This study investigated the effect of rural–urban migration on skill development in South Africa, focusing on the role of educational attainment and occupational development in shaping individuals’ skill profiles.
National Income Dynamics Study (NIDS) data was used. Propensity score matching (PSM) was employed to obtain a matched group of rural residents and rural-to-urban migrants with similar probabilities of migrating from rural to urban areas. Linear regression analyses were performed after the PSM analysis.
The results of the regression analysis after PSM indicated that rural-to-urban migrants reported enhanced skills development compared to rural residents. Sensitivity analysis using different logarithms yielded similar results. Promoting access to education and vocational training in rural areas can mitigate migration and enhance local skill development.
This study addresses a significant gap in the literature by examining the impact of rural–urban migration on skills development, a topic rarely explored in existing literature. PSM provides robust insights into how migration influences human capital by focusing on education and occupational development.
The peer review history for this article is available at: https://publons.com/publon/10.1108/IJSE-09-2024-0745
1. Introduction
The movement of populations from rural to urban landscapes is a widespread trend that has significantly shaped the socioeconomic dynamics of many nations (Burlacu et al., 2022; Ying, 2024). Throughout history, people have moved from rural areas to cities in search of better jobs, more educational opportunities, and a higher quality of life (Nagendraiah and Ravindra, 2019). This trend is not confined to any specific region but is a hallmark of economic development worldwide (Tacoli et al., 2015; Hossainand Tasnim, 2023). As nations industrialise, the shift from agrarian-centred economies to more diversified urban economies becomes inevitable, with migration being the catalyst for this transformation (Wu et al., 2018).
In developing countries, this movement is particularly pronounced because of the rapid pace of industrialisation and urbanisation (Andersson and Molinder, 2024). However, these prospects often encounter significant challenges, such as the strain on urban infrastructure, the rise of informal settlements, and the persistence of economic inequalities. Understanding the dynamics of rural-urban migration, with a specific emphasis on elements such as education, unemployment, livelihoods, and economic impacts, can help tackle the root causes of migration and reduce inequalities in rural areas (Farrell, 2017; Mukhtar et al., 2018).
South Africa presents a unique case in this broader context. The country has witnessed a substantial increase in internal migration, with individuals moving from rural to urban areas in search of better opportunities (Adetiba, 2022). This migration is deeply intertwined with the country’s historical, economic, and social fabric. The legacy of apartheid, characterised by forced removals and restricted mobility, has left enduring imprints on migration patterns and urban development (Bakker et al., 2020). In the post-apartheid era, the lifting of these restrictions has led to accelerated urbanisation, with significant implications for both rural and urban communities (Mthiyane et al., 2022).
Rural-urban migration can also be understood through the lens of the Human Capital Theory, which suggests that individuals migrate to urban areas to invest in skills, education, and work experience, expecting higher returns in terms of earnings and employment opportunities (Selod and Shilpi, 2021). Research indicates that rural-urban migration can reduce the chances of unemployment for migrants compared to those who remain in rural areas (Mbu and Chuo, 2020). Additionally, migration often leads to the redistribution of labour from less productive agricultural sectors to more dynamic industrial and service sectors, contributing to economic development (Liao and Yip, 2018). However, while the economic impacts of migration, such as increased household incomes through remittances, have been well-documented, less attention has been paid to how rural-urban migration specifically affects skill development among migrants in South Africa (Mitra, 2019; Keopasith and Neng, 2020).
The transition from rural to urban areas often necessitates new skills that align with the demands of more industrialised economies (Andersson and Molinder, 2024). The existing literature suggests that migrants may acquire new skills and experience in urban areas, potentially enhancing their employability and economic prospects. However, the extent to which this occurs, particularly in the context of South Africa, remains underexplored and is investigated in this study. This gap is significant given the country’s unique socio-economic landscape and the potential for rural-urban migration to influence skill acquisition and economic growth. Understanding the relationship between migration and skill development is particularly important in South Africa, where disparities in skills and employment opportunities contribute significantly to economic inequality (McGrath and Akoojee, 2007; Reddy and Mncwango, 2021).
This study aims to fill this gap by examining the relationship between rural-urban migration and skill development, focusing on South Africa. Using data from the National Income Dynamics Study (NIDS) and employing propensity score matching (PSM), this study provides a nuanced understanding of how modern rural-urban migration influences skill development. These findings contribute to formulating policies that address the inequalities faced by those remaining in rural areas and promote more inclusive and balanced economic development.
2. Methodology
2.1 Data
The NIDS, a quantitative dataset, the only nationally representative individual-level panel dataset in South Africa, was used to examine the relationship between rural-urban migration and skill development. The dataset covered a range of variables related to migration, income, education, demographic characteristics, employment status, household assets, and living conditions. Comprising a sample size of more than 28,000 individuals across 7,300 households (Brophy et al., 2018), the NIDS survey has been conducted every two years since 2008 by the Southern African Labour and Development Research Unit (SALDRU) at the University of Cape Town (Brophy et al., 2018). The primary objective of the NIDS is to monitor individuals longitudinally, making it an ideal source for investigating the long-term impacts of rural-urban migration.
For this study, all waves of the NIDS, from the first wave in 2008 to the most recent wave in 2017/2018, were used. The sample was restricted to individuals aged >18 years in the first wave. Skill development was the primary outcome variable in this study and was generated using a composite index that included education level, occupation, income, and employment status. Principal Component Analysis (PCA) was used to derive the skill development score, as education, occupation, income, and employment status are well-recognised indicators of human capital and skill acquisition (Becker, 1964; Mincer, 1974; Rohmah et al., 2023; Schultz, 1961).
2.2 Empirical analysis
To estimate the impact of rural-urban migration on skill development, PSM, a statistical technique that allows for comparing groups (in this case, migrants and non-migrants) as if they are randomly assigned, was employed. This approach helped control for selection bias by ensuring that the groups being compared were similar in terms of their observed characteristics, thus making the comparison more credible (Garcia Iglesias, 2022).
A logistic regression model was used to estimate the propensity scores, representing the probability of an individual being a rural-urban migrant given their observed characteristics, as described in a previous study by Nyoni and Kollamparambil (2022). The model is specified as follows:
Where Λ denotes the logistic cumulative distribution, migration is an indicator variable for rural-urban migration, Xi is a vector of covariates, α is the intercept, β is the vector of coefficients, and εi is the error term.
A migrant is defined as someone who was living in a rural area in Waves 1 (2008) and 2 (2010) and moved to an urban area in Wave 3 (2012). Individuals who did not move or who moved in other patterns (e.g. urban to rural, urban to urban) were excluded from the analysis to maintain focus on rural-urban migration. The treatment group included individuals who relocated from rural to urban areas between 2008 and 2010. By contrast, the control group comprised individuals who remained in rural areas in 2008 and 2010.
The covariates included in the PSM model were carefully selected based on their relevance to the likelihood of migration and skill development. These covariates included age, gender, household size, race, marital status, health status, perceptions of crime frequency, socioeconomic status, and access to basic services such as sanitation and electricity. The selection of variables included in the model was supported by previous studies, which identified similar factors as important in understanding migration patterns and their effects on skill acquisition (Selod and Shilpi, 2021).
Several matching algorithms were used to ensure the robustness of the model, including nearest-neighbour matching, radius matching (calliper of 0.1), and kernel matching (bandwidth of 0.06). Using multiple algorithms helped to cross-validate the findings, ensuring that the estimated effects of migration on skill development were not overly dependent on a single matching technique (Becker and Ichino, 2002). Consistency across different matching methods strengthens confidence in the findings obtained.
The average treatment effect on the treated (ATT) for each algorithm was calculated to assess the impact of migration on skill development:
Where Y1 is the skill development outcome for migrants, and Y0 is the counterfactual outcome for individuals who did not migrate.
2.2.1 Post-PSM regression analysis
A regression analysis of the matched sample was conducted to further explore the relationship between rural-urban migration and skill development. A model specification similar to that by Nagendraiah and Ravindra (2019) was adopted in a study on rural migration, education attainment, and human capital. The regression model is as follows:
Where skillDevi is the skill development index for individual i, migrationi is a binary variable indicating whether individual i is a rural-urban migrant or non-migrant (0), Xi represents other covariates previously included in the migration model. The regression model (3) helped to isolate the effect of rural-urban migration on skill development while controlling for other relevant factors.
A placebo test was conducted to address the potential impact of omitted and unobservable variables on the accuracy of the estimated treatment effects in the PSM model. This involved estimating the treatment effects using an alternative outcome variable that was not expected to be affected by migration. The placebo test helped verify that unobserved cofounders did not drive the estimated effects.
Post-matching, the standardised mean bias across covariates was examined to ensure that the treated and control groups were comparable. The balance of covariates indicates that the matching process effectively reduced bias, making the treated and control groups more similar regarding observed characteristics.
3. Empirical results
3.1 Descriptive statistics
The descriptive summary provides an overview of the socio-economic characteristics and outcomes in South Africa, with a focus on rural-urban migration and skill development (Table 1). The sample consisted of individuals with an average age of 36 years and a gender distribution of approximately 59% female. Most of the sample were Black Africans, comprising 87.7% of the respondents. The household size varied, with an average of six members.
Sociodemographic characteristics of the sample
| Mean | SD | Min | Max | |
|---|---|---|---|---|
| Rural-urban migration | 0.044 | 0.205 | 0 | 1 |
| Skill development | −0.002 | 1.23 | −2.331 | 2.739 |
| Female | 0.589 | 0.49 | 0 | 1 |
| Age | 36.089 | 14.442 | 18 | 73 |
| Marital status | ||||
| Married | 0.245 | 0.43 | 0 | 1 |
| Living with partner | 0.09 | 0.286 | 0 | 1 |
| Widow/widower | 0.067 | 0.251 | 0 | 1 |
| Divorced/separated | 0.018 | 0.133 | 0 | 1 |
| Never married | 0.58 | 0.494 | 0 | 1 |
| Race | ||||
| Black Africans | 0.877 | 0.328 | 0 | 1 |
| HH size | 5.752 | 3.736 | 1 | 41 |
| Access to electricity | 0.789 | 0.408 | 0 | 1 |
| Access to sanitation | 0.27 | 0.444 | 0 | 1 |
| Health status | ||||
| Excellent | 0.308 | 0.462 | 0 | 1 |
| Very good | 0.311 | 0.463 | 0 | 1 |
| Good | 0.266 | 0.442 | 0 | 1 |
| Fair | 0.083 | 0.276 | 0 | 1 |
| Poor | 0.032 | 0.175 | 0 | 1 |
| Crime frequency | ||||
| Never happens | 0.18 | 0.384 | 0 | 1 |
| Very rare | 0.229 | 0.42 | 0 | 1 |
| Not common | 0.179 | 0.384 | 0 | 1 |
| Fairly common | 0.174 | 0.379 | 0 | 1 |
| Very common | 0.238 | 0.426 | 0 | 1 |
| Socioeconomic status | ||||
| Poorest | 0.221 | 0.415 | 0 | 1 |
| Poor | 0.214 | 0.41 | 0 | 1 |
| Middle | 0.184 | 0.388 | 0 | 1 |
| High | 0.205 | 0.404 | 0 | 1 |
| Highest | 0.176 | 0.381 | 0 | 1 |
Source(s): Authors’ own work
Regarding rural-urban migration, 4.4% of the sample migrated from rural to urban areas. Skill development measured on a standardised scale had a mean close to zero, indicating a balanced distribution of skill levels among the sample. A significant proportion of the sample was married or had never married, with smaller percentages living with a partner, widowed, or divorced/separated. Access to basic services revealed that 78.9% of the households had electricity, while only 27% had sanitation.
Most participants rated their health positively, with a significant proportion rating their health as excellent or very good. Perceptions of crime frequency varied, with notable percentages believing that the crime was either very rare or very common. Socioeconomic status was distributed across quintiles, with a relatively balanced representation.
3.2 Determinants of rural-urban migration
The logistic regression model used to estimate propensity scores identified several key predictors of rural-urban migration, as presented in Table 2. The results indicate that age significantly affect the likelihood of migration, with older individuals being more likely to migrate. The positive coefficient of age and the negative coefficient of age squared indicate a nonlinear relationship, suggesting that propensity increases at a decreasing rate. Additionally, the analysis revealed that individuals from larger households were less likely to migrate, and being African significantly increased the likelihood of rural-urban migration.
Logistic regression predicting propensity score of being a rural-to-urban migrant
| Migration status | ||
|---|---|---|
| Coeff | S.E | |
| Age | 0.106*** | (0.063) |
| Age2 | −0.002*** | (0.000) |
| Female | 0.090 | (0.083) |
| HH size | −0.188*** | (0.019) |
| African | 0.976*** | (0.140) |
| Marital status | ||
| Living with partner | −0.662*** | (0.162) |
| Widow/Widower | −0.507 | (0.358) |
| Divorced or separated | 0.092 | (0.273) |
| Never married | −0.104 | (0.113) |
| Health status | ||
| Fair | −0.095 | (0.406) |
| Good | −0.292 | (0.369) |
| Very good | −0.296 | (0.367) |
| Excellent | −0.363 | (0.367) |
| Frequency of crime | ||
| Very rare | 0.057 | (0.129) |
| Not common | 0.006 | (0.138) |
| Fairly common | −0.056 | (0.138) |
| Very common | 0.008 | (0.126) |
| Socioeconomic status | ||
| Poor | 0.068 | (0.106) |
| Middle | −0.244** | (0.121) |
| High | −0.457*** | (0.128) |
| Highest | −0.848**** | (0.208) |
| Access to sanitation | 1.183*** | (0.094) |
| Access to electricity | −0.131 | (0.119) |
| _cons | −3.850*** | (0.728) |
| N | 1,506 | |
| Pseudo R2 | 0.1502 | |
Note(s): *** Significant at 1% level, ** Significant at 5% level, *Significant at 10% level; Std errors in parentheses; Ref group: no schooling, excellent health status, crime never happens, male
Source(s): Authors’ own work
Socioeconomic factors also played a role, with wealthier individuals, particularly those in higher wealth categories, being identified as less likely to migrate than the poorest individuals. Marital status was found to influence migration decisions, with individuals living with a partner being less likely to migrate than married individuals. Access to sanitation was also a significant predictor, with a positive coefficient indicating that individuals with better sanitation were more likely to migrate.
3.3 Common support and matching quality test
The substantial overlap between the two groups indicates a good region of common support, suggesting that the matching process effectively balanced the covariates between migrants and non-migrants (Figure 1). This balance enhances the accuracy of estimations concerning the effect of treatment on skill development.
The two-panel density plot titled “Caliper equals 0.25 asterisk S D” presents side-by-side distributions of propensity scores. The left panel is titled “Rural-to-urban migrants” and shows a horizontal axis labeled “Pr (treatment)” ranging from 0 to 0.5 in increments of 0.1, and a vertical axis labeled “Density” ranging from 0 to 8 in increments of 2 units. In the left panel, a shaded histogram appears with the data as follows: Pr (treatment): 0 to 0.05, Density: 6.52. Pr (treatment): 0.05 to 0.1, Density: 4.726. Pr (treatment): 0.1 to 0.15, Density: 3.331. Pr (treatment): 0.15 to 0.2, Density: 1.822. Pr (treatment): 0.2 to 0.25, Density: 1.082. Pr (treatment): 0.25 to 0.3, Density: 0.769. Pr (treatment): 0.3 to 0.35, Density: 0.911. Pr (treatment): 0.35 to 0.4, Density: 0.512. Over the histogram, a smooth density curve appears that starts near the point (0.00, 3.87), rises sharply to a peak around (0.031, 6.52), then declines steadily while passing through (0.118, 3.331) and (0.273, 0.769), shows a small secondary rise around (0.324, 0.968), and finally approaches zero density near (0.441, 0). The right panel is titled “Rural non-migrants” and shows a horizontal axis labeled “Pr (treatment)” ranging from 0 to 0.5 in increments of 0.1, and a vertical axis labeled “Density” ranging from 0 to 30 in increments of 10 units. In the right panel, a shaded histogram appears with the data as follows: Pr (treatment): 0 to 0.05, Density: 15.161. Pr (treatment): 0.05 to 0.1, Density: 3.011. Pr (treatment): 0.1 to 0.15, Density: 1.075. Pr (treatment): 0.15 to 0.2, Density: 0.001. A smooth density curve overlays the histogram, starting near (0.00, 16.774), rising sharply to a pronounced peak near (0.006, 27.742), then declining steeply while passing through (0.076, 2.796) and (0.127, 0.86), and flattening out close to zero density by the time it reaches around (0.227, 0). Note: All numerical data values are approximated.Density distributions of property scores for rural-urban migrants and non-rural migrants
The two-panel density plot titled “Caliper equals 0.25 asterisk S D” presents side-by-side distributions of propensity scores. The left panel is titled “Rural-to-urban migrants” and shows a horizontal axis labeled “Pr (treatment)” ranging from 0 to 0.5 in increments of 0.1, and a vertical axis labeled “Density” ranging from 0 to 8 in increments of 2 units. In the left panel, a shaded histogram appears with the data as follows: Pr (treatment): 0 to 0.05, Density: 6.52. Pr (treatment): 0.05 to 0.1, Density: 4.726. Pr (treatment): 0.1 to 0.15, Density: 3.331. Pr (treatment): 0.15 to 0.2, Density: 1.822. Pr (treatment): 0.2 to 0.25, Density: 1.082. Pr (treatment): 0.25 to 0.3, Density: 0.769. Pr (treatment): 0.3 to 0.35, Density: 0.911. Pr (treatment): 0.35 to 0.4, Density: 0.512. Over the histogram, a smooth density curve appears that starts near the point (0.00, 3.87), rises sharply to a peak around (0.031, 6.52), then declines steadily while passing through (0.118, 3.331) and (0.273, 0.769), shows a small secondary rise around (0.324, 0.968), and finally approaches zero density near (0.441, 0). The right panel is titled “Rural non-migrants” and shows a horizontal axis labeled “Pr (treatment)” ranging from 0 to 0.5 in increments of 0.1, and a vertical axis labeled “Density” ranging from 0 to 30 in increments of 10 units. In the right panel, a shaded histogram appears with the data as follows: Pr (treatment): 0 to 0.05, Density: 15.161. Pr (treatment): 0.05 to 0.1, Density: 3.011. Pr (treatment): 0.1 to 0.15, Density: 1.075. Pr (treatment): 0.15 to 0.2, Density: 0.001. A smooth density curve overlays the histogram, starting near (0.00, 16.774), rising sharply to a pronounced peak near (0.006, 27.742), then declining steeply while passing through (0.076, 2.796) and (0.127, 0.86), and flattening out close to zero density by the time it reaches around (0.227, 0). Note: All numerical data values are approximated.Density distributions of property scores for rural-urban migrants and non-rural migrants
Following the estimation of propensity scores, PSM was conducted to compare skill development outcomes between those who migrated and those who did not. Results from balance checking indicate that 707 treated (migrated) individuals and 11,091 untreated (non-migrated) individuals were within the common support region, ensuring that matching was effective.
The matching results showed that the covariates were well-balanced after matching (Table 3). The reduction in the standardised mean bias for all covariates indicates this. The post-matching bias values were close to zero, and the overall pseudo-R squared value was low, suggesting that the model successfully achieved a covariate balance.
Matching quality test: balance property
| Matching algorithm | Pseudo R2 | LR Chi2 (p-value) | Rubin’s B | Rubin’s R | Mean bias after |
|---|---|---|---|---|---|
| Nearest four neighbours matching | 0.1446 | 0.000 | 21.0 | 0.95 | 3.2 |
| Radius matching (caliper = 0.1) | 0.1446 | 0.000 | 21.0 | 0.95 | 3.2 |
| Kernel matching (bandwidth = 0.06) | 0.1446 | 0.000 | 21.0 | 0.95 | 3.2 |
Source(s): Authors’ own work
The standardised bias measured the difference in the mean values of covariates between the treated (migrants) and control (non-migrant) groups, expressed as a percentage of the average standard deviation (Figure 2).
The dot plot titled “Caliper equals 0.25 asterisk S D” shows standardized percent bias across covariates. The horizontal axis is labeled “Standardized percent bias across covariates” and ranges from negative ten to ten in increments of 5 units. The vertical axis is labeled “Covariates” and ranges from top to bottom as “refrem”, “enrgelec”, “5 dot n b t h f”, “4 dot n b t h f”, “3 dot n b t h f”, “2 dot n b t h f”, “5 dot Health underscore Status”, “4 dot Health underscore Status”, “3 dot Health underscore Status”, “2 dot Health underscore Status”, “5 dot marital underscore status”, “4 dot marital underscore status”, “3 dot marital underscore status”, “2 dot marital underscore status”, “5 dot wealth”, “4 dot wealth”, “3 dot wealth”, “2 dot wealth”, “african”, “h h sizer”, “best underscore gen”, “age 2”, and “best underscore age underscore yrs” A legend on the right side shows circle markers labeled “Unmatched” and x-shaped markers labeled “Matched”. A vertical dashed reference line is placed at 0 on the horizontal axis. The unmatched dots are distributed across a wider range, extending from negative 5.396 to 8.119 percent across different covariates, with several points clearly positioned far from the zero line. Examples of unmatched point coordinates include (negative 5.446, best underscore age underscore yrs), (0.693, 3 dot Health underscore Status), and (5.248, african). The matched dots are concentrated closer to the vertical reference line, generally falling within a narrower range from about negative 3.02 to 3.02 percent across most covariates. Examples of matched point coordinates include (negative 0.594, best underscore gen), (0.495, 4 dot wealth), and (0.941, 2 dot Health underscore Status). The matched points cluster tightly around the vertical reference line compared to the unmatched points across the full set of covariates. Note: All numerical data values are approximated.Standardised deviation of each variable
The dot plot titled “Caliper equals 0.25 asterisk S D” shows standardized percent bias across covariates. The horizontal axis is labeled “Standardized percent bias across covariates” and ranges from negative ten to ten in increments of 5 units. The vertical axis is labeled “Covariates” and ranges from top to bottom as “refrem”, “enrgelec”, “5 dot n b t h f”, “4 dot n b t h f”, “3 dot n b t h f”, “2 dot n b t h f”, “5 dot Health underscore Status”, “4 dot Health underscore Status”, “3 dot Health underscore Status”, “2 dot Health underscore Status”, “5 dot marital underscore status”, “4 dot marital underscore status”, “3 dot marital underscore status”, “2 dot marital underscore status”, “5 dot wealth”, “4 dot wealth”, “3 dot wealth”, “2 dot wealth”, “african”, “h h sizer”, “best underscore gen”, “age 2”, and “best underscore age underscore yrs” A legend on the right side shows circle markers labeled “Unmatched” and x-shaped markers labeled “Matched”. A vertical dashed reference line is placed at 0 on the horizontal axis. The unmatched dots are distributed across a wider range, extending from negative 5.396 to 8.119 percent across different covariates, with several points clearly positioned far from the zero line. Examples of unmatched point coordinates include (negative 5.446, best underscore age underscore yrs), (0.693, 3 dot Health underscore Status), and (5.248, african). The matched dots are concentrated closer to the vertical reference line, generally falling within a narrower range from about negative 3.02 to 3.02 percent across most covariates. Examples of matched point coordinates include (negative 0.594, best underscore gen), (0.495, 4 dot wealth), and (0.941, 2 dot Health underscore Status). The matched points cluster tightly around the vertical reference line compared to the unmatched points across the full set of covariates. Note: All numerical data values are approximated.Standardised deviation of each variable
Before matching, there was a substantial bias across several covariates, indicating significant differences between the treated and control groups. However, after matching, the bias was significantly reduced for all covariates, with values approaching zero. This reduction in bias demonstrates the effectiveness of the propensity score matching process in balancing the covariates between the two groups.
The decrease in bias for each covariate confirms that the matching procedure successfully created comparable groups. This allowed for a more accurate estimation of the treatment effect. This balanced covariate distribution between migrants and non-migrants enhances the reliability of the findings regarding the impact of rural-urban migration on skill development.
3.4 Treatment effects
The ATT was estimated to assess the impact of migration on skill development (Table 4). Before matching, there was a significant difference in skill development between the treated and untreated groups, with a difference of 0.3816. After matching, this difference remained significant but was reduced to 0.2241. This indicates that rural-urban migration had a positive, and significant impact on skill development. The results showed a positive, significant impact of rural-urban migration on skill development across all matching algorithms used. The ATT values indicated that the treated group (migrants) consistently had higher skill development outcomes than the control group (non-migrants). Specifically, the nearest four neighbours matching, radius matching with a calliper of 0.1, and kernel matching with a bandwidth of 0.06, showed significant differences, reinforcing the positive effect of migration on skill development.
Average treatment effects of rural-urban migration on skill development
| Matching algorithm | Treatment group | Control group | Difference | t-value | Change (%) |
|---|---|---|---|---|---|
| Nearest four neighbours matching | 0.358 | 0.190 | 0.168 | 3.22 | 88.42 |
| Radius matching (caliper = 0.1) | 0.358 | 0.184 | 0.174 | 2.69 | 94.57 |
| Kernel matching (bandwidth = 0.06) | 0.358 | 0.184 | 0.174 | 2.69 | 94.57 |
Source(s): Authors’ own work
3.4.1 Placebo test
A placebo test addressing the potential influence of omitted and unobservable variables on the estimated treatment effects in the rural-urban migration model was implemented to ensure the robustness of the PSM estimates. Applying the PSM method to a placebo outcome assessed whether any significant effects of rural-urban migration on skill development would appear when no such effect was expected. The placebo test should have yielded no significant differences if the original estimates were unbiased. As shown in Table 5, the results of the placebo test confirmed the absence of bias in the main findings, as no significant effects were found.
3.5 Effect of rural-urban migration on skill development
The regression analysis after PSM further underscored several significant predictors of skill development among individuals who migrated from rural to urban areas compared to those who did not. The model explained approximately 16.6% of the variability in skill development, as indicated by the R-squared value (Table 6).
Results of regression analysis after PSM
| Skill development | ||
|---|---|---|
| Coeff | S.E | |
| Rural-urban migration | 0.205*** | (0.045) |
| Age | −0.008 | (0.006) |
| Age2 | 0.000* | (0.000) |
| Female | 0.007 | (0.021) |
| HH size | −0.029*** | (0.003) |
| African | 0.130*** | (0.030) |
| Marital status | ||
| Living with partner | −0.633*** | (0.036) |
| Widow/Widower | −0.458*** | (0.053) |
| Divorced or separated | 0.077 | (0.068) |
| Never married | −0.326*** | (0.029) |
| Health status | ||
| Fair | 0.395*** | (0.097) |
| Good | 0.458*** | (0.090) |
| Very Good | 0.554*** | (0.089) |
| Excellent | 0.701*** | (0.090) |
| Frequency of crime | ||
| Very rare | 0.203*** | (0.032) |
| Not common | 0.151*** | (0.034) |
| Fairly common | 0.217*** | (0.035) |
| Very common | 0.170*** | (0.032) |
| Socioeconomic status | ||
| Poor | 0.073** | (0.031) |
| Middle | 0.342*** | (0.033) |
| High | 0.657*** | (0.032) |
| Highest | 0.719*** | (0.037) |
| Access to sanitation | 0.339*** | (0.024) |
| Access to electricity | 0.105*** | (0.030) |
| _cons | −0.478*** | (0.157) |
| N | 1,506 | |
| R squared | 0.166 | |
| Adjusted R-squared | 0.1642 | |
Note(s): *** Significant at 1% level, ** Significant at 5% level, *Significant at 10% level; Std errors in parentheses; Ref group: no schooling, excellent health status, crime never happens, male
Source(s): Authors’ own work
Rural-urban migration had a significant, positive impact on skill development, with migrants showing an improvement of approximately 20.5%. This suggests that moving from rural to urban areas facilitates greater skill acquisition. Age had a nonlinear relationship with skill development. While each additional year of age was associated with a small decrease in skill development, the squared term indicated a slight increase in skill development as individuals grew older, reflecting the nuanced age effect.
Household size negatively affected skill development. Each additional household member was associated with a 2.9% decrease in skill development, indicating that individuals from larger households may face more constraints in developing skills. Race showed a positive association with skill development for African individuals, who experienced 13.0% higher skill development than others.
Regarding marital status, living with a partner was associated with a 63.3% decrease in skills development. Being a widow/widower was associated with a 45.8% decrease, and never married was associated with a 32.6% decrease in skill development. Health status was a strong positive predictor of skill development. Individuals with a better health status experienced progressively higher skill development than those with poor health.
Perceptions of crime frequency were positively correlated with skill development. Individuals who perceived any level of crime had higher skill development than those in the reference group (where crime had never occurred). Socioeconomic status was a critical factor. Individuals with a higher socioeconomic status, ranging from poor to rich, experienced higher skill development than those in the lowest socioeconomic status group. Access to basic services, such as sanitation and electricity, further enhances skill development.
4. Discussion
Using longitudinal data from the NIDS, the primary aim of this study was to examine the impact of rural-urban migration on skill development in South Africa, with a specific focus on understanding the underlying factors that drive migration and how these movements influence skill development.
The study found that the relationship between age and rural-urban migration is nonlinear. This aligns with previous studies which suggest that younger people may be less inclined to migrate because of financial obligations, while older individuals might move for better prospects or life-related reasons (Nagendraiah and Ravindra, 2019; Rufai et al., 2019; Rohmah et al., 2023). While previous studies, such as Megersa and Tafesse (2024), have found that larger family sizes positively impact migration decisions by providing family support and access to farmland, this study presents a contrasting perspective. These findings suggest that the costs and challenges of relocating family members outweigh the benefits, resulting in lower migration rates for larger households. This is supported by another study that also observed a negative relationship between household size and migration, although the results were not statistically significant (Rufai et al., 2019).
The significant impact of being African on the likelihood of rural-urban migration underscores potential demographic variations in migration trends. In terms of socioeconomic factors, the findings are consistent with those of previous studies, indicating that individuals with higher socioeconomic status may be less motivated to migrate because of reduced financial pressures (Haer and Yuniarti, 2023; Tsapenko and Tsapenko and 2023). Moreover, marital status also plays a role in migration choices; individuals cohabiting with a partner are less likely to migrate than others (Guntoro et al., 2019). In addition, the positive impact of having access to sanitation on the likelihood of rural-urban migration may stem from access to resources and better overall living conditions, which makes it easier for people to move. This suggests that individuals with access to essential services may feel empowered to relocate to urban areas.
The significant positive influence of moving from rural to urban areas on skill development, as shown by a 20.5% increase among migrants, emphasises the important role of migration in improving human capabilities. This finding is consistent with the Human Capital Theory, which proposes that people migrate in search of opportunities for acquiring skills, education and employment that are more abundant in urban settings (Becker, 1964; Tese and Aga, 2024). The improvement in skills among migrants likely results from the availability of resources, diverse job markets, and training opportunities offered by urban areas compared with rural environments.
Nevertheless, while migration is a pathway for skill development and economic progress, national policymakers should prioritise improving access to quality education and vocational training in rural areas to reduce reliance on migration for skill acquisition. By aligning training programs with the economic needs of rural communities, individuals can acquire valuable skills locally, thereby contributing to the development of human capital without the need for relocation.
Local governments play a critical role in improving rural infrastructure and ensuring access to service, such as electricity, clean water, sanitation, and healthcare. Enhancing these essential services can improve living standards, making rural areas more appealing for residents and businesses (Chaurey and Le, 2022; Sharma and Kumar, 2023). These enhancements help retain skilled individuals and stimulate local economic growth, thus nurturing the development of human resources within rural communities.
Based on Human Capital Theory, this balanced approach indicates that while promoting rural-urban migration is crucial for skill development, there is also great value in creating an environment where individuals can boost their skills and economic prospects within their rural settings. Policymakers can encourage more inclusive and sustainable economic progress by addressing underlying reasons for migration and supporting development initiatives. This ensures that everyone has a chance to succeed, whether they decide to move or stay in their rural hometowns.
5. Conclusion
This study examined the effect of rural-urban migration on skill development in South Africa, revealing that migration significantly influences skills acquisition. These findings indicate the positive role of urban environments in providing opportunities for skill development driven by access to diverse job markets, better infrastructure, and training opportunities. These results underscore the importance of policies that support skill development, while addressing the underlying drivers of migration. Such policies could focus on improving access to education and vocational training in rural areas to ensure that migration is a choice, and not a necessity, for accessing better opportunities.
Like any other research endeavour, this study has its limitations. Endogeneity remains a concern, as skill development could influence an individual’s migration decision, a factor was not controlled for due to the lack of an instrumental variable. Additionally, while the skill development measure was carefully constructed using existing data, it did not fully capture aspects of skill development, such as training and career advancement initiatives. Addressing these gaps through future studies, which incorporate expansive datasets and alternative methodologies could provide a more nuanced understanding of the relationship between rural and urban migration and skill development.
By addressing these challenges, policymakers can formulate strategies to promote inclusive growth, reduce inequalities between rural and urban areas, and ensure equitable access to opportunities for all individuals in South Africa.
Funding: This research received no external funding
Data availability statement: All datasets are publicly available in DataFirst portal at https://www.datafirst.uct.ac.za/dataportal/index.php/catalog/NIDS/?page=1&sort_by=title&sort_order=asc&ps=15&repo=NIDS (accessed on 20 February 2024).
Conflicts of interest: The authors declare no conflict of interest.
