Checking if a nominal variable is important in a GLM model Is it possible to put a nominal explanatory variable as a continuous variable in a GLM in R (instead of telling R that its a factor with different levels) to see if this variable (treatment) is a significant driver of the response variable that we are interested in before looking at the levels?
Thanks!
 A: Treating a nominal variable as continuous is never a good idea because a simple relabeling of the numerical value of the levels can lead to different estimates and different hypothesis test decisions, which is a very undesirable property. 
It sounds like you want to know how to test the significance of a nominal variable without examining the difference between levels of the nominal variable - since this is a GLM (fit by maximum likelihood) you can do this with the likelihood ratio test. Suppose you have a number of 'other' covariates $X_{1}, ..., X_{p}$ and an $m$-level categorical variable with dummy variables $B_{1}, ..., B_{m-1}$ indicating which level (excluding the reference level) of the nominal variable you are in (in R, these variables are generated automatically and incorporated in the fit if the predictor is a factor). So, then you fit two models 
$$ Y_{i} = \beta_{0} + \sum_{j=1}^{p} X_{ij} \beta_{j} + \sum_{k=1}^{m-1} B_{k} \alpha_{k} + \varepsilon_{i} $$ 
and 
$$ Y_{i} = \beta_{0} + \sum_{j=1}^{p} X_{ij} \beta_{j} + \varepsilon_{i} $$ 
where the second model reflects the belief that the nominal predictor has no effect (irrespective of the effect of particular levels). Extract the maximized log-likelihoods from the first and second model $L_{1}, L_{0}$, respectively, (using, for example the logLik() function in R) and calculate
$$ 2 \Big( L_{1} - L_{0} \Big) $$ 
which has an approximate $\chi^{2}_{m-1}$ distribution, which you can use for significance testing.  
