ARE THERE SKILLS PAYOFFS IN LOW AND MIDDLE-INCOME COUNTRIES?

Transcription

1 ARE THERE SKILLS PAYOFFS IN LOW AND MIDDLE-INCOME COUNTRIES? Namrata Tognatta SKILLS GSG SEMINARS WEEK Earnings Returns to Schooling and Skills December 7, 2015

2 Outline Motivation and Research Questions The STEP Skills Measurement program Scope Skills Measures Methodology Findings from cross-country analysis Implications

3 Motivation Schooling/Education levels key to measuring productivity and labor market success Recent evidence shows declining returns while educational attainment increases Can other dimensions of skills (namely, cognitive and noncognitive skills measures) give us additional information?

4 Research Questions 1. What are the returns to schooling in select low and middleincome countries? 2. Is there an association between cognitive skills and earnings in these countries? 3. Is there an association between noncognitive skills and earnings in these countries? 4. Are there differences by gender? Valerio, A., Sanchez Puerta, M.L., Tognatta, N. and Monroy Taborda, S. (2015). Are there skills payoffs in low and middle-income countries? Empirical evidence using STEP Skills Surveys. Working Paper. The World Bank. Washington, DC. (Forthcoming). Tognatta, N., Valerio, A. and Sanchez Puerta, M.L. (2015). What can cognitive and non-cognitive skills measures tell us about the gender wage gap in middle-income countries? The World Bank. Washington, DC. (Forthcoming).

5 Measures of Cognitive and Non-cognitive Skills Cognitive Skills Reading literacy Direct assessment of reading skills (same scale as PIAAC) Use of foundational skills (at work or in daily life) Reading Writing Numeracy Computer use at work Non-cognitive Skills Personality (Big Five) Openness Conscientiousness Extraversion Agreeableness Neuroticism Behavior Risk & time preference Grit Hostile attribution bias Decision-making Choices on hypothetical scenarios involving, risks, returns and time

6 Data & Methodology Wage Equation wi = a0 +a1controlsi +a2yearsofschoolingi +ei wi = a0 +a1controlsi +a3readingscoresi +ei wi = a0 +a1controlsi +a4complexityofcomputerusei +ei wi = a0 +a1controlsi +a5noncognitiveskillsscoresi +ei Controls: gender, experience, type of employment, occupation Heckman s correction for selection bias Instruments used in the first stage include shocks before age 15, socioeconomic status at age 15, and household asset index. Sample restrictions 25 to 64 year olds including wage workers and selfemployed workers, excluding part-time workers, employers, and unpaid workers. Top 1% earners excluded.

7 16.0 Distribution of schooling and earnings Men Women Earnings-Men Earnings-Women

8 Distribution of reading proficiency scores and earnings Women Men Earnings-Men Earnings-Women

9 Percentage Points Results: Returns to Schooling wi =a0 +a1controlsi +a2yearsofschoolingi +ei Findings consistent with recent literature Substantial heterogeneity across countries Controlling for occupation, type of employment and potential experience, returns are lower for women than men * * * * * * *

10 Percentage Points Results: Association between Cognitive Skills (Reading Scores) and Earnings wi =a 0 +a1controlsi +a 3ReadingScoresi +ei wi =a 0 +a1controlsi +a 2Schoolingi +a 3ReadingScoresi +ei Net association between cognitive skills and earnings is substantial Each standard deviation increase is equivalent to raising proficiency to the next level Reduction in magnitude of association when schooling is added to the model * Without Ed * With Ed * No change in association between schooling and earnings

11 Percentage Points Results: Association between Cognitive Skills (Computer use) and Earnings wi =a 0 +a1controlsi +a 4ComplexityOfComputerUsei +ei wi =a 0 +a1controlsi +a 2YearsOfSchoolingi +a 4ComplexityOfComputerUsei +ei As the level of complexity of computer use on the job increases, so do earnings This association stays stable (the magnitudes are quite large) even after controlling for schooling * * * * * * * * * * * * * * * * * Number of workers with and jobs requiring high complexity of computer use are relatively few Browser-based Tasks MS Off. Tasks Basic Prog.

12 Results: Association between Noncognitive Skills and Earnings wi =a 0 +a1controlsi +a 5NonCognitiveSkillsScoresi +ei wi =a 0 +a1controlsi +a 2YearsOfSchoolingi +a 5NonCognitiveSkillsScoresi +ei Openness Conscient iousness Extraversi on Agreeable ness Emotional Stability Grit Decisionmaking Armenia ** * Bolivia Colombi a * ** Georgia Kenya 0.117* Ukraine 0.085** ** Vietnam ** Magnitude of the associations between personality traits and earning is similar to those found in the literature Controlling for noncognitive skills alters the relationship between schooling and earnings in Armenia, Georgia, Ghana and Vietnam.

13 Results: Association between schooling, Cognitive Skills, Noncognitive Skills and Earnings Armenia Bolivia Colombia Georgia Kenya Ukraine Vietnam Years of education *** *** *** *** *** *** *** Reading proficiency *** *** *** *** *** *** *** MS Office Tasks *** *** *** *** *** *** *** Basic Programming Tasks *** *** *** *** *** *** *** Openness *** *** *** *** *** *** *** Observations Returns to schooling controlling for all cognitive and noncognitive skills are around 4 percentage points Above and beyond schooling, cognitive skills as measured by reading skills scores matter in Ukraine Computer-related skills continue to have large associations with earnings Noncognitive skills like Openness are significantly related to earnings in some of the countries

14 Results: Gender Differences Returns to Schooling Payoff to Cognitive Skills Women Men Women Men Armenia 0.07 *** 0.02 *** Ukraine 0.07 *** 0.03 *** Bolivia 0.07 *** 0.04 *** Georgia 0.07 *** 0.06 *** Ghana 0.06 *** 0.04 *** Kenya 0.07 *** 0.07 *** Vietnam 0.05 *** 0.07 *** Colombia 0.04 *** 0.07 *** Kenya 0.22 *** 0.12 *** Georgia 0.17 *** 0.08 *** Vietnam 0.18 *** 0.10 *** Armenia 0.02 *** 0.01 *** Ukraine 0.00 *** 0.12 *** Colombia 0.03 *** 0.14 *** Ghana 0.17 *** 0.23 *** Bolivia 0.11 *** 0.13 *** Note: Weighted OLS estimates controlling for potential experience, occupation, and type of employment. *p<0.1, **p<0.05, ***p<0.01

15 Implications and Questions for Policy Schooling or years of education continues to be the most stable predictor of earnings Is it perhaps important to consider measures of schooling quality? Significant heterogeneity in associations between skills and earnings across countries What is the role of labor market characteristics/institutional factors in explaining some of this heterogeneity? Findings on computer-related skills could be suggestive of a complementarity between education and computer use/skills Noncognitive skills findings should be considered lower-bound estimates

16 Thank you. Questions? STEP Skills Measurement Program:

17 Pooled Regression Results (Country fixed effects; Noncognitive skills excluded) (1) (2) (3) (4) (5) Female *** *** *** *** *** Self-Employed *** ** * Hi Skill White Collar 0.461*** 0.283*** 0.234*** Lo Skill White Collar * ** ** Schooling 0.054*** 0.047*** 0.043*** Reading proficiency 0.126*** 0.056*** 0.044*** Browser-based Tasks 0.357*** 0.282** 0.275*** MS Office-based Tasks 0.443*** 0.340*** 0.332*** Basic Progmming Tasks 0.566*** 0.463*** 0.456*** Observations 6,699 6,699 6,699 6,699 6,699 R-squared

18 Results: Gender Differences Armenia Ukraine Bolivia Georgia Kenya Ghana Returns to Schooling Women Returns to Schooling Men Payoff to Cognitive Skills Women Payoff to Cognitive Skills Men