We report percent certainty (more widely known as McFadden's rho-squared) because it is the standard for logit models. When we run HB, we think it also makes sense to think in terms of RLH, because it provides a respondent-level measure of fit of the respondent's model to her data. We don't report RLH for aggregate logit or LC, but of course we can compute it from the log likelihood.

First, divide your model's log likelihood by the number of choices it reflects (number of choices is typically number of respondents times number of questions). Now take the anti-log (or exponential) which is the EXP function on Excel. For example, let's say your LL is -1,204 and that your experiment includes 10 choice questions from each of 100 respondents. First, divide -1,204 by (10 x 100) to get -1.204. Now take the exponential function to see that your RLH is 0.30 (of course in an aggregate model like this, what you have is the average RLH, not the respondent-level RLH, right?)