Skip to content

Nonparametric Bounds on Returns to Schooling: Overcoming Ability and Selection Bias

Literature › Works

Martine Mariotti; Jürgen Meinecke (2010). Nonparametric Bounds on Returns to Schooling: Overcoming Ability and Selection Bias. Australian National University (draft, version 29 January 2010).
Type working paper
Scope Post-1994 South African economy
Method econometric
Period studied 1995–2000
Regions South Africa
Themes education human capital, labour markets wages, methods data sources, inequality poverty unemployment, mining migrant labour, transition post apartheid
Citation key mariotti2010a

Argument

The paper estimates the average treatment effect of schooling on earnings for employed African men in South Africa, where omitted ability bias and low labour force participation (selection into work) make OLS unreliable. It adopts Manski and Pepper's (2000) monotone treatment response (MTR) and monotone treatment selection (MTS) bounds. It extends them with monotone instrumental variables (migrant-worker status and pensioner-household status) to handle selection into employment. The resulting upper bounds on the per-year return to schooling (12.64% in 1995 and 10.68% in 2000 for a high-school education) lie below OLS estimates (15.59% and 15.31%). The authors conclude that parametric estimates are probably biased upward by unobserved heterogeneity.

Evidence and method

Merged Statistics South Africa household surveys: OHS 1995 with IES 1995, and LFS September 2000 with IES 2000. The sample is employed African men aged 18-65 (11,702 observations in 1995 and 9,143 in 2000 in Table V; 9,527 and 8,837 in the data-section text). Bootstrap with 2,000 repetitions gives 95% quantiles. Robustness check on men aged 18-39. OLS Mincerian regressions serve as the benchmark. The literature review summarises earlier studies of returns to schooling.

Findings

  • Headline upper bounds on the high-school return fall below OLS in both years. (p. 1)

    We estimate an upper bound of 12.64 percent in 1995 and 10.68 percent in 2000. Compared to parametric estimation our bounds are informative

  • African male labour force participation is low and falling, creating serious sample selection. (p. 6)

    Table III shows that for both 1995 and 2000, African male labor force participation is remarkably low and is declining from 1995 to 2000.

  • Without selection correction, the 1995 upper bound for eight years of schooling is 11.74%. (p. 15)

    The upper bound estimate on the return to education is equal to 11.74%.

  • The bounds sit below the 1995 OLS estimate of 15.59%. (p. 16)

    In all cases, the MTR–MTS bound lies below the OLS estimator of 15.59%.

  • The selection-corrected 1995 upper bound for high school is 12.64% per year, with a 95% quantile of 14.91%. (p. 17)

    The ATE of high–school education over zero education is at most 12.64% per year of education with a 95% quantile of 14.91%.

  • In 2000 the high-school bounds are 8.39% (uncorrected) and 10.68% (corrected), well below OLS of 15.31%. (p. 17)

    the estimate for the ATE of high–school education on earnings is at most 8.39% (without sample selection correction) and 10.68% (with correction).

  • Upper bounds do not permit inference about change over time in the true return. (p. 17)

    By the same reasoning we cannot claim that on average the ATE of education on earnings has decreased between 1995 and 2000.

  • Restricting to men aged 18-39 to test for cohort effects gives higher bounds but the same qualitative result, against OLS of 18.48%. (p. 19)

    the ATE of high–school education is at most 12.84% (without correction for sample selection) and 13.06% (with correction) while the OLS point estimate is 18.48%.

  • Migrant workers earn more than non-migrants and the pension variable is used to justify the lower-bound restriction, so these are plausible monotone instruments. (p. 8)

    Table IV shows that, indeed, migrant workers earn more tan non-migrants on average, suggesting this variable is suitable as a restriction on the upper bound.

  • Mean employed African male income fell between 1995 and 2000 and educational attainment barely changed. (p. 5)

    Table I shows that the mean real income of African males has declined between 1995 and 2000