Assignment 1

ECON7440 Public Economics, 2026

The University of Queensland

This take-home assignment (worth 15% of the final grade) has three questions, each with multiple

parts. In total, there are 60 marks available. Question 1 is worth 32 marks, question 2 is worth 10

marks and question 3 is worth 18 marks.

Question 1 builds on some of the Stata exercises we have completed to date in the tutorial questions.

There are two accompanying Stata files, “ECON7440 assignment.do” and “ECON7440 assignment.dta”.

You will need to write some additional Stata code to complete the exercises. I have provided a CSV

file if you would prefer to use another software program (e.g., R).

The assignment is due at 5 PM (Brisbane time) on September 14, 2026. Late assignments will

receive a penalty of 10% of the maximum mark per day.

This is an open-book non-invigilated assignment. You may review any course materials to assist

your assignment. However, you must complete the assignment individually without discussing the

questions with anyone else. Any possible violation of academic integrity will be investigated and may

be reported to the university for disciplinary action.

Please submit your answers in a PDF file via Blackboard by the due time. You are strongly

encouraged to allow at least 30 minutes in case of technical issues over submission. It would be more

convenient if you type your answers with a text editing application (such as Microsoft Word, Google

Docs, Apple Pages, etc.) and save the file as a PDF document (rather than handwriting your answers

and scanning them).

Questions

Teen driving

[32 marks]

1. The following empirical questions are based on Moore & Morris’ 2024 paper “Shaping the Habits

of Teen Drivers”, which we have studied in the tutorials and is contained in the Assignment 1

folder on Blackboard. Stata files for this question are included in Blackboard, although you will

have to write additional code for several questions to summarise the data and estimate regressions.

You can use an alternative software program (e.g., R) if you prefer.

The paper examines a restriction on first-year drivers in New South Wales, Australia. These

are drivers who have recently started driving independently for the first time. The restriction

prevents first-year drivers from carrying two or more (2+) passengers between 11:00 pm and

4:59 am. The restriction was announced in January 2007 and introduced in July 2007. Other

road-safety policies affecting young drivers were introduced at the same time. The restriction is

actively enforced by police. The same policy still exists in Queensland and was introduced at the

same time.

1

Moore & Morris (2024) use linked administrative data on driver licences and road crashes from

2004 to 2014. The dataset supplied for this assignment contains quarterly crash rates for first-

year drivers in NSW from 2004 Q3 to 2014 Q3. Crash rates are classified by the time of day and

the number of passengers (0–1 vs 2+) carried by the first-year driver. The day is split into four

periods (daytime: 8:00 am–7:59 pm; evening: 8:00 pm–10:59 pm; late night: 11:00 pm–4:59 am;

and morning: 5:00 am–7:59 am). The restriction only applies in the late-night period for driving

with 2+ passengers.

For the empirical parts of this question, use “ECON7440 assignment.dta” and the accompanying

“ECON7440 assignment.do” do-file. There are 328 observations in the dataset (328 = 41 quarters

× 4 time-of-day periods × 2 types of passengers).

Note on the Stata regressions:

For every regression in this question, calculate heteroskedasticity-robust standard errors by in-

cluding the , robust option. The general form of a Stata regression command is:

reg ${outcome} var1 var2 var3 if sample_condition, robust

For example, a regression restricted to crashes involving 2+ passengers during daytime or late-

night hours would take the form:

reg ${outcome} var1 var2 var3 if two_plus_passengers == 1 & (daytime == 1 | late_night == 1), robust

You will need to replace var1, var2, var3, etc. with the necessary explanatory variables and

update the sample restrictions in some cases.

(a) Explain why late-night driving with multiple passengers by inexperienced teen drivers may

be subject to particularly large:

(i) negative externalities; and

(ii) negative internalities.

Draw on empirical evidence presented or discussed in Moore & Morris (2024) to support your

arguments. [4 marks]

(b) Moore & Morris use a differences-in-differences approach for identifying the impact on late-

night crash rates. What is the key identifying assumption? Explain (i) whether the assump-

tion is based on absolute or relative changes in crash rates and (ii) why this distinction is

important. [3 marks]

(c) The outcome used in the following analysis is the natural logarithm of the crash rate per

100,000 first-year drivers. Using Stata, calculate the average log crash rate before and after

the restriction separately by time of day and number of passengers. Fill in the missing entries

in the table below and place the completed table in your assignment. [1 mark]

2

Table 1: Mean of log crash rates by time of day, number of passengers and before/after restriction

Number of passengers Time of day Pre-restriction Post-restriction Change

2+ Daytime 6.304 5.995 -0.309

2+ Evening

2+ Late night 5.036 3.788 -1.248

2+ Morning

0 or 1 Daytime 8.098 7.861 -0.237

0 or 1 Late night

(d) Focus on crashes involving 2+ passengers — those targeted by the restriction. Using the

values in part (c), calculate the differences-in-differences estimate for late-night crash rates

using daytime hours as the comparison group. Show your working. [2 marks]

(e) Estimate the corresponding differences-in-differences regression. (Hint: the regression should

contain 41 daytime and 41 late-night observations, and you can verify the estimate using your

answer to part (d).)

(i) Report the differences-in-differences coefficient, its robust standard error and p-value.

(ii) Calculate the implied percentage change in late-night crash rates with 2+ passengers.

(iii) State whether you can reject the null hypothesis that the restriction had no effect.

[3 marks]

(f) Your friend argues that crash rates in the evening hours provide a better comparison group

than crash rates in daytime hours because evening driving is more similar to late-night driving:

(i) Restrict the sample to crashes involving 2+ passengers during evening and late-night

hours. Estimate a differences-in-differences regression using evening hours as the com-

parison group. Report the differences-in-differences coefficient, its robust standard error

and p-value. Calculate the percentage change implied by the coefficient and compare it

to your estimate in (e).

(ii) Discuss whether crash rates during evening hours are likely to be a suitable comparison

group for estimating the effect of the restriction on late-night crash rates. What is a

concern of this approach?

[4 marks]

(g) Continue to focus on crashes involving 2+ passengers. Using daytime as the omitted compar-

ison group, estimate a revised differences-in-differences model that allows the post-restriction

change in crash rates to differ across the evening, late night, and morning.

(i) Report the three differences-in-differences coefficients, their robust standard errors and

p-values, and calculate the percentage changes implied by the coefficients. (Hint: the

regression should contain 164 observations, and you can use the means in Table 1 to

hand-check each estimate.)

3

(ii) Interpret what the pattern of estimates suggests about:

• the effect of the restriction during the restricted late-night hours;

• possible spillovers into the evening and morning hours

[6 marks]

(h) Restrict the sample to crashes involving 0 or 1 passenger during late-night or daytime hours.

Estimate the same differences-in-differences specification used in part (e).

(i) Report the differences-in-differences coefficient and its p-value. Is the coefficient statis-

tically significant?

(ii) Explain why this regression can be viewed as a placebo test. Would a statistically sig-

nificant estimate here necessarily invalidate the main research design for 2+ passengers?

[4 marks]

(i) Estimate the following triple-differences model using late-night and daytime observations for

both passenger groups:

ln(Yit ) =α + β1 Latei + β2 T woP lusi + β3 P ostt

+ β4 (Latei × T woP lusi ) + β5 (Latei × P ostt )

+ β6 (T woP lusi × P ostt )

+ β7 (Latei × T woP lusi × P ostt ) + εit

where:

• Latei equals one for late-night observations and zero for daytime observations;

• T woP lusi equals one for crashes involving 2+ passengers and zero for crashes involving

0 or 1 passenger; and

• P ostt equals one after the restriction was introduced and zero before its introduction.

(i) Report the estimate of β7 , its robust standard error and p-value. (Hint: the model

should have 164 observations.) Calculate the percentage change implied by the estimate

and interpret it.

(ii) Verify that the triple-differences estimate is equal, subject to rounding, to the differences-

in-differences estimate in part (e) minus the differences-in-differences estimate in part

(h).

[4 marks]

(j) What additional source of confounding can the triple-differences model above address relative

to the baseline differences-in-differences model in part (e)? [1 mark]

4

Education and child maltreatment

[10 marks]

2. In week 4, Dr Adam Dzulkipli presented a guest lecture on “School and Child Maltreatment”.

The following questions are based on his presentation:

(a) Explain two mechanisms through which requiring children to spend more time in school

could reduce child maltreatment or other harmful outcomes. [2 marks]

(b) Dzulkipli et al. study a South Australian reform that increased the minimum school leaving

age from 16 to 17. Explain how the authors use individuals’ dates of birth to construct a

regression-discontinuity design. State the key identifying assumption required for this design.

[2 marks]

(c) The lecture also examined a South Australian school-entry rule. Children born before 1 May

were eligible to begin school one year earlier than children born on or after 1 May. Eligibility

increased actual school enrolment by 80.4 percentage points and reduced the probability of

a child-protection notification by 3.3 percentage points.

(i) Explain why the effect of eligibility on child-protection notifications is a reduced-form

or intention-to-treat estimate. [1 mark]

(ii) Explain why a fuzzy regression-discontinuity design is required to estimate the effect

of actual school enrolment. [1 mark]

(iii) Use the estimates above to calculate the implied effect of school enrolment on the

probability of a child-protection notification. Interpret the estimate carefully, including

whose treatment effect it identifies. [2 marks]

(d) In the figure below showing the effect of the compulsory-schooling reform on “any emergency-

department visit at age 16”, there is a positive trend in the outcome to the right of the

date-of-birth threshold. What, if anything, does this trend tell us about whether the policy

became less effective for later birth cohorts? Explain your answer, referring to the local

nature of the regression-discontinuity design. [2 marks]

5

Commodity taxation [18 marks]

3. Suppose the market for oranges is subject to the demand curve QD = 18 − 4P and supply curve

QS = 6P − 2. Assume Q is measured in kilos and P is measured in dollars.

(a) Show that the equilibrium quantity is 10 kg and the equilibrium price is $2. [1 mark]

(b) Calculate the elasticity of demand (ϵD ) and the elasticity of supply (ϵS ) at the market

equilibrium. Show that ϵS = −1.5 × ϵD . [2 marks]

(c) Assume the government imposes an excise tax of $0.20 per kg of oranges, and the tax is

levied on consumers. Based on the relative elasticities of demand and supply, show that

producers bear only 40% of the incidence of the tax. What is the change in (i) the total

amount paid by consumers (including the tax) and (ii) the amount received by producers?

[3 marks]

(d) Suppose that a consumer lobby group, Choice magazine, argues that the tax is unfair on

consumers, particularly those on low incomes. They suggest that the government should

impose the tax on producers of oranges instead, since these are large businesses that can

afford to pay the tax. Do you agree with Choice’s argument? Explain why. [2 marks]

(e) Suppose the government requires additional revenue and thus wants to raise the tax on

oranges to $0.40. Suppose also that apples are not taxed at all. What would you advise the

government to do? [2 marks]

(f) Suppose that bananas are subject to a similar excise tax as oranges (i.e., the tax is a similar

percentage of the competitive equilibrium price). Your analysis of the market suggests that

the elasticities of supply and demand are larger for oranges than for bananas. How would

you advise the government to adjust the taxes on oranges and bananas if they are seeking

to raise revenue? What about if the current amount of revenue is sufficient? [3 marks]

(g) Suppose now that consumers have just two choices of fruit: oranges and mandarins. Both

of these fruits are subject to identical demand and supply curves, and have a $0.20 excise

tax per kg on producers. However, orange farmers use harmful pesticides that pollute the

environment, while mandarins are produced organically. Independent experts have quantified

the marginal damage to society per kg of oranges produced at $0.40. Explain how the

government could improve social welfare by adjusting the taxes on oranges and mandarins.

[5 marks]

6