Dear all, I'm quite sure that this is a stupid question, but I'll ask anyway. I want to perform an ANCOVA with two continuous factors and three categorical factors. Plant population growth rate (GR) = dependent variable Seed reduction due to herbivory (SR) = continuous explanatory variable Herbivore species (HS, 2 levels) = categorical explanatory variable Population (Pop, 24 levels) = categorical explanatory variable Population size (Popsize) = continuous explanatory variable Year (Year, 16 levels) = categorical explanatory variable My model is technically simple: model<-aov(GR~SR*HS*Pop*Popsize*Year) However, R is not returning any F and P values ? only Df, Sum Sq and Mean Sq. I have to remove either Year or Pop in order to get the test statistics. Why is this? Thank you in advance! Johan A. Stenberg, Umea University, Sweden
On Thu, 2007-11-15 at 16:36 +0100, Johan A. Stenberg wrote:> Dear all, > > I'm quite sure that this is a stupid question, but I'll ask anyway. > I want to perform an ANCOVA with two continuous factors and three > categorical factors. > > Plant population growth rate (GR) = dependent variable > Seed reduction due to herbivory (SR) = continuous explanatory variable > Herbivore species (HS, 2 levels) = categorical explanatory variable > Population (Pop, 24 levels) = categorical explanatory variable > Population size (Popsize) = continuous explanatory variable > Year (Year, 16 levels) = categorical explanatory variable > > My model is technically simple: > > model<-aov(GR~SR*HS*Pop*Popsize*Year) > > However, R is not returning any F and P values ? only Df, Sum Sq and > Mean Sq. I have to remove either Year or Pop in order to get the test > statistics. Why is this? > > Thank you in advance! > Johan A. Stenberg, Umea University, SwedenSee ?summary.aov which is referenced in the "See Also" section of ?aov and is used in the examples therein. HTH, Marc Schwartz
On Nov 15, 2007 4:36 PM, Johan A. Stenberg <johan.stenberg at emg.umu.se> wrote:> Dear all, > > I'm quite sure that this is a stupid question, but I'll ask anyway. > I want to perform an ANCOVA with two continuous factors and three > categorical factors. > > Plant population growth rate (GR) = dependent variable > Seed reduction due to herbivory (SR) = continuous explanatory variable > Herbivore species (HS, 2 levels) = categorical explanatory variable > Population (Pop, 24 levels) = categorical explanatory variable > Population size (Popsize) = continuous explanatory variable > Year (Year, 16 levels) = categorical explanatory variable > > My model is technically simple: > > model<-aov(GR~SR*HS*Pop*Popsize*Year) > > However, R is not returning any F and P values ? only Df, Sum Sq and > Mean Sq. I have to remove either Year or Pop in order to get the test > statistics. Why is this? > > Thank you in advance! > Johan A. Stenberg, Umea University, Sweden >Hi, How much data do you have? a model on the data with all interactions included ,as in your example, requires estimating well over 500 parameters. Even the largest data sets might be strained by this. Supposedly, that's why the aov doesn't give p-values. (though since you don't give a reproducible example, I can't be certain) I'd suggest reducing the number of interactions. I'm fairly certain that you don't want 4- or 5-way interactions for example-they tend to be hard to interpret. /Gustaf -- Gustaf Rydevik, M.Sci. tel: +46(0)703 051 451 address:Essingetorget 40,112 66 Stockholm, SE skype:gustaf_rydevik
Your model is fully saturated. It specifies terms that use up all degrees of freedom. There are no degrees of freedom left over for a Residual term and therefore there is no denominator for the tests. When you drop one term, then those degrees of freedom are left over, that is they form the Residual, and are used as the denominator for the tests. The usual practice is to suppress the high-order interactions, in your example by model <- aov(GR ~ SR*HS*Pop*Popsize*Year - SR:HS:Pop:Popsize:Year) Please use spaces around the arrow, tilde, and + and - signs for legibility. Rich