Hello everybody! I use R 1.6.2 in Windows, and have a problem controlling the na.action. In a dataset with twelve trials, one of the trials lack any readings of the variable "STS.SH" (standing power at harvest) Fitting an aov() object with the call: led1t7sts.aov <- aov(STS.SH ~ Trial/Block + Treatment + Treatment:Trial, data = led1t7, na.action=na.exclude) seems to work as it produces an object with 10 df for the factor "Trial". But when I use model.tables or TukeyHSD on the object I get this:> model.tables(led1t7sts.aov, "means")Error in replications(paste("~", paste(names(tables), collapse = "+")), : na.action must be a function I have tried to use "na.action=na.exclude" inside the model.tables call as well, without any bettering. I can naturally cope with the problem by taking the whole trial away from the dataset, but it doesn?t feel very sophisticated...;-) (Prof. Ripley answered a similar question from me two weeks ago. The answer was good but didn?t work as the reason of the error was the same as this time: a whole trial with only na:s in it). Thanks /CG CG Pettersson cg.pettersson at evp.slu.se
This ia already fixed in R-devel. The answer is the same: don't use na.omit implicitly: use it explicitly. On Wed, 26 Feb 2003, CG Pettersson wrote:> Hello everybody! > > I use R 1.6.2 in Windows, and have a problem controlling the na.action. > > In a dataset with twelve trials, one of the trials lack any readings of the variable "STS.SH" (standing power at harvest) > > Fitting an aov() object with the call: > led1t7sts.aov <- aov(STS.SH ~ Trial/Block + Treatment + Treatment:Trial, data = led1t7, na.action=na.exclude) > seems to work as it produces an object with 10 df for the factor "Trial". > > But when I use model.tables or TukeyHSD on the object I get this: > > model.tables(led1t7sts.aov, "means") > Error in replications(paste("~", paste(names(tables), collapse = "+")), : > na.action must be a function > > I have tried to use "na.action=na.exclude" inside the model.tables call as well, without any bettering. > > I can naturally cope with the problem by taking the whole trial away from the dataset, but it doesn?t feel very sophisticated...;-) > (Prof. Ripley answered a similar question from me two weeks ago. The answer was good but didn?t work as the reason of the error was the same as this time: a whole > trial with only na:s in it). > > Thanks > /CG > CG Pettersson > cg.pettersson at evp.slu.se > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > http://www.stat.math.ethz.ch/mailman/listinfo/r-help >-- Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595
In 27/2, I got the following answer from Prof. Ripley: (The question is at the bottom)>This ia already fixed in R-devel. The answer is the same: don't use >na.omit implicitly: use it explicitly.I feel rather stupid for the moment, as I don?t understand an answer that looks very simple. What?s the code to do the trick using na.omit explicitly? (Preferably starting with my code in the question) I can?t get it to work, so my tries are not worth printing here... Thanks /CG On Wed, 26 Feb 2003, CG Pettersson wrote: > Hello everybody!> > I use R 1.6.2 in Windows, and have a problem controlling the na.action. > > In a dataset with twelve trials, one of the trials lack any readings of the variable "STS.SH" (standing power at harvest) > > Fitting an aov() object with the call: > led1t7sts.aov <- aov(STS.SH ~ Trial/Block + Treatment + Treatment:Trial, data = led1t7, na.action=na.exclude) > seems to work as it produces an object with 10 df for the factor "Trial". > > But when I use model.tables or TukeyHSD on the object I get this: > > model.tables(led1t7sts.aov, "means") > Error in replications(paste("~", paste(names(tables), collapse = "+")), : > na.action must be a function > > I have tried to use "na.action=na.exclude" inside the model.tables call as well, without any bettering. > > I can naturally cope with the problem by taking the whole trial away from the dataset, but it doesn?t feel very sophisticated...;-) > (Prof. Ripley answered a similar question from me two weeks ago. The answer was good but didn?t work as the reason of the error was the same as this time: awhole> trial with only na:s in it). > > Thanks > /CG > CG Pettersson > cg.pettersson at evp.slu.seCG Pettersson cg.pettersson at evp.slu.se