Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

cognitive skills presentation EF 2

No description
by

David Newhouse

on 27 April 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of cognitive skills presentation EF 2

4. Results How important are cognitive skills in determining youth employment outcomes? 1. Motivation 5. Conclusions Are cognitive skills the key to improving youth employment outcomes? Test scores more strongly associated with agricultural work But cohorts with higher educational attainment are more likely to work in “High-Status” occupations.
Share working in Agriculture Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Comparable effects on wage work and sectoral productivity Education negatively related to productivity in TIMSS subsample
But productivity data only available for a limited set of countries Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Both scores and years of education predictive of employment outcomes Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Adding countries and looking at more recent time period raises questions about robustness The conditional correlation between growth and cognitive skills is reduced when 24 countries for which GDP in 1960 is unavailable are added Data Coverage Note: Cohorts are defined separately by gender. On average countries administered each test roughly twice, to four different cohorts. Association between TIMSS and lower unemployment ratios and rates Note: Regressions also include regional dummies and a linear time trend. s.e’s clustered on country A-M scores associated with higher enrollment, lower employment, and lower idleness and unemployment Note: Regressions also include regional dummies and a linear time trend. s.e’s clustered on country Many results fragile to adding country fixed effects, excepting reductions in unemployment Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, and a linear time trend, as well as country fixed effects. Standard errors clustered on country. And to some extent effects on job quality indicators Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, and a linear time trend, as well as country fixed effects. Standard errors clustered on country. In PISA data, higher scores significantly associated with greater subsequent education Note: Regressions also include regional dummies and a linear time trend. s.e’s clustered on country More similar effects by gender in the Altinok-Murseli test score database Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. If anything, slightly stronger effects on job quality outcomes for men Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. More dispersion in test scores within countries leads to lower rates of continued enrollment Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Summary Statistics Higher PISA scores associated with much higher school enrollment, less idleness and unemployment Note: Regressions also include regional dummies and a linear time trend. s.e’s clustered on country Cohorts with higher scores are less likely to work in agriculture, more likely to work in wage employment Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Summary of average effects on five outcomes PISA unemployment and employment effects potentially larger for women Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. TIMSS unemployment effects also driven by women Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. What about all youth? Consider relationship between test score and youth employment outcomes for all youth in the cohort
Potential indirect effects on all youth of higher cognitive scores for eligible students
Compare estimated effects of average years of education with test scores. Conclusion High test scores and high educational attainment have very different effects on youth employment outcomes
Higher test scores associated with less unemployment, and less agricultural employment
Higher education if anything associated with slightly more unemployment and agricultural employment

Questions for further research
Are there particular types of cognitive skills that are more relevant to future youth labor market outcomes?
How important is youth unemployment and agricultural employment in developing countries?
For future youth labor market outcomes and social cohesion?
What institutions and policies can improve scores on cognitive skills exams? Cognitive Skills and Youth Employment
Subtitle Jean N. Lee
David Newhouse
World Bank Human Development Network
Labor Markets and Youth Team Are cognitive skills the key to prosperity? “There is strong evidence that the cognitive skills of the population have powerful effects on individual earnings, on the distribution of income, and on economic growth.”
Hanushek and Woessman (JEL 2008)
But evidence is largely based on strong cross-country correlations between test scores and past growth
Individuals with higher measured skills earn somewhat more
Cognitive test scores (between 1960 and 2003) is highly correlated with GDP growth (between 1960 and 2010)
Difficult to interpret causally Evidence for a causal effect is far weaker HW offer three main arguments to support a causal effect of skills on growth
The relationship between cognitive skills and growth survives excluding East Asian countries
Skills differentials as measured by international tests are rewarded in the US labor market
Countries’ test scores are not related to school expenditure, making it less likely that growth is causing skills
But none of these arguments convincingly show that improved skills cause growth rather than the reverse.
Per pupil spending is often weakly correlated with educational outcomes in developed counties
More developed economies likely have greater returns to cognitive skill and better-educated teachers and parents Contributions of this paper Revisits the relationship between cognitive skills and growth
By focusing on the relationship between test scores and subsequent growth
By looking at growth since 1990 and adding countries

Extends analysis to examine how cognitive skills relate to youths’ future employment outcomes
Which labor market outcomes are affected?
How do effects vary by gender?
Are effects stronger at different levels of development? Main findings Somewhat fragile econometric evidence that increasing cognitive skills raises growth
Strong positive unconditional correlation
Weak conditional correlation when adding new countries and focusing on recent time period (1990-2010)
Between 2001 and 2009, reading skills improved faster in rapidly growing countries.
Unlikely that increases in scores among 15 year olds between 2001 and 2009 contributed much directly to higher GDP growth throughout economy
More likely explanation: Kids in fast-growing countries have greater incentives to develop cognitive skills and are exposed to higher-quality education
We look at intermediate outcomes between education quality and growth to try to fill in the causal chain: labor markets and job quality Main findings (cont.) Cognitive skills increase school enrollment, decrease youth idleness, reduce unemployment, increase wage work and facilitate transitions out of agriculture.
Individuals in higher scoring cohorts are much more likely to remain enrolled in school
Cohorts that scored higher are less likely to be idle or unemployed
Cohorts that scored higher are less likely to work in agricultural sector
Higher scores associated with greater share of employment in wage work.
Some evidence for effects on working in a higher-status occupation. Data: Cognitive Skills Data mainly from three main sources:
PISA
OECD Programme for International Student Assessment, administered in 2000, 2003, 2006 and 2009
TIMMS
Trends in International Math and Science Study, administered in 1995, 1999, 2003, 2007 and 2011
Altinok-Murseli
Alternative collection of standardized test-score data developed by Altinok and Murseli (2007)
Includes PISA and TIMSS, as well as IAEP, PASEC, IEA, LLECE and SACMEQ
In all cases, average scores across all subjects
All cognitive test scores also partly reflect non-cognitive skills
Also use country cognitive skill measure in growth analysis
From Hanushek and Woessman (2009)
Covers 69 countries using tests from 1963 to 2003 Data on labor market outcomes International Income Distribution Database (I2D2 revision 2)
Standardized labor market data from many household surveys

Includes information on individuals’ age, gender, educational attainment, and labor market outcomes

Two main sets of outcomes:
1. Activity status
Share of population employed, student, idle, unemployed
Unemployment rate

2. Job quality proxies
Share of employment in wage or salaried job
Share of employment in agricultural sector
“High-status occupation”
Manager, professional, technician, clerical, service, or sales
Average productivity across three sectors (agriculture, industry, services)
Caveat: Productivity data available for very few countries Link labor data to test scores Define cohorts based on country, birth year, and gender

Assign test score to cohorts based on expected age at test
Assume enter kindergarten at age 5

Examine cohort’s subsequent labor market outcomes
Limit to youth aged 15 to 24 Method 2: All youth Divide all youth into cohorts based on cohort and survey year
Estimate


Where:
is the average labor market outcome of cohort c in year t
is the average cognitive test score associated with cohort c
is the average years of education of cohort c
contains the age of cohort c in year t and its square.
is country c’s log per capita GDP in 1990
is a vector of dummies containing cohort c’s region
is an error term clustered by country Results look somewhat similar in middle and low-income countries Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Eligible youth benefit from high scores Higher score cohorts enjoy better youth employment outcomes along key dimensions.
Higher school enrollment and lower rates of working
Especially for women
Less unemployment and idleness
Less agricultural employment, more wage employment and possibly higher occupational status
Especially for men
Results generally sensitive to including country fixed effects and often to restricting sample to low and middle-income countries Conclusion Somewhat fragile evidence that higher-score countries enjoy faster subsequent growth
Higher scores benefit youth along important dimensions
Higher school enrollment rates
Lower unemployment ratios, especially for women
Lower chance of agricultural employment
Increases in wage employment relative to uncompensated work
No discernible impact on productivity, but possible effects on occupational status
Results are somewhat robust
When restrict sample to low and middle-income countries
When consider all youth and include years of education
Results are often sensitive to the inclusion of country fixed effects, with the exception of the outcomes of unemployment and agricultural employment Main Contribution of this paper Examines how cognitive skills relate to youths’ future employment outcomes
Consider variety of labor market outcomes
Estimate effects separately by gender
Unlike growth, effects may be apparent quickly Main findings 1. High average scores associated with reduced unemployment and increased school enrollment.
Generally robust to inclusion of country fixed effects.

2. Some indication that higher scores help create better jobs
Higher incidence of wage employment
Lower share in agricultural sector
Scattered evidence of increase in higher-status occupations.

3. Greater dispersion in PISA scores reduces school attendance and increases employment
Suggests steep decline in net returns to university attendance among low-scoring youth

4. Years of education remains an important determinant of employment outcomes when controlling for cognitive skills. Data on Cognitive Skills Three main sources:

1. PISA
OECD Programme for International Student Assessment (2000, 2003, 2006 and 2009)

2. TIMSS
Trends in International Math and Science Study (1995, 1999, 2003, 2007 and 2011)

3. Altinok-Murseli Metadata
Alternative collection of standardized test-score data developed by Altinok and Murseli (2007)
Includes PISA and TIMSS, as well as IAEP, PASEC, IEA, LLECE and SACMEQ
Unlike PISA and TIMSS, scores not disaggregated by gender

In all cases, average scores across all subjects
All cognitive test scores also partly reflect non-cognitive skills Relationship between cognitive skills and recent growth is not entirely robust The conditional correlation disappears when 24 countries for which GDP in 1960 is unavailable are added Suggestive evidence that economic growth improves reading scores Countries that grew faster improved more on their PISA reading score between 2000 and 2009
Test participants too young to contribute to growth in nine years, so causality likely runs from growth to skills Despite strong relationship between test score and per capita GDP… H-W country score vs. Per capita GDP in 1990 Method 1: Eligible subsample Growth is limited as an indicator Impact of skills measured at age 15 on growth may only appear 30 years later
Level of observation is the country
The strong relationship between growth (between 1965 and 2000) and tests (between 1960 and 2010) could results from mutual causality
This work investigates relationship between test scores and subsequent labor market outcomes
Examine future outcomes based on past tests
Examine outcomes at the cohort level Labor market outcomes International Income Distribution Database
Standardized data from 350 household surveys
Includes information on individuals’ age, gender, educational attainment, and labor market outcomes
We examine several outcomes
Activity status
Share of population employed, student, idle, unemployed
Unemployment rate
Job quality
Share of employment in wage or salaried job
Share of employment in agricultural sector
“High-status occupation”
Manager, professional, technician, clerical, service, or sales
Average productivity across three sectors (agriculture, industry, services)
Caveat: Productivity data available for very few countries Link labor data to test scores Define cohorts based on country, birth year, and gender
Assign test score to cohorts based on expected age at test
Assume enter school at age 5
Examine cohort’s subsequent labor market outcomes between ages 15 and 24 using two methods
First, focus on eligible subsample of youth
Examine outcomes of youth with sufficient education to take exam
For example, only youth that completed 9th grade for PISA and TIMSS
To ensure sufficient data on outcomes for eligible subsample, average with outcomes of youth one year older and younger
Then, examine outcomes for all youth in cohort
Allows for inclusion of cohorts’ average years of education
Sufficient data to examine outcomes for the single test cohort 2. Data and Specification Divide youth with sufficient education to be eligible for test into cells
Cells defined by cohort (country, gender, birth year) and survey year
Estimate


Where:

is the average labor market outcome in year t of eligible youth, across three cohorts: cohort c and one year older and younger.
is the average cognitive test score associated with cohort c.
is a vector containing the age of cohort c in year t and its square.
is the log per capita GDP of cohort c’s country in 1990.

is a vector of dummies containing the region of cohort c’s country (high-income countries classified as separate region).

is the error term, clustered by country. Good reasons to be skeptical about direction of causality

H-W evidence is largely based on strong cross-country correlations between past test scores and past growth
Average cognitive scores (between 1960 and 2003) are highly correlated with average GDP growth (between 1960 and 2003)
Causality likely goes partly from growth to skills acqusition Is improving cognitive skills the key to creating jobs for youth? Average labor market outcome of cohort c in year t Cohort c's average test score Age of cohort c in year t Female dummy Vector of predetermined characteristics for cohort c's country:
1. The country's log per capita GDP in 1990
2. Imputed gender-specific 1991 youth unemployment rate and
labor force participation rate (from ILO KILM)
3. Share of GDP in natural resources in 1990 Linear time trend Error term clustered on country Main Specification Data coverage Note: cohorts are defined separately by gender 3. Theoretical Framework Jean N. Lee
David Newhouse
World Bank Human Development Network
Labor Markets and Youth Team Link labor data to test scores

First, focus on eligible subsample of youth
Calculate outcomes only for youth with enough education to take exam
Also Include youth one year older and younger to ensure sufficient numbers of eligible youth

Then, examine outcomes for all youth in cohort
Allows for inclusion of cohorts’ average years of education
Sufficient data to examine outcomes for the single test cohort How could cognitive skills affect unemployment and job quality? Add cognitive skills to standard search model
Lifetime payoff if employed: Discounted expected lifetime payoff if employed at job with with wage w Separation probability Wage this period Expected lifetime payoff of being unemployed
rather than working at wage w How could cognitive skills affect unemployment and job quality? Lifetime payoff if unemployed Discounted expected lifetime payoff of being unemployed payoff from being unemployed Index of cognitive skills Arrival rate of offers Index of cognitive skills Expected lifetime payoff of option to take a job if unemployed Eligible Youth Descriptive statistics Higher PISA scores associated with much higher school enrollment, less idleness and unemployment Note: Regressions also include regional dummies and a linear time trend. s.e’s clustered on country Cognitive skills associated with increased schooling, reduced unemployment, and better job quality Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Effects on unemployment and ag employment most robust in subsample of developing countries Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Developing countries are low and middle-income countries as defined by the World Bank. Cognitive skills are an important tool to improve youth employment outcomes Evidence on employment outcomes stronger than for growth
Higher school enrollment and less unemployment
Especially for women and in higher-income countries
Less agricultural employment and increased wage employment
Especially in middle-income countries

Insufficient evidence to conclude that cognitive skills are the most important input. Cognitive scores vs. Attainment Both test scores (among high school students) and educational attainment are important determinants of job quality

Cognitive skills more strongly associated with reduced unemployment and agricultural employment than attainment Potential mechanisms through which cognitive skills could affect outcomes 1. If cognitive skills help unemployed workers generate more offers
Expect less unemployment and better job quality

2. If high dispersion in cognitive skills makes workers more likely to separate
Expect greater unemployment

3. If cognitive skills help generate better offers
Expect improved job quality
Could reduce unemployment if youths' wage expectations adjust with a lag. Interpretation Both achievement and attainment improve jobs for youth
Youth in higher skill economies may be better equipped to search for jobs
Greater use of IT
Better signalling institutions
Reservation wage adjust slowly to greater productivity from improved skills Additional research possibilities Are there particular types of cognitive (or non-cognitive) skills that are more relevant to future youth labor market outcomes?
Do the declines in youth unemployment and agricultural employment brought about by cognitive skills have long-term implications later in life?
What institutions and policies can improve scores on cognitive skills exams? PISA scores increase studying in higher-income countries and improve job quality in MICs TIMSS reduces unemployment and increases wage employment in MICs Note: Graphs display coefficients and CIs on test score from kernel local regressions. Weights are based on per capita GDP in 1990 using a bandwidth of one. Regressions also control for per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. In A-M data, strong wage and ag effects in MICs Note: Graphs display coefficients on test score from kernel local regressions. Weights are based on per capita GDP in 1990 using a bandwidth of one. Regressions also control for per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. This is a background paper for the 2013 WDR on Jobs. We thank Martin Rama, Jesko Hentschel, for their support and helpful comments, and Claudio Montenegro, Emiliana Vegas, Alejandro Ganimian and Nadir Altinok for providing data. Distribution of wage offers
depends on cognitive skills Rapid growth in both unemployment and educational attainment may be a key risk factor Source: Campante and Chor (2012), from World Values Survey, Barro-Lee, and WDI Weak cognitive skills could be the problem Fixed effects specification Many results not robust to adding country fixed effects, except reductions in unemployment Note: Regressions also include gender, age, and a linear time trend, as well as country fixed effects. Standard errors clustered on country. Some positive results on job quality in TIMSS and A-M data Note: Regressions also include gender, age, and a linear time trend, as well as country fixed effects. Standard errors clustered on country. Timss and A-M span longer time period All youth Both Scores and years of education predicted of labor market outcomes Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Dispersion Consider outcomes for all youth in cohort Captures indirect effects of high skills among better educated on all youth
and correlation between scores and cognitive skills of less educated youth

Allows comparison between measured skills (among eligible students) and years of education Test scores much more strongly associated with working outside of agriculture Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Alhough educational attainment slightly increases share working in "high-status" occupations Both scores and achievemnt increase wage employment substantially Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. Alhough educational attainment slightly increases share working in "high-status" occupations Returns to university likely concave in skills
Or greater disperson lowers signal value of continued schooling Inequality in test scores reduces studying Note: Regressions also include per capita GDP in 1990, gender, age, youth unemployment rates by gender in 1991, youth labor force participation rates by gender in 1991, natural resource to GDP ratios, regional dummies, and a linear time trend. Standard errors clustered on country. In PISA scores Results fairly robust Particularly effects on unemployment, which are most robust to country fixed effects
Effects of skills on job quality/economic structure may take longer to realize

Results generally strongest among middle-income countries Inequality in scores More unequal test scores associated with increased employment and less studying
But not necessarily increased unemployment
Full transcript