3
$\begingroup$

Does anyone know an R package for nearly orthogonal designs?

I would like to create an experimental design, using 12 runs, up to 10 factors, and with mixed levels (e.g. a combination of 2 and 3 level factors). I would like to explore some nearly-orthogonal designs.

There are lots of packages for orthogonal fractional designs. For example, I have looked at the documentation for AlgDesign, planor, FrF2, support.CEs, DoE.Base.

In SAS, there exist a set of macros for creating orthogonal and nearly orthogonal fractional factorial experimental designs: http://support.sas.com/techsup/technote/ts723.html

Does anyone know if something similar exists in R? The CRAN task view does not mention nearly orthogonal designs. http://cran.r-project.org/web/views/ExperimentalDesign.html

Many thanks Tim

Update: The Federov algorithm implemented in AlgDesign optFederov lets you create nearly orthogonal designs for mixed factors, as shown in the documentation

$\endgroup$
  • 1
    $\begingroup$ Note, I asked this on stackoverflow and was voted off topic, for asking for a recommendation. Please let me know if the same holds here stackoverflow.com/questions/25056315/… $\endgroup$ – psychonomics Jul 31 '14 at 12:51
  • 2
    $\begingroup$ Did you take a look at the task view for experimental design. If so, can you specify how this does not provide the desired information? $\endgroup$ – Henrik Jul 31 '14 at 12:59
  • $\begingroup$ Yes, asking for packages / code is off-topic here. You might ask on the r-help listserv. $\endgroup$ – gung - Reinstate Monica Jul 31 '14 at 12:59
  • 1
    $\begingroup$ This question appears to be off-topic because it is about asking for r packages / if r can do certain things. $\endgroup$ – gung - Reinstate Monica Jul 31 '14 at 13:01
  • 1
    $\begingroup$ With 12 runs and 10 factors, you can have at most one 3-level factor involved before you start confounding main effects. $\endgroup$ – rvl Jul 31 '14 at 13:26
2
$\begingroup$

Update: The Federov algorithm implemented in AlgDesign optFederov lets you create nearly orthogonal designs for mixed factors, as shown in the documentation

The example in the post below is a non-orthogonal fractional factorial design https://stackoverflow.com/questions/5044876/how-to-create-a-fractional-factorial-design-in-r

$\endgroup$
  • $\begingroup$ D-optimal designs tend to be close to orthogonal, yes. $\endgroup$ – kjetil b halvorsen Mar 25 '17 at 18:12
1
$\begingroup$

by going to suggested link i created below code and though it might be useful.

###Define factor and level in below code
cand.list = expand.grid(Storage = c("8 GB", "16 GB"),
                        Brand = c("Samsung", "Apple", "Nokia"),
                        RAM = c("1 GB", "2 GB"),
                        BrowseTime = c("24 hour", "36 hour"),
                        Weight = c("3.95 oz OR 111 gram", "5.04 oz OR 142gram"),
                        ScreenSize = c("4.7", "5.5", "5.7"))

###same as SPSS orthogonal design 'seed'. Can put any number. Does not matter.
set.seed(69)

###Generate 16 alternatives in an optimal orthogonal design
optFederov( ~ ., data = cand.list, nTrials = 16)

###End of code

I have a question for you all though. I got below values for design efficiency D =0.2519353; A = 5.462121; Ge = 0.748; Dea = 0.714. Which value should we be looking at for D-Eficiency? I assume it's D and how much it should be in order for this design to be usable in a choice experiment as alternatives? is current D value of 0.2519353 good enough for use?

$\endgroup$
  • 1
    $\begingroup$ Thank you for sharing your solution. Your question, though, must be posted in a new thread for it to be answered. $\endgroup$ – whuber Feb 14 '15 at 15:10
  • $\begingroup$ Well, never mind. I found answer myself after posting this question. stats.stackexchange.com/questions/137695/… $\endgroup$ – Enthusiast Apr 10 '15 at 21:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.