YouTube video by J. Krishnamurti - Official Channel

A lesson on Unlearning and unconditioning from a great saint Ashtavakra.

From Vita Activa to Vita Contemplativa? Rethinking the Active Life in a World of Busyness Imagine a life where every action, every thought, every breath is dictated by the relentless pursuit of doing...

Side comment: a compound symmetric correlation structure such as that induced by including random intercepts in the model can be approximated by conditioning on the average of all previous responses. Unconditioning on this will be very difficult. Also, using your method with large M will be computationally very expensive. We should consider a parallel path where we make the simulation of longitudinal processes from fitted models more efficient than the current `Hmisc` function. By simulating an N=100,000 sample we can get whatever unconditional estimates are needed very easily. It would be nice to also find a way to do posterior sampling in the same spirit, to capture all the uncertainties of parameter estimates.

Side comment: a compound symmetric correlation structure such as that induced by including random intercepts in the model can be approximated by conditioning on the average of all previous responses. Unconditioning on this will be very difficult. Also, using your method with large M will be computationally very expensive. We should consider a parallel path where we make the simulation of longitudinal processes from fitted models more efficient than the current `Hmisc` function. By simulating an N=100,000 sample we can get whatever unconditional estimates are needed very easily. It would be nice to also find a way to do posterior sampling in the same spirit, to capture all the uncertainties of parameter estimates.

Side comment: a compound symmetric correlation structure such as that induced by including random intercepts in the model can be approximated by conditioning on the average of all previous responses. Unconditioning on this will be very difficult. Also, using your method with large M will be computationally very expensive. We should consider a parallel path where we make the simulation of longitudinal processes from fitted models more efficient than the current `Hmisc` function. By simulating an N=100,000 sample we can get whatever unconditional estimates are needed very easily. It would be nice to also find a way to do posterior sampling in the same spirit, to capture all the uncertainties of parameter estimates.

Side comment: a compound symmetric correlation structure such as that induced by including random intercepts in the model can be approximated by conditioning on the average of all previous responses. Unconditioning on this will be very difficult. Also, using your method with large M will be computationally very expensive. We should consider a parallel path where we make the simulation of longitudinal processes from fitted models more efficient than the current `Hmisc` function. By simulating an N=100,000 sample we can get whatever unconditional estimates are needed very easily. It would be nice to also find a way to do posterior sampling in the same spirit, to capture all the uncertainties of parameter estimates.