Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~pgreen/Papers/wea98.ps Äàòà èçìåíåíèÿ: Tue Oct 20 17:30:22 1998 Äàòà èíäåêñèðîâàíèÿ: Mon Oct 1 23:58:29 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: http astrokuban.info astrokuban

An investigation of the relation between the spectral energy
distributions and the emission lines in low­redshift quasars. 1
Belinda J. Wilkes, Joanna Kuraszkiewicz 2 , Paul J. Green,
Smita Mathur and Jonathan C. McDowell
Harvard­Smithsonian Center for Astrophysics, Cambridge, MA 02138
Received ; accepted
To appear in ApJ
1 Observations reported here were obtained at the Multiple Mirror Telescope Observatory,
a facility operated jointly by the University of Arizona and the Smithsonian Institution.
2 also N. Copernicus Astronomical Center, ul. Bartycka 18, 00­716 Warsaw, Poland

-- 2 --
ABSTRACT
We investigate the relations between the observed emission line strengths,
widths and continuum properties of a sample of 41 low redshift (z ! 1) quasars
for which contemporaneous IR­soft­X­ray spectral energy distributions (SEDs)
are available. This includes investigating correlations between optical/UV lines
with both the luminosity and the shape of the quasars' continuum, as well as
correlations between the various lines. The sample is heterogeneous, primarily
selected upon the existence of good quality Einstein X­ray data and includes 18
radio­loud and 23 radio­quiet quasars.
We find anticorrelations between the equivalent­width (EW) and various
UV luminosities (the Baldwin effect) for the Lyff and Hfi lines and a marginal
anticorrelation for CIII]. Exclusion of narrow line, low luminosity AGN
reveals a significant Baldwin effect for the CIV and CIII] lines. A significant
anticorrelation of EW(CIV) with ff ox is also present. We find no correlations
between any lines and the X­ray luminosity or X­ray slope. The FeII optical
multiplet shows no simple relationship with luminosity or continuum slope,
however there is a tendency for objects with flat X­ray spectra and/or strong
X­ray luminosities to have weak FeII.
Our data do not favor a model in which changes in continuum shape (due to
e.g. a decreasing ionization parameter) cause the Baldwin effect. The data can
instead be explained by an accretion disk (AD) model in which limb darkening
and the projected surface area of an optically thick, geometrically thin disk
combine to cause a viewing­angle dependent UV luminosity and a more isotropic
X­ray luminosity. The scatter in our correlations is larger than that expected
from this AD model, suggesting the presence of dust, which reddens both
the continuum and the broad emission lines. The CIV and CIII] carbon lines

-- 3 --
show flatter slopes and larger scatter in the line ­ continuum relations than
predicted by the AD + dusty torus model. This may be due to a significant
contribution from collisional excitation which is not directly related to the
ionising continuum.
Subject headings: quasars:general --- quasars:emission lines

-- 4 --
1. Introduction
The strong, broad emission lines which characterize quasar spectra are generally
believed to be generated in a large number of small gas clouds photoionized by the central
continuum source of a quasar. This region is known as the broad emission line region
(BELR). To date photoionization models have been reasonably successful in predicting the
average emission line properties of a quasar using an average continuum shape. However,
it has become clear that, while the emission line properties are largely similar from quasar
to quasar, the observed spectral energy distributions (SEDs) are not (Elvis et al. 1994,
hereafter E94). If photoionization models are generally applicable, we would expect
systematic relations between the observed lines and continuum in different objects (Krolik
& Kallman 1988) and, at first glance at least, the dichotomy between continuum and the
line behavior looks surprising.
The most enduring relation between the lines and continua in quasars is the inverse
correlation between line equivalent width (EW) and continuum luminosity known as the
Baldwin effect (Baldwin 1977). This relation is not predicted in a simple photoionization
scenario and explanations range from luminosity­dependent continuum shape, which affects
the BELR ionization, to geometric effects. Another intriguing set of correlations, which
have caused a flurry of activity in recent years, centers around an inverse relation between
the strengths of the broad FeII–4570 and narrow [OIII]–5007 lines and includes soft X­ray
slope and luminosity and broad line width (Boroson and Green 1992, Boller, Brandt and
Fink 1996, Lawrence et al. 1997). Again a simple photoionization scenario fails. The
current popular explanation involves the mass flow into the central black hole, this is
attractive since it would imply an observational link into the central power house (Pounds,
Done and Osbourne 1996). Despite the general success of photoionization models, these
results suggest that other factors also play a role in determining the emission line properties

-- 5 --
of quasars.
Most studies of line and continuum properties of quasars to date have centered around
large samples, often including non­uniform datasets from a variety of sources and generally
utilising global continuum parameters such as L opt , LX and ff ox . Since the photoionization
of any individual ion and its resulting relation with the ionizing continuum is unique to that
ion, the use of global parameters has obvious limitations for investigating photoionization
models. The current study takes a different, complementary approach in concentrating
on uniform, high­quality continuum and emission line data for a small sample of quasars.
This dataset facilitates a study of the relation of each emission line to various parts of the
continuum and, potentially, allows differentiation between ionization and other effects.
We have obtained far­infrared (IR) through soft X­ray (100 ¯m -- 3.5 keV) continuum
data for a sample of 41 quasars. Spectral energy distributions (SEDs) for 30 of these objects
were presented in E94, the remaining 11 are compiled here from a combination of our own
and published data. We also present low­resolution (5­20 š A), optical spectrophotometry
which was obtained within a few months of the optical/IR continuum measurements.
Ultra­violet (UV) spectra from IUE and/or HST have been obtained either by ourselves
or from their respective archives. The range in shapes of the IR--UV continuum is large
even in this small, low­redshift sample (E94), allowing investigation of whether the range in
continuum shapes produces any corresponding range in the emission line properties.
2. Observations and data reduction
2.1. The sample
Our sample is a heterogeneous sample of 41 low redshift quasars (z!1), where 23
are radio­quiet (RQQ) and 18 are radio­loud (RLQ). The radio­quiet objects dominate at

-- 6 --
lower redshifts, while the radio­loud dominate at higher redshifts (see Fig.1). All but three
quasars (PG1012+008, PG1121+422, PG2304+042) have Einstein X­ray observations with
sufficient counts to define the X­ray spectral index. This selection introduces a bias towards
objects with strong X­ray emission relative to the optical (small ff ox ). The spectral energy
distributions (SEDs) for 30 objects were presented in E94. For the remaining 11 objects,
the SEDs were assembled from existing data as described below. The list of all quasars
with their redshifts is given in Table 1.
2.2. Optical Spectroscopy
The quasars were observed during the period 1985­91 mostly with the Multiple Mirror
Telescope (MMT) on Mt. Hopkins in Arizona. Both blue and red spectrographs were used
as far as possible to provide full coverage (3200--8400 š A) of the optical continuum. Southern
objects were observed with the CTIO 1.5m and CCD spectrograph. The observational
details are given in Table 1, including the instrument and wavelength range covered in each
case.
In photometric conditions, spectrophotometry was carried out by observing each quasar
twice, first through a large aperture, ¸ ? 5'', and second through a small aperture ¸ 1:5'' for
a longer time in order to obtain high spectral resolution and signal­to­noise. A standard
star was observed through the wide aperture, at similar air mass, immediately before/after
the quasar observation, to provide flux calibration. The data were reduced in the standard
manner using IRAF. The continuum of the small aperture data was then normalized to
match the shape and absolute flux level of the large aperture observation yielding a final
spectrum with resolution ¸ 5 \Gamma 20 š A (depending on the instrument) and a photometric
accuracy of ¸ 5% . Except where noted in Table 1, all spectra have been flux calibrated
using a large aperture spectrum, observed in photometric conditions as described above.

-- 7 --
The date of the spectrophotometry, sometimes different from that of the higher S/N, small
aperture spectrum, is also noted in Table 1.
In some cases several spectra covering largely overlapping wavelength ranges were
obtained. If no significant change in continuum shape was apparent, these spectra were
combined to improve the final S/N. Spectra with continuum shapes which did not agree
within the errors were not combined, instead the discrepant or lower quality spectrum was
discarded. Blue and red spectra of the same object were not combined since this adds
uncertainty in the small overlap region of the spectra and is not required for our analysis.
Table 1 lists observational details for all the individual spectra (i.e. before being combined).
The final, combined spectra are presented in Figure 2 in order of increasing right­ascension
with blue and red spectra displayed separately.
The fluxes, equivalent widths (EW) and full­width at half maximum
intensity (FWHM) were measured for all the prominent optical and UV emission
lines: OVI–1034+Lyfi–1025, Lyff–1216+NV–1240, SiIV–1397+OIV]–1402,
CIV–1549, CIII]–1909+SiIII]–1892+AlIII–1857, MgII–2798, [NeV]–3426, [OII]–3727,
[NeIII]–3869+HeI–3889, [NeIII]–3967, Hffi–4102, Hfl–4340, FeII–4570, Hfi–4861,
[OIII]–5007. The EW and fluxes were measured by the summation procedure detailed in
Robertson (1986), which provides a better estimate of the line wings than Gaussian fitting.
All fluxes and EW were re­measured using the splot task in IRAF. Rather than fitting a
Gaussian function to these lines, which frequently does not work well, we measured the
actual data by fitting a linear continuum and integrating across the observed emission line
(keystroke 'e'). If the EW and fluxes derived by both methods agreed within 10% the
values obtained by the summation procedure were used for the subsequent analysis, if they
disagreed the values from the splot task were used. For the FeII optical multiplet, which
is especially difficult to measure if the line is weak, the summation procedure was used to

-- 8 --
determine the EW. The FWHM were measured using only the IRAF splot task by fitting
a linear continuum, measuring the line peak flux above the continuum, determining the
half­maximum intensity and measuring the width at that level. By measuring all the data
ourselves we minimize the scatter generally introduced by combining datasets from different
measuring techniques and different authors. The equivalent widths of optical and UV lines
are listed in Tables 4 and 5, the UV and optical line fluxes in Tables 6,7 and the UV and
optical FWHM in Tables 8,9. All values are in the rest­frame.
2.3. Spectral Energy Distributions (SEDs).
The SEDs for the 11 quasars, not presented in E94, were compiled here from a
combination of our own ground­based optical and near­infrared photometry (from the 0.6 m
FLWO telescope, the MMT, the Mount Lemmon 61'' and the IRTF, for details see Table 3
and E94) as well as from the Einstein X­ray data, archival IUE or HST ultra­violet data and
IRAS far­infrared data. The various datasets were combined following the method described
in E94. When no optical photometry was available the MMT spectrophotometry was used.
For inclusion in the SEDs, prominent emission lines were subtracted from the optical and
UV spectrum and the spectrum binned into broader wavelength bands. A correction for
galactic extinction was applied to the SEDs based upon the Galactic neutral hydrogen
column and assuming a fixed conversion of N(HI)=E(B \Gamma V ) = 5:0 \Theta 10 21 cm 2 mag \Gamma1
(Burstein & Heiles 1978). The Galactic HI column has been accurately measured by Elvis
et al. (1986) for 7 of the 11 objects. For the remaining 4 objects it was determined using the
dereddening values from Neugebauer et al. (1987). After correcting for galactic extinction
the data were blueshifted to the rest frame using a cosmological model
with\Omega o = 1 and
H o = 50 kms \Gamma1 Mpc \Gamma1 . No k­corrections and no assumptions about the intrinsic spectrum
were required since we were working with the complete spectral energy distributions. The

-- 9 --
contribution from the host galaxy was then subtracted using the method and template of
E94 (based on the Sbc galaxy model of Coleman, Wu and Weedman 1980) and normalized
by the host galaxy monochromatic luminosity in the H band, which was taken from the
literature or, when not available, set to 1/3 of the luminosity of the quasar in that band
(McLeod and Rieke 1994a,b). The final SEDs for the 11 objects are displayed in Figure 3
in order of right­ascension.
To allow comparison with earlier studies, in Table 2 we list several standard continuum
parameters taken from the literature and determined in the usual way (i.e. not from the
SEDs). These include: L opt (luminosity at 2500 š A), LX (X­ray luminosity at 2 keV), ff ox
(effective optical to X­ray slope) from Wilkes et al. (1994), ff x (X­ray energy index, where
f š / š \Gammaff x ) from Wilkes and Elvis (1987), ff ouv (optical slope, f š / š \Gammaff ouv taken between
1285 š A and 5100 š A) from Kuhn (1996) and radio class (quasar considered radio­loud (L) if
RL ? 1, after E94). We also list the blue­bump strength measured from the SEDs, CUV=IR
(E94). In order to better characterize the SEDs, we directly measured the decade and
octave luminosities, as defined in E94 (given in Table 10,11). In Table 12 we give several
broad­band luminosities: LUVOIR , LBOL , Lycon, L Ion (ionizing continuum), all defined as in
E94. All SEDs parameters were measured using the TIGER software package written by J.
McDowell (Wilkes & McDowell 1995). We have also measured the driving continuum for
each line following Krolik & Kallman (1988; Table 4).
3. Correlation Analysis
We studied correlations between the various line and continuum properties in
Tables 14,15 using the ASURV statistical package (Isobe, Feigelson & Nelson 1986) which
includes allowance for the presence of upper limits in the sample. Specifically we applied
the following tests to each pair of parameters: the Generalized Kendall Rank test and the

-- 10 --
Spearman Rank test, which is insensitive to outlying points. We considered a correlation
real only if the probability of it occurring by chance was !2% in both of these tests. We
present the percentage probability of a chance correlation in the Generalized Kendall Rank
test followed by that for the Spearman Rank test in Tables 14,15 for probabilities ! 5%.
Significant correlations (P ! 2%) are given in bold text. When multiple correlations were
found we tested for the primary relation using ASURV bivariate Spearman ranks as input to
partial Spearman rank (PSR) analysis (Kendal & Stuart 1976). The primary correlations,
i.e. the strongest pairwise correlations when the other variables are held constant, are
underlined in Tables 14 and 15.
Since we are considering a large number of parameters (3 parameters for each of 15
emission lines: EW, FWHM, flux; and 33 continuum parameters ­ see previous section), we
divide our results up by emission line parameters, first discussing the equivalent width vs
continuum luminosity correlations (Section 3.1) followed by the flux (Section 3.2) and the
FWHM (Section 3.3). We then discuss the correlations between the various line parameters
(Section 3.4) and between the various continuum parameters (Section 3.5). The number
of correlations studied and found and the number of spurious correlations expected with
P ! 2% are shown in Table 13.
3.1. Equivalent width ­ continuum luminosity correlations
3.1.1. UV lines
The best known line­continuum correlation for quasars is the inverse correlation
between the emission line equivalent widths (EW) and the continuum commonly known
as the Baldwin effect. It was detected for the first time by Baldwin (1977) in a sample of
high­redshift quasars in which the EW of the CIV–1549 line decreased with increasing UV

-- 11 --
continuum luminosity, L(1450 š A). The Baldwin effect has since been reported for a large
number of lines (OVI, NV, HeII, CIII], MgII and Lyff) in a number of different samples
(eg. Tytler & Fan 1992, Zamorani et al. 1992, Green 1996). Baldwin et al. (1978) showed,
however, that the correlation is much stronger in a sample of flat­spectrum, radio­loud
quasars (RLQs). This was later confirmed by Wampler et al. (1984), and Baldwin, Wampler
& Gaskell (1989). On the other hand, the analysis of optically­selected quasar samples
(Osmer 1980, Crampton et al. 1990, Boyle et al. 1991) shows that the Baldwin effect is
less pronounced. It has been suggested that there exist two kinds of the Baldwin effect:
a global, or object­to­object, effect and an intrinsic effect due to variability of the object
(Pogge & Peterson 1992, Murdoch 1983). The slope of the intrinsic effect is different from
that of the global one, adding scatter to the global relationship. However as the result of
variability and selection effects some regions of the EW­L diagram may be left unpopulated
and hence result in a more pronounced Baldwin effect for the radio­loud quasars.
The correlation results, between EW and continuum luminosity, are presented in
Table 14 with coding explained in Section 3. The primary correlations are underlined and
are displayed for the Lyff, Hfi and CIII] lines in Figure 4 a,b,c.
The Baldwin effect is present in our sample for the Lyff line, for which the EW
anticorrelates with all the various OUV luminosities. The primary correlation is with
L(0.1­0.2¯m) luminosity.
The EW(CIII]) shows a very strong (primary) correlation with L(1 \Gamma 2keV ) and a
marginal correlation with L(0.1­0.2¯m). A CIII]­X­ray correlation was reported by Green
(1996) in a sample of quasars observed by the Einstein and IUE satellites. However our CIII]
line sample has more than 50% upper limits, the correlation looks unconvincing (Fig.4b),
and no related correlation between the CIII] flux and X­ray flux is seen (Section 3.2). This
suggests that the EW(CIII]) vs L(1­2keV) correlation is spurious and we cannot confirm

-- 12 --
the correlation reported by Green (1996).
The traditional Baldwin effect for the CIV line is not present in our sample (Figure 5a).
However, a significant correlation (P ! 1% in all cases) is found for both CIV and CIII]
lines if we omit seven objects with low logšL š (0:1 \Gamma 0:2¯) i.e. IZw1, NAB0205+024,
PG 0844+349, PG 1012+008, PG1211+143, PG 1351+640, IIZw136. These objects have
relatively weak carbon lines (EW(CIV)/EW(Lyff) !0.55) as can be seen in Figures 5a,b,c
where they are indicated by filled circles. They also have narrow FWHM of Hfi (! 2500km
s \Gamma1 , Table 9) falling into the range of narrow line Sy1 objects (NLSy1, Osterbrock & Pogge
1985; exept for PG 0844+349 and PG 1351+640 which are broad­absorption line QSOs).
This suggests that NLSy1s may have systematically lower carbon line EW and so do not
follow the general Baldwin effect for the rest of the sample.
The only significant correlation involving ff ox is a strong anticorrelation with EW(CIV)
relation (Fig. 6a). A marginal correlation with EW(Hfi) becomes significant when one
object, PG1001+054, is omitted (Table 14). For the OVI line, marginal anticorrelations
of EW(OVI) with L opt (PK =1.7%, P S =3.8% and with ff ox (PK = 3:1%, P S = 4:1%) are
present in our sample. The latter correlation was reported by Zheng, Kriss & Davidsen
(1995) in their sample of 30 QSOs and 2 Sy1 (z?0.15), which had both IUE and Einstein
X­ray fluxes. However with only 10 points in our sample we cannot draw strong conclusions
from our lack of a correlation here.
3.1.2. Balmer lines
The primary correlation for EW(Hfi) is with L(0.4­0.8¯m) (Figure 4c). The EW(Hfi)
also anticorrelates with the L(100­10¯m) infra­red luminosity, L(0.1­0.2¯m), OUV
luminosities and X­ray L(1­10keV) luminosity (Table 14). A closer look shows that the

-- 13 --
``correlation'' between EW(Hfi) and L(1­10keV) is due to a systematically higher L(1­10keV)
for RLQs (Figure 4d). The L(1­10keV) luminosities are determined by extrapolating the
Einstein soft­X­ray (0.1­3.5 keV) flux and energy index. The systematically flatter ff x
and stronger X­ray flux of the RLQs (Wilkes and Elvis 1987) exaggerates the systematic
difference between L(1­10keV) in the two classes.
Similar to Hfi, the equivalent­widths of the other Balmer lines anticorrelate with the
far­IR luminosity, L(10­100¯).
3.1.3. FeII line
No significant correlations are found for the EW of the FeII–4570 optical multiplet.
The one with the lowest probability of chance occurrence (PK =2.1%, P S =3.5%) is an
anticorrelation with L(1­2keV) (reported as significant by Boroson & Green 1992 and
others). However Fig. 7a does show a trend for objects with larger L(1­2keV) luminosity
and smaller EW(FeII) to be radio­loud quasars, while radio­quiet objects have a much
larger range in EW(FeII) and concentrate towards smaller X­ray luminosities.
No correlation between the EW(FeII) line and ff x is present in our sample (see Fig.
7b) nor in the RQQ and RLQ subsamples. Such correlations have been found by Wilkes
et al. (1987) and Shastri et al. (1993) in samples observed by the Einstein satellite.
However Boroson (1989) studying a radio selected sample, Walter and Fink (1993), who
studied the Seyfert galaxies observed in the ROSAT All Sky Survey (RASS), and Zheng &
O'Brien (1990) failed to confirm this correlation. 31 objects in our sample are also included
in the Shastri et al. (1993) sample, however only 16 of them have FeII measurements.
A comparison of our Fig. 7b, with their Fig. 5a shows small disagreements between the
measured values and one incorrect slope. PG0844+349 has an X­ray energy index ff x of 1.6

-- 14 --
in Shastri et al. (1993) whereas the correct value from the Einstein data is 0.6 (E94).
Given the apparently discrepant results in the literature, we conclude (as did Lawrence
et al. 1997) that there is no direct correlation between FeII and ff x , but rather a zone
of avoidance such that objects with flat X­ray spectra do not have strong FeII. A similar
effect, that objects with stronger X­ray luminosities do not have strong FeII, would also
explain our marginal correlation between FeII and L(1­2keV) (Fig. 7a). This also agrees
with the long­established tendency towards weak FeII in RLQs (Peterson, Foltz, and Byard
1981, Phillips 1977, Osterbrock 1977).
Lawrence et al. (1997) note a much stronger correlation between FeII and the
near­IR­X­ray slope ff ix . We tested for this in our sample and found no significant
correlation (PK =5.0%, P S = 6:9%). However our data cover only about 60% of the range
in ff ix and a smaller range in log(FeII/Hfi) than that of Lawrence et al. (1997) (­1.5 to 0 in
our sample cf. ­1 to 0.7 in theirs).
3.2. Line flux vs continuum flux correlations.
The optical/UV line flux -- continuum flux correlations are presented in Table 15,
with similar coding to Table 14. This analysis was performed identically in both flux and
luminosity space to guard against purely redshift­induced results. However, given the small
range in redshift (z!0.4 for all but 5 sources, Figure 1), there is no significant difference
between the two sets of results. All the primary correlations for these lines are shown
in Figure 8. For Lyff and CIV lines the primary correlations are with the continuum
regions closest to the driving continuum, as expected. However we note that simultaneous
measurements of both UV lines and UV continuum (as is the case in our sample) will
result in smaller scatter and lower probabilities for these correlations. In other bands small

-- 15 --
continuum variations or systematic shifts between the measurements will add scatter.
For the CIII] line the primary correlation is with the optical flux, far from its driving
continuum. No conclusions can be drawn for the SiIV line as it is not clear what the
driving continuum is for this line (T. Kallman, private communication). Correlations
for Lyff flux and EW are the most significant (usually P !! 0:1% cf. P ?? 0:3%; see
Table 15) and have the smallest scatter among the UV lines (Table 16). The predominance
of the EW(Lyff) correlations reported in the previous section may be due to the primary
correlation being with the local continuum for this line (Table 15). Thus, the EW is
determined by only two parameters, the line flux and the local continuum, whereas for lines
whose primary correlation is not with the local continuum, non­simultaneity and calibration
uncertainties between the different continuum regions can induce additional scatter.
3.2.1. Line flux vs. driving continuum correlations
We investigate the correlations between the strengths of the broad emission lines such
as Lyff, OVI, CIV, CIII], Balmer lines, FeII and their driving continua. To determine the
driving continua, the SEDs were linearly interpolated between observational points in the
OUV. In the X­rays the observed ff x was used and the EUV continuum was determined
by a linear interpolation between the lowest energy point in the X­ray range and the
highest in the UV. Then the spectrum was integrated over the energy range of the driving
(ionizing) continuum of each line (see Table 15) following the definitions of Krolik and
Kallman (1988; see their Table 4). The probability of a correlation between each line
and its driving continuum, as given by the Kendall and Spearman rank tests, is shown in
Table 15. The slopes of the linear regression, and the scatter of the observational points
around the regression line are shown in Table 16. All correlations between the lines and
driving continuum are shown in Fig 9. While these correlations are strong, surprisingly in

-- 16 --
no case were they the primary line vs continuum correlations. In addition the slopes of the
regression lines were always smaller than the expected value of 1. Possible explanations
include differing line and continuum reddening, optically thin (matter bounded) clouds and
continuum emission from an accretion disk (see discussion in Section 4.1.3).
3.3. Line FWHM vs. Continuum Luminosity Correlations
Only two significant correlations were found between opt/UV line widths and
continuum luminosity parameters: FWHM(NeHe) with far­IR (10­100¯m) luminosity
(PK =1.9%, P S =1.5%), and second for FWHM([OIII]) with L Ion (1.3%,1.3%). NeHe also
correlates but less significantly with L(0.1­0.2¯) (2.4,2.7%) and LBOL (2.2%,1.7%).
As has been noted in Table 13, the number of spurious correlations with a probability
of a chance correlation less than 2% is expected to be 6 for the 315 correlations tested. Since
we found less than 6 significant correlations we consider it likely that they are spurious.
3.4. Correlations between emission line parameters
We have also studied correlations between the various parameters of each emission line.
We shall present here only those correlations which are significant and/or interesting.
The equivalent widths of the Balmer lines correlate with each other and the FWHM(Hfi)
correlates with FWHM(Hfl). This is consistent with photoionization models, as the Balmer
lines are expected to form in the same region.
The Lyff and CIV line EW and FWHM also correlate with each other (Fig 10). These
correlations are easily understood in the light of standard photoionization models, as it is
generally believed that these lines originate in clouds located at the same distance from the

-- 17 --
continuum source. However the presence of a correlation, also reported by Corbin (1991)
and Corbin & Boroson (1996), between the EW(Lyff), EW(CIV) and EW(Hfi) as well
as FWHM(Lyff), FWHM(CIV) and FWHM(Hfi) (and also FWHM of Hfl) is interesting.
Hfi (and Hfl) is a low­ionization line and is expected to form in a different region than
the CIV and Lyff lines. If both the high and low­ionization components scale similarly
with luminosity, two models are applicable. First, the Collin­Souffrin et al. (1988) model,
where low­ionization lines (LILs) are formed in the atmosphere of the accretion disk,
while high­ionization lines (HILs) form in a distinct spherical component. Second, the
standard model, where LILs and HILs are formed in the same cloud but in zones of differing
ionization.
We see no correlations between the EW and FWHM of individual lines in our sample.
An anti­correlation has been reported for CIV in some samples (eg. Francis et al. 1992,
Wills et al. 1993, Corbin & Francis 1994).
The EW([OIII]) anticorrelates with EW(FeII). This relation was noted by Boroson
and Green (1992) and dominates their first eigenvector. They rule out an interpretation
based on the standard beaming model partly due to the strong dependence on [OIII], which
is thought to be isotropic, and partly to the positive correlation of absolute continuum
magnitude with the magnitude of [OIII], which is not expected if beaming causes the range
in optical continuum flux. Instead they favor a model based on a geometric effect intrinsic
to the quasar such as shadowing of the NLR by a co­planar, toroidal­shaped BLR so that
the covering factor, perhaps determined by the ratio of accretion rate to the Eddington
rate, drives the correlations. Eddington­limited flow was also suggested to explain those
objects with extremely strong FeII emission (Pounds, Done and Osbourne 1996) based
primarily upon their unusually steep, soft X­ray spectra.
Since [OIII] and [OII] originate in similar parts of the NLR, we expect their emission

-- 18 --
line profiles to be similar. We find a correlation between their equivalent widths, but we do
not see a correlation for their FWHM. However we do not place much weight on this latter
result since the narrow lines are often close to the spectral resolution, [OII] is weak and we
have only 19 measurements of variable quality.
3.5. Continuum­continuum correlations
All the continuum luminosities correlate with one another. The regression slopes, for
L 1 vs. L 2 and L 2 vs L 1 relations, are presented in Table 17. Most of the relations have
regression slopes near unity. The exception is the L(0.1­0.2¯m) vs. L(0.2­0.4¯m) relation
whose slope is less than 1, indicating that L(0.1­0.2¯m) increases faster with increasing
bolometric luminosity than any other luminosity. This could be explained if the peak of the
big blue bump (BBB) moves from the EUV into the UV as the luminosity increases.
Correlations with the spectral indices were also investigated. The only correlations
found were those between LX , L(1­2keV) and ff ox (regression slope = --1.51\Sigma0.83 and
--1.88\Sigma0.7 respectively) showing the steepening of ff ox with decreasing X­ray luminosity
and the correlation between LX and L opt (regression slope = 0.74\Sigma0.15 in agreement with
the Wilkes et al. 1994 slope of 0.71). We do not find the correlation between ff ouv and ff x
reported by Puchnarewicz et al. (1996) or that of ff ox with ff x also reported by Laor et al.
(1997) (see Section 4.1.5).
The mean SEDs for high and low luminosity quasars in our sample are shown
separately in Fig.12. The quasars that contribute to the high luminosity mean are:
0637­752, 1226+023, 1407+265, 1704+608 (L(0.8­1.6¯m)?45.5) and those that contributed
to the low luminosity mean are: 0007+106, 0026+129, 0804+761, 1416­129, 1501+106
(L(0.8­1.6¯m)!44.2 and EW(Hfi) ? 70 š A. The low luminosity object 1001+054 was omitted

-- 19 --
while making the mean, as we had no information on the X­ray slope and the large errors in
starlight caused large uncertainties in the ¸ 1¯m region. The higher luminosity SEDs show
less dispersion around the median in the IR to UV bands than the lower luminosity SEDs.
The range in X­rays is systematically shifted to lower X­rays for the higher luminosity SEDs
and the big blue bump (BBB) is stronger (in the UV) for lower luminosity SEDs (meaning,
as was also noticed from the slope analysis, that the peak of the BBB moves from the EUV
into the UV as the luminosity increases). However the range in the SEDs present is broad
and overlaps significantly between the two samples.
4. Discussion
4.1. The Baldwin Effect
In a simple, radiation­bounded photoionization model, one would expect the line flux
in the high­ionization lines, such as CIV, Lyff, to be proportional to the strength of the
ionizing continuum as a larger fraction of emitting material reaches the ionization state of
the line. Since EW is the ratio of line flux to that in the local continuum (i.e. in the same
wavelength region as the line), the Baldwin effect indicates that the line flux is increasing
more slowly than the local continuum (or is constant). One possible interpretation is
that the ionizing/heating continuum for the emission line (i.e. its driving continuum) is
increasing more slowly than the local continuum, suggesting that the continuum shape
correlates with the luminosity of a quasar. A number of papers have reported such a
correlation. Tananbaum et al. (1986), Wilkes et al. (1994), as well as Green et al. (1995)
note that the power law between UV and soft X­rays, ff ox , increases significantly with
luminosity. This results in the soft X­ray luminosity being weaker relative to the UV in
higher luminosity quasars.

-- 20 --
4.1.1. Zheng & Malkan model
Zheng & Malkan (1993), studying optical and UV properties of quasars and Seyfert
galaxies, interpreted their color­luminosity relations as the result of a shift of the BBB
towards lower energies in higher OUV luminosity objects. In this scenario higher luminosity
quasars are predicted to have a lower fraction of higher energy ionizing photons available
relative to the opt/UV continuum. This change in continuum shape would cause the
equivalent widths of high ionization emission lines sensitive to the X­ray continuum, such
as CIV and HeII, to decrease at high luminosities relative to lower­ionization emission lines,
such as Lyff, CIII] and the Balmer lines. The scenario predicts a stronger and more easily
detectable Baldwin effect for higher­ionization lines. We do see a shift in the SEDs, but our
data do not favor this model as we see the Baldwin effect in both high and low­ionization
lines.
4.1.2. Mushotzky and Ferland model
Mushotzky and Ferland (1984) explain the Baldwin effect as due to a systematic
decrease in ionization parameter, U, as the luminosity increases. As the ionization
parameter depends on the number of ionizing photons, U depends on the continuum shape.
Their model assumes a single zone broad line region i.e. determines the ionization for one
``slab'' at a given distance from the central source and assumes spherical symmetry. The
model predicts the presence of a Baldwin effect for the CIV line, as the CIV luminosity
increases rapidly with increasing U (decreasing UV luminosity and ff ox ) in the range 30
to 32 in L 1450 . This relation flattens around L 1450 ! 30 and so predicts a weaker/or no
Baldwin effect for lower luminosity AGN (see Fig.1 in Mushotzky and Ferland 1984). In
addition their calculation showed that there should not exist a Baldwin effect in Lyff, CIII],
Hff and Hfi lines as the luminosity of these lines decreases with increasing U (i.e. decreasing

-- 21 --
luminosity) in the same luminosity range. It also follows that the equivalent widths of Lyff,
CIII], CIV should, in this scenario, be relatively independent of both ff ox and ff x as the
lines do not originate in the X­ray heated zones deep in the emission line clouds.
In our sample, this model matches only the CIV line for which we see the Baldwin effect
for L 1450 ? 29:7 (L(0:1 \Gamma 0:2¯m) ? 45) . The four narrow line low luminosity AGN which
have L(0:1 \Gamma 0:2¯m) ! 45 destroy the anticorrelation (with these objects the probablity
of a chance correlation is P ? 5%, without P ! 1%), and fall into the region where the
model predicts no Baldwin effect (as it is the region where CIV remains fairly constant
with increasing U). However we see a strong correlation for EW(CIV) with ff ox , as well as
less significant correlations for Lyff, Hfi and OVI with spectral shape, none of which are
consistent with the Mushotzky and Ferland model. Our data also show a Baldwin effect for
Lyff and the Balmer lines which the model does not predict. The model also predicts that
the CIV/Lyff ratio should decrease with increasing luminosity but no such relation is found
in our sample. We conclude that this model does not provide a good explanation for the
relations present in our sample and so reject a pure photoionisation scenario for the BELR.
4.1.3. The Accretion Disk (AD) Model
Netzer (1985,1987) and Netzer, Laor and Gondhalekar (1992), using an optically
thick and geometrically thin accretion disk, explain the Baldwin effect in a different way.
The model assumes a distribution of identical clouds at distances from the center of the
disk greater than the size of the UV­emitting part of the disk. The incident ionizing flux
from the accretion disk is composed of two components: the UV continuum, which is
viewing angle­dependent due to limb darkening and change in projected surface area, and
the X­ray component, which is viewing angle independent. The UV emission is strongest
when the disk is viewed face­on. Because the BELR radiates more isotropically than

-- 22 --
the UV disk, a random selection of objects differing only in viewing angle would result
in measurements of constant line luminosities but varying UV continuum luminosities,
producing an anticorrelation of EW with LUV . In addition, an anticorrelation between EW
and ff ox is expected as objects viewed face­on would have steeper ff ox (due to relatively
higher UV to X­ray emission) and smaller EW than those seen edge­on. Correlations of the
more isotropic X­ray emission with line EW would be much weaker.
The continuum viewed by the BLR clouds themselves is a function of viewing angle.
As a result, clouds at small viewing angles produce strong Lyff, OVI and Balmer continuum
emission due to the stronger UV ionising continuum. Clouds at large viewing angles, where
the fraction of high­energy X­ray photons is largest, emit low­excitation lines such as FeII
and MgII. The model predicts that, in clouds at these viewing angles, FeII emission can
reach an intensity comparable to that of Lyff and more than 10 times that of Hfi. Under
our assumption of optically thin emitting clouds, this viewing angle dependence will not
affect the observed line strengths unless external effects, such as viewing­angle dependent
reddening, are also present.
In general the results presented here show good agreement with the predictions of the
AD model. We see an anticorrelation of EW with UV luminosity for Lyff, Hfi, Hffi and a
less significant relation for CIII], although this becomes significant after excluding the low
luminosity narrow line AGN (probably NLSy1 ­ see Section 3.1.1). A similar relation for
CIV (the traditional Baldwin effect) also appears when these NLSy1s are omitted. We
see a strong correlation between EW and ff ox for CIV only and less significant relations
for Lyff, Hfi and probably for OVI (although we do not have enough data to draw strong
conclusions here). None of the lines correlate with X­ray luminosity, agreeing with the
model predictions.
To make a more quantitative comparison between the model and our data, we assume

-- 23 --
that the AD and BELR are hidden by a dusty torus similar to that postulated for Seyfert 1
and Seyfert 2 galaxies. Adopting an opening angle of ¸ 60 ffi (Phillips, Charles & Baldwin
1983) and following Netzer et al. (1992), the AD model predicts that the range in disk
inclination yields a range in OUV luminosity of a factor of 2 (0.3 in log). This modifies the
slope ff, of the line luminosity vs driving continuum correlations in the following way:
ff = 1 \Gamma 0:3=(logF max
cont \Gamma logF min
cont )
The range in flux continuum F cont in our sample is ¸ 2.5 orders­of­magnitude so the
expected slope is ¸ 1 \Gamma 0:12=0.88. Most of the lines fit into this scenario (Lyff, Hfl, and
probably Hfi, Hffi, FeII, OVI although the errors are large, see Table 16). However CIV and
CIII] have flatter slopes (0.57\Sigma0.14 for CIV and 0.69\Sigma0.19 for CIII]). Exclusion of the seven
discrepant NLSy1 objects (see Section 3.1.1) yields slopes still not steep enough (0.64\Sigma0.14
for CIV and 0.64\Sigma0.25 for CIII]). (We note that the CIII] line sample has ? 50% upper
limits so the errors in the regression slopes are probably under­estimated.) This may be due
to a significant contribution from collisional excitation, which is not directly related to the
ionising continuum. In contrast, the FeII line flux vs continuum flux correlations are too
steep (¸ 1:3; see Table 16) to fit this scenario. One possible explanation for this behavior is
reddening by dust which increases with increasing viewing­angle. Given the predominance
of FeII emission from large viewing­angles described above, this line would be reddened
more than the continuum. This could lead to a stronger viewing­angle dependence than the
continuum and so a steeper relation, as observed.
Another discrepancy between our results and the predictions involves the scatter
around the mean slope in the line vs. continuum correlations which is larger than that
expected (i.e. ?0.15 around the mean, Table 16). Effects that could increase the scatter
include a non­spherical geometry of the BLR clouds, random obscuration by dust in the
NLR or a different opening angle for the dusty torus. These possibilities are discussed in

-- 24 --
the following sections.
4.1.4. The two­zone BLR
The BLR is sometimes suggested to include two distinct regions: the high ionization
line (HIL) region (Lyff, CIV, CIII]) and low ionization line (LIL) region (Balmer, FeII
and MgII lines). Collin­Soufrin et al. (1988) assumed, for example, that the LILs have
a disk geometry while the HILs do not. In this scenario, a random selection of objects
differing only in disk inclination would lead to a scatter around the mean in a plot of F line
vs FDRIV which is larger and a slope which is flatter for the HILs (as these lines have a
spherical symmetry), than for the LILs. In our sample the mean slopes for the LIL Balmer
and FeII lines are indeed steeper than those for the HILs (i.e. Lyff, CIV and CIII]) (see
Table 16). However the scatter in the correlations is larger for LILs than HILs, contrary
to the predictions. This scenario potentially offers an explanation for the steepness of the
FeII vs OUV flux correlations if distributions of FeII and OUV continuum emission from
the AD differ. However, given the partial agreement between model predictions and the
observations and the large uncertainties in the derived slopes, we are unable to draw strong
conclusions on the applicability of this model.
4.1.5. The role of dust
The presence of dust in quasars has been discussed by many authors (eg. Webster
et al. 1995, Rawlings et al. 1995, Sprayberry & Foltz 1992, Low et al. 1989, McLeod &
Rieke 1994a,b). Baker (1997), for example, studied a complete sample of radio­loud quasars
and found correlations between optical continuum and emission lines and the orientation
indicator R (the ratio of radio­core to lobe flux density). These correlations and the result

-- 25 --
that core­dominated quasars are 3­5 mag brighter in the optical region than lobe­dominated
quasars were explained as due to viewing­angle dependent dust extinction. In addition
strong correlations of the equivalent widths of [OII] and MgII and the [OII]/[OIII] ratio with
R, the lack of significant trends for CIII], CIV, Hfi and a weaker trend for [OIII] suggest
that the broad line region and the inner part of the narrow line region suffer extinction
similar to the continuum. The Balmer decrements (Hff/Hfi and Hfi/Hfl) also correlate with
R, although for Hfi/Hfl the correlation is less significant, probably due to contamination
of Hfl by [OIII]–4363. Given the lack of measured R values for the quasars in our sample,
we can provide only weak, direct support for this scenario by confirming the lack of any
relation between the Balmer decrement, Hfi/Hfl, and the EW of the broad emission lines
expected from the above correlations with R.
Puchnarewicz et al. (1996), studying a sample of X­ray selected Seyfert 1 galaxies
and quasars covering a wide range in redshift and X­ray spectral slope, reported that the
optical to X­ray spectra change from convex to concave as the X­ray slope, ff x , hardens
(i.e. ff x and ff ox anticorrelate). This was interpreted in terms of an intrinsic convex
spectrum being absorbed by differing amounts of cold gas and dust, possibly depending on
viewing­angle and so similar to the Baker (1997) scenario. Given the small range of ff X in
our sample (\Gamma0.3 to 0.9 cf. \Gamma2 to 3 for Puchnarewicz et al. 1996) and the large scatter in
the correlation, we do not see a significant correlation here. However, dividing our sample
according to ff X and applying the median test reveals a significant trend (? 99 %) for ff ox
to be smaller when ff x is hard. We do not find a significant correlation between ff opt (i.e.
ff ouv in our sample) and ff x , again this is not surprising given our small range of ff X . We
also note that Puchnarewicz et al. (1996) show a weak correlation for the high redshift
objects only (z ? 1:0, P S = 4%), their lower redshift objects do not show a correlation,
consistent with our sample.

-- 26 --
There are a number of other properties of our sample which could potentially be
explained by dust reddening. These include: the larger than canonical ``case B'' Balmer
decrements: 2.85 ! Hff/Hfi !7 (case B Hff/Hfi=2.85), and 2.1 ! Hfi/Hfl ! 5.3 (case B
Hfi/Hfl=2.1) implying a reddening of A v ! 2:04 mag; the existence of primary correlations
for optical and UV lines which are not with the driving continuum (see Section 4.1.3); the
increased scatter in the line luminosity vs driving continuum relations above that predicted
by the disk model; the larger scatter within the line/continuum correlation of Hfl compared
to Hfi (see Table 16). We investigate dust as a contributor to the correlations in detail in
the next section.
4.1.6. Simulations of the Effects of Reddening.
We investigate the effects of dust on the results of our analysis by applying random
amounts of dust redenning to a simulated sample of quasars with properties similar to our
actual sample. First we selected the quasar in our sample with the largest BBB i.e. the
one with largest difference: L(0.1­0.2¯m)\GammaL(0.8­1.6¯m) (PG1426+015). Since this object
also has a ratio Hfi/Hfl equal to the unreddened ``case B'' value (2.1), we assume it is
unreddened. A hypothetical sample of 26 objects was then constructed with the continuum
shape of PG1426+015 but covering the range in bolometric luminosity present in the
current sample (2.5 in L(0.8­1.6¯m)), spaced by 0.1 in logšL š . The maximum reddening
was determined from the difference in strength of the strongest and weakest BBBs in the
real sample. Assuming the dust lies outside the BELR, a random amount of reddening up
to the maximum was applied to both the SED and the two extreme emission lines (in –):
Lyff and Hfi. The process was performed twice, using two extinction curves: the standard
Galactic extinction curve (Seaton 1979) and the Small Magellanic Cloud (SMC) extinction
curve (Pr'evot et al. 1984), which corresponds well to the extinction by dust dominated

-- 27 --
by amorphous carbon grains (Czerny et al. 1995). The estimated maximum reddening
was E(B \Gamma V ) = 0:32 for the Seaton extinction curve and E(B \Gamma V ) = 0:20 for the SMC
extinction curve. We note that this is much less than that estimated from the Balmer
decrement, implying a different amount of obscuration. We then performed a correlation
analysis similar to that applied to the real sample to look for the primary correlations
between Lyff and Hfi and various parts of the SED and to estimate the scatter in the
resulting line vs. driving continuum correlations. The results are as follows: for the Seaton
extinction curve the primary correlation for Lyff was with the driving continuum and for
Hfi with L(0.2­0.4¯m), the scatter for the line vs. driving continuum correlation for Lyff
was oe = 0:12 and for Hfi oe = 0:42; for the SMC extinction curve the primary correlation
for Lyff was with L(0.1­0.2¯m) and for Hfi was with L(0.2­0.4¯m), the line vs. driving
continuum correlation scatter for Lyff was oe = 0:12 and for Hfi oe = 0:20.
A comparison of these results with the current sample (Table 15,16) reveals that the
simulation using the SMC extinction curve has the same primary correlations for Lyff and
Hfi lines as the real sample and the scatter in the line vs. driving continuum correlations is
comparable to the additional scatter required to bring the results into agreement with the
AD model (see Table 16). The scatter introduced by the Netzer AD model (Section 4.1.3)
for both Lyff and Hfi is oe = 0:15. The addition of dust yields an overall scatter, oe ¸ 0:27
for Lyff and oe ¸ 0:35 for Hfi which is slightly larger than in the real sample (compare with
oe = 0:22 and oe = 0:33 respectively). However this slightly larger scatter is probably due to
the simplifying assumption that the difference between objects with larger BBB and those
with smaller BBB is entirely due to dust, neglecting any component due to the viewing
angle of the accretion disk. Taking this into account would reduce the maximum E(B­V)
and improve agreement with the observations.
We conclude that the Netzer AD disk model surrounded by an optically thick

-- 28 --
dusty torus, with addition of small amounts of dust outside the BELR fits well into our
line­continuum correlations. Our conclusion that the SMC dust extinction curve, which
corresponds well to extinction by dust dominated by small amorphous carbon grains (see
Czerny et al. 1995), fits better than the Seaton extinction curve, is consistent with previous
results that the dust in Seyfert galaxies and quasars may differ from the dust in our Galaxy,
showing a depletion of silicates (Czerny, Loska, Szczerba 1991, Laor & Draine 1993).
4.1.7. The Torus
The assumption that quasars have the same dusty torus as the Sy1/Sy2 galaxies
(Section 4.1.3) may not be valid. Perhaps the opening angle and radius or thickness of
a torus depends on the luminosity of the central engine. More luminous quasars would
then have larger opening angles and larger radius/thinner tori. A large torus opening
angle would naturally explain the flat slope and large scatter for the carbon CIV and CIII]
line vs. driving continuum correlations. This scenario could also explain why we see the
Seyfert 1/Seyfert 2 dichotomy, while there is nothing similar for quasars. However, the other
lines would then be too steep for the AD model so this scenario would require a different
emitting region for the carbon lines. This seems contrived and an unlikely possibility given
the generally similar behavior of the UV lines.
4.1.8. LOC model
As mentioned earlier, the majority of existing BELR models have been based on one
or two­zones of clouds. In recent papers, Baldwin et al. (1995) and Korista et al. (1996)
model the BELR from many clouds covering a wide range in distance r from the ionizing
source and with a wide range in gas density nH at each r. The integrated BELR spectrum

-- 29 --
is determined by integrating over the full distribution function of clouds at all radii and
densities. This is called the ``locally optimally emitting cloud'' (LOC) model. This model
predicts that the integrated spectrum depends only weakly on the shape of the ionizing
spectrum, the column density of the clouds and the cloud distribution with r (see Fig.2 in
Baldwin 1997). This implies that the BLR spectrum is determined more by selection effects
than by details of the cloud properties. Since we have concluded that the behavior of the
emission lines in our sample is not purely due to photoionization effects, it seems likely
that a combination of a BELR based on the LOC model, an accretion disk providing the
ionising continuum and a distribution of dust external to both AD and BELR would be
able to fully describe our sample.
Conclusions
We investigate the relations between various emission line parameters and continuum
parameters in a sample of low redshift (z ! 1) quasars. The sample consists of radio­loud
and radio­quiet objects for which IR to soft­X­ray SEDs are available, and is biased towards
AGN with strong X­ray emission relative to the optical. Eleven newly compiled SEDs as
well as optical spectra for all the quasars are presented. We have measured all the emission
line parameters uniformly to minimize the scatter generally introduced when combining
datasets from different techniques and different authors. We use survival analysis which
allows for the presence of upper limits.
We find anticorrelations between the line equivalent­widths and UV luminosities, i.e.
the Baldwin effect, for the Lyff and Hfi lines. Exclusion of narrow line low luminosity
AGN reveals the Baldwin effect also for the CIV and CIII] lines. This suggests that NLSy1
objects may have systematically lower carbon EW. We also find a significant correlation
between CIV and ff ox .

-- 30 --
No correlations between the EW and the X­ray luminosity or slope are reported.
However FeII–4750 shows a tendency for objects with flat X­ray spectra and/or strong
X­ray luminosities to have weak FeII (consistent with Peterson et al. 1981, Phillips 1977,
Osterbrock 1977). The anticorrelation between the EW([OIII]) and EW(FeII) (i.e. the first
Boroson and Green 1992 eigenvector) is present in our sample.
Correlations between the various parameters for each line were also studied. We confirm
the correlations between the FWHMs and EWs of CIV, Lyff and Hfi lines previously found
by Corbin (1991) and Corbin & Boroson (1996), indicating that the high and low­ionization
line components in the BLR scale similarly with luminosity. This is consistent with both
a model where low and high­ionization lines are formed in the same clouds but in zones of
differing ionization as well as a model were low­ionization lines are formed in the atmosphere
of an accretion disk, while high­ionization lines form in a distinct spherical component.
The continuum­continuum correlations reveal that the peak of the big blue bump
strengthens and perhaps shifts from the EUV into the UV as the luminosity increases. We
confirm that the X­rays are systematically lower relative to the optical for higher luminosity
objects (Wilkes et al. 1994). However both high and low luminosity objects show a broad
range in continuum shapes and the distributions of SEDs for these subsamples overlap.
We investigate correlations between the line flux and various continuum fluxes including
the driving continuum flux i.e. the part of the continuum responsible for ionizing each line.
While the line ­ driving continuum correlations are strong, in no case were they the primary
line vs. continuum correlations. In a pure photoionisation scenario, this is surprising but
can be explained by the presence of dust which reddens both the lines and continuum
similarly.
All the above correlations fit into the accretion disk model of Netzer et al. (1992 and
references therein) where an optically thick, geometrically thin accretion disk is composed

-- 31 --
of two components: a viewing­angle dependent UV continuum, due to limb darkening and
the change in projected surface area, and a viewing­angle independent X­ray continuum.
The model is consistent with our data as both the predicted EW anticorrelations with
UV luminosity and ff ox and the lack of correlations with the X­ray continuum are seen
in our sample. A quantitative look at the line flux vs. continuum flux correlations (i.e.
the primary correlations, regression slopes and scatter) reveals discrepancies which can
be explained if dust is present, reddening both the lines and the continuum. Simulations
suggest that the composition of the dust matches that of the SMC rather than our Galaxy
i.e. showing a depletion of silicates (consistent with previous studies: Czerny et al. 1995,
Laor & Draine 1993). The slopes of the line flux vs. driving continuum flux correlations for
CIV and CIII] lines are flatter than predicted and with too large a scatter to fit into the
AD + dust model. This can be explained if, as is typical, there is a significant contribution
from collisional excitation. The correlation would then be flatter as only a part of the line
flux would be directly related to the ionizing continuum.
We are grateful to Adam Dobrzycki for providing us with the HST data, Kim McLeod
for precise estimates of starlight for a few objects in our sample and Ken Lanzetta for
the IUE Atlas of AGN spectra. We thank Bo—zena Czerny and Martin Elvis for valuable
discussions and the referee for comments which improved the paper. BJW, JCM, PJG
gratefully acknowledge support provided by NASA through Contract NAS8­39073 (ASC).
JK acknowledges the support of a Smithsonian pre­doctoral fellowship at Harvard­
Smithsonian Center for Astrophysics and KBN grant 2P03D00410. SM acknowledges
support by NASA grant NAG5­3249 (LTSA) and PJG grant HF­1032.01­92A awarded by
the Space Telescope Science Institute, which is operated by the Association of Universities
for Research in Astronomy, Inc., under NASA contract NAS5­26555.

-- 32 --
REFERENCES
Baker, J., 1997, MNRAS, 286, 23
Baldwin, J.A.,1997 in ``Emission Lines in Active Galaxies: New Methods and Techniques''
IUE Colloquium 159
Baldwin, J.A., Ferland, G., Korista, K., Verner, D., 1995 ApJ, 455, L119
Baldwin, J.A., Burke, W.L., Gaskell, C.M. and Wampler, E.J., 1978, Nature, 273, 431
Baldwin, J.A., Wampler, E.J., Gaskell, C.M. 1989 ApJ 338, 630
Baldwin, J.A., 1977 ApJ, 214, 679
Bentley, M., Haves, P., Spencer, R.E., Stannard, D., 1975, MNRAS, 173, 93P
Boller, T., Brandt, W.N., Fink, H., 1996, AA, 305, 53
Boroson, T.A. & Green, R.F., 1992, ApJS, 80, 109
Boroson, T.A., 1989, ApJ, 343, 9
Boyle, B.J., Jones, L.R. and Shanks, T., 1991 MNRAS 251, 482
Burstein, D., Heiles, C. 1978, ApJ, 225,40
Coleman, G.D., Wu, C.C., Weedman, D.W. 1980, ApJS, 43, 393
Collin­Soufrin, S., Dyson, J.E., McDowell, J.C., Perry, J.J., 1988, MNRAS, 232, 539
Comastri, A., Setti, G., Zamorani, G., Elvis, M., Giommi, P., Wilkes, B.J., McDowell, J.C.,
1992, ApJ, 384, 62
Condon, J.J., O'Dell, S.L., Puschell, J.J., Stein, W.A., 1981, ApJ, 246, 624
Corbin, M.R., 1991, ApJ, 371, 51
Corbin, M.R. & Boroson T.A., 1996, ApJS, 107, 69
Corbin, M.R. & Francis, P.J. 1994, AJ, 108, 2016

-- 33 --
Crampton, D., Cowley, A.P. and Hartwick, F.D.A, 1990 AJ, 100, 47
Cutri, R.M., Wi'sniewski, W.Z., Rieke, G.M., Lebofsky, M.J., 1985, ApJ, 296, 423
Czerny, B., Loska, Z., Szczerba, R., Cukierska, J., Madejski, G., 1995, Acta Astron. 45, 623
Czerny, B., Loska, Z., Szczerba, R., 1991, in Physics of Active Galactic Nuclei, Eds. W.J.
Duschl and S.J. Wagner (Berlin: Springer­Verlag), p.198
Elvis, M., Wilkes, B. J., McDowell, J. C., Green, R. F., Bechtold, J., Willner, S. P.,
Cutri, R., Oey, M, S., and Polomski, E. 1994 ApJS 95, 1 (E94)
Elvis, M., Green, R.F., Bechtold, J., Schmidt, M., Neugebauer,G., Soifer, B.T., Matthews,
K., Fabbiano, G., 1986, ApJ, 310, 291
Ennis, D.J., Neugebauer, G., 1982, ApJ, 262,460
Francis, P.J., Hewett, P.C., Foltz, C.B., & Chaffee, F.H., 1992, ApJ, 398, 476
Green, P.J., 1996, ApJ, 467, 61
Green, P.J., Schartel, N., Anderson, S.F., Hewett, P.C., Foltz, C.B., Fink, H., Brinkmann,
W., Trumper, J, Margon, B., 1995, ApJ, 450, 51
Hyland, A.R., Allen, D.A., 1982, 199, 943
Isobe, T., Feigelson, E.D., Nelson P.I., 1986, 306, 490
Kellermann, K.I., Schmidt, M., Shaffer, D.B., Green, R., 1989, AJ 98, 1195
Kendal, M. and Stuart, A. 1976, The Advanced Theory of Statistics, Vol. II (New York:
Macmillan)
Krolik, J.H. and Kallman, T.R., 1988, ApJ 324, 714
Korista, K., Baldwin, J.A., Ferland, G., Verner, D., 1997, ApJS, 108, 401
Kuhn, O., 1996, PhD Thesis, Harvard University
Lanzetta, K.M., Turnshek, D.A., Sandoval, J., 1993, ApJS, 84, 109

-- 34 --
Laor, A., Fiore, F., Elvis, M., Wilkes, B.J., McDowell, J.C. 1997, ApJ, 477, 93
Laor, A., Draine, B.T., 1993, ApJ. 402, 441
Lawrence, A., Elvis, M., Wilkes, B.J., McHardy, I., Brandt N., 1997, MNRAS, 285, 879
Low, F.J., Cutri, R.M., Kleinmann, S.G., Huchra, J.P., 1989, ApJ, 340, 1L
McDowell, J.C., Canizares, C., Elvis, M., Lawrence, A., Markoff, S., Mathur, S., Wilkes,
B.J, 1995, ApJ, 450, 585
McLeod, K.K. & Rieke, G.H., 1994a, ApJ, 431, 137
McLeod, K.K. & Rieke, G.H., 1994b, ApJ, 420, 58
Miley, G.K., Hartsuijker, A.P., 1978, AApS, 34, 129
Murdoch, H.S., 1983, MNRAS, 202, 987
Mushotzky, R., Ferland, G.J. 1984, ApJ 278, 558
Netzer, H., Laor, A., Gondhalekar, P.M., 1992, MNRAS, 254, 15
Netzer, H. 1987, MNRAS 225, 55
Netzer, H. 1985, MNRAS, 216, 63
Neugebauer, G., Green, R.F., Matthews, K., Schmidt, M., Soifer,B.T., Bennett, J., 1987,
ApJS, 63, 615
Neugebauer, G., Miley, G.K., Soifer, B.T., Clegg, P.E., 1986, AJ, 308, 815
Neugebauer, G., Oke, J.B., Becklin, E.E., Matthews, K., 1979, ApJ 230,79
Osmer, P.S., 1980, ApJS 42,523
Osterbrock, D.E. & Pogge R.W., 1985, ApJ 297, 166
Osterbrock, D.E., 1977, ApJ, 215, 733
Owen, F.N., Puschell, J.J., 1982, AJ, 87, 595

-- 35 --
Owen, F.N., Porcas, R.W., Mufson, S.L., Moffett, T.J, 1978, AJ, 83, 685
Parley, R.A., 1982, AJ, 82, 859
Penfold, J.E., 1979, MNRAS, 186, 297
Peterson, J.G., Foltz, C.B., Byard, P.L. 1981, ApJ, 251,4
Phillips, M.M., 1977, ApJ, 215,746
Phillips, M.M., Charles, P.A. & Baldwin, J.A. 1983, ApJ, 266, 485
Pogge, R.W, & Peterson, B.,M., 1992, AJ, 103, 1084
Pooley, G.G., Henbest, S.N., 1974, MNRAS, 169, 477
Pounds, K.A., Done, C., Osborne, J.P., 1996, MNRAS, 277, 1430
Pr'evot, M.L., Lequeux, J., Maurice, E., Pr'evot, E., Rocca­Volmerange,B., 1984, AAp, 132,
389
Puchnarewicz, E.M., Mason, K.O., Romero­Colmenero, E., Carrera, F.J., Hasinger, G.,
McMahon, R., Mittaz, J.P.D., Page, M.J., Carballo, R., 1996 MNRAS, 281, 1243
Rawlings, S., Lacy, M., Sivia, D.S., Eales, S.A., 1995, MNRAS, 273, 821
Robson, E.I., Gear, W.K., Smith, M.G., Ade, P.A.R., Nolt, I.G., 1985, MNRAS, 213, 355
Romanishin, W. Hintzen, P., 1989, ApJ, 341, 41
Sanders, D.B., Phinney, E.S. Neugebauer, G., Soifer, B.T., Matthews, K., 1989, ApJ, 347,
29
Seaton, M.J., 1979, MNRAS, 187, 73P
Shastri,P., Wilkes, B.J., Elvis, M., McDowell, J. 1993, ApJ, 410, 29
Sitko, M.L., Stein, W.A., Zhang, Y.­X., Wi'sniewski, W.Z., 1982, ApJ, 259, 486
Smith, E.P., Heckman, T.M., Bothun, G.D., Romanishin, W., Balick, B, 1986, ApJ 306, 64

-- 36 --
Sprayberry, D., Foltz, C.B., 1992 ApJ, 390, 39
Steppe, M., Salter, C.J., Chini, R., Kreysa, E., Brunswig, W., Perez,J.L., 1988, AApS, 75,
317
Swarup, G., Sinha, R.P., Hilldrup, K., 1984, MNRAS, 208, 813
Taylor, G.L., Dunlop, J.S., Hughes, D.H., Robson, E.I., 1996, MNRAS, 283, 930
Tananbaum, H., Avni, Y., Green, R. F., Schmidt, M., & Zamorani, G. 1986, ApJ, 305, 57
Tytler, D., and Fan, X.­M., 1992, ApJS, 79, 1
Walter, R., & Fink, H.H., 1993, AA, 274, 105
Wampler, E.J., Gaskell, C.M., Burke, W.L., Baldwin, J.A. 1984, ApJ 276, 403
Webster, R.L., Francis, P.J., Peterson, B.A., Drinkwater, M.J., Masci, F.J., 1995, Nature,
375, 469
Williams, O.R. et al. 1992, ApJ, 389, 157
Wilkes, B.J., Tananbaum, H., Worrall, D.M., Avni, Y., Oey, M.S, Flanagan, J., 1994, ApJS,
92, 53
Wilkes, B.J., and McDowell, J., C. 1995 in ``Astronomical Data Analysis and Software
Systems III'' ed. D. Crabtree, R. Hanisch and J. Barnes [ASP Conference Series] 61,
423
Wilkes, B.J., Elvis, M., McHardy, I., 1987, ApJ, 321, 23
Wilkes, B.J., Elvis, M. 1987, ApJ, 323, 243
Wills, B.J., Brotherton, M.S., Fang, D., Steidel, C.C. & Sargent, W.L.W. 1993, ApJ, 415,
563
Zamorani, G., Marano, B., Mignoli, M., Zitelli, V. and Boyle, B. J., 1992, MNRAS, 256,
238

-- 37 --
Zheng, W., and Malkan, M.A., 1993, ApJ, 415, 517
Zheng, W., Kriss G.A., Davidsen A.F, 1995, ApJ 440, 606
Zheng, W., O'Brien, P.T., 1990, ApJ, 353, 433
This manuscript was prepared with the AAS L A T E X macros v4.0.

-- 38 --
Fig. 1.--- L opt (2500 š A luminosity) vs. redshift for our sample. In this and all subsequent
figures stars indicate radio­loud quasars and circles radio­quiet quasars.
Fig. 2.--- Optical spectrophotometry of the quasars in our sample in order of increasing right
ascension and on a scale of F š (erg cm \Gamma2 s \Gamma1 Hz \Gamma1 ) as a function of – ( š A). Observational
details are given in Table 1.
Fig. 3.--- Radio­X­ray Spectral energy distribution (SEDs) for the 11 new objects (Table 3)
on a logšL š vs logš scale.
Fig. 4.--- Primary correlations for UV and optical line equivalent widths vs continuum
luminosities. Radio­loud quasars are indicated by stars and radio­quiet by circles.
Fig. 5.--- a) The lack of Baldwin effect for EW(CIV). b) CIV/Lyff and c) CIII]/Lyff vs.
UV luminosity relations. Filled circles denote low luminosity, narrow­line objects (NLSy1s)
which decrease the significance of the CIV Baldwin effect in our sample
Fig. 6.--- a) EW(CIV) vs. ff ox correlation. b) EW(Hfi) vs ff ox relation. Notice quasar
1001+054 in the upper­right corner, which drastically spoils this correlation
Fig. 7.--- a) The relation between the EW(FeII) and the soft­X­ray luminosity b) The relation
between EW(FeII) and ff x . We see no correlations but rather zones of avoidance such that
objects with stronger X­ray luminosities and/or flat spectra do not have strong FeII
Fig. 8.--- Primary correlations for the line vs. continuum fluxes.
Fig. 9.--- Correlations for UV and optical line fluxes vs driving continuum fluxes.
Fig. 10.--- Correlations between the EW (in š A) and FWHM (in units of 10 \Gamma3 kms \Gamma1 ) of Lyff,
CIV and Hfi lines
Fig. 11.--- The correlation between [OIII] and optical FeII equivalent widths (the first

-- 39 --
Boroson & Green eigenvector)
Fig. 12.--- a) Mean energy distributions for high luminosity quasars, normalized at 1.5 ¯m
and the 68, 90, and 100 (dashed line) Kaplan­Meier percentile envelopes, b) similar mean
energy distributions for low luminosity quasars

-- 40 --