Cheat Sheet: Multinomial, Multivariate, and Ordered Logit/Probit
Purpose:
When trying to determine the odds, likelihood, or probability of an event, use logit or probit.
o When the dependent variable is not continuous, use logit/probit to eliminate the
heteroskedasticity (nonlinear distribution of residual).
Logit is appropriate when the categories of the dependent are well dispersed. Probit is
recommended when the middle categories have greater frequencies than the high and low tail
categories.
Similarities and Differences Between Models
Logit Probit
Multinomial Multivariate Ordered Multinomial Multivariate Ordered
-No orders -≥2 outcome -Has -No orders -≥2 outcome -Has
-Any # of variables are meaningful -Any # of variables are meaningful
Dependent
categories binary & orders categories binary & orders
Variable
-mutually exclusive correlated -mutually exclusive correlated
categories categories
-Need one - n number of -one equation -Need one - n number of -one equation
reference category equations (n: in total w/ set reference category equations (n: in total w/ set
Setting up
- (n-1) number of number of ranges - (n-1) number of number of ranges
the
equations (n:the dependent equations (n:the dependent
Equation
number of variables) number of variables)
categories) categories)
Predicted
Use 1/(1+e-Σy) to calculate probability Use the Z-statistic table to find probability
Values
-Positive and significant beta: increases the likelihood of being in certain category
Coefficients -Negative significant beta: decreases the likelihood of being in certain category
-Insignificant beta: has little explanatory power in the model
Examples of Equations (taken from readings):
Multinomial Logit: Log (PA/PF) = aAF+XbAF+uAF Multinomial Probit: Pr (PA/PF) = aAF+XbAF+uAF
Five categories in dependent variable (grade: A, B, C, D, E, F), need four equations.
The probabilities of each grade need to be calculated using the predicted values.
Ordered Logit: Log (grade) = a0 + a1GPA' + a2ACT + Ordered Probit: Pr (grade) = a0 + a1GPA' + a2ACT +
a3ABSENCE + a4GVALUE a3ABSENCE + a4GVALUE
Use the results in SPSS/Stata to calculate threshold of the probabilities
Multivariate logit: Log [JOINTAcctii=1]=αj +βj BPi +γjfi+ Multivariate probit: Pr [JOINTAcctii=1]=αj +βj BPi
φjxi+ εji +γjfi+ φjxi+ εji
An equation for each dependent variable.
The probability for each dependent variable needs to be calculated using predicted values.
The relationship between the dependent variables can also be calculated using the predicted values.
Web Resources!!!
1. Probit and Logit Response Models: http://faculty.chass.ncsu.edu/garson/PA765/probit.htm
2. Stata 11 help for logit: http://www.stata.com/help.cgi?logit
3. Also please look under “Help with SPSS and Stata” on the class website for great resources from
Indiana University and UCLA.
