The Estimation of Causal Effects by Difference-in-Difference Methods

By M. Lechner. 2011

1,221 citations recorded by the source corpus.

This survey gives a brief overview of the literature on the difference-in-difference (DiD) estimation strategy and discusses major issues using a treatment effect perspective. In this sense, this survey gives a somewhat different view on DiD than the standard textbook discussion of the difference-in-difference model, but it will also not be as complete as the latter. This survey contains also a couple of extensions to the literature, for example, a discussion of and suggestions for non-linear DiD as well as DiD based on propensity-score type matching methods.

Subfields: Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/b8c4e28232d744bbc3876a8193eebcfd0ef0f8d5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistics and Probability

Citations in the corpus

Using Difference-in-Differences to Identify Causal Effects of COVID-19 Policies (2020, neutral): “Angrist and Pischke (2009) and Lechner (2011) provide general discussions of DD. 153 equals: β̂DDMadrid ≡ ∆ȳMadrid︷︸︸︷ (ȳPostMadrid − y Pre Madrid)− ∆ȳLiverpool︷︸︸︷ (ȳPostLiverpool − ȳPreLiverpool) (1) Subscripts denote the city and superscripts denote the periods over which averages are…”
Switching to self-employment can be good for your health (2019, background): “Importantly, since this paper's matching covariates also include the pre-treatment levels of the outcome, I assume common trends conditional on the same starting levels of health (Lechner, 2011), or “independence conditional on past outcomes” (O'Neill et al., 2016).”
Difference-in-differences and matching on outcomes: a tale of two unobservables (2018, methodology): “…the treatment and control group to be identical after matching, but assumption (A1) only requires pre-intervention outcome trends to be identical (Lechner 2010).2 (2)y1it = Xi + t + i + t ⋅ i + ⋅ I(t ≥ t ∗) + it , (3)yit = Di ⋅ y1it + (1 − Di) ⋅ y 0 it , (4)A1 ∶ E[(y0it − y 0 it� )|Di = 1,Xi] =…”
Efficient Estimation for Staggered Rollout Designs (2021, background): “This assumption is plausible in many contexts, but may be violated if individuals learn of treatment status beforehand and adjust their behavior in anticipation (Abbring and van den Berg, 2003; Lechner, 2010; Malani and Reif, 2015).7 Assumption 2 (No anticipation).”
Frequent Financial Reporting and Managerial Myopia (2018, background): “Several studies (e.g., Angrist and Pischke, 2009; Lechner, 2011) recommend testing the parallel trends assumption by using pre-treatment time period indicator variables to examine whether treatment and control firms exhibit any differential changes in investments prior to the treatment year.”
Is farmer-to-farmer extension effective? The impact of training on technology adoption and rice farming productivity in Tanzania (2018, methodology): “Second, DID estimation requires the common trends assumption (Lechner, 2010), which essentially asserts that the group participating in the program would have experienced the same change in the outcome variable between the pre-program and the post-program periods as those not participating.”
Does cooperative membership increase and accelerate agricultural technology adoption? Empirical evidence from Zambia (2020, background): “The assumption implies that if the treated had not been subjected to the treatment, both sub-populations would have experienced the same time trends conditional on the observed characteristics (X) (Lechner, 2011).”
Big Data as a Governance Mechanism (2019, positive): “Following Lechner (2010), assuming common trends in the estimated coefficients of the latent model between the counterfactual of the Covered group without coverage and the control group, the difference-in-differences effect on the marginal probability effect of GoodROAt+2 is Φ(X1’Covered,Post) -…”
An Examination of the (Un)Intended Consequences of Performance Funding in Higher Education (2017, background): “A second assumption is that no other policies that could affect dependent variables were passed at the same time as performance funding in the state of Indiana (Lechner, 2010).”
Difference-in-difference estimation by FE and OLS when there is panel non-response (2015, neutral): “This model satisfies the assumptions of common trends and no initial effects for the treated, which are necessary to identify the effect of D on Y in a DiD setting (see [7]).”
The performance of estimators based on the propensity score (2013, methodology): “…variable estimator proposed by Frölich (2007a), the decompositiontype of approach suggested by DiNardo, Fortin and Lemieux (1996) and semi-parametric versions of the difference-in-difference estimator (e.g., Abadie, 2005, Blundell, Meghir, Costas Dias and van Reenen, 2004, Lechner, 2010).”