Skip to content

What are the long run effects of labor migration on human capital? Evidence from Malawi

Literature › Works

Taryn Dinkelman; Martine Mariotti (2014). What are the long run effects of labor migration on human capital? Evidence from Malawi.
Type working paper
Scope Wider apartheid literature
Method econometric
Period studied 1950–1998
Regions Malawi, South Africa (Witwatersrand gold mines), Southern Africa
Themes mining migrant labour, education human capital, agriculture land, labour markets wages, demography urbanisation, methods data sources
Citation key dinkelman2014

Argument

The paper asks whether adult male labour migration to the South African gold mines raised or lowered the schooling of the next generation in Malawian sending districts. It exploits two shocks: the 1967 treaty with WNLA, which removed recruitment quotas, and the 1974 plane crash, which led Banda to ban recruiting until 1977. A difference-in-differences design compares districts and sub-districts with and without pre-1937 WNLA recruiting stations, across cohorts that were or were not of primary-school age during the shocks. It finds large positive long-run education effects, so the income effect of migrant earnings outweighed the substitution of child labour for missing men. These effects appear only in non-estate districts, where child labour was a poor substitute for adult male labour. The authors also present it as a first step in studying the regional (non-South African) economic history of apartheid-era mine labour.

Evidence and method

Newly digitized 1977 Census aggregate reports (district, five-year age group, sex) matched to the 1998 Census (individual-level, aggregated to district and to 210-227 sub-district Traditional Authorities). Also historical Censuses (1931, 1945, 1966), archival data on WNLA station locations by 1937 (Malawi National Archives, Rhodes House), colonial memoranda, Malawi Statistical Yearbooks, Department of Labour reports, and secondary literature on WNLA and the mine labour system (Wilson, Lucas, Crush et al., Jeeves, Paton, Prothero).

Place in the debates

Findings

  • The 1967 treaty sharply expanded recruitment of Malawians for South African mines. (p. 3)

    In 1967, a newly signed labor treaty between Malawi and South Africa opened the way for a 300% increase in the flow of Malawians to South African mines.

  • Malawian mine migration rose and then collapsed over 1950-1990. (p. 9)

    rose from just over 10,000 men per year to a high of 120,000 men per year, and back down to almost zero

  • The centralized WNLA recruitment system kept mine wages uniform and suppressed competition for labour. (p. 9)

    The centralized and highly bureaucratic procedure of recruitment of foreign labor allowed the mines to maintain a uniform wage across the industry thereby eliminating labor competition within the industry.

  • The 1974 plane crash triggered the abrupt end of the expansion and the Malawian ban on recruiting. (p. 10)

    The expansion in labor recruitment came to an abrupt halt in April 1974, when a WNLA plane transporting miners back to Malawi crashed, and 74 miners were killed.

  • After the ban, South African mines shifted their recruitment strategy inward and did not restore pre-1974 foreign employment levels. (p. 11)

    recruitment inwards, substituting local labor for what they saw as unreliable foreign supplies (Crush 1986; Crush et al 1991, pg 129).

  • Mine earnings far exceeded the best local alternative, which helps explain sustained migration across districts. (p. 12)

    average miner earnings were at least twice as much as the average agricultural wage in the late 1960s.

  • The preferred specification gives positive and statistically significant effects on schooling for cohorts exposed to both the expansion and the ban. (p. 22)

    cohorts eligible for school during the expansion and contraction of mining employment gained an additional 0.25 to 0.37 years of education and were between 2.2 and 3.4 percentage points more likely to have ever been in primary school.

  • The migration-induced enrollment gain explains a large share of national enrollment growth in the period. (p. 4)

    The primary school enrollment effects account for between one third and one half of the total increase in enrollment rates between 1967 and 1978.

  • The authors conclude that income effects outweighed substitution of child labour at the national level. (p. 29)

    The income effects associated with mining employment seemed to have outweighed any substitution effects for child labor at national level.

  • The education gains are concentrated in non-estate districts, where child labour substitutes poorly for adult men. (p. 31)

    All of the positive long run education effects we estimate for exposed cohorts in exposed districts is driven by variation in access to mine work among districts without large agricultural estates.

  • The paper positions itself as filling a gap in the economic history of apartheid. (p. 6)

    The vast majority of this literature focuses on the South African experience of apartheid and does not consider the impacts of the structures of apartheid on South Africa

Evidence ledger

Sources used

Source Kind Period How used
Malawi 1977 Population Census (aggregate reports, district x five-year age group x sex), newly digitized population census 1977 Supplies education outcomes for older comparison and exposed cohorts, child labor shares, ever-worked-abroad counts, and net internal migration rates.
Malawi 1998 Population Census (individual-level data) population census 1998 Provides adult education outcomes (years of schooling, any primary school, bilingualism) for younger and exposed cohorts at district and sub-district level.
Historical administrative records on locations of WNLA recruiting stations by 1937 (by district and Traditional Authority) state archives pre-1937 Constructs the pre-determined measure of migration costs (number of stations, any-station indicator) defining high and low shock exposure.
Colonial documents: Governor's Memorandum on Labor Migrancy in Malawi (1956) and Provincial Office Memo (7 Dec 1961); Government of Nyasaland 1961 Aide Memoire on WNLA activities, Provincial Reports PCN1-21-16 state archives 1956-1961 Documents the procedure and costs of obtaining a mine job via WNLA stations.
Malawi Department of Labour Annual Reports (foreign travel documents issued, 1958-60 and 1963-67) official statistical series 1958-1967 Flow measure of migration (men issued foreign travel documents) to validate that WNLA stations predict outmigration.
Historical Censuses of Malawi (1931, 1945, 1966) digitized for district baseline variables population census 1931, 1945, 1966 Baseline controls (1931 population density, 1945 literacy, 1966 wage employment) and balance tests.
Treaty Series No. 10/1967, Agreement between South Africa and Malawi on employment and documentation of Malawi nationals legislation gazette 1967 Establishes the 1967 removal of recruitment quotas and deferred-pay provisions (the expansion shock).
Malawi Statistical Yearbook 1972 (Table 7.9, deferred pay and remittances) and 1995 official statistical series 1967; 1992-1995 Value of deferred pay/remittances in 1967 used to compute income per returning miner; 1995 yearbook gives number of primary schools in 1992.
Malawi/Nyasaland Ministry of Education Annual Report 1960 and Malawi Ministry of Education 1977 'Major trends in education in Malawi 1974-1976' official statistical series 1960-1977 Shows primary school numbers did not grow, ruling out school supply expansion as confounder.
Chamber of Mines of South Africa Annual Report 1973 company records or reports 1973 Used to argue that the 1974 recall was unexpected because no concerns about recruiting were noted.
Newspaper: The Observer-Reporter, 5 April 1974, '77 Persons Killed in Plane Crash' newspapers periodicals 1974 Documents the April 1974 WNLA plane crash that triggered the labor ban.
Heyneman 1980, World Bank Staff Working Paper 420 (enrollment data, school fees) secondary literature compilation 1959-1978 National enrollment numbers and school fee levels used for background and to compute enrollment-effect share.
Pryor and Chipeta 1990 (estate earnings) and Wilson 1972 / Crush et al 1991 (miner earnings) secondary literature compilation 1966-1974 Compares average estate wages with miner earnings to show mines paid at least double the next best alternative.
Annual WNLA recruitment series for Malawian miners, 1950-1994 (Figure 1) secondary literature compilation 1950-1994 Documents the rise from about 10,000 to 120,000 recruits and collapse to near zero that defines the two shocks.
Christiansen (1983) district classification of estate districts, corroborated with FAO crop suitability measures secondary literature compilation 1960s-1970s Defines estate versus non-estate districts for heterogeneity analysis.

New data

Newly digitized 1977 Census district-age-sex aggregates (matched to 1998 Census microdata aggregated to district and TA), digitized 1931/1945/1966 census district variables, and a new geocoded dataset of WNLA recruiting station locations by district and Traditional Authority as of 1937 (27 districts, 227 TAs).

Inference

Quantification: Econometric (causal design).

Difference-in-differences combining spatial variation in historical WNLA station presence (district and TA) with cohort eligibility for primary school during the 1967-73 expansion and 1974-77 ban, using district and region fixed effects and region trends; supported by event-study-style age-group figures, internal-migration bounding exercise, and split-sample heterogeneity by estate versus non-estate districts and cross-sectional child labor correlations.

Key claims

Claim Basis Strength
Cohorts age-eligible for primary school during the mining expansion and ban in districts with more WNLA stations gained about 10-15% more schooling and 5-8% higher any-primary share. (p. 22) Table 5 DiD regressions on 1977 and 1998 Census district-age-sex cells (N=480) with district FE and region trends. Systematic data
Results are robust at the sub-district level using only younger comparison cohorts, though point estimates are smaller. (p. 24) 1998 Census microdata aggregated to TA-age-sex cells, Table 6. Systematic data
Selective internal migration does not explain the results; bounded estimates remain positive and significant. (p. 28) Net migration rates computed from 1977 Census birthplace data, assumed constant in 1998, applied in worst-case imputation (Table 7). Partial data
School supply expansion cannot explain the results because the number of primary schools did not grow between 1960 and 1990. (p. 5) Ministry of Education Annual Report 1960, Statistical Yearbook 1995, Ministry of Education 1977. Partial data
WNLA stations predict higher labor outmigration and are a valid proxy for lower migration costs. (p. 13) Table 1 regressions of 1977 Census ever-worked-abroad counts and Dept of Labour travel documents on station measures. Systematic data
Positive education effects arise only in non-estate districts, where child labor is a poor substitute for male labor. (p. 31) Split-sample regressions (Table 9) and 1977 Census child labor correlations (Table 8), with estate districts defined from Christiansen (1983). Partial data
Income effects from deferred pay dominated substitution effects; returning miners received lump sums equal to about three years of local income, so effects were larger for ban-period cohorts. (p. 29) Deferred pay figures (Statistical Yearbook 1972), miner and estate earnings from secondary sources, treaty terms; cohort DiD coefficients. Partial data
The 1967 treaty produced a roughly 300% expansion, with 20% of adult men missing at the peak, followed by a 1974 ban after the plane crash. (p. 9) WNLA recruitment series (Figure 1), treaty text, secondary accounts (Lucas, Chirwa), newspaper report. Partial data

The author's caveats

Census data lack district of birth, so exposure is measured with error from internal migration (addressed with bounds assuming constant net migration rates and equal education of migrants). The 1977 Census gives only five-year age groups, making cohort eligibility fuzzy; child labor is observed only in 1977 so mechanism evidence is cross-sectional and 'suggestive'; no pre-period child labor data exist; mortality selection concerns prevent using older 1998 cohorts; WNLA stations were non-randomly placed.

Gaps in the evidence

The design lacks individual-level birthplace and 1967-77 child labor or school enrollment panel data, so mechanisms (income vs substitution, child labor) rest on a single 1977 cross-section and unverified deferred-pay magnitudes. Mine-level or district-level recruitment counts, actual deferred-pay disbursements by district and school-level enrollment/fee records from archives (existing in the Malawi National Archives and WNLA/TEBA records) would test the exposure and income channels directly. The parallel-trends and no-concurrent-shock assumptions (e.g. estate growth) are only addressed with fixed effects and not tested with pre-1967 education microdata.

Replicability: Data described but not public.