:: Visual Choice ::

Behavioural Choice Modelling Software

:: Navigation
:: Coefficient Estimation

For estimating discrete choice logit model coefficients

Visual Choice Estimation estimates the coefficients of the logit model utility function using maximum likelihood discrete choice logit model estimation. It can be used for both revealed and stated preference data. Model types can be switched-in or out with the click of the mouse. It takes as input a file of attributes, another of decisions, a problem specification file and it outputs a log file containing the coefficients, and statistics for the model being estimated. Attributes can be separation variables which are specific for each alternative, those which are not specific for each alternative such as socio-economic variables as well as alternative specific constants (ASCs). It can estimate the following types of model:

Visual Choice Estimation's graphical user interface helps you set up the estimation specification as well as visually select different model types with a variety of features. It uses two Panes:

The input files are entered onto the property pane text boxes using the ellipses' (circled in red in figure 1) and file contents can be viewed by pressing the view button. The properties pane also contains details about the problem to be estimated (eg the number of alternatives, separation variables, alternative specific constants (ASC) etc); the units, the fitting precision and other details depending upon the problem.

Having set up the properties, the estimations themselves are specified and run from the Estimation Pane. Set up the type of model to run from the drop down list box (see figure 2). Select which alternatives to omit by unticking its tick box. Select which attributes to include using their tick boxes. You can fix certain coefficients by providing a coefficient file. You should specify one time and one cost coefficient and Visual Choice will use these to compute the values of time for you.

Other options are available depending upon the type of model to be estimated (the greyed-out text and controls in figure 2). Nested logit needs the nest of each alternative. For cross nested all alternatives are in all nests. Latent class needs the number of classes. Mixed logit needs the attributes you want the mean and/ or standard deviation coefficients for. For the standard deviation coefficient you can select from: Log-normal, Uniform, Triangular, Johnston, Normal and you can specify if you will allow the negative part of the Log normal distribution to be used.

Having set up your run, you press the run button at the top of the pane and when you get the "run complete" message, you press the View log file button to view the results in notepad (see figure 3). These give the coefficient for each attribute together with its standard error, 't' statistic, precision and (if necessary) its value of time. Summary statistics include the null and model log likelihood, the rho squared, rho bar squared, Bayesian Information Criteria (BIC), Akaike Information Criteria (AIC), Consistent Akaike Information Criteria (CnAIC) statistics as well as the number of observations, convergence iteration and other statistics depending on the model type. For mixed logit, you get the mean and standard deviation coefficients, for nested logit you get the coefficients for each nest and the logsum coefficient. For latent class you get the coefficients for each class and for cross nested logit you get the contributions from each parent.

Example of a Property Pane showing how you can view and change the details of your estimation run.

Figure 1 Example of a Property Pane showing how you can view and change the details of your estimation run.

Example of an Estimation Pane showing the alternatives, attributes and details of what you wish to estimate in the next estimation run.

Figure 2 Example of an Estimation Pane showing the alternatives, attributes and details of what you wish to estimate in the next estimation run.

Example of an estimation results log file showing the estimated coefficients and their statistics.

Figure 3 Example of an estimation results log file showing the estimated coefficients and their statistics.