Hui Xie and Yi Qian, "Measuring the Impact of Nonignorability in Panel 
Data with Non-Monotone Nonresponse", Journal of Applied Econometrics, 
Vol. 27, No. 1, 2012, pp. 129-159.

The data used in this paper are those used in

Preisser JS, Galecki AT, Lohman KK, and Wagenknecht, LE (2000). "Analysis of
Smoking Trends With Incomplete Longitudinal Binary Responses", Journal of
the American Statistical Association, 95:1021-31.

The dataset can be also downloaded at Statlib-- JASA Data Archive, under the
directory named "pglw".

The dataset is in one plain ASCII file called pglwlong.dat, which is an
ASCII file in DOS format. It is zipped in the file pglwlong.zip. Unix/Linux
userrs should use "unzip -a".

The size of the file is approximately 397KB.  The file contains data for
5078 subjects. The data matrix, if read in correctly, should have 20312
observations (rows) and have 7 variables (columns). The names of variables,
from left to right in the data matrix, are as follows:

id = person identifier code

time = the year of followup measurement with 
		year 0 referencing 1986.
		year 2 referencing 1988.
		year 5 referencing 1991.
		year 7 referencing 1993.

age = baseline (time=0) age in years

birth      1 = 1963-1967
           2 = 1959-1962 
           3 = 1955-1958

education: 1 = High School or less
	   2 = Some College 
	   3 = College Degree

racesex:   1 = Black Males
           2 = Black Females
           3 = White Males
           4 = White Females

y  = reported smoker (1=yes, 0=no, .=missing)