dr birnbaum is absolutely correct and it's nice to see the same citation i
was planning on sending (it's a classic for survey work)
btw 60% is very generous, for US government work the agencies usually like
80-90% or higher [private companies may require even higher levels]
the problem at that level is the costs involved in follow-up on
non-respondents such surveys that i have conducted typically require 3 to 4
major follow-up contacts and may add telephone tracing of survey
non-respondents
a high nonn-response rate makes it very difficult to estimate non-response
bias
all of this speaks to the need for a well designed study, as rigorous as
any RCT, with well defined sampling frames, preparation of the respondents,
well targeted and well designed instruments, and diligence on the part of
the researcher
bill
William D. Grant, Ed.D.
Executive Vice Chair, Research Professor
Director, Center for Evidence Based Practice
Department of Family Medicine
SUNY Upstate Medical University
475 Irving Ave., #200
Syracuse, NY 13210
315.464.6997
315.464.6982 (f)
>>> David Birnbaum <[log in to unmask]> 04/25 1:27 PM >>>
You posted:
> Dear list,
> I am looking for literature about critical appraisal of questionnaire
> surveys especially for the problems with non-respondents (nonresponse
> bias).
> I've heard that the JAMA only accepts surveys with a response-rate of >=
> 60%. But I could not find such a minimum rate in the literature
> (searched 3 survey research textbooks and 2 clinical epidemiology
> textbooks).
>
> Three questions:
> Is there any evidence for such a minimum rate or is it "only" a
> convention like a p-value of 0,05?
> If it is a convention is anyone aware of refernces about it?
> Is such a minimum rate reasonable?
The answer to your three questions is yes, yes, and yes. For a good
discussion of the impact of nonresponse rates on width of confidence
intervals, as well as appropriate strategies, see pp. 359-63 in William
Cochran's Sampling Techniques (3rd edition, 1977, John Wiley & Sons).
Surveys often ignore implications of non-response; however, at response
rates below 60-70% there are distinct problems with width of adjusted
confidence intervals becoming excessive relative to precision required for
decisions, as well as increasing potential for self-selection bias to
distort conclusions.
David Birnbaum, PhD, MPH
Clinical Assistant Professor
Dept. of Health Care & Epidemiology
University of British Columbia, Canada
|