Prior studies examining audit quality typically regress a dummy variable for the actual type of auditors and other client characteristics with the proxy of audit quality.However,this setup implicitly assumes that client firms are randomly assigned to auditors.In addition,these studies restrict the slope coefficients of client characteristics on the proxy of audit quality to be the same across Big N and non-Big N clients.We re-investigate the auditor quality by incorporating self-selection of auditors in this paper.In addition,we allow the slopes of client characteristics to differ across the clients of the two types of auditors.We use the property-liability insurance companies operating in the United States between year 2002 and 2006 as a sample.We proxy audit quality by earning conservatism,measured as the signed loss reserve error.Our empirical results stand in sharp contrast with previous findings,i.e.,Big N clients are not systematically different from non-Big N clients in reserve-estimation but the former are more conservative than the latter for financially weak insurers.Our results indicate that Big N clients are less conservative in reserve-estimation than non-Big N clients after adjusting for self-selection bias,regardless of their financial conditions.