Appropriate Statistics for Plant Study?

#1
This post is lengthy, but I hope someone can take the time to look it over and offer some suggestions. Thanks in advance.

A colleague has asked for some advice on a statistical approach to a study which she inherited at her place of work. The study was intended to examine the effect of mycorrhizae inoculation on plant growth. Following is the study design:

Twenty (20) plants per species (14 species) were randomly assigned to each of the group's combination of treatments. The combination of treatments were:

1. drought conditions, inoculated, full fertilizer
2. drought conditions, non-inoculuated, full fertilizer
3. permanent flooding, inoculated, full fertilizer
4. permanent flooding, non-inoculated, full fertilizer
5. seasonal flooding, inoculated, full fertilizer
6. seasonal flooding, non-inoculated, full fertilizer
7. mesic conditions, inoculated, full fertilizer
8. mesic conditions, non-inoculated, full fertilizer
9. mesic conditions, inoculated, half fertilizer
10. mesic conditions, non-inoculated, half fertilizer
11. mesic conditions, inoculated, no fertilizer
12. mesic conditions, non-inoculated, no fertilizer

The specific objectives of this study were:

1)Look at the effect of mycorrhizal inoculation on the growth and survival of wetland woody plants when plants are subjected to seasonal flooding (mid-Jan through mid-April) and permanent flooding (mid-Jan through the end of October);

2)Assess the benefits of mycorrhizal inoculation on the growth and survival of wetland woody plants when plants are subjected to mesic conditions (under irrigation) and varying levels of fertilizer;

3) Determine the benefits of mycorrhizal inoculation to wetland trees and shrubs when plants are subjected to drought (no irrigation)

The hypothesis tested in this study was that mycorrhizae have a positive effect on the growth and survival of native woody wetland plants when the study plants are subjected to stressful conditions (manipulation of inundation and fertilizer levels). It was expected that those plants treated with mycorrhizal fungi would have better survival and better growth (height) than the non-inoculated plants when subjected to the same conditions of watering and fertilizer.

Naive statistical person that I am, my first thought was that we should handle each species independently of the other 13. Otherwise, we’d be comparing the proverbial apples to oranges. Then, since the objective is to examine the effects of inoculation, the non-inoculated group would serve as the control for each set of conditions and we could simply do t-tests between non-inoculated and inoculated; i.e. drought inoculated vs. drought non-inoculated; flooded inoculated vs. flooded non-inoculated. Another colleague, more experienced in statistics than I, cautioned that this could lead to flagging false significances, and suggested an approach that would utilize a multiple comparison statistic, such as Dunnett’s. My feeling is that since this study was not set up as a typical toxicology or effects study (e.g., different dose levels of a common treatment), that might not be the best approach either. Comments/suggestions?
 

JohnM

TS Contributor
#2
As I mentioned in a previous post, if your "a priori" theories / hypotheses call for a study involving many comparisons, then do them - you're basing the study on a scientific theory in your field, and not just doing a bunch of statistical tests in the hopes of finding a few significant results ("fishing expedition").

If the theory is specified and formulated prior to the study and data collection, then there is little concern over an increasing Type I error rate. Along with the individual results of "significant" / "non-significant", also look at the overall pattern of results.

Good luck!