Dummy secondary occasions for Robust design

[TimG]

There are a number of posts on this forum suggesting that even if your data were not collected under the Robust Design, you can use these models by adding a dummy secondary occasion (a column of 0's) to each primary occasion, and set p2=c=0 for the dummy occasions.

Does anyone know of a reference justifying this approach, particularly using a dummy secondary occasion for all primary occasions? I found 1 paper using this approach for some primary periods http://onlinelibrary.wiley.com/doi/10.1002/jwmg.380/full, but I'm wondering about the theory behind its validity.

[TimG]

After thinking about this some more, I believe the only reason adding dummy secondary occasions was suggested is when another issue (e.g. state uncertainty) required a MARK model with robust design format (e.g. CRDMSOHug). Adding the dummy occasions seems valid, but I can't imagine it would provide the same benefits as a true robust design.

To answer my own question, using the same dataset and analogous model structures, I tried 1) adding dummy secondary periods to CJS-type data and used a robust design model, fixing p2=0; and 2) retaining the CJS data format and used a multistate model with 2 states (observable and unobservable). These returned basically identical results. However in agreement with Kendall and Nichols (2002) and Schaub et al (2004), there seems to be issues with identifiability and poor precision when using these approaches for certain cases, which would be alleviated by true robust design data.

[Bill Kendall]

Tim,
Looks like you figured it out yourself. Yes, the main reasons you would use dummy time periods for the robust design in MARK is if (1) you wanted to use a parameterization only available in the robust design format (e.g., state uncertainty models as you point out) or (2) where some of your primary periods have multiple secondary periods and some don't.