Our study is looking at annual plant
Biomass (ANPP) across several
sites and years.
Each site has four
treatment plots (N, P, NP, C) and within each treatment plot is a blocking factor called
patchtype (whether the biomass was collected under a plant or next to a plant). We only want to compare interplant patchtype with other interplant and only want to compare underplant patchtype with other underplant. EX: N plot underplant with p plot underplant.
We have 6 years of data and would like to account for precipitation (as a covariate). We want to know if sites are varying across treatment and across year within treatment.
I was advised to use a correlation structure of AR1 to account for repeated measures of
year. However, I am not confident in how to structure the model, whether i need to do anything else to account for time.
- I only have one replicate at each site (n=15 sites) when you break down the dataset by year->site->treatment->patchtype.
- patchtype is nested in treatment, but i am not sure if i should treat this nested term as a fixed or random factor (i think fixed).
- Am i structuring the covariate of
rainfallcorrectly by simply putting it as the first predicting variable in my model? The goal here is to ask, is biomass varying across and/or within treatment across year when you account for variation caused by rainfall (known to control lover 70% of biomass variability)?
- Do i also need a nested term for treatment within year to test for how each biomass under each treatment may be varying across years?
- Should i be using ML or REML method?
model <- lme(Biomass ~ rainfall+year*treatment/patchtype, random = ~1 | sites, data = mydata, correlation = corAR1(), method = "REML")
About the data:
Year: 2008, 2009, 2010, 2013, 2015, and 2016
Treatments: N, P, NP, C (nutrient additions)
Patchtype: Interplant and Underplant
Biomass: ANPP in grams
Rainfall: in cm, unique for each site and each year (not different between treatment plots)
Thanks for the help!