Estimating the "impact" of out-of-home placement on child well-being: approaching the problem of selection bias.