Google Sign In
Facebook Sign In
Non-response, which is inter alia affected by overburdened respondents, is a pressing problem of survey research.
One of the ways of lessening respondent burden is to shorten the questionnaire and to use randomized groups. For less important questions on a scale from 1 to 5, it is sufficient to collect 30 responses (sometimes even 10 to 15). In such cases, the confidence interval is 3.0±0.4, which is completely adequate to detect low (a great deal under 3) and high (a great deal over 4) values, e.g. of (un)satisfaction. With 100 responses, the confidence interval is 3.0 ± 0.2, and with 400 responses, the confidence interval is 3.0 ± 0.1, which is mostly sufficient for almost all practical needs.
Why would we then present such a question to all respondents (for example 1000) and risk a higher non-response rate due to a longer and more difficult questionnaire? With the use of randomized subgroups we can randomly assign less important questions to respondent subsamples. Random group creation can be achieved with the use of conditions (IF statements).
Important questions and questions that must be analysed simultaneously (e.g. correlations) are of course assigned to all respondents, while less important questions can be presented to only a half (even/odd), a third, a quarter, etc. The described approach is the key method of efficiently avoiding excessively long questionnaires and dramatic increase in the number of questions.
