Diff-in-diff II March 11, 2020 Fill out your reading report PMAP - PowerPoint PPT Presentation

Diff-in-diff II March 11, 2020 Fill out your reading report PMAP 8521: Program Evaluation for Public Service on iCollege! Andrew Young School of Policy Studies Spring 2020

Plan for today Quick talk about COVID-19 DiD review DiD full example

Quick talk about COVID-19

What is all this? New virus in the coronavirus family Officially named SARS-COV-2 Causes respiratory disease named COVID-19

What is all this? Originated in Wuhan, Hubei Province, China Don’t call it “Chinese Coronavirus” or “Kung Flu” or other xenophobic names!

Symptoms Fever and dry cough initially; pneumonia-like respiratory failure later for vulnerable people Up to two weeks can pass between exposure and symptoms Asymptomatic transmission likely possible

Lethality

Lethality

Why is everything shutting down? Flattening the curve

If you’re young and healthy, all these cancellations and precautions are not about you! Social distancing, staying home, washing your hands, etc. protects the vulnerable Huge collective action problem!

What you can do Wash hands for 20 seconds, disinfect phone, don’t touch your face Stay home if you’re sick Practice social distancing Limit non-essential travel Don’t buy masks Stock up on essentials but don’t hoard

flattenthecurve.com

What does this mean for our class? I HAVE NO IDEA YET GSU hasn’t made any official decisions I’m committed to helping you all succeed and keep learning! I’ll continue to stream class via WebEx 2-week late work window is eliminated

Two wrongs make a right

Raising the minimum wage What happens if you raise the minimum wage? Economic theory says there should be fewer jobs New Jersey in 1992 $4.25 → $5.05

Before vs. after Average fast food jobs in NJ Before: 20.44 After: 21.03 ∆ : 0.59 Is this the causal effect?

Treatment vs. control Average fast food jobs in states PA after : 21.17 NJ after : 21.03 ∆ : − 0.14 Is this the causal effect?

Problems Comparing only before/after Impossible to know if growth happened because of treatment or just naturally Comparing only treatment/control Impossible to know if any changes happened because of natural growth

Time Minimum wage Jobs Being in New Jersey

Pre mean Post mean ∆ (post − pre) A B Treatment B − A (not yet treated) (treated) C D Control D − C (never treated) (never treated) ∆ (trtmt − ctrl) A − C B − D (B − A) − (D − C)

Pre mean Post mean ∆ (post − pre) A B Treatment B − A (not yet treated) (treated) C D Control D − C (never treated) (never treated) Growth! ∆ (trtmt − ctrl) A − C B − D (B − A) − (D − C)

Pre mean Post mean ∆ (post − pre) A B Treatment B − A (not yet treated) (treated) C D Control D − C (never treated) (never treated) ∆ (trtmt − ctrl) A − C B − D (B − A) − (D − C) Within-group effects

Pre mean Post mean ∆ (post − pre) A B Treatment B − A (not yet treated) (treated) C D Control D − C (never treated) (never treated) ∆ (trtmt − ctrl) A − C B − D (B − A) − (D − C) Growth of treatment − growth of control (DiD!)

<latexit sha1_base64="(nul)">(nul)</latexit> <latexit sha1_base64="(nul)">(nul)</latexit> <latexit sha1_base64="(nul)">(nul)</latexit> <latexit sha1_base64="(nul)">(nul)</latexit> �� =(¯ x ���������� ���� − ¯ x ���������� ��� ) − (¯ x �������� ���� − ¯ x �������� ��� )

Pre mean Post mean ∆ (post − pre) A B B − A NJ 20.44 21.03 0.59 C D D − C PA 23.33 21.17 − 2.16 (0.59) − A − C B − D ∆ (trtmt − ctrl) ( − 2.16) = − 2.89 − 0.14 2.76

B A D C

C D A B

Finding all the group means is tedious though! What if there are other backdoors to worry about? Regression to the rescue!

Time Minimum wage Jobs Being in New Jersey

<latexit sha1_base64="tPctye95c/TWQ7QaqhaC+TK3+c=">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</latexit> Y it = ↵ + � Group i + � Time t + � (Group i × Time t ) + ✏ it Group = 1/TRUE if treatment Time = 1/TRUE if after

<latexit sha1_base64="tPctye95c/TWQ7QaqhaC+TK3+c=">ACcXicbVHRahNBFJ3dWluj1bT6IqIMDS2VQthtC/pSKPqgjxGatpINy93JTJ0ZneYuVsaln3+/rmT/jiDzibBLSpFwYO59zDnDmTGSUdRdHPIFx7tP54Y/NJ6+mzrecv2ts7F64orcC+KFRhrzJwqGSOfZKk8MpYBJ0pvMyuPzf65Q1aJ4v8nGYGhxomuRxLAeSptP3je1pJqvnpPk9AmSnwQ5kSMATnhDeUvXFqWpU9kIE9D6r3IuNdYpNULS8v4RqrnvYMWYkF90903vGxcaJ5VP0SRI252oG82HPwTxEnTYcnp+y4ZFaLUmJNQ4NwgjgwNK7AkhcK6lZQODYhrmODAwx8hmE1b6zme54Z8XFh/cmJz9l/HRVo52Y685saOpWtYb8nzYoafxWMnclIS5WFw0LhWngjf185G0KEjNPABhpc/KxRQsCPKf1PIlxKtPfgujrxcfo20n7NOyjk32hu2yAxazD+yMfWU91meC/QpeBW+Dd8Hv8HXIw93FahgsPS/ZvQkP/wASQrtI</latexit> Y it = ↵ + � Group i + � Time t + � (Group i × Time t ) + ✏ it α = Mean of control, pre-treatment β = Increase in outcome across groups γ = Increase in outcome across time δ = Difference in differences!

<latexit sha1_base64="5lDjK/m0Q2HKJ3tx76YpE80WMco=">ACbHicbVHLahsxFNVM0zZ1H3GaLlJCQcQ0JBTMTBJoNoXQLNqlC3EeMxwR762RaQZId0JNcOs+ofd9RO6TdUYxuaOL0gOJwHujrKjJKOouhXED5ae/zk6fqz1vMXL19tDdfX7itAL7olCFvcrAoZI59kmSwitjEXSm8DK7OWv0y1u0Thb5Oc0MDjVMcjmWAshTafvHdVpJqvmnPZ6AMlPgH3iSIQFPeEL4naovtihNncpGmIDW/5RzqbFOqRFGqOaR/ZVMQt7j7vsPmgAaJ5VfoLk8bXeibjQf/hDES9Bhy+ml7Z/JqBClxpyEAucGcWRoWIElKRTWraR0aEDcwAQHubgdxhW87Jq/t4zIz4urD858Tl7N1GBdm6mM+/UQFO3qjXk/7RBSeOTYSVzUxLmYnHRuFScCt40z0fSoiA18wCElX5XLqZgQZD/n5YvIV598kNwcdiNj7qH347p5+XdayzHbL9lnMPrJT9pX1WJ8J9jvYCLaDt8Gf8E24E75bWMNgmdli9ybc+wsKGboO</latexit> Y it = ↵ + � Group i + � Time t + � (Group i × Time t ) + ✏ it Pre mean Post mean ∆ (post − pre) Control α α + γ γ Treatment α + β α + β + γ + δ γ + δ ∆ (trtmt − ctrl) β β + δ δ

R time!