In quantitative causal inference, we do power analysis to determine the minimum sample size to achieve (1 - $\beta$) statistical power of our hypothesis test with a specific $\alpha$ statistical significance and a minimum, meaningful effect size.

In qualitative studies, when using Grounded Theory, how should we ensure the saturation to stop collecting more data? For example, after interviewing twenty people, when I observe that the next four as saying very similar things that I heard before and there is no new theme being added, is that an indication of saturation? How should I guarantee that if I interview another fifty people, I will not find very interesting new themes that I did not observe in the previous twenty people?

In our specific study, we provide each student participant with two similar passages with two types of knowledge representations in random order. Then ask them questions about each of the two passages, in addition to questions about how they compare their learning through each knowledge representation format. We have interviewed 40 people and thematically coded their recorded thoughts. We identified 73 themes, and many of them were new to us. We are not sure how many more participants we should interview. If we interview five more people and cannot find any new themes, does that indicate a reasonable saturation to stop our interviews? How can we guarantee that if we interview ten more people, we will not be able to find new themes that we did not encounter before?

  • 1
    $\begingroup$ By interviewing a more diverse set of people? $\endgroup$ Commented Apr 23, 2022 at 14:43
  • $\begingroup$ So, how do you formulate it in an objective way? $\endgroup$ Commented Apr 25, 2022 at 0:44
  • $\begingroup$ I'm sorry to say that I do not know, but, if you can add some more details&context about your study (please do it as ab edit and not only in comments) then maybe somebody can! $\endgroup$ Commented Apr 25, 2022 at 2:49
  • $\begingroup$ I added about our specific study. Thank you for your feedback. $\endgroup$ Commented Apr 25, 2022 at 12:45
  • $\begingroup$ If the people are not chosen randomly (or, in some cases, systematically) for interviews, there's almost nothing you can say and no guarantees you can make apart from the mathematically limiting ones (e.g., if you interview 30 more people you won't observe more than 30 new answers). Otherwise, you might conceive of this as a kind of quality control application, where there are many techniques you could deploy; or possibly as a sequential experiment. The exploratory nature of your investigation suggests considerations of power are premature. $\endgroup$
    – whuber
    Commented Apr 25, 2022 at 13:45


Your Answer

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