Can spatial mobility insure families against long-term impacts of economic shocks? Evidence from drought and disability in South Africa¶
Taryn Dinkelman (2012). Can spatial mobility insure families against long-term impacts of economic shocks? Evidence from drought and disability in South Africa. Dartmouth College (working paper, 13 July 2012).
| Type | working paper |
| Scope | Wider apartheid literature |
| Method | econometric |
| Period studied | 1948–1996 |
| Regions | South African homelands (TBVC and non-TBVC), rural South Africa |
| Themes | bantustans removals influx, mining migrant labour, agriculture land, demography urbanisation, inequality poverty unemployment, methods data sources, labour markets wages |
| Citation key | dinkelman2012 |
Argument¶
Dinkelman uses apartheid-era variation in mobility restrictions, with the older TBVC homelands facing the tightest and longest-lasting limits, together with district-year drought shocks. She asks whether spatial mobility insures families against the long-run health effects of childhood drought. A triple-difference design shows that early-childhood drought raises male disability by about 20% in TBVC areas relative to non-TBVC areas. Drought induces much more adult outmigration from less restricted areas than from TBVC areas. Remittance evidence from 1995-96 suggests migrants send money home in drought years. The author concludes that barriers to mobility impose long-run human-capital costs.
Evidence and method¶
1996 South African Census 10% individual sample, used to build a pseudo-panel of outmigration from 'last move' data for 1948-1986 and to measure disability among cohorts born 1948-1986. Rainfall data from over 1,000 weather stations (the appendix says over 1,600), converted to a Standardized Precipitation Index drought indicator. Maize output data from the South African Maize Board for 1964-84. GIS homeland boundaries. Census household remittance and migrant-worker data for 1995-96. Secondary literature on pass laws and homelands (Lemon, Simkins, Savage, Wilson).
Place in the debates¶
- Migration as insurance against income shocks (development economics) (see Living standards, health and social policy). Extends the literature by arguing that mobility also insures long-term health outcomes, implying an even larger role for mobility than previously emphasised.
- Long-run economic costs of apartheid influx control and homeland confinement (see Living standards, health and social policy). Finds that mobility restrictions imposed measurable long-run health and human-capital costs in the homelands, which is a hidden cost of the labour-control system.
- Fetal and early-life origins of adult health (see Living standards, health and social policy). Confirms that drought in utero and at ages 1-2 raises adult disability, especially for males, consistent with the fragile-male literature.
Findings¶
-
Drought exposure at birth raises male disability in TBVC areas by about 20% relative to the male mean. (p. 4)
Drought raises disability rates by 20% relative to the mean in TBVC areas
-
The male birth-year drought-by-TBVC effect is about one percentage point and statistically significant at 1%; the female effect is insignificant. (p. 18)
Drought in the birth year raises the rate of male disability by about 1 percentage point; it raises the number of serious disabilities by almost 0.01.
-
Drought raises outmigration from non-TBVC areas by 0.088%, a 36% rise over the district-year mean. (p. 20)
the estimate of 1 indicates that drought induces 0.088% more outmigration. Relative to mean outmigration at the district-year level (0.24%), this is a 36% increase in adults moving out.
-
The net outmigration response to drought in TBVC areas is much smaller, about a 10% increase over the mean. (p. 21)
The total effect of drought on outmigration from TBVC areas is the sum of the main effect and interaction term: 0.021% (0.088 - 0.067) or a 10% increase in outmigration relative to the mean.
-
Outmigration was lower from TBVC than non-TBVC districts, consistent with tighter restrictions on mobility in TBVC districts. (p. 12)
the percent of adults moving out of a TBVC area in any given year is significantly lower (0.21%) compared with non-TBVC area outmigration (0.25%)
-
Remittances respond to drought only in households with a migrant worker (2.68 percentage points, a 13% increase). This evidence is from 1995, after apartheid. (p. 23)
migrant worker households in drought-exposed districts were a significant 2.68 percentage points more likely to receive any remittances in 1995 (column (1)). This is a 13% increase
-
Drought in utero and at ages 1-2 raises male disability in all homeland areas, so families could not fully insure anywhere. (p. 3)
drought exposure in utero and at ages one and two increases disability rates among males in all areas, implying that families were not able to fully insure against drought shocks in any areas.
-
Fertility, mortality and cohort-size changes do not explain the disability results; any survivor bias would work against the finding. (p. 25)
Any survivor bias in the male sample from TBVC drought cohorts would lead us to underestimate the differential impact of drought on disability rates in TBVC areas.
-
The apartheid state used the homelands to hold surplus labour and dependants far from urban centres, and labour migration was granted in response to white-economy labour demand. (p. 7)
These states provided a space to locate excess labor and non-labor population resources (women, children and the aged) far from urban centers, in rural parts of the country.
-
Over 30% of rural homeland households still had a migrant worker in 1996, which supports the long persistence of migrant networks. (p. 23)
In 1996, just over 30% of all households in rural TBVC and non-TBVC districts reported at least one migrant worker
Evidence ledger¶
Sources used¶
| Source | Kind | Period | How used |
|---|---|---|---|
| 1996 South African Census, 10% individual record sample (African individuals, disability, migration, fertility, remittance questions) | population census | Census 1996; cohorts born 1948-1986; pseudo-panel of residence 1948-1986 | Provides disability outcomes, last-move migration histories, birth/prior district, fertility and child mortality, which underpin the health and outmigration regressions. |
| 1996 Census household-level data on remittance receipts and migrant workers (1995 remittances) | population census | 1995-1996 | Used to estimate whether drought in 1995 raises remittance receipts, especially in households with a migrant worker, as a proxy for apartheid-era remittance links. |
| South African Weather Service monthly rainfall data, weather station level | official statistical series | 1920-2009 | Aggregated to district-year rainfall totals to construct the Standardized Precipitation Index and the binary drought indicator (SPI below -1.5). |
| Maize Board province-level maize production data (Transvaal and Orange Free State) | official statistical series | 1964-1984 | Validates the SPI drought measure by showing maize output is more sensitive to low SPI values. |
| Statistics South Africa GIS boundary data, 1996 and 2001 Census district councils | maps spatial data | 1996 and 2001 boundaries | Defines the district unit of analysis and is overlaid with homeland maps to identify former homeland districts. |
| Political map of South Africa 1986, Perry-Castañeda Library Map Collection (homeland maps) | maps spatial data | 1986 | Overlaid with Census boundaries in ArcGIS to assign each district a TBVC indicator capturing mobility restriction intensity. |
| Secretary for Bantu Administration and Development, General Circular No. 25 (1967) | commission or official report | 1967 | Quoted as evidence of government policy that Africans were only temporarily resident in white areas and expected to return to homelands. |
| 2007 South African Community Survey and 2007 Cape Area Panel Study (multiple movers) | household or labour survey microdata | 2007-2008 | Provides an upper bound (0.01 to 0.13) on the share of repeat movers used to argue measurement error in last-move data biases results downward. |
| Posel and Casale (2006) analysis of 2002 household survey data on remittances | secondary literature compilation | 2002 | Supports the importance of migrant-remittance linkages in South Africa in recent years. |
| Mariotti (2011) study of living standards in former homelands (National Income Dynamics Panel data) | secondary literature compilation | mid-1970s labor demand shock | Used to describe the 1975-1979 labour demand shock to TBVC areas and justify excluding those years in a robustness test. |
| Wilson (1996) data on calorie consumption of rural African adults | secondary literature compilation | 1990s | Documents poor nutritional living standards in homelands so drought could affect health. |
| Coovadia et al (2009) homeland doctor:patient ratio | secondary literature compilation | 1970s | Illustrates dire public health provision in all homelands. |
New data¶
Author-built district-year panel of drought (SPI from >1,000 weather stations, 1948-1986) matched to Census-derived individual outcomes and a constructed pseudo-panel of adult outmigration histories (624 district-year observations, 16 districts); TBVC indicator built via GIS overlay of homeland maps on district boundaries.
Inference¶
Quantification: Econometric (causal design).
Triple-difference (difference-in-difference-in-differences) regressions with district and year/birth-year fixed effects, comparing drought effects across TBVC vs non-TBVC homelands (differential mobility restriction), on individual disability, district-year outmigration constructed from last-move Census data, and household remittances; supported by balance tables, composition (fertility, cohort size) checks, and measurement-error analysis.
Key claims¶
| Claim | Basis | Strength |
|---|---|---|
| Birth-year drought exposure raises male disability in TBVC (more mobility-restricted) areas by about 1 percentage point, roughly 20% of the mean, relative to non-TBVC areas. (p. 18) | 1996 Census 10% sample (298,475 males), SPI drought indicator, triple-difference regression with district and birth-year fixed effects (Table 2). | Systematic data |
| Drought exposure in utero and at ages one and two raises male disability in all homeland areas, so families did not fully insure anywhere. (p. 19) | Extended specification with drought at each age up to four (Table 3), 1996 Census male sample. | Systematic data |
| Drought induces much more adult outmigration from non-TBVC areas (about 36% over mean) than from TBVC areas (about 10%), confirming differential mobility restrictions. (p. 20) | Pseudo-panel of district-year outmigration built from Census last-move data matched to SPI drought (Table 5), with narrow-TBVC and 1960-1986 robustness checks. | Partial data |
| Remittances respond to drought in households with a migrant worker, linking outmigration to household resources; the effect is stronger in non-TBVC areas. (p. 23) | 1996 Census household data on 1995 remittances and 1995 drought; used as a proxy since no apartheid-era remittance data exist (Table 6). | Partial data |
| The disability results are not explained by drought-induced changes in fertility, child mortality, cohort size or sex composition (or HIV). (p. 25) | Census fertility/mortality of women aged 40-60 (Table 7) and district-year cohort-size regressions (Table 8). | Partial data |
| TBVC and non-TBVC areas are similar in observable socio-economic and demographic characteristics, supporting TBVC status as capturing mainly mobility restrictions. (p. 12) | Table 1 balance of means from the 1996 Census sample. | Partial data |
| Measurement error from last-move Census data likely biases outmigration estimates downward, so results are unlikely to overstate the drought effect on migration. (p. 17) | Analytical measurement-error appendix, external survey estimates of multiple movers (2007 Community Survey, Cape Area Panel Study), and 1960-1986 subsample robustness. | Partial data |
The author's caveats¶
No historical data on migration or remittances exist for apartheid; migration is reconstructed from Census last-move data assuming no one moves more than once, and the Census does not record birth district (prior district assumed) or month of birth, so drought exposure is mismeasured. Census does not capture temporary migration; remittance evidence is from 1995-96, post-apartheid. TBVC status is time-invariant, and the sample is restricted to survivors alive in 1996.
Gaps in the evidence¶
The mobility-restriction mechanism rests on a TBVC/non-TBVC dichotomy and on remittance evidence from 1995 rather than apartheid-era remittance or income flows; contemporaneous household or labour-bureau and pass-law/influx-control records, mine recruiting records (e.g. Chamber of Mines) or archival remittance data could test it directly, though largely not available now. Measures of health at birth or childhood (rather than adult disability) and the actual birth district would reduce reliance on assumptions about repeat movers. Alternative differences between TBVC and non-TBVC homelands (labour demand, agricultural conditions, health services) are only addressed through fixed effects and means comparisons.
Replicability: Data described but not public.