Background
I am working on a paired randomized experiment, I have 32 schools of different sizes, matched on a few characteristics into 16 pairs. Treatment was randomized at the school level. Data is collected at the student level. (1) I want to use pair fixed effects to estimate differences at the student level within pairs, (2) I want to cluster standard errors at the school level, and (3) I want to bootstrap standard errors.
I believe that the following commands should produce equivalent estimates:
(a) bootstrap, r(50) seed(10101) cluster(Entidad) idcluster(school): xtreg sex_EL treatment, fe i(Par)
and
(b) xtreg sex_EL treatment, fe i(Par) vce(bootstrap, r(50) seed(10101)) cluster(Entidad) nonest dfadj
However, command (a) runs, but command (b) returns an error: "insufficient observations to compute bootstrap standard errors"
I have 32 schools, 16 pairs, 3,654 students. The smallest school has 14 students, the smallest pair: 61 students.
Questions
1) Why is command (b) returning this error?
2) How is command (a) different than command (b)?
3) In "A Practitioner's Guide to Cluster-Robust Inference", Cameron and Miller write, "In Stata the pairs cluster bootstrap for OLS without fixed effects can be implemented in several equivalent ways including: ... ". Why "without fixed effects"?
I am working on a paired randomized experiment, I have 32 schools of different sizes, matched on a few characteristics into 16 pairs. Treatment was randomized at the school level. Data is collected at the student level. (1) I want to use pair fixed effects to estimate differences at the student level within pairs, (2) I want to cluster standard errors at the school level, and (3) I want to bootstrap standard errors.
I believe that the following commands should produce equivalent estimates:
(a) bootstrap, r(50) seed(10101) cluster(Entidad) idcluster(school): xtreg sex_EL treatment, fe i(Par)
and
(b) xtreg sex_EL treatment, fe i(Par) vce(bootstrap, r(50) seed(10101)) cluster(Entidad) nonest dfadj
However, command (a) runs, but command (b) returns an error: "insufficient observations to compute bootstrap standard errors"
I have 32 schools, 16 pairs, 3,654 students. The smallest school has 14 students, the smallest pair: 61 students.
Questions
1) Why is command (b) returning this error?
2) How is command (a) different than command (b)?
3) In "A Practitioner's Guide to Cluster-Robust Inference", Cameron and Miller write, "In Stata the pairs cluster bootstrap for OLS without fixed effects can be implemented in several equivalent ways including: ... ". Why "without fixed effects"?
