Migration in Kenya: beyond Harris-Todaro

ABSTRACT

This paper examines the impact of agrarian structures on the migration behavior and destination of rural household heads and individuals in Kenya. To explore the complexity of migration we extend the standard Harris-Todaro framework to account for land inequality and size as well as type of destination. Using probit regressions, we show that Kenyan household heads born in districts with higher land inequality, smaller per capita land and lower per capita rural income are more likely to migrate. We show that for individuals whose incomes are squeezed by larger land inequality, migration from villages to smaller cities, and villages in different districts could be a preferable strategy to migrating to Greater Nairobi. The impact of land inequality is larger for male than female migration and insignificant for females’ rural-to-rural migration. Moreover, the level of education, age, marital status, gender, religion and distance to Nairobi play a role in migration behavior.

KEYWORDS:

• Migration
• distribution
• agrarian structures
• Kenya
• education

JEL CLASSIFICATION:

• O15
• Q15
• O12
• O55

Acknowledgements

The authors would like to thank Richard Agesa, Michael Ash, Leonce Ndikumana, Lynda Pickbourn, and Mehmet Uğur for their helpful comments that greatly improved the paper. All remaining errors are of course ours.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Over recent years, people working in Nairobi have often had to live outside its administrative boundaries. Therefore, we defined Nairobi as Greater Nairobi, which includes Nairobi district and Thika and Kiambu, which are less than 1-h distance to Nairobi. This is because, Thika and Kiambu increasingly become bedroom communities for Nairobi. While there are other bedroom communities in Kajiado and Machakos districts they were not individually identified in the data, and are only a small proportion of the urban population in those districts, which have other urban centers that are not bedroom communities to Nairobi.

2. According to a number of studies, the migration behavior is more complex than the Harris and Todaro (1970) model. In the Harris and Todaro model, the expected urban and rural incomes converge to an equilibrium point as an outcome of rural-to-urban migration. On the other hand, Faini (1996) shows that rural-to-urban migration increases the gap between urban and rural incomes, when the factors of production have increasing returns to scale. In addition, Poot (2008) exhibits that migration leads to aging population in smaller areas, which could increase the regional gaps of productivity. In contrast, Stark, Helmenstein, and Prskawetz (1997) show that migration could raise the average income in the relatively backward regions through the human capital gains of return migration. Last, the New Economics of Labor Migration literature (for example, Stark and Bloom 1985; Stark and Taylor 1991) explains the migration behavior with relative deprivation rather than absolute incomes, which will also be discussed in this paper. However, the empirical evidence on Kenya (for example, Agesa 2001; Agesa and Agesa, 1999; Bigsten 1996; Gray 2011) and on other developing economies (for example, Schultz 1982; Tunali 1996; Zhu 2002) strongly supports Harris and Todaro’s main claim that expected urban and rural incomes’ affect rural-to-urban migration. For this reason and also for the simplicity reasons, our model is mainly based on Harris and Todaro (1970). However, we will also consider the direct impact of relative deprivation in our discussion.

3. In our study as in most studies using cross-sectional data, we are only able to analyze the most recent migration. We can make no claims as to whether this is a temporary, circular or permanent migration. The respondents in the survey used, however, have been surveyed at what they presently consider their permanent address.

4. Our share of agglomeration is from World Bank (2016)’s World Development Indicators (WDI) database. WDI does not explicitly report the cities that were included in their agglomeration classification. Nevertheless, both WDI’s share of agglomeration for 2009 and Nairobi and Mombasa total population share in Kenya estimated from Kenya National Bureau of Statistics (KNBS) (2012)’s 2009 Kenya Population and Housing Census are consistently 10.5%. Moreover, WDI’s share of agglomeration data is smooth and continuous, which shows that the estimates in WDI consider the same cities. Following these two outcomes, we can conclude that share of agglomeration values in WDI data considers Nairobi and Mombasa.

5. Urban migration has been found in the past to be dominated by secondary school-educated individuals (wa Gĩthĩnji 2000). In 1960 there were only 91 secondary schools, which increased to 142 on the eve of independence in 1962, by 1968, 5 years after independence there were 601 a more than 6-fold increase from 1960 (Government of Kenya (GOK) 1969).

6. Amsden (2001) and Chang (2008) provide good summaries on how import substitution and export-oriented industrial policies stimulated the growth of industry in the developing world.

7. Z-goods is a term defined by Hymer and Resnick (1969) for non-agricultural traditional activities in an agrarian economy. These activities consist of food processing, handicraft activities and services for local needs.

8. Similarly, the urban sector is also formed by underemployed agents and agents that receive identical wages.

9. Based on 1989 and 1999 Kenya Census data, Archambault, de Laat, and Zulu (2012) show that only 15.7% and 13.7% of married household heads have an access to electricity and sewage systems, respectively.

10. Emina et al. (2011)’s work specifically focuses on Korogocho and Viwandani slums of Nairobi.

11. wa Gĩthĩnji, Konstantinidis, and Barenberg (2014) test the empirical relationship by using the logarithm of land productivity as a dependent and the logarithm of farm size as an independent variable. The empirical analysis based on household data finds coefficients of −0.32 and −0.40 between two variables. By taking the second derivative of land productivity with respect to farm size, we can show that the relationship between the land productivity and farm size is convex.

12. Using probit model, Agesa and Agesa (2005) test expected earnings’ impact on probability of rural-to-urban migration. They find the coefficient for the difference between logarithms of expected wage for migrants in urban areas and non-migrants in rural areas of male as 2.8742. By taking the second derivative of migration probability with respect to ratio between expected urban to rural earnings, we can show that the relationship between expected urban to rural earnings to probability of migration is concave.

13. Land inequality is associated with differing socioeconomic and geographical characteristics (see Appendix 2, Tables 2.1–2.5). Areas with relatively low land inequality (Gini of lower than 0.50) are found mostly in the west of the country and the coastal area. These areas have both a more equal income and educational distribution. Ecologically they are dominated by a higher proportion of flat land, with a significant proportion of this land having rich or loamy soils. At the other extreme high land inequality areas are dominated by very small and also large farms with a relatively thin middle. These districts are found mainly in the former Rift Valley Province in areas dominated by settler farming during the colonial era and pastoral production prior to that. Ecologically these areas are flatter and more amenable to mechanized farming and have less loamy soils. Educational inequality is very high in these areas. Whereas all other areas have more than 30% completing high school or university, the proportion in these areas is only 20% with the proportion completing university being half of that of the areas with lower land inequality. The middle part of the distribution in terms of land inequality is found towards the center of the country with relatively more equal population shares across income and education distributions. With the educational distribution being quite close to that of areas with low land inequality. The soil distribution has a significant proportion in loamy soils but also have a higher proportion in hilly areas compared both to the low and high land inequality areas.

14. According to KIHBS 2005/2006, among the 70 districts in Kenya, Nairobi and Mombasa are entirely urban.

15. The per capita land measure does not measure the population density. It rather reflects the total cultivated land over population in each district. We predict this using 1988 Rural Labor Force Survey (RLFS). First, we attribute a piece of land to each individual in the sample through dividing the total land of household to the household size. Then, we take the weighted average of each individual’s land using the population weights of each individual in the sample.

16. A number studies including Deb and Seck (2009) for Mexico and Indonesia, Gupta and Mitra (2002) for India and Tunali (1996) for Turkey show that distance to larger cities affects the migration behavior.

17. Authors’ calculations from the Kenya Integrated Household Budget Survey (KIHBS) – 2005/2006.

18. The urban areas that are less than 1-h distance to Nairobi are regarded as Greater Nairobi.

19. We also estimated the marginal effects in Table 9 by redefining Nairobi by considering Thika and Kiambu as a part of ‘other urban’ rather than Greater Nairobi. The results are similar to Table 9. For household heads, the coefficients for land Gini, logarithm of per capita land and logarithm of per capita income on migration to Nairobi are also insignificant at 10%. For all adults, the coefficient for land Gini explaining migration to Nairobi is insignificant at 10% and the coefficients for logarithm of per capita land and logarithm of per capita income are negative. Moreover, the land Gini coefficients explaining migration to ‘other urban’ are significant at 1% and are slightly higher compared to Table 9. The coefficients for the land Gini are 0.2334 and 0.2126 for household heads and all adults, respectively. The estimates are available upon request.

20. We also estimated a separate probit model with four categories: no migration = 0, migration to rural in another district = 1, migration to other urban = 2, migration to Mombasa, Nakuru, Kisumu, Nakuru and Eldoret (Uasin Gishu) = 3, migration to Greater Nairobi = 4. We included Mombasa, Nakuru, Kisumu, Nakuru and Eldoret (Uasin Gishu) in a separate category, since they are the largest four urban areas after Nairobi. The estimates reflect that the marginal effects for land, income, and education variables are very similar for migration to other urban and migration to Mombasa, Nakuru, Kisumu, Nakuru and Eldoret (Uasin Gishu). For all adults, the difference between the marginal effects of both categories is 0.0128 for land Gini, 0.0014 for logarithm of per capita land, 0.0006 for logarithm of per capita rural income, and 0.0005 for years of education. The estimates are available upon request.

21. We also estimated the probit regression in Table 9 by taking migration from rural to Greater Nairobi as the base of the regression. For both all adults and household heads, the coefficients for years of education were significantly negative at 1% level. This shows that the education’s coefficient in Table 8 is significantly larger for migration to Greater Nairobi. The estimates are available upon request.

22. Authors’ calculations from the Kenya Integrated Household Budget Survey (KIHBS) – 2005/2006.

23. Authors’ calculations from the Kenya Integrated Household Budget Survey (KIHBS) – 2005/2006.