There is a lively debate currently in the HCI community about the statistical tools to be used when analyzing our experimental data. It is recommended to use confidence intervals rather than NHST. See for example : http://dragice.fr/fairstats-last.pdf
I am currently working on the analysis of experimental data from a repeated measures experiment. Are there any statistical tools (R packages) available to perform this type of analysis? Thank you in advance
Translated with www.DeepL.com/Translator
You did not mention for which kind of analyses you want to get confidence intervals, nor did you say anything about the parameters for which you want to have them, e.g. means, differences, effect sizes....
Please elaborate more on your question, so that we are able to help you.
I am comparing 4 interaction techniques. I measured task completion times over 10 repetitions for each participant (20 participants) with each technique. I would like to know if one of the technique leads to shorter completion times.
Oh, sorry. I'd like to use CI on the medians, since it allows to handle outliers efficiently. However, I don't know how to take into account the repetitions. Should I simply calculate a mean value for each participant with each technique?
Thank you for your help!
I dont know, but you do not seem to understand my question. Maybe it is not common for your research area, but if you have a hypothesis (e.g. at least one interaction technique leads to shorter repetition times than the others), you typically use frequentistic or bayesian inferential analyses to estimate if the H0 is likely to be true (all techniques have the same repetition time) or the H1 may be accepted (repetition times differ), instead of descriptive analyses ("calculating a mean value for each participant").
Maybe you should consult a text book to get a basic knowledge in inferential statistics, e.g. Field, A. (2013). Discovering Statistics Using R. Sage Publishing. This is a good book for beginners and will most likely cover the problem you have. I would guess that you could use a 4 (techniques) X 10 (measurements) ANOVA with repeated measures on both factors.
P.S.: Outliers may also be tackled by robust techniques for parametric data (e.g. trimmed means) instead of switching to non-parametric tests, which would massively limit the possibilities for analyzing the data.
Check our repeated measures; there are online tools for calculation of CI.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Human-Computer Interaction