Naureen Karachiwalla and Albert Park - Promotion Incentives in the Public Sector: Evidence from Chinese Schools

Promotion Incentives in the Public Sector: Evidencefrom Chinese Schools

Naureen Karachiwalla and Albert Park

IFPRI (PHND) and HKUST

[email protected]

November 10, 2015

Naureen Karachiwalla and Albert Park Promotion Incentives in the Public Sector November 10, 2015 1 / 17

Motivations

Low public service delivery in low-income countries, including forteachers.

Much interest in incentivizing teachers.

The public sector faces greater incentive problems.

Thus far, focus on bonuses.

China is one of few countries with promotion incentives, verysophisticated.


Overview

1 Motivations

2 Related Literature

3 Promotions in China and Data

4 Predictions from Theoretical Model

5 Empirical Specification

6 Results

7 Implications


Related Literature

Theory: Promotions as tournaments

Lazear and Rosen (1981) - promotions also induce effortMacleod and Malcolmson (1988) - employees that differ by abilityGibbs (1989) - multi-person tournaments with heterogeneous employees

Empirics:

Wage determination in firms - Baker, Gibbs and Holmstrom, 1994;Medoff and Abraham, 1980; Medoff and Abraham, 1981Little direct evidence on effort - Gibbs, 1995; Campbell, 2008; Kwon,2006; Haeck and Verboven, 2011


Promotions in China

Four ranks at each of Primary and Middle school.

To apply for a promotion, need:

To wait a certain number of years (depending on education); and,Favourable evaluation scores (one ‘excellent’ or two ‘good’) in the last5 years.

Promotion depends on the number of spaces available in a township.

Wages are higher at higher rank levels, and increase every five years.


Teacher Evaluations

Annual evaluations on a four point scale: excellent, good, pass, fail.

Based on four criteria: student test scores (34%), attendance (13%),preparation (30%), and ‘attitude’ (23%).

Township committee chooses weights and conducts evaluations(classroom observation, questionnaires, principal reports).

Top 10-15% (by rank and township) get ‘excellent’, next 30-40% get‘good’ scores. Rest get a ‘pass’.

Results announced at annual teacher meetings.


Data

Gansu Survey of Children and Families (GSCF), focussed on ruralschools.3 waves, use 2007 and construct teacher panel for 2003-2006.Sampled 100 villages in 42 townships in 20 counties.Sampled the main primary and middle school in each village.Sample of 2,350 teachers.


Theoretical Model Basics

School offers promotions, teachers hired in lowest rank, n teacherscompete for k promotion slots (p∗ = k/n).

School offers an increase in wages if promoted (∆EV ).

p∗ and ∆EV are fixed - heirarchy is in a steady state.

There is a (known) distribution of skill across teachers. Teachers knowtheir skill but not that of others. Beliefs on skill updated over time.

Output: qi = si + ei + πi where πi = εi + µ. ’Luck’ is distributedsymmetrically around a mean of zero.

Effort has a cost, which is increasing at an increasing rate(C ′(e) > 0,C ′′(e) > 0)

Teacher chooses effort so that the marginal cost of effort is equal tothe gain from promotion (change in wages) weighted by howresponsive promotion is to effort - C ′(e) = ∆EV ∗ dp/de


Theoretical Model Predictions

Incentives are highest for the marginal skilled teacher (s = 1 − p∗).

When n increases but p∗ stays the same, incentives increase for thoseclose with skill percentile close to 1 − p∗ (and decrease for those withvery high or very low skill).

Teacher effort is expected to be zero if it is 5 or more years before theteacher is eligible for promotion.

Teacher effort is expected to increase in the years leading up topromotion eligibility.

Higher discount rate.More years in which current evaluation score counts.

Teacher effort is expected to decrease if the teacher is passed over forpromotion several times:

Retirement becomes closer.Update beliefs on skill.






































Empirical Specification

evit = δ1Dit +δ2a10,it +δ3a90,it +δ4nit +δ5nit ∗a90,it +δ6nit ∗a90,it +ϕi +vit(1)

evit - evaluation scores for 2003, 2004, 2005, 2006

a - ability index, dummies for top and bottom 10% of abilitydistribution

n - number of teachers in rank in township (log), and interacted withability dummies

ϕ - individual teacher fixed effect

D - dummies for:

t = X − 4, t = X − 3, t = X − 2, t = X − 1, t = X , t = X + 1,t = X + 2, ..., X = X + 15 or greatert ≤ 5 - omitted category








D - dummies for:









D - dummies for:









D - dummies for:









D - dummies for:



Results - time dummies


Results

Variable Coefficient SE

Number of teachers in same rank in township (log) 0.013 0.082Number of teachers * ability bottom 10 per-cent -0.091* 0.052Number of teachers * ability top 10 per-cent -0.043 0.062Ability in bottom 10 per-cent 0.320* 0.193Ability in top 10 per-cent 0.038 0.200

Number of observations 3,683R2 0.022


What if evaluations are not a good proxy for effort?

Evaluation scores may capture both ability and effort.

Fixed effect and ability index mitigate this problem.

However, evaluation scores are related to measures of teacher timeuse and test scores. Time Use Test scores


What if principals are manipulating evaluation scores orteachers are learning?

Teachers could be learning and that would also produce an upwardtrend pre-eligibility.

The teachers in the sample have an average of 12 years of experience.

What if principals are just awarding high scores to teachers who arenearing eligibility for promotion?

Again, evaluation scores are related to time use and test scores.Ranks strongly predict test scores (Park and Hannum, 2001; Ding andLehrer, 2001)Principals are also evaluated, results announced every year, done by anentire committee at the township.


What about selection?

What if the downward trend is just bad teachers?

Fixed effect - within teacher regression.

Older teachers are just slowing down.

Similar regression for Primary high rank, there is a strong downwardtrend. Primary High

Primary high teachers also spend much less time working.Primary High time

The average age of Primary high teachers is 47 years, for those 10-15years into the rank it’s 50.

The average age of the teachers in the downwards sloping part is 44years.


Do higher evaluatin scores increase promotion probability?

Yes! One unit increase in evaluation score increases the probability ofpromotion by almost 15% (significant at the 5% level).

Regression of promotion on evaluation scores (instrumented bychange in log wages), with many controls, and county fixed effects.

Controls include: education, experience, gender, promotion rate,ability, number of teachers, county variables (education, experience,gender), time, pre-teaching test scores.


Implications

Effort responds to promotion incentives

Implications for design of incentives in the public sector

Optimal contest size and promotion rateTiming of incentivesIncentivizing teachers falling behindCombining pay for performance (within-rank incentives) withpromotion incentives (happening in China!)


Evaluation Scores and Time Use

Back


Evaluation Scores and Student Test Scores

Back


Primary High Level Regression

Back


Primary High Time Use

Back


Education

Naureen Karachiwalla and Albert Park - Promotion Incentives in the Public Sector: Evidence from Chinese Schools