[R-sig-ME] Removing random intercepts before random slopes

[Maarten Jung]

> Now, if you're only interest in the marginal model, that is in the fixed effects and their standard errors, it is perfectly fine to exclude the random intercepts even if you have slopes in your model, because you just see it as a particular (perhaps more parsimonious) choice for your marginal covariance matrix.

This seems to be an interesting point: If one defines the covariance
matrix as in [1], the off-diagonal elements of this matrix should be
zero for a model without correlation parameters (by which I mean the
|| syntax) and without random intercepts, i.e. just (uncorrelated)
random slopes: (0 + c1 + c2 || Worker). This is, of course, a more
parsimonious model compared to
(1 + c1 + c2 || Worker); however, from a conceptual point of view it
does not look like an appropriate model for most situations I can
think of.
On the other hand (as Jake pointed out and I agree with that) this
model should still be OK if one is only interested in the fixed
effects, their SEs and t values.

[1] https://stats.stackexchange.com/a/348572/136579