In a project that I am doing to evaluate customer behaviour, I have asked a set of questions such as whether they read the unit pricing before they decide to purchase a particular product or do they only buy a particular product if it is on 'special'.

In a population of 60 stores (with unequal customer base but the total number of customers being 85,000), I used probability proportional to size (PPS) sampling method to constructed a sample of 1,000 customers.

A total of 900 customers responded to the survey. The response rate is 90%.

Conversely, the non-response rate is 10%.

How 'acceptable' is this non-response rate? (I know the response rate for this example is impressive!)

What factors need to be considered when making a decision on whether this non-response rate is acceptable or not?

  • 7
    $\begingroup$ These are questions to be considered, read about, and discussed for the long haul. The answers depend on situation, personal preference, research purpose, audience, etc. Woe unto those who seek shortcuts to all learning on a Q&A site! $\endgroup$
    – rolando2
    Feb 1, 2012 at 4:05
  • 5
    $\begingroup$ I don't think that last comment was intended personally, Adhesh, but was meant to emphasize the point that general questions without a clear application are not usually appropriate for this site. It is possible to learn systematically by asking a series of related questions. However, because our focus is on the immediate question, it would be difficult even for a dedicated learner to gain the perspective and knowledge that come from studying a text, taking a course, or engaging expert services. Help us help you: please refine this question to focus on a definite problem you currently face. $\endgroup$
    – whuber
    Feb 1, 2012 at 20:08
  • 4
    $\begingroup$ I would second @whuber's comment which points to an important aspect when asking question: Without some context, it is often too hard to provide a definitive and arguable answer, let alone some relevant pointers. In particular, what kind of questionnaire is it (e.g., when validating a questionnaire, I would first ask if this doesn't reflect a problem with face validity); does the response rate (RR) depend on respondents' characteristics like gender, age, or geographical location; what do you intend to do with the results (30% RR on a 2-year follow-up with youth might be seen as acceptable)? $\endgroup$
    – chl
    Feb 1, 2012 at 21:02

2 Answers 2


I won't offer a rule of thumb, although I don't think 80 percent would be achievable in many situations.

I will suggest that a crucial thing to consider is why some people are not responding. If you are certain non-response is completely random (which is vanishingly unlikely I'm afraid) and unrelated to any variables of interest than almost any response rate is satisfactory. If not, then you need to consider the whole range of issues of bias, imputation, acceptable tolerance of inaccuracy, etc. before you can venture what is acceptable.

I also strongly recommend Lumley's book.


Pace the commenter, in my field people often talk about 80% as being the rock-solid level as a rule of thumb.

I read this book recently on Survey Analysis and found it extremely useful (based on R which is a big plus from my point of view, don't know what flavour software you prefer):



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.