5.5 The Myth of Unbiased Selection
Given the strong selection effects introduced by relativistic beaming and obscuration, it is very difficult to find useful unbiased samples. In the ideal case, surveys would be designed to select AGN by some completely isotropic property. Since surveys are generally flux-limited, this effectively means selecting sources at a wavelength not strongly affected by obscuration or beaming, such as in the low-frequency radio, far-infrared, or hard X-ray bands.
At present, however, it is not possible to select a useful AGN sample in an unbiased way. The 3CR radio sample, selected at 178 MHz where steep-spectrum emission from the lobes dominates, contains only a few core-dominated objects (i.e., blazars); in the 3CR catalog (Spinrad et al. 1985), only 2/298 (0.7%) objects are BL Lac objects (0% in the complete subsample of Laing et al. 1983). That is, the 3CR catalog may be (essentially) unbiased by beaming but it does not include enough blazars to test their participation in unified schemes. In contrast, in the 1 Jy radio source catalog (Stickel et al. 1994), selected at 5 GHz, there are 37/527 BL Lacs (7%; there are 34 in the complete sample, with V 20 mag), and 214/527 (41%) FSRQ, i.e., about 50% of the sources are blazars. Low-frequency-selected samples that are larger than the 3CR, like the Molonglo or B2 surveys, are still not completely identified: a complete sample of 550 sources selected from the Molonglo catalog is in the process of being identified (Kapahi et al. 1994), while two relatively small but completely identified sub-samples of the B2 catalog exist (Fanti et al. 1987) but are defined by a cut in optical magnitude, which is certainly not an isotropic property. Finally, the emission at low radio frequencies need not be entirely isotropic, as it can be affected somewhat by beaming in the cores and radio-lobe hot spots.
Far-infrared selection does avoid the bias introduced by dust obscuration of optical/ultraviolet emission but introduces a bias against dust-free AGN. Furthermore, the selection effects at far-infrared wavelengths depend on the properties of the obscuring torus or other dusty structures, which may not generate isotropic emission (Pier and Krolik 1992) and which may differ from source to source by more than just orientation. The far-infrared may also be affected by beaming insofar as there is a contribution from an infrared nonthermal source. Finally, the sample identification for the IRAS survey is a major burden, not to mention that AGN are not found efficiently (less than 1% of the high-latitude sources are AGN).
Hard-X-ray surveys are theoretically useful but no deep large-area surveys are currently planned. Also, X-ray spatial resolution at 10 keV is not as good as radio or infrared resolution, making the confusion limit correspond to an effectively higher flux; i.e., hard X-ray surveys can not go as deep. The lower resolution also makes optical identification of the sample more difficult. In the future, an all-sky survey with arcsecond resolution at 10-20 keV would be very valuable for evaluating unified schemes in a model-independent way.
In short, to address the key question of unified schemes - whether the numbers and luminosities of objects in the classes one proposes to unify are as expected by the scheme - means using well-defined (i.e., flux-limited) samples with known biases, and accounting for those selection effects quantitatively. This allows us to comment on whether a specific unified scheme is plausible and what parameter values are needed to make it so. It also avoids the pitfall of defining samples post facto, based on ``isotropic quantities'' like narrow-line emission or host galaxy type, which may well be incomplete because the detection threshold is not related to the sample definition. Fortunately, allowing for selection bias head-on, at least in the case of relativistic beaming, is a robust process which is not terribly sensitive to the details of the beaming model.