Sample Size for Qualitative Studies: Two Recent Approaches

Setting sample sizes for quantitative studies can be done in an explicit way by means of calculations based on the concept of precision and a specified effect size (ideally based on a loss function).[1] But qualitative studies are vague, based on notions of ‘theoretical saturation’. Problems:

  1. How much information is needed to satisfy the theory?
  2. How do you know how much information will be provided in the average encounter with an informant?

A small contribution to ameliorating the latter problem comes from the idea of the expected ‘information content’ of each encounter.[2] The authors identify five factors that determine this quantity:

  1. The broader the aim, the larger the sample of encounters needed.
  2. Salient knowledge of people in the sample. The greater their knowledge, the smaller the required sample.
  3. The extent to which theory is already established. The more developed the theory, the smaller the required sample.
  4. The quality of likely dialogue. The more articulate the respondents, the smaller the sample size.
  5. Analysis type. An exploratory cross-case study requires larger samples than an in-depth analysis of a few, well selected, respondents.

Certainly a useful paper and aide mémoire. However, translation into the actual sample size required remains, let’s be honest, informed guesswork.

The CLAHRC WM Director is attracted to an earlier paper by Fugard and Potts,[3] not referenced in the Malterud, et al. paper. The earlier paper proposed a logical calculation based on the three critical determinants:

  1. The expected prevalence, among respondents, of the least prevalent theme – this should be based on an explicit estimate of the prevalence of the least prevalent theme that the study should be capable of uncovering.
  2. The number of desired instances of the theme.
  3. The power of the study – the probability of detecting sufficient themes of the desired prevalence.

For example, to have an 80% power to detect two instances of a theme with a prevalence of 10% among encounters, 29 informants are required. Now that makes sense. This method has the considerable advantage of requiring the researcher to specify the prevalence of the theme that should not be missed in the sample. The CLAHRC WM Director would like to see this quantitative thinking incorporated and routinely used in planning qualitative research. He will write more generally on some of the problems in the way qualitative research is routinely framed in a future post.

— Richard Lilford, CLAHRC WM Director

References:

  1. Girling AJ, Lilford RJ, Braunholtz DA, Gillett WR. Sample-size calculations for trials that inform individual treatment decisions: a ‘true-choice’ approach. Clin Trials. 2007; 4(1): 15-24.
  2. Malterud K, Siersma VD, Guassora AD. Sample Size in Qualitative Interview Studies: Guided by Information Power. Qual Health Res. [ePub].
  3. Fugard AJB, & Potts HWW. Supporting Thinking on Sample Sizes for Thematic Analyses: a Quantitative Tool. Int J Soc Res Methodol. 2015; 18(6).
Advertisements

3 thoughts on “Sample Size for Qualitative Studies: Two Recent Approaches”

  1. We recently published a qualitative study that used a quantitative approach to sample size determination (Phys Ther 2016;96:313-323). In this study, we sought to identify response categories that may potentially be common in the broader population, so that we could build a scale that was inclusive of these common responses. It was not the aim of our study to identify how common each category was, rather, to just identify what common categories might be.

    We found the quantitative approach to sample size determination intuitively appealing, however, it was quite a fight to get this past the reviewers of this paper. Although we feel that similar approaches are justifiable, one would need to be prepared for a fight when seeking to use similar approaches in future to get them past reviewers who may not be comfortable with blending qualitative research with quantitative thinking.

    A small exert of our text from our paper relating to sample size is below.

    For this study, we arbitrarily determined
    that for a response category to
    be considered “common,” and worth
    including in the resultant item set, it
    would be present in at least 20% of the
    broader older adult population. With an
    a priori plan for 20 participants, this
    study was designed to have only a 1.2%
    chance of not observing a “minimally
    common” response in the broader adult
    population at least once in our sample of
    older adults. To illustrate, if we recruited
    a sample of one participant, there would
    be an 80% [(100% — prevalence in
    broader older adult population)]11 =
    (100% — 20%)’] probability that we
    would not observe a “minimally common”
    response in our sample. For 2 participants,
    there would be a 64% probability
    (100% — 20%)2 and so on.

    Prof Terry Haines
    Director, Allied Health Research Unit, Monash Health & Monash University

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s