Short Paper
Does Migration Push Exit from Agriculture in Mexico?
2022 Volume 58 Issue 2 Pages 90-97
Details
2022 Volume 58 Issue 2 Pages 90-97
The decrease in the number of agricultural workers is a serious issue in Mexico. We hypothesized that migration might push the head of the household to exit from agriculture. The effect differs between farming and non-farming households because the former could increase farm income by devoting part of their remittance income to agricultural production. This study examines the relationship between the migration of household heads and their engagement in agriculture after returning home and their use of agricultural inputs. We found no significant relationship between migration and engagement in farming for farming households but a negative and significant relationship for non-farming households. In addition, we found a positive and significant relationship between migration and input use in farming households. This suggests that migration is likely to cause the exit of household heads from agriculture, but the possibility of increased farm income by purchasing more inputs may mitigate this effect.
The decrease in the number of agricultural workers has become a major concern in many countries. The decline in Mexico is shown in Figure 1, and it is a notable issue because it is related to the sustainability and competitiveness of agriculture in Mexico, and the agriculture of the United States relies heavily on the labor supply from Mexico (Taylor et al., 2012).
The rate of labor force in agriculture (%)
Source: Mexican Migration Project (MMP) “commun174.csv”, World Bank
1) The rate in the MMP community is obtained by calculating the sum of the product of the labor force rate in agriculture of the village and the population of the municipality, and then dividing it by the total municipal population of all villages.
Previous studies have suggested that farmers tend to leave agriculture when nonfarm income is higher than farm income or when the farm income is insecure (Sharma and Bhaduri, 2009; Taylor, et al., 2012). Moreover, Mexico has a long history of supplying migrants to the U.S. The migration of household members may affect the exit from agriculture of each member in Mexico. The off-farm experience reduces transaction costs involved in the occupational shift to other industries (Pfeffer, 1989; Weiss, 1997). In rural Mexico, Fitting (2011) reported that migrants have become accustomed to the wage payment system and the culture of the destination country, and do not prefer to engage in farming in the local area after their returning. Thus, the migration experience pushes people exit from agriculture. However, such effects may depend on whether the agricultural households with migrants have operated their farms, or the migrant members have engaged in agriculture without farms, mainly agricultural workers. Given the weak credit market in rural areas, as with the case of rural Mexico, the remittance from migrant members and the savings mitigate the budget constraints of migrant households and thus may increase investment in agriculture if they own farms. The increased investment helps the household stay in agriculture.
The existing literature that examined the relationship between migration and exit from agriculture did not consider the possible heterogeneity by farm holdings, which makes migration and exit be related by enabling households to invest in agriculture (Fitting, 2011; Pfeiffer et al., 2009).
The purpose of this study is to fill the gap in the literature, using a rich individual and household data in Mexico provided by the Mexican Migration Project (MMP). First, we examine whether the migration experience of agricultural household heads makes them continue farming or exit, and the relationship differs by the status of farm holdings. The study focuses on the occupational choice of the household heads because they constitute most migrants in Mexico and are the core workers in the agricultural household (Fitting, 2011). Second, we examine whether the migration experience of household heads increase the use of agricultural inputs and machinery.
The destination of 89.4% of Mexican migrants was the United States in 2010 (Mexican Census of Population and Housing 2010). There is a long history of Mexican migration to the U.S. According to Massey et al. (2002), with the restriction of migration from Asia and the increased domestic demand for labor in the U.S., Mexican migrants began to increase. Between 1942 and 1964, the Bracero program granted working visas to Mexicans, and the number of Mexican migrants to the U.S. increased dramatically. After 1980, the migration rate from rural Mexico continued to increase, but it dropped sharply to less than half from 2005 to 2009 (Conover, et al., 2018).
According to Fitting’s survey (2011), young males usually migrate in rural Mexico. Most migrants worked in the primary industry in the destination country for 8 to 9 months and returned to the local area to help harvest crops in 1980s. However, after the mid-1990s, more and more migrants preferred to work in the secondary and tertiary industries in the destination country for 2 to 5 years, and most of them held a factory job and received wages after their returning. Thus, the migration period was recently getting longer, and migrants were shifting to nonagricultural jobs.
We conducted the empirical analysis using the MMP published by Princeton University. MMP(174) is a dataset of survey results from 1987 to 2018 and in 1982, consisting of a sample of 28,331 households. MMP nonrandomly selects a different community in Mexico each year and randomly samples about 200 households in each community. If the head was absent in the surveyed household, the spouse was interviewed. MMP provides information on the year of acquisition or lease of farmland and the migration experience for household heads from their birth year to the survey year. This enabled us to identify the year when the household head exited farming. MMP also provides the information on the input use in the survey year. We constructed the pooled cross-section dataset from 1987 to 2018 and in 1982 to analyze the relationship between migration and input use.
(2) Definition of subsamplesTo examine the different effect of migration, which is caused by farm households’ gaining higher wage to purchase agricultural inputs, we focus on small family farms, which are more likely to be subject to the budget constraint, and we restricted the original sample using the following definitions and constructed a sample of farm households (n=1,695).
1. A household that has ever been owned or rented in farmland before the survey year.
2. A household that owned or rented in less than 5 ha of land or did not own or rent in land in the survey year. According to the observation of Fitting (2011), most rich households owned more than 5 ha of land. We identified small farms based on her observation. In our sample, 40.2% of farm households owned more than 5 ha of land and they were excluded in the analysis.
3. The age of the household head is less than 65 years
4. A household that was surveyed after 1998. This definition is added because information about the cultivated area is available only after the 1998 survey. We distinguished between cultivating and owning or renting. If a household cultivates land, it owns or rents in land, but the reverse is not necessarily true.
5. A household that was surveyed in Mexico.
To compare the estimates and examine the different effects of migration, we constructed a sample of nonfarm households (n=4,675). This sample consists of those households whose heads have ever worked in the agriculture but have never owned or rented in land, and satisfies the definitions 3, 4 and 5 explained above.
Table 1 shows the descriptive statistics for the sample of farm households and the sample of nonfarm households. Seventy-four percent of households have never owned or rented in land and were categorized in the sample of nonfarm households. Also, most nonfarm households did not use inputs in the survey year. Therefore, we assume nonfarm households usually do not invest in input.
| farm | nonfarm | |
|---|---|---|
| Head characteristics | ||
| Main occupation in Mexico in the survey year | ||
| Agriculture | 0.63 | 0.49 |
| Non-agriculture | 0.32 | 0.47 |
| Job experience of migrant heads in the U.S. | ||
| Any job | 0.98 | 0.98 |
| Only agriculture | 0.25 | 0.31 |
| Migration experience and the timing | ||
| Yes, after started agriculture, migafter | 0.23 | 0.33 |
| Yes, but the timing is unclear, migother | 0.04 | 0.03 |
| Head was in Mexico | 0.97 | 0.93 |
| Education (year) | 6.05 | 5.81 |
| Age (year) | 48.7 | 44.2 |
| Inheritance | ||
| Land (ha) | 1.35 | 0 |
| House (amount) | 0.36 | 0.22 |
| Migration experience of relatives to the U.S. | ||
| Father | 0.02 | 0.03 |
| Mother | 0.01 | 0.01 |
| Brother | 0.43 | 0.40 |
| Sister | 0.13 | 0.16 |
| HH characteristics | ||
| Cultivation | ||
| Yes/No | 0.88 | 0 |
| Land size (ha) | 1.46 | 0 |
| Farm management and input | ||
| Financed by U.S.$ (yes/no) | 0.10 | 0.00 |
| Fertilizer (yes/no) | 0.71 | 0.02 |
| Insecticide (yes/no) | 0.64 | 0.02 |
| Family labor (num) | 1.51 | 0.05 |
| Daily labor (num) | 0.85 | 0.03 |
| Machinery (yes/no) | 0.38 | 0.01 |
| Truck (num) | 0.38 | 0.27 |
| Tractor (num) | 0.01 | 0.00 |
Source: MMP.
We ran the following two regression models, where i denotes household.
We were interested in whether a household head who had already engaged in agriculture continues to have an agricultural job after experiencing the first migration. Thus, the main variable of interest is migafter, which is a dummy variable meaning that the household head has migrated after working in agriculture or having owned or rented in land. We controlled for a dummy variable migother, which captures all other cases than migafter, such that the household head has migrated before participating in agriculture or in the case where the order of engagement in agriculture and migration cannot be identified. In farm households, 23% satisfied migafter=1, and 4% satisfied migother=1. In nonfarm households, 33% satisfied migafter=1, and 3% satisfied migother=1. The first regression model is shown below.
(1) jobagiri=α0+α1migiafter+α2migother+controli+communi*yeari+εi.
Jobagrii is a dummy variable, indicating if the household head’s main job in the survey year is agriculture in Mexico. Model 1 is employed to examine whether migration makes household heads more likely to engage in agriculture in Mexico. The parameters were estimated by OLS while controlling for community and survey year dummy.
We assume migration effect on the engagement of household head in agriculture is decomposed into the negative effect of experiencing off-farm job and the positive effect of gaining higher wage to purchase inputs. Farm households may get both effects, but nonfarm households may get only former negative effect because they do not cultivate land in which purchased inputs were used. Therefore, we expect α1 can be negative for nonfarm households, and can be either 0 or positive for farm households.
(2) inputi=β0+β1migiafter+β2migiother+controli+communi*yeari+εi
Model 2 was employed to examine what kind of input had increased after migration. The parameters were estimated by OLS.
The dependent variable in Model 2 contains a dummy variable that indicates whether money, especially US dollars, was paid for purchasing some equipment or renting in land (farm_dol), the usage of variable input, and fixed input, all of which indicates the values in the survey year. The variable input includes the use of fertilizers (fertil), the use of insecticides (insect), the number of family workers (familab), and the number of daily workers (dailab). Fixed input includes the holding status of agricultural machinery (machine), the number of trucks and tractors (truck, tractor), and total farmland area (land).
We assume β1 is positive or at least 0 for all inputs except family labor because migration should usually generate the cash transfer to households, and it enables them to purchase inputs.
Several heads, whose household was in Mexico in the survey year, stayed in the U.S. in the survey year, 3% in the sample of farm households and 7% in the sample of nonfarm households. Thus, we controlled for a dummy indicating that the household head was in Mexico in the survey year.
There is also the endogeneity issue of migration. To deal with that, we controlled for the education years of household heads in the survey year in models 1 and 2. Only 0.5% of heads for farm households and 0.4% of heads for nonfarm households changed education years after their first migration. Thus, we assumed that migration may not have affected the attainment of education on the part of household heads in our sample. Also, we controlled for the amount of inherited house and land as the proxy of wealth in models 1 and 2. However, we still could not control for other unobserved characteristics such as entrepreneurship and the health conditions of household heads. The list of all control variables is shown in the footnote of Table 2.
| Dependent variable | job agri | job agri | job nonagri | job nonagri | Cultivation (yes/no) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| migafter | −0.03 (0.03) | −0.17 (0.14) | −0.09*** (0.02) | −0.28*** (0.10) | 0.03 (0.03) | 0.12 (0.14) | 0.09*** (0.02) | 0.34*** (0.10) | 0.04* (0.02) | −0.07 (0.10) |
| method | OLS | IV | OLS | IV | OLS | IV | OLS | IV | OLS | IV |
| sample | farm | nonfarm | farm | nonfarm | farm | |||||
| adjR2 | 0.30 | 0.26 | 0.25 | 0.23 | 0.21 | 0.20 | 0.21 | 0.17 | 0.20 | 0.18 |
| over-id score | 11.4 | 11.1 | 9.13 | 14.2 | 12.7 | |||||
| N | 1695 | 4675 | 1695 | 4675 | 1695 | |||||
Source: MMP.
1) Heteroskedasticity robust standard errors were shown in the parentheses.
2) *** means the significance level in 1%, ** for 5 % and * for 10%.
3) F statistic in the 1st stage was 23.2 (farm) and 26.5 (nonfarm), F statistic for IV variables in the 1st stage was 9.7 (farm) and 23.5 (nonfarm). Also, no overidentification scores were significant.
4) In both model 1 and 2, we controlled for education years, sex, age, inherited land area and houses of the household head, a dummy indicating household head changed education after migration, a dummy indicating household head was in Mexico when whose household was surveyed, community and survey year fixed effects.
The MMP provides information about the migration experiences of the fathers, mothers, brothers, and sisters of the head. We can use those family networks as IV (de Brauw, 2019). We also can use the information of co-residence status with those head’s family in the survey year. However, it may be hard to completely satisfy the exclusion restriction. Therefore, as the robustness check, we complementarily conducted the IV estimation, where endogenous variables are migafter and migother, and instruments are the migration experience of the heads’ fathers, mothers, brothers, and sisters, the co-residence status with them, and the interaction terms. We used 12 instruments.
Table 2 and 3, respectively show estimation results of models 1 and 2. The estimates by OLS and IV are shown, respectively. The IV estimates are for the robustness check. The results of the F-statistics in the first stage are significant. It suggests that our instrumental variables are not too weak.
| Dependent variable | farm dol | fertil | insect | machine | truck | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| migafter | 0.17*** (0.03) | 0.45*** (0.12) | 0.06** (0.03) | 0.15 (0.12) | 0.06** (0.03) | 0.30** (0.13) | 0.09*** (0.03) | 0.30** (0.13) | 0.11*** (0.04) | 0.12 (0.21) |
| unit | yes/no | yes/no | yes/no | yes/no | num | |||||
| method | OLS | IV | OLS | IV | OLS | IV | OLS | IV | OLS | IV |
| adjR2 | 0.12 | 0.01 | 0.23 | 0.23 | 0.22 | 0.18 | 0.28 | 0.25 | 0.16 | 0.05 |
| over-id score | 9.38 | 11.3 | 8.69 | 10.2 | 8.25 | |||||
1) Results that were not significant were skipped, such as family labor, day labor, cultivated land area and tractor.
The results of model 1 for farm households and nonfarm households are shown in Table 2. We also regressed a dummy variable, indicating that the household head had a nonagricultural job, and a dummy variable, indicating that the heads and/or some of household members were cultivating land in the survey year.
In Table 2, we found no significant relationships between the migration of farm household heads and their engagement in the agricultural and nonagricultural job after returning. However, for nonfarm household heads, we found a significant and negative relationship for engagement in agriculture, and positive relationship for the engagement in nonagricultural job. These results were supported by the IV estimates. Therefore, it is suggested the migration usually pushes workers to exit from agriculture, and the effects are different between migrants from farm households and nonfarm households.
This is in line with previous studies that argue that the migration experience pushes people to exit from agriculture (Fitting, 2011; Pfeffer, 1989; Weiss, 1997).
Table 2 also shows, in the sample of farm households, positive and significant relationships between migration and the probability that a household continues cultivating. However, the IV estimate does not show a significant result, and the effect of migration is not robust.
If household heads with migration experience got only an agricultural job in the U.S., they should not have received the negative effect of migration by experiencing nonagricultural jobs. To examine this in our sample, we additionally regressed the same dependent variables in Table 1, controlling for the same variables, on the dummy variable indicating the household head had only an agricultural job in the U.S. with restricting the sample to migafter=1. The result, shown in Table 4, suggests if a household head experienced only an agricultural job in the U.S., they may not receive the negative migration effect, and it was more significant in the sample of farm households. This supports the existence of migration effect by experiencing nonagricultural jobs. However, the job selection in the U.S. is also endogenous, and this result may involve the bias by omitted unobservable variables. The future study must examine this relationship in detail.
Table 3 shows results of Model 2, which examines the relationship between the migration experience of household heads and the input use in the farm household.
| Dependent variable | job agri | job agri | job nonagri | job nonagri | Cultivation (yes/no) |
|---|---|---|---|---|---|
| only agri in US | 0.16** (0.07) | 0.04 (0.03) | −0.13** (0.07) | −0.04 (0.03) | −0.01 (0.05) |
| sample | farm | nonfarm | farm | nonfarm | farm |
| adjR2 | 0.34 | 0.27 | 0.20 | 0.30 | 0.25 |
| N | 393 | 1535 | 393 | 1535 | 393 |
We found the positive and significant relationship between migration and farm_dol, which suggests the household that sent a migrant is significantly more likely to purchase inputs with U.S. dollars. The estimates on the use of fertilizer, insecticides, agricultural machinery and the number of trucks are positive and significant. The IV estimates also supported OLS estimates except concerning the number of trucks. Our result suggests households that sent a migrant used more variable and fixed inputs.
Comparing the results in Tables 2 and 3, it is suggested that household heads from nonfarm households who migrated is usually more likely to exit from agriculture. However, farm households whose heads migrated may purchase agricultural inputs by the remittance income, which may change heads’ minds about quitting farming. One of the important concerns that has been discussed in previous literatures is whether the increased input by remittances is productive (De Brauw, 2019) or for labor substitutive purposes (Fitting, 2011; Pfeiffer, et al., 2009). The former decision-making of input use may promote head’s engagement in farming, but the latter may help them exit. Fitting (2011) reported agricultural machinery, such as an electric corn grinder (molino), and trucks are “income-generating equipment” in the rural Mexico. We found a significant and positive relationship between migration and the use of machinery. Moreover, we found no significant relationship between migration and labor and land. Thus, it suggests that the farm management became capital intensive after the head returned from migration. Migration may have supported rural agriculture in Mexico.
This paper examined the relationship between the migration of the household head and their engagement in agriculture after their return, and compared the results between farm and nonfarm household. It also shows the relationship between the head’s migration and the input use of the farm household.
We conducted a regression analysis with the sample of the MMP from 1998 to 2018 surveys.
The first suggestion is that the migration experience of household heads may make them more likely to exit from agriculture after their return if they don’t have land. Second, it is suggested that households that sent a migrant tend to have more fixed and variable inputs, such as agricultural machinery, trucks, fertilizer, and pesticides.
Considering these results and the discussion of previous studies, we can interpret that migration may affect the job selection of the household head, who is likely to be the core worker in the farm household in rural Mexico (Fitting, 2011) directly by providing the experience in off-farm jobs, and this effect may vary if the head has an option to purchase inputs. Our result infers the possibility that migration helps household heads shift into off-farm jobs and exit from agriculture. However, farm households that sent a migrant are more likely to purchase fixed and variable inputs that may generate income, which may make the head continue to engage in agriculture by increasing the profitability of agriculture after the head’s return. Finally, the migration effect on the exit of farm household heads may be offset.
The first contribution of this study is to specifically examine the household head’s exit, who is likely to be the core worker. The decrease in the number of core workers may have a big impact on Mexican agriculture. The second contribution is to compare the relationship between the migration experience of the household head and the continuation of farming and input use to understand why agricultural labor recently decreased in Mexico. In our sample, seventy-four percent of households haven’t owned or rented in land, and those household heads are more likely to exit agriculture. Therefore, we may say migration can be a factor that pushed the Mexican agricultural labor exit.
There are remaining issues with the empirical method used in this study. First, we did not completely solve the endogeneity issue related to migration, although we tried to control for important covariates and conducted an IV estimation as the robustness check. Second, we did not explicitly examine the simultaneous decision-making between input use and the job selection of household heads, and the relationship between possession of fixed capital purchased in the past and head’s current job selection because of data construction. Third, because this study used a sample mostly consisting of household heads who had returned to their village from the destination and the decision-making of returning can be endogenous, the selection by returning may have biased our estimates. The future study must more explicitly resolve the sample selection issue. Fourth, the future study needs to divide the effects of migration more explicitly into the effects of gaining skills and gaining money.
The authors would like to thank Enago (www.enago.jp) for the English language review.
