Technical STATA assignment - Experts only
EC338 assignment 2017-18
You can work on the assignment in groups of up to three students. You need to register to let us know who
you are working with, even if you are working alone. You should receive an email via Tabula to let you know
once the group sign-in has opened. Please register your group by Friday 17th November 2017.
The deadline for submission of the final assignment is 23.55 on Friday 15th December 2017. Late assignments
will be penalised. All groups should submit just one copy of their answers. The group number and student ID
numbers of all group members should be stated clearly on your submission.
You should answer all 10 questions in the assignment. The weight of each question is as indicated. The entire
assignment is worth 100 marks in total.
Your submission should comprise two elements:
1) A pdf document containing written answers to the questions and tables of any coefficient estimates you are
asked to produce.
2) A stata dofile containing all the commands used to produce these estimates.
The word limit (excluding tables and stata output and code) is 1500 words. Please state the final word count
clearly on your submission. Assignments that go over this limit may be penalised.
A document addressing Frequently Asked Questions will be added to the module webpage and will be regularly
updated to ensure that all groups have access to the same information when completing the assignment. You may
therefore want to check the page regularly up to the submission deadline.
Background
In recent years, many countries – including England – have moved from more centralised to more decentralised
education systems: specifically, they have introduced reforms aimed at devolving more power (giving greater
autonomy) to schools. The rationale for doing so is to increase school performance, usually measured in terms of
students’ exam results. This may occur directly – by enabling those with the greatest knowledge about their
pupils to make decisions to maximise their exam performance – or indirectly, e.g. by creating greater
competition between schools.
In the US, such reforms have led to the creation of “charter” schools; in Sweden, to “free” schools. In England,
there have been two major reforms, leading to the introduction of “grant maintained” schools in the 1990s and
“academies” in the 2000s. There is a growing literature on the effects of greater school autonomy on student and
school performance (e.g. see Epple, Romano and Zimmer (2016) for a recent summary of the literature on the
effects of charter schools), including some papers studying the effects of these reforms in England.
This assignment considers Clark (2009), who studied the effects of grant maintained schools, and Eyles and
Machin (2015), who studied the effects of academies. You will find both papers and two related datasets on the
module webpage: clark.dta contains the actual school-level data used by Clark (2009); academies.dta contains a
school-level version of the type of data used by Eyles and Machin (2015). As such, it contains panel data for
schools that converted to academies between 2002 and 2010, including information up to four years before and
three years after the year in which each school converted.1 Each variable represents a school-level average of all
pupils who took their GCSEs (end of secondary school exams) in the academy. The variables in each dataset are
clearly labelled and should be self-explanatory.
1 We make similar sample restrictions to those imposed by Eyles and Machin (2015), e.g. requiring four years of data prior
to conversion (i.e. omitting new academies) and dropping those that were previously independent (private) schools. For
academies formed from two or more predecessor schools we follow Eyles and Machin (2015) in creating weighted versions
of the characteristics included in the dataset in the years leading up to conversion.
Questions
1) Briefly outline the approach taken by Clark (2009) to estimate the impact of greater school autonomy on
exam performance. (Focus on his school-level estimates, e.g. those reported in Columns 5 and 6 of Table 1.)
Why does he adopt this approach?
(7 marks)
2) What assumptions are required for this to be a causal (consistent) estimate of the effect of greater school
autonomy on exam performance? Do you think they are likely to hold in this case? Explain why or why not.
(8 marks)
3) Replicate the results in the third row (Base+3) of Table 1. (Note that Clark includes controls for GCSE
performance in the year of the vote, and dummy variables for the term and year in which the vote took place
(gm_attempt1_ballot_year_term) and school type (school_type) from the second specification (column)
onwards. He also uses robust standard errors throughout.)
(12 marks)
4) Briefly outline the approach taken by Eyles and Machin (2015) to estimate the impact of greater autonomy
on exam performance.
(4 marks)
5) What assumptions are required for this to be a causal (consistent) estimate of the effect of greater school
autonomy on exam performance? Do you think they are likely to hold in this case? Explain why or why not.
(6 marks)
6) Use academies.dta to produce school-level estimates of the effect of academy status on exam performance
using similar specifications to those in Columns 1, 4 and 7 of Table 6 in Eyles and Machin (2015). (You can
regard schks2_eng_exp, schks2_eng_abv, schks2_mat_exp, schks2_mat_abv, schks2_sci_exp and
schks2_sci_abv as equivalent to their controls for Key Stage 2 standardised score. Note that to be consistent
with Eyles and Machin’s estimates, you should not use school fixed effects and will need to restrict attention
to years up to and including 2009.) Explain briefly why these restrictions are necessary.
(15 marks)
7) Discuss the results produced in Question 6) above, highlighting what we learn from each specification and
any concerns you might have about the findings.
(14 marks)
8) Both Clark (2009) and Eyles and Machin (2015) discuss the potential endogeneity of their estimates to
school enrolment decisions. Explain what they do to try to overcome their concerns. Given your answer,
what concerns, if any, does this give you about the estimates produced in Question 6) above?
(12 marks)
An alternative identification strategy to that used by Eyles and Machin (2015) would be to exploit variation in
the timing of academy conversion for all schools included in academies.dta.
9) Produce a table with estimates of the effect of academy status on exam performance based on this approach,
using four model specifications: the first three should be similar to those in Question 6); the fourth should
add school fixed effects to the last of these three models. Would you feel more or less confident about using
this approach to estimate the impact of academy status on exam performance than using the approach in
Question 6)? Explain your answer.
(15 marks)
10) All schools in England can now become an academy if they wish. On the basis of the results discussed in
these two papers – and those produced in Question 9) – would you be confident that all schools would, on
average, see a boost to their exam performance in the years following conversion? Explain why or why not.
(7 marks)
References
Clark, D. (2009), The performance and competitive effects of school autonomy, Journal of Political Economy,
Vol. 117, pp. 745-783.
Epple, D, R. Romano and R. Zimmer (2016), Charter Schools: a survey of research on their characteristics and
effectiveness, Handbook of the Economics of Education, Vol. 5, pp. 139-208.
Eyles, A. and S. Machin (2015), The introduction of academy schools to England’s education, CEP Discussion
Paper No. 1368, Centre for Economic Performance, London School of Economics.