Hi, I am new to SAS with limited stats knowledge. I recorded bats at 4 different types of habitats and I visited the same four types of habiats at 5 different sites. I wanted to know whether bat counts differed among habitat types so my model was:
class habitat site;
model bats=habitat site habitat*site; dist=poisson type3 scale=deviance;
and the data was input as 3 comlumns: bats site habitat
My output is below and my question is:
WHY ARE THERE BLANKS IN THE OUTPUT?
I have read the SAS manual and it really wasn't helpful, it just mentions aliasing but does not say what I am doing wrong or if it is ok to have the blanks. Does it mean anything in terms of interpreting the data?
By "blanks" do you mean the periods "." that indicate missing values and are associated with df=0 and Estimate=0?
If so, then those are not indicative of a problem per se (although you do, I believe, have a problem with your model specification, which I'll address below). The short answer is that by default GENMOD includes an intercept and uses GLM (reference-cell) coding where the last level of each factor is the reference. The parameter for the reference level (e.g., habitat 4) is *set to zero*; thus, the parameter estimate has SE=0, and the other statistics fall out accordingly. See the "Parameterization Used in PROC GENMOD" section in the SAS/STAT GENMOD documentation for a wealth of useful details.
However, I think you still have a problem: You have 20 observations. The MODEL statement instructs GENMOD to estimate the following parameters:
20 parameters total (as listed in your output)
which leaves you with zero df for any stat tests.
To obtain a significance test of whether means are equal for all levels of some fixed-effects factor (like HABITAT), you need appropriate replications of HABITAT. I might guess that you are thinking of SITEs as replications of HABITAT. If so, then your model statement should look more like
model bats=habitat / dist=poisson type3 scale=deviance;
Something else to think about is whether your count data follow the Poisson distribution well. For the Poisson, the mean is equal to the variance; often counts have variance greater than the mean, which is known as over-dispersion. Sometimes a negative binomial is appropriate. Sometimes the counts are high enough that a normal distribution fits well. Sometimes there are too many zeros ("zero-inflated"). Counts are not always Poisson; you just have to dive in and find out.