Class 7 – Randomization* *(how to design an RCT that accommodates real world constraints)

Slides:



Advertisements
Similar presentations
AFRICA IMPACT EVALUATION INITIATIVE, AFTRL Africa Program for Education Impact Evaluation Muna Meky Impact Evaluation Cluster, AFTRL Slides by Paul J.
Advertisements

Standardized Scales.
Choosing the level of randomization
Girls’ scholarship program.  Often small/no impacts on actual learning in education research ◦ Inputs (textbooks, flipcharts) little impact on learning.
How to randomize? povertyactionlab.org. Quick Review: Why Randomize Choosing the Sample Frame Choosing the Unit of Randomization Options How to Choose.
Presented by Malte Lierl (Yale University).  How do we measure program impact when random assignment is not possible ?  e.g. universal take-up  non-excludable.
Chapter 11 What Works and What Doesn’t. Are Hospitals Good for You? From Angrist and Pischke, Mostly Harmless Econometrics.
Povertyactionlab.org How to Randomize? Abhijit Banerjee Massachusetts Institute of Technology.
Benefits and limits of randomization 2.4. Tailoring the evaluation to the question Advantage: answer the specific question well – We design our evaluation.
What could go wrong? Deon Filmer Development Research Group, The World Bank Evidence-Based Decision-Making in Education Workshop Africa Program for Education.
Progress out of Poverty Index
Selection of Research Participants: Sampling Procedures
How do Vouchers Work? Evidence from Colombia Eric Bettinger, Case Western U Michael Kremer, Harvard Juan Saavedra, Harvard 7 June 2007 World Bank.
Impact Evaluation: The case of Bogotá’s concession schools Felipe Barrera-Osorio World Bank 1 October 2010.
Chapter 12 Sample Surveys
Course Content Introduction to the Research Process
Impact Evaluation Session VII Sampling and Power Jishnu Das November 2006.
Conditional Cash Transfers for Improving Utilization of Health Services Health Systems Innovation Workshop Abuja, January 25 th -29 th, 2010.
1 Targeting the Ultra Poor: An Impact Assessment.
Experiments and Observational Studies.  A study at a high school in California compared academic performance of music students with that of non-music.
Goal Paper  Improve our understanding on whether business training can improve business practices and firm outcomes (sales, profits, investment) of poor.
Cross-Country Workshop for Impact Evaluations in Agriculture and Community Driven Development Addis Ababa, April 13-16, 2009 AIM-CDD Using Randomized Evaluations.
Sampling UAPP 702 Research Methods for Urban & Public Policy
1 Randomization in Practice. Unit of randomization Randomizing at the individual level Randomizing at the group level –School –Community / village –Health.
COLLECTING QUANTITATIVE DATA: Sampling and Data collection
Research and Evaluation Center Jeffrey A. Butts John Jay College of Criminal Justice City University of New York August 7, 2012 How Researchers Generate.
Measuring Impact: Experiments
AADAPT Workshop South Asia Goa, December 17-21, 2009 Nandini Krishnan 1.
Adjustment of benefit Size and composition of transfer in Kenya’s CT-OVC program Carlo Azzarri & Ana Paula de la O Food and Agriculture Organization.
POLS 7170X Master’s Seminar Program/Policy Evaluation Class 5 Brooklyn College – CUNY Shang E. Ha.
Randomized Controlled Trials in Rural Finance: An Example from India Michael Faye and Sendhil Mullainathan Harvard University March 2007
Assessing the Distributional Impact of Social Programs The World Bank Public Expenditure Analysis and Manage Core Course Presented by: Dominique van de.
How do we Randomize? Jenny C. Aker Tufts University.
Sampling “Sampling is the process of choosing sample which is a group of people, items and objects. That are taken from population for measurement and.
Tahir Mahmood Lecturer Department of Statistics. Outlines: E xplain the role of sampling in the research process D istinguish between probability and.
The World Bank Human Development Network Spanish Impact Evaluation Fund.
Nigeria Impact Evaluation Community of Practice Abuja, Nigeria, April 2, 2014 Measuring Program Impacts Through Randomization David Evans (World Bank)
Why Use Randomized Evaluation? Isabel Beltran, World Bank.
Applying impact evaluation tools A hypothetical fertilizer project.
Measuring Impact 1 Non-experimental methods 2 Experiments
Africa Impact Evaluation Program on AIDS (AIM-AIDS) Cape Town, South Africa March 8 – 13, Steps in Implementing an Impact Evaluation Nandini Krishnan.
TRANSLATING RESEARCH INTO ACTION What and How to randomize? July 9, 2011 Dhaka Raymond Guiteras, Assistant Professor University of Maryland povertyactionlab.org.
CHOOSING THE LEVEL OF RANDOMIZATION. Unit of Randomization: Individual?
Africa Program for Education Impact Evaluation Dakar, Senegal December 15-19, 2008 Experimental Methods Muna Meky Economist Africa Impact Evaluation Initiative.
Comments on: The Evaluation of an Early Intervention Policy in Poor Schools Germano Mwabu June 9-10, 2008 Quebec City, Canada.
Bangor Transfer Abroad Programme Marketing Research SAMPLING (Zikmund, Chapter 12)
Implementing an impact evaluation under constraints Emanuela Galasso (DECRG) Prem Learning Week May 2 nd, 2006.
5 Education for All Development Issues in Africa Spring 2007.
Chapter 3 Surveys and Sampling © 2010 Pearson Education 1.
Global Workshop on Development Impact Evaluation in Finance and Private Sector Rio de Janeiro, June 6-10, 2011 Using Randomized Evaluations to Improve.
HPTN Ethics Guidance for Research: Community Obligations Africa Regional Working Group Meeting, May 19-23, 2003 Lusaka, Zambia.
Africa Impact Evaluation Program on AIDS (AIM-AIDS) Cape Town, South Africa March 8 – 13, Randomization.
Slide 7.1 Saunders, Lewis and Thornhill, Research Methods for Business Students, 5 th Edition, © Mark Saunders, Philip Lewis and Adrian Thornhill 2009.
Copyright © 2015 Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution-Non Commercial-No Derivatives (CC-IGO.
© 2010 Jones and Bartlett Publishers, LLC. Chapter 12 Clinical Epidemiology.
Cross-Country Workshop for Impact Evaluations in Agriculture and Community Driven Development Addis Ababa, April 13-16, Causal Inference Nandini.
Formulation of the Research Methods A. Selecting the Appropriate Design B. Selecting the Subjects C. Selecting Measurement Methods & Techniques D. Selecting.
Randomized Control Trials
How to Randomize.
Explanation of slide: Logos, to show while the audience arrive.
Impact Evaluation Toolbox
Randomization This presentation draws on previous presentations by Muna Meky, Arianna Legovini, Jed Friedman, David Evans and Sebastian Martinez.
Randomization This presentation draws on previous presentations by Muna Meky, Arianna Legovini, Jed Friedman, David Evans and Sebastian Martinez.
Sampling and Power Slides by Jishnu Das.
Impact Evaluation Designs for Male Circumcision
Explanation of slide: Logos, to show while the audience arrive.
Sampling for Impact Evaluation -theory and application-
Steps in Implementing an Impact Evaluation
Steps in Implementing an Impact Evaluation
Presentation transcript:

Class 7 – Randomization* *(how to design an RCT that accommodates real world constraints)

Randomly sample from area of interest

Randomly sample from area of interest Randomly assign to treatment and control Randomly sample From both treatment and control

BEFORE YOU EVEN TALK ABOUT RANDOMIZATION DESIGN…  Explain why randomization is necessary  Talk about attribution  Articulate why (non-random) comparison group may be different and specify way this may bias results  Explain that the intervention may have unintended (good or bad) consequences worth measuring  Can measure spillover  Cost-benefit analyses

 Unit of Randomization  Ways to Randomize  Simple Randomization (eg: lottery)  Randomization “in the bubble”  Phase-in  Rotation  Encouragement Design  Multiple treatments, 2 stage  (if time) Stratification

 Options  Individual  Cluster  Which level to randomize?  Considerations  What unit does the program target for treatment?  What is the unit of analysis?

“Groups of individuals”: Cluster Randomized Trial

 In the example discussed last week, students were tutored after school   how did we randomize last week?  What if students are taken out of class to be tutored, say by a teaching assistant?   how would you randomize?

 Randomizing at the child-level within classes  Randomizing at the class-level within schools  Randomizing at the community-level

 Nature of the Treatment  How is the intervention administered?  What is the catchment area of each “unit of intervention”  How wide is the potential impact?  Aggregation level of available data  Do you want to measure spillover?  Power requirements  Generally, best to randomize at the level at which the treatment is administered.

 Sometimes a program is only large enough to serve a handful of communities  Primarily an issue of statistical power  Will be addressed next week

 Unit of Randomization  Ways to Randomize  Simple Randomization (eg: lottery)  Randomization “in the bubble”  Phase-in  Rotation  Encouragement Design  Multiple treatments, 2 stage  (if time) Stratification

“high rate” “low rate” Final sample included 132 branch offices in 80 geographic clusters

 School Voucher Program in Colombia (PACES)  Student must be entering 6th grade and under 15 years old  Students must provide evidence that they live in poor neighborhood  Renewable through graduation unless student is retained in a grade  Vouchers awarded by lottery if demand exceeds supply  Covered about 60% of fees  Key Findings (after 3 years) on Voucher Recipients:  Increased Usage of Private Schools  Higher Educational Attainment  No Difference in Drop-out Rates  Less Grade Repetition  Higher Test Scores  Less Incidence of Teen-age Employment  Further research on peer effects

 Lotteries are simple, common and transparent  Randomly chosen from applicant pool  Participants know the “winners” and “losers”  Simple lottery is useful when there is no a priori reason to discriminate  Perceived as fair  Transparent

 What if you have 500 applicants for 500 slots?  Could increase outreach activities  (but think of external validity)  Sometimes screening matters  Suppose there are 2000 applicants for 500 slots  Screening of applications produces 500 “worthy” candidates  A simple lottery will not work - What are our options?

What are they screening for? Which elements are essential? Selection procedures may exist only to reduce eligible candidates in order to meet a capacity constraint If certain filtering mechanisms appear “arbitrary” (although not random), randomization can serve the purpose of filtering and help us evaluate

 Organizations may not be willing to randomize among eligible people.  But might be willing to randomize those at the margin – ie, those who are borderline in terms of eligibility  Just above the threshold  not eligible, but almost  What treatment effect do we measure? What does it mean for external validity?  (hint: review RDD from last week)

 Dean Karlan and Jonathan Zinman worked with a bank in South Africa  Loan officers ranked applicants as “egregiously uncreditworthy” or “marginally uncreditworthy”  Randomly selected marginal applicants to be reconsidered  53% of those were offered a loan  Look at impact of credit on those randomized into the “reconsider group”  Not just those offered loan… more on this next week.  Found that access to credit increased likelihood that clients would retain job, increase income, feel less food insecurity  Marginal loans are profitable, but less than regular loans

 Take advantage of operational constraints  Typical during expansion phase  Everyone gets program eventually  Figure out what determines order of expansion  Examples: Progresa (Mexico), Deworming (Kenya)

Phase-in design Round 1 Treatment: 1/3 Control: 2/3 Round 2 Treatment: 2/3 Control: 1/3 Round 3 Treatment: 3/3 Control: Round 1 Treatment: 1/3 Control: 2/3 Round 2 Treatment: 2/3 Control: 1/3 Randomized evaluation ends

Advantages Everyone gets something eventually Provides incentives to maintain contact Concerns Can complicate estimating long-run effects Care required with phase-in windows Do expectations of change actions today?

Round 1 Treatment: 1/2 Control: 1/2 Rotation design Round 2 Treatment from Round 1  Control —————————————————————————— Control from Round 1  Treatment Round 1 Treatment: 1/2 Control: 1/2

Groups get treatment in turns – Group A gets treatment in first period – Group B gets treatment in second period Advantages – Perceived as fairer; easier to get accepted Concerns – If people in Group B anticipate they’ll receive the treatment the next period, they can have a different behavior in the first period – Impossible to measure long-term impact since no control group after first period

Group Year 1Year 2Year 3 AGrade 3Grade 4Grade 3 BGrade 4Grade 3Grade 4 Schools in Varodara, India were divided into two groups: ( A and B) and teaching assistant offered in schools according to the following schedule:

 Sometimes it’s practically or ethically impossible to randomize program access  But most programs have less than 100% take- up  Randomize encouragement to receive treatment

Something that makes some folks more likely to use program than others Not itself a “treatment” For whom are we estimating the treatment effect? Think about who responds to encouragement Do not choose an encouragement that affects those who are different than the entire population

 Unit of Randomization  Ways to Randomize  Simple Randomization (eg: lottery)  Randomization “in the bubble”  Phase-in  Rotation  Encouragement Design  Multiple treatments, 2 stage  (if time) Stratification

Treatment 1 Treatment 2 Treatment 3 Multiple treatments

 Spillover: when the control group, although “untreated”, is affected (positively or negatively) by the treatment  Choose unit that contains spillover (ie randomize at school or village rather than individual level)  Measure Spillover: TUP (Bangladesh, Honduras, Peru, Pakistan, Ghana, Ethiopia, Yemen)

 Program which consists of targeting poorest families in a village and providing consumption support ($$ or food), asset transfer, livelihood training  Eventual graduation to microfinance

Communities (Total = 80) 40 treatment 40 control C 20 cont Households (Total = 1600) A 20 treat Ho do we determine impact? Program direct impact: A-C

Villagers may benefit from neighbors being treated Or they may be negatively affected Why would it be desirable (from a policy perspective) to measure spillover?

Communities (Total = 80) 40 treatment 40 control C 20 cont Households (Total = 2400) A 20 treat B 20 cont Ho do we determine impact? Program direct impact: A-C Program indirect impact: B-C

Objective: when you have a small sample, make sure key variables are balanced between Treatment and Control What is it: – dividing the sample into different subgroups – selecting treatment and control from each subgroup Stratify on variables that could have important impact on outcome variable (bit of a guess) Stratify on subgroups that you are particularly interested in (where may think impact of program may be different) Can get complex to stratify on too many variables Makes the draw less transparent the more you stratify

 Sampling!