I am implementing a Factorial Survey tool using R programming where researchers will be able to experiment with a confounded fractional D-efficient design. The implementation idea came from this article https://journals.sagepub.com/doi/10.1177/0049124115582269 (The Factorial Survey: Design Selection and its Impact on Reliability and Internal Validity), where on page 14-15 "Confounded Factorial D-efficient design" has been introduced. My task is to provide the researchers with a portal where they will be able to construct the dimensions and levels and then will be able to exclude illogical cases, and then look for an optimal design which has a minimum of 90% D-efficiency and minimum confoundings.
In the article, confounded factorial D-efficient design consists of two steps to construct which are the following:
In the first step, it is recommended to generate a number of D-efficient design and to select one to be used for the survey.
My plan to achieve this step is to generate the vignette universe first (gen.factorial), exclude illogical cases and then, using optFederov function from AlgDesign package. Regarding using optFederov, I have some questions:
Q1. How to find the D-efficiency of the given design in optFederov? There are D, A, I, Ge, Dea as an output but I could not find any output which shows the D-efficiency within a scale of 0-100. i,e: D-efficiency 97%. Is there any extra function call or operation that I need to perform for getting a score of D-efficient design? I have also checked eval.design function and passing X as the same design but the output is Geff, Deffbound, diagonality and others which I think does not mention as a value of D-efficiency.
Q2. The first step recommends generating a number of D-efficient design, which I think is meant to call optFederov with different formula in the parameter or is there a way to get different designs based on one formula in the parameter?
Q3. In chapter 4.2.2.1 Two Level Design (https://cran.r-project.org/web/packages/AlgDesign/vignettes/AlgDesign.pdf), for a full fraction or half fractional design, the estimable term (or nTrails) were 34 for two level desing and 35 (but chosen 40 for orthogonal contrast) for the example in chapter 4.2.2.2 Mixed level designs. My question is how can I find out, how many terms or nTrails are there for a given design. I could not figure out how to calculate for it.
Q4. In Chapter 3: Confounding from the same paper, what I have understood is that measuring confounding is based on the diagonality of the design, which means the nearer the diagonality is 1, the minimized the confounding is. Did I understand it correct? And by using eval.design, when I set confoundings to true what does the value represent? what does it mean when the values are above or below 0.00 or even 0.00?
Q5. Why do I have to use eval.design after getting optFederov design? For bigger vignette universe, it takes alot of time to process optFederov and after getting the optimal design I have to check the D-efficiency, Confounding of the design which is time-consuming. Isn't there any way to directly pass the vignette universe (Candidate list after excluding the illogical cases) to the eval.design function or some other function and get D-efficiency, diagonality and suggested rows of the design?
The second step is to further optimize the D-efficiency, this time across different vignette sets (blocks).
I think after finding the optimal design with optFederov, I will again find for optimal D-efficient design by using optBlock from the package.
From Borhan Safa (October 5, 2019):
I am implementing a Factorial Survey tool using R programming where researchers will be able to experiment with a confounded fractional D-efficient design. The implementation idea came from this article https://journals.sagepub.com/doi/10.1177/0049124115582269 (The Factorial Survey: Design Selection and its Impact on Reliability and Internal Validity), where on page 14-15 "Confounded Factorial D-efficient design" has been introduced. My task is to provide the researchers with a portal where they will be able to construct the dimensions and levels and then will be able to exclude illogical cases, and then look for an optimal design which has a minimum of 90% D-efficiency and minimum confoundings.
In the article, confounded factorial D-efficient design consists of two steps to construct which are the following:
My plan to achieve this step is to generate the vignette universe first (gen.factorial), exclude illogical cases and then, using optFederov function from AlgDesign package. Regarding using optFederov, I have some questions:
I think after finding the optimal design with optFederov, I will again find for optimal D-efficient design by using optBlock from the package.