in the season. Similarly, some type_IV farmers will be re-assigned to type_I if what makes them a labor-sharing participant is their use of the arrangement only for harvesting and threshing in the season. According to this deﬁnition, the household needs to use labor sharing for land preparation or weeding to be considered as using labor sharing for the season. These activities happen early in the season before a bumper harvest increases the demand for a labor-sharing party or a poor harvest reduced the demand for one. Current season’s labor-sharing participation is deﬁned using this new deﬁnition, but previous seasons’ participation can be for any activity including harvesting and threshing. As discussed in the results section, the deﬁnition of participation in labor sharing with or without those who exclusively used it for harvesting and threshing give qualitatively similar results (on the presence and/or absence of synergy and learning eﬀects), though the magnitudes are lower under the more conservative deﬁnition where we exclude those who used labor sharing only for harvesting and threshing. (d) Contemporaneous learning versus learning from past laborsharing participation The fourth issue deals with potential learning that may happen within the same season of labor sharing, as is the case for the synergy eﬀect. The identiﬁcation of learning eﬀects by comparing type_III farmers (who have past labor-sharing experience but not current season) and type_I farmers (who have never used labor sharing) mainly refers to learning from past labor-sharing participation, for instance, learning on new skills or technologies after looking at the input uses and returns to those investments from past seasons’ production decisions of their labor-sharing partners. Though this lagged learning is consistent with the gradual nature of technology adoption and diﬀusion, it is possible for labor-sharing participation to have a contemporaneous learning eﬀect within the same season of labor sharing (for instance, informational exchanges on input prices, input sources, or meteorological forecasts). We concede that our identiﬁcation strategy may not diﬀerentiate synergy eﬀects (which also happen within the same season of labor-sharing use) from contemporaneous learning eﬀects, if any. We can still disentangle the combined

eﬀect of synergy and contemporaneous learning from the eﬀect of learning from past seasons’ labor-sharing participation. (e) Free-riding or shirking behavior The ﬁfth concern deals with the issue of potential free-riding or shirking behavior by some labor-sharing partners as some farmers may reduce their eﬀorts in work parties with the belief that others work at their maximum eﬀorts. This issue is not necessarily an identiﬁcation problem but begs for discussion of the synergy eﬀects in the context of potential free-riding behavior. The potential negative eﬀects of labor sharing due to free-riding and shirking behavior of labor-sharing partners are expected to be lower for households who use labor-sharing multiple times, and/or use labor sharing with similar people within and across seasons since repeated interactions can reveal shirking and provide an opportunity for the person who calls for the party to stop working with someone known for shirking behavior or known to be a very bad farmer or not a team player. Our identiﬁcation strategy, provides the opportunity to test for such negative impacts of labor-sharing arrangements. Both type_II and type_IV farmers have used labor sharing in the current season (hence can beneﬁt from synergy eﬀect and any contemporaneous learning or information exchange that happens within the current season). However, type_IV farmers, who use labor sharing only in the current season but not in previous survey rounds, have limited labor-sharing interactions compared to type_II farmers (who used labor sharing multiple time both in previous and current seasons). Type_II farmers have a chance in previous seasons to identify shirking behavior and positively sort labor-sharing partners so that they only call for partners without such shirking and freeriding problem. That is partly why, in 2004 for instance, 91% of the people invited to a labor-sharing party were invited before for the same purpose by the same household. Type_IV farmers do not have such an opportunity because they are using labor sharing for the ﬁrst time. Thus, if the eﬃciency of type_II is greater than type_IV farmers (or type_II farmers are more eﬃcient than type_IV farmers over type_I and type_III farmers), that indicates the presence of shirking and a free-riding problem (a speciﬁc form of transaction cost that we discuss in the theoretical framework), especially for those who use labor sharing for the ﬁrst time and cannot sort labor-sharing partners by their characteristics in a shared labor environment based on previous experience. 3. THEORETICAL FRAMEWORK The theoretical framework below draws on Gilligan’s (2004) model of labor exchange and is used to guide our empirical model. Consider a farmer that produces output, Q, using land, labor, fertilizer, oxen and other purchased inputs. Let us represent land, fertilizer, oxen, and other purchased inputs by the vector X to be able to focus our attention on labor (N). Farm labor, N, is comprised of family labor ðN f Þ, hired labor ðN h Þ, and shared labor ðN s Þ. Thus, the technological possibilities can be summarized using the transformation function Qmax ¼ f ðX ; N f þ N h þ N s Þ

ð1Þ

where f is non-negative, non-decreasing, and concave in its arguments. If we subtract from the transformation function a non-negative random variable, U, associated with technical

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS

ineﬃciency and add a stochastic term, V, for remaining randomness, then the actual production, Q, can be re-written as Q ¼ f ðX ; N f þ N h þ N s Þ U þ V :

ð2Þ

In addition to the direct impact through increased labor days, labor sharing can aﬀect production in two other important ways. The ﬁrst is by reducing the amount of family labor, N f , to be dedicated to own farms as farmers have to reciprocate on their partners’ plots. Thus, we redeﬁne family labor days as a function of shared labor. Secondly, labor sharing can inﬂuence farmers’ productivity and eﬃciency through its synergy and learning eﬀects. Thus, a farmer’s ineﬃciency, U , will be a function of whether the farmer is engaged in labor sharing along with a vector of other household and farm characteristics, K, that determine technical ineﬃciency. We, therefore, redeﬁne Q and U as follows: Q ¼ f ðX ; N f ðN s Þ þ N h þ N s Þ U þ V

ð3Þ

utility subject to the time constraint, the liquidity constraint, and the non-negativity constraints on the choice variables. argMax f ðX ; N f ðN s Þ þ N h þ N s Þ þ wo N o wN h

X ;R;N o ;N f ;N s ;N h

cf N s þ wðRÞ gðN s ; KÞ þ V

ð4Þ

ð8Þ

s:t: N F R N s ð1 þ cs Þ N o P 0 M þ wo N o P wN h þ cf N s R > 0; X > 0; N o P 0; N h P 0; N s P 0: Leisure, R, is assumed strictly positive because w0 ð0Þ is inﬁnite, implying that some leisure is always reserved as its marginal utility is inﬁnite at R = 0. The Lagrangian for this optimization problem becomes ‘R;N o ;X ;N h ;N s ;b;a ¼ f ðX ; N f ðN s Þ þ N h þ N s Þ þ wo N o wN h cf N s þ wðRÞ gðN s ; KÞ þ b½N F R N s ð1 þ cs Þ N o

and U ¼ gðN s ; KÞ:

5

þ a½M þ wo N o wN h cf N s :

ð9Þ

The Kuhn–Tucker ﬁrst-order conditions are as follows.

Though labor sharing doesn’t involve cash payments, there is a per-head search and transaction cost of organizing a labor-sharing party, cs . In addition, reciprocity reduces labor supply on one’s own farm by the amount of labor days on the partners’ farms, N s . If we represent the initial family labor endowment of the household by N F , leisure time by R, labor days spent on oﬀfarm activities by N o , then the household time constraint can be shown as

@‘ @N f ¼ f 0N s þ f 0N f : cf g0N s bð1 þ cs Þ acf ; @N s @N s @‘ N s P 0; N s ¼ 0: @N s

N F R N f N s N o cs N s P 0:

@‘ ¼ f 0R þ w0 ðRÞ b; @R

ð5Þ

The ﬁrst N s in Eq. (5) refers to labor days spent on others’ farms to reciprocate labor received from partners while the second N s is multiplied by cs to get the total search and transaction cost of organizing a labor-sharing party since cs is deﬁned in per-head terms. In addition to the time constraint, there is also a liquidity constraint as farmers have to pay for hired labor. Labor sharing enters the liquidity constraint because it is customary in many labor-sharing parties that the person calling for such a work party has to provide food and drink. The per-head cost of providing food and drink to the partners is given as cf . Farmers’ initial ﬁnancial endowment is represented by M, and their oﬀ-farm income by wo N o where wo and N o are the wage rate and labor days associated with oﬀ-farm activities. Farm income does not enter as a source of liquidity because it is not earned until the end of the season. If we deﬁne the wage rate per labor day for the amount of labor hired-in by w, then the liquidity constraint is given as M þ wo N o P wN h þ cf N s :

ð6Þ

We also follow Gilligan (2004) in assuming that households have identical preferences over income, y, and leisure, R, and that utility, W, is additive in income and the utility derived from leisure, WðY ; RÞ ¼ y þ wðRÞ 0

00

ð7Þ 0

where W > 0; W < 0, and w ð0Þ is inﬁnite. The assumption of constant marginal utility of income in Eq. (7) is reasonable for subsistence farmers who are the focus of this study. We have also assumed the price of agricultural production to be normalized to one. The farmer, then, maximizes his household

@‘ ¼ f 0N h wð1 þ aÞ; @N h

@‘ ¼ f 0X ; @X

X > 0;

N h P 0; R > 0;

X

R

Nh

@‘ ¼ 0: @N h

@‘ ¼ 0: @R

ð13Þ

N o P 0;

@‘ ¼ N f R N s ð1 þ cs Þ N o ; @b

ð11Þ ð12Þ

@‘ ¼ 0: @X

@‘ ¼ f 0N o þ wo ð1 þ aÞ b; @N o

ð10Þ

N0

b P 0;

@‘ ¼ 0: @N o

ð14Þ

@‘ ¼ 0: @b

ð15Þ

b

@‘ @‘ ¼ M þ wo N o wN h þ cf N s ; a P 0; a ¼ 0: ð16Þ @a @a If a farmer is involved in a labor-sharing party ðN s > 0Þ, the complementarity conditions of the Khun–Tucker ﬁrst-order @‘ ¼ 0. Thus conditions in Eq. (10) imply that @N s f 0N s g0N s ¼ ð1 þ aÞcf þ bð1 þ cs Þ f 0N f :

@N f @N s

ð17Þ

where f 0N s is the direct impact of labor sharing which we capture by including the amount of labor days from labor-sharing partners in the production function along with family and hired labor days, g0N s is the marginal impact of labor sharing @N on farmers’ technical ineﬃciency, and f 0N f : @Nfs is the marginal cost of reciprocating on the labor-sharing partners’ plots. cf and cs are both non-negative as they refer to per head costs of transaction and food provision while organizing labor sharing. b and a are also non-negative since they are Lagrange multipliers, which will be strictly positive if the time and liquidity constraints are binding. Thus, the left hand side of Eq. (17) can be considered as the marginal beneﬁt of

6

WORLD DEVELOPMENT

participation in labor-sharing arrangements in terms of the sum of increased production ðf 0N s Þ and increased eﬃciency ðg0N s Þ. The right hand side of Eq. (17) is the marginal cost of participation in labor-sharing arrangements in terms of the sum of marginal transaction costs of establishing or joining labor-sharing parties ðcs Þ, the cost of providing food to partners ðcf Þ, the reduced level of output due to reciprocity @N ðf 0N f : @Nfs Þ, and the extra pressure it creates on farmers’ liquidity and time constraints (a and b). Thus, farmers beneﬁt from participation in labor-sharing arrangements if the marginal beneﬁts from participation in terms of increased output and productivity is greater than its marginal cost. In terms of the diﬀerent labor-sharing types discussed in the preceding section, Eq. (17) implies that the marginal beneﬁt of laborsharing participation can be higher among farmers with more reliable and continued use of labor-sharing parties (such as type_II farmers) since they are likely to enjoy lower transaction cost of forming or joining labor-sharing parties and can manage to sort themselves into a group of more able and motivated farmers with higher synergy and/or learning eﬀects. 4. EMPIRICAL MODEL In the empirical application of the model, we employ a translog input distance function to estimate the technical eﬃciency of the small-scale farmers in Ethiopia. The distance function is chosen since it accommodates the multiple-input multiple-output nature of production in the farming system under consideration without the need for monetary aggregation of outputs (which ignores the technical interdependence of the range of crops farmers cultivate) or the need for using the dual cost approach which requires an assumption of cost-minimizing behavior that is unlikely to hold as agricultural outputs are endogenously determined.

tance function by deﬁnition is linearly homogenous in inputs (Nemoto & Furumatsu, 2014; O’Donnell & Coelli, 2005), which we have imposed in the model by dividing the righthand-side inputs by the numeraire input as shown below. Let X be a vector of inputs X ¼ ðx1 ; . . . ; xL Þ 2 RLþ and let Y be a vector of outputs denoted by Y ¼ ðy 1 ; . . . ; y M Þ 2 RM þ . The input distance function is deﬁned as X 2 LðY Þ ð18Þ DI ðY ; X Þ ¼ max h h where LðY Þ is the input set of production technology that describes the set of input vectors that are feasible for each output vector Y 2 RM þ (Kumbhakar & Lovell, 2000). The input distance function gives the maximum amount by which a producer’s input vector can be radially contracted and still remain feasible for the output vector it produces. Hence, DI ðX ; Y Þ ¼ h, takes a value greater than or equal to 1, where a value of 1 means that the farmer is operating at the lower boundary of the input requirement set. The input distance function is non-decreasing, concave and homogenous of degree one in inputs, and non-increasing and quasi-concave in outputs (O’Donnell & Coelli, 2005). An input distance function deﬁned over L inputs and M outputs takes the form DIit ¼ DI ðx1it ; . . . ; xLit ; y 1it ; . . . ; y Mit Þ

where i ¼ 1; . . . ; N represents farmers; t ¼ 1; . . . ; T denotes time; m ¼ 1; . . . ; r; . . . ; M are the diﬀerent crops produced by the farmer; and l ¼ 1; . . . ; p; . . . ; L are the applied inputs. Following Nemoto and Furumatsu (2014), we have chosen a hybrid translog Box–Cox model to represent the input distance function where we use logarithmic transformation of inputs and Box–Cox transformation of outputs as follows. lnDIit ¼ co þ

M X

M X M L X X cm y kmit þ :5 cmr y kmit y krit þ cl lnxlit

m

(a) Input distance function The choice between input versus output distance functions mainly depends on whether outputs or inputs can be perceived as exogenously given. The productivity literature suggests the use of input distance functions when output is given as exogenous and the use of output distance function when inputs are given as exogenous. As Kumbhakar (2011) shows in detail, in input distance functions where outputs are assumed exogenous, the ﬁrst-order conditions for proﬁt maximization is the same as those of cost minimization and the solution of input ratios from these ﬁrst-order conditions are exogenous (not aﬀected by ineﬃciency). Similarly, if inputs are exogenously given, output ratios for a proﬁt-maximizing (revenue maximizing) ﬁrm will be exogenous, making all the regressors in an output distance function model exogenous. However, in our speciﬁc application of a distance function among small-scale producers in Ethiopia, both production inputs and the type of crops produced are choice variables under farmers’ decision, while the amount of production is a function of observed inputs and unobserved household-speciﬁc eﬀects. Our choice to use an input distance function, therefore, is informed not so much because we think output is exogenous but because we believe we have a more convincing set of instruments to address the endogeneity of outputs in the generalized method of moments (GMM) estimation that we are following. Translog input distance functions, in particular, lead to consistent estimates of the regressors if the underlying technology is linearly homogenous (Kumbhakar, 2011). The input dis-

ð19Þ

m

r

l

L X L M X L K X X X þ :5 cpl lnxpit lnxlit þ cml lnxlit y kmit þ cc C K ð20Þ p

where

y kmit

m

l

m

l

is the Box–Cox transformations of outputs, deﬁned y k 1

by Box and Cox (1964) as y kmit ¼ mitk and k is the transformation parameter to be estimated. C K in Eq. (20) refers to K location dummy variables. The hybrid Box–Cox transformation is continuous around zero and hence allows us to include output variables with zero values for which log transformation is not possible. This is an important feature of the model because it is likely that farmers produce only some of the crops, resulting in a signiﬁcant share of zero values in the output variables. Linear homogeneity of the input distance function with respect to inputs implies that DIit ðY ; wX Þ ¼ wDIit ðY ; X Þ 8 w > 0. A convenient way of imposing linear homogeneity is to divide all the inputs by a numeraire input, say x1 (Coelli & Perelman, 2000). lnðDIit =x1 Þ ¼ co þ

M M X M X X cm y kmit þ :5 cmr y kmit y krit m

þ

L1 X

m

r

L1 X L1 X cl ln xlit þ :5 cpl ln xpit ln xlit p

l

l

M X L1 K X X þ cml ln xlit y kmit þ cc C K m

l

m

ð21Þ

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS lit where xlit ¼ xx1it and the coeﬃcients in Eq. (21) are diﬀerent from (20) to note that they are normalized by c1 , the coeﬃcient of the input transferred to the left hand side. ln DIit is assumed to be an additive error term composed of a nonnegative half-normal random term ðuit Þ that accounts for possible technical ineﬃciency of the farm and a normal statistical noise ðvit Þ with zero mean that captures the noise in production. Noting that DIit P 1 and rearranging the terms gives,

ln x1 ¼ ½co þ

M M X M L1 X X X cm y kmit þ :5 cmr y kmit y krit þ cl ln xlit m

m

r

l

L1 X L1 M X L1 X X þ :5 cpl ln xpit ln xlit þ cml ln xlit y kmit p

m

l

K X þ cc C K uit þ vit

l

ð22Þ

m

The following symmetry restrictions are imposed on the distance function in Eq. (22). cmr ¼ crm 8 m; r; clp ¼ cpl 8 l; p;

and

ð23Þ

cml ¼ clm 8 l; m Factors that aﬀect the ineﬃciency of farmer i are incorporated in the model by deﬁning uit in terms of household- and farm-speciﬁc variables X uit ¼ bj Z ijt ð24Þ j

where Z ijt refers to j = 1,. . .,J diﬀerent variables for farmer i at time t that are believed to aﬀect the productivity and eﬃciency of a farmer. It is here in Eq. (24), to be estimated jointly with (22), that we estimate the impacts derived in the theoretical model, speciﬁcally many of the terms shown in Eq. (17). Atkinson, Cornwell, and Honerkamp (2003) have shown that the Generalized Method of Moments (GMM) can be used to address the possibility of endogeneity of either outputs or inputs with the composite error term inherent in distance functions. The GMM approach has an additional advantage as it doesn’t require distributional assumption on the error term. In the empirical estimation that follows, we adopt the GMM approach of Atkinson et al. (2003) aiming to accommodate the multiple output nature of production in the farming system under study through distance functions as well as address the endogeneity inherent in input distance function estimation by instrumenting the endogenous right-hand-side outputs. Following Atkinson et al. (2003) and Atkinson and Dorfman (2005), the non-negativity of the uit is imposed after estimation by adding and subtracting from the ﬁtted model ^ ut ¼ mini ð^uit Þ, which deﬁnes the frontier intercept. Letting b i ðY ; X ; tÞ denote the part of the ﬁtted distance function other D than the composite error term vit uit , adding and subtracting ^ ut yields b i ðY ; X ; tÞ þ ^vit ^ uit þ ^ ut ^ ut ln x1 ¼ D b ðY ; X ; tÞ þ ^vit ^ ¼D uit i

ð25Þ

b i ðY ; X ; tÞ ^ b ðY ; X ; tÞ ¼ D ut is the estimated frontier where D i distance function in period t and ^ uit ¼ ^ uit ^ ut P 0. Given these estimates, Atkinson and Dorfman (2005) showed that we can compute farmer i’s level of technical eﬃciency in period t, TEit :

TEit ¼ expð^ uit Þ

7

ð26Þ

^it guarantees that 0 < TEit 6 1. where the normalization of u (b) Descriptions of variables and instruments In the ﬁnal estimated model, the log of land under cultivation in the season is the input used as the dependent variable (ln x1 in Eq. (22)), while the remaining inputs on the right hand side of the equation, the ln xlit ’s, that are normalized by the numeraire input ðln x1 Þ include the amount of fertilizer used, amount of labor days in the season including family, hired, and shared labor, as well as number of livestock the household owns in which the diﬀerent types of livestock are converted into tropical livestock units (TLUs). The main reason for including livestock units in the distance function is to account for draft power as a factor of production since less than two percent of households in the Ethiopian Rural Household Survey own tractors and, instead, use draft animals for plowing, threshing, and transporting harvests. The set of indicator variables in C K include dummy variables for the region the farmer resides in. The output variables on the right hand side of Eq. (22), the y mit ’s, are production levels of teﬀ, wheat, maize, barley, and sorghum. Ineﬃciency eﬀects that describe the environment under which production takes place, the Z ijt ’s in Eq. (24), include informal labor-sharing arrangements such as whether the farmer has called for a work party on at least one of his plots, as well as whether he is involved in other informal social networks such as funeral associations (idir) and oﬀ-farm activities. Labor sharing is included in the estimation in terms of the four diﬀerent labor-sharing types to identify its synergy and learning eﬀects. This allows us to directly estimate the terms from Eq. (17) that come from the theoretical model. In addition, Z ijt includes other major sources of information and education such as whether the farmer has access to government extension services, the highest level of education among members of the household, the average slope and soil fertility of the farmer’s plots, as well as the household head’s age, gender, education, and soil conservation practices. Plotlevel variables, such as soil fertility and slope, are aggregated to the household level using the weighted average of the values of the variables across plots where plot size is used as weight. Unobserved individual heterogeneity can aﬀect some variables in ln xlit ; y mit , or Z ijt . However, our use of a distance function, the richness of our data set, along with the use of generalized method of moments (GMM) has allowed us to address the issues well. As discussed earlier in the empirical model, the input ratios ðln xlit Þ in an input distance function can be considered as exogenous (see Kumbhakar, 2011 for the detail). However, the levels of output which appear in the distance function are going to be aﬀected by unobserved heterogeneity. Our instruments for these output variables include the price of each crop- and weather-related variables that show whether there was enough rain on the farmer’s ﬁelds during the cropping season, whether the rain stopped on time, whether it rained during harvest time, whether there were ﬂood, wind, and disease infestations in the season, and information on the type of crop most aﬀected by such adverse weather conditions. We have allowed the interaction of the instruments among themselves as well as with the included exogenous variables as instruments of interaction terms that involve endogenous variables. In addition, some of the Z ijt variables can potentially be correlated with unobserved individual heterogeneity. Speciﬁcally, history of participation in labor-sharing use that deﬁnes

8

WORLD DEVELOPMENT

farmers into the four labor-sharing types, education of the household head and the maximum level of education in the family, membership in funeral associations (idir), oﬀ-farm income, soil conservation practices, and access to extension services can be considered as endogenous. As described earlier while discussing the potential endogeneity of labor-sharing participation in Section (2), labor sharing is instrumented with the ratio of male adults to total number of adults in the household, whether the household head was born in the village, and whether the spouse of the household head was born in the village. The membership status of the household head in funeral associations is instrumented with two variables that show whether the mother and/or father of the head of the household are/were a member of any idir. Whether parents of the head of the household head are/were members of any idir is conceivably independent of whether their son/daughter grow up to be part of such an association, but it is likely to inﬂuence his/her decision to be a member in one. The education of the household head and the maximum level education in the family are instrumented with whether there exists a primary school in the village in 1992, 1997, 1999, and 2004. Oﬀ-farm income is instrumented with the distance of the village from the nearest town and whether there were productive safety net-related public works (food-forwork or cash-for-work programs) in the village. Availability of productive safety net programs is also considered to be a good instrument for farmers’ soil conservation practices because the program focuses on rehabilitation of forest, grazing, and agricultural lands as well as construction of wells, ponds, dams, terraces, and roads. Access to extension services is instrumented with a variable from the community-level survey that shows whether there are extension agents assigned and residing in the peasant association. As shown below in the result’s section, these instruments passed the Sargan–Hansen or J-test of overidentiﬁcation. Though formal tests of weak identiﬁcation for a non-linearin-parameters GMM are not yet available in the econometric literature, our instruments have passed the pathologies that GMM estimators exhibit in the presence of weak identiﬁcation as suggested by Stock, Wright, and Yogo (2002). 5. DATA AND DESCRIPTIVE STATISTICS We use the Ethiopian Rural Household Survey (ERHS) data (Hoddinott & Yohannes, 2016), which is a longitudinal household data set covering households in 15 peasant associations in rural Ethiopia. Data collection started in 1989 while the decision to make the data panel was made in 1994. The survey was expanded and redesigned, and samples rerandomized to encompass 15 peasant associations across the country, yielding a sample of 1,477 households. A new round was conducted in late 1994, with further rounds in 1995, 1997, 1999, 2004, and 2009. These surveys have been supervised by the Economics Department at Addis Ababa University, the Center for the Study of African Economies (CSAE) at the University of Oxford, and the International Food Policy Research Institute (IFPRI). We focus on the ﬁve major cereals in Ethiopia: teﬀ (both white and black mixed), maize, wheat, barley, and sorghum. The choice of these crops is due to their critical importance in food security eﬀorts in Ethiopia and their high share in total land area harvested and amount of production. The data show that the ﬁve crops under consideration are also important crops, mimicking the national average. In our sample, 70% and 65% of an average household’s land were covered under

these ﬁve crops. Furthermore, we found that most of the variation in the average land size under these crops is across region/states rather than across villages within a region. Thus, the region dummy variable we have included in the estimation, can take care of diﬀerences in the importance of these crops across the diﬀerent areas. Ethiopia has two rainy seasons: the main (Meher) rains between June and September and the second small showers (Belg rains) between February and May. Our study focuses on the main rainy season. This helps to reduce the noise in the data as the agricultural production system in terms of crops in the ﬁeld, intensity of the rain, and utilization of inputs are markedly diﬀerent in the two seasons. In addition, the ﬁve cereals that are the focus of this study are mainly produced during the main rainy season. According to data from the Central Statistical Agency of Ethiopia, during 1995–2008, close to 99% of teﬀ production, 98% of wheat production, 90% of barley production, and 89% of maize production was done during the main rainy season (CSA, 2009). Further, some variables are only available for the main rainy season in the data set. For instance, the data do not contain family and hired labor for the belg season, the fertilizer data for this season do not report the crop on which it was applied, and variables related to major events during the belg season are highly aggregated while they are detailed for the main season. Labor sharing is a prevalent practice in rural Ethiopia despite diﬀerences from region to region. In our sample, 65% of households in 1999 and 47% in 2004 called for labor-sharing work parties on their farming plots. Despite being called by diﬀerent names in diﬀerent parts of the country, labor-sharing arrangements usually follow two types of structures: debo and wonfel, as they are called in the Amhara region. Wonfel refers to a labor-sharing group that works in rotation for each group member and reciprocity is within the same season while debo refers to a labor-sharing group in which reciprocity to members is upon demand either within the same season or in the future (Krishnan & Sciubba, 2009). In this article, we refer to a household head being engaged in labor sharing if he or she participates in either of the two labor-sharing arrangements. The 2004 survey round of the ERHS data set shows an 86% reciprocity in laborsharing parties. Of the total farmers who participated in labor sharing in 2004, 78% have either already reciprocated or will reciprocate in the same season and 8% will reciprocate in the future, while 14% said they do not have to reciprocate. Households use labor sharing for land preparation and plowing (50% in 1999 and 33% in 2004), weeding (66% in 1999 and 26% in 2004), and harvesting and threshing (85% in 1999 and 56% in 2004) (Table 3). There are some farmers who use labor sharing only for harvesting and threshing (21% in 1999 and 43% in 2004) but the majority also use it for plowing and weeding, providing opportunity for laborsharing partners to inﬂuence productivity in all stages of production. Labor-sharing use for land preparation and plowing can make required labor available for on-time sowing of crops, while its use for weeding means pest and weeds that can aﬀect crop productivity are checked early on. Laborsharing use during harvest means quick completion of harvest, which could save a lot of harvest loss, particularly in season’s where untimely rain during harvest periods can destroy crops on the ﬁeld. Reasons for calling such work parties in the 2004 survey round include quick completion of task (47.8%), group being the best way of completing the tasks (18.3%), only way to get large amount of labor (15.3%), and because it was customary for the task (10.2%). Based on farmers’ responses, labor

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS

9

Table 3. Activities for which labor-sharing arrangements were used in 1999 and 2004

Plowing and land preparation Weeding Harvesting and threshing Labor sharing only for harvesting and threshing

1999

2004

Total

50.1% 66.1% 84.7% 20.6%

33.0% 25.6% 56.0% 42.7%

43.1% 49.6% 73.0% 29.6%

661

457

1118

Observations Note: The same household can use labor sharing for multiple tasks.

market failure in terms of unavailability of paid labor accounts for only 2.6% of the reasons for farmers to engage in labor sharing. However, 5.4% of the farmers get into labor sharing partly because they cannot aﬀord paid labor. See Table 4 for details. Farmers who called for labor parties also reported that 57% of the partners they called for the work party are as good a farmer as they are while 27% are better farmers and 16% are worse. The fact that about 84% of the farmers they called for a work party are either as good a farmer or better than they are suggests an opportunity for potential productivity gains from learning or synergy by working with others. As shown in Table 5, there is a fair distribution of farmers among the diﬀerent types based on their labor-sharing experiences. The share of farmers with access to the public extension system has almost doubled in the ﬁve years during 1999–2004. More than three quarters of the household heads are male and less than 10% completed primary school. More than three quarters of the household heads are members of idir (funeral associations), and 24 to 38% of the farmers are engaged in oﬀ-farm activities in 1999 and 2004. All the output variables (teﬀ, wheat, barley, maize, and sorghum) and chemical fertilizers are measured in Kilograms. Average production is the highest for teﬀ and the lowest for sorghum but the averages would obviously be higher if we consider only the farmers that produce the speciﬁc crop. Average fertilizer use is 57 kg in 1999 and 42 kg in 2004. The highest level of education in the household is around four years on average and it is expected to capture intra-household schooling externality. The average age of a household head is about 50 years. Soil fertility is measured in a 1–3 scale where 1 refers to fertile, 2 medium fertile, and 3 infertile soil, and it is averaged among the diﬀerent plots of the farmer. Labor is measured in labor days and it includes family, hired, and shared labor. Thus, the direct impact of labor sharing on production in terms of increased labor supply has been

accounted for in the distance function. While computing labor days, we follow Storck, Emana, Adnew, Borowiccki, and W/ Hawariat (1991) to account for the physical hardship in crop production by giving adult men a weight of 1, adult women a weight of 0.8, and child labor a weight of 0.35. See Table 5 for further details. 6. RESULTS The model is estimated using heteroscedasticity and autocorrelation consistent iterated GMM. Using the Sargan–Hansen or J-test of overidentiﬁcation (Baum, Schaﬀer, & Stillman, 2003; Wooldridge, 2002), we fail to reject the validity of the over-identifying restrictions. The J-test resulted in a GMM criterion function value of 91.6 which has a v2 distribution of 83 degrees of freedom, which gives a p-value of 0.24. A rejection of this test would have cast doubt on the validity of our instruments. Other than the validity of instruments, the other pillar in GMM estimation is that the instruments are suﬃciently related to the endogenous variables. When instruments are weak, the orthogonality conditions hold even at non-optimal values of the estimated parameters when in fact they should hold or get close to zero only at the optimal values. That is why Stock et al. (2002) suggested that in non-linear GMM, the problem is better termed as the weak identiﬁcation problem than the weak instruments problem. The instruments we used in the estimation have passed the pathologies that GMM estimators exhibit in the presence of weak identiﬁcation as suggested by Stock et al. (2002), since a formal test of weak identiﬁcation for a non-linear-in-parameters GMM is not yet available in the econometric literature. For instance, twostep GMM estimators and iterated GMM point estimators can be quite diﬀerent and can yield quite diﬀerent conﬁdence sets in the presence of weak identiﬁcation. The two step

Table 4. Why labor sharing? Reasons for calling a work party Quick completion of task Group is best way of completing task Only way to get large amount of labor Customary for this task Cannot aﬀord paid labor No paid labor available Others

Frequency

Percent

Cumulative

961 368 308 206 109 53 6

47.8 18.3 15.3 10.2 5.4 2.6 0.3

47.8 66.1 81.4 91.6 97.0 99.6 100.0

Results are based on the 2004 survey only. The question was asked for each activity labor sharing was called for. Respondents can give more than one reason. Source: Authors’ computation from ERHS.

10

WORLD DEVELOPMENT Table 5. Descriptive statistics of variables used in estimation Variable

Type_II (1/0) Type_III (1/0) Type_IV (1/0) Type_I (1/0) Soil conservation (1/0) Head’s age (years) Male head (1/0) Head completed primary school (1/0) Highest years of schooling (all members) Oﬀ-farm income (1/0) Idir membership (1/0) Extension (1/0) Average Soil Fertility (1 (fertile) to 3 scale) Average slope of plots (1 (ﬂat) to 3 scale) Teﬀ (kg) Wheat (kg) Barley (kg) Maize (kg) Sorghum (kg) Land (Ha) Fertilizer (kg) Labor (labor days) Tropical Livestock Units (number) Observations

1999

2004

Mean

Standard deviation

Mean

Standard deviation

0.452 0.151 0.205 0.192 0.361 49.07 0.792 0.0344 3.767 0.238 0.757 0.0959 1.683 1.304 225.2 158.7 188.5 135.6 77.91 0.919 57.87 134.4 4.819

(0.498) (0.359) (0.404) (0.394) (0.481) (14.99) (0.406) (0.182) (3.243) (0.426) (0.429) (0.295) (0.664) (0.543) (481.0) (323.5) (392.1) (323.7) (209.3) (0.814) (78.92) (135.7) (3.895)

0.438 0.370 0.0473 0.144 0.569 51.05 0.751 0.0896 4.022 0.368 0.788 0.173 1.687 1.291 215.8 159.1 187.0 183.9 45.74 1.229 44.81 147.0 4.752

(0.496) (0.483) (0.212) (0.352) (0.496) (15.00) (0.433) (0.286) (3.458) (0.483) (0.409) (0.379) (0.690) (0.516) (476.8) (324.2) (390.1) (441.3) (259.4) (2.469) (85.68) (158.6) (4.331)

845

845

Note: Units of measurements of the variables in parenthesis. Source: Authors’ computation from ERHS.

Table 6. Ineﬃciency eﬀects Labor-sharing use

For all agricultural activities

Not exclusively for harvesting and threshing

Parameter

Estimate

Standard error

Estimate

Standard error

Type_II Type_IV Type_I Age of household head Male head Poor quality soil Conservation Extension Head completed primary school Household members’ highest education Steep plots Steeper plots Oﬀ-farm income Idir membership

0.26** 0.08 0.06 0.12* 0.16*** 0.08 0.14 0.11 0.58*** 0.04** 0.10** 0.04 0.26** 0.19

0.11 0.11 0.13 0.07 0.05 0.10 0.10 0.12 0.21 0.02 0.05 0.10 0.11 0.13

0.24** 0.04 0.05 0.09 0.17*** 0.10 0.18* 0.07 0.51** 0.04** 0.12 0.01 0.25** 0.17

0.10 0.12 0.11 0.06 0.05 0.10 0.10 0.12 0.21 0.02 0.05 0.10 0.11 0.13

Notes: ***p < 0:01, **p < 0:05, *p < 0:1. A negative coeﬃcient means that an increase in that variable lowers ineﬃciency.

GMM and the iterated GMM estimators are almost identical in our case, which diﬀer only after two digits for almost all the coeﬃcients. We have presented the variables that explain farmers’ ineﬃciency in Table 6 even though they were estimated simultaneously in one step with the full set of the distance function variables presented in the appendix (Table 9). As shown in Eq. (24), these variables are explaining farmers’ ineﬃciency and hence negative signs imply that technical eﬃciency

increases as the explanatory variable increases and positive signs are associated with eﬃciency reducing eﬀects. We have included dummy variables for type_I, type_II, and type_IV farmers in the estimation and used type_III as a base group against which the eﬃciency of the other types of farmers are to be compared. The results show that the eﬃciency of type_III farmers, who have participated in labor sharing in the past but not in the current season, is not diﬀerent from the eﬃciency of type_I

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS

farmers (those who have no labor-sharing history in the current season and in the past). Thus, there is no evidence for the presence of learning eﬀects from participating in labor sharing in previous seasons. Type_II farmers, on the other hand, are more technically eﬃcient than type_III farmers, indicating the presence of a synergy eﬀect from labor sharing. That is, labor sharing improves the productivity and eﬃciency of farmers but the source of this productivity gain is synergy from working together. These gains from synergy are not trivial, but amount to an approximate 33% in 1999 and 29% in 2004. As the theoretical section predicts, the eﬀect of labor sharing, if it exists, is stronger for type_II farmers with the opportunity to get labor sharing at appropriate time, with reduced search and transaction costs of employing labor sharing, and a higher chance of gathering or joining a more motivated group of partners that are less susceptible to free-riding and shirking behavior, because of their continued use of labor sharing. The results conﬁrm that type_II farmers are more eﬃcient than type_IV farmers despite both types having used labor sharing in the production season. Moreover, the eﬃciency differential between type_II and type_III is statistically signiﬁcant while the diﬀerence between type_IV and type_III is not, albeit having the expected negative sign. Redeﬁning labor-sharing participation in the season by excluding those who use labor sharing only for harvesting and threshing (after they get the chance to evaluate the crop stand on the ﬁeld) results in qualitatively similar results. That is, the eﬃciency of type_I and type_III are not signiﬁcantly different, while type_II are more eﬃcient than both type_III and type_IV (Table 6). This implies, as before, the presence of synergy eﬀect from current season’s labor-sharing use, no learning eﬀect from past seasons’ labor-sharing participation, and the synergy eﬀect accrues to those who use labor sharing more often. However, with this new deﬁnition, the synergy eﬀect is lower (26% in 1999 and 14% in 2004). A possible explanation for the lack of learning eﬀects is the limited possibility for labor-sharing partners to transfer new skills either because of a lack of heterogeneity among laborsharing partners or because they are related in a number of other ways that give them an opportunity to learn from each other in those venues, hence decreasing the importance of labor sharing as a venue for learning. The 2004 round of the ERHS survey shows that about 91% of the people invited to the labor sharing were invited before for the same purpose by the same household and the farmers who called for the work party have worked before for 86% of the people they invited as part of a working party. In addition, 55% of labor-sharing partners are relatives, 68% are neighbors, 69% belong to the same idir (a funeral association), 27% have plots

11

next to each other, 27% are partners in oxen sharing or similar arrangements, and another 27% are members of the same mahiber (a religious association). Though we did not ﬁnd cumulative learning eﬀects that aﬀect productivity at times when a labor party is not present, it is possible for the synergy eﬀect to contain contemporaneous learning such as information exchanges that do not transcend current harvest period to aﬀect future period productivity. Male-headed households are found to be more eﬃcient than female-headed households. This calls for better understanding of the constraints of female headed households. Farming experience as captured by the age of the head of the household is also eﬃciency enhancing. The results show evidence for intra-household schooling externality as households in which members’ years of schooling is higher are found to be more eﬃcient than less educated households. However, we ﬁnd an unexpected positive eﬀect of primary school completion of the household heads on ineﬃciency. Households who make soil and water conservation practices are found to be more eﬃcient. Farmers harvesting in steeply sloped plots are found to be less eﬃcient. In addition, farmers’ engagement in oﬀ-farm income, though beneﬁcial in terms of an alternative source of income, appears to come at the expense of reduced eﬃciency of agricultural production. This is expected if oﬀ-farm income reduces the use of farm inputs and incentives to invest in conservation (Holden, Shiferaw, & Pender, 2004) or competes with crop intensiﬁcation (Mathenge, Smale, & Tschirley, 2015) or reduces the use of family labor on the farm (Pfeiﬀer, LopezFeldman, & Taylor, 2009), relative to its positive eﬀect on other purchased inputs (Pfeiﬀer et al., 2009). Table 7 presents the input and output elasticities of the distance function. The elasticity of the input distance function with respect to any output is equal to the negative of the cost elasticity of that output (Irz & Thirtle, 2004). It is theoretically expected to be negative for all desirable outputs and, in absolute value, reﬂects the relative importance of each output. The elasticities of the distance function with respect to input quantities are equal to the cost shares and therefore reﬂect the relative importance of the inputs in the production process (Irz & Thirtle, 2004). As shown in Table 7, in both 1999 and 2004, fertilizer is the most important input in the production process in terms of its contribution to cost, followed by livestock units (as a proxy for mechanization). The other elasticities are not statistically signiﬁcantly diﬀerent from zero. The average technical eﬃciency of the sampled farmers is found to be 55.4% in 1999 and 48.3% in 2004 (Table 8). These ﬁgures imply that an average farmer requires almost twice as

Table 7. Average Elasticities of the input distance function Evaluated at 1999 values

Teﬀ Wheat Maize Barley Sorghum Fertilizer Labor Livestock

Evaluated at 2004 values

Estimate

Standard error

Estimate

Standard error

0.119 0.013 0.009 0.312 0.008 0.319*** 0.430 0.297***

0.362 0.390 0.355 0.414 0.340 0.342 0.766 0.255

0.050 0.018 0.151 0.232 0.010 0.387*** 0.135 0.228***

0.459 0.464 0.476 0.498 0.345 0.305 0.944 0.300

12

WORLD DEVELOPMENT Table 8. Mean technical eﬃciency scores Labor-sharing use

Full sample Type_II Type_III Type_IV Type_I

For all agricultural activities

Not exclusively for harvesting and threshing

1999

2004

Average annual eﬃciency change (%)

1999

2004

Average annual eﬃciency change (%)

55.4 64.0 48.1 51.5 44.8

48.3 55.0 42.8 46.9 42.7

2.64 2.67 2.59 2.51 2.30

55.9 63.2 50.3 51.2 47.9

51.5 55.2 48.4 49.8 49.1

2.47 2.63 2.23 2.12 0.29

Source: Authors’ computation.

much land as the most eﬃcient farmer to produce the same level of output. Given that average land holding in Ethiopia is less than one hectare and with little suitable land available for the expansion of crop cultivation, especially in the highlands (Taﬀesse, Dorosh, & Gemessa, 2012), the return on eﬃciency enhancing investments is expected to be quite high. Farmers’ average technical eﬃciency showed a decline during 1999–2004, by an average of 2.6 percentage points per year, calling for more work to be done to improve farmers’ eﬃciency. In 1999 and 2004, farmers using labor sharing in the current year with previous labor-sharing experience (type_II farmers) are found to be 16 and 13 percentage points more eﬃcient than those who do not use labor sharing in the current year but have used it in the past (type_III farmers). Thus, the synergy from labor sharing is not just statistically signiﬁcant, but economically signiﬁcant as well, able to boost output by approximately 33% in 1999 and 29% in 2004. Even when we exclude those who exclusively use labor sharing for harvesting and threshing activities (to account for potential upward bias for those who increased labor demand after a good crop stand on the ﬁeld), type_II farmers are 13 and 7 percentage points more eﬃcient than type_III farmers. 7. CONCLUSION Informal labor-sharing arrangements, other social interactions, and educational opportunities provide possible avenues for gains in farmers’ productivity. Using data from rural Ethiopia, we estimate a distance function of grains production to test for agricultural productivity gains from labor sharing, social club memberships, farming experience, education, gender, and exposure to agricultural extension programs. We exploited the richness of our data to construct a set of instruments to address potential endogeneity in a

number of decision variables including labor-sharing participation, access to extension, education, oﬀ-farm income, and soil conservation practices. We estimate the distance function using generalized method of moments (GMM) and identify the learning and synergy eﬀects from labor sharing by segmenting the farmers according to their history of participation in labor-sharing parties. We ﬁnd large gains (approximately 33 and 29% in 1999 and 2004) from labor sharing, with the gains due to synergy eﬀects that boost labor productivity. Soil and water conservation practices, farming experience, male household headship, and education levels of household members are found to improve farmers’ technical eﬃciency. However, previous seasons’ labor-sharing participation does not lead to learning as the productivity gains from labor sharing in earlier years disappear in subsequent years if the farmers do not continue to engage in labor sharing. Labor-sharing partners are either neighbors, relatives, members of the same funeral and religious associations, or have plots next to each other, which together reduce labor sharing as a single venue for learning. In other words, the absence of learning from labor sharing suggests that, at least in the context of our study, farmers already know most or all of what they might learn simply by observing other farmers. On the other hand, the synergy eﬀect (which at least theoretically may also include contemporaneous learning within a season such as information exchanges on prices, input sources, and meteorological forecasts), is strong enough to warrant the design of extension and outreach policies that recognize and utilize farmers’ informal social networks such as labor-sharing arrangements. For instance, labor sharing can be used as a platform for extension workers and input suppliers to target several farmers at a time. These results do not encourage policies based on passive learning, and provide some evidence in favor of hands-on training and need for introduction of new techniques and technologies, as well as improving overall education of household members.

REFERENCES Atkinson, S. E., Cornwell, C., & Honerkamp, O. (2003). Measuring and decomposing productivity change: Stochastic distance function estimation versus data envelopment analysis. Journal of Business and Economic Statistics, 21, 284–294, URL: http://proxy-remote.galib.uga. edu/login?url=http://search.ebscohost.com/login.aspx?direct=true& db=ecn&AN=0671460&site=ehost-live. Atkinson, S. E., & Dorfman, J. H. (2005). Bayesian measurement of productivity and eﬃciency in the presence of undesirable outputs: Crediting electric utilities for reducing air pollution. Journal of Econometrics, 126, 445–468. http://dx.doi.org/10.1016/j.jeconom.2004.05.009, URL: http://www.sciencedirect.com/science/article/B6VC0-4CPVPX6-1/2/7c36100f3304dad6e7bcafa1f9dadf14. Current developments in productivity and eﬃciency measurement.

Bandiera, O., & Rasul, I. (2006). Social networks and technology adoption in Northern Mozambique. Economic Journal, 116, 869–902, URL: http://proxy-remote.galib.uga.edu/login?url=http://search.ebscohost.com/login.aspx?direct=true&db=ecn&AN=0892738&site= ehost-live. Baum, C. F., Schaﬀer, M. E., & Stillman, S. (2003). Instrumental variables and gmm: Estimation and testing. The Stata Journal, 3, 1–31. Box, G. E. P., & Cox, D. R. (1964). An analysis of transformations. Journal of the Royal Statistical Society. Series B (Methodological), 26, 211–252, URL: http://www.jstor.org/stable/2984418. Coelli, T., & Perelman, S. (2000). Technical eﬃciency of european railways: A distance function approach. Applied Economics, 32, 1967–1976, URL: http://proxy-remote.galib.uga.edu/login?url=

SYNERGY AND LEARNING EFFECTS OF INFORMAL LABOR-SHARING ARRANGEMENTS http://search.ebscohost.com/login.aspx?direct=true&db=ecn&AN= 0588620&site=ehost-live. Conley, T. G., & Udry, C. R. (2010). Learning about a new technology: Pineapple in Ghana. American Economic Review, 100, 35–69, URL: http://proxy-remote.galib.uga.edu/login?url=http://search.ebscohost.com/login.aspx?direct=true&db=ecn&AN=1084826&site= ehost-live. CSA (2009). Statistical Abstracts and Statistical Bulletins. Various years. Technical Report Central Statistics Agency: Federal Democratic Republic of Ethiopia. Addis Ababa: CSA. Dercon, S., & Krishnan, P. (2003). Risk sharing and public transfers. The Economic Journal, 113, C86–C94, URL: http://www.jstor.org/stable/ 3590049. Fafchamps, M., & Lund, S. (2003). Risk-sharing networks in rural philippines. Journal of Development Economics, 71, 261–287, URL: http://proxy-remote.galib.uga.edu/login?url=http://search.ebscohost.com/login.aspx?direct=true&db=ecn&AN=0657371&site= ehost-live. Falco, S. D., & Bulte, E. (2013). The impact of kinship networks on the adoption of risk-mitigating strategies in Ethiopia. World Development, 43, 100–110. http://dx.doi.org/10.1016/j.worlddev.2012.10.011, URL: http://www.sciencedirect.com/science/article/pii/S0305750X12002550. Foster, A. D., & Rosenzweig, M. R. (1995). Learning by doing and learning from others: Human capital and technical change in agriculture. Journal of Political Economy, 103, 1176–1209, URL: http:// proxy-remote.galib.uga.edu/login?url=http://search.ebscohost.com/ login.aspx?direct=true&db=ecn&AN=0374154&site=ehost-live. Genius, M., Koundouri, P., Nauges, C., & Tzouvelekas, V. (2014). Information transmission in irrigation technology adoption and diﬀusion: Social learning, extension services, and spatial eﬀects. American Journal of Agricultural Economics, 96, 328–344. http://dx. doi.org/10.1093/ajae/aat054, URL: http://ajae.oxfordjournals. org/content/96/1/328.abstract. arXiv:http://ajae.oxfordjournals. org/content/96/1/328.full.pdf+html. Gilligan, D. O. (2004). The Economics of Agricultural Labor Exchange with Evidence from Indonesia (Ph.D. thesis). College Park: University of Maryland. Hoddinott, J., & Yohannes, Y. (2016). Ethiopian rural household surveys (erhs), 1989–2009. URL: http://hdl.handle.net/1902.1/15646. Holden, S., Shiferaw, B., & Pender, J. (2004). Non-farm income, household welfare, and sustainable land management in a lessfavoured area in the Ethiopian highlands. Food Policy, 29, 369–392, URL: http://search.ebscohost.com/login.aspx?direct=true&db=eoh& AN=0754570&site=ehost-live. Irz, X., & Thirtle, C. (2004). Dual technological development in botswana agriculture: A stochastic input distance function approach. Journal of Agricultural Economics, 55, 455–478. http://dx.doi.org/10.1111/j.14779552.2004.tb00110.x. Kondylis, F., Mueller, V., & Zhu, J. (2017). Seeing is believing? Evidence from an extension network experiment. Journal of Development Economics, 125, 1–20. http://dx.doi.org/10.1016/j.jdeveco.2016.10.004, URL: http://www.sciencedirect.com/science/article/pii/ S0304387816300827. Krishnan, P., & Patnam, M. (2014). Neighbors and extension agents in ethiopia: Who matters more for technology adoption?. American Journal of Agricultural Economics, 96, 308–327. http://dx.doi.org/ 10.1093/ajae/aat017, URL: http://ajae.oxfordjournals.org/content/96/ 1/308.abstract. arXiv:http://ajae.oxfordjournals.org/content/96/1/ 308.full.pdf+html. Krishnan, P., & Sciubba, E. (2009). Links and architecture in village networks. Economic Journal, 119, 917–949, URL: http://proxyremote.galib.uga.edu/login?url=http://search.ebscohost.com/login. aspx?direct=true&db=ecn&AN=1042937&site=ehost-live. Kumbhakar, S. C. (2011). Estimation of multiple output production functions. Kumbhakar, S. C., & Lovell, C. A. K. (2000). Stochastic frontier analysis. Cambridge University Press.

13

Liverpool-Tasie, L. S. O., & Winter-Nelson, A. (2012). Social learning and farm technology in Ethiopia: Impacts by technology, network type, and poverty status. Journal of Development Studies, 48, 1505–1521. http://dx.doi.org/10.1080/00220388.2012.693167, URL: http:// www.tandfonline.com/doi/abs/10.1080/00220388.2012.693167. arXiv: http://www.tandfonline.com/doi/pdf/10.1080/00220388.2012.693167. Maertens, A., & Barrett, C. B. (2012). Measuring social networks’ eﬀects on agricultural technology adoption. American Journal of Agricultural Economics, 1–7. http://dx.doi.org/10.1093/ajae/aas049, URL: http:// ajae.oxfordjournals.org/content/early/2012/05/14/ajae.aas049.short. arXiv:http://ajae.oxfordjournals.org/content/early/2012/05/14/ajae. aas049.full.pdf+html. Mathenge, M. K., Smale, M., & Tschirley, D. (2015). Oﬀ-farm employment and input intensiﬁcation among smallholder maize farmers in kenya. Journal of Agricultural Economics, 66, 519–536, URL: http:// search.ebscohost.com/login.aspx?direct=true&db=eoh&AN= 1520602&site=ehost-live. Munshi, K. (2004). Social learning in a heterogeneous population: Technology diﬀusion in the indian green revolution. Journal of Development Economics, 73, 185–213, URL: http://proxy-remote.galib.uga.edu/login?url=http://search.ebscohost.com/login.aspx?direct= true&db=ecn&AN=0677472&site=ehost-live. Nemoto, J., & Furumatsu, N. (2014). Scale and scope economies of Japanese private universities revisited with an input distance function approach. Journal of Productivity Analysis, 41, 213–226. O’Donnell, C. J., & Coelli, T. J. (2005). A bayesian approach to imposing curvature on distance functions. Journal of Econometrics, 126, 493–523, URL: http://proxy-remote.galib.uga.edu/login?url=http:// search.ebscohost.com/login.aspx?direct=true&db=ecn&AN= 0776996&site=ehost-live. Pfeiﬀer, L., Lopez-Feldman, A., & Taylor, J. E. (2009). Is oﬀ-farm income reforming the farm? Evidence from Mexico. Agricultural Economics, 40, 125–138, URL: http://search.ebscohost.com/login.aspx?direct= true&db=eoh&AN=1036401&site=ehost-live. Santos, P., & Barrett, C. B. (2010). Identity, interest and information search in a dynamic rural economy. World Development, 38, 1788–1796, URL: http://proxy-remote.galib.uga.edu/login?url= http://search.ebscohost.com/login.aspx?direct=true&db=ecn&AN= 1154390&site=ehost-live. Songsermsawas, T., Baylis, K., Chhatre, A., & Michelson, H. (2016). Can peers improve agricultural revenue?. World Development, 83, 163–178, URL: http://search.ebscohost.com/login.aspx?direct=true&db=eoh& AN=1568703&site=ehost-live. Stock, J. H., Wright, J., & Yogo, M. (2002). Gmm, weak instruments, and weak identiﬁcation. Journal of Business and Economic Statistics Symposium on GMM. Storck, H., Emana, B., Adnew, B., Borowiccki, A., & W/Hawariat, S. (1991). Farming systems and resource economics in the tropics: Farming system and farm management practices of small holders in the Hararghe Highland. volume II. Wissenschaftsverlag Vauk. Taﬀesse, A. S., Dorosh, P., & Gemessa, S. A. (2012). Crop production in Ethiopia: Regional patterns and trends. In Food and agriculture in Ethiopia: Progress and policy challenges (pp. 53–83). University of Pennsylvania Press, URL: http://www.jstor.org/stable/j.ctt3fh6vv.12. Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. MIT Press.

APPENDIX A

14

WORLD DEVELOPMENT Table 9. Full set of iterated GMM coeﬃcients Labor-sharing use

For all agricultural activities

Parameter

Estimate

Teﬀ Wheat Barley Maize Sorghum Teﬀ-Wheat Teﬀ-Barley Teﬀ-Maize Teff-Sorghum Teﬀ square Wheat-Barley Wheat-Maize Wheat-Sorghum Wheat square Barley-Maize Barley-Sorghum Barley square Maize-Sorghum Maize square Sorghum square Livestock Fertilizer Labor Fertilizer-labor Fertilizer-Livestock Fertilizer square Labor-Livestock Labor square Livestock square Wheat-Livestock Wheat-Fertilizer Wheat-Labor Barley-Livestock Barley-Fertilizer Barley-Labor Maize-Livestock Maize-Fertilizer Maize-Labor Sorghum-Livestock Sorghum-Fertilizer Sorghum-Labor Teﬀ-Livestock Teﬀ-Fertilizer Teﬀ-Labor k Constant Region dummies Tigray Amhara SNNP

Not exclusively for harvesting and threshing

Standard error

Estimate

Standard error

0.039 0.030 0.004 0.036* 0.046* 0.001 0.001 0.001 0.001 0.000 0.000 0.000 0.002 0.001 0.000 0.002 0.000 0.001 0.002* 0.002 0.241*** 0.439*** 0.005 0.004 0.013 0.099*** 0.041*** 0.049** 0.137*** 0.002 0.001 0.005 0.001 0.011*** 0.004 0.001 0.008*** 0.000 0.002 0.004* 0.013*** 0.003 0.007*** 0.002 0.356*** 0.403

0.02 0.02 0.03 0.02 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.06 0.05 0.12 0.01 0.01 0.02 0.01 0.02 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.07 0.44

0.033 0.027 0.012 0.025 0.038 0.001* 0.001 0.001 0.001 0.001 0.000 0.000 0.001 0.001 0.001 0.002 0.000 0.001 0.002* 0.002 0.235*** 0.428*** 0.011 0.008 0.010 0.101 *** 0.040*** 0.052** 0.143*** 0.002 0.001 0.005 0.001 0.011*** 0.005 0.001 0.008*** 0.001 0.002 0.004** 0.013*** 0.003 0.007*** 0.002 0.356*** 0.296

0.02 0.02 0.03 0.02 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.07 0.05 0.12 0.01 0.01 0.02 0.01 0.02 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.07 0.42

0.505** 0.287** 0.184

0.25 0.12 0.12

0.566** 0.249** 0.203*

0.24 0.12 0.12

**

Oromiya is the base group for the region dummies. The ineﬃciency eﬀects in Table (6) are estimated simultaneously with these coeﬃcients.

Available online at www.sciencedirect.com

ScienceDirect

**

Synergy and Learning Effects of Informal Labor-Sharing Arrangements

Synergy and Learning Effects of Informal Labor-Sharing Arrangements

Recommend Documents