Overview (PROBIT command)

PROBIT can be used to estimate the effects of one or more independent variables on a dichotomous dependent variable (such as dead or alive, employed or unemployed, product purchased or not). The program is designed for dose-response analyses and related models, but PROBIT can also estimate logistic regression models.

Options

The Model. You can request a probit or logit response model, or both, for the observed response proportions with the MODEL subcommand.

Transform Predictors. You can control the base of the log transformation applied to the predictors or request no log transformation with the LOG subcommand.

Natural Response Rates. You can instruct PROBIT to estimate the natural response rate (threshold) of the model or supply a known natural response rate to be used in the solution with the NATRES subcommand.

Algorithm Control Parameters. You can specify values of algorithm control parameters, such as the limit on iterations, using the CRITERIA subcommand.

Statistics. By default, PROBIT calculates frequencies, fiducial confidence intervals, and the relative median potency. It also produces a plot of the observed probits or logits against the values of a single independent variable. Optionally, you can use the PRINT subcommand to request a test of the parallelism of regression lines for different levels of the grouping variable or to suppress any or all of these statistics.

Basic Specification

  • The basic specification is the response-count variable, keyword OF, the observation-count variable, keyword WITH, and at least one independent variable.
  • PROBIT calculates maximum-likelihood estimates for the parameters of the default probit response model and automatically displays estimates of the regression coefficient and intercept terms, their standard errors, a covariance matrix of parameter estimates, and a Pearson chi-square goodness-of-fit test of the model.

Subcommand Order

  • The variable specification must be first.
  • Subcommands can be named in any order.

Syntax Rules

  • The variables must include a response count, an observation count, and at least one predictor. A categorical grouping variable is optional.
  • All subcommands are optional and each can appear only once.
  • Generally, data should not be entered for individual observations. PROBIT expects predictor values, response counts, and the total number of observations as the input case.
  • If the data are available only in a case-by-case form, use AGGREGATE first to compute the required response and observation counts.

Operations

  • The transformed response variable is predicted as a linear function of other variables using the nonlinear-optimization method. Note that the previous releases used the iteratively weighted least-squares method, which has a different way of transforming the response variables. See the topic MODEL Subcommand (PROBIT command) for more information.
  • If individual cases are entered in the data, PROBIT skips the plot of transformed response proportions and predictor values.
  • If individual cases are entered, the chi-square goodness-of-fit statistic and associated degrees of freedom are based on the individual cases. The case-based chi-square goodness-of-fit statistic generally differs from that calculated for the same data in aggregated form.

Limitations

  • Only one prediction model can be tested on a single PROBIT command, although both probit and logit response models can be requested for that prediction.
  • Confidence limits, the plot of transformed response proportions and predictor values, and computation of relative median potency are necessarily limited to single-predictor models.