These patches fix the issues raised by James Stewart regarding GLM regressions and the Gaussian family. Now when family="gaussian" is used in glm, the results from add_{pi, probs, quantile} will appear as if a linear model has been fit.
Since we assume Normally distributed errors, the outputs of add_{pi, probs, quantile}.lm and add_{pi, probs, quantile}.glm agree when the link function is identity, even though the models are fit with two distinct algorithms.
These patches fix the issues raised by James Stewart regarding GLM regressions and the Gaussian family. Now when
family="gaussian"
is used inglm
, the results fromadd_{pi, probs, quantile}
will appear as if a linear model has been fit.Since we assume Normally distributed errors, the outputs of
add_{pi, probs, quantile}.lm
andadd_{pi, probs, quantile}.glm
agree when the link function is identity, even though the models are fit with two distinct algorithms.Closes #14