Have an experimental design with four time points and 6 treatments (one control), sample size =4/group. I have an outcome (DV) and two IVs - time and dose. For the first time point (0 days), thera are only the data for controls (so 4 values). At the other time points, 24 values. In paralel, I have another variable (also dependent of time and dose) which I intend to use it as an covariate. Any ideas if this design can be used for a two-way Ancova.
Is it the same point at most of the time points, or is it different people at each time point? Either way, with just 4 values at the first time point it may be wise to focus on the other time points (and if it is the same people, exclude those 4 people).
Yes, I have thought... It is a between subjects compariosn so different specimens at each time point.
If it is different specimens then I recommend not using the time with just 4 values since that does not provide much information. However, is that time point important for your theoryof the relationship between time, dose, and outcome? Why are there so many fewer in this condition. Was there a large "dropout"?
No, it was the time point 0, so it was taken into account just as some kind of an additional benchmark for the controls at the other points.
So you have three predictor variable (two categorical and I assume the covariate is continuous, but that really isn't a big thing). What is the outcome variable like (e.g., is it a proportion, a count, or something else, and what is the distribution like [though the important distribution assumptions are of the residuals). Assuming the assumptions of a linear regression are acceptable, then you might just regress the response variable onto these three (and perhaps their interactions, and if the covariate is continuous perhaps allowing a flexible curve to be used). Do you have theories about how the covariate may interact with the categorical variables?
Yes this was the idea. In priciple the two IVs should influence both DV and the predictor, and there is some correlation between the predictor and DV - my assumption is that there is some interaction between predictor and DV, most probably with the predictor having a stronger influence. The correlation between them is 0.57.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size