Truncated Regression

Example: Truncated Regression

To create this example:
  1. Create the Work.Cigar data set. For more information, see CIGAR Data Set.
  2. In the Tasks section, expand the Econometrics folder, and then double-click Panel Data Models. The user interface for the Panel Data Models task opens.
  3. On the Data tab, select the WORK.CIGAR data set.
    Tip
    If the data set is not available from the drop-down list, click Select a table icon. In the Choose a Table window, expand the library that contains the data set that you want to use. Select the data set for the example and click OK. The selected data set should now appear in the drop-down list.
  4. Assign columns to these roles:
    Role
    Column Name
    Panel Structure
    Cross-sectional ID
    state
    Roles
    Dependent variable
    sales
    Continuous variables
    price
    cpi
    ndi
  5. On the Model tab, complete these steps:
    1. Select Truncated regression as the model type.
    2. Select the Set the lower bound check box. From the Lower bound method drop-down list, select Specify by value. In the Lower bound value box, enter 90.
  6. To run the task, click Submit SAS Code Icon.
Here are the results:
Summary Statistics of Continuous Responses, Model Fit Summary, and Parameter Estimates

Assigning Data to Roles

To perform a truncated regression analysis, you must assign an input data set. To filter the input data source, click Filter Icon.
You also must assign variables to the Cross-sectional ID and Dependent variable roles.
Role
Description
Panel Structure
Cross-sectional ID
specifies the cross section for each observation. The task verifies that the input data is sorted by the cross-sectional ID.
Note: For the truncated regression model, character variables are not supported.
Time ID
specifies the time period for each observation. For each cross section, the values of the time ID must be unique.
Note: For the truncated regression model, a time ID is not required and is ignored in the analysis.
Roles
Dependent variable
specifies the numeric variable for the analysis.
Continuous variables
specifies the independent covariates (regressors) for the regression model. If you do not specify a continuous variable, the task fits a model that contains only an intercept.
Categorical variables
specifies the classification variables. The task generates dummy variables for each level of the categorical variable.
Additional Roles
Group analysis by
enables you to obtain separate analyses of observations for each unique group.

Setting the Model Options

To create a truncated regression model:
  1. From the Model type drop-down list, select Truncated regression.
  2. Specify the model effects.
    You can display the main effects model or create a custom model. To create a custom model, select the Custom Model option, and then click Edit. The Model Effects Builder opens. All continuous variables and categorical variables are listed in the Variables pane.
    • To create a main effect, select the variable in the Variables pane, and then click Add.
    • To create a crossed effect, select the variables in the Variables pane. (You can use Ctrl to select multiple variables.) Then click Cross.
    When you finish, click OK. The effects that you specified now appear on the Model tab.
    Here is an example of model effects on the Model tab.
    price and price*productLine effects
    Note: Random effects are automatically included in the model. This functionality is experimental.
  3. Set the upper and lower bounds of the censored variables.
    Note: If you do not specify an upper bound or a lower bound, the result is a linear regression model.

Setting the Options

Option Name
Description
Methods
Covariance matrix estimator
specifies the method to calculate the covariance matrix of parameter estimates.
You can use the default value, or you can choose from these covariance types:
  • Inverse Hessian matrix – the covariance from the inverse Hessian matrix.
  • Outer product matrix – the covariance from the outer product matrix.
  • Outer product and Hessian matrices – the covariance from the outer product and Hessian matrices (the quasi-maximum likelihood estimates).
Optimization
Method
specifies the optimization method to use.
Maximum number of iterations
specifies the maximum number of iterations in the optimization process. You can use the default value, or you can specify a custom value.
Statistics
Select the statistics to display in the results.
Here are the additional statistics that you can include in the results:
  • correlations of the parameter estimates
  • covariances of the parameter estimates
  • iteration history of the objective function and parameter estimates

Creating the Output Data Sets

You can create these output data sets:
  • an output data set that contains the default statistics from the analysis and additional statistics, such as predicted values, residuals, the error standard deviation, and the linear predictor
  • a parameter estimates data set