Simpsons paradox and mixed models - large number of groups

[Benjamin]

@cluhmann

I am not sure, I guess I am overlooking something, cf. below where x_hat does not occur. I just noticed this because my editor (PyCharm) notified me about it earlier, but again I might be just overlooking something.

Thanks a lot for helping out here @cluhmann .

@jessegrabowski

mu = (intercept[group_idx]
            + slope[group_idx] * x
            + mx_slope * mx_hat[group_idx]
    )
    sigma = pm.HalfNormal("sigma", sigma=2)
    
    pm.Normal("y", mu=mu, sigma=sigma, observed=data.y, dims="obs_id")