Constant variables in a bambi model

1

It seems bambi does not allow constant variables as input:

df = pd.DataFrame({"y": [1,3,5,8],
                   "x1": [25,8,6,9],
                   "x2": [5,5,5,5]})

model = bmb.Model(
    formula="y ~ x1 + x2", 
    data=df, 
    noncentered=True
).fit()

ValueError: The term ‘x2’ is constant!

Is there a way to make it ignore this check for constants?
IMO a warning would be fine but it should still compute the model (brms allows this).

The workaround I found so far was adding small amounts of noise to x2.

1 Like
2

CC @tcapretto

3

Hi @Harryg, it’s not possible to do that for now. But I don’t think it’s a good idea to have a constant predictor, it will be confounded with the intercept. What is the reason for having a constant term?

1 Like
4

Thanks for your reply @tcapretto . My use case is that I need to apply a multivariate mixture prior to my linear model, derived from the posterior of an earlier model with a different dataset. The earlier model lacks one variable, so I’ll introduce a fixed dummy variable to make this prior transferable to the new model.
It’s probably a very unique use case, but maybe one can imagine a scenario where you have a constant variable in your data but want to define a very strong prior shifting this constant in a certain direction.