[Stephen Jenkins]

I have no idea what you actually did (details really matter). For instance, I would need to be shown that your use of mixlogit does implement the Haan/Uhlendorf MNL model with frailty that I referred to last year. (But, please, don't even think of trying to show me or writing to me personally. Sorry, but I don't have time to look at it.) And, related, it's unclear why you clustered your standard errors. Ultimately, you need to persuade the reviewer that you did do "event history analysis" but used the fact that a particular sort of estimator that is typically used for other applications can be applied to estimate the event history model of interest here (subject to the data being organised in a particular way, etc. etc.) The simplest event history (discrete time) model with one spell per subject and no frailty can be estimated by organising the data into subject-period form (one row per period at risk of experiencing the event) and using a binary dependent model estimator such as logit or cloglog (what I called the "Easy Estimation" method in my 1995 Oxford Bulletin of Economics and Statistics paper). Although there are repeated observations (row) per subject in this data set-up, the SEs should not be clustered. The reason is that the Easy Estimation method is a convenient "trick" to maximize the correct likelihood. It gets more complicated outside the basic model, but the basic intuition may persist. This may be what the reviewer is asking you about.

Good luck.