ECU3091 Econometrics A

Differences-in-differences

Dr. Barra Roantree

Trinity College Dublin, 2022
Recap: quasi-(natural) experiments
A quasi-experiment or natural experiment generates “as
if” random assignment, but that is not the result of an explicit
randomized treatment and control design.

Rather, “as if” random assignment occurs because of

vagaries in legal institutions, location, timing of policy or
programme implementation

Or natural randomness such as birth dates, rainfall, or other

factors unrelated to causal effect under study
 Differences-in-differences estimator

Differences-in-differences estimator exploits this variation:

estimates treatment effect as the difference between the pre- &
post-treatment values of Y for the treatment and control groups.

Let:
Yi before = value of Y for subject i before the expt
Yi after = value of Y for subject i after the expt

ˆ1diffs −in −diffs = (Y treat ,after − Y treat ,before ) − (Y control ,after − Y control ,before )
Differences-in-differences estimator
ˆ1diffs −in −diffs = (Y treat ,after − Y treat ,before ) − (Y control ,after − Y control ,before )
 John Snow & the first diff-in-diff
• 10,000s died of cholera between 1830–1850 in London
• At the time, prevailing wisdom was that cholera spread by
inhaling ‘bad air’ (miasma), but in fact cholera was spread
through drinking contaminated water from pumps/wells
• Parts of London served by competing water companies, one
of which ("Lambeth") moved their intake pipes upstream &
above the main sewage discharge points earlier than others
• This provided John Snow – a doctor who theorised cholera
spread through unclean water – with a quasi-experiment
• Snow collected data on households' water provider and
matched this with data on household mortality held by the city
• Compared mortality of Lambeth’s customers to that of others
John Snow & the first diff-in-diff
Deaths per 10,000: based on Table XII from Snow (1855)
Company name 1849 1854
Southwark & Vauxhall 135 147
Lambeth 85 19

Source: https://mixtape.scunning.com/09-difference_in_differences#tbl-snow1

ˆ1diffs −in −diffs = (Y treat ,after − Y treat ,before ) − (Y control ,after − Y control ,before )

• = (19-85) - (147-135)
• = 78 fewer deaths per 10,000 in Lambeth than S&W
 Diff-in-diff and the minimum wage

• Seminal application of diff-in-diff by Card & Krueger (1994) to

question of impact minimum wage has on employment
• An empirical question given competing theoretical predictions
of competitive & monopsonistic models of the labour market
• Surveyed c.400 fast-food restaurants in NJ & neighbouring
PA before & after 1992 NJ min wage rise from $4.25 to $5.05
Diff-in-diff and the minimum wage
• Distribution of wages in firms surveyed suggests
minimum wage increase was binding

Figure 9.1: Distribution of wages for NJ and PA.

Reprinted from Card and Krueger (1994).
Diff-in-diff and the minimum wage
• Distribution of wages in firms surveyed suggests
minimum wage increase was binding

Figure 9.1: Distribution of wages for NJ and PA.

Reprinted from Card and Krueger (1994).
Diff-in-diff and the minimum wage

Source: https://www.nobelprize.org/uploads/2021/10/popular-economicsciencesprize2021-3.pdf
Diff-in-diff and the minimum wage
Dependent Variable PA NJ NJ – PA
FTE before 23.3 20.44 −2.89
(1.35) (0.51) (1.44)
FTE after 21.147 21.03 −0.14
(0.94) (0.52) (1.07)
Change in mean FTE −2.16 0.59 2.76
(1.25) (0.54) (1.36)

Note: standard errors in parenthesis.

Source: https://mixtape.scunning.com/09-difference_in_differences#tbl-minwage-dd

ˆ1diffs −in −diffs = (Y treat ,after − Y treat ,before ) − (Y control ,after − Y control ,before )
= (21.03 – 20.44) – (21.147 – 23.3)
= (0.59) – (–2.16)
= 2.76
Diff-in-diff and the minimum wage

Source: https://github.com/ScPoEcon/ScPoEconometrics-Slides
Diff-in-diff and the minimum wage

Source: https://github.com/ScPoEcon/ScPoEconometrics-Slides
Aside: diff-in-diff regression estimator

Regression formulation for 2-groups and 2-periods:

ΔYi = 0 + 1X i + ui
where
Yi = Yi after − Yi before
X i = 1 if treated, = 0 otherwise

ˆ1 is the diffs-in-diffs estimator

Why use regression formulation? Allows for:

– more than 2-groups and/or 2-periods
– statistical inference with clustered standard errors
– inclusion of control variables to improve precision of estimates
Aside: diff-in-diff regression estimator

More general regression formulation:

𝐾

𝑌𝑖𝑡 = 𝛽0 + 𝛽1 𝑇𝑟𝑒𝑎𝑡𝑖 + 𝛽2 𝑃𝑜𝑠𝑡𝑖 + 𝛿 𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡𝑖 + ෍ 𝛽𝑘 𝑊𝑖,𝑘 + 𝑢𝑖

𝑘=3

where:
▪ 𝑇𝑟𝑒𝑎𝑡𝑖 = 1 if in treatment group, 0 otherwise
▪ 𝑃𝑜𝑠𝑡𝑖 = 1 if post-treatment period, 0 otherwise
▪ 𝑇𝑟𝑒𝑎𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡𝑖 = 1 for treatment group post-treatment, 0 otherwise
▪ 𝑊𝑖,𝑘 = set of control variables to improve precision of estimates

Estimate of treatment effect given by 𝛿

Common/parallel trends
Implicit so far is crucial assumption of common trends
▪ e.g. absent any minimum wage increase, PA trend in fast-food
employment is what we would have expected to see in NJ
▪ i.e. PA trend in fast-food employment between February and
November 1992 provides a reliable counterfactual employment
trend New Jersey's fast-food industry would have experienced had
New Jersey not increased its minimum wage.

Impossible to completely validate assumption

▪ … but we can test and probe in quite intuitive way by comparing
trends before the policy change with additional years of data
Common/parallel trends

Source: https://github.com/ScPoEcon/ScPoEconometrics-Slides
Common/parallel trends

Source: https://github.com/ScPoEcon/ScPoEconometrics-Slides
Further reading
• Chapter 5, Angrist J. D. & Pischke J.-S. (2014). Mastering'
metrics: the path from cause to effect. Princeton University
Press.
• Scientific Background on the Sveriges Riksbank Prize in
Economic Sciences in Memory of Alfred Nobel 2021
• Card, David & Krueger, Alan B (1994). "Minimum Wages
and Employment: A Case Study of the Fast-Food Industry in
New Jersey and Pennsylvania," American Economic
Review, American Economic Association, vol. 84(4), pages
772-793, September.
• Coleman, Thomas, (2019) Causality in the Time of Cholera:
John Snow As a Prototype for Causal Inference.
http://dx.doi.org/10.2139/ssrn.3262234