Syndicate content

A Curated List of Our Postings on Technical Topics – Your One-Stop Shop for Methodology

David McKenzie's picture
This is a curated list of our technical postings, to serve as a one-stop shop for your technical reading. I’ve focused here on our posts on methodological issues in impact evaluation – we also have a whole lot of posts on how to conduct surveys and measure certain concepts that I’ll leave for another time. Updated August 20, 2015.
Random Assignment
Allocating treatment and control with multiple applications per applicant and ranked choices
Is optimization just re-randomization redux? Thoughts on the "don't randomize, optimize" papers
Be an optimista, not a randomista, when you have small samples
Tips for randomization in the wild: adding a waitlist
How to randomize in the field
Stratified randomization and the FIFA world cup
Doing stratified randomization with uneven numbers in the Strata
How to randomize using many baseline variables
Public randomization ceremonies
Designing experiments to measure spillover effects
Mechanism experiments and opening up the black box
Sample weights and RCT design
Pre-analysis plans and reporting
Pre-registration of studies to avoid fishing and allow transparent discovery
A joint test of orthogonality when testing for baseline balance
A pre-analysis plan check-list
The New Trial Registries
What isn’t reported in impact evaluations but maybe should be
Randomization checks: testing for joint orthogonality
Propensity Score Matching
Guido Imbens on clustering standard errors with matching
Testing different matching estimators as applied to job training programs
The covariate balanced propensity score
The often unspoken assumptions behind diff-in-diff
Regression Discontinuity
Curves in all the wrong places: Gelman and Imbens on why not to use higher-order polynomials in RD
Regression discontinuity with an implicit index
Other Evaluation Methods
Evaluating an Argentine tourism policy using synthetic controls: tan linda que enamora?
Impact as narrative
The synthetic control method, as applied to regulatory reforms
Using spatial variation in program performance to identify impacts
Small n impact evaluation methods
Can we trust shoestring evaluations?
Another reason to prefer Ancova: dealing with measurement changes between baseline and follow-up
Endogenous stratification: the surprisingly easy way to bias your heterogeneous treatment effects and what to do instead
Why is difference-in-difference estimation still so popular in experimental analysis?
Regression adjustment in randomized experiments (part one, part two)
When to use survey weights in analysis
Adjustments for multiple hypothesis testing
Bounding approaches to deal with attrition
Linear probability models versus probits
Dealing with multiple lotteries
Estimating standard errors with small clusters (part one, part two)
Decomposition methods
Estimation of treatment effects with incomplete compliance
Power Calculations and Improving Power
Should I work with only a subsample of my control group if I have take-up problems?
Power calculations: what software should I use?
Does the intra-cluster correlation matter for power calculations if I am going to cluster my standard errors?
Power calculations for propensity score matching
Power calculations 101: dealing with incomplete take-up
Collecting more rounds of data to boost power
Improving power in small samples
On External Validity
Getting beyond the mirage of external validity
All those external validity issues with impacts? They apply to costs too
External validity as seen from other quantitative social sciences and the gaps in our practices
Towards a more systematic approach to external validity: understanding site selection bias
Weighting for external validity
Will that successful intervention over there get results here?
Learn to live without external validity
Why the external validity of regression estimates can be less than you think
Why similarity is the wrong concept for external validity
A rant on the external validity double standard
Jargony Terms in Impact Evaluations
A proposed taxonomy of behavioral responses to evaluation
Quantifying the Hawthorne effect
The Hawthorne Effect
The John Henry Effect
Placebo effects
Clinical Equipoise
Stata Tricks
Generating regression and summary statistics tables in Stata
Graphing impacts with Standard Error Bars
Calculating the intra-cluster correlation
Generating regression and summary statistics tables in Stata: A checklist and code
Worm wars: the anthology
Worm wars: a review of the reanalysis of the Miguel and Kremer deworming study
Response to Brown and Wood's response
Brown and Woods response on "how scientific are scientific replications"
how scientific are scientific replications?
Systematic reviews and meta-analysis
how systematic is that systematic review? The case of learning outcomes
How standard is a standard deviation? A cautionary note on using SDs to compare across impact evaluations
should we give up on SDs for measuring effect size?
What do 600 papers on 20 types of interventions tell us about what types of interventions generalize?


Submitted by jleavitt on

I wish that other bloggers would do something like this. It is always nice to have a mini "database" of articles about a certain topic. You never know when something like that will come in handy.

Submitted by Gaurav D. Joshi on

Fully Agree with Homira. Thanks a ton David for showing the way. Hope other interesting topics will also have one such 'one-stop-shop'.

Submitted by GG on

Great list - knew many of these posts but not others, great to have them all together in one place. Now David, a follow-up request. I knew your small N larg T paper and its ANCOVA recommendation, but I hadn't seen the posts by Lin on regression adjustment:
A linear combination of both would be great, since "your" ANCOVA is a regrssion adjustment on a special case, averaged pre-treatment outcomes, and those may vary substantially with the outcomes and the treatment effects...

Submitted by D Phillips on

David - I defer absolutely to the high level of sophistication of your impact benefits analysis (!) - i.e. the calculation of the numerator of your rate of return. However I could not get a good understanding of the cost - i.e. the denominator. I.e. what is the (social) cost of the impact benefits.

If we don't take the cost into account then we of course we might be saying simply that if you give an entrepreneur $50,000 his business is likely to be in better shape in three years time than one that does not get the $50,000 under most scenarios - which would not be surprising.

Please could you explain What is the denominator?
1. just the value of the business grants?
2. 1. plus BPC management costs?
3. 2. plus transaction costs to winning (and losing) entrepreneurs?
4. 3. plus matching investment from the entrepreneur?
5. 4. plus contributed investment funds (funded by outside loans, equity)?

And if it is 1. or 2. how much of the overall return benefits are attributable only to the grants? Could we assume a crowding-in effect which attributes the complementary investment also to some extent to the grants ?? (I see that you have found no crowding in effect overall but it seems likely to have happened in some cases).

Getting back to the numerator how have you accounted for benefit externalities that might show up in linkages and associated technological spillover effects, which are arguably key to the development of industry, banking and business services etc? (That is, indirect effects that cannot be accounted for through performance of the treatment or control group enterprises). OR how do you allow for possible contamination of the control group by technological spillovers generated by enterprises that receive awards?? (Apart from the fact hat Nigeria ia a big country and businesses may not meet face to face).

Id really appreciate understanding this since it is key to the validity of the rate of return analysis and by extension the value of business plans.

This should really be posted as a comment to my post on the Nigeria business plan competition, not to the general technical topic links. But I couldn't figure out how to move where you had put the comment, so will answer here.

Short answer is that:
- the measures of impact take into account transaction costs to the winners, and any crowding out or crowding in of other funding, including matching funding provided by them and/or equity/loan finance that is crowded out. The paper shows a little bit of crowd-out of this other finance, and no evidence of crowding in.
- I am not able to measure the spillovers to the control group, and have to rely on it being a small number of winners spread throughout the country.
- the measure of the cost of the program includes both the cost of the grants and the cost of operating the program (approx $2 million in costs to hand out $58 million in grants).

Submitted by Ty on

How is this "curated" exactly? It's just a freaking list. Stop saying things are "curated." You can't "curate" a list of things you already published. I'd contend you can't "curate" a list of anything at all. What an obnoxious trend it is becoming to use the word "curated" for "a bunch of stuff a person happens to like." Ugh!

Curate, verb meaning to select, organize, and look after the items in a collection. This precisely describes the process used to go through almost 5 years of old blog posts, select the ones related to technical topics, and organize them in a way that is hopefully useful for people so that they form a coherent collection.

Add new comment