Ngene is capable of generating design types for a wide range of discrete choice experiments and model types. Ngene assists you in getting the most out of a survey by maximising information and making choice tasks more realistic and familiar to respondents.
Ngene is syntax based in order to allow maximum flexibility in specifying the dimensions of your design. You can generate designs with any number of labelled and/or unlabelled alternatives (including a status quo alternative or a ‘no choice’ alternative), any number of choice tasks, any number of attributes, and any number of attribute levels. Further, in Ngene you can control design properties such as attribute level balance, orthogonality, correlation structure, minimum overlap, and blocking.
Ease of use
All outputs in Ngene can be easily copied and pasted into other software such as Excel for further analysis and use. Further, Ngene can read in externally generated designs or data and evaluate efficiency, correlations, choice probabilities, etc. Syntax files, data files, and design outputs can be saved in a project. Multiple algorithms exist for generating designs that run without the need to change the default settings, although more advanced users may wish to take advantage of additional control. Further, inspecting an experimental design matrix is made easy in Ngene by being able to show quick mockups of formatted choice tasks. Ngene includes an extensive manual that details how to use Ngene, and introduces stated choice experimental design theory.
In order to select the best design, Ngene allows the user to fully specify the model(s) that would most likely be estimated. This includes selecting the model type and formulating the utility functions with main effects and interaction effects, linear and nonlinear effects (using dummy or effects coding). Current supported model types are the multinomial logit model as well as (cross-sectional and panel) mixed logit and error component models. A range of efficiency criteria is available in order to maximise information when estimating the model parameters or willingness-to-pay, including D-error and A-error. Prior information about the parameters can be provided in order to further optimise the design. Ngene supports fixed priors as well as uniformly and normally distributed priors (i.e. Bayesian efficient designs) to account for uncertainty in the prior values. When priors are provided, Ngene automatically calculates the expected sample size required for estimating the model parameters at a given level of statistical significance. In order to minimise computation time for Bayesian and mixed logit designs, Ngene can take smart draws from random distributions, including quasi-random draws (such as Halton draws, Sobol draws, MLHS) and Gaussian quadrature. Ngene also supports optimising over multiple model specifications and efficiency measures at the same time.
Ngene gives the user complete control with respect to the combinations of attribute levels that are considered feasible in order to enhance realism in the design. This is done by including conditional constraints (if…then), through setting requirements on attribute level combinations that cannot occur or that should occur, or by introducing scenario variables. For very complex and heavily constrained designs, the user also has the option to externally create candidate sets with feasible choice tasks (e.g. in Excel) and read them into Ngene. This method can also be used for other advanced designs in which the user wishes to limit the number of alternatives or attributes shown to a respondent in each choice task in order to reduce choice task complexity.
It is important that problematic choice tasks, such as choice tasks in which there are no trade-offs or where one of the alternatives is strictly dominant, are not present in the experimental design in order to avoid possible parameter biases in model estimation. While this is often considered a manual job, Ngene can automatically detect such problematic choice tasks and avoids including them in the experimental designs.
No silly choices
Instead of using a fixed set of choice tasks shown to respondents, Ngene can create designs in which the attribute levels are pivoted around the reference levels of the individual respondents. This makes the choice context more familiar and reduces hypothetical bias. To achieve this, Ngene allows the inclusion of a reference alternative while the levels of the other alternatives are absolute or relative differences to the reference levels. Furthermore, it is also possible to create different designs for different segments within the population by including one or more socio-demographic variables in the utility functions.