taalbrecht / discretechoicedesigner

D-optimal design creation tool for discrete choice experiments

DiscreteChoiceDesigner

=======

DiscreteChoiceDesigner is a shiny app user interface that uses the MultiEqOptimizer package to create single experimental designs jointly optimized for multiple candidate models as described in:

https://github.com/taalbrecht/MultiEqOptimizer

Currently optimized multiple discrete choice conditional logistic models.

Installation

TBD

devtools::install_github('taalbrecht/MultiEqOptimizer')

Documentation

TBD

Example

TBD

})

discretechoicedesigner's Issues

Level naming for attribute parameters

Allow users to enter comma separated values to name all levels for attribute parameters.

Effect and difference to detect fields clear values when switching tabs

If the effect size and difference to detect fields are populated, they clear their values when switching between tabs in some cases. The other dynamically sized input fields on the same page are not affected (min and max, number of levels, etc)

Add recommended minimum number of questions

Can calculate as lowest common denominator of all possible variable levels in model matrix

Continuous parameter with 2 levels causes error using Federov sampling

Tried to make an experiment with three variables, A, B, and C with B as attribute and A and C as continuous with 2 levels and min/max of 0,1 and -1, 1 respectively. Two formulas were used:

A+B+C - weight 0.4
A+B+I(A^2) - weight 0.6

All parameter priors set to 0

2 alternates
6 questions
1 randomized start

If A has 2 levels, this will cause an error as A^2 and A must be perfectly correlated in the second function. C can be set to two levels with no problem.

Add error statement when this occurs instead of causing a random untraceable error from the users perspective

Add design finder feature

Add feature to "find" best design with regards to number of questions & number of alternates. After running, plot d-efficiency of each model vs. the number of runs and vs the number of alternates. Ask user to estimate maximum number of alternates, questions, and questions per respondent.

Add ability to optimize between discrete choice and linear regression model

Add ability for user to select which type of optimization to use for each formula. Jointly optimize between weighting parameter and optimization type for each equation.

Unrecoverable error when optimizing for 3 or more formulas in single design search

Unrecoverable error that can be replicated by using the following settings

2 continuous inputs with 5 levels from -1 to 1 with A+B+A*B as the input formula for each one. Error does not occur if only 2 formulas are used.

Occasional difficulties where runs won't process on first execution per session

Trend of error occurring during the first model run for each session, both in the single and multi design tabs. Cause is unknown. Subsequent runs beyond the first attempt seem to be successful and do not have the same issue. Specified design parameters were identical between the first attempt and subsequent attempts in the observed occurrences of this bug.