[Feature Request]: Increase distribution fitting capabilities

[kochjo-dev]

Description

The distributions platform allows to fit various continous distributions and to determine scale and shape, however no 3rd parameter e.g. offset or bias. It also lags the capability to try various (all?) distributions and select the one with the best goodness of fit. Last not least it lags grouping and censoring capabilties.

Purpose

Determine the best fitting distribution to real world reliability data

Use-case

Need to fit 3 parameter Weibull to censored data

Is your feature request related to a problem?

The data contains timestamps of fails from more than one supplier. The question is if there is a difference in the reliabilty between those and to predict the failure probability after 10a lifetime

Is your feature request related to a JASP module?

Distributions

Describe the solution you would like

One would need to have 3 parameter Weibull introduced, not just lambda and k. There should be a (left, right and interval) censoring capababilty (there is one in the survival analysis). There should be batch capabilties aka run all distributions and see which one gives the best fit (AICc, BICc, p-values, ..)

Describe alternatives that you have considered

Alternatively the dataset would need to be split up the data, either throw away the censored datapoints and use the existing platform, or the use Weibayes - which is not available in JASP?, all hoping for no bias being present.

Additional context

No response

[EJWagenmakers]

Good suggestions, and very much on our radar!

[patc3]

Possibly relevant: #3178 both Weibull and censoring are available in GAMLSS

[FBartos]

I think that this is more relevant for survival (at least in the short-medium term). I don't think there is any immediate plan of adding censoring to the distributions module itself.

[kochjo-dev]

Thx Frantisek. The main feature is indeed fitting reliability data and getting the right distribution parameters. This may include non-parametric, but should also allow to check for established/accepted lifetime distributions (exponential, SEV, weibull, lognormal, for a start) and do model comparisons as described.