Hello everyone,
As a part of my mixed-methods thesis, I ran a small survey about neurologists' practice on breaking bad news. I'm relatively inexperienced with quantitative studies, so, although I have managed to derive all the relevant descriptive statistics, I was wondering if I could perform any additional analyses to such a small sample size?
I have tried to consult google and several booked but got a bit confused. Do I have to test normality and then maybe perform non-parametric tests? As an example, I would like to look at whether their age, experience, perceived difficulty of breaking bad news etc are associated with their attitudes and practice.
Thank you!
Hello Lefteris,
The answer depends on: (a) the specific variables involved and how they are quantified; and (b) the specific research question each analysis is intended to address. Since this is for your thesis, might I suggest that you confer with your thesis advisor and committee members (if there is a committee), since these are the folks who must approve your work?
Unfortunately, without more specific information (about a, b, above), I can't really offer specific recommendations.
Good luck with your work.
Hello Lefteris,
The answer depends on: (a) the specific variables involved and how they are quantified; and (b) the specific research question each analysis is intended to address. Since this is for your thesis, might I suggest that you confer with your thesis advisor and committee members (if there is a committee), since these are the folks who must approve your work?
Unfortunately, without more specific information (about a, b, above), I can't really offer specific recommendations.
Good luck with your work.
You can find out how large a sample size should be to answer the question.
One helpful link is
https://www.qualtrics.com/experience-management/research/determine-sample-size/
There are a couple of things to consider here. SPSS is usually considered reliable with sample sizes of 30 and above. But you need to match your data and your analysis, for example if your data is categorical you need to use non parametric analysis tools, if you have interval or scaled responses you can use parametric tests. If you use parametric tests your data should be normally distributed which you may find challenging with such a small sample. Again non parametric analysis tools may be more appropriate. But as David Morse mentioned above, without knowing more about your questionnaire and research design, more definite answers are difficult.
You could be able to do group comparisons using T-tests (urban neurologist responses vs rural, etc.., depending on what metadata you have tracked). As mentioned above, you need the minimum number of participants to provide statistical power to the test. (Above mentioned was N=30 for SPSS; I was taught 21 for M-PLUS SEM tests, but for normalization etc as also mentioned, the number I have seen is N=100. There are mathematical tests to learn compute actual numbers.)
Also, if you have spare time, you might explore certain questions that you have no intention to publish about, but that can help guide you in follow-up or future similar studies. I.e., Data from T-test questions without strong statistical power would then be anecdotal. So it would not be reliable inferential knowledge, but you could still use it to get a clearer idea of what such a study would look like, and use it as inspiration to design a statistically powerful survey to investigate it, as part of your private personal brainstorming.
It can also be informative to play around with the data, as if you had enough statistical size, just to see what kind of structure a longitudinal study, for example, might look like if in future you wanted to design one. There is again no inferential power, but it still serves as a rehearsal to become familiar with the statistical tools should you want to investigate properly in the future. That knowledge can help inform research design decisions, so your current research can be used again in the future. Then when you go to seek the assistance of experienced statistical researchers, support from advisers or collaborators, to actually proceed, you then have a better idea of what questions and assistance to ask about.
So there what I am saying is so long as you have an SPSS subscription paid up, to make the most of it!
The major thing is to take the advisor of your supervisor. 49 is above the lower limit for SPSS reliability check. I would suggest that use only non parametric tests by categorically matching data and analysis. For further use of parametric analysis, there is need for me to understand on more missing information as indicated by David and Antje.
Additionally, please remember that repeated analyses of the same small set of data can increase the likelihood of thinking that your results will be statistically significant at your chosen level, as 0.05, where in reality you may need to adjust downward to account for multiple analyses.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size