I've trawled the web for a day now with no success in finding a guide for a within-subjects ANCOVA; there seems to only be guides for between subjects or mixed analyses out there.
My design is:
2 within conditions - anger memory vs resentment memory
There are higher ratings of envy in the resentment memory condition and I believe it is messing with my paired t-test results on the DVs, so I want to control for envy.
I am highly uncertain I've input things right and I am even more confused how to read the output, so I was hoping someone could point me to a book since nothing is on the web. A. Field and Howitt/Cramer doesn't have it in their books.
https://www4.stat.ncsu.edu/~davidian/st732/notes.html
The early chapters of Prof Davidian's notes should help. The basic model for repeated measure ANCOVA is
Y= a+(post-pre)*b + rest of the model.
Hope this helps. Best wishes, David Booth
Apply the univariate repeated measures option and enter "envy" as your covariate. Your two levels of the within-subjects DV should appear in two columns, respectively. Enter them as fixed factors. When reading the output, check for violation of sphericity. If it is not violated, use the outputs from the within-subjects effects table. Note the presence of the covariate effect posed by "envy", if any. The within-subjects effect would have been corrected for the covariate effect, regardless of the significance of the latter. You need to report BOTH the covariate and the within-subject effects.
Hello Anki,
A proposed covariate is measured before time of the treatment/stimulus condition (or the collection of the responses), otherwise there is a confounding that will impair your ability to make sense of the results as if they were from a standard covariate analysis. Your question makes it sound as if the envy scores were collected at the time of the responses under the two conditions. That won't work (well). If I've misunderstood your design, I apologize.
Good luck with your work!
No, that is correct, David - all data was collected on one occasion (participants recalled & rated the responses they had in a past anger vs resentment situation). I am not sure I understand why this would cause a confound? And, if so, is there any other analysis I should use? Ultimately I am really just trying to remove a disproportionate influence of envy in one condition (resentment), to re-check if my current t-test results will hold or change.
I don't think that adding envy as a simple covariate will be useful. Envy is (I think) a single score for each participant - if so it is completely confounded with the subject term in a within subjects design. Thus adding it as a simple covariate has no impact on the F tests for the within-factors and interaction.
However, you could look at whether the envy covariate interacts with the two factors you have. That is does envy have a disproportionate effect between conditions. If you do this analysis you need to center envy before adding it as a covariate and before compute product terms to test interactions. Some software gives weird results if covariates are not centre (and in any case covariate effects are usually interpreted at their mean value). If using SPSS I would create the centered covariate and product terms using compute before enter them into your model.
Hello Anki,
I also have the same confusion about explaining the output of mixed between-whithin ANCOVA. I don't know if it necessary to run twice ANOVA.
" Run the first model with the covariates, but only report the between-subjects portion of that analysis. Run the second model without the covariates, but only report the within-subjects (WS) portion of that model. "
http://www-01.ibm.com/support/docview.wss?uid=swg21477023
Have you find the guide book or the answer of the question?
Hello, Fan! Ha, no - in the end my supervisor advised me to do a single level multi-level analysis (MLA) instead to control for envy. This meant creating a separate SPSS data sheet in which the participants become a variable & the whole analysis was done on syntax. It is easy to do but complex to understand and (tbh) I very much relied on my supervisor's guidance.
I will send his lecture slides to you. :-)
Thank you so much Anki. I've read the slides you send to me, but it's really not easy for me to understand completely. Considering the limitation of small sample and the correlation results in my study, i still chose to do ANOVA. I do want to control some variables such as gender or stage of disease, but it seems that the covariate must be continuous in ANCOVA. i'm not sure whether this requirement is also necessary for mixed between ANOVA.
Thanks again for your help!
Hello Anki! I have an analysis very similar to yours (2 moments), but I have 3 covariates. Could you please send me by PM the materials for the multilevel analysis? Thank you!
Anki Wikman Hi Anki, I also have a similar experiment could you please send me the materials for the multilevel analysis. Thank You!
Anki Wikman Hi Anki, I'm working on an analysis that I think could benefit from a very similar solution to the one you found - it would be great to have a look at the slides you used if possible?
- Badges
- Science topic
- Similar topics
- Books