I want to do a statisitical test to compare two very large independent samples. I believe a direct comparison using a t-test or ANOVA would not be valid since the sample baselines are different.
I was thinking about using ANCOVA and taking the baseline as a covariate but this baseline would only be relevant to one of the two groups that I want to compare.
Perhaps I should approach the question in the following way:
- How much of the variance of Group 1 is accounted for by the covariate Baseline 1?
- How much of the variance of Group 1 that is not accounted for by the Baseline 1 could be accounted for by the difference in means of Group 1 and Group 2?
I hope the question makes sense but please let me know if further explanation is needed. Any help would be much appreciated!
Hello Eleni,
When I saw your question, I immediately thought that ANCOVA would be the appropriate method to use. I am unsure why you say that the baseline for only one of your groups would be relevant. Wouldn't participants in both groups have a baseline? And that could be fed into the ANCOVA, surely.
In essence, I think you might not have a problem - but maybe I'm not understanding something.
Hello Robert Trevethan ,
Thank you for your reply. The issue I have with using ANCOVA is that the baselines of the two groups differ to the extent that the assumption 'the treatment effect and covariate are independent' is violated.
I have run an ANOVA with the baseline as the outcome and the independent variable as the predictor. The result was significant, which I believe means that the baseline is significantly different for Group 1 and Group 2 and therefore it is not appropriate to use the baseline as a covariate in the analysis.
However, it is only Group 1 that is recieving an intervention, so whilst the baselines of both groups are significantly different, only one group has a different baseline to the outcome i.e.
Baseline 1 -> Group 1 outcome
Baseline 2 = Group 2 outcome
Do you think it would be possible to use 2 x one-way ANOVA, first comparing Group 1 to it's baseline, then comparing Group 1 to Group 2?
I hope I've managed to make the problem a bit clearer but please let me know if not. Any thoughts/suggestions would be much appreciated.
Many thanks!
Hello Eleni Davidson,
I'm sorry, but I can't "picture" your problem. Please feel free to get back with more information concerning the exact nature of your investigation, but in the meantime, I wonder whether looking at some YouTube presentations about ANCOVA might give you some reassurance about using that kind of analysis. Not all of those YouTube presentations are good, but from memory one by Andy Field is pretty helpful.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size