Start jomo adjustment without burn-in

[nextpagesoft]

Burn-in is done in the jomo precalculation step. Start imputation without burn-in.

[nextpagesoft]

jomo produces data resulting in errors in the main report if nburn is set to 0. Most likely it will have to be closed as it is.

[nextpagesoft]

Matteo Quartagno contacted to advise on the best strategy.

[nextpagesoft]

Response from Matteo Quartagno received:

There are two options that I can think of: a. To do as you did, passing all of the starting values for the parameters (e.g. beta.start and l1cov.start in this case) as the last values drawn during the burn-in. The only issue is that it is not possible to pass the starting values for the imputations in function jomo. You need, as you say, to run a second burn-in for jomo, although this can be much smaller, as you only need to reach the convergence distribution for the imputations. But I see that this is not optimal… So the second option is.. b. To switch to the jomoImpute interface in package mitml. This uses the jomo.MCMCchain functions recursively, does the burn-in and the imputations and stores the parameter values at each step at the same time. The only issue with this is that, since it saves info on a lot of parameters at potentially thousands of steps, it sometimes runs into lack-of-memory issues. But it depends on the dimension of your dataset and on the number of imputations. I think I will make it possible to pass the starting value for the imputations with function jomo as well in the future. Possibly from the next version of the package.

Closing for now