Picking an appropriate ANOVA, assigning factors

Hi all. I recently performed a simple ANOVA on my data, but I think doing so was not quite right. What I'm attempting to do is (1) determine whether two species have different average stem densities during a point early in the growing season, and during a point late in the growing season, and (2) whether each species has different stem densities early in the season compared to its density late in the season. Essentially, we see one species start at higher density in the early season, and it decreases through midseason, while the other species starts at a lower density and increases through midseason. I would like to know if these dynamics are significant, and if they result in significant differences between species at both time points.

I have measurements of early and late-season stem densities from 4 experimental conditions, each with 6 plots. These measurements were repeated each season from 2005 until 2013. The conditions are ambient precipitation+ambient temperature, drought+ambient temperature, ambient preciptiation+heated, and drought+heated. I would like to test also whether any differences listed above depend on the treatment conditions. In my initial analysis, I simply performed ANOVA separately within each condition, but I do not think this is appropriate either.

I appreciate any guidance you have to give. Thanks!


TS Contributor
This sounds at a minimum like a two factor, repeated measures ANOVA. Depending on how your plots are physically laid out, you may have introduced a split-plot element to your design. If you are interested in including year, you also have early/late nested within year.
Thank you. I do not think I wish to include year as a factor, but simply acknowledge that measurements were taken from the same plots both within and between years. I think you may also be right about split-plot, as the heat treatments are nested within precipitation treatments.


TS Contributor
The reason I mention year as a factor is that by treating year as just a replicate, you will have more variation in your residual error and a less sensitive test. By using year as a factor, you account for that variation, reduce the residual error and have a more sensitive test. Another option would be to block on years. Each are similar to what is done by using a covariate or by appropriately using a paired t-test vs. a 2-sample t-test.