8
$\begingroup$

What is a good and relatively brief overview of good practice for getting valid results from surveys.

I'm particularly interested in something about good survey design and analysis.

$\endgroup$
2
  • 2
    $\begingroup$ This is unanswerable. There are several components of good survey practice: (1) sample design: the classic designs are covered by Cochran (1977) and Kish (1995) that were recommended by Michael Chernick, but the modern reality of phone and address-based sampling are way more complicated, and these books will tell you nothing about practical aspects; (2) questionnaire design and psychometrics of getting valid answers from people -- a totally unrelated area entrenched in psychology rather than in probability theory; (3) survey management -- the art and science of data collection itself; ... $\endgroup$
    – StasK
    Commented Jun 19, 2012 at 14:34
  • 3
    $\begingroup$ (4) survey post-processing, weighting and variance estimation, which is a return to statistics -- again, this is a rather poorly covered topic in terms of existing books. My brief learning of this took about two years of two classes per semester. If you are determined to do a survey on your own, you will likely screw up the first one completely, you will miss a lot of good practices on the second one, and your third one will probably be somewhat usable. As a statistical consultant, I've seen all of that, trust me. $\endgroup$
    – StasK
    Commented Jun 19, 2012 at 14:39

3 Answers 3

7
$\begingroup$

On the more qualitative side of things in survey methodology, see the books mentioned in https://blogs.rti.org/surveypost/2012/05/15/surveying-on-a-deserted-island-a-bakers-dozen-list-of-resources-to-take-along/. If you are totally new to sampling, Lohr (2009) is a more modern treatment covering additionally some of the aspects of survey data analysis, replicate variance estimation, and some practical aspects, as compared to the somewhat more formal classics mentioned by Michael Chernick.

Update: I posted the original answer in 2012. Since then, another good book came out: Valliant, Dever and Kreuter (2013). Practical Tools for Designing and Weighting Survey Samples (Amazon). This is a good practical book only to the extent that you know the basics, so my advice to learn from Lohr's book first still applies.

$\endgroup$
2
  • 1
    $\begingroup$ I agree with Stask on Lohr's book. $\endgroup$ Commented Jun 19, 2012 at 15:09
  • $\begingroup$ First link says something like "page not found". $\endgroup$
    – Learner
    Commented May 24, 2017 at 5:20
3
$\begingroup$

I recommend Cochran's Sampling Techniques. It provides the fundamentals and is very clear. Leslie Kish's Survey Sampling is another classic that I can recommend. An advantage of going to amazon is that there are often many user generated book reviews there for the OP to look at. I personally have written a lot of reviews there.
I can think of nothing better than an amazon link for textbook because once you get to the site you have user and publisher reviews that you can read and amazon often provides look ins to the Table of Contents,Preface and excerpts to chapters. You are not going to get a link that will give you a free electronic copy of the book. Now reference articles that fit the bill might be possible to recommend but I think the book recommendations are better references in this case. Perhaps a brief monograph like one of the little green SAGE books would suit the OP but I like these books better.

$\endgroup$
0
1
$\begingroup$

Previous answers seem to have well addressed OP's question. However, I will add, for the benefit of future readers, that Thomas Lumley has provided a wealth of information on "complex surveys", which can be loosely characterized as surveys with often thousands to tens of millions or more of observations (perhaps data larger than your machine's memory), often implementing complex sampling methods (e.g. National Health Interview Survey or Nationwide Inpatient Sample).

Lumley has contributed to "complex survey" analysis through his R package survey. See here for a list of presentations about the survey package, and here for a number of very good vignettes on using the survey package.

$\endgroup$
1
  • 1
    $\begingroup$ "Complex" surveys in Lumley's book (or rather in general in survey statistics) are not complex because of the sample size (they would be called "big" rather than "complex" if that were the case), but rather because the design creates samples that are very far from i.i.d., and hence need to be analyzed by special tools. I can fit the largest data set of this kind that I am aware of (10M cases from 5 year worth of American Community Survey, see impus.org, just fine on my computer, although that is indeed a 2Gb file). $\endgroup$
    – StasK
    Commented Sep 23, 2014 at 2:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.