users of scale = T likely assume this scales the outcome also

[tbates]

Problem Students use scale and assume that this scales the outcome (when it only scales the rhs).

The problem is that the standardized coefficients are then interpreted as what an SD of x causes in SD of y, but the output is actually in raw units of y. I can't see a conventional use for scaling x without scaling y...

Solution you'd like It would be nice to wind back time and have transform.response default to TRUE. But perhaps the output notification could say:

Continuous predictors are mean-centered and scaled by 1 s.d (the output remains in raw units)

i.e., notify users when y is raw. Otherwise they are reporting scaled effects often way bigger than the expected beta.

[jacob-long]

Yes, I'm open to being more explicit that the outcome is not being transformed when it is not. The current wording says nothing about the outcome and so I can see how any user who might be expecting the outcome to be scaled not getting that information from the current message.