If we assume that a plant growing or not is a simple random event that can happen or not (eg a Bernoulli trial), the number of plants grown in each room is Poisson distributed. What we want to know is the chance of a specific tile getting grown on in a time period.
Let's also assume that plants and trees are all identical, e.g. the only thing we care about is the total number of plants plus trees. Then, we can give confidence intervals around the chance of getting grown on.
Dusting: 14 grown, p=.0014, 95% interval: .0007 - .0023
Small pile: 18 grown, p=.0018, 95% interval: .0011 - .0028
Pile: 19 grown, p=.0019, .0011 - .0030
So far the effect of more than a dusting of mud is clearly not yet significant as the upper bound of the dusting is well above the lower bound of the pile. If the growing is indeed an independent stochastic process for each tile, it should not be needed to repeat the trial, it would be just as good to have a bigger surface. If it is not, the confidence intervals are nonsense anyway :-)
I'm curious in the "after 1 year" results, as increasing the p should also make it a lot easier to get significant results...
ooo ninja'd, results are in. The above was written after the 'galena' data. Lemme read the new data and update :-)
Hmm, the effect is still quite small, from 84 to 94, and obviously not significant yet (confidence interval around .0084 goes up to .0104....). Quick chi squared test confirms this, p value is around .5 (so 50% chance that the result is due to chance).
Edit: must learn to read. Just saw that the totals I quoted are for the trees only. If I add trees and shrubs it is 168 (room 1) vs 202 (room 3), confidence intervals .0144 - .0195 vs .0175 - .0232, chi squared (df=1) is 2.999, p=.083. So, not significant under the silly (or at least arbitrary) convention of requiring .05 but the effect is certainly starting to show...