FRT Problem Set estimator and Lin’s estimator

Problem Set 3
Due October 18th, 11pm
Note that October 21st will be our in-class mid-term.
1. A predictive estimator and Lin’s estimator
Consider a completely randomized experiment. Let Zi
, xi and Yi be the binary treatment, centered
covariates, and outcome for unit i, i = 1, . . . , n. We can use Lin’s estimator ˆτL to estimate the
average treatment effect.
We also discussed a strategy to impute all missing potential outcomes. From the treatment
group, we can use the OLS to fit a linear predictor for the potential outcome under treatment:µˆ1(xi) = ˆγ1 + βˆT1 xi. From the control group, we can use the OLS to fit a linear predictor for the
potential outcome under control: ˆµ0(xi) = ˆγ0 + βˆT0 xi. Then we can use these predictors to impute
the missing potential outcome, leading to a predictive estimator

Show the above identities using the properties of the OLS.
2. Data re-analyses
Re-analyze three datasets from matched-pair designs.
(1) In FRTDarwinMP.R, I analyze Darwin’s data using the FRT based on the test statistic ˆτ .
Re-analyze this dataset using the FRT with the Wilcoxon signed rank sum statistic.
Re-analyze this dataset based on the Neymanian inference: unbiased point estimator, conservative
variance estimator, 95% confidence interval.
(2) In NeymanMPstar.R, I analyze the data from based on Neymanian inference.
Re-analyze this dataset using the FRT with different test statistics.
Re-analyze this dataset using the FRT with covariate adjustment, e.g., you can define test
statistics based on residuals from the OLS fit of the observed outcome on covariates. Will the
conclusion change if you do not include an intercept in your OLS fit?
(3) Use the data from Angrist and Lavy (2009). The original analysis is quite complicated. We
focus only on Table A1 viewing the schools as experimental units. Then we have a matchedpair
design on the schools. For simplicity, we drop pair 6 and all the pairs with noncompliance.
This results in 14 complete pairs. The outcome is the Bagrut passing rates in 2001 and 2002,
with the Bagrut passing rates in 1999 and 2000 as pretreatment covariates.
Re-analyze the data using the FRT with and without covariate adjustment.
Re-analyze the data based on the Neymanian inference with and without covariates.
3. Covariance estimator in matched-pair designs
In a matched-pair design, we define the within-pair differences of outcome and covariate as
τˆi = (2Zi − 1)(Yi1 − Yi2), τˆxi = (2Zi − 1)(xi1 − xi2),
and the averages of them as
Show that an unbiased estimator of cov(ˆτ, τˆx) isˆθ =1n(n − 1)∑ni=1(ˆτxi − τˆx)(ˆτi − τˆ).
4. Data analysis: stratification and regression
Use the dataset homocyst in the R package senstrat. The outcome is homocysteine, the homocysteine
level, and the treatment is z, where z = 1 for a daily smoker and z = 0 for a never smoker.
Covariates are female, age3, ed3, bmi3, pov2 with detailed explanations in the R package. st
is a stratum indicator, defined by all the combinations of the discrete covariates.
(1) How many strata have only treated or control units? What is the proportion of the units in
these strata? Drop these strata and perform a stratified analysis of the observational study.
Report the point estimator, variance estimator and 95% confidence interval for the average
treatment effect.
(2) Run OLS of the outcome on the treatment indicator and covariates without interactions. Report
the result.
(3) Apply Lin’s estimator of the average treatment effect. Report the result.
(4) Compare the results in the above three analyses. Which one is more credible?
5. More results on observational studies
The Hajek estimator differs from the Horvitz–Thompson estimator in the numerators.
6. Re-analysis of Rosenbaum and Rubin (1983)
Use Table 1 of this paper. If you are interested, you can read the whole paper. It is a canonical
paper. But for this problem, you only need Table 1.
Rosenbaum and Rubin (1983) fitted a logistic regression model for the propensity score and
stratified the data into 5 subclasses. Because the treatment (Surgical versus Medical) is binary and
the outcome is also binary (improved or not), they represented the data by a table.
Based on this table, estimate the average treatment effect, and report the 95% confidence
Angrist, J. and Lavy, V. (2009). The effects of high stakes high school achievement awards: Evidence
from a randomized trial. The American Economic Review, 99:1384–1414.
Rosenbaum, P. R. and Rubin, D. B. (1983). Assessing sensitivity to an unobserved binary covariate
in an observational study with binary outcome. Journal of the Royal Statistical Society, Series
B (Methodological), 45:212–218.

因为专业,所以值得信赖。如有需要,请加QQ99515681 或邮箱:[email protected]