Stata Command: reporterror

Stata Command: reporterror.ado

The Stata output displayed below shows estimated probability masses of the true years of education (Ui) using two noisy measurements, namely self-reported years of education (Xi) and sibling-reported years of education (Yi), with possibly nonclassical and nonseparable measurement errors (Hu and Sasaki, 2017). Also displayed are the conditional probabilities of correct reports by Xi and Yi given the true level of education Ui. We observe that the self reports tend to be accurate when the true years of education is greater (e.g., 16-18), while sibling reports tend to be accurate in typical diploma years (e.g., 12, 16 and 18) - see discussions in Hu and Sasaki (2017). These results can be automatically produced by the Stata command reporterror as follows.

reporterror

Installation:

. ssc install reporterror

Usage:

. use "example_education.dta"

. reporterror selfreport siblingreport, minu(12)

Help:

. help reporterror

Reference: Hu, Y. and Y. Sasaki (2017) Identification of Paired Nonseparable Measurement Error Models. Econometric Theory, 33 (4), pp. 955-979. Paper.

Title
 reporterror -- Executes estimation of the probability masses of an unobserved discrete random variable using two measurements with possibly nonclassical and nonseparable measurement errors.
Syntax
 reporterror x y [if] [in] [, minu(real) maxu(real) nounderreport(real) boot(real)]
Description
 reporterror executes estimation of the probability masses of an unobserved discrete random variable using two measurements with possibly nonclassical and nonseparable measurement errors based on Hu and Sasaki (2017). The command takes two measurements, x (e.g., self-reported years of education) and y (e.g., sibling-reported years of education), of an unobserved discrete latent variable u (e.g., true years of education), and estimates the probability masses f(u) of u. Also reported are the conditional probabilities of correct reports x and y given u.
Options
 minu(real) sets the lower bound of the support of u. Not calling this option automatically sets it to the minimum value of the common support of x and y.
 maxu(real) sets the upper bound of the support of u. Not calling this option automatically sets it to the maximum value of the common support of x and y.
 nounderreport(real) can be used to toggle between the assumption of no under-reporting and no over-reporting. Not calling this option or setting a non-zero argument in this option implies the assumption of no under-reporting. Calling this option with zero as an argument implies the assumption of no over-reporting
 boot(real) sets the number of bootstrap iterations. The default value is boot(2500).
Examples
 (u true years of education, selfreport first measurement of u, siblingreport second measurement of u)
 Estimation focusing on individuals with high-school diploma u=12 or higher degrees:
 . use "example_education.dta" . reporterror selfreport siblingreport, minu(12)

Stored results
 reporterror stores the following in r():
 Scalars r(N) observations
 Macros r(cmd) reporterror
 Matrices r(suppU) support of U r(fU) mass f(U) of U r(fX) mass f(X) of X r(fY) mass f(Y) of Y r(pXU) conditional probability Pr(X=U|U) r(pYU) conditional probability Pr(Y=U|U)
Reference
 Hu, Y. and Y. Sasaki. 2017. Identification of Paired Nonseparable Measurement Error Models. Econometric Theory, 33 (4), pp. 955-979. Link to Paper.
Authors
 Yingyao Hu, Johns Hopkins University, Baltimore, MD. Yuya Sasaki, Vanderbilt University, Nashville, TN.