🚀 go-pugleaf

This note is in response to Archer Gravely's message asking
how to deal with some sample survey data where the response
rate is below 60%.  He is right to be concerned.  There is no
real substitute for a high response rate but there are steps
that can and should be taken.

First of all, if this has not already been done, a "weight"
should be attached to each student in the sample (both
respondents and nonrespondents).  In the example, 500 students
were representing a population of about 3000, so the weights
would average about 6 (the number of students in the
population that each student in the sample represents).

Now divide the students into categories ("cells") that are
(relatively) homogeneous.  In the example, I would use
race-age-cum gpa cells.  Then "adjust" the weights of the
responding sampled students so they continue to add to the
cell count for the population.  Weighted estimates should now
be more representative of the population.

Lastly (for Rich Thompson), the Dillman reference (a book) is
D.A. Dillman, Mail and Telephone Surveys.  New York:  Wiley,
1978.


Michael P. Cohen
National Center for Education Statistics
Washington, DC 20208-5654
xic@cu.nih.gov