Dennis, Thanks for the response. For clarification, the treatments are assigned to tanks at random. For example, all 18 tanks were randomly assigned a nutrient treatment (1, 2, or 3) and randomly assigned a clip treatment (a, b, or c). That gives 9 combinations, each replicated once in a total of 18 tanks. After trying the model you suggest (m1 <- lme(ch4 ~ t * spp * nut * clip, random = ~ 1 | tank/nut * clip/spp,
data = ch4data) I get another error - "Error in getGroups.data.frame(dataMix, groups) : Invalid formula for groups"
If I run the model without specifying nut*clip nested within tank, the model runs leaving me to still suspect that I am still unclear how to specify my nesting correctly. This is the model that runs - ch4.model <- lme(ch4 ~ t * spp * nut * clip, random = ~ 1 | tank, data = ch4data) I will look through Pinheiro and Bates (2000), thanks for pointing me to it. Anthony
On Fri, Aug 31, 2012 at 6:50 PM, Dennis Murphy <djmuser at gmail.com> wrote:
Hi: It's not clear to me how nut and clip are assigned. Are two tanks assigned to each nut/clip combination at random? If so, then the df breakdown in the ANOVA should look something like Whole plot: (tank level) nut 2 clip 2 nut * clip 4 Error(WP) 9 Split plot: (section level) variety 3 variety * nut 6 variety* clip 6 3fi 12 Error (SP) 27 Split-split plot: (occasion level) Occasion 1 (Cross occasion with everything above, same degrees of freedom) Occasion is equivalent to a split-split plot factor when the #occasions = 2, since you can only get a compound symmetric covariance structure in that case. That's how I'd analyze it, something like m1 <- lme(ch4 ~ t * spp * nut * clip, random = ~ 1 | tank/nut * clip/spp, data = ch4data) For reference, see the split-plot example starting on p.45 of Pinheiro and Bates (2000). I'm not sure about the correctness of the random effects specification, but check out the example to see how split-plot designs are programmed in lme(). I'd try it out myself, but no data were provided... HTH, Dennis On Fri, Aug 31, 2012 at 3:54 PM, Anthony Rietl <anthony.rietl at gmail.com> wrote:
Tom, Thanks for your input. My treatments (nut & clip) were applied to the entire tank and all 4 species within that tank, making it a true split plot. Each treatment combination also had a replicate - for example, 2 tanks would have the treatment combination of 1(nut) and a(clip), and so on for other treatments. I was not aware that spp*nut*clip*t would give me main effect, 2-way, and 3-way interactions. Thanks for pointing that out. However, if I run the model with spp*nut*clip*t, removing all of the other interactions, I still get the same error regarding singularity, making me think that the nested term is still a problem. Thanks, Anthony On Fri, Aug 31, 2012 at 5:29 PM, <Tom_Philippi at nps.gov> wrote:
Anthony-- You did not specify whether your nutrient and clipping treatments are applied to entire tanks, quadrants within tanks, or individual plants within quadrants, so I don't think anyone can suggest the proper model specification. Nonetheless, I suspect that the error message is because your formula specifies each main effect several times. Note that spp*nut*clip*t is not just the 4-way interaction, but each main effect, 2-way, and 3-way interaction. spp:nut:clip:t would be the 4-way interaction. See the documentation for formula, or any of relevant resources listed at: http://cran.r-project.org/other-docs.html Tom Anthony Rietl <anthony.rietl at gm ail.com> To Sent by: r-sig-mixed-models at r-project.org r-sig-mixed-model cc s-bounces at r-proje ct.org Subject [R-sig-ME] Split plot design with repeated measures - model errors 08/31/2012 05:12 (nlme) PM EST Forgive me if this has been answered, but after extensive searching I have found nothing to help me. Here is a brief overview of the experiment: -I have 18 tanks, each of which are split into 4 sections, each section has a different specie of wetland plant. Tank is my whole plot factor. Variety, or species (spp), is my split plot factor. -I have 2 treatments, each at three levels. Nutrient addition (nut) and clipping (clip) -I am measuring methane (ch4) emission as my response variable. I measured methane on 2 occasions, represented by variable t for time. Data Structure
str(ch4data)
'data.frame': 144 obs. of 6 variables: $ tank: Factor w/ 18 levels "1","2","3","4",..: 1 1 1 1 2 2 2 2 3 3 ... $ spp : Factor w/ 4 levels "Emac","Ewal",..: 4 2 3 1 3 4 1 2 1 4 ... $ nut : Factor w/ 3 levels "1","2","3": 2 2 2 2 1 1 1 1 3 3 ... $ clip: Factor w/ 3 levels "a","b","c": 3 3 3 3 2 2 2 2 2 2 ... $ t: Factor w/ 2 levels "1","2": 1 1 1 1 1 1 1 1 1 1 ... $ ch4 : num 0.382 1.642 1.529 0.245 11.482 ... Model ch4.model = lme(fixed = ch4 ~ spp+nut+clip+t + spp*nut+spp*clip+spp*t + nut*clip+nut*t + clip*t + spp*nut*clip + spp*nut*clip*t +(nut*clip)%in%tank, random= ~1|tank, data=ch4data) this gives the following error Error in MEEM(object, conLin, control$niterEM) : Singularity in backsolve at level 0, block 1 Specifically, my questions are as follows: What is wrong with the fit of my model and what does this error mean? What in my model is singular? Have I over specified the model? I think the problem is with nesting the nut*clip interaction within tank, but I'm not sure. Additional help with fitting this model correctly would be much appreciated. I am new to R, so please take that into account. Thanks, Anthony
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
-- Anthony Rietl Graduate Student Louisiana State University Renewable Natural Resources Baton Rouge, LA 70803 Office: (225) 578-1540
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
Anthony Rietl Graduate Student Louisiana State University Renewable Natural Resources Baton Rouge, LA 70803 Office: (225) 578-1540