Developer's Meeting Notes 7/20/12

At the Developers Meeting on 7/20/12 we discussed the following:

• Mike Hunter made some minor modifications to LISREL adding a passing test case for FIML with LISREL that has FIML rearrange the observed data columns instead of forcing the user to do so.
• Mike Hunter also discussed his experience at the recent Psychometrics Conference where he presented the LISREL functionality to OpenMx. In particular there was a good discussion at the conference of the confounding between a model specification with the software used to estimate the model. Mike's work was well received and three of the five talks presented were using OpenMx in some way. It appears that many conference attendants are interested in using OpenMx specifically for the optimizer. In the future the group expects that these users will return to OpenMx when they have other SEM needs.
• The group discussed a slow-running script Hermine had circulated to OpenMx Developers list. After profiling the script throughout the week it appears that the use of rbind and cbind within the script are not causing the inefficiency. Instead, it appears that the inefficiency is caused by a large number of small algebras. The group discussed replacing these algebras with naming functions to calculate contiguous blocks of data needed. Adding helper functions to make generating the naming scheme would be an interesting addition to OpenMx but with everyone busy on other development tasks, it may have to wait for a new OpenMx developer to come into the fold.
• Joshua Pritkin discussed his progress implementing Item Response model. He is working translating a text book algorithm for dichotomous items into R for dichotomous and polytomous items. Everyone agreed the 2PL case of this algorithm will be particularly useful. The group also agreed that the 3PL case is significantly less useful because it is less common and unstable.
• The group discussed the possibility of mirroring the OpenMx website at VCU and somewhere outside of the country.
• In the near future, Mike Hunter will be incorporating some old code into the trunk of OpenMx to get the maximum likelihood of a state space model.