Interaction terms for an incomplete design R I'm trying to fit a GLM on some data and I feel like there should be an interaction term between two of the explanatory variables (one categorical and one discrete) but all the non-zero instances of the discrete variable occur on the "1" state of the categorical variable (partly why I feel like there should be an interaction). When I put the interaction in the glm (var1*var2), it just shows N/A for the interaction term (var1:var2) in the summary ANOVA.
I have Included a mock example below
a <- data.frame("y" <- c(0,1,2,3),
                "var1" <- c(0,1,1,1),
                "var2" <- c(0,0,1,2))
a.glm <- glm(y ~ var1*var2, family=poisson, data = a)
summary(a.glm)

and then this shows up in the console:
Call:
glm(formula = y ~ var1 * var2, family = poisson, data = a)

Deviance Residuals: 
       1         2         3         4  
-0.00002  -0.08284   0.12401  -0.04870  

Coefficients: (1 not defined because of singularities)
             Estimate Std. Error z value Pr(>|z|)
(Intercept)   -22.303  42247.166    0.00     1.00
var1           22.384  42247.166    0.00     1.00
var2            0.522      0.534    0.98     0.33
var1:var2          NA         NA      NA       NA

(Dispersion parameter for poisson family taken to be 1)

    Null deviance: 4.498681  on 3  degrees of freedom
Residual deviance: 0.024614  on 1  degrees of freedom
AIC: 13.63

Number of Fisher Scoring iterations: 20

This is the table giving the mean of y for each combination in my actual data.
|    |   0   |   1   |   2   |   3   |
| 0  | 1.592 |  N/A  |  N/A  |  N/A  |
| 1  | 1.859 | 1.759 | 1.543 | 0.813 |
|mean| 1.721 | 1.759 | 1.543 | 0.813 |
I'd rather not make var2 categorical as there clearly seems to be a negative correlation between var2 and y which is being overshadowed by the var1 = 0 values. (there are relatively few observations of var2 = 2 and 3 which does not help overcome this effect)
Any help would be appreciated!
Thank you!
 A: One way to deal with this is to set up a new categorical predictor based on the combinations of predictor values for which you have data. A full interaction term between a binary predictor and a 3-level categorical predictor would require fitting 5 coefficients. You only have 4 combinations with values. So you could define a new 4-level categorical predictor $x_{ij}$ with $i$ being the level of var1 and $j$ being the corresponding level of var2. You would define $x_{00}$, $x_{10}$, $x_{11}$ and $x_{12}$ in your example. Then use $x_{ij}$ as the predictor in your model instead of var1 and var2.
That allows you to evaluate the overall association of the predictors with outcome and differences among particular combinations of var1 and var2. That's not the same as a standard interaction term, but it's something you can do with the data you have.
If you want to treat var2 as a linear predictor (with only a few levels), and if the nature of the subject matter is such that you can have non-zero values for var2 only when var1=1 (with var2=0 still possible when var1=1), then you can use the approach on this page. Write your model without the interaction term:
glm(formula = y ~ var1 + var2, family = poisson, data = a)

The Intercept of the model is the estimated outcome (in the log-link scale) for var1=0 (for all of which var2=0 also). The coefficient for var1 is the estimated outcome at var2=0 when var1=1. The coefficient for var2 then evaluates your hypothesis that there is a linear relationship between var2 and outcome.
If there could theoretically be non-zero values of var2 when var1=0 you can still model var2 continuously as above, but then you have to be very careful when interpreting your data. You only have information about the association of var2 with outcome when var1=1, you can't say anything about the association of var2 with outcome when var1=1, and you thus have no information about a possible interaction between var1 and var2.
