In Section 2.2 we commented on the importance of having a sufficiently large range in luminosity to detect and study the BE. Selection effects and intrinsic scatter in the properties of quasar spectra can mask the appearance of the BE. We also mentioned that the KRK study provided some of the most conclusive evidence for the BE because their sample spanned ~ 7 orders of magnitude in luminosity.
The fact that the BE is seen over such a large luminosity range has
theoretical implications, in addition to observational uses. This
finding provides another challenge to models that rely solely on disk
inclination effects or other sources of continuum beaming to produce a
BE (KRK;
Francis 1993);
these scenarios can generally reproduce a
BE-like correlation over only 1-2 orders of magnitude in L.
Variability was suggested early on as the source of the BE
(Murdoch 1983),
but the typical amplitude of variability seen to date in Seyfert nuclei
and normal QSOs is 
1
order of magnitude, rather than 7, and
when AGNs do vary, they appear to trace out a locus in the
W
-
L plan that is distinct from that of the ensemble BE
(KRK;
see also Section 5).
Given the existence of large scatter in the Baldwin relations, is it possible to understand the origins of this dispersion? Inclination or other beaming phenomena, if not responsible for the BE, may provide a source of scatter in the correlation (Netzer et al. 1992); however, this effect is unlikely to dominate the scatter, since predictions of the resulting equivalent width distributions appear to be inconsistent with the observations (Francis 1993). Variability clearly is a source of scatter in the BE, at least at low luminosities. Korista et al. (1998) have recently suggested that metallicity is another probable source of scatter in the overall trend. This last hypothesis may be testable via use of the N V line strength as a crude indicator of abundances within the broad-line media.