ESTIMATING MULTINOMIAL LOGIT MODELS WITH SAMPLES OF ALTERNATIVES

Area of Science:

• Econometrics
• Urban Studies
• Statistical Modeling

Background:

• Bruch and Mare provided advice on sampling choice sets for conditional logistic regression in residential mobility.
• Previous econometric research indicated no statistical correction is needed for simple random sampling of unchosen alternatives.
• This study reconsiders the necessity of statistical corrections in specific sampling scenarios.

Purpose of the Study:

• To evaluate the impact of statistical corrections for simple random sampling of choice sets in residential mobility models.
• To determine if following Bruch and Mare's advice leads to accurate coefficient estimates.
• To assess the bias introduced by sampling corrections using real-world data.

Main Methods:

• Utilized data on stated residential preferences from the Los Angeles component of the Multi-City Study of Urban Inequality.
• Applied conditional logistic regression models.
• Compared coefficient estimates with and without the recommended statistical correction for simple random sampling.

Main Results:

• Implementing the statistical correction suggested by Bruch and Mare resulted in biased coefficient estimates.
• Omitting the sampling correction significantly reduced or eliminated the observed bias.
• Simple random sampling of unchosen alternatives does not require a statistical correction in these models.

Conclusions:

• The advice to implement a statistical correction for simple random choice set sampling in residential mobility models is flawed.
• Omission of the statistical correction leads to more accurate coefficient estimates.
• Econometric research supports the non-necessity of such corrections, aligning with findings from this study.