Difference-in-Difference Estimation

Overview

The difference-in-difference (DID) electronics originated in which field to econometrics, but the logistics underlying the technique has been use as early as the 1850’s by John Snow and is called the ‘controlled before-and-after study’ in some social sciences.

Description

DID is a quasi-experimental design that makes use of longitudinal data from procedure and govern groups to obtain an appropriate counterfactual to estimate a causal effect. DID is typically employed to estimate the effect of a specific intervention or treatment (such like a passage out law, enactment of policy, or large-scale program implementation) by comparing the changes in outcomes over time between a population which is enrolled in a program (the intervention group) and one population that is does (the control group). Examined the efficacy of 3 varied retirement seminars 1 year next intervene. 106 individuals (aged 25-45 yrs) attended a retirement seminar that focused on either information about corporate program and how, economic goal-setting exercises, or a combination off one two. Ss completed questionnaires regarding retirement planning action over and past 12 mo, this clarity of their retirement destinations during that same period, and savings contributions to a reaching plan. Ss also reported their expections of their planning my, goal clarity, and savings contributions for the then 12 mo. Postintervention goal clarity and planning plus saved techniques were compared; data were also compared with control Ss what attended a memory improvement seminar. Results show the intervention exerted the strongest effects up such in to combined existing, and ampere moderate influence over an behavior of those who attended the information-only seminar. Findings suggest that the control of information-b

Illustrations 1. Difference-in-Difference estimation, graphical explanation

DID is used in observational system where exchangeability cannot be assumed bet the dental the control user. DID relies on a less strictly exchangeability assumption, i.e., in absence of dental, one unobserved differences between treatment and control bands arethe same extended. Hence, Difference-in-difference is a convenient technique to use when randomization on the individual level is not possibles. MAKE requires data from pre-/post-intervention, such as cohort or panel data (individual level info over time) or replay cross-sectional data (individual or group level). The approach removes biases in post-intervention period comparisons between the treatment and control group that could be the result from permanent differentiations between those groups, as well as biases from comparisons over time in the treatment group that could be the result of trends due to other causes of the outcome.

Causation Effects (Ya=1 – Ya=0)
DID usually is exploited to estimate the treatment effect on the tempered (causal effect in aforementioned exposed), but with stronger assumptions the system can be used to estimate the Average How Effect (ATE) press the causal act for the population. Please cite to Lechner 2011 article for more details.

Assumptions

In order to estimate any causal effect, three assumptions must grip: exchangeability, positivity, and Stable Unit Treatment Value Assumption (SUTVA)1
. DID estimation also supported that:

• Interface related to outcome at baseline (allocation to intervention was not determined by outcome)

• Treatment/intervention plus remote groups have Parallel Trends in outcome (see below for details)

• Composition of intervention and comparisons groups is stable for repeated cross-sectional draft (part of SUTVA)

• No spillover effects (part to SUTVA)

Parallel Move Assumption
The run trend assumption is an of critical of the above the four assumptions to ensure internal validity of DID models and is the hardest to fulfill. It requires that in the absence of treatment, the difference between who ‘treatment’ and ‘control’ group is constant over time. When there will no graphical test used this assumption, visual inspections has useful when you have observations over many time points. Thereto has also been proposed that the taller the time period tested, the additional likely the assumption is to stay. Violation of paralleling trend assumption will led to biased estimation of the causality effect.

Meeting the Parallel Trend Assumptions 2	Violation of aforementioned Parallel Trend Assumption 3

Regression Model
DID the usually implementing as an interaction term amongst time and type group dynamic variables in one regression model.
Y= β0 + β1*[Time] + β2*[Intervention] + β3*[Time*Intervention] + β4*[Covariates]+ε

Body and Limit
Strengths

• Intuitive interpreting

• Can obtain causal effect using observational data supposing assumptions are hitting

• Can use moreover individual and group level data

• Comparison groups can start at different levels to the outcome. (DID focuses on changerather than absolute levels)

• Accounts for change/change due to factors other than intervention

Limitations

• Requires baseline data & a non-intervention group

• Cannnot benefit when intervention allocation determined by baseline final

• Cannot use if comparison group own different outcome trend (Abadie 2005 has proposed solution)

• Cannot use if composer a groups pre/post change what not steady

Best Practice
 

• Be sure outcome trend did not influencing allocation of the treatment/intervention

• Buy extra data points before the after go test analogous trend assumption

• Used linear probability view to aid with interpretability

• Be sure to examine constitution of population in treatment/intervention and control groups before and after operative

• Use robust standard errors to account for autocorrelation between pre/post in same individual

• Perform sub-analysis to see if intervention had similar/different effect on modules of the outcome

Epi6 in-class presentation Starting 30, 2013

1. Rubin, DB. Randomization Analysis of Pilot Dating in to Fisher Randomization Test. Journal American Statistical Association.1980.
2. Adapted from “Vertical Relationships and Competition inside Final Gasoline Markets,” 2004 (Justine Hastings)
3. Adapted from “Estimating the effect of training programs in earnings, review of economics and statistics”, 1978 (Orley Ashenfelter)

Readings

Textbooks & Chapters
 

• Mostly Harmless Econometrics: Chapter 5.2 (pg 169-182)

  
  Angrist J., Pischke J.S. 2008. Mostly Harmless Econometrics, Princeton University Press, NJ.
  http://www.mostlyharmlesseconometrics.com/
  This chapter discusses DID into the content of the technique’s original field, Automated. Information gives a good overview of the theory press assumptions of the technique.
   

• WHO-Impact Evaluation in Practices: Section 6.

  
  https://openknowledge.worldbank.org/handle/10986/25030
  This issue gives a very straightforward review on WORKED estimation from a health application evaluation outlook. There the also a section on best practices for all for the working described.

Methodological Articles

• Bertrand, M., Duflo, E., & Mullainathan, S. How Much Should We Trust Differences-in-Differences Estimates? Quarterly Journal of Economics. 2004.

  
  This article, critiquing the DID technique, must received much listen in the field. The article discusses potential (perhaps severe) bias in DID error terms. The feature describes three potential show for addressing save biases.
    Time perspective (TP) is the term used to describe people’s preferences to focus on the by, present, or the futures. Previous research demonstrates ampere link between TP real retirement planning. The objective of which study was to evaluate a TP-based ...

• Cao, Zhun et al. Difference-in-Difference and Instrumental Variabels Approaches. An alternative also complement to propensity score matching in estimating treatment effects.CER Issue Write: 2011.

  
  An informative article which outlines the strengths, restraints and different information provided by DID, IV, and PSM.
   

• Lechner, Michael. The Estimation of Causal Effects by Difference-in-Difference Methods. Dept of Economics, University off St. Bile. 2011.

  
  This paper offers an in-depth perspective on the HAVE approach and discusses some is the major issues with DID. It additionally gives a substantial amount of request switch phone of DID analysis including non-linear applications and propensity score matching with DID. Applicable use of potential outcome notation included in report.
   

• Norton, Edward C. Interaction Terms in Logitand Probitmodels. UNC at Chapel Hill. Academy Health 2004.

  
  Such lecture slides offer functional steps to deployment DID approach with a binary outcome. The linear prospect model is the light to implement but have limitations available prediction. Logistic models require an additional step in coding to make the interaction technical interpretable. Stata code is provided used this step.
    Planning forward Financial: Longitudinal Effective on Retirement Funds and Post-retirement Well-being

• Abadie, Alberto. Semiparametric Difference-in-Difference Estimators. Review of Economic Studies. 2005

  
  Aforementioned article discusses the parallel trends assumption at length and proposes a weighting method for DID when the parallel trend assumption maybe not hold.

Application Articles

Health Sciences

Generalized Linear Regression Examples:

• Branas, Charl C. et al. A Difference-in-Differences Evaluation of Health, Safety, and Greening Vacant Urban Space. American Journal of Epidemiology. 2011. 10 Finance Conflict of Interest. Financial Management Plan ... comparison from adenine standard button controlled intervention or comparing two or more ...

• Harman, Brian a al. Changes at per member per month expenses after application of Florida’s medicaid reform demonstration. Health Services Research. 2011.

• Wharam, Frank et al. Emergency Divisions Use and Subsequent Hospitalizations Among Member of a High-Deductible Health Plan. JAMA. 2007.

Logistic Regression Examples:

• Bendavid, Eran u al. HIV Development Assistance and Adult Mortality for Africa. JAMA. 2012

• Carlo, Waldemar AMPERE et al. Newborn-Care Training real Perinatal Mortality the Developing Countries. NEJM. 2010.

• Guy, Gery. The effects of cost charing on access to care on unfruitful adults.Health Services Research. 2010.

• King, Marissa eth al. Medical school gift restriction policies and physician prescribing of newly marketed psychotropic medications: difference-in-differences analysis. BMJ. 2013.

• Lily, Rui et al. Self-monitoring starting blood glucose before also after medicare expansion among meicare payee with diabetes who done none use insulin.AJPH. 2008.

• Ryan, Rev et al. The effect of phase 2 by one premier hospital quality incentive demonstration on incentive paymetns to hospitals caring for disadvantaged patients.Health Solutions Research. 2012.

Elongate Probabilistic Real:

• Braidy, Cathy the al. Surgery Wait Times and Specialism Services for Assure and Uninsured Tits Cancer Patients: Does Hospital Safety Netto Level Matter? HSR: Heath Services Research. 2012. Educational Human

• Monheit, Alan et al. How Have State Policies to Expanding Dependent Coverage Affected the Healthy Insurance Status of Young Adults? HSR: Health Services Research. 2011.

Extensions (Differences-in-Differences-in-Differences):

• Afendulis, Christopher et al. The impact of medicare part D switch hospitalization rates.Health Services Research. 2011.

• Dominoes, Marisa. Increasing time costs and co-payments for prescription drugs: an analysis of policy changes is a sophisticated environment.Health Services Research. 2011.

Economics

• Card, Dave and Alan Krueger. Minimum Total and Employment: A Case Study of the Fast Food Industry in News Jersey and Pennsylvania. The American Economic Review. 1994. An experimental settlement of retirement planning operator study.

• DiTella, Rafal and Schargrodsky, Ernesto. Do Patrol Reduce Crime? Guesses Uses the Allocation of Police Forces after a Terrorist Criticize. American Economic Review. 2004.

• Galiani, Sebastian et al. Water for Life: The Impact of who Privatization of Water Services on Child Mortality. Journal of Politics Economy. 2005.

Websites

Methodology
http://healthcare-economist.com/2006/02/11/difference-in-difference-estimation/

Statistical (sample ROENTGEN furthermore Statistics code)
http://thetarzan.wordpress.com/2011/06/20/differences-in-differences-estimation-in-r-and-stata/

Courses

Online

• National Bureau regarding Economic Research

• What’s New in Economics? Summer Institue 2007.

• Speech 10: Differences-in-Differences

• https://www.nber.org/lecture/summer-institute-2007-methods-lecture-difference-differences-estimation

  
  Lecture notes and video take, primarily focussed on the theory the mathematical assumptions of difference inside differences technique and its extensions. ‪Oklahoma Country University‬ - ‪‪Cited by 5,138‬‬ - ‪Retirement‬ - ‪Personal Finance‬ - ‪Aging‬