The site
My study area is located in a typical Norwegian mountain landscape with a long history of grazing activity. Since year 2000, a still on-going sheep grazing experiment has been run there. In the 2.7 km2 field area (see figure), ranging from 1100-1300 m.a.s.l., nine separate fenced areas (subfields AB.. .I), ordered into three geographically adjacent blocks (ABC - DEF – GHI), were each (in a stratified block wise manner) subjected to either of 3 treatments, i.e. differences in sheep grazing intensity:
1) control (no sheep)
2) low grazing intensity (25 sheep/km2)
3) high grazing intensity (80 sheep/km2)
My hypothesis
The question I ask is simply: Does differences in grazing intensity translate into differences in bumble bee abundance (mediated through grazing induced changes in plant communities or otherwise)?
Data collecting
To find out, I established five 30m transects in each of the nine subfields. I did this by using a semi-random method for site allocation. My field assistant and I went on to register a total of 649 bumble bees during a 10 day period in early July 2014. These 649 observations sum up from the fundamental unit of registration, which is each and every “transect walk”; they make up the response variable in my statistical analysis. Since there were 45 transect and each was visited 15 times, that totals up to 675 “transect walks”.
An infield weather station recorded data to be used in the statistical analysis.
Statistics
I'm using a mixed effects statistical model (glmmADMD with negative binomial probability distribution and BIC model selection criterion) in R to sort out the contributions of different registered explanatory variables on the observed variation in bumble bee abundance between “transect walks”.
Translated into English language, this means that I'm using statistics to help me find out which of many factors (effects/explanatory variables) makes more or fewer bees appear during one transect walk than another. Those factors that significantly contribute to a better understanding of the system, are kept in the final best model, while those that merely complicate matters are left out in the end. Thus the “experimental treatment” effect, the one that will answer my initial question, is just one explanatory variable among many that are treated equally by the model selection criterion.
The investigated explanatory variables are:
“experimental treatment”, “person”, “transect”, “subfield”, “block”, “time of day”, “day number”, “temperature”, “wind speed”, “solar radiation”, “m.a.s.l.”
Results
At this point in the analysis, the final best model contain “transect”, “time of day” and “day number” as the only explanatory variables. This despite the fact that the observed between treatment difference in bumble bee abundance was rather pronounced, with control = 223 bees, low grazing intensity = 260 bees and high grazing intensity = 166 bees (see figure).
The next step now is to try improving the model by investigating the possible addition of interaction between effects.
In the discussion, I will look for biological significance in the absence of statistical evidence.