reliability/standard error

[International DivisionH]

Basically, I recall reading in my Thorndike textbook many decades ago
that reliability/standard error is less an issue when you are talking
about group data, than when you are talking about an individual score.

In the 2nd edition of Ed. Measurement, Stanley states: A low level of
reliability that would make scores, especially difference scores,
largely meaningless can be tolerated when one is interested in
comparing mean differences. Here, the emphasis should be on the
standard (or the variance) or error of measurement of the mean, not on
the reliability coefficient. The variance error of a mean due to
errors of measurement is 1/nth the variance error of measurement of
the single scores. ...where n is the number of examinees. P. 392

1. Has this changed? If so why?
2. If not, then wouldn't this mean that when looking at large groups
(i.e. several hundred respondents), one could have greater faith in a
smaller set of items? Wouldn't this logic explain why groups like
Harris or Gallup can make assertions based on responses to single
questions?

Thank you for any responses you can provide.