|
NLOGIT has become the standard package for estimation and simulation of multinomial discrete choice models. Version 3.0 is a full information maximum likelihood estimator for, among other models, up to four level nested logit models. Many other formulations are included in NLOGIT, including random parameters (mixed logit) models, multinomial probit, many forms of the nested logit model, and several new formulations for panel data. NLOGIT is the only large package for discrete choice analysis that contains the full set of features of an integrated econometrics program (LIMDEP).
NLOGIT 3.0 includes the following features:
Econometrics Studio - NLOGIT and LIMDEP
NLOGIT is a superset of LIMDEP Version 8.0. The full program consists of all of LIMDEP 8.0 plus the full set of additional features in NLOGIT, including the additional data management features, estimators for many types of discrete choice models, and the model simulator.
Documentation
The documentation for NLOGIT consists of the full set of manuals for LIMDEP plus a separate 200 page reference guide specifically for NLOGIT. This reference guide documents all features that are specific to the NLOGIT package.
Data Analysis
NLOGIT will typically be used to analyze individual, cross section data on consumer choices and decisions among multiple alternatives. But, the program is equally equipped for market shares or frequency data, data on rankings of alternatives, and, for several of the estimators, panel data from repeated observation of choice situations. There are several data handling procedures for NLOGIT in addition to all those available in LIMDEP.
These are some of the features of the program operation of importing data and exporting data from NLOGIT to other programs - the READ (import) and WRITE (export) operations. You can also transform and examine the data within the program.
- Read (import data)
- ASCII, XLS, WKS, Binary, DIF, CSV
- Supported by DBMS Copy and Stat Transfer
- Up to 3,000,000 observations
- Merge choice specific (attribute) and generic (characteristic) data in a panel data set format
- Multiple line data input or single line (e.g., from Gauss)
- Write (export data)
- ASCII
- CSV (can be read directly into Excel, Lotus)
- DIF
- Binary
- WKS
- NLOGIT system files can be read by Stat Transfer and DBMS Copy
- Transformations
- All algebraic transformations
- Recoding
- Sorting
- Existing or new variables
- Spreadsheet style data editor
- Convert data sets from single to multiple line observations
- Scaling for merging revealed and stated preference data sets
- Data Formats for Discrete Choice Models
- Individual choice data
- Frequencies or market shares (proportions)
- Ranks
- Automatic bypass of observations with missing data
- Panel data estimators for multinomial choice
- Choice Sets
- Fixed number of choices
- Variable number of choices out of universal choice set
- Variable number of choices from individual choice sets
- Weighting
- Individual observation weights
- Choice based sampling corrections in all models
Model Specification
NLOGIT's estimation programs are accessed as LIMDEP model commands. Since discrete choice models are often more complicated to specify than other single equation models in LIMDEP, the command setup includes many specifications that are specific to NLOGIT.
Commands for NLOGIT use LIMDEP's standard command syntax. NLOGIT also provides command builder dialog boxes for most model specifications
- Model Sizes
- Up to 100 choices in choice sets
- Up to 125 parameters in utility functions
- Tree structures may have up to 25 branches, 10 limbs, 5 trunks in 4 levels
- Data Setup
- Individual choice, frequencies, proportions, or ranks
- Scaling of data for merging revealed and stated preference data
- Size variables for multiplicative models
- Weighting for choice based sampling
- Automatic handling of missing data
- Multiple observations (panel data) for several models
- Utility Specifications
- LIMDEP standard command structure for equation specification
- Choice specific constants
- Interactions of choice specific constants and characteristics
- Within and across equation parameter constraints
- Fixed parameters in utility functions
- Logs and Box-Cox transformed variables
- Size variables for multiplicative models
- Scaling of sets of variables and observations
- Choice Sets
- Up to 100 choices
- Fixed or variable number of choices
- Universal choice set or individual specific choice sets
- Restricted choice during estimation
- Results
- Compute and save utilities
- Fitted probabilities
- In sample and out of sample predictions
- Log sums (inclusive values) for all tree levels
- Display tree structure
- Descriptive statistics by choice and attribute
- Crosstab of predicted and actual choice
- List of fitted probabilities with predicted and actual choice
- Elasticities and marginal effects
- All standard model output, coefficients, standard errors, fit measures, diagnostic statistics
- All post estimation analysis tools for hypothesis tests (LR, LM, Wald)
Model Estimation
NLOGIT supports a greater range of models for discrete choice than any other package. These include the basic multinomial logit model, nested logit models up to four levels, the multinomial probit model and the state of the art estimator for the mixed (random parameters) logit model. In all cases, there are a variety of different forms of the model available. The following lists the variety of discrete choice specifications available in NLOGIT.
- Binary Choice
- Binomial logit: cross section, fixed effects, random effects, heteroscedastic, random parameters, latent class
- Binomial probit: cross section, fixed effects, random effects, heteroscedastic, random parameters, latent class
- Bivariate probit
- Partial observability models
- Multivariate probit with unrestricted correlation matrix
- Ordered Choice
- Ordered probit, logit, Gompertz
- Panel data models: random effects, fixed effects, random parameters, latent class
- Sample selection models
- Multinomial Choice: Multinomial Logit
- Multinomial logit specification
- Choice specific constants
- Choice specific attributes and interactions of characteristics with constants
- Marginal effects
- Test procedure for IIA
- Restricted choice sets
- Estimation using revealed preference or sets of ranks
- Generalized maxiumum entropy
- Heteroscedastic Extreme Value
- Choice specific variances in MNL model
- Equality restrictions and grouping choices
- IIA test
- Homogeneity of variances test
- Nested Logit
- Up to four levels in nested logit models
- Command builder for tree specification
- Constrained IV parameters
- Marginal effects decomposed at the levels in the tree
- Save utilities, inclusive values, probabilities
- FIML or two step estimation
- Random utility specifications to constrain the model
- Covariance Heterogeneity
- Extends two level nested logit model
- Individual specific heteroscedasticity and heterogeneity in IV parameters
- Multinomial Probit
- Up to 20 choices
- GHK simulator
- Unrestricted or restricted correlation matrix
- IIA test
- Heteroscedasticity and covariance heterogeneity
- Panel data - multinomial, multiperiod probit
- Random Parameters (Mixed) Logit
- Up to 100 random parameters
- Maximum simulated likelihood estimation
- Pseudorandom draws or Halton sequences
- Mixture of random and nonrandom parameters
- Panel data structures
- Time invariant random effects
- AR(1) specification for random components
- Freely correlated random parameters
- Unrestricted mixture of normal, lognormal, tent, uniform parameters
- Restrictions on means and/or variances of random parameters
- Individual heterogeneity in means of random parameters
- Individual specific parameter estimates
- Latent Class
- Multinomial logit structure
- MNL sub model for class probabilities
- Panel data structure
- Up to five latent classes
Inference Tools for Hypothesis Testing
The full set of post estimation and analysis tools in LIMDEP is accessed by NLOGIT. This includes the Wald, likelihood ratio and Lagrange multiplier tests as well as all the matrix algebra and scientific calculator tools. NLOGIT also provides tools specific for discrete choice analysis, including a built-in procedure for testing the IIA assumption of the multinomial logit model.
Simulation
Any model estimated by NLOGIT can be subjected to 'what if' analyses using the model simulation package. The base case model produces a set of fitted probabilities for the sample data which aggregate to a prediction of the sample shares for the alternatives in the choice set. The simulator is then used, with the estimation data set or any other compatible data set, to recompute these shares under scenarios that you specify, such as a change in the price of a particular alternative or a change in household incomes.
After estimation of any model, you can simulate the probabilities computed by the model using the same or a different data set. The simulation can restrict the choice set or use the original one. Scenarios in the simulations involve changing attributes and recomputing probabilities and sample shares so as to examine the effect of the change on aggregate. The simulator may be used with any model.
Options for Constructing the Scenarios to be Simulated
- Simulate full choice set or only a subset of the choices
- Simulate any set of appropriately structured observations, whether used to fit the model or not
- Merge revealed and stated preference data for simulations
- Vary specific attributes of specific choices for simulations, proportional
- Changes or add or subtract specific amounts
- Construct scenarios of one or more changes in specific attributes
- Compare two or more different scenarios in the same simulation
|
