Perceptions versus Behavior: New Insights into South African Internal - - PDF document

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 1

Perceptions versus Behavior: New Insights into South African Internal Migration Tyler W. Myroniuk, George Mason University Department of Sociology and Anthropology Robinson Hall B, Room 305, 4400 University Drive, 3G5, Fairfax, VA 22030 Acknowledgments: I would like to thank Michael White, Chantel Pheiffer, Becca Wang, Jason Davis, Maria Abascal, Casey Miller, and Lindsey Reynolds for their critiques of this manuscript at different stages of development. The research was funded indirectly by grant R24 HD041020, Brown University Population Studies & Training Center. Further, the National Income Dynamics Study has been implemented by the Southern Africa Labour and Development Research Unit (SALDRU) based at the University of Cape Town’s School of Economics. I would also like to extend my thanks to the research team currently led by Murray Leibbrandt, Ingrid Woolard, Cecil Mlatsheni, Nicola Branson, and Samantha Richmond.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 2

Abstract Throughout sub-Saharan Africa, individuals tend to engage in frequent internal migration episodes, often in efforts to provide for their households and survive. The perception that one will move and the disjuncture between acting upon such perceptions is at the foundation of migration

• research. In the case of South Africa, men and women have varying abilities to move, face different

consequences, and might judge the benefits of moving not worthwhile after evaluating their initial migratory intentions. This paper goes beyond conventional migration studies by seeking to understand whether the factors associated with the perceived likelihood of migration are similar to those associated with migration itself (including the extent to which prior conceptions of migrating influence behavior) for Black South African residents. The 2008, 2010, and 2012 waves of the nationally representative South African National Income Dynamics Study are exploited, and fixed- effects regressions are employed to examine the perceived chances of migrating and migration

• behavior. There are marked gender differences, and migration perceptions are only linked to

migration behavior for women. Keywords: gender, internal migration, longitudinal data, perceptions, South Africa

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 3

Introduction Population growth, urbanization, and increasing disparities between cities and rural areas in the past half-century have made internal migration a central “economic survival strategy” for many households in sub-Saharan African (SSA) countries.1 This type of migration is often, although not exhaustively, a household decision that attempts to optimize financial and/or non- financial returns on the investment to send someone to a city or other rural area to look for work.2 Such migration strategies will almost certainly continue over the course of the 21st century in SSA, where much of the world’s population growth3 and continued urbanization will come from In South Africa, where this study takes place, individuals tend to frequently, and internally, migrate between urban and rural areas, often between provinces too, in such economic survival efforts.4 In the years immediately after the end of apartheid in 1994—and, thus, the end of internal travel restrictions for non-Whites—Black South Africans, in particular, rapidly took advantage of this new era by migrating to urban areas in greater numbers than ever before to seek out

1 Mberu, B. U. (2016). African Migration and Population Distribution: Recent Trends,

Methodological Challenges and Policy Issues. In, MJ White (ed.), International Handbook of Migration and Population Distribution (pp. 245-267). Springer Netherlands, p. 248.

2 Stark, O., & Taylor, J. E. (1991). Migration incentives, migration types: The role of relative

• deprivation. The economic journal, 101 (48), 1163-1178.

3 Gerland, P., Raftery, A. E., Ševčíková, H., Li, N., Gu, D., Spoorenberg, T., et al. (2014). World

population stabilization unlikely this century. Science, 346(6206), 234-237.

4 Collinson, M., Tollman, S. M., Kahn, K., Clark, S. J., & Garenne, M. (2006). Highly prevalent

circular migration: households, mobility and economic status in rural South Africa. In M. Tienda et al. (Eds.), Africa on the move: African migration and urbanisation in comparative perspective (pp. 194-216). Johannesburg: Wits University Press. Collinson, M. A. (2010). Striving against adversity: the dynamics of migration, health and poverty in rural South Africa. Global Health Action, 3, 5080.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 4

employment and a better life for themselves and their families.5 However, in the years since, labor migration throughout the country has declined, somewhat inexplicably.6 Scholarly work uncovering the determinants of internal migration in South Africa, and elsewhere in SSA, is often marred by a lack of detailed, longitudinal data—thereby increasing the extent to which unobserved heterogeneity influences results—and/or the dearth of nationally representative data. This paper goes beyond conventional migration studies by seeking to understand whether the factors associated with the perceived likelihood of migration are similar to those associated with migration itself (including the extent to which prior conceptions of migrating influence behavior) for Black South Africans. Further, the analyses assess if there are important differences in the likelihood of migrating between individuals who believe that they will move in the near future and those who do not. The perception that one will move and the disjuncture between acting upon such perceptions is at the foundation of migration research. In the case of South Africa, men and women have varying abilities to move, face different consequences, and might judge the benefits of moving not worthwhile after evaluating their initial migratory intentions. The distinctions of migration perceptions versus behavior are virtually always overlooked or unmeasured aspects of migration, in large part due to the difficulties in conducting panel studies that follow-up with migrants in SSA nations. I seek to overcome prior limitations that migration research has faced by exploiting the nationally representative National Income Dynamics Study (NIDS) in South Africa, and employing quasi-experimental techniques to identify the temporal ordering of perceptions of

5Reed, H.E. (2013). Moving across boundaries: migration in South Africa, 1950–

• 2000. Demography, 50(1), 71-95.

6 Posel, D. (2010). Households and labour migration in post-apartheid South Africa. Journal of

Studies in Economics and Econometrics, 34(3), 129-41.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 5

migration and behavior in addition to minimizing bias stemming from underlying, fixed, unobserved factors that often hamper such analyses.7 In doing so, I offer a more-nuanced understanding of the critical, life-altering, decision to migrate, that individuals of all ages in SSA face. Why Move? Within South Africa, as in other SSA nations, individuals generally migrate or plan to migrate with the belief that this will improve their and their families’ economic well-being, even if this is not always the case.8 A robust literature has shown that this type of migration among adults comes in many forms, although all are typically due to employment/wages, marriage, and/or health.9 Most commonly, this research has shown that, 1) the opportunity to take a job will spur

7 Southern Africa Labour and Development Research Unit. (2016). National Income Dynamics

Study 2008, Wave 1 [dataset]. Version 6.1. Cape Town: Southern Africa Labour and Development Research Unit [producer], 2016. Cape Town: DataFirst [distributor]. Southern Africa Labour and Development Research Unit. (2016). National Income Dynamics Study 2010-2011, Wave 2 [dataset]. Version 3.1. Cape Town: Southern Africa Labour and Development Research Unit [producer], 2016. Cape Town: DataFirst [distributor]. Southern Africa Labour and Development Research Unit. (2016) National Income Dynamics Study 2012, Wave 3 [dataset]. Version 2.1. Cape Town: Southern Africa Labour and Development Research Unit [producer], 2016. Cape Town: DataFirst [distributor].

8 Oosthuizen, M., and P. Naidoo 2004 “Internal Migration to the Gauteng Province.” Working

Paper, Development Policy Research Unit, University of Cape Town. Posel, D., and D. Casale 2006 “Internal Labor Migration and Household Poverty in Post- Apartheid South Africa.” In Poverty and Policy in Post-Apartheid South Africa. Eds. H. Bhorat, and R. Kanbur. Cape Town: Human Sciences Research Council Press. Pp. 351–365.

9 Although there are too many examples to cite here, for relevant and recent micro- and macro-

level research, see: Anglewicz, P. (2012). Migration, marital change, and HIV infection in Malawi. Demography, 49(1), 239-265. Camlin, C. S., Snow, R. C., & Hosegood, V. (2014). Gendered patterns of migration in rural South Africa. Population, space and place, 20(6), 528-551. Choe, C., & Chrite, E. L. (2014). Internal migration of blacks in south africa: An application of the roy model. South African Journal of Economics, 82(1), 81-98.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 6

individuals to move, even if only to return home several months later (if the result is not permanent migration); 2) depending on the kinship system, both women and men will move to their new spouse’s home, and in the case of divorce or widowhood they might return to origin household prior to the marriage; and 3) healthy individuals tend to migrate, HIV infection often leads to expulsion from a household (often accompanied by a union dissolution), and, in some cases, unhealthy individuals return home to die.10 Lifestyle migration—such as for the weather and recreation opportunities, or to be with friends—has not been empirically identified in SSA. Gender plays a large, stratifying role in migration. In southern Africa, women have a higher prevalence of migrations to/within rural areas than men, while men have a higher prevalence of Collinson, M., Tollman, S. M., Kahn, K., Clark, S. J., & Garenne, M. (2006). Highly prevalent circular migration: households, mobility and economic status in rural South Africa. In M. Tienda et al. (Eds.), Africa on the move: African migration and urbanisation in comparative perspective (pp. 194-216). Johannesburg: Wits University Press. Collinson, M. A., Tollman, S. M., & Kahn, K. (2007). Migration, settlement change and health in post-apartheid South Africa: Triangulating health and demographic surveillance with national census data1. Scandinavian Journal of Public Health, 35(69_suppl), 77-84. Collinson, M. A. (2010). Striving against adversity: the dynamics of migration, health and poverty in rural South Africa. Global Health Action, 3, 5080. Crush, J., Dodson, B., Williams, V., & Tevera, D. (2017). Harnessing Migration for Inclusive Growth and Development in Southern Africa. Southern African Migration Programme; Waterloo, Ontario, Canada. Englund, H. (2002). The village in the city, the city in the village: migrants in Lilongwe. Journal

• f Southern African Studies, 28(1), 137-154.

Ginsburg, C., Bocquier, P., Béguy, D., Afolabi, S., Augusto, O., Derra, K., Odhiambo, F., Otiende, M., Soura, A., Zabré, P. and White, M.J., 2016. Human capital on the move: Education as a determinant of internal migration in selected INDEPTH surveillance populations in

• Africa. Demographic Research, 34, pp.845-884.

Lurie, M. N., & Williams, B. G. (2014). Migration and health in Southern Africa: 100 years and still circulating. Health Psychology and Behavioral Medicine: an Open Access Journal, 2(1), 34- 40. Vearey, J. (2012). Learning from HIV: exploring migration and health in South Africa. Global public health, 7(1), 58-70.

10 Clark, S. J., Collinson, M. A., Kahn, K., Drullinger, K., & Tollman, S. M. (2007). Returning

home to die: Circular labour migration and mortality in South Africa 1. Scandinavian Journal of Public Health, 35(69_suppl), 35-44.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 7

migrations to/within urban areas than women.11 These patterns imperfectly reflect employment- based and marriage-related migration differences between men. Evidence from rural, northeastern South Africa indicates that between 1994 and 2003, men of working age (mid-20s to mid-50s) had three times higher rates of temporary migration than women.12 These patterns are emblematic of decades of efforts prior to the end of apartheid, in 1994, of White South Africans encouraging Black, male, labor migration away from their rural homes—to mining and other natural resource extraction settlements, even if present-day migration patterns are much more urbanized.13 Prior to the end of apartheid, out-migration (excluding marriage migration) from rural areas—although less prevalent for women than men—was dependent on established social networks in urban destinations.14 More recent research has shown that female migration patterns have become increasingly urbanized in the post-apartheid era, and for labor, in addition to historically typical temporary rural-rural migrations.15 While there is some evidence to support the notion that men have a much easier time than women in simply “getting up and going” to a new location, these gender differences are no longer stark.

11 Camlin, C. S., Snow, R. C., & Hosegood, V. (2014). Gendered patterns of migration in rural

South Africa. Population, space and place, 20(6), 528-551.

12 Collinson, M. A., Tollman, S. M., & Kahn, K. (2007). Migration, settlement change and health

in post-apartheid South Africa: Triangulating health and demographic surveillance with national census data1. Scandinavian Journal of Public Health, 35(69_suppl), 77-84.

13 Feinstein CH. 2005. An Economic History of South Africa: Conquest, Discrimination and

• Development. Cambridge University Press: London, United Kingdom.

14 Bozzoli, B. (1991) Women of Phokeng: Consciousness, Life Strategy, and Migrancy in South

Africa, 1900–1983. London: James Currey.

15 Camlin, C. S., Snow, R. C., & Hosegood, V. (2014). Gendered patterns of migration in rural

South Africa. Population, space and place, 20(6), 528-551.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 8

Migration Perceptions versus Behavior As is the case virtually anywhere, geographic space is critical in migration decision-making in post-apartheid South Africa. The trend of urbanization has long been in motion, but migration to the most heavily urbanized centers—Gauteng Province (where the Johannesburg-Pretoria corridor is located), Cape Town, Durban, and Port Elizabeth—is largely due to migration from adjacent provinces.16 The perception of being able to easily move to an urban center, would seemingly be more prevalent among those who live nearby. More broadly, as has been long- documented in neoclassical and new economics of labor migration theoretical perspectives, any perception of migration is conditional on the belief that migrating would make an individual’s life,

• r their households’ livelihood, better.

Very little demographic research has assessed the relationship between these perceptions and migration behavior; most recently, the case of international migration from Mexico to the US, and within Mexico, shows that the aspiration of moving predicts migration behavior.17 Dated, and

• nly cross-sectional, research from Kenya provides a challenge to the idea that planning to migrate

and following-through are closely linked.18 In South Africa, important differences in individuals who intend to migrate within a year and five years have been identified, with several common threads: greater levels of spousal pressure to migrate, being divorced, and prior migration experience are linked to higher chances of migrating internally, whereas those with more life

16 Kok, P. C., O’Donovan, M., Bouare, O., & van Zyl, J. (2003). Post-apartheid patterns of

internal migration in South Africa. HSRC Press: Cape Town.

17 Creighton, M. J. (2013). The role of aspirations in domestic and international migration. The

Social Science Journal, 50(1), 79-88.

18 Sly, D. F., & Wrigley, J. M. (1985). Migration decision making and migration behavior in

rural Kenya. Population & Environment, 8(1), 78-97.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 9

satisfaction have lower changes of migrating internally.19 These migratory intentions are complicated by the commonness of dual residences for many Black South Africans who consider themselves residents of their origin and migratory household, because of aforementioned established circular migration patterns.20 This paper examines the relationship between perceptions of migrating in the next two years, and migration behavior two years later, among Black South African men and women between 2008-2010, 2010-2012, and 2008-2012, separately. These analyses differ from all prior research on this linkage by not only employing nationally representative data, but utilizing prospective regressions to identify the temporal order of migration perceptions and behaviors, and fixed-effects regression techniques with identify changes in migration perceptions and behaviors

• ver time whilst accounting for all time-invariant (observable, unobserved, and unobservable)

individual characteristics that induce considerable bias into regression estimates. The result is an empirically rigorous investigation into an oft-undervalued and unstudied determinant of migration. Data and Methods NIDS contains information on demographic characteristics, economic circumstance and behaviors, health, and other aspects of individuals and households from over 28000 respondents from all racial groups and ages. The public version of the NIDS data—the 2008, 2010, and 2012 waves—and its detailed module on migration, are exploited here. Since this paper focuses on Black Africans (aged 15 years and above), particularly those who completed the survey in all three waves

19 De Jong, G. F., & Steinmetz, M. (2006). Migration intentions in South Africa and

• elsewhere. Migration in South and Southern Africa: Dynamics and Determinants, eds (Pieter

Kok, Derik Gelderblom, John O. Oucho, and Johan van Zyl. HSRC Press: Cape Town. 249-265.

20 Posel, D., & Marx, C. (2013). Circular migration: a view from destination households in two

urban informal settlements in South Africa. The Journal of Development Studies, 49(6), 819-831.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 10

(2008, 2010, and 2012), the analytic sample consists of 2741 men (36%) and 4985 women (63%), which results in 23178 observations over the three waves. Although a fourth wave, 2014-2015 could be included, in principle it would not benefit these analyses due to additional respondent attrition and already sufficient variation in the main predictors and migration outcomes over time. Although there is attrition between waves due to the panel design, weights are employed to

• vercome these concerns and maintain the representativeness of the analytic sample. Additionally,

while secure data including details of migration at fine levels of geography are accessible at the University of Cape Town, the public data are appropriate here. The outcomes of interest— respondents’ perceived chances of migrating in the next two years (0=no/unlikely; 1=possibly/definitely), as well as migration behavior (0=did not move in the past 2 years; 1=moved in the past two years)—are effectively captured by the public data. These do not require modification via the secure data for the purposes of this paper. The analyses begin with cross-sectional logistic regressions in each wave to test for basic associations between migration outcomes and predictors. As a next step, prospective regressions ask the question: do respondents’ characteristics from an earlier period (e.g., wave 1) predict the

• utcome of a future (e.g., wave 2 or 3)—including prior perception of migration and migration

behavior? Still, since only a small handful of observed characteristics are estimated, it is likely that a substantial amount of bias is induced in such models since a number of important factors related to predictors are not measured nor observable. Two additional analyses seek to understand and/or eliminate such potential biases. First, another set of prospective regressions assesses the differences in observable characteristics between those who perceived that they would migrate in two years compared to those who did not perceive that they would move, and the probability of migrating for each group. Second, fixed-

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 11

effects regressions are estimated to account for any time-invariant characteristic such as personality traits or locational characteristics that, might otherwise bias estimates. These models examine changes in predictors and subsequent changes in outcomes by gender. Since these data are nationally representative, but include attrition due to the analytic methods, appropriate panel weights are applied to each set of regressions. Equation 1 shows the cross-sectional logistic regressions and basic associations between

• bserved predictors and perceived chances of migrating in the next two years/migration behavior,

Mit: 1) ln⁡ [𝑄(𝑁𝑗𝑢⁡|⁡𝑦𝑗𝑢)] = 𝛾0 + 𝑦𝑗𝑢𝛾1 + 𝑨𝑗𝛾2 Mit, the outcome, is the log odds of the perceived chance of migration in the next two years/migration behavior for an individual, i, at time, t. The vector xit represents a set of time- varying characteristics—age, household assets (0-26 scale), employment status (employed, unemployed, not active in economy), and marital status (married, widowed/divorced/separated, never married). The vector zi contains fixed characteristics of respondents’ locations—nine provinces with each divided into urban or rural, totaling 18 additional parameters in order to minimize bias stemming from different migratory opportunities throughout South Africa. The β0 term is the intercept. The second estimating procedure, Equation 2, builds on Equation 1 through prospective logistic regressions in efforts to establish some temporal order in our efforts to understand the relationship between observed predictors and perceived chances of migrating in the next two years/migration behavior, Mit. Here, observed characteristics from the previous wave are used to predict outcomes in the present: 2) ln⁡ [𝑄(𝑁𝑗𝑢⁡|⁡𝑦𝑗𝑢)]= 𝛾0 + 𝑦𝑗𝑢−𝑙𝛾1 + 𝑨𝑗𝛾2

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 12

Mit, the outcome, is the log odds of migration behavior for an individual, i, at time, t; xit-k represents a set of time-varying characteristics, which are the same as above and also include the perceived chance of migrating from prior waves, from k previous waves (k-1 or k-2); zi again represents locational fixed characteristics; and β0 is the intercept. The basic structure of Equation 2 is used again to estimate the log odds of migration behavior within the subsamples of: a) men who did not believe they would move within two years; b) men who believed they would move within two years; c) women who did not believe they would move within two years; and d) women who believed they would move within two years. Equation 3 depicts the estimation procedure for fixed-effects models on perceived chances

• f migrating (all waves)/migration (waves 2 and 3) behavior between , Mit:

3) ln⁡ [𝑄(𝑁𝑗𝑢 = 1⁡|⁡𝑦𝑗𝑢, 𝑥𝑗𝑢−1, 𝛽𝑗 ) - ln⁡ [1 − 𝑄(𝑁𝑗𝑢 = 1⁡|⁡𝑦𝑗𝑢, 𝑥𝑗𝑢−1, 𝛽𝑗 )] = ⁡𝜀𝑦𝑗𝑢 +⁡𝛽𝑗 Mit, the outcome, is the log odds of the perceived chance of migration in the next two years/migration behavior for an individual, i, at time, t. The vector xit represents a set of time- varying characteristics (excluding age, which is functionally a fixed characteristic here)— household assets (0-26 scale), employment status (employed, unemployed, not active in economy), and marital status (married, widowed/divorced/separated, never married). Prior perceptions of migrating are captured in the term wit-1 for models where migration behavior is the outcome; αi consists of all observable and unobservable/unobserved individual and contextual fixed effects. The parameter, δ, is estimated by the conditional maximum likelihood.21

21 Chamberlain, G. (1984). Panel data. In Z. Griliches and M.D. Intriligator (Eds.), Handbook of

econometrics, Volume II (pp.1247-1318). New York: Elsevier Science Publishers BV.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 13

Results-Descriptive Statistics Table 1 depicts the characteristics of Black men and women who participated in the 2008, 2010, and 2012 waves of NIDS. Most notably, in each wave, more men reported perceiving that they would move within the next two years compared to women. These perceptions are reflected in migration behavior in 2010 and 2012, as well.

• Insert Table 1 about here-

Table 2 provides bi-variate descriptive statistics of the disjuncture between perceptions of moving and behavior. Chi-square tests indicate that migration behavior for men and women in 2010 and 2012 are both dependent on perceived migration in 2008 (although the relationship for men between 2008 perception and 2012 behavior is marginal [p<.10]). Similarly, migration behavior for men and women in 2012 is dependent on 2010 perceptions. Thus, there is further empirical need to assess how migration perceptions and behavior are related.

• Insert Table 2 about here-

Of course, there is a longstanding demographic literature, from nearly all global contexts, indicating that those who migrate are different from those who do not migrate, on a number of key

• characteristics. Among Black South Africans, in this sample, this persists. Apart from household

assets, those who migrate are significantly different from non-movers on virtually all observed measures across these 2008, 2010, and 2012 NIDS waves (results not presented, but available upon request).

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 14

Results-Multivariate Regressions Table 3 shows relationships between age and education that conform to most migration

• hypotheses. Nevertheless, and most interestingly, prior perceptions of migration—being more

likely to think one will migrate—are only positively, and significantly, associated with migration behavior for women. However, the estimation procedures in Table 3 do not account for a considerable amount of potential bias in these relationships.

• Insert Table 3 about here-

In response to the inherent bias stemming from between-individual differences in the regressions found in Table 3, sub-sample analyses in Table 4 parse out distinctions between those who did not perceived that they would migrate, but did, in the coming years, and those who perceived that they would migrate, and followed through, in the coming years. Although the coefficients for each predictor are differentially associated with migration for each sub-sample by year, and by gender, a common thread emerges—men appear to be able to get up and migrate much more easily. For men, being older is seemingly, and consistently, less of a deterrent to the log odds of migration for those who thought they would migrate, compared to women (see models 2, 4, and 6 compared to models 7-12). Also, for men, having completed a high school education

• r more, are associated with higher log odds of migration for those who did not believe that they

would migrate and did so, compared to women (see models 1 and 3 versus 7, 9, and 11). It look as if men and women have differing migration decision-making processes, and based on their perceptions of migration at an earlier period.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 15

• Insert Table 4 about here-

Given the observable differences in those who moved but did not believe that they would, and those who believed that they would move and did (Table 4), Table 5 improves the extent to which cause and effect can be estimated (although certainly not completely) by measuring changes within individuals through fixed effects regressions. Thus, the time-varying observable and all time invariant, between individual differences found in Table 4 are eliminated from the estimating

• procedure. The results from Table 5 indicate that an increase in household assets is associated with

a lower perceived likelihood of migrating and actually migrating for men and women. Surprisingly for men, and relative to transitioning into employment, transitioning into unemployment is linked to lower perceived chances of migrating, but not for migration behavior. Finally, transitioning into not being active in the economy—likely due to reaching retirement/grant-eligible age—is associated with women being more likely to believe that they will move in the next two years, whereas both not being active in the economy and becoming unemployed are linked to lower chances of migrating. Consistent with Table 3, only women’s changing beliefs about the likelihood

• f migrating in the next two years are positively linked to migration behavior.
• Insert Table 5 about here-

Discussion and Conclusion These nationally representative NIDS data offer a rare opportunity to assess the extent to which the predictors of migration, and perceived chances of migration, differ, as well as the

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 16

relationship between perceiving moving in two years and doing so. The distinct gender results re- affirm that the process in which men and women entertain the idea of migration, and follow- through, are markedly different in important ways. Most notably, perceiving that one will move in the next two years is strongly associated with the chances of moving two years later for women— underscoring the idea that it might take more planning for women to migrate, due to disadvantages, marital norms, and varying labor market opportunities compared to men. Additionally, differences in age and education are associated with men’s seemingly more-fluid migratory decisions, than

• women. Although increases in one’s household assets are linked to lower chances of thinking about

moving and moving, for both men and women, the types of changes that men experience with regards to employment and their migration decisions are different from those of women. Despite not being able to identify causal mechanisms in Black South African residents’ migration decisions, the fixed effects analyses account for changes within individuals—and thus, fixed (and

• ften confounding) characteristics are eliminated—which offers tremendous insight into push-pull

factors of migration in South Africa, as well as supporting the results found in the theoretically, more-biased, first set of regressions in these analyses. These analyses only pertain to South Africa, but the decision to migrate is an important

• ne in all southern African nations. Whether a young, northern Malawian man engaged in

subsistence agriculture is contemplating taking a chance on finding a wage-based job harvesting tea in the southern district of Thyolo, or a Zimbabwean grandmother seeks to reunite with her family in Limpopo Province, or a rural Tanzanian woman needs to acquire additional technical training in Dar es Salaam to improve her family’s livelihood, the process of thinking about migrating and doing so is consequential. Identifying the disjuncture between these two gives further insight into individuals’ and households’ migration calculi—certainly not a new strategy

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 17

to improve one’s personal and/or financial circumstances. While experimental designs are unethical, and natural experiments—typically in the form of disasters or conflict leading to forced migration—are not ideal to understand this calculus, expanding extant migration modules in

• bservational research is necessary throughout SSA contexts. Future research ought to consider

the qualitative characteristics of migration—types of locations residing in and moved to—in addition to distance, in efforts to expand our assessment of migration in SSA.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 18

Table 1: Unweighted Univariate Descriptive Statistics, Black South Africans 2008-2010-2012 (Percentages, Means, and Standard Deviations Where Appropriate) Men Women Perceived Move in 2 Years 2008 18.9 16.0 2010 19.1 18.2 2012 18.8 14.7 Moved Last 2 Years 2010 7.9 7.2 2012 13.4 11.5 Age-2008 35.1 (17.0) 39.8 (17.6) Education 2008-Less than Grade 12 80.3 81.4 2008-Only Grade 12 15.7 14.4 2008-More than Grade 12 4.1 4.2 2010-Less than Grade 12 77.7 79.8 2010-Only Grade 12 16.0 14.2 2010-More than Grade 12 6.3 6.0 2012-Less than Grade 12 73.6 77.2 2012-Only Grade 12 20.3 16.4 2012-More than Grade 12 6.1 6.4 Marital Status 2008 Not Married 35.4 33.5 Married 64.6 66.5 Marital Status 2010 Not Married 35.5 32.2 Married 64.5 67.8 Marital Status 2012 Not Married 36.5 32.2 Married 63.5 67.8 Household Assets (0-26 score) 2008 5.8 (3.5) 5.7 (3.4) 2010 5.8 (3.6) 5.7 (3.6) 2012 7.1 (3.6) 6.9 (3.6) Employment Status 2008 Not Active in the Economy 41.0 47.0 Unemployed 16.2 21.7 Employed 42.9 31.3 Employment Status 2010 Not Active in the Economy 46.3 59.3 Unemployed 14.5 15.3 Employed 39.2 25.4 Employment Status 2012 Not Active in the Economy 34.9 51.6 Unemployed 18.1 18.2 Employed 47.0 30.2 N= 2741 4985 Locational fixed effects not presented. Percentages might not add to 100 due to rounding.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 19

Table 2: Unweighted Descriptive Statistics of Observed and Expected Differences in Migration Perceptions and Behavior, Black South Africans 2008-2010-2012 (Percentages) Stayed Moved In Prior 2 Years χ2 Perceived Move in 2 Years 2008, Men 2008-2010 2008-2010 *** No/Unlikely 93.0 7.0 Possibly/Definitely 87.8 12.3 Perceived Move in 2 Years 2008, Women 2008-2010 2008-2010 *** No/Unlikely 94.3 5.7 Possibly/Definitely 84.7 15.4 Perceived Move in 2 Years 2010, Men 2010-2012 2010-2012 * No/Unlikely 87.5 12.5 Possibly/Definitely 84.1 15.9 Perceived Move in 2 Years 2010, Women 2010-2012 2010-2012 *** No/Unlikely 89.7 10.3 Possibly/Definitely 85.0 15.0 Perceived Move in 2 Years 2008, Men 2010-2012 2010-2012 # No/Unlikely 87.2 12.8 Possibly/Definitely 84.4 15.6 Perceived Move in 2 Years 2008, Women 2010-2012 2010-2012 *** No/Unlikely 89.7 10.3 Possibly/Definitely 81.8 18.2 Note: # p<.10, * p<.05, ** p<.01, *** p<.001.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 20

Table 3: Weighted Logistic Regressions of the log odds that respondents will move during the next two years versus migration in the previous two years, key variables only Men Women Chance of Moving in 2 Yrs. Moved Last 2 Yrs. Chance of Moving in 2 Yrs Moved Last 2 Yrs. 2008 2010 2012 2010 2012a 2012b 2008 2010 2012 2010 2012a 2012b Age ‘08

• 0.02***
• 0.02#
• 0.03**
• 0.03***
• 0.05***
• 0.04***

Educ ‘08 (ref. <gr.12) gr.12 0.56** 0.68* 0.34# 0.33* 0.12 0.21 >gr.12 0.76* 1.13* 0.43 0.80** 0.37 0.58* Age ‘10

• 0.01#
• 0.03**
• 0.01***
• 0.05***

Educ ‘10 (ref. <gr.12) gr.12 0.75*** 0.28 0.23 0.12 >gr.12 0.67* 0.94** 0.13 0.37 Age ‘12

• 0.04***
• 0.02***

Educ ‘12 (ref. <gr.12) gr.12 0.29 0.50*** >gr.12 0.43 0.61** Chance of Moving in 2 Years? (2008 value) 0.15 0.21 0.93*** 0.38* Chance of Moving in 2 Years? (2010 value) 0.27 0.41* Observations 2606 2555 2684 2606 2549 2606 4800 4754 4911 4800 4754 4800 Pseudo R2 0.072 0.076 0.065 0.064 0.070 0.060 0.074 0.076 0.080 0.139 0.087 0.085 Note: # p<.10, * p<.05, ** p<.01, *** p<.001. Models also include location fixed effects (province by urban or rural location), marital status, and employment

• status. Weights to account for sample attrition are employed. Equations 1 and 2, above, depict the estimating procedure.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 21

Table 4: Weighted Logistic Regressions of the log odds that respondents migrated in the previous two years by perceptions of migration sub-samples

Men Women Moved 2008-2010 Moved 2008-2010 Moved 2010-2012 Moved 2010-2012 Moved 2010-2012 Moved 2010-2012 Moved 2008-2010 Moved 2008-2010 Moved 2010-2012 Moved 2010-2012 Moved 2010-2012 Moved 2010-2012 Did Not Perceive Migration 2008-2010 Perceived Migration 2008-2010 Did Not Perceive Migration 2010-2012 Perceived Migration 2010-2012 Did Not Perceive Migration 2008-2010 Perceived Migration 2008-2010 Did Not Perceive Migration 2008-2010 Perceived Migration 2008-2010 Did Not Perceive Migration 2010-2012 Perceived Migration 2010-2012 Did Not Perceive Migration 2008-2010 Perceived Migration 2008-2010 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Age ‘08

• 0.02#

0.02

• 0.03**
• 0.02
• 0.05***
• 0.04**
• 0.05***
• 0.02*

Educ ‘08 (ref. <gr.12) gr.12 0.87** 0.61 0.31 0.35 0.20 0.24 0.19 0.19 >gr.12 1.57***

• 0.44

0.64

• 0.31

0.38 0.66 0.86* 0.00 Marital Status ‘08 (ref. Married) Not Married 0.10 0.10 0.59* 0.97#

• 0.32
• 0.42

0.05 0.26 HH Assets ‘08

• 0.05

0.03

• 0.01

0.05

• 0.01
• 0.08

0.02 0.00 Employment ’08 (ref. Employed) Not Active in Econ.

• 0.37
• 0.38
• 0.22
• 0.74#

0.43 0.22 0.41#

• 0.12

Unemployed

• 0.70
• 0.16
• 0.05
• 0.03

0.37 0.90* 0.47*

• 0.00

Age ‘10

• 0.03**
• 0.02
• 0.05***
• 0.04**

Educ ‘10 (ref. <gr.12) 0.00 0.00 0.00 0.00 gr.12 0.06 0.60 0.19

• 0.13

>gr.12 0.97** 0.57 0.30 0.59 Marital Status ‘10 (ref. Married) 0.00 0.00 0.00 0.00 Not Married 0.44 0.89 0.33

• 0.26

HH Assets ‘10 0.03

• 0.00
• 0.01

0.03 Employment ’10 (ref. Employed) Not Active in Econ.

• 0.33
• 0.08
• 0.26

0.36 Unemployed

• 0.29

0.24 0.13 0.79# Constant

• 1.60
• 1.74
• 0.52
• 3.19#
• 0.87
• 2.98*
• 0.32
• 2.97**
• 0.84
• 0.29
• 0.10
• 1.38

Observations 2104 475 2060 469 2111 480 4041 750 3892 835 4041 759 Pseudo R2 0.102 0.097 0.079 0.083 0.063 0.093 0.105 0.166 0.086 0.105 0.093 0.062

Note: # p<.10, * p<.05, ** p<.01, *** p<.001. Models also include location fixed effects (province by urban or rural location). Weights to account for sample attrition are employed.

Myroniuk, Tyler W. 2018. IUSSP working paper. Do not cite without author’s permission. 22

Table 5: Weighted Fixed-Effects Logistic (Conditional Logit) Regressions of the log odds that respondents will move during the next two years versus migration in the previous two years, key variables only Men Women ∆ Chance of Moving in 2 Yrs. ∆ Moved Last 2 Yrs. ∆ Chance of Moving in 2 Yrs. ∆ Moved Last 2 Yrs. HH Assets

• 0.05*
• 0.11***
• 0.08***
• 0.13***

Labor Force (ref. Employed) Not Active in Economy

• 0.20
• 0.97***

0.22#

• 0.57*

Unemployed

• 0.29#
• 0.42

0.02

• 0.54*

Marital Status (ref. Married) Not Married

• 0.42#
• 0.22

0.00

• 0.07

Chance of Moving in 2 Years?

• 0.05

0.32# Observations 3287 838 5457 1336 Pseudo R2 0.010 0.073 0.016 0.068 Note: # p<.10, * p<.05, ** p<.01, *** p<.001. Models also include location fixed effects (province by urban or rural location). Weights to account for sample attrition are employed. Models produce estimates only from cases within respondents where changes are observed in the outcome variable between waves. Equation 3, above, depicts the estimating procedure.