Stéphanie Gascón Garcia
2007-Nov-02 19:38 UTC
[R] lme model with replicates within a random factor
Dear all, I wonder if anyone can help me with specifying a right model for my analysis. I am a beginner to lme methods. I was unfortunately not able to find a solution to my problem on my own. Data structure: I have sampled monthly 6 basins during two hydrological cycles, and I have taken several (2 to 4) samples (“replicate”) for each basin and month. I’m trying to relate Shannon diversity to some environmental variables, taking away pseudoreplication. Thus, I have tried an lme model with “time” and “basin” as random factors. I have put hydrological cycle as a fixed factor, since I’m interested in quantifying their effect. But I’m not sure if there are other possibilities to do this. This is my model structure: Model1<-lme(shannon<-log(chla+1)*hyd_cycle+toc*hyd_cycle, random=~time|basin) Note that I have not put “replicate” within the random structure (random=~time|basin/replicate), and this is my major concern. I’m not sure if I have to include “replicate” in the random structure, since replicates are randomly taken each time, and so are not repeated measures within the basins. All books and internet information that I have looked at the moment are examples with repeated measures within blocks (e.g. repeated measures of the same plant and treatment). Can anyone help me with this? I would appreciate any hints and suggestions Stéphanie [[alternative HTML version deleted]]
1. Consult the Pinheiro and Bates, MIXED EFFECTS MODELS ... book, especially if you will continue to use such methodology as part of your professional work. 2. I think the model you want is: Model1<-lme(shannon ~ log(chla+1)*toc*hyd_cycle,random = ~time|basin) You don't need to include the replicate term because that is automatically included as the error term in the model. Note that this model probably makes most sense if time is a factor.If time is a numeric vector, then the model implies that you expect some kind of random trend with time in each basin, and you would probably therefore want to include it as a fixed effect, too, something like: Model1<-lme(shannon ~ log(chla+1)*toc*hyd_cycle + time,random=~time|basin). This postulates some overall time trend with random slope and intercept offsets from the overall trend in each basin. Bert Gunter Genentech Nonclinical Statistics -----Original Message----- From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of St?phanie Gasc?n Garcia Sent: Friday, November 02, 2007 12:39 PM To: r-help at stat.math.ethz.ch Subject: [R] lme model with replicates within a random factor Dear all, I wonder if anyone can help me with specifying a right model for my analysis. I am a beginner to lme methods. I was unfortunately not able to find a solution to my problem on my own. Data structure: I have sampled monthly 6 basins during two hydrological cycles, and I have taken several (2 to 4) samples (replicate) for each basin and month. Im trying to relate Shannon diversity to some environmental variables, taking away pseudoreplication. Thus, I have tried an lme model with time and basin as random factors. I have put hydrological cycle as a fixed factor, since Im interested in quantifying their effect. But Im not sure if there are other possibilities to do this. This is my model structure: Model1<-lme(shannon<-log(chla+1)*hyd_cycle+toc*hyd_cycle, random=~time|basin) Note that I have not put replicate within the random structure (random=~time|basin/replicate), and this is my major concern. Im not sure if I have to include replicate in the random structure, since replicates are randomly taken each time, and so are not repeated measures within the basins. All books and internet information that I have looked at the moment are examples with repeated measures within blocks (e.g. repeated measures of the same plant and treatment). Can anyone help me with this? I would appreciate any hints and suggestions Stiphanie [[alternative HTML version deleted]]