NYU Abu Dhabi Conference on Education Media and

NYU Abu Dhabi Conference on Education Media and Human Development. Quantitative Analysis of Education Policy in the UK Peter Dolton Royal Holloway College, University of London and Centre for Economics of Education, London School of Economics Peter. Dolton@lse. ac. uk

‘I’m giving education just one more try… if I fail again, I’m entering politics!’

Outline • Examples of Analysis (and failures of analysis) of Policy in the UK (and data). • My take on the ‘causal’ v ‘observational’ debate - IDENTIFICATION • Brief follow up on the Andreas Scheicher presentation on a paper I am writing.

Examples of Key Education Policy Reform Questions • What are effects of National Curriculum from 1988? • What has been the effect of the Literacy and Numeracy Hour? • What is the effect of National KS tests at age 7, 11, 13, 16, 18: – Have educational standards been rising – Has publication of school results encouraged competition? • What has been the effect of the Introduction of a School Teacher Performance threshold on pay in 2000?

Some outputs are easier to observe than others!

Recent Policy Questions • Effect of Class size on outcomes • Why boys are doing so much worse than girls.

The proportion of boys and girls achieving 5 good GCSEs Source: Df. ES (2003)

Data to Answer these Questions Administrative Data on: • National Pupil Database on every child with all their scores on all KS tests. • Database of Teacher Records. • School level data on performance in KS tests. • Assorted other Admin data on House Prices, (Land Registry), Deprivation etc

Other Data Sources • Loads of good surveys – cohort data etc • BUT

• NO LINK BETWEEN ADMIN DATA • Hence impossible to find out which teacher taught which class.

Some Real Effects of these Policies Which we don’t need data to tell us. • Teaching to the test to push up school scores. • Educational ‘improvement’ by government edict. • Squashing of teacher initiative to teach – 9/11 example.

IDENTIFICATION • OLS – observational • RCT – ‘causal’ Many other techniques • Panel, Longitudinal, Cohort, Spatial • Statistical Matching • Difference-in-Differences • Regressional Discontinuity Design • Instrumental Variables - LATE

• Often involve the creative use of: • Some administrative change or rule like Miamonides Rule (Angrist and Lavy) • Changes in Policy

• Above techniques may give us as close an estimate of causal effects as you are going to get with RCTs.

If You Pay Peanuts do You Get Monkeys? A Cross Country Comparison of Teacher Pay and Pupil Performance. Peter Dolton[1] Oscar D. Marcenaro-Gutierrez[2] [1] Royal Holloway University of London & Centre for Economic Performance, LSE. [2] University of Malaga.

UK ADVERT – Make a Difference – Become a Teacher!!!!

‘To save democracy, is it? I have been hurling stones thinking it’s about teachers’ pay!’

What Makes a Good Teacher? ? Not sure we really know the answer BUT

1. Central Motivation • Why do teachers in Holland Earn 4 times teachers in Israel (after allowing for PPP adjustments)? • Kids in some countries do 2 -6 times as well as kids in other countries. • Is there a link between these 2 facts – • If we take relative salary as a measure of teacher quality, is it the case that kids perform better?

Motivation cont’d • Think of two possible basic causal mechanisms: • You pay teachers better gives them an incentive to work harder and be more effective in teaching kids. • You pay teachers better and this raises the status of the job and induces more able young people to want to be teachers in the future.

3. Data We have new data to do this with • OECD data on teachers salaries • PISA, TIMSS data on pupil performance. • PISA 2000, 2003, 2006 for Maths, Science and Reading • TIMSS 1995, 1999, 2003 Maths and Science

Data cont How do we measure teacher salary? • In terms of real PPP $ • Relative to country’s standard of living - In terms of $PPP divided by GDP per head. • Relative point in the income distribution of the country. (Assuming Income is lognormal, and we have Gini coeff and Ave earnings)

Figure 1. a. Actual and fitted Upper Secondary school teachers’ salaries after 15 years experience in 2007 $ PPP (2005)

Figure 1. c. Actual and fitted Upper Secondary school teachers’ salaries after 15 years experience relative to the earnings distribution of the whole population (2003)

Teacher Salaries We also have data on • Starting • After 15 years • At top of scale. AND • Primary • Lower Secondary • Upper Secondary

Figure 2. Relative position (percentile) of teachers’ salaries in the earnings distribution of the pop’n (Upper Secondary Education)

Data Conclusions • Most countries pay their teachers between 5075%ile. • Some countries have flat career salaries: Denmark, Finland, Sweden, Peru • Other significant advancement: Austria, Belgium, France etc • Some countries teacher wages are falling back: Indonesia, Chile, Thailand, Australia • Others teachers are being paid better: Brazil, Czech, Uruguay

Figure 3. a. Standardised Average Scores (8 th grade students) by country (2006)

Figure 4. Score’s percentile at 8 th grade students as a function of teachers’ salaries after 15 years experience.

4. Econometric Estimation • Teacher Salaries -> function of: Supply of Teachers Demand for Teachers • Production Function for Pupil Outcomes: function of : Teacher Hours, Pupil Teacher Ratios Educational Expenditure GDP Growth

4. Econometric Estimation: Identification • Use panel data therefore: • With fixed country effects we arguing that there are not systematic influences on pupil outcomes which are: – Not measured i. e. in u – and correlated with teacher earnings. • Then identification of ‘causal effects’ would rely on changes.

Controls • • Teaching Hours Pupil/Teacher ratios Fraction of Women GDP growth Educational Expenditure Growth in size of teacher cohort. Growth in size of pupil pop’n

5. Results Teachers salaries vary across country: • -ve With supply • +ve with Pupil/teacher ratios • -ve with size of pupil cohort.

Marginal Effects • $5000 or 15% rise in teachers earnings • OR • 5% shift up the wage distribution for teachers • will mean. 20 of a SD in test score and hence around 8% rise in student performance.

6. Implications If we wish to improve pupil scores we need to: • Pay teachers more – further up income distn • Reduce pupil/teacher ratios To reduce inequality of student performance we need: • Reduce pupil/teacher ratios NOT Increase teaching hours as ambiguous effect.