WT Week 7 Flashcards
Motivation
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“Do financial incentives coupled with monitoring improve teacher attendance and learning in India?
Increasing access to education has been among key objectives for developing countries.
Quality of education has remained low due to high absenteeism. Getting teachers to come to work has to be policy priority to make universal primary education meaningful.”
Special settings
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“India: NGO conducts experiment with objective to reduce absenteeism
Government schools have very high absenteeism rates, but government teachers are hard to incentivise/monitor
Government teachers are political force and have secure jobs/wages
Increasing number of schools are taught by para-teachers
Para-teachers have flexible contracts, limited tenures and no political connections
Para-teachers might be lower quality -> setting is ideal because para-teachers can be financially incentivised and face the risk of being fired”
Theory
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“Teachers choose between going to school (at a cost) and their outside option: they attend when the opportunity cost of absenteeism is large
Once in school, teachers choose whether and how well to teach: they teach when the marginal cost of teaching conditional on attending is low. Teacher not sibject to yearly renewal contracts and there are strong unions.
Students decide whether to attend school: depends on opportunity cost
Monitoring effect:
- Advantage: could lead to higher attendance by increasing fear of being detected
- Disadvantage: works only if punishmet is credible
Financial incentives:
- Advantage: increase benefit of attending, i.e. opportunity cost of absenteeism; can improve learning if marginal cost of teaching is low
- Disadvantage: i) incentives may be too weak or mis-targeted (multitasking); ii) may crowd out intrinsic motivation; iii) may not work when target income reached
Salaries as non-linear function of attendance:
* Flat rates for attending 10 days and less in a month
* - Extra pay for each additional day attended in the same month
* Calculation re-starts every month
* In-money: person has already worked 10 days in the month on the 21st day, will receive extra pay for each additional day he works
* Out-money: person has only worked 5 days in the month of the 21st day, will only be paid the flat rate even if he works every day until the 26th
Predition 1: in-money person has high incentive to work with no change at the start of the next month unless he has already reached his target income, in which case there would be a drop in incentive
Prediction 2: out-money person has no incentive to work but high incentive to work at start of next month; motivated to work in order to be in-money next month”
Empirical design
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“Experiment:
Randomly allocate schools to treatment and control
Treatment schools receive cameras with instructions and financial incentives
Camera: students have to take a photo of teacher at start and end of day, with 5 hours in-between
Financial incentive: fixed pay for 10 and less days in month, extra pay for each additional day
Control schools have random checks only and a flat salary
Tests for performance are designed by the researchers
Absenteeism: regress number of days the school is open on treatment
Performance: regress test scores on treatment, control for previous performance
Test scores and previous performance highly autocorrelated, so adding it to regression
reduces standard errors on all variables and improves precision of treatment effect
Measurement error: compare effect of attendance on scores in different specifications
OLS specification with random checks data at baseline
OLS specification with random checks at endline (i.e. more data)
OLS specification with camera data (more accurate)
- Structural model
Allows to separate the response to financial incentive from independent effect on monitoring
Regression discontinuity:
Variation in financial incentive depending on number of days left until end of month and number of days already attended
Treatment effect: Change in probability to attend at the first day of the new month
Identification assumption: Other factors that influence attendance do not change discontinuously at the start of the month
Monitoring does not jump, outside options do not jump
Additional advantage from model: allows to identify most cost-effective incentive scheme and alternative policies”
Regression discontinuity
Child attendance and exam results
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
Data
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“120 schools (60 treated and 60 control).
Data on attendance: from random checks conducted by independent team, from cameras at treatment schools.
Data on student performance: from pre-programme tests, from mid-term tests and post-tests, on written and oral test”
Key findings
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“1) The treatment reduces teacher absenteeism
Teachers had absence rate of 21% in treatment, compared to 42% in control and 44% at baseline
Figure 1: immediate and long-lasting improvement in attendance rates
Figure 2: reduce absence everywhere in distribution
- Fewer teachers attend less than 50% in treatment
- More teachers have a very high attendance rate in treatment
Table 2: greater effect for below-median teachers because they are brought to same level of attendance as above-median teachers (78%)
Fig. 3: the change in incentive at the beginning of a month is important for teachers who were not in the money since, in the data, we see that 65 percent of the teachers who are out of the money in a month will be in the money the following month. The teachers who were in the money, however, have a 95 percent chance to be in the money again.
2) The treatment increases instructions per student
The student attendance conditional on the school being open is the same for treatment and control
The number of days per month that the school is open increases, so each student receives more instructions per month
Children in the treatment schools, on average, received about 30 percent more instruction time than children in the comparison school –> implies that there is no lack in student demand (in which case attendance rates would fall in treatment because there will be more days with low student attendance) result could be biased (see limitations)
3) The treatment improves test performance
Conditional on previous test scores, treatment improves test performance very early on test scores are 0.17 standard deviations higher
Most able students have highest effect (since attendance unchanged, this is because these students are most able to benefit from more instructions)
26 percent of students in the treatment schools graduated to the government schools, compared to only 16 percent in the comparison schools –> improved teacher performance is inconsistent with multitasking and crowding-out of motivation
result could be biased (see limitations)
4) A weak correlation between absenteeism and performance can be due to measurement error
The correlation is weak for the OLS with random checks at mid-test because only little data was available (attenuation bias)
Correlation gets stronger as more accurate data (daily camera data) is used
Difference is no longer large at post-test because more observations from random checks available –> implies that the correlation between teacher absenteeism and performance is stronger than suggested in many studies
5) The financial incentives are the main driver of the result
The RDD shows a discontinuous jump in probability to show up at school when the teacher is out-money and no jump when he is in-money
65% of the out-money teachers are in-money next month, 95% of in-money are in-money again
Monitoring unlikely to drive the result because then the person with fewer days attended should have higher motivation to attend (which he has not) and the person with already 10 days attended should have no motivation to attend more days (which he has)
In control group no discontinuous jumps –> implies that the financial incentives work independently of the monitoring
relies on identification assumption (see limitations”
Interpretation / policy implications
Incentives Work: Getting Teachers to Come to School (Duflo et al., 2012)
“Limitations
Teacher performance: if teachers with high outside options (so less present without treatment) also teach less when at school, getting them to work reduces average effort in treatment group (more observations drawn among low performance teachers in treatment group)
Student attendence:
- If treatment induced schools with high student attendence to open more often, the student attendence rate would be biased upwards
- If treatment induced schools with low student attendance to open more often, treatment effect of attendance rate is downard biased
- Researchers cannot condition on student attendance rates for all schools prior to treatment because they can only measure attendance for schools that were open –> effect on attendance likely negative biased as treatment causes weaker schools to open more often
Regression discontinuity:
- Identification assumption requires that the teacher’s outside option and the monitoring effect do not change discontinuously at the beginning of a new month
- If supervisor sums missed days each month and punishes for fewer than 10 days, probability of being punished changes discontinuously at first day of month
Other:
- Long-term effect may be larger if social norms about teacher absenteeism change
- Does not give explanation for unchanged student attendance
Policy implications and external validity:
- Financial incentives work to increase teacher attendance and student test performance
- The barriers that prevent teachers to go to school seem to be easily resolved
- Programme could fail with government teachers
- Since para-teachers no less effective (test performance), hiring para-teachers and incentivising them could be effective policy to improve education system
- Scale-up depends on appicability in other schools and cost-effectiveness”
Theory
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
Design a two-sided experiment involving workers and firms which allows comparison of supply and demand-side interventions
Special settings
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
“Young people face a higher risk of unemployment than adults in every country. Significant funding from World Bank on skills training programmes for young people, but with limited evidence on the relative effectiveness of vocational training.
Unclear which model of training can be effective at increasing youth employment as depends on relative importance of providing formal certified training versus reducing search costs and hiring constraints for firms.”
Empirical design
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
“Worked with NGO BRAC Uganda to randomise applicants (through oversubscription design) for 6 month nationwide vocational training scholarships across 3 groups: vocational training (VT), firm-provided training throguh apprenticeships (FT) and control group.
VT selected ampong high demand vocations and received certificate of completion
FT matched workers to sample of SMEs in same sectors and offered subsidy to firm owner to hire and train the worker (no certificate provided)
BRAC conducted extensive monitoring of the training activities at both the institutes and the firms
Baseline survey and randomisation took place in 2012 and training programmes in 2013, then follow-up surveys annually for 3 years after end of training (through written sector-specific skills test)
Estimated job ladder model of worker search to show different effects on transitions between jobs and unemployment for each track; estimated using data on worker skills and monthly labour market histories of workers in the follow-up period (employment status, earnings, labour market transitions)”
Motivation
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
To understand which active labour market policies are effective at facilitating the transition of youth into remunerative employment (through vocational training and apprenticeships)
Data
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
“Firm surveys to compare firm outcomes between those offered FT and control group
Worker survey to track skills progress from application for vocational training to 12, 24, 36 month follow-up surveys
Tracks 1,700 workers and 1,500 firms over four years”
Key findings
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
“Significantly higher take-up rates in VT arm than FT due to limited interest by firms in taking apprentices
VT and FT had similarly large gains (0.4 SDs over control group) –> both prove effective in generating productive human capital
However, long-lasting effect for VT but not for FT (materialised quickly but faded after the apprenticeship, while some retained employment, the employment rates of FT group were similar to control)
VT improvements did not decline over the 3 year study period and maintained higher number of months worked per quarter
Model shows difference between VT and FT effects largely explained by unemployment-to-job transition rates, certification plays a large role as FT couldn’t credibly signal skills (i.e. FT were no more likely to find a job after the experiment than the control group)”
Interpretation / policy implications
Tackling Youth Unemployment: Evidence from a labour market experiment in Uganda (Alfonsi et al., 2020)
“VT found to be more cost-effective than FT; highlights human capital alone is not sufficient and need for credible way to show skills to potential employers
Programmes targeting human captial accumulation and skills certification are therefore likely to be most promising
Cost of VT and FT were significantly higher than workers’ monthly incomes, showing that the interventions help workers overcome credit constraints”
Motivation
Caicedo et al. (2022)
Labor market opportunities for young people are a significant concern in developing countries, where formal employment opportunities are limited, especially for those without college education. Apprenticeships, combining formal vocational education and training within firms, are seen as a solution to upgrade skills and raise productivity among young workers. However, firms’ willingness to train apprentices is a limiting factor. This study investigates firm responses to apprenticeship regulation in Colombia, where a reform required firms to train apprentices based on firm size. The analysis aims to understand the heterogeneity in firm responses, estimate training costs, and evaluate the overall benefits of apprenticeship expansion.
Special settings
Caicedo et al. (2022)
The Colombian context offers several advantages for studying apprenticeship regulation:
Youth Labor Market Issues: In the early 2000s, youth unemployment rates were around 30%, and youth informality rates exceeded 70%.
2003 Reform: Introduced apprentice quotas based on firm size, adjusted minimum wages for apprentices, and allowed firms to pay a fee to opt-out of training apprentices.
Administrative Records: High-quality administrative data at the firm level and a firm census with detailed firm characteristics.
These factors provide a unique setting to analyze firm behavior and the impact of apprenticeship regulation.
Theory
Caicedo et al. (2022)
The theoretical framework considers an economy with multiple sectors and heterogeneous firms characterized by training costs and managerial ability. Firms produce using labor from workers and apprentices, with apprentices requiring costly training to become productive. The model captures differential responses to the reform based on training costs, highlighting that high-skill sector firms with high training costs avoid apprentices, while low-skill sector firms with low training costs seek them. The model also underscores the importance of apprentices’ minimum wages in influencing training levels.
Empirical design
Caicedo et al. (2022)
Data
Caicedo et al. (2022)
The study utilizes high-quality administrative records from Colombia, including:
Firm-level Administrative Data: Provides detailed information on firm size, sector, and apprenticeship training.
Firm Census: Contains additional survey information on firm characteristics.
Pre-reform and Post-reform Data: Enables comparison and analysis of firm responses to the apprenticeship regulation.
Key findings
Caicedo et al. (2022)
- Aggregate Increase in Apprentices: The reform significantly increased the number of apprentices, expanding training opportunities for young individuals.
- Sectoral Heterogeneity: Firms in high-skill sectors tended to avoid training apprentices by reducing firm size to stay below regulation thresholds or paying fees. In contrast, firms in low-skill sectors sought to maximize apprentice intake.
- Training Costs: High training costs in high-skill sectors deterred firms from training apprentices. In low-skill sectors, training costs were lower, making apprentices more attractive.
- General and Dynamic Equilibrium Effects: In general equilibrium, the reform increased overall production by up to 3.7% dynamically, benefiting apprentices but potentially displacing incumbent workers.
Interpretation / policy implications
Caicedo et al. (2022)
The study suggests that expanding apprenticeship programs can have substantial welfare benefits, particularly when the supply of trained workers increases. However, policies should account for sectoral heterogeneity to minimize distortions in firm size and resource allocation. Sector-specific apprenticeship regulations, such as differentiated minimum wages for apprentices, can enhance welfare gains and encourage training in high-skill sectors. This approach can balance the distribution of training benefits and mitigate adverse effects on firms and incumbent workers. The findings highlight the importance of considering firm-specific training costs and sectoral differences in designing effective apprenticeship policies.
emperical settings
Caicedo et al. (2022)
Bunching methods are used to gauge firm size responses to discontinuities in apprentice quotas, while structural estimation techniques uncover training costs. The analysis covers partial and general equilibrium scenarios to assess the overall impact of the regulation and potential counterfactual policies.
