Ø When the individual has to choose among several discrete alternatives, use multinomial regression models (MRM), which assume independence of irrelevant alternatives (IIA). Some examples are: Ø 1. Transportation choices: Car, bus, railroad, bicycle Ø 2. Religion choices: Christianity, Judaism, Islam, Hinduism, Buddhism, other Ø 3. Choice of education: High school, college, post-graduate Ø 4. Choice of Job: Do not work, work part time, or work full time Ø 5. Buying a car: American, Japanese, European MULTINOMIAL REGRESSION MODELS (MRM) Ø Consider the nominal or unordered MRM. Ø For transportation choice, use the nominal MRM because there is no particular (natural) order among the various options. Ø Three types of models: Ø 1. Nominal MRM for chooser-specific data Ø 2. Nominal MRM for choice-specific data Ø 3. Nominal MRM for chooser-specific and choice-specific data, or mixed nominal MRM Ø Chooser represents an individual who has to choose among several alternatives. Ø Choice represents the alternatives or options that face an individual. MULTINOMIAL LOGIT (MLM) OR MULTINOMIAL PROBIT MODELS (MPM) Ø These models are used for chooser-specific data. Ø These models answer: How do the choosers’ characteristics affect their choosing a particular alternative among a set of alternatives? Ø MLM or MPM is suitable when regressors vary across individuals. MULTINOMIAL LOGIT (MLM) Ø Generalize logit model as follows: Ø Choose a base category and set the coefficients equal to zero. 1 p i1 = a2 +b2 X i a 3 + b3 X i 1+ e +e a2 +b2 X i e p i2 = a2 +b2 X i a 3 + b3 X i 1+ e +e a 3 + b3 X i e p i3 = a2 +b2 X i a 3 + b3 X i 1+ e +e MULTINOMIAL LOGIT (CONT.) Ø Take log of odds ratios and estimate equations simultaneously using maximum likelihood (ML): æ p i2 ö lnçç ÷÷ = a 2 + b 2 X i è p i1 ø æ p i3 ö lnçç ÷÷ = a 3 + b 3 X i è p i1 ø p i1 = 1 - p i 2 - p i 3 CONDITIONAL LOGIT (CLM) OR CONDITIONAL PROBIT (CPM) MODELS Ø These models are used for choice-specific data. Ø These models answer: How do the characteristics or features of various alternatives affect individuals’ choice among them? Ø CLM or CPM is appropriate when regressors vary across alternatives. CONDITIONAL LOGIT MODEL (CLM) Ø Generalize the logit model as follows: ØUnlike with MLM, the coefficients α and β do not vary across choices, yet the added subscript j for an individual varies across the alternatives. a + bX ij e p ij = J
åe m =1 a + bX im
ØEstimated using maximum likelihood.
MIXED MRM
Ø Models used when we have data on both chooser-
specific and choice-specific characteristics. Ø Such models can also be estimated by the conditional logit model by adding appropriate dummy variables. Ø For example, in choosing cars, features of the cars as well as income and age of individuals may affect their choice of car. MIXED LOGIT (MXL) Ø To incorporate subject-specific characteristics in the analysis, MXL proceeds as follows: Ø Interact the subject-specific variables with the choice-specific characteristics. Ø Estimate model using CLM. Reference
Damodar Gujarati Econometrics by Example, second edition