No. Whereas I partly agree that it's Ross ``fault'' trying to
use too smart LaTex (and using outdated \bf instead of \mathbf),
;-)
The bug is really there, since we are talking about the Rd "language",
not LaTeX, an in Rd, \eqn and \deqn are defined to have either
one or two arguments -- where Ross used the 2-argument version
correctly (in principle at least) --> See the manual "Writing R
Extensions".
Forgive me for not reading R-ext carefully, but Ross's Rd code is
still "obviously" wrong in the lights of the two-argument \eqn:
(really doesn't differ from the 1-arg interpretaion of \eqn)
\eqn{{\bf\beta}_j}{\bf\beta}_jnormal-bracket5bracket-normal{b(j)}
In other words,
\eqn{...}{...}_...
and the "_" is still outside of any maths environment, which is most
probably not Ross's intention.
But that is Latex code produced by R, not Rd code produced by Ross.
The bug is in the Latex production (which I think is done by
share/perl/R/Rdconv.pm, but I don't know Perl well enough to attempt
to fix it).