Are most low-luminosity active galactic nuclei really obscured?

At low Eddington ratios ((m)over dot), two effects make it more difficult to detect certain active galactic nuclei (AGN) given a particular set of selection methods. First, even allowing for fixed accretion physics, at low (m)over dot AGN become less luminous relative to their hosts, diluting their emission; the magnitude of the dilution depends on host properties and, therefore, on luminosity and redshift. Secondly, low-(m)over dot systems are expected and observed to transition to a radiatively inefficient state, which changes the spectral energy distribution ( SED) shape and dramatically decreases the luminosity at optical through infrared (IR) wavelengths. The effects of dilution are unavoidable, while the precise changes in accretion physics at low (m)over dot are somewhat uncertain, but potentially very important for our understanding of AGN. These effects will have different implications for samples with different selection criteria, and generically lead to differences in the AGN populations recovered in observed samples, even at fixed bolometric luminosity and after correction for obscuration. Although the true Eddington ratio distribution may depend strongly on mass/luminosity, this will be seen only in surveys robust to dilution and radiative inefficiency, such as X-ray or narrow-line samples; by contrast, selection effects imply that AGN in optical samples will have uniformly high Eddington ratios, with little dependence on luminosity, even at low L-bol where the median 'true' (m)over dot less than or similar to 0.01. These same selection effects also imply that different selection criteria pick out systems with different hosts: as a result, the clustering of low- luminosity optical/IR sources will be weaker than that of X-ray sources, and optical/IR Seyferts will reside in more disc-dominated galaxies, while X-ray-selected Seyferts will be preferentially in early-type systems. Taken together, these effects can naturally explain longstanding, apparently contradictory claims in the literature regarding AGN Eddington ratio distributions, host populations and clustering. Finally, we show that if current observed Eddington ratio distributions are correct, a large fraction of low- luminosity AGN currently classified as 'obscured' are in fact radiatively diluted and/or radiatively inefficient, not obscured by gas or dust. This is equally true if X-ray hardness is used as a proxy for obscuration, since radiatively inefficient SEDs near (m)over dot similar to 0.01 are characteristically X-ray hard. These effects can explain most of the claimed luminosity/redshift dependence in the 'obscured' AGN population, with the true obscured fraction as low as similar to 20 per cent.