I would appreciate if you could help me choose different R packages and writing model to analyze what I believe to be unequally spaced repeated measurement data. My data consists of data on PAR (photosinthetically active radiation) -average and maximum daily PAR, collected successively every 10 min with data loggers. Sensors were positioned above different mulches (sub factor in three levels) with differently fertilized plant canopy (main factor in 4 levels), so there are two different factors influencing AVG and MAX PAR. Also since sensors weren't positioned in repetitions I divided data in every 10 min, 15 min and 25 min sequence for AVG and MAX for each sensor. So the experiment is two-factorial (fertilizers (4) and mulch (3) ) but has time dimension too. More like a split-split plot. What kind of package should I choose, longitudinal? And how to write a model ? I'm so very sorry for my poor knowledge on statistics, I'm not a statistician. I' would be very grateful if you could guide me to the write way to handle my data and to write a model. Sincerely, Josipa Perkovi? Dipl.ing.agr Science novice Institute for Agriculture and Tourism, Porec Croatia
help on choosin right model for data
2 messages · Josipa Perković, Ben Bolker
7 days later
Josipa Perkovi? <josipa at ...> writes:
I would appreciate if you could help me choose different R packages and writing model to analyze what I believe to be unequally spaced repeated measurement data. My data consists of data on PAR (photosinthetically active radiation) -average and maximum daily PAR, collected successively every 10 min with data loggers. Sensors were positioned above different mulches (sub factor in three levels) with differently fertilized plant canopy (main factor in 4 levels), so there are two different factors influencing AVG and MAX PAR. Also since sensors weren't positioned in repetitions I divided data in every 10 min, 15 min and 25 min sequence for AVG and MAX for each sensor. So the experiment is two-factorial (fertilizers (4) and mulch (3) ) but has time dimension too.
Are there multiple sensors within each mulch*fertilizer combination? If so, I would suggest starting with something lme, and using something like avg_PAR ~ mulch*fertilizer , random = ~1|sensorID, correlation=corAR1(form=~1|time) Don't know how much of the day your data covers, but you might want a term like cos(2*pi*time)+sin(2*pi*time) or ns(time,5) [a 5-knot spline; you'll need library(splines) first where time is fraction-of-day [0,1] in the first case, scaling doesn't matter in the second case. You might also need to consider a long-term time trend (linear or quadratic or spline). You should definitely graph your data. Pinheiro and Bates (2000) might be useful. Ben Bolker