We measure bias and efficiency of parameter estimates in the conditional logit (CL) and independent availability logit (IAL) models. Our Monte Carlo experiments consider both no choice set formation where individuals choose from the full set of alternatives, and when choice sets are stochastically formed and individuals choose from a subset of all alternatives. We also compare the performance of the two models using empirical data on paddlefish angler preferences and catch-and-release regulations in Oklahoma. Both the CL and IAL work well when their own assumptions hold, but not under the alternative’s assumptions. The IAL approximates the attribute-based cutoff well in empirical data. While neither the IAL nor the CL is universally preferred, based on our findings, we recommend the IAL when the true consideration sets are unknown.

我们在条件 logit (CL) 和独立可用性 logit (IAL) 模型中测量参数估计的偏差和效率。我们的蒙特卡罗实验既考虑了个体从完整的备选集合中进行选择的无选择集的形成，也考虑了随机形成的选择集以及个体从所有备选的子集中进行选择的情况。我们还使用俄克拉荷马州白鲟垂钓者偏好和捕获和释放规定的经验数据比较了两个模型的性能。当他们自己的假设成立时，CL 和 IAL 都工作得很好，但在替代方案的假设下则不然。IAL 在经验数据中很好地近似了基于属性的截止点。虽然 IAL 和 CL 都不是普遍首选，但根据我们的发现，当真正的考虑集未知时，我们推荐 IAL。