On the Treatment of Repeated Observations in Panel Data: Efficiency of Mixed Logit Parameter Estimates