Which statistical test works for paired and grouped data? I have a small set of samples of plants where I've made some changes to some but not others and want to know if there was any significant effect on the number of leaves as a result of this change. My data looks like this: 

I was wondering what statistical analysis I should use for this? I would usually use a GLMM with condition and timepoint as fixed effects and plantID as a random effect. However, in this case the sample size is quite small so it seems that I should use a non-parametric test, but can't find one that deals with the effect of plantID while also dealing with condition and timepoint. Should I just use GLMM or is something else better?
 A: A famous adage states that sample size is where you randomize.  It seems that you have randomly assigned (?) 4 plants to treatment and 5 plants to control. 
So even if you go for the suggested Wilcoxon's rank sum test on the differences, you would have a fairly low overall sample size of 9 plants. Additionally, it it usually the case that nonparametric tests have less statistical power than parametric tests, meaning that that you are less likely to detect a significant effect of treatment when one truly exists if employing a nonparametric test such as the suggested Wilcoxon's rank sum test. 
The same power considerations would apply if you were to replace your default Poisson mixed effects model with a nonparametric counterpart - you would be less likely to detect a significant effect of treatment when one truly exists.
Contrary to popular belief, nonparametric procedures are not a panacea for a low sample size. Being able to relax assumptions in a nonparametric test comes at a cost (i.e., reduced power).
The best panacea for a low sample size is to actually conduct a formal sample size calculation prior to collecting your data. This way you can feel assured that you have a large enough sample size to detect an effect of the desired magnitude at your chosen power level and significance level. More often than not, studies where no formal sample size calculation was conducted up front end up being under-powered, producing inconclusive findings.
If you are interested in subject-specific effects (with the subject being a plant in your case), I would say your best bet is to fit a mixed effects Poisson model along the lines you suggested and report a confidence interval for the interaction effect. (Given your small sample size, you might choose to report a 90% confidence interval).  Then, after exponentiating the interval, monitor how far the point estimate around which the interval is centered is from 1 and also how far the endpoints of the interval are from 1.  This will enable you to say if it's likely that there is a treatment effect and describe the uncertainty involved in trying to estimate that effect.
