Difference-in-Differences Methods

Size: px
Start display at page:

Download "Difference-in-Differences Methods"

Transcription

1 Difference-in-Differences Methods Teppei Yamamoto Keio University Introduction to Causal Inference Spring 2016

2 1 Introduction: A Motivating Example 2 Identification 3 Estimation and Inference 4 Diagnostics and Extensions

3 Example: Minimum Wage and Employment Do higher minimum wages decrease employment? Card and Krueger (1994) consider impact of New Jersey s 1992 minimum wage increase from $4.25 to $5.05 per hour Compare employment in 410 fast-food restaurants in New Jersey and eastern Pennsylvania before and after the rise Survey data on wages and employment from two waves: Wave 1: March 1992, one month before the minimum wage increase Wave 2: December 1992, eight month after increase Teppei Yamamoto Difference in Differences Causal Inference 1 / 30

4 Location of Restaurants Teppei Yamamoto Difference in Differences Causal Inference 2 / 30

5 Wages Before Rise in Minimum Wage Teppei Yamamoto Difference in Differences Causal Inference 3 / 30

6 Wages After Rise in Minimum Wage Teppei Yamamoto Difference in Differences Causal Inference 4 / 30

7 1 Introduction: A Motivating Example 2 Identification 3 Estimation and Inference 4 Diagnostics and Extensions

8 Setup: Groups, Periods and Treatments Data structure: Two waves of randomly sampled cross-sectional observations Either panel or repeated cross sections Cross-sectional units: i {1,..., N} Time periods: t {0 (pre-treatment), 1 (post-treatment)} { 1 (treatment group) Group indicator: G i = 0 (control group) Treatment indicator: Z it {0, 1} Units in the treatment group receive treatment in t = 1: Time Period Group t = 0 t = 1 G i = 1 (treatment group) Z i0 = 0 (untreated) Z i1 = 1 (treated) G i = 0 (control group) Z i0 = 0 (untreated) Z i0 = 0 (untreated) Teppei Yamamoto Difference in Differences Causal Inference 5 / 30

9 Setup: Potential Outcomes Potential outcomes Y it (z): Y it (0): potential outcome for unit i in period t when not treated Y it (1): potential outcome for unit i in period t when treated Teppei Yamamoto Difference in Differences Causal Inference 6 / 30

10 Setup: Potential Outcomes Potential outcomes Y it (z): Y it (0): potential outcome for unit i in period t when not treated Y it (1): potential outcome for unit i in period t when treated Causal effect for unit i at time t is τ it = Y it (1) Y it (0) Teppei Yamamoto Difference in Differences Causal Inference 6 / 30

11 Setup: Potential Outcomes Potential outcomes Y it (z): Y it (0): potential outcome for unit i in period t when not treated Y it (1): potential outcome for unit i in period t when treated Causal effect for unit i at time t is τ it = Y it (1) Y it (0) Observed outcomes Y it are realized as Y it = Y it (0)(1 Z it ) + Y it (1)Z it Because Z i1 = G i in the post-treatment period, we can also write Y i1 = Y i1 (0)(1 G i ) + Y i1 (1)G i Teppei Yamamoto Difference in Differences Causal Inference 6 / 30

12 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Problem: Missing potential outcome: Teppei Yamamoto Difference in Differences Causal Inference 7 / 30

13 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Problem: Missing potential outcome: E[Y i1 (0) G i = 1], i.e. what is the average post-period outcome for the treated group in the absence of the treatment? Teppei Yamamoto Difference in Differences Causal Inference 7 / 30

14 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Before vs. After Use E[Y i1 G i = 1] E[Y i0 G i = 1] for τ ATT Assumes Teppei Yamamoto Difference in Differences Causal Inference 8 / 30

15 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Before vs. After Use E[Y i1 G i = 1] E[Y i0 G i = 1] for τ ATT Assumes E[Y i1 (0) G i = 1] = E[Y i0 (0) G i = 1] (No change in average potential outcome over time) Teppei Yamamoto Difference in Differences Causal Inference 8 / 30

16 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Treated vs. Control in Post-Period Use E[Y i1 G i = 1] E[Y i1 G i = 0] for τ ATT Assumes Teppei Yamamoto Difference in Differences Causal Inference 9 / 30

17 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Treated vs. Control in Post-Period Use E[Y i1 G i = 1] E[Y i1 G i = 0] for τ ATT Assumes E[Y i1 (0) G i = 1] = E[Y i1 (0) G i = 0] (Mean ignorability of treatment assignment) Teppei Yamamoto Difference in Differences Causal Inference 9 / 30

18 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Difference-in-Differences (DD) { } { } Use: E[Y i1 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 1] E[Y i0 G i = 0] Teppei Yamamoto Difference in Differences Causal Inference 10 / 30

19 Identification Strategies Estimand: ATT in the post-treatment period τ ATT = E[Y i1 (1) Y i1 (0) G i = 1] = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] Pre-Period (t = 0) Post-Period (t = 1) Treatment Group (G i = 1) E[Y i0 (0) G i = 1] E[Y i1 (1) G i = 1] Control Group (G i = 0) E[Y i0 (0) G i = 0] E[Y i1 (0) G i = 0] Control Strategy: Difference-in-Differences (DD) { } { } Use: E[Y i1 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 1] E[Y i0 G i = 0] Assumes: E[Y i1 (0) Y i0 (0) G i = 1] = E[Y i1 (0) Y i0 (0) G i = 0] (Parallel trends) Teppei Yamamoto Difference in Differences Causal Inference 10 / 30

20 Graphical Representation: Difference-in-Differences E[Y i1 G i = 1] E[Y i0 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 0] t = 0 t = 1 Teppei Yamamoto Difference in Differences Causal Inference 11 / 30

21 Graphical Representation: Difference-in-Differences E[Y i1 G i = 1] E[Y i0 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 0] t = 0 t = 1 Teppei Yamamoto Difference in Differences Causal Inference 11 / 30

22 Graphical Representation: Difference-in-Differences E[Y i1 G i = 1] E[Y i1 (0) G i = 1] E[Y i0 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 0] t = 0 t = 1 Teppei Yamamoto Difference in Differences Causal Inference 11 / 30

23 Graphical Representation: Difference-in-Differences E[Y i1 G i = 1] E[Y i1 (0) G i = 1] E[Y i0 G i = 1] E[Y i1 G i = 0] E[Y i0 G i = 0] τ ATT t = 0 t = 1 Teppei Yamamoto Difference in Differences Causal Inference 11 / 30

24 Identification with Difference-in-Differences Under the parallel trends assumption: E[Y i1 (0) Y i0 (0) G i = 1] = E[Y i1 (0) Y i0 (0) G i = 0] The ATT can be nonparametrically identified as: { } τ ATT = E[Y i1 G i = 1] E[Y i1 G i = 0] { } E[Y i0 G i = 1] E[Y i0 G i = 0] Teppei Yamamoto Difference in Differences Causal Inference 12 / 30

25 Identification with Difference-in-Differences Under the parallel trends assumption: E[Y i1 (0) Y i0 (0) G i = 1] = E[Y i1 (0) Y i0 (0) G i = 0] The ATT can be nonparametrically identified as: { } τ ATT = E[Y i1 G i = 1] E[Y i1 G i = 0] { } E[Y i0 G i = 1] E[Y i0 G i = 0] Proof: {E[Y i1 G i = 1] E[Y i1 G i = 0]} {E[Y i0 G i = 1] E[Y i0 G i = 0]} = {E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 0]} {E[Y i0 (0) G i = 1] E[Y i0 (0) G i = 0]} = E[Y i1 (1) G i = 1] E[Y i1 (0) G i = 1] +E[Y i1(0) G i = 1] }{{} = τ ATT E[Y i1 (0) G i = 0] E[Y i0 (0) G i = 1] + E[Y i0 (0) G i = 0] = τ ATT + {E[Y i1 (0) Y i0 (0) G i = 1] E[Y i1 (0) Y i0 (0) G i = 0]} }{{} = 0 under parallel trends = τ ATT Teppei Yamamoto Difference in Differences Causal Inference 12 / 30

26 Notes on the Parallel Trends Assumption What type of confounding does DD make us robust to? Teppei Yamamoto Difference in Differences Causal Inference 13 / 30

27 Notes on the Parallel Trends Assumption What type of confounding does DD make us robust to? Parallel trends is satisfied if unobserved confounding is time-invariant and additive Parallel trends is violated if there is unobserved time-varying confounding Teppei Yamamoto Difference in Differences Causal Inference 13 / 30

28 Notes on the Parallel Trends Assumption What type of confounding does DD make us robust to? Parallel trends is satisfied if unobserved confounding is time-invariant and additive Parallel trends is violated if there is unobserved time-varying confounding Parallel trends may be more plausible with pre-treatment covariates: E[Y i1 (0) Y i0 (0) G i = 1, X i = x] = E[Y i1 (0) Y i0 (0) G i = 0, X i = x] This assumes parallel trends within strata Under the conditional parallel trends assumption, the ATT is identified as τ ATT = [ x {E[Y i1 G i = 1, X i = x] E[Y i1 G i = 0, X i = x]} ] {E[Y i0 G i = 1, X i = x] E[Y i0 G i = 0, X i = x]} Pr(X i = x G i = 1) Teppei Yamamoto Difference in Differences Causal Inference 13 / 30

29 Notes on the Parallel Trends Assumption What type of confounding does DD make us robust to? Parallel trends is satisfied if unobserved confounding is time-invariant and additive Parallel trends is violated if there is unobserved time-varying confounding Parallel trends may be more plausible with pre-treatment covariates: E[Y i1 (0) Y i0 (0) G i = 1, X i = x] = E[Y i1 (0) Y i0 (0) G i = 0, X i = x] This assumes parallel trends within strata Under the conditional parallel trends assumption, the ATT is identified as τ ATT = [ x {E[Y i1 G i = 1, X i = x] E[Y i1 G i = 0, X i = x]} ] {E[Y i0 G i = 1, X i = x] E[Y i0 G i = 0, X i = x]} Pr(X i = x G i = 1) Note the parallel trends assumption is not invariant to nonlinear transformation of the outcome scale For example, parallel trends in Y it (z) implies non-parallel trends in log Y it (z) and vice versa Teppei Yamamoto Difference in Differences Causal Inference 13 / 30

30 1 Introduction: A Motivating Example 2 Identification 3 Estimation and Inference 4 Diagnostics and Extensions

31 Plug-in Estimation for Panel Data Estimand: { } τ ATT = E[Y i1 G i = 1] E[Y i1 G i = 0] { } E[Y i0 G i = 1] E[Y i0 G i = 0] Teppei Yamamoto Difference in Differences Causal Inference 14 / 30

32 Plug-in Estimation for Panel Data Estimand: { } τ ATT = E[Y i1 G i = 1] E[Y i1 G i = 0] { } E[Y i0 G i = 1] E[Y i0 G i = 0] A plug-in estimator ( difference in difference-in-means ): { 1 N 1 i=1 N G i Y i1 1 N 0 = { 1 N 1 i=1 N i=1 } { 1 (1 G i )Y i1 N G i {Y i1 Y i0 } 1 N 0 N 1 N i=1 N G i Y i0 1 N 0 i=1 N i=1 (1 G i ){Y i1 Y i0 } (1 G i )Y i0 } where N 1 and N 0 are the number of treated and control units respectively }, Teppei Yamamoto Difference in Differences Causal Inference 14 / 30

33 Plug-in Estimation for Panel Data Estimand: { } τ ATT = E[Y i1 G i = 1] E[Y i1 G i = 0] { } E[Y i0 G i = 1] E[Y i0 G i = 0] A plug-in estimator ( difference in difference-in-means ): { 1 N 1 N G i Y i1 1 N 0 i=1 = { 1 N 1 N i=1 } { 1 (1 G i )Y i1 N G i {Y i1 Y i0 } 1 N 0 i=1 N 1 N i=1 N G i Y i0 1 N 0 i=1 N i=1 (1 G i ){Y i1 Y i0 } (1 G i )Y i0 } where N 1 and N 0 are the number of treated and control units respectively Standard errors can be estimated by extending the diff-in-means variance formula using the same logic, assuming no clustering }, Teppei Yamamoto Difference in Differences Causal Inference 14 / 30

34 THE AMERICAN ECONO Example: Card and Krueger TABLE 3-AVERAGE EMPLOYMENT PER STO IN NEW JERSEY MINIM Variable PA (i) Stores by state NJ (ii) Difference, NJ-PA (iii) Wag $4. (iv 1. FTE employment before, all available observations 2. FTE employment after, all available observations 3. Change in mean FTE employment 4. Change in mean FTE employment, balanced sample of storesc Teppei Yamamoto Difference in Differences Causal Inference 15 / 30

35 Plugin-Estimation for Repeated Cross Sections Repeated cross-sectional data require slight change in notation: Period indicator is now a variable: T i {0, 1} Estimand: τ ATT = E[Y i (1) Y i (0) G i = 1, T i = 1] Identified as: τ ATT = E[Y i G i = 1, T i = 1] E[Y i G i = 0, T i = 1] {E[Y i G i = 1, T i = 0] E[Y i G i = 0, T i = 0]} N now refers to the size of the pooled sample Teppei Yamamoto Difference in Differences Causal Inference 16 / 30

36 Plugin-Estimation for Repeated Cross Sections Repeated cross-sectional data require slight change in notation: Period indicator is now a variable: T i {0, 1} Estimand: τ ATT = E[Y i (1) Y i (0) G i = 1, T i = 1] Identified as: τ ATT = E[Y i G i = 1, T i = 1] E[Y i G i = 0, T i = 1] {E[Y i G i = 1, T i = 0] E[Y i G i = 0, T i = 0]} N now refers to the size of the pooled sample The plug-in estimator is then written as: ˆτ ATT = { N i=1 G } it i Y N i i=1 N i=1 G (1 G i)t i Y i N it i i=1 (1 G i)t i { N i=1 G } i(1 T i )Y N i i=1 N i=1 G (1 G i)(1 T i )Y i i(1 T i ) N i=1 (1 G i)(1 T i ) Covariates X i can be incorporated via subclassification Teppei Yamamoto Difference in Differences Causal Inference 16 / 30

37 Regression Estimator for Repeated Cross Sections Because G i and T i are both binary, the same estimator can be calculated via regression: Ŷ i = ˆµ + ˆγG i + ˆδT i + ˆτG i T i where ˆµ, ˆγ, ˆδ and ˆτ are OLS regression estimates Teppei Yamamoto Difference in Differences Causal Inference 17 / 30

38 Regression Estimator for Repeated Cross Sections Because G i and T i are both binary, the same estimator can be calculated via regression: Ŷ i = ˆµ + ˆγG i + ˆδT i + ˆτG i T i where ˆµ, ˆγ, ˆδ and ˆτ are OLS regression estimates Easy to show that ˆτ = ˆτ ATT : After (T i = 1) Before (T i = 0) After - Before Treated G i = 1 ˆµ + ˆγ + ˆδ + ˆτ ˆµ + ˆγ ˆδ + ˆτ Control G i = 0 ˆµ + ˆδ ˆµ ˆδ Treated - Control ˆγ + ˆτ ˆγ ˆτ Teppei Yamamoto Difference in Differences Causal Inference 17 / 30

39 Regression Estimator for Repeated Cross Sections Because G i and T i are both binary, the same estimator can be calculated via regression: Ŷ i = ˆµ + ˆγG i + ˆδT i + ˆτG i T i where ˆµ, ˆγ, ˆδ and ˆτ are OLS regression estimates Easy to show that ˆτ = ˆτ ATT : After (T i = 1) Before (T i = 0) After - Before Treated G i = 1 ˆµ + ˆγ + ˆδ + ˆτ ˆµ + ˆγ ˆδ + ˆτ Control G i = 0 ˆµ + ˆδ ˆµ ˆδ Treated - Control ˆγ + ˆτ ˆγ ˆτ Covariates (X i ) can be added to the right-hand side, with the risk of possible misspecification bias Don t include X i that can be affected by the treatment! (post-ttt bias) Cluster standard errors at the level the treatment is assigned (Card & Krueger get this wrong) Teppei Yamamoto Difference in Differences Causal Inference 17 / 30

40 Regression Estimator for Panel Data For panel data, consider an additive linear model for potential outcomes: Y it (z) = α i + γt + τz + ε it where α i is a time-invariant unobserved effect for unit i that may be correlated with treatment. Teppei Yamamoto Difference in Differences Causal Inference 18 / 30

41 Regression Estimator for Panel Data For panel data, consider an additive linear model for potential outcomes: Y it (z) = α i + γt + τz + ε it where α i is a time-invariant unobserved effect for unit i that may be correlated with treatment. We can show: τ = τ ATE = τ ATT Parallel trends imply: E[Y i1 (0) Y i0 (0) G i = 1] = E[Y i1 (0) Y i0 (0) G i = 0] E[ε i1 ε i0 G i = d] = 0 for d {0, 1} Teppei Yamamoto Difference in Differences Causal Inference 18 / 30

42 Regression Estimator for Panel Data For panel data, consider an additive linear model for potential outcomes: Y it (z) = α i + γt + τz + ε it where α i is a time-invariant unobserved effect for unit i that may be correlated with treatment. We can show: τ = τ ATE = τ ATT Parallel trends imply: E[Y i1 (0) Y i0 (0) G i = 1] = E[Y i1 (0) Y i0 (0) G i = 0] E[ε i1 ε i0 G i = d] = 0 for d {0, 1} Therefore, the first-differenced regression of Y i1 Y i0 on G i can unbiasedly estimate τ ATT = τ ATE Notice that panel data allow for unit-level unobserved confounding unlike the repeated cross-section case, but it must be additive and time-invariant Can include time-varying covariates (X it ) with possible risk of post-ttt bias Teppei Yamamoto Difference in Differences Causal Inference 18 / 30

43 Example: Minimum Wage and Employment ck_data <- plm.data(ck_data, indexes = c("id", "postperiod")) firstdiff.mod <- (plm(emptot ~ postperiod * nj, data = ck_data, model = "fd")) summary(firstdiff.mod) Oneway (individual) effect First-Difference Model Call: plm(formula = emptot ~ postperiod * nj, data = ck_data, model = "fd") Unbalanced Panel: n=410, T=1-2, N=794 Coefficients : Estimate Std. Error t-value Pr(> t ) (intercept) * postperiod1:nj * Teppei Yamamoto Difference in Differences Causal Inference 19 / 30

44 1 Introduction: A Motivating Example 2 Identification 3 Estimation and Inference 4 Diagnostics and Extensions

45 Parallel Trends Violations: Selection The parallel trends assumption can be violated for various reasons 1 Selection and targeting: Treatment assignment may depend on time-varying factors Examples: Self-selection: participants in worker training programs experience a decrease in earnings before they enter the program Targeting: policies may be targeted at units that are currently performing best (or worst). Teppei Yamamoto Difference in Differences Causal Inference 20 / 30

46 More Parallel Trends Violations 2 Compositional Differences Across Time In repeated cross-sections, the composition of the sample may change between periods, i.e. due to migration. This may confound any DD estimate since effect may be attributable to change in population. 3 Long-term effects versus reliability Parallel trends assumption most likely to hold over shorter time-period In the long run, many things may happen that could confound effect of treatment. Teppei Yamamoto Difference in Differences Causal Inference 21 / 30

47 Functional form dependence 4 Functional form dependence: Magnitude or even sign of the DD effect may be sensitive to the functional form, when average outcomes for controls and treated are very different at baseline Training program for the young Employment for the young increases from 20% to 30% Employment for the old increases from 5% to 10% Positive DD effect: (30-20) - (10-5) = 5% increase Teppei Yamamoto Difference in Differences Causal Inference 22 / 30

48 Functional form dependence 4 Functional form dependence: Magnitude or even sign of the DD effect may be sensitive to the functional form, when average outcomes for controls and treated are very different at baseline Training program for the young Employment for the young increases from 20% to 30% Employment for the old increases from 5% to 10% Positive DD effect: (30-20) - (10-5) = 5% increase But if you consider log changes in employment, the DD is [log(30) log(20)] [log(10) log(5)] = log(1.5) log(2) < 0 DD estimates may be more reliable if treated and control are more similar at baseline Teppei Yamamoto Difference in Differences Causal Inference 22 / 30

49 Diagnostics for Parallel Trends 1 Pre-treatment trends in the outcome: Check if the trends are parallel in the pre-treatment periods Requires data on multiple pre-treatment periods (the more the better) Note this is only diagnostics, not a direct test of the assumption! 2 Placebo test using previous periods: Suppose DD with time periods t 1, t 2, t 3, where treatment occurs in t 3 Exclude data from t 3, assign t 2 as placebo" treatment period, and re-estimate DD 3 Placebo test using alternative groups: Re-code some control groups as treated Re-estimate DD with the placebo treated units & without actual treated units 4 Placebo outcomes: Find outcomes that, theoretically, should be unaffected by the treatment, but might Re-estimate DD on these outcomes Teppei Yamamoto Difference in Differences Causal Inference 23 / 30

50 Checking for Pre-Treatment Trends FTE Employment New Jersey Pennsylvania Time Teppei Yamamoto Difference in Differences Causal Inference 24 / 30

51 Checking for Pre-Treatment Trends FTE Employment New Jersey Pennsylvania Time Teppei Yamamoto Difference in Differences Causal Inference 24 / 30

52 Checking for Pre-Treatment Trends FTE Employment New Jersey Pennsylvania Time Teppei Yamamoto Difference in Differences Causal Inference 24 / 30

53 Checking for Pre-Treatment Trends FTE Employment New Jersey Pennsylvania Time Teppei Yamamoto Difference in Differences Causal Inference 25 / 30

54 Chiefs and Voting (de Kadt and Larreguy 2015) ANC Vote Share KwaZulu Bantustans excl. KZ Rest of South Africa Ethnic shift in ANC elite Election Year Teppei Yamamoto Difference in Differences Causal Inference 26 / 30

55 Extension: Triple Differences Triple differences (or difference in differences in differences ): Use a placebo DD to make parallel trends more plausible Teppei Yamamoto Difference in Differences Causal Inference 27 / 30

56 Extension: Triple Differences Triple differences (or difference in differences in differences ): Use a placebo DD to make parallel trends more plausible Example: In state A, all 9th-grade girls are given a free bicycle. 1st DD: G i {girl, boy} and T i {9, 10} in state A 2nd DD: G i {girl, boy} and T i {9, 10} in state B (placebo) The DDD estimator: ˆτ DDD = Ê[Y i girl, 10, A] Ê[Y i girl, 9, A] } {Ê[Yi boy, 10, A] Ê[Y i boy, 9, A] (Ê[Yi girl, 10, B] Ê[Y i girl, 9, B] } ) {Ê[Yi boy, 10, B] Ê[Y i boy, 9, B] Can use regression with a triple interaction May eliminate time-varying confounding that are common in states A and B (e.g. girls change more from grade 9 to 10 than boys) Teppei Yamamoto Difference in Differences Causal Inference 27 / 30

57 Muralidharan and Prakash (2013) Teppei Yamamoto Difference in Differences Causal Inference 28 / 30

58 Muralidharan and Prakash (2013) Teppei Yamamoto Difference in Differences Causal Inference 29 / 30

59 Summary and Remarks DD: An extremely popular strategy when there is longitudinal data (panel or repeated cross-sections) and the treatment is one-shot Parallel trends = a form of ignorability assumption, i.e., unobserved confounding must be additive and time-invariant An important generalization of DD is fixed effects regression with both time and unit fixed effects: y it = x it b + α i + η t + ε it where x it includes the treatment variable Z it Other extensions include: Nonlinear and semi-/nonparametric DD Difference in discontinuities: Combine RD with DD Matching before DD: Make estimates more robust to functional form misspecification Teppei Yamamoto Difference in Differences Causal Inference 30 / 30

150C Causal Inference

150C Causal Inference 150C Causal Inference Difference in Differences Jonathan Mummolo Stanford Selection on Unobservables Problem Often there are reasons to believe that treated and untreated units differ in unobservable characteristics

More information

PSC 504: Differences-in-differeces estimators

PSC 504: Differences-in-differeces estimators PSC 504: Differences-in-differeces estimators Matthew Blackwell 3/22/2013 Basic differences-in-differences model Setup e basic idea behind a differences-in-differences model (shorthand: diff-in-diff, DID,

More information

Statistical Models for Causal Analysis

Statistical Models for Causal Analysis Statistical Models for Causal Analysis Teppei Yamamoto Keio University Introduction to Causal Inference Spring 2016 Three Modes of Statistical Inference 1. Descriptive Inference: summarizing and exploring

More information

Gov 2002: 9. Differences in Differences

Gov 2002: 9. Differences in Differences Gov 2002: 9. Differences in Differences Matthew Blackwell October 30, 2015 1 / 40 1. Basic differences-in-differences model 2. Conditional DID 3. Standard error issues 4. Other DID approaches 2 / 40 Where

More information

Lecture 9. Matthew Osborne

Lecture 9. Matthew Osborne Lecture 9 Matthew Osborne 22 September 2006 Potential Outcome Model Try to replicate experimental data. Social Experiment: controlled experiment. Caveat: usually very expensive. Natural Experiment: observe

More information

What s New in Econometrics. Lecture 1

What s New in Econometrics. Lecture 1 What s New in Econometrics Lecture 1 Estimation of Average Treatment Effects Under Unconfoundedness Guido Imbens NBER Summer Institute, 2007 Outline 1. Introduction 2. Potential Outcomes 3. Estimands and

More information

Causal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies

Causal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies Causal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies Kosuke Imai Department of Politics Princeton University November 13, 2013 So far, we have essentially assumed

More information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? Kosuke Imai Department of Politics Center for Statistics and Machine Learning Princeton University

More information

Causality and Experiments

Causality and Experiments Causality and Experiments Michael R. Roberts Department of Finance The Wharton School University of Pennsylvania April 13, 2009 Michael R. Roberts Causality and Experiments 1/15 Motivation Introduction

More information

Statistical Analysis of Causal Mechanisms

Statistical Analysis of Causal Mechanisms Statistical Analysis of Causal Mechanisms Kosuke Imai Princeton University April 13, 2009 Kosuke Imai (Princeton) Causal Mechanisms April 13, 2009 1 / 26 Papers and Software Collaborators: Luke Keele,

More information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? Kosuke Imai Princeton University Asian Political Methodology Conference University of Sydney Joint

More information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data? Kosuke Imai Department of Politics Center for Statistics and Machine Learning Princeton University Joint

More information

Econometrics of causal inference. Throughout, we consider the simplest case of a linear outcome equation, and homogeneous

Econometrics of causal inference. Throughout, we consider the simplest case of a linear outcome equation, and homogeneous Econometrics of causal inference Throughout, we consider the simplest case of a linear outcome equation, and homogeneous effects: y = βx + ɛ (1) where y is some outcome, x is an explanatory variable, and

More information

STOCKHOLM UNIVERSITY Department of Economics Course name: Empirical Methods Course code: EC40 Examiner: Lena Nekby Number of credits: 7,5 credits Date of exam: Friday, June 5, 009 Examination time: 3 hours

More information

Introduction to Causal Inference. Solutions to Quiz 4

Introduction to Causal Inference. Solutions to Quiz 4 Introduction to Causal Inference Solutions to Quiz 4 Teppei Yamamoto Tuesday, July 9 206 Instructions: Write your name in the space provided below before you begin. You have 20 minutes to complete the

More information

Propensity Score Methods for Causal Inference

Propensity Score Methods for Causal Inference John Pura BIOS790 October 2, 2015 Causal inference Philosophical problem, statistical solution Important in various disciplines (e.g. Koch s postulates, Bradford Hill criteria, Granger causality) Good

More information

Difference-in-Differences Estimation

Difference-in-Differences Estimation Difference-in-Differences Estimation Jeff Wooldridge Michigan State University Programme Evaluation for Policy Analysis Institute for Fiscal Studies June 2012 1. The Basic Methodology 2. How Should We

More information

Instrumental Variables

Instrumental Variables Instrumental Variables Teppei Yamamoto Keio University Introduction to Causal Inference Spring 2016 Noncompliance in Randomized Experiments Often we cannot force subjects to take specific treatments Units

More information

Empirical approaches in public economics

Empirical approaches in public economics Empirical approaches in public economics ECON4624 Empirical Public Economics Fall 2016 Gaute Torsvik Outline for today The canonical problem Basic concepts of causal inference Randomized experiments Non-experimental

More information

Statistical Analysis of Causal Mechanisms

Statistical Analysis of Causal Mechanisms Statistical Analysis of Causal Mechanisms Kosuke Imai Luke Keele Dustin Tingley Teppei Yamamoto Princeton University Ohio State University July 25, 2009 Summer Political Methodology Conference Imai, Keele,

More information

Lecture 4: Linear panel models

Lecture 4: Linear panel models Lecture 4: Linear panel models Luc Behaghel PSE February 2009 Luc Behaghel (PSE) Lecture 4 February 2009 1 / 47 Introduction Panel = repeated observations of the same individuals (e.g., rms, workers, countries)

More information

Randomized Experiments

Randomized Experiments Randomized Experiments Teppei Yamamoto Keio University Introduction to Causal Inference Spring 2016 1 Introduction 2 Identification 3 Basic Inference 4 Covariate Adjustment 5 Threats to Validity 6 Advanced

More information

Quantitative Economics for the Evaluation of the European Policy

Quantitative Economics for the Evaluation of the European Policy Quantitative Economics for the Evaluation of the European Policy Dipartimento di Economia e Management Irene Brunetti Davide Fiaschi Angela Parenti 1 25th of September, 2017 1 ireneb@ec.unipi.it, davide.fiaschi@unipi.it,

More information

Front-Door Adjustment

Front-Door Adjustment Front-Door Adjustment Ethan Fosse Princeton University Fall 2016 Ethan Fosse Princeton University Front-Door Adjustment Fall 2016 1 / 38 1 Preliminaries 2 Examples of Mechanisms in Sociology 3 Bias Formulas

More information

Problem set - Selection and Diff-in-Diff

Problem set - Selection and Diff-in-Diff Problem set - Selection and Diff-in-Diff 1. You want to model the wage equation for women You consider estimating the model: ln wage = α + β 1 educ + β 2 exper + β 3 exper 2 + ɛ (1) Read the data into

More information

Matching. Quiz 2. Matching. Quiz 2. Exact Matching. Estimand 2/25/14

Matching. Quiz 2. Matching. Quiz 2. Exact Matching. Estimand 2/25/14 STA 320 Design and Analysis of Causal Studies Dr. Kari Lock Morgan and Dr. Fan Li Department of Statistical Science Duke University Frequency 0 2 4 6 8 Quiz 2 Histogram of Quiz2 10 12 14 16 18 20 Quiz2

More information

Difference-in-Difference estimator. Presented at Summer School 2015 by Ziyodullo Parpiev, PhD June 9, 2015 Tashkent

Difference-in-Difference estimator. Presented at Summer School 2015 by Ziyodullo Parpiev, PhD June 9, 2015 Tashkent Difference-in-Difference estimator Presented at Summer School 2015 by Ziyodullo Parpiev, PhD June 9, 2015 Tashkent Today s Class Non-experimental Methods: Difference-in-differences Understanding how it

More information

Quantitative Methods in Economics Causality and treatment effects

Quantitative Methods in Economics Causality and treatment effects Quantitative Methods in Economics Causality and treatment effects Maximilian Kasy Harvard University, fall 2016 1 / 23 5) Difference-in-Differences (cf. Mostly Harmless Econometrics, chapter 5) 2 / 23

More information

Causal Mechanisms Short Course Part II:

Causal Mechanisms Short Course Part II: Causal Mechanisms Short Course Part II: Analyzing Mechanisms with Experimental and Observational Data Teppei Yamamoto Massachusetts Institute of Technology March 24, 2012 Frontiers in the Analysis of Causal

More information

ECON Introductory Econometrics. Lecture 17: Experiments

ECON Introductory Econometrics. Lecture 17: Experiments ECON4150 - Introductory Econometrics Lecture 17: Experiments Monique de Haan (moniqued@econ.uio.no) Stock and Watson Chapter 13 Lecture outline 2 Why study experiments? The potential outcome framework.

More information

Balancing Covariates via Propensity Score Weighting

Balancing Covariates via Propensity Score Weighting Balancing Covariates via Propensity Score Weighting Kari Lock Morgan Department of Statistics Penn State University klm47@psu.edu Stochastic Modeling and Computational Statistics Seminar October 17, 2014

More information

Job Training Partnership Act (JTPA)

Job Training Partnership Act (JTPA) Causal inference Part I.b: randomized experiments, matching and regression (this lecture starts with other slides on randomized experiments) Frank Venmans Example of a randomized experiment: Job Training

More information

Gov 2002: 5. Matching

Gov 2002: 5. Matching Gov 2002: 5. Matching Matthew Blackwell October 1, 2015 Where are we? Where are we going? Discussed randomized experiments, started talking about observational data. Last week: no unmeasured confounders

More information

Lecture 10 Regression Discontinuity (and Kink) Design

Lecture 10 Regression Discontinuity (and Kink) Design Lecture 10 Regression Discontinuity (and Kink) Design Economics 2123 George Washington University Instructor: Prof. Ben Williams Introduction Estimation in RDD Identification RDD implementation RDD example

More information

Controlling for Time Invariant Heterogeneity

Controlling for Time Invariant Heterogeneity Controlling for Time Invariant Heterogeneity Yona Rubinstein July 2016 Yona Rubinstein (LSE) Controlling for Time Invariant Heterogeneity 07/16 1 / 19 Observables and Unobservables Confounding Factors

More information

Potential Outcomes Model (POM)

Potential Outcomes Model (POM) Potential Outcomes Model (POM) Relationship Between Counterfactual States Causality Empirical Strategies in Labor Economics, Angrist Krueger (1999): The most challenging empirical questions in economics

More information

Combining Difference-in-difference and Matching for Panel Data Analysis

Combining Difference-in-difference and Matching for Panel Data Analysis Combining Difference-in-difference and Matching for Panel Data Analysis Weihua An Departments of Sociology and Statistics Indiana University July 28, 2016 1 / 15 Research Interests Network Analysis Social

More information

Structural Nested Mean Models for Assessing Time-Varying Effect Moderation. Daniel Almirall

Structural Nested Mean Models for Assessing Time-Varying Effect Moderation. Daniel Almirall 1 Structural Nested Mean Models for Assessing Time-Varying Effect Moderation Daniel Almirall Center for Health Services Research, Durham VAMC & Duke University Medical, Dept. of Biostatistics Joint work

More information

Addressing Analysis Issues REGRESSION-DISCONTINUITY (RD) DESIGN

Addressing Analysis Issues REGRESSION-DISCONTINUITY (RD) DESIGN Addressing Analysis Issues REGRESSION-DISCONTINUITY (RD) DESIGN Overview Assumptions of RD Causal estimand of interest Discuss common analysis issues In the afternoon, you will have the opportunity to

More information

Causal Inference with General Treatment Regimes: Generalizing the Propensity Score

Causal Inference with General Treatment Regimes: Generalizing the Propensity Score Causal Inference with General Treatment Regimes: Generalizing the Propensity Score David van Dyk Department of Statistics, University of California, Irvine vandyk@stat.harvard.edu Joint work with Kosuke

More information

Unpacking the Black-Box: Learning about Causal Mechanisms from Experimental and Observational Studies

Unpacking the Black-Box: Learning about Causal Mechanisms from Experimental and Observational Studies Unpacking the Black-Box: Learning about Causal Mechanisms from Experimental and Observational Studies Kosuke Imai Princeton University Joint work with Keele (Ohio State), Tingley (Harvard), Yamamoto (Princeton)

More information

New Developments in Econometrics Lecture 11: Difference-in-Differences Estimation

New Developments in Econometrics Lecture 11: Difference-in-Differences Estimation New Developments in Econometrics Lecture 11: Difference-in-Differences Estimation Jeff Wooldridge Cemmap Lectures, UCL, June 2009 1. The Basic Methodology 2. How Should We View Uncertainty in DD Settings?

More information

Gov 2002: 4. Observational Studies and Confounding

Gov 2002: 4. Observational Studies and Confounding Gov 2002: 4. Observational Studies and Confounding Matthew Blackwell September 10, 2015 Where are we? Where are we going? Last two weeks: randomized experiments. From here on: observational studies. What

More information

Selection on Observables: Propensity Score Matching.

Selection on Observables: Propensity Score Matching. Selection on Observables: Propensity Score Matching. Department of Economics and Management Irene Brunetti ireneb@ec.unipi.it 24/10/2017 I. Brunetti Labour Economics in an European Perspective 24/10/2017

More information

Identification and Inference in Causal Mediation Analysis

Identification and Inference in Causal Mediation Analysis Identification and Inference in Causal Mediation Analysis Kosuke Imai Luke Keele Teppei Yamamoto Princeton University Ohio State University November 12, 2008 Kosuke Imai (Princeton) Causal Mediation Analysis

More information

Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook)

Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook) Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook) 1 2 Panel Data Panel data is obtained by observing the same person, firm, county, etc over several periods. Unlike the pooled cross sections,

More information

Statistical Analysis of Causal Mechanisms

Statistical Analysis of Causal Mechanisms Statistical Analysis of Causal Mechanisms Kosuke Imai Princeton University November 17, 2008 Joint work with Luke Keele (Ohio State) and Teppei Yamamoto (Princeton) Kosuke Imai (Princeton) Causal Mechanisms

More information

Structural Nested Mean Models for Assessing Time-Varying Effect Moderation. Daniel Almirall

Structural Nested Mean Models for Assessing Time-Varying Effect Moderation. Daniel Almirall 1 Structural Nested Mean Models for Assessing Time-Varying Effect Moderation Daniel Almirall Center for Health Services Research, Durham VAMC & Dept. of Biostatistics, Duke University Medical Joint work

More information

PSC 504: Instrumental Variables

PSC 504: Instrumental Variables PSC 504: Instrumental Variables Matthew Blackwell 3/28/2013 Instrumental Variables and Structural Equation Modeling Setup e basic idea behind instrumental variables is that we have a treatment with unmeasured

More information

Causal Inference Basics

Causal Inference Basics Causal Inference Basics Sam Lendle October 09, 2013 Observed data, question, counterfactuals Observed data: n i.i.d copies of baseline covariates W, treatment A {0, 1}, and outcome Y. O i = (W i, A i,

More information

Statistical Analysis of Causal Mechanisms for Randomized Experiments

Statistical Analysis of Causal Mechanisms for Randomized Experiments Statistical Analysis of Causal Mechanisms for Randomized Experiments Kosuke Imai Department of Politics Princeton University November 22, 2008 Graduate Student Conference on Experiments in Interactive

More information

Applied Quantitative Methods II

Applied Quantitative Methods II Applied Quantitative Methods II Lecture 10: Panel Data Klára Kaĺıšková Klára Kaĺıšková AQM II - Lecture 10 VŠE, SS 2016/17 1 / 38 Outline 1 Introduction 2 Pooled OLS 3 First differences 4 Fixed effects

More information

Econ 2148, fall 2017 Instrumental variables I, origins and binary treatment case

Econ 2148, fall 2017 Instrumental variables I, origins and binary treatment case Econ 2148, fall 2017 Instrumental variables I, origins and binary treatment case Maximilian Kasy Department of Economics, Harvard University 1 / 40 Agenda instrumental variables part I Origins of instrumental

More information

Inference for Regression Inference about the Regression Model and Using the Regression Line, with Details. Section 10.1, 2, 3

Inference for Regression Inference about the Regression Model and Using the Regression Line, with Details. Section 10.1, 2, 3 Inference for Regression Inference about the Regression Model and Using the Regression Line, with Details Section 10.1, 2, 3 Basic components of regression setup Target of inference: linear dependency

More information

Econometric Analysis of Cross Section and Panel Data

Econometric Analysis of Cross Section and Panel Data Econometric Analysis of Cross Section and Panel Data Jeffrey M. Wooldridge / The MIT Press Cambridge, Massachusetts London, England Contents Preface Acknowledgments xvii xxiii I INTRODUCTION AND BACKGROUND

More information

Applied Microeconometrics (L5): Panel Data-Basics

Applied Microeconometrics (L5): Panel Data-Basics Applied Microeconometrics (L5): Panel Data-Basics Nicholas Giannakopoulos University of Patras Department of Economics ngias@upatras.gr November 10, 2015 Nicholas Giannakopoulos (UPatras) MSc Applied Economics

More information

Gov 2000: 13. Panel Data and Clustering

Gov 2000: 13. Panel Data and Clustering Gov 2000: 13. Panel Data and Clustering Matthew Blackwell Harvard University mblackwell@gov.harvard.edu November 28, 2016 Where are we? Where are we going? Up until now: the linear regression model, its

More information

Impact Evaluation Workshop 2014: Asian Development Bank Sept 1 3, 2014 Manila, Philippines

Impact Evaluation Workshop 2014: Asian Development Bank Sept 1 3, 2014 Manila, Philippines Impact Evaluation Workshop 2014: Asian Development Bank Sept 1 3, 2014 Manila, Philippines Session 15 Regression Estimators, Differences in Differences, and Panel Data Methods I. Introduction: Most evaluations

More information

ESTIMATING AVERAGE TREATMENT EFFECTS: REGRESSION DISCONTINUITY DESIGNS Jeff Wooldridge Michigan State University BGSE/IZA Course in Microeconometrics

ESTIMATING AVERAGE TREATMENT EFFECTS: REGRESSION DISCONTINUITY DESIGNS Jeff Wooldridge Michigan State University BGSE/IZA Course in Microeconometrics ESTIMATING AVERAGE TREATMENT EFFECTS: REGRESSION DISCONTINUITY DESIGNS Jeff Wooldridge Michigan State University BGSE/IZA Course in Microeconometrics July 2009 1. Introduction 2. The Sharp RD Design 3.

More information

An Introduction to Causal Analysis on Observational Data using Propensity Scores

An Introduction to Causal Analysis on Observational Data using Propensity Scores An Introduction to Causal Analysis on Observational Data using Propensity Scores Margie Rosenberg*, PhD, FSA Brian Hartman**, PhD, ASA Shannon Lane* *University of Wisconsin Madison **University of Connecticut

More information

Empirical Likelihood Methods for Two-sample Problems with Data Missing-by-Design

Empirical Likelihood Methods for Two-sample Problems with Data Missing-by-Design 1 / 32 Empirical Likelihood Methods for Two-sample Problems with Data Missing-by-Design Changbao Wu Department of Statistics and Actuarial Science University of Waterloo (Joint work with Min Chen and Mary

More information

PSC 504: Dynamic Causal Inference

PSC 504: Dynamic Causal Inference PSC 504: Dynamic Causal Inference Matthew Blackwell 4/8/203 e problem Let s go back to a problem that we faced earlier, which is how to estimate causal effects with treatments that vary over time. We could

More information

Dynamics in Social Networks and Causality

Dynamics in Social Networks and Causality Web Science & Technologies University of Koblenz Landau, Germany Dynamics in Social Networks and Causality JProf. Dr. University Koblenz Landau GESIS Leibniz Institute for the Social Sciences Last Time:

More information

The Economics of European Regions: Theory, Empirics, and Policy

The Economics of European Regions: Theory, Empirics, and Policy The Economics of European Regions: Theory, Empirics, and Policy Dipartimento di Economia e Management Davide Fiaschi Angela Parenti 1 1 davide.fiaschi@unipi.it, and aparenti@ec.unipi.it. Fiaschi-Parenti

More information

ESTIMATION OF TREATMENT EFFECTS VIA MATCHING

ESTIMATION OF TREATMENT EFFECTS VIA MATCHING ESTIMATION OF TREATMENT EFFECTS VIA MATCHING AAEC 56 INSTRUCTOR: KLAUS MOELTNER Textbooks: R scripts: Wooldridge (00), Ch.; Greene (0), Ch.9; Angrist and Pischke (00), Ch. 3 mod5s3 General Approach The

More information

Write your identification number on each paper and cover sheet (the number stated in the upper right hand corner on your exam cover).

Write your identification number on each paper and cover sheet (the number stated in the upper right hand corner on your exam cover). STOCKHOLM UNIVERSITY Department of Economics Course name: Empirical Methods in Economics 2 Course code: EC2402 Examiner: Peter Skogman Thoursie Number of credits: 7,5 credits (hp) Date of exam: Saturday,

More information

Causal Mediation Analysis in R. Quantitative Methodology and Causal Mechanisms

Causal Mediation Analysis in R. Quantitative Methodology and Causal Mechanisms Causal Mediation Analysis in R Kosuke Imai Princeton University June 18, 2009 Joint work with Luke Keele (Ohio State) Dustin Tingley and Teppei Yamamoto (Princeton) Kosuke Imai (Princeton) Causal Mediation

More information

Statistical Analysis of Randomized Experiments with Nonignorable Missing Binary Outcomes

Statistical Analysis of Randomized Experiments with Nonignorable Missing Binary Outcomes Statistical Analysis of Randomized Experiments with Nonignorable Missing Binary Outcomes Kosuke Imai Department of Politics Princeton University July 31 2007 Kosuke Imai (Princeton University) Nonignorable

More information

Clustering as a Design Problem

Clustering as a Design Problem Clustering as a Design Problem Alberto Abadie, Susan Athey, Guido Imbens, & Jeffrey Wooldridge Harvard-MIT Econometrics Seminar Cambridge, February 4, 2016 Adjusting standard errors for clustering is common

More information

Lecture Module 8. Agenda. 1 Endogeneity. 2 Instrumental Variables. 3 Two-stage least squares. 4 Panel Data: First Differencing

Lecture Module 8. Agenda. 1 Endogeneity. 2 Instrumental Variables. 3 Two-stage least squares. 4 Panel Data: First Differencing Lecture Module 8 Agenda 1 Endogeneity 2 Instrumental Variables 3 Two-stage least squares 4 Panel Data: First Differencing 5 Panel Data: Fixed Effects 6 Panel Data: Difference-In-Difference Endogeneity

More information

Propensity Score Weighting with Multilevel Data

Propensity Score Weighting with Multilevel Data Propensity Score Weighting with Multilevel Data Fan Li Department of Statistical Science Duke University October 25, 2012 Joint work with Alan Zaslavsky and Mary Beth Landrum Introduction In comparative

More information

BOOTSTRAPPING DIFFERENCES-IN-DIFFERENCES ESTIMATES

BOOTSTRAPPING DIFFERENCES-IN-DIFFERENCES ESTIMATES BOOTSTRAPPING DIFFERENCES-IN-DIFFERENCES ESTIMATES Bertrand Hounkannounon Université de Montréal, CIREQ December 2011 Abstract This paper re-examines the analysis of differences-in-differences estimators

More information

Regression Discontinuity Designs in Stata

Regression Discontinuity Designs in Stata Regression Discontinuity Designs in Stata Matias D. Cattaneo University of Michigan July 30, 2015 Overview Main goal: learn about treatment effect of policy or intervention. If treatment randomization

More information

Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies

Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies Kosuke Imai Princeton University February 23, 2012 Joint work with L. Keele (Penn State)

More information

Robustness to Parametric Assumptions in Missing Data Models

Robustness to Parametric Assumptions in Missing Data Models Robustness to Parametric Assumptions in Missing Data Models Bryan Graham NYU Keisuke Hirano University of Arizona April 2011 Motivation Motivation We consider the classic missing data problem. In practice

More information

Discussion of Papers on the Extensions of Propensity Score

Discussion of Papers on the Extensions of Propensity Score Discussion of Papers on the Extensions of Propensity Score Kosuke Imai Princeton University August 3, 2010 Kosuke Imai (Princeton) Generalized Propensity Score 2010 JSM (Vancouver) 1 / 11 The Theme and

More information

Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies

Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies Unpacking the Black-Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies Kosuke Imai Princeton University January 23, 2012 Joint work with L. Keele (Penn State)

More information

AGEC 661 Note Fourteen

AGEC 661 Note Fourteen AGEC 661 Note Fourteen Ximing Wu 1 Selection bias 1.1 Heckman s two-step model Consider the model in Heckman (1979) Y i = X iβ + ε i, D i = I {Z iγ + η i > 0}. For a random sample from the population,

More information

Applied Econometrics Lecture 1

Applied Econometrics Lecture 1 Lecture 1 1 1 Università di Urbino Università di Urbino PhD Programme in Global Studies Spring 2018 Outline of this module Beyond OLS (very brief sketch) Regression and causality: sources of endogeneity

More information

Linear Modelling in Stata Session 6: Further Topics in Linear Modelling

Linear Modelling in Stata Session 6: Further Topics in Linear Modelling Linear Modelling in Stata Session 6: Further Topics in Linear Modelling Mark Lunt Arthritis Research UK Epidemiology Unit University of Manchester 14/11/2017 This Week Categorical Variables Categorical

More information

Introduction to causal identification. Nidhiya Menon IGC Summer School, New Delhi, July 2015

Introduction to causal identification. Nidhiya Menon IGC Summer School, New Delhi, July 2015 Introduction to causal identification Nidhiya Menon IGC Summer School, New Delhi, July 2015 Outline 1. Micro-empirical methods 2. Rubin causal model 3. More on Instrumental Variables (IV) Estimating causal

More information

Telescope Matching: A Flexible Approach to Estimating Direct Effects

Telescope Matching: A Flexible Approach to Estimating Direct Effects Telescope Matching: A Flexible Approach to Estimating Direct Effects Matthew Blackwell and Anton Strezhnev International Methods Colloquium October 12, 2018 direct effect direct effect effect of treatment

More information

AAEC/ECON 5126 FINAL EXAM: SOLUTIONS

AAEC/ECON 5126 FINAL EXAM: SOLUTIONS AAEC/ECON 5126 FINAL EXAM: SOLUTIONS SPRING 2013 / INSTRUCTOR: KLAUS MOELTNER This exam is open-book, open-notes, but please work strictly on your own. Please make sure your name is on every sheet you

More information

Flexible Estimation of Treatment Effect Parameters

Flexible Estimation of Treatment Effect Parameters Flexible Estimation of Treatment Effect Parameters Thomas MaCurdy a and Xiaohong Chen b and Han Hong c Introduction Many empirical studies of program evaluations are complicated by the presence of both

More information

Gov 2000: 9. Regression with Two Independent Variables

Gov 2000: 9. Regression with Two Independent Variables Gov 2000: 9. Regression with Two Independent Variables Matthew Blackwell Fall 2016 1 / 62 1. Why Add Variables to a Regression? 2. Adding a Binary Covariate 3. Adding a Continuous Covariate 4. OLS Mechanics

More information

Assessing Studies Based on Multiple Regression

Assessing Studies Based on Multiple Regression Assessing Studies Based on Multiple Regression Outline 1. Internal and External Validity 2. Threats to Internal Validity a. Omitted variable bias b. Functional form misspecification c. Errors-in-variables

More information

150C Causal Inference

150C Causal Inference 150C Causal Inference Instrumental Variables: Modern Perspective with Heterogeneous Treatment Effects Jonathan Mummolo May 22, 2017 Jonathan Mummolo 150C Causal Inference May 22, 2017 1 / 26 Two Views

More information

EMERGING MARKETS - Lecture 2: Methodology refresher

EMERGING MARKETS - Lecture 2: Methodology refresher EMERGING MARKETS - Lecture 2: Methodology refresher Maria Perrotta April 4, 2013 SITE http://www.hhs.se/site/pages/default.aspx My contact: maria.perrotta@hhs.se Aim of this class There are many different

More information

The returns to schooling, ability bias, and regression

The returns to schooling, ability bias, and regression The returns to schooling, ability bias, and regression Jörn-Steffen Pischke LSE October 4, 2016 Pischke (LSE) Griliches 1977 October 4, 2016 1 / 44 Counterfactual outcomes Scholing for individual i is

More information

II. MATCHMAKER, MATCHMAKER

II. MATCHMAKER, MATCHMAKER II. MATCHMAKER, MATCHMAKER Josh Angrist MIT 14.387 Fall 2014 Agenda Matching. What could be simpler? We look for causal effects by comparing treatment and control within subgroups where everything... or

More information

Differences-in- Differences. November 10 Clair

Differences-in- Differences. November 10 Clair Differences-in- Differences November 10 Clair The Big Picture What is this class really about, anyway? The Big Picture What is this class really about, anyway? Causality The Big Picture What is this class

More information

multilevel modeling: concepts, applications and interpretations

multilevel modeling: concepts, applications and interpretations multilevel modeling: concepts, applications and interpretations lynne c. messer 27 october 2010 warning social and reproductive / perinatal epidemiologist concepts why context matters multilevel models

More information

Econ 582 Fixed Effects Estimation of Panel Data

Econ 582 Fixed Effects Estimation of Panel Data Econ 582 Fixed Effects Estimation of Panel Data Eric Zivot May 28, 2012 Panel Data Framework = x 0 β + = 1 (individuals); =1 (time periods) y 1 = X β ( ) ( 1) + ε Main question: Is x uncorrelated with?

More information

Combining multiple observational data sources to estimate causal eects

Combining multiple observational data sources to estimate causal eects Department of Statistics, North Carolina State University Combining multiple observational data sources to estimate causal eects Shu Yang* syang24@ncsuedu Joint work with Peng Ding UC Berkeley May 23,

More information

The Distribution of Treatment Effects in Experimental Settings: Applications of Nonlinear Difference-in-Difference Approaches. Susan Athey - Stanford

The Distribution of Treatment Effects in Experimental Settings: Applications of Nonlinear Difference-in-Difference Approaches. Susan Athey - Stanford The Distribution of Treatment Effects in Experimental Settings: Applications of Nonlinear Difference-in-Difference Approaches Susan Athey - Stanford Guido W. Imbens - UC Berkeley 1 Introduction Based on:

More information

Covariate Balancing Propensity Score for General Treatment Regimes

Covariate Balancing Propensity Score for General Treatment Regimes Covariate Balancing Propensity Score for General Treatment Regimes Kosuke Imai Princeton University October 14, 2014 Talk at the Department of Psychiatry, Columbia University Joint work with Christian

More information

Lecture Note 2 Causal Inference in Economics, with an Application to the Minimum Wage Debate

Lecture Note 2 Causal Inference in Economics, with an Application to the Minimum Wage Debate Lecture Note 2 Causal Inference in Economics, with an Application to the Minimum Wage Debate David Autor, MIT Department of Economics 14.03/14.003 Microeconomic Theory and Public Policy, Fall 2016 1 1

More information

Extending causal inferences from a randomized trial to a target population

Extending causal inferences from a randomized trial to a target population Extending causal inferences from a randomized trial to a target population Issa Dahabreh Center for Evidence Synthesis in Health, Brown University issa dahabreh@brown.edu January 16, 2019 Issa Dahabreh

More information

CEPA Working Paper No

CEPA Working Paper No CEPA Working Paper No. 15-06 Identification based on Difference-in-Differences Approaches with Multiple Treatments AUTHORS Hans Fricke Stanford University ABSTRACT This paper discusses identification based

More information

Gov 2002: 13. Dynamic Causal Inference

Gov 2002: 13. Dynamic Causal Inference Gov 2002: 13. Dynamic Causal Inference Matthew Blackwell December 19, 2015 1 / 33 1. Time-varying treatments 2. Marginal structural models 2 / 33 1/ Time-varying treatments 3 / 33 Time-varying treatments

More information