Dust biasing of damped Lyman alpha systems: a Bayesian analysis

If damped Lyman alpha systems (DLAs) contain even modest amounts of dust, the ultraviolet luminosity of the background quasar can be severely diminished. When the spectrum is redshifted, this leads to a bias in optical surveys for DLAs. Previous estimates of themagnitude of this effect are in some tension; in particular, the distribution of DLAs in the (NHI, Z) (i.e. column density-metallicity) plane has led to claims that we may be missing a considerable fraction of metal-rich, high column density DLAs, whereas radio surveys do not unveil a substantial population of otherwise hidden systems. Motivated by this tension, we perform a Bayesian parameter estimation analysis of a simple dust obscuration model. We include radio and optical observations of DLAs in our overall likelihood analysis and show that these do not, in fact, constitute conflicting constraints. Our model gives statistical limits on the biasing effects of dust, predicting that only 7 per cent of DLAs are missing from optical samples due to dust obscuration; at 2 sigma confidence, this figure takes a maximum value of 17 per cent. This contrasts with recent claims that DLA incidence rates are underestimated by 30-50 per cent. Optical measures of the mean metallicities of DLAs are found to underestimate the true value by just 0.1 dex ( or at most 0.4 dex, 2 sigma confidence limit), in agreement with the radio survey results of Akerman et al. As an independent test, we use our model to make a rough prediction for dust reddening of the background quasar. We find a mean reddening in the DLA rest frame of log(10) < E( B - V)> similar or equal to -2.4 +/- 0.6, consistent with direct analysis of the Sloan Digital Sky Survey (SDSS) quasar population by Vladilo et al., log(10) < E( B - V)> = -2.2 +/- 0.1. The quantity most affected by dust biasing is the total cosmic density of metals in DLAs, Omega(Z,DLA), which is underestimated in optical surveys by a factor of approximately 2.