Products > LOGIT > Product Overview > Features & Capabilities

Features & Capabilities

LOGIT can produce the results most analysts need with just three simple commands.Yet many advanced features are available for sophisticated research and projects. LOGIT will estimate binary, multinomial, conditional logistic regression models, ranked logit and discrete choice models.


Discrete Choice Modeling

The discrete choice model introduced to econometrics by McFadden and to psychology by Luce is available. Data may be organized by case or by choice (alternatives) and results include prediction success and elasticity tables. Elasticities may be evaluated at sample means of the covariates or calculated for each case and averaged over the sample.


Prediction Success (Confusion Matrix) and Derivative Tables

Classification tables using predicted probabilities or class assignment rules can be printed for binary or multinomial models using McFadden's prediction success table. Derivative tables showing how probabilities change when an independent variable changes are printed for all models, multinomial as well as binary.


Stepwise Regression

Forward, backward, mixed and interactive stepping can be conducted with provision for forcing variables into the model, restricting variable addition to a pre-specified list of variables, and searching over interactions. Interactive stepping allows the user to determine variable addition or deletion at each step.


Hypothesis Testing: Wald, Score and LR

Hypothesis testing is especially easy. Wald tests are specified with natural algebraic expressions, and tables testing all possible pairs of levels of categorical variables can be printed automatically for all occurrences of categorical variables, main effects and interactions. The SCORE test is also available for testing any hypothesized coefficient values. Likelihood Ratio testing is facilitated with the built-in CDF command: evaluate any statistic for twenty different distributions from the command line.


Separate Learning & Test Samples

Data may be separated into a learning sample on which the model is estimated and a test sample. Predicted probabilities and prediction success tables will be produced for both subsamples, allowing more accurate assessment of model performance.


Start Values

Optional start values may be specified for any model. This allows previously-obtained results to be used to calculate predicted probabilities on new data sets, and to request further output for original data sets.


Odds Ratios, Deciles of Risk, Diagnostics

LOGIT produces odds ratios with upper and lower bounds based on a user-specified confidence interval. Deciles of risk tables and the Hosmer-Lemeshow, Pearson Chi-square and Deviance statistics are optionally produced for the binary logit, and a full set of Pregibon regression diagnostics can be saved for each case in the data set. Output also includes predicted probabilities and predicted classification.


Simulation

Simulation evaluates the logit and predicted probability for any set of values of the covariates along with odds ratios and confidence bounds. Simulation of logit differences is ideal for assessing the effects of interactions, and simulation over a range of different values of the covariates allows graphing of the response curves and surfaces.


Integrated BASIC

LOGIT contains a full BASIC interpreter allowing you to create new variables on the fly. BASIC can also be used to exclude cases from analysis based on complex conditions.


New to LOGIT

Fractional dependent variable allowed to record market shares, adjustments to standard errors to account for repeated measures (multiple responses from the same respondent); store choice and demographic data in separate files. Constrained estimation allows you to fix parameters at predetermined values.