mlefit: Maximum likelihood regression on Weibull and Lognormal...

In WeibullR: Weibull Analysis for Reliability Engineering

Maximum likelihood regression on Weibull and Lognormal distributions

Description

Maximum likelihood regression on weibull and lognormal distributions

Usage

mlefit(x, dist="weibull", npar = 2, debias="none", optcontrol=NULL)

Arguments

x	A dataframe such as generated by mleframe with column names 'left', 'right' and optionally 'qty'. Exact failure data (occurrences) have same time entered in both 'left' and 'right' columns. Suspension data has last known life[time] entered in 'left' column and -1 entered in 'right' column. The left(early) interval bound for left-censored data must be entered as zero. (NA is not accepted).
dist	A string defining a distribution to be fit. Implemented distributions are "weibull" (default),"lognormal", "weibull3p" and "lognormal3p".
npar	An optional argument for specifying 3p optimization. Priority is given to any suffix to the distribution name.
debias	An optional string argument indicating the adjustment to be applied to the shape or standard deviation parameter of the fitted data. Adjustments recognized are: "rba" (default), "mean", and "hrbu" (for hirose-ross beta unbias). Any entry for debias with the lognormal distribution will be processed as "rba" with a warning if it was not clearly specified.
optcontrol	An optional list of arguments for control of the MLE optimization. The strict naming convention for the list items are: 'vstart', a vector holding starting estimate for parameters in the order used by the underlying R distribution [CAUTION: for weibull the order of parameters is (shape, scope)], 'limit', the convergence limit, 'maxit', the maximum number of interations permitted in the simplex optimization, and 'listout' a logical indicating whether output should list both the fit and the optimization progress as a dataframe. Additional list items 'num_points', 'err_t0_limit', and 'err_gor_limit' are handled for 3rd parameter processing -see details.

Details

The negative log-likelihood is minimized by the Nelder-Meade, simplex, algorithm. This algorithm requires a reasonable starting point for the estimate. Such a reasonable estimate is attempted by default, but in some cases this may be inadequate for proper function. the optcontrol argument has been provided to give debug developers a means of altering and examining the progress of the optimization. For 3rd parameter processing defaults are used for the simplex options, while 'num_points' can be used to alter the seek points, 'err_t0_limit', and 'err_gor_limit' control cut-offs for unstable optiimizations, while 'list_out' provides a different output on the optimization progress.

Value

Returns a named vector with parameters of the distribution fit and goodness of fit in terms of log likelihood.

References

William Q. Meeker and Luis A. Escobar, (1998) "Statistical Methods for Reliability Data", Wiley-Interscience, New York

Robert B. Abernethy, (2008) "The New Weibull Handbook, Fifth Edition"

John I. McCool, (2012) "Using the Weibull Distribution: Reliability, Modeling and Inference"

Marie Laure Delignette-Muller, Christophe Dutang (2015). "fitdistrplus: An R Package for Fitting Distributions". Journal of Statistical Software, 64(4), 1-34. URL http://www.jstatsoft.org/v64/i04/.

Examples

failures<-c(90,96,30,49,82)
suspensions<-c(100,45,10)
weibull_fit<-mlefit(mleframe(failures,suspensions))

WeibullR documentation built on June 26, 2022, 1:06 a.m.