At the Developers Meeting on 8/30/12 we discussed the following:
- The group discussed the mxFit and mxExpectation proposal. After ironing out some light confusion due to a few typos in the document, the group agreed that it looks very good. Implementation will start on it in the upcoming weeks.
- The group also discussed the possibility of modification indices in the presence of definition variables. The issue was brought up due to twin models. Employing modification indicies, modifying one and not the other is meaningless. In particular employing modification indicies and definition variables in this area is an unexplored area.
- The group also discussed the IRT User interface proposal that was distributed on the OpenMx Developers email list. In particular the issues discussed were:
- The group discussed different methods for users to specify the parameter matrix. Two options were offered up: (1) specifying everything in one large matrix vs (2) a mxPath style specification using a new construct called mxItem().
- Implementing Bayesian priors in IRT as mxAlgebras. This would require implementing a number of distributions (via keywords) in OpenMx.
- The group keeping items in the same, consistent order as the thresholds matrix by listing items in column order instead of row order.