© Michael Lechner, 2006, p. 1 (Non-bayesian) Discussion (translation) of Principal Stratification for Causal Inference with Extended Partial Complience.

Slides:



Advertisements
Similar presentations
Modelling partial compliance through copulas in a principal stratification framework Bartolucci F. and Grilli L. (2011) To appear in JASA. Leonardo Grilli.
Advertisements

Designing an impact evaluation: Randomization, statistical power, and some more fun…
Introduction to Propensity Score Matching
REGRESSION, IV, MATCHING Treatment effect Boualem RABTA Center for World Food Studies (SOW-VU) Vrije Universiteit - Amsterdam.
Econometrics Session 1 – Introduction Amine Ouazad, Asst. Prof. of Economics.
Methods of Economic Investigation Lecture 2
AP Statistics Section 5.2 B More on Experiments
The counterfactual logic for public policy evaluation Alberto Martini hard at first, natural later 1.
Copyright EM LYON Par accord du CFC Cession et reproduction interdites Research in Entrepreneurship- The problem of unobserved heterogeneity Frédéric Delmar.
Improving health worldwide George B. Ploubidis The role of sensitivity analysis in the estimation of causal pathways from observational.
 1  Outline  Model  problem statement  detailed ARENA model  model technique  Output Analysis.
Econometric Modeling More on Experimental Design.
Pooled Cross Sections and Panel Data II
4. Multiple Regression Analysis: Estimation -Most econometric regressions are motivated by a question -ie: Do Canadian Heritage commercials have a positive.
Impact Evaluation: The case of Bogotá’s concession schools Felipe Barrera-Osorio World Bank 1 October 2010.
Chapter 2 – Tools of Positive Analysis
1 Difference in Difference Models Bill Evans Spring 2008.
TRADUIRE LA RECHERCHE EN ACTION Employment RCTs in France Bruno Crépon.
Single and Multiple Spell Discrete Time Hazards Models with Parametric and Non-Parametric Corrections for Unobserved Heterogeneity David K. Guilkey.
Opportunities and Challenges in a Multi-Site Regression Discontinuity Design Stephen W. Raudenbush University of Chicago Presentation at the MultiLevel.
1 Comment on Zabel/Schwartz/Donald: An Analysis of the Impact of SSP on Wages Alexander Spermann Mannheim 28 October 2006.
JDS Special program: Pre-training1 Carrying out an Empirical Project Empirical Analysis & Style Hint.
AADAPT Workshop Latin America Brasilia, November 16-20, 2009 Non-Experimental Methods Florence Kondylis.
IC-based value creation process of firms: cluster approach Grigorii Teplykh Marina Oskolkova The research is carried out in the framework of "Science Foundation.
Quasi Experimental Methods I Nethra Palaniswamy Development Strategy and Governance International Food Policy Research Institute.
Instrumental Variables: Problems Methods of Economic Investigation Lecture 16.
SUTVA, Assignment Mechanism STA 320 Design and Analysis of Causal Studies Dr. Kari Lock Morgan and Dr. Fan Li Department of Statistical Science Duke University.
Private involvement in education: Measuring Impacts Felipe Barrera-Osorio HDN Education Public-Private Partnerships in Education, Washington, DC, March.
Methods of explanatory analysis for psychological treatment trials workshop Methodology Research Group Funded by: MRC Methodology Grant G MHRN Methodology.
Comments on “Partial Identification by Extending Subdistributions” by Alexander Torgovitsky Frank A. Wolak Department of Economics Director, Program on.
Michael Lechner, 2005 Discussion of Out-of-Pocket Health Care Expenditures by Edward Norton, Hua Wang, Sally C. Stearns Michael Lechner University of St.
CAUSAL INFERENCE Presented by: Dan Dowhower Alysia Cohen H 615 Friday, October 4, 2013.
ECON 3039 Labor Economics By Elliott Fan Economics, NTU Elliott Fan: Labor 2015 Fall Lecture 21.
Discussion of: The Impact of a Temporary Help Job on Participants in Three Federal Programs by Carolyn J. Heinrich, Peter H. Muser and Kenneth R. Troske.
RCTs and instrumental variables Anna Vignoles University of Cambridge.
Mediation: Solutions to Assumption Violation
Application 2: Minnesota Domestic Violence Experiment Methods of Economic Investigation Lecture 6.
1 Statistics in Research & Things to Consider for Your Proposal May 2, 2007.
Treatment Heterogeneity Cheryl Rossi VP BioRxConsult, Inc.
Applying impact evaluation tools A hypothetical fertilizer project.
Marginal Treatment Effects and the External Validity of the Oregon Health Insurance Experiment Amanda Kowalski Associate Professor, Department of Economics,
Instrumental Variables: Introduction Methods of Economic Investigation Lecture 14.
Discussion of „Employment risk and the living arrangements of young adults“ Martin Biewen, University of Mainz.
Labour class Cost and Benefits of Danish Active Labour Market Programmes Lars Skipper Anvendt KommunalForskning (+Svend & Jakob)
Experimental and Ex Post Facto Designs
Discussion Tobias J. Klein: „College Education and Wages in the U.K.: Estimating Conditional Average Structural Functions in Nonadditive Models with Binary.
MATCHING Eva Hromádková, Applied Econometrics JEM007, IES Lecture 4.
STRUCTURAL MODELS Eva Hromádková, Applied Econometrics JEM007, IES Lecture 10.
The Evaluation Problem Alexander Spermann, University of Freiburg 1 The Fundamental Evaluation Problem and its Solution SS 2009.
Alexander Spermann University of Freiburg, SS 2008 Matching and DiD 1 Overview of non- experimental approaches: Matching and Difference in Difference Estimators.
INSTRUMENTAL VARIABLES Eva Hromádková, Applied Econometrics JEM007, IES Lecture 5.
ENDOGENEITY - SIMULTANEITY Development Workshop. What is endogeneity and why we do not like it? [REPETITION] Three causes: – X influences Y, but Y reinforces.
Copyright © 2015 Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution-Non Commercial-No Derivatives (CC-IGO.
IMPACT EVALUATION PBAF 526 Class 5, October 31, 2011.
Statistical Experiments What is Experimental Design.
The Evaluation Problem Alexander Spermann, University of Freiburg, 2007/ The Fundamental Evaluation Problem and its Solution.
Survival time treatment effects
Quasi Experimental Methods I
Threats and Analysis.
Quasi Experimental Methods I
Difference-in-Differences
Introduction to Design
Matching Methods & Propensity Scores
Matching Methods & Propensity Scores
Matching Methods & Propensity Scores
Instrumental Variables
1/18/2019 1:17:10 AM1/18/2019 1:17:10 AM Discussion of “Strategies for Studying Educational Effectiveness” Mark Dynarski Society for Research on Educational.
Inferential Statistics
Alternative Scenarios and Related Techniques
Measuring the Wealth of Nations
Presentation transcript:

© Michael Lechner, 2006, p. 1 (Non-bayesian) Discussion (translation) of Principal Stratification for Causal Inference with Extended Partial Complience by Hui Jin and Don Rubin Mannheim, ZEW, October 2006 Michael Lechner SIAW, ZEW, CEPR, IZA

© Michael Lechner, 2006, p. 2 A statistics paper in an econometric perspective  Perspective 1: An exercise in IV estimation (à la IA ’94 and AIR ’96) Complication: There is only a binary instrument, but we are interested in the effects of multiple treatments in the form of dose responses. A binary instrument is not powerful enough for such comparisons.  Perspective 2: We ‘must’ condition on an endogenous variable (an intermediate outcome) to estimate the effect of interest  Paper shows how to recover the causal effect of interest (!) in such a framework  These problems occur although there is underlying experiment that assigns people to different treatment states  However: Paper uses different language than econometricians do...

© Michael Lechner, 2006, p. 3 An artifical example from the training literature... as translation device  Unemployed want to attend a training programme  UE is randomized in one of 2 programmes, called T and C (Z)  T is tough programme – not much fun, a lot of work, add. human capital  C is a leasurly social experience, no human capital  Each programme has duration of 4 weeks, participants may leave programmes any time (even before they start)  UE have a taste for leasure  Programmes have heterogenous effects

© Michael Lechner, 2006, p. 4 An artifical example from the training literature... as translation device (2)  We want to understand the effect of the programmes on the employment rate 2 years after the start of the programme.  Even more: We may want to understand the effects of a completed programme compared to the other completed programme.  Problem: If we base the analysis on the subsamples of those who complete the programme, we may contaminate the causal inference, because those who realised that they have a low return may have dropped out already.  This type of selection problem is more likely to occur with the tough programme.

© Michael Lechner, 2006, p. 5 Solutions of the identification problems  The paper provides two solutions to this problem (and is VERY clear about the underlying assumptions) 1)Instead of conditioning on the endogenous observable intermediate outcome (programme duration), condition on the potential intermediate outcomes. For example: Compare the person that completed to the tough programme to somebody who participated in the easy programme but would have completed the tough programme and average (unobservable  find other restrictions!). Here, require also same propensity to complete the easy programme 2)Device a hypothetical random experiment that would identify the effects

© Michael Lechner, 2006, p. 6 Z as something like an instrument of D and d Assumptions used in paper  Standard assumptions: SUTVA, Z randomized  Exclusion of direct effect of Z on Y:If a change in Z does not affect (potential) terminating behavior in both programmes, than potential outcomes Y(Z) are the same (plausible in example)  Strong access monotonicity (D(C)=0, d(T)=0): If randomised into the tough programme, there is no way of participating in parts of the easy programme, and conversly... (plausible in strict experiment)  removes 2 of the 4 (partially) unobservables from the playing field... Negative side effect monotonity ( ): If UE would have left nice programme, UE would have left tough programme as well (???) [behavioral assumption]  restricts the remaining unobservable in terms of the observable

© Michael Lechner, 2006, p. 7 Identification and estimation How does identification work without a Bayesian perspective? The missing equation...  Paper shows a Bayesian estimation strategy  For a Non-Bayesian, there remain a couple of open points that center around the equation that is missing in the paper:  Issues: - What is the role of the different assumption in the identification step ? - More specific: For example, what happens if we assume weak (insted of strong) access monotonicity? In this case, do we identify an interval or a point? - Frequentist estimation... which moments of the data are required?

© Michael Lechner, 2006, p. 8 Next target: Dose response  Additional assumption - Dose depends (only) on single index which is observable for control group, but unobservable for controls - Example: There is some underlying variable which influences length of participation. However, for the nice programme the UE follow this ‚desire‘, but for the nasty programme they deviate towards. This deviation is influenced by the ‚desire‘ only and is otherwise random (hard to justify in this example)  Same questions as before...

© Michael Lechner, 2006, p. 9 Conclusion  Principal stratification could be a very helpful concept in econometrics  It is clearly related to IV estimation – relation could be made even more explicit  Taking account of the non-Bayesian perspective would greatly enhance its value for econometricians