1. Introduction
Labor migration from rural areas, particularly rural-urban migration, and remittances sent by migrants to rural households are becoming increasingly important features of many developing countries (Taylor and Lopez-Feldman, Reference Taylor and Lopez-Feldman2010; International Organization for Migration [IOM] 2019; Marta et al., Reference Marta, Fauzi, Juanda and Rustiadi2020). Ghana is a prime example of this trend, with migration having been key to its development since the colonial times (Tutu, Reference Tutu and Tutu1995). Agriculture-related migration in Ghana includes rural-rural, rural-urban, and urban-rural migration, with rural-urban migration being the most prevalent, especially amongst the poorest and most agriculturally dependent regions (GSS, 2021).
The role of migration has received considerable attention in development economics. Research shows that remittances from migrants can help rural households overcome credit and risk constraints (Taylor and Lopez-Feldman, Reference Taylor and Lopez-Feldman2010). For instance, farmers frequently use remittances sent by migrants to purchase new and improved farm inputs as well as hire labor, which increases agricultural productivity (GSS, 2014; IOM, 2019; Msokwe and Mbonile, Reference Msokwe and Mbonile2021; Tabetando et al., Reference Tabetando, Matsumoto and Fani2022).
Rural-urban migration, on the other hand, can result in the loss of rural workers from agriculturally dependent regions, which can lead to planting and harvesting delays (Amfo et al., Reference Amfo, Aidoo and Mensah2022; IOM, 2022) and land abandonment (Gyamera et al., Reference Gyamera, Duncan, Kuma, J.S. and Arko-Adjei2018; Msokwe and Mbonile, Reference Msokwe and Mbonile2021; de Brauw et al., Reference de Brauw, Kramer and Murphy2021). All these factors have the potential to harm agricultural productivity and rural development (Awumbila et al., Reference Awumbila, Owusu and Teye2014; Adaku, Reference Adaku2019). Other studies, such as Rozelle et al. (Reference Rozelle, Taylor and DeBrauw1999), find that migration has both negative labor loss and positive remittance effects on agricultural production in rural China. The effects of migration and remittances on the income and productivity of households are complicated and hard to predict, as Taylor and Lopez-Feldman (Reference Taylor and Lopez-Feldman2010) argue; hence, robust framework econometric approaches that account for possible self-selection issues of migration.
The aim of this study is to examine the impact of rural-urban migration on technical efficiency changes in maize production in Ghana. Technical efficiency change refers to how well a farm uses its inputs to produce outputs over a given period (Coelli et al., Reference Coelli, Rao, O’donnell and Battese2005). We concentrate on Ghana because it has one of the highest levels of remittances in Sub-Saharan Africa and has seen a significant increase in rural-urban migration in recent years (GSS, 2014; IOM, 2019). Furthermore, farmers in Ghana have been reported to be technically inefficient (Asante et al., Reference Asante, Villano and Battese2017; Asante et al., Reference Asante, Temoso, Addai and Villano2019; Danso-Abbeam and Baiyegunhi, Reference Danso-Abbeam and Baiyegunhi2020; Selorm et al., Reference Selorm, Sarpong, Egyir, Mensah Bonsu and An2023; Yitayew et al., Reference Yitayew, Abdulai and Yigezu2023), with farmers producing only 71% of the country’s potential maize yield of 5.2 million metric tons between 2017 and 2021 (Food and Agriculture Organization (FAO), 2022; Ministry of Food and Agriculture [MOFA] 2020; ISSER 2020). As a result, evidence on the factors influencing farm technical efficiency, including the impact of migration, is of interest to researchers and policymakers in Ghana.
However, existing evidence on the impact of migration on farm technical efficiency and productivity in Ghana and globally is mixed, with some studies reporting negative effects (Sauer et al., Reference Sauer, Gorton and Davidova2015), others reporting positive effects (Agza et al., Reference Agza, Alamirew and Shibru2021), and others finding no relationship (Wouterse, Reference Wouterse2010). One possible explanation for these findings is that most studies use cross-sectional data, which fails to account for self-selection in migration. According to Taylor and Lopez-Feldman (Reference Taylor and Lopez-Feldman2010), it is important for studies to seek out possible instruments to control for the non-randomness of the process that allocates households across migration regimes, as well as to identify specific ways in which migration influences efficiency in rural households.
As a result, our research contributes to the literature in several ways. First, by utilizing panel data from the Ghana Socioeconomic Panel Survey (2013/14 and 2017/18), we were able to examine how the impact changes over time (Agza et al., Reference Agza, Alamirew and Shibru2021; Sauer et al., Reference Sauer, Gorton and Davidova2015; Wouterse, Reference Wouterse2010). We employ an econometric strategy that combines three approaches: propensity score matching (PSM) and difference-in-differences (DID) techniques in conjunction with stochastic production frontiers (SPF), which provides a more comprehensive framework to address the issue of selectivity bias associated with rural-urban migration. These approaches allow us to account for both observed and unobserved heterogeneities in our estimations. Thus, while the findings are focused on Ghana, the econometric strategy used is quite novel in this field; thus, the findings can be replicated in other countries and settings where migration is common.
Furthermore, as far as we know, no research has been conducted to evaluate the impact of rural-urban migration on the technical efficiency of maize farmers in Ghana, specifically using panel data. Most of the research that has been done in Ghana on how migration affects different economic indicators has used cross-sectional methods (Agza et al., Reference Agza, Alamirew and Shibru2021, Reference Agza, Alamirew and Shibru2023; Sauer et al., Reference Sauer, Gorton and Davidova2015; Wouterse, Reference Wouterse2010). Teye et al. (Reference Teye, Boakye-Yiadom, Asiedu, Litchfield, Wilson, Kubi and Awumbila2019) conducted a notable study on the impact of migration on household welfare in Ghana, albeit without considering technical efficiency. As a result, this study contributes directly to ongoing efforts aimed at improving rural livelihoods and building resilience in one of Ghana’s most disadvantaged regions. In addition, the findings can help determine whether migration acts as a constraint or a catalyst for efficiency, offering evidence to guide agricultural policy, rural development planning, and social protection programs in Ghana and other developing economies with similar characteristics.
The rest of the paper is organized as follows: Section 2 presents and discusses the trends in rural-urban migration in Ghana. Section 3 discusses the data and study area as well as the econometric procedures used in this paper. The results are presented and discussed in Section 4, while the final section concludes.
1.1. Trends of rural-urban migration in Ghana
Rural-urban migration is a common type of migration in many SSA countries, including Ghana, where trends have been significant since colonial times. As SSA countries transition from subsistence agriculture to cash crop cultivation, manufacturing, and services, a significant number of people, particularly young males, migrate to urban areas for better-paying jobs (Kutor et al., Reference Kutor, Annan-Aggrey, Poku, Kyeremeh and Arku2022; Fengbo et al., Reference Fengbo, Lucas, Bloom and Shijun2016; Aguilar-Støen et al., Reference Aguilar-Støen, Taylor and Castellanos2016; Deaton and Anschel, Reference Deaton and Anschel2015). In 2008, the global urban population surpassed the global rural population for the first time in human history (Todaro and Smith, Reference Todaro and Smith2014). By 2030, cities are expected to house 61.1% of the world’s population, increasing to two-thirds (66%) by 2050 (United Nations, 2014).
Migration has been an important part of Ghana’s development (Tutu, Reference Tutu and Tutu1995; GSS, 2014; IOM, 2019). According to GSS (2021), 5.2 million people migrated from urban to rural areas, 4.1 million migrated between urban areas, 2 million migrated from rural to urban areas, and 1.9 million migrated from rural to rural areas. Migration patterns also differ by gender, with male primarily migrate for job opportunities, while females often migrate to join their husbands and partners (GSS, 2014).
Rural-urban migration is more prevalent than urban-to-urban migration and has been a key driver of Ghana’s rapid urbanization (GSS, 2014). The country’s urbanization rate rose steadily from 9% in 1931 to 14% in 1948, 23% in 1960, 44% in 2000, and 51% in 2010 (GSS, 2014). By 2021, the urban population reached 56%, largely due to growth in the Greater Accra and Ashanti regions, which together accounted for 47% of this urban expansion (GSS, 2021). Approximately 28.9% of the population were internal migrants, with 33.9% of rural residents migrating to urban areas, compared to 22.2% of urban residents relocating to rural areas.
Regionally, Greater Accra, Western, Ashanti, and Brong Ahafo received the most migrants, while the rest of Ghana lost people (GSS, 2014). Greater Accra gained nearly twice as many people as the other three regions combined (Western, Ashanti, and Brong Ahafo), while the Volta Region lost the most people due to migration, followed by the Northern and Eastern regions (GSS, 2014). Greater Accra also has the highest rate of urbanization in Ghana, at 91%, followed by Ashanti at 61% (GSS, 2014). This implies that migration from rural to urban areas plays a significant role in the growth of these capitals. By 2050, 26.5 million Ghanaians are expected to live in cities, potentially affecting public health, infrastructure, and social stability in the destination cities (UN Habitat, 2014). In terms of agricultural migrants, historically, many Ghanaians migrated to other regions in search of land to cultivate cocoa (GSS, 2021). However, recent data (GSS, 2021) indicates that the era appears to be over, and people are now moving short distances within their regions for agricultural purposes (GSS, 2021).
Some policymakers and researchers see recent rural-urban migration trends as a key source of economic growth through remittances sent to origin regions and as key labor suppliers in destination regions (Agyemang and Raqib, Reference Agyemang and Raqib2013; Kutor et al., Reference Kutor, Annan-Aggrey, Poku, Kyeremeh and Arku2022). However, some policy experts and urban developers are concerned that it will disrupt urban development, create urban peripheries, and increase crime rates (Agyemang and Raqib, Reference Agyemang and Raqib2013; Ge et al., Reference Ge, Long, Qiao, Wang, Sun and Yang2020; Marta et al., Reference Marta, Fauzi, Juanda and Rustiadi2020; Deaton and Anschel, Reference Deaton and Anschel2015). According to some studies, rural-urban migration contributes to the loss of a needed labor force in rural areas, which may have long-term consequences for agricultural productivity and rural development (Awumbila et al., Reference Awumbila, Owusu and Teye2014; United Nations, 2014). Some empirical evidence suggests that rural-urban migration can lead to a decrease in farming activities due to population decline (Amfo et al., Reference Amfo, Aidoo and Mensah2022; Kutor et al., Reference Kutor, Annan-Aggrey, Poku, Kyeremeh and Arku2022).
Despite this, research on the impact of rural-urban migration on farm technical efficiency is limited and inconclusive. As a result, the primary goal of this research is to make a scholarly contribution by investigating the relationship between rural-urban migration and its impact on the technical efficiency of maize farmers in Ghana.
2. Materials and methods
2.1. Study sites
The study focused on Northern region of Ghana, which was selected due to its well-known high rate of rural-urban migration. Farming households in this area face widespread poverty, climate vulnerability, and heavy reliance on agriculture. As migration reshapes labor availability and resource flows in rural areas, understanding its impact on farm productivity in this region is crucial. According to the International Organization for Migration (IOM, 2019), the Northern region accounted for one of the largest volumes of rural-urban migrants, totaling 339,687 individuals. Additionally, the region is a significant contributor to the total maize crop production in Ghana. The Northern region is in the northern part of Ghana and covers a total area of 70,384 km2. It is situated within latitude N 9° 32' 38.1372'' and longitude W 0° 54' 20.3832. The region shares boundaries with the Savannah region to the west, the Togo international border to the east, the Oti region to the south, and the North East region to the north. The region has a population of 1,948,913 inhabitants (GSS, 2021), with more than two-thirds of the population engaged in agriculture. Figure 1 illustrates the study area. The major crops grown in the region include yam, rice, sorghum, cowpea, millet, maize, guinea corn, groundnuts, beans, and soybeans. The region falls within the Guinea savannah agroecological zone, known for its arid conditions and vegetation predominantly composed of grasslands and drought-resistant trees. Temperatures in the region range from 14 °C at night to as high as 40 °C during the day. The rainfall pattern is unimodal, with average annual rainfall ranging between 750 to 1050 mm. Many farmers in the region have migrated from rural villages to the southern part of Ghana. According to the GSS (2021), there were 433,121 declared out-migrants and 100,524 in-migrants in the region.
Map of Northern region. Source: Author’s design, 2023.
2.2. Data
We utilized data collected from nationally representative surveys, known as the Ghana Socioeconomic Panel Survey, conducted between November 2013 and April 2018. The survey is a collaboration between Institute of Statistical, Social and Economic Research (ISSER), Global Poverty Research Lab at Northwestern, and Economic Growth Center (EGC) at Yale. The data collection covered all ten regions of Ghana, namely Greater Accra, Ashanti, Eastern, Western, Central, Northern, Upper East, Upper West, Brong Ahafo, and Volta. The data captures information from different crop production seasons from 2013 to 2018 production seasons. To ensure a representative sample, a two-stage stratified sampling design was employed. In the first stage, geographic clusters were purposively selected from an updated master sampling frame based on the 2000 Ghana Population and Housing Census. The second stage involved sampling 334 areas from the list of enumeration areas (EAs) in each region. In the final stage, a simple random sampling technique was used to select 15 listed households from each selected cluster. In total, 5,010 households were interviewed as part of the survey. The Ghana Socioeconomic Panel Survey provides a robust dataset that captures the dynamics and characteristics of farm households across Ghana’s various regions, enabling comprehensive analysis and insights into the agricultural sector and related socioeconomic factors.
The analysis focuses on a subset of maize farm households located in the Gushegu, Karaga, Savelugu, Nanton, and Tolon districts in northern Ghana (Figure 1). These districts were selected due to their high maize production (ISSER, 2020) and significant rural-urban migration rates observed in the region over the past decades (IOM, 2022). Each district contributed 291 farm households to the sample. For the analysis, we utilized data from the second and third waves of the Ghana Socioeconomic Panel Survey. A total of 1,109 households were successfully interviewed during the second wave. In the third wave, the incomplete response rate was 4.8%, resulting into 1,056 households available for re-sampling. Thus, a balanced panel of 1,056 maize farm households was used for the analysis. The study also examined the systematic attrition rate and found no significant difference in the regression analysis when comparing the results with and without considering attrition. The follow-up rounds and the baseline survey questionnaire provided extensive information on various aspects, including household characteristics, production variables, immigration, and migration history.
2.3. Conceptual and empirical framework
The conceptual framework (Figure 2) illustrates the hypothesized pathways through which rural-urban migrationFootnote 1 influences technical efficiency in maize production. At the core of the framework is the recognition that migration alters household labor dynamics and resource flows, which can have both negative and positive effects on farm-level efficiency. On one hand, migration reduces the availability of household labor, potentially disrupting critical farm operations such as land preparation, planting, and weeding. This labor loss may lead to delays, suboptimal input use, and reliance on less experienced household members or hired labor, which could reduce technical efficiency.
Conceptual framework. Source: Author’s design.
Conversely, migration can enhance efficiency through remittance flows. These financial transfers can alleviate liquidity constraints and enable investment in productivity-enhancing technologies, such as improved seeds, fertilizers, or mechanization. However, it is important to recognize that remittances do not necessarily translate directly into agricultural investment. Households may channel remittances into non-agricultural activities or income diversification strategies, potentially reducing reliance on farming. This diversification may both mitigate risks associated with agriculture and shift labor away from farming, thereby having a deliberating effect on influencing technical efficiency. Most rural households incur significant costs supporting migrants, such as money or food sent out to family members, reducing the resources available for agricultural inputs, hired labor, or land improvements, especially in the short term. This dual nature of remittance flows both inward and outward, creating a dynamic resource environment that shapes the net impact of migration on technical efficiency.
Additionally, migration influences the household’s farm management capacity. The departure of a key decision-maker can affect the quality and timeliness of farm decisions, especially when responsibilities shift to individuals with limited experience or access to agricultural support services. The impact on technical efficiency thus depends on how well households adapt to these changes. The framework also captures the enabling role of technology adoption and input use as mediating factors. When migration leads to improved access and application of modern technologies, the negative effects of labor loss can be offset, resulting in enhanced efficiency.
Overall, this framework underscores the complex and context-dependent relationship between migration and technical efficiency. It provides a structured basis for empirical analysis, highlighting key pathways and intervening variables that should be considered in assessing the net impact of migration on maize productivity.
Previous studies (Lai et al., Reference Lai, Polachek and Wang2009; Kumbhakar et al., Reference Kumbhakar, Tsionas and Sipilainen2009; Greene, Reference Greene2010) have used the SPF model with sample selection correction which considers unobserved factors that may lead to selection bias. This bias can result from the correlation between the selection equation term ϵ i and μ i in the traditional SPF model. Greene (Reference Greene2010) posits that the correlation between ϵ i and v i in the traditional SPF model is a source of selection bias. To address this issue, Greene (Reference Greene2010) extended the Heckman sample selection method to the SPF model using adjusted linear and nonlinear models (Terza, Reference Terza2009). Unlike the computationally demanding log likelihood functions used by Kumbhakar et al. (Reference Kumbhakar, Tsionas and Sipilainen2009), this study adopts multi-step method (Bravo-Ureta et al., Reference Bravo-Ureta, Greene and Solís2012) to address selection bias resulting from unobserved factors.
Greene (Reference Greene2010) argues that the Heckman selectivity correction method is not appropriate for non-linear models and developed a new estimator for SF in cross-sectional data that addresses selectivity. The current study uses the extended model from Greene (Reference Greene2010) and Bravo-Ureta et al. (Reference Bravo-Ureta, Greene and Solís2012) to integrate two-round panel data (baseline and endline) and DID estimator. The study is conducted in four steps. Step 1 involves estimating the decision to migrate using a logit model, where the dependent variable is binary (1 for migrant and 0 for non-migrant farm households) and expressed as a function of baseline covariates. This model is used to derive propensity scores. Step 2 uses PSM to select similar treatment and control units based on their propensity scores, except for their treatment status. Step 3 uses the matched samples to estimate a selectivity-corrected SPF-DID using the panel data, which allows for the measurement of the impact of rural-urban migration on maize output value and TE changes in a unified step, as opposed to relying on endline cross-sectional data. Step 4 analyzes the shift in the frontier between the baseline and the endline due to common trends for all farm households and to the impact of rural-urban migration. The study also identifies efficiency disparities for each period and observation for both treated and control groups. The DID estimator accounts for baseline differences that may persist after matching the treated and control groups.
The SF model based on panel data (Aigner et al., Reference Aigner, Lovell and Schmidt1977) can be written as:
$$Y_{it}=\beta ^{`}X_{it}+v_{it}-u_{it};$$
$$ u_{it}=\left| \sigma _{u}{\rm U}_{it}\right| =\sigma _{u}\left| {\rm U}_{it}\right| ; {\rm U}_{it}\sim N\left(0, 1\right) {\rm \ and\ } v_{it}=\sigma _{v}V_{it}, V_{it}\sim N\left(0, 1\right) $$
Maximum simulated likelihood estimator (Greene, Reference Greene2010), which addresses the issue of selectivity in stochastic frontier models with cross-sectional data, is of particular interest. The study utilizes the model presented by Bravo-Ureta et al. (Reference Bravo-Ureta, González-Flores, Greene and Solís2020) to examine the separate impact of rural-urban migration on technical efficiency change. Following Heckman (Reference Heckman1979) sample selection model for linear regression is expressed as:
$$D_{i}=1\left[\propto^{\! \prime} Z_{i}+w_{i}\gt 0\right], w_{i}\sim N\left(0, 1\right); $$
$$Y_{it}=\beta^{\! \prime}X_{it}+\omega _{it}, \omega _{it}\sim N(0, \sigma _{\omega }^{2}];$$
$$\left(w_{i},\omega _{it}\right)\sim N_{2}\left\{\left(0, 0\right), \left(1, \rho \sigma _{\omega },\sigma _{\omega }^{2}\right)\right\};$$
(Y it , X it ) observed only when D i =1.
The Heckman’s framework is used to estimate the model (Heckman, Reference Heckman1979). The framework consists of two steps: The first step involves estimating a Probit model using maximum likelihood estimation to generate an Inverse Mill’s Ratio (IMR) for each observation. The second step involves using OLS to estimate a linear regression of Y it , incorporating the IMR as a regressor in equation (2) while using the observed subsample. To account for the IMR being a constructed regressor, standard errors are adjusted, and the Full Information Maximization Likelihood Estimation is used. However, Bravo-Ureta et al. (Reference Bravo-Ureta, González-Flores, Greene and Solís2020) extends the Heckman’s model by combining the models in (1) and (2) and (3) in a stochastic frontier (SF) model for a panel data setting, rather than a linear regression as:
$$D_{i}=1\left[\propto^{\! \prime} Z_{i}+w_{i}\gt 0\right], w_{i}\sim N\left(0, 1\right); $$
$$Y_{it}=\beta^{\! \prime}X_{it}+\omega _{it}, \omega _{it}\sim N(0, \sigma _{\omega }^{2}];$$
(Y it , X it ) observed only when D i =1;
$${\varepsilon _{it}}=v_{it}-u_{it};$$
$$u_{it}=| \sigma _{u}{\rm U}_{it}| =\sigma _{u}| {\rm U}_{it}| {\rm \; where} \; {\rm U}_{it}\sim N(0, 1);$$
$$v_{it}=\sigma _{v}V_{it} {\rm \; where \; } V_{it}\sim N(0, 1);$$
$$(w_{i},v_{it})\sim N_{2}\{(0, 0), (1, \rho \sigma _{v},\sigma _{v}^{2})\}.$$
For selection bias control stemmed from the unobserved factors, SPF sample selection (SS) model alongside the error terms is employed (Greene, Reference Greene2010; Bertrand et al., Reference Bertrand, Duflo and Mullainathan2004) and is specified as:
$${\rm sample} \;{\rm selection}\!:\, M_{it}=1\left[\tau{'} H_{it}+\delta _{it}\gt 0\right], \;\delta _{it} \; N(0, 1)$$
$${\rm SPF}:\, y_{it}=\beta {'} X_{it}+\omega _{it}, \omega _{it} \;N(0, \sigma _{\omega }^{2})$$
$$M_{it} =1 \, \rm if\ observed \, (y_{it}, X_{it})$$
$$ {\rm Error\ terms\!:\,}\mu _{it}=V_{it}-U_{it}$$
$$\mu _{it}=| \sigma _{\mu }U_{it}| =\sigma _{\mu }|U_{it}|, {\rm where} \;U_{it} \; N(0, 1)$$
$$V_{it}=\sigma _{v}v_{it}, \;{\rm wheren} \;\sigma _{\mu } \;N(0, 1)$$
$$\delta _{it}, V_{it} N_{2} \;\{\left(0, 1\right), \left(1, \gamma \sigma _{V}, \sigma _{V}^{2}\right)\}$$
where M it = rural-urban migration statusFootnote 2 , H it = set of independent factors in the SS model, and δ it = unobservable random term, y it = maize output value, X it =set of inputs in the SPF model, ω it =composed random term, τ′ and β′= unknown coefficients to be computed. Mostly, the elements in the random terms structure agree with those normally added in the SPF. The coefficient γ implies that selection bias may be present or absent due to unobserved variables (Greene, Reference Greene2010).
Moreover, the study considered the Cobb-Douglas (CDFootnote 3 ) and Translog (TL) functional forms to compute efficiency parameters (Bravo-Ureta et al., Reference Bravo-Ureta, Solis, Lopez, Maripani, Thiam and Rivas2007). To determine which model to use, a maximum likelihood ratio test (MLRT) is used to justify the acceptance of CD model over TL model. In order words, the 5% significance level of the MLRT rejects the TL model. Therefore, the likelihood ratio test was 89.17 (p = 0.005) indicated that use of CD over TL.
Following Bravo-Ureta et al. (Reference Bravo-Ureta, González-Flores, Greene and Solís2020), the CD empirical specifications adequately fits the data, given its assumption and expressed as:
$$\ln Y_{it}^{k}=\ln\beta _{0}^{k}+\sum _{j=1}^{5}\beta _{it}^{k}\ln X_{jit}+ V_{it}-U_{it}$$
where Y it =i th farmer’s log total value product, X lit = amount of the l th input, β=unknown coefficients to be computed, V it and U it = consist random term.
The measurement and definition of the explanatory variables are presented in Table 1, while the mean differences in variables are reported in Table A1.
Explanatory variables employed in the models
3. Empirical results and discussion
3.1. Definitions of variables used in the production estimates
Tables 2 and 3 shows the summary statistics of production variables. It also indicated the matching options for baseline and endline across the treatment status. Overall, treated farmers had a higher total value of maize in Ghana cedis than untreated both at the baseline and endline survey. Among the treated farmers, the highest average TVP was GHS2004.45 appears in the unmatched panel. Similar results are reported in the untreated farmers group. Comparing baseline to the endline results of the treated farmers, the results showed that endline results are greater than baseline results. However, the reverse was observed among untreated farmers group. For instance, a typical untreated maize farmer in the common support had a reduction in TVP at the endline survey. Turning on the conventional factors (i.e., LAND and LABOR), the study results revealed that untreated farmers tend to operate under large farms compared to the treated farmers at the baseline. At the endline, the treated farmers expanded land operations resulting in large farms. Most treated farmers spent considerable number of days on their farms greater than untreated farmers across the two matching options.
Summary statistics of production variables
Source: Author’s computations, 2023, GHS is the Ghanaian cedi which is the local currency. Exchange rate was GHS1.00 = US$ 0.0878, on 06/13/23.
Summary statistics of production variables
Source: Author’s computations, 2023.
Obviously, the treated farmers spent relatively more on seed and agrochemicals than the untreated in all observations. With regards to the climate-smart agricultural technologies (CSATs) including IMPROVED SEED, ZERO TILLAGE, AND ROW PLANTING, the results showed that most farmers sought for these technologies to improve maize output value. Specifically, treated farmers in the common support and NN 1-1 groups adopted most of the CSATs. Among the three CSATs, majority of the farmers adopted the improved seed technology. Furthermore, the location variables showed that 6 districts were captured as geographic differences in production. It was evident that most farmers came from GUSHEGU district, and this was followed by the SAVELUGU and NANTON.
3.2. Factors influencing rural-urban migration
The results of the first stage of the logit estimates of the PSM involving 466 baseline observations are presented in Table 4. Prior to the estimations, the distribution of propensity scores (Figure A1) and test of balancing covariates was obtained using both nearest neighbor matching (Table A2) and common support techniques (Table A3) and these covariates were found not to be significant implying substantive similarities between migrants and non-migrants; hence, differences in outcomes can be attributed to the treatment. Furthermore, we employed the logit model for the purpose of computing the propensity scores, which was further applied to obtain matched samples of treated and untreated households. To achieve this goal, we employ two extreme matching procedures, notably, treated and untreated samples which are on common support and nearest neighbor (NN) 1-1 without replacement. Whereas common support employs all overlapping observations based on the PSs, the NN 1-1 significantly decreases the number of observations entering the analysis. These two matching techniques were selected to test the robustness of the impact estimates.
Logit results for rural-urban migration using baseline data
Saboba is used as reference group.
Source: Author’s computations, 2023.
1% = ***, 5% = **, 10% = *.
Our results show the probability of farm household migrating was influenced by educational level, livestock value, off-farm activity, remittance, farm plot, lag maize yield, access to electricity, migration history, time trend, distance, and location. There is a negative and statistically significant relationship between the age of the household head and the likelihood of rural-urban migration. Specifically, each additional year of age is associated with a 9.4% decrease in the probability of migration. This suggests that older farmers, who are more accustomed to the rural lifestyle, exhibit a higher level of reluctance to migrate compared to their younger counterparts. Alarima (Reference Alarima2018) indicated that younger people are more propelled to migrate than older people in Nigeria. Educational level plays a significant role in rural-urban migration, and it was categorized as basic, secondary, and tertiary education, with no formal education as the reference group. In comparison to those with no formal education, farmers with secondary and tertiary education exhibit a positive and significant influence on the probability of migration. Farmers who have completed secondary education are 33.6% more likely to migrate to urban areas, while those with tertiary education show a 41.6% higher probability of migration. Farmers who have completed basic education also demonstrate an increased inclination to migrate to urban centers. This finding aligns with previous studies conducted in China (Mullan et al., Reference Mullan, Carolina, Grosjean and Kontoleon2011; Shi, Reference Shi2020; Tay et al., Reference Tay, Tai and Tan2022), Ghana (Alhassan Reference Alhassan2017; Kutor et al., Reference Kutor, Annan-Aggrey, Poku, Kyeremeh and Arku2022; Marta et al., Reference Marta, Fauzi, Juanda and Rustiadi2020) and Nigeria (Alarima, Reference Alarima2018) which have indicated that educated people have a higher tendency to migrate to urban areas. The results show that GHS1 increase in remittance received by farm households corresponds to a 7.9% increase in the likelihood of migration. This suggests that higher remittances received by the household head tend to serve as an incentive for migration. It is possible that the higher income obtained through migration, which often exceeds the earnings from farming, motivates household heads to migrate themselves or encourage other family members to migrate. The results align with the findings of Kapri and Ghimire (Reference Kapri and Ghimire2020) and Fengbo et al. (Reference Fengbo, Lucas, Bloom and Shijun2016).
The study findings indicate that both livestock value and farm plot ownership have a positive and significant influence on migration. This suggests that household heads who possess additional livestock or farm plots are more likely to migrate to urban areas. One possible explanation for this is that farm household heads can readily convert livestock and farm plots into cash, providing them with the financial means to pursue migration. Similar finding was observed in Nigeria (Nwaru and Iheke, Reference Nwaru and Iheke2015) and Ghana (Ackah and Medvedev, Reference Ackah and Medvedev2012; Tanle et al., Reference Tanle, Ogunleye-Adetona and Arthor2020). Off-farm activity shows a negative significant influence on the likelihood of migration. Specifically, farmers who engage in off-farm activities are discouraged from migrating, with a decrease of 8.2% in the probability of migration. This is because farmers who participate in off-farm work may earn higher incomes compared to what they could potentially earn through urban employment. This contradicts the findings of Darko (Reference Darko2013) that migration in Ghana is positively associated with off-farm work.
Access to basic amenities, such as electricity, has a negative and significant influence on the probability of migration. Specifically, farmers who have access to electricity are 9.8% less likely to migrate to urban areas. The availability of electricity in rural areas may contribute to improved living conditions and economic opportunities, reducing the need for farmers to migrate in search of better amenities or livelihoods. This concur with findings of past studies (Ackah and Medvedev, Reference Ackah and Medvedev2012; Alhassan, Reference Alhassan2017; Kutor et al., Reference Kutor, Annan-Aggrey, Poku, Kyeremeh and Arku2022). Migration history of farm households has a positive and significant effect on the probability of migration. This suggests that farm households with a previous history of migration have 11.6% higher likelihood of migrating to urban centers. The migration history of a household can influence future migration decisions, as past experiences and networks established in urban areas may create a familiarity and preference for urban living.
Lag maize output has a negative and significant influence on the likelihood of migration. Specifically, a decrease in maize output by 1 kg in the previous year corresponds to a 1.6% increase in the probability of migration. This finding is consistent with previous studies (Ackah and Medvedev, Reference Ackah and Medvedev2012; Atta-Ankomahand and Osei Reference Atta-Ankomah and Osei2021), indicated that the level of maize output value among farmers directly influences rural-urban migration. Ackah and Medvedev (Reference Ackah and Medvedev2012) found that lower maize yields in the previous year indicate a reduction in agricultural productivity and potential financial constraints for farmers. In such circumstances, farmers may seek alternative livelihood opportunities in urban areas, where they perceive better economic prospects and stability. Atta-Ankomahand and Osei (Reference Atta-Ankomah and Osei2021) also indicated that the need to secure a reliable income source and overcome the challenges associated with low agricultural productivity can motivate farmers to consider migration as a viable option. Relative to Saboba district as the based group, farmers from Karaga and Tolon districts has a positive significant effect on rural-urban migration. The results indicated that farmers are more propelled to migrate compared to the reference group.
Moreover, we explore the role of remittance in the impact of rural-urban migration on technical efficiency (Table A4). The results demonstrate that rural-urban migration significantly reduces farm households’ TE (total effect = −0.038, p < 0.05) in the short term, mainly due to labor loss (direct effect = −0.048, p < 0.001). However, remittances have a positive indirect effect (0.010, p < 0.05), offsetting about 26% of this impact by enabling investments in productivity. This duality aligns with the “migration paradox” literature (e.g., Rozelle et al., Reference Rozelle, Taylor and DeBrauw1999), where short-term labor losses conflict with long-term capital accumulation. The results support integrated rural development strategies by addressing the direct labor shock through mechanization subsidies or training for remaining household members, while amplifying indirect benefits via financial inclusion programs.
3.3. Difference in difference using traditional OLS
The difference in difference estimates using the traditional OLS is presented in Table 5. The study results revealed that R-squared indicated that explanatory variables predicted variations in the dependent variable. The F-test of all the models suggested the simultaneous influence of the explanatory variables on the dependent variable. Variables such as AGROCHEMICALS, LABOR, TIME, IMPROVED SEED, and location (GUSHEGU) significantly influenced the different matching options. The positive and significant coefficient of AGROCHEMICAL suggest that 1% increase in agrochemicals, farmers TVP increased by 0.58–0.106 across matching options. LABOR variable positively and significantly influences TVP of the farmers. This implies that 1% change in LABOR in man days, the value output of maize increased from 0.067 to 0.116. The time variable showed the TVP growth of the farmers. It can be observed that TVP of maize production, decreases overtime among the farmers. However, the impact of rural-urban migration on TVP (i.e., interaction of time and migration status) was positively significant at 1%. This suggested that the casual effect of rural-urban migration was a 2.8% increase in TVP for NN group of farmers. Although the rural-urban migration variable (treatment) was not statistically significant, the DID model incorporates adjustments for baseline differences. The study further observed that adoption of IMPROVED SEED increased TVP from 9.8 to 11.7% at 1% significance level. This means that there is a positive association between the adoption of improved seed and TVP.
Difference in difference estimates using traditional OLS with different matching
Note: ***, **, * are 1, 5, 10% level.
Source: Author’s computations, 2023.
SE = standard errors. Saboba used as base group. NNM = Nearest neighbor matching. RUM = rural-urban migration.
The positive significant coefficient of land implies that a unit change land size, the TVP of maize farmers increased by 7% among the NNM group. While zero tillage has a positive and significant effect on TVP for common support groups of farmers, adoption of MECHANIZATION positively and significantly influenced TVP for unmatched and NNM groups of farmers. Compared to the SABOBA district, the study revealed that GESHUGU has a positive significant impact on TVP across the three options. However, farmers located in KARAGA district tend to have negative and significant effects on TVP for unmatched groups but positive significant effect for common support groups. At the significance level of 1%, other location variables positively and significantly influenced TVP for only the common support group of farmers.
3.4. Different matching options using conventional SPF
Table 6 presented the stochastic production frontier (SPF) with different matching options. The study showed that the λ was statistically significant different from zero across various matching options. This means that there is technical inefficiency in the observed maize output. Also, the results suggested that the SPF model best fits the data than the standard production function. The study results revealed that similar findings were observed from the SPF compared to the traditional OLS model. The LAND variable only showed a positive and significant effect on value output in column 4. This implies that farmers who owned additional hectare under maize production are better off increasing maize output value by 7.1%. The findings confirmed the significant role of agrochemicals in increasing maize output value across all the matching options. A unit change in the amount spent on AGROCHEMICALS increase output value from 0.59 to 0.107.
Stochastic frontier estimates with different matching
Note: ***, **, * = 1, 5, 10% level.
Source: Author’s computations, 2023.
SE = standard errors. Saboba used as base group. NNM = Nearest neighbor matching.
Regarding the TIME variable, the results indicated that output value dwindled from 0.65 to 0.144 across all the matching options. This suggested that the output value growth in the study area declined over time. Nevertheless, the impact of rural-urban migration on maize output value was positively significant at 1% in column 4. Adoption of IMPROVED SEED has a positive significant impact on TVP, implying an increase from 0.098 to 0.118 across all matching options. In columns 2 and 4, adoption of MECHANIZATION positively and significantly influenced the output value. But farmers who are in the common support had output value increased by 0.011. Location variables specifically GUSHEGU district showed a positive significant effect on value of maize output value across all matching options compared to the SABOBA district. In addition, other districts except KARAGA positively and significantly influenced value of maize output.
3.5. SPF with sample selection
Given the nature of the self-selection bias from Table 7, the study addressed this issue by employing the sample selection SPF. Table 7 presents the estimations of the sample selection SPF for all the matching options. Based on Table 7, the estimates of the selectivity parameter ρ are statistically different from zero for all three datasets, indicating the presence of sample selection bias. Therefore, it is preferable to implement the sample selection framework rather than the conventional SPF, as the latter may lead to biased estimates that could affect the technical efficiency scores. In addition, the study’s findings indicated that more stringent matching techniques can help reduce biases arising from self-selection, not only from observable characteristics, but also from unobservable factors (example motivations and production skills) (Rahman, Reference Rahman2011; Villano et al., Reference Villano, Bravo-Ureta, Solís and Fleming2015).
Sample selection SPF estimates with different matching
Note: ***, **, * = 1, 5, 10% level.
Source: Author’s computations, 2023.
SE = standard errors. Saboba used as base group. NNM = Nearest neighbor matching.
As anticipated, all the models estimated showed a positive and partial production elasticities which indicate the proportionate contribution of each input to the percentage change in yield. Turning on conventional inputs such as LAND, LABOR, and AGROCHEMICALS, the results revealed positive and significant values that are consistent with previous studies. LAND and LABOR are the inputs that make the largest contribution to maize output among the groups of farmers studied, as evidenced by their expected positive and significant effects, significant at least at the 5% level. This finding aligns with earlier findings (Abdul-Rahaman and Abdulai, Reference Abdul-Rahaman and Abdulai2018; Abate et al., Reference Abate, Francesconi and Getnet2014; Tleubayev et al., Reference Tleubayev, Bobojonov and Götz2022; Jayne et al., Reference Jayne, Chamberlin and Headey2014; Rahman, Reference Rahman2011; Tirkaso and Hansson, Reference Tirkaso and Hansson2023). Abdul-Rahaman and Abdulai (Reference Abdul-Rahaman and Abdulai2018) assert that land and chemicals increase rice output and efficiency among farmers in Ghana. In Africa, Jayne et al. (Reference Jayne, Chamberlin and Headey2014) highlighted the fact that farmers in Africa are challenged with land and credit availability. LABOR makes a significant contribution to the value of output across all matching options, likely due to increasing marginal output value emanating from a shortage of labor in farm households due to migration. This finding is consistent with the results of Mensah and Brümmer (2016), who found a positive impact of labor supply in mango farmers’ yield in Ghana. However, the result contradicts with the findings of previous studies (González-Flores et al., Reference González-Flores, Bravo-Ureta, Solís and Winters2014; Rahman et al., Reference Rahman2011) which found decreasing marginal productivity of labor.
Also, AGROCHEMICALS make a significant contribution to maize output value across the matching options, while SEED plays a minor role, thus insignificant. This agree with other findings which found that fertilizer and pesticide expenditure to increase farmers’ crop output in Indonesia (Alwarritzia et al., Reference Alwarritzia, Nansekib and Chomeic2015), and Ghana (Danso-Abbeam and Baiyegunhi, Reference Danso-Abbeam and Baiyegunhi2020; Denkyirah et al., Reference Denkyirah, Okoffo, Adu, Aziz, Ofori and Denkyirah2016).
Regarding CSATs adoption (IMPROVED SEED, ZERO TILLAGE, and ROW PLANTING), the study results revealed a positive significant contribution of these technologies to enhance maize output value. Among these technologies, IMPROVED SEED exhibits the highest partial elasticity hovering from 0.10 to 0.117, followed by ZERO TILLAGE. This finding is in line to Keil et al. (Reference Keil, D’souza and McDonald2017) and Meena et al. (Reference Meena, Rajesh and Beer2016).
The parameter for adoption of MECHANIZATION exhibited a highly significant and positive relationship in all of the models, confirming the importance of mechanization in agriculture (Cossar, Reference Cossar2019; Li et al., Reference Li, Ma, Botero-R and Quoc Luu2023). The study’s findings is consistent with the results of Bravo-Ureta et al. (Reference Bravo-Ureta, González-Flores, Greene and Solís2020) which found that farmers who use mechanical or animal power are more likely to achieve higher output than those who use traditional farming methods.
Location variables, which account for variations in the environment, biophysical factors, and socio-economic conditions, such as the impact of neighboring farms, the adoption of new technologies, and access to information and inputs, all of which are known to have an influence on crop yield. Compared to the reference group (SABOBA district), farmers located in GUSHEGU, KARAGA, NANTON, SAVELUGU and TOLON districts significantly obtained higher output value. Similar findings found that difference in location impacted crop output (Abdul-Rahaman and Abdulai, Reference Abdul-Rahaman and Abdulai2018).
Moreover, all the models feature negative and significant coefficient for the common time trend variable (TIME), indicating a decrease in output over time. This is likely because of labor shortages as a result of out-migration that occurred in the area during the period of analysis (GSS, 2014; Mahama, Reference Mahama2013; Wouterse, Reference Wouterse2010a, Reference Wouterse2010b). However, the interaction coefficient between RUM and TIME, denotes the impact (ATET) of rural-urban migration, is positive significant in all matching options with estimates ranging from 0.146 to 0.287.
3.6. Impact of rural-urban migration on maize TVP
Table 8 presents the impact of rural-urban migration on maize total value products (TVP), as well as the frontier estimates obtained from three datasets. It also presents the frontier estimates from conventional and sample selection models. These findings confirm the negative shock on output resulting from high rural-urban migration and drought conditions in the study area. Tambo (Reference Tambo2016) indicated that the Northern region of Ghana is known for experiencing frequent harsh climate change, which have a negative impact on agriculture. However, holding the negative TIME trend and other variables constant, the results showed that rural-urban migration had a positive impact, as indicated by the positive DID estimates. The highest DID values are obtained from common support matching at GHS160.59 and GHS155.14 for conventional and sample selection SPF, respectively. Additionally, the estimated impact is lower for all three datasets when the sample selection SPF correction is implemented, underscoring the importance of accounting for selectivity bias.
Impact of rural-urban migration on maize output value
Source: Author’s computations, 2023.
3.7. Impact of rural-urban migration on TE
Table 9 presents the mean changes in technical efficiency (TE) at baseline and endline for both migrant and non-migrant groups of farmers, with no significant difference in the distribution of TE across models, time, and treatment status. Also, Figure 3 presents the kernel density estimates for these scores. It indicates that the distributions of TEs across models, time, and treatment status are not significantly different. This resonates with the idea that the impacts of an event require time to be fully felt or productive (Teye et al., Reference Teye, Boakye-Yiadom, Asiedu, Litchfield, Wilson, Kubi and Awumbila2019). While farm households were generally technically inefficient, migrant households exhibited greater technical efficiency compared to non-migrant households. These results support the idea that rural-urban migration requires time for farm households to become fully productive. Furthermore, there is evidence suggesting that rural-urban migration initially had a negative effect on the TE of treated farmers, but as they gradually sought new technologies and productive practices, TE scores improved. Additionally, the targeted and continuous financial assistance provided by the migrants could have contributed to keeping TE levels similar to the untreated group. However, the average TE scores for both treated and control groups are lower compared to the findings of Bravo-Ureta et al. (Reference Bravo-Ureta, Solis, Lopez, Maripani, Thiam and Rivas2007). These results imply that improving managerial performance could have a substantial positive impact on treated farmers and help enhance TE in the Northern region of Ghana.
Mean technical efficiency scores with different matching among baseline and endline
Source: Author’s computations, 2023.
Kernel distributions of TE scores for alternative matched samples.
4. Conclusion and policy recommendations
This paper evaluates the impact of rural-urban migration on technical efficiency change in maize production. We used two rounds balanced panel data from 1056 maize households in northern Ghana between 2014 and 2018. To account for selectivity bias arising from observed and unobserved factors, propensity score matching combined with differences in differences with stochastic production frontier is employed.
The results show that the self-selection SPF estimates were significantly lower, confirming the correction of unobserved factors. The average values from the sample selection SPF and conventional SPF were GHS160.59 and GHS155.14, respectively, providing evidence that migrant farm households achieved notably better results than the control group. This suggests that unobserved factors were affecting the accuracy of SPF estimates and that correcting for them can lead to more accurate results. These findings highlight the importance of accounting for unobserved factors when estimating SPF and can inform policy decisions related to agricultural production and rural-urban migration in Ghana. Also, farm households were technically inefficient, but migrants exhibited lower levels of technical inefficiency compared to non-migrants.
The study provides evidence of selection bias, highlighting the need for a more comprehensive approach to improved agricultural output value based on rural-urban migration dynamics. The study recommended that there should be a focus on capacity building. Technical capacity development for farmers can be complemented with effective extension services, particularly in input applications, to increase yield and technical efficiency. Continuous provision of agricultural extension services and ensuring a reliable supply of agricultural technologies such as improved seedlings and agrochemicals, and adoption of climate smart technologies could replace labor and improve agricultural output. The significant difference between the sample selection SPF and conventional SPF estimates highlights the role of unobserved factors such as behavioral traits, risk preferences, local knowledge, and contextual constraints in shaping farm performance. This implies that policies designed solely around observable characteristics (e.g., farm size, education level, or input use) may overlook critical drivers of productivity and thus risk poor targeting or limited effectiveness. To improve outcomes, agricultural policy must move beyond one-size-fits-all solutions and adopt a more holistic approach that explicitly considers both observed and unobserved heterogeneity among farm households. This includes incorporating behavioral insights into program design, tailoring interventions to local conditions, and developing tools (such as participatory extension models or digital advisory platforms) that can adapt to context-specific needs. In migration-affected communities, such a nuanced approach will better support the diverse pathways through which households adjust and sustain agricultural productivity.
While our dataset records only monetary remittances received, it does not capture the value of in-kind transfers such as food or other goods. As such, our analysis focuses solely on cash remittances. Also, the data lack detailed information on the timing, duration, or cyclicality of migration episodes. It is possible that some household members migrated before 2010 and returned prior to baseline or migrated between waves and returned before endline. In such cases, lingering effects such as accumulated remittances or changes in household decision-making could still influence technical efficiency, even if the migrant was present during the survey. Future research would benefit from more comprehensive data that capture both in-kind remittances and detailed migration histories to better understand their influence on agricultural productivity and household welfare.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/aae.2025.10019.
Data availability statement
Data used for this research can be obtained from the Economic Growth Centre (ECG) at Yale University and is available at: https://yale.app.box.com/s/ww9pqfb1lf9bwvhtocs51ixd7v4qpv2s/folder/135707578976.
Acknowledgements
We acknowledge the Economic Growth Centre (ECG) at Yale University, the Global Poverty Research Lab at Northwestern University, and the Institute of Statistical Social and Economic Research (ISSER), at the University of Ghana, Legon for providing the Ghana Socioeconomic Panel Survey (2013/14 and 2017/18) which was used in the analyses.
Author contribution
Conceptualization, S.P., B.O.A., O.T., and R.V.; Methodology, S.P., B.O.A., O.T., and R.V.; Formal Analysis, S.P., B.O.A. and O.T.; Data Curation, S.P., B.O.A., O.T.; Writing – Original Draft, S.P., B.O.A., O.T.; Writing – Review and Editing, S.P., B.O.A., O.T., and R.V.
Financial support
The authors declare that there is no funding associated with this study.
Competing Interests
The authors declare that there are no known conflicts of interest with this submission.
Appendix.
Histogram distribution of propensity score.
Open access
