What inspired this Wish List Request?
This is a request from Neptune, a customer.
Right now, the competing risk model does not allow inference for the aggregate model and also does not allow fitting a mixture on top of a competing risk model. They would also like to get model diagnositcs (log-like, AICc, BIC) to compare various models.
Please describe the new or improved feature or solution you would like to see.
They would like to be able to get custom estimation of quantiles and probabilities when fitting a competing risk or mixture distribution in the Life Distribution platform. This should be updated whenever the "Update Model" button is clicked for competing risk and be available when fitting a mixture. Also, the customer would like to be able to fit a mixture distribution and competing risk model simultaneously.
From the customer:
Add aggregate likelihood scores (-2Loglikelihood, AIC, AICc, BIC) for a given selection as above. This is simple when all failures are either uncensored or right-censored, but it’s a little more complicated when a labeled failure is left-censored or interval-censored. That is, the likelihood of an observed failure is the joint probability that the given mode fails in that interval, and that the given mode fails before any other failure mode. Mathematically, this might work out to an integration, and it might not scale well.
When running a Life Distribution analysis with failure cause provided, allow an option for a mixture (as opposed to a competing cause analysis). This is relatively simple to implement: the likelihood for a failure labeled with failure mode i is just pi∙Likelihoodi , and the likelihood of a suspension is just 1-Σ i (p i∙F i). Such an implementation would reasonably include custom estimation and aggregate likelihood scores as above. There’s a little bit of question as to how to handle Σ i p i – if the sum of probabilities is allowed to be less than 1, then DS models should not be included. If the sum of probabilities is constrained to equal 1, then DS models could be allowed, but (a) it doesn’t make sense to include more than 1, and (b) it doesn’t matter which mode gets the DS parameter. In general I’d probably never allow DS models, but allow comparison between the constrained case and the “DS” case.
When running either a Competing Cause analysis or a mixture analysis with failure cause provided, allow comparison between models with two selected failure modes analyzed as separate modes vs. lumped together.
What would the value be to you and/or other users if this new feature is provided?
It would be helpful to know if a multi-modal failure distribution fits better than a competing risk model. It would also be helpful to predict quantiles, etc., if for competing risk, mixture distributions or any combination.
Is this an external customer request?
This is an external customer request. The customer is Neptune. Contact Jeff Fowler at jfowler@neptunetg.com.
Is there a competitive software package that has this feature or solution that we should be aware of?
Unknown.
Will this impact a renewal, expansion, or adoption at a customer, and does this put an account at risk if we do not address this?
Unknown.
