wqm_cdfest: Compute Nonparametric CDF Estimates

In Auburngrads/SMRD: Statistical Methods For Reliability Data

Description

Computes nonparametric estimates of the cumulative distribution function using maximum likelihood. If possible, estimation this is done with a generalized version of the Kaplan-Meier estimate, and otherwise, by using Turnbull's e-m algorithm

Usage

WQMCDFEST(
  y,
  ny,
  codes,
  weight,
  ty,
  nty,
  tcodes,
  n,
  nstart,
  dscrat,
  scrat,
  iscrat,
  kprint,
  maxit,
  tol,
  maxmsd,
  p,
  q,
  prob,
  sd,
  m,
  pchmax,
  lsd,
  ier
)

Arguments

y	Numeric matrix [n x ny] containing the lower and upper limits for group censored observations.
ny	Number of columns in y (either 1 or 2)
codes	Integer vector of censor codes (see Details)
weight	Vector of observation weights or multiplicities
ty	Numeric matrix containing lower and upper truncation limits
nty	Number of columns in ty
tcodes	Integer vector of truncation codes (see details)
n	Number of rows in y
nstart	If nstart = 0 set automatic start values, otherwise send down nstart values in prob for restart
dscrat	Numeric scratch vector (length = 3 * n + 2)
scrat	Numeric scratch array (length = max((7 * n), maxmsd * (maxmsd - 1) / 2))
iscrat	Integer scratch array (length = 6 * n + 4)
maxit	Maximum number of iterations for s-c algorithm
tol	Desired estimation accuracy (0 < tol < 0.1)
maxmsd	Maximum m for which the full information matrix estimates of the standard errors can be computed
iprint	Print level for debug dump

Details

• 0dummy observation

• 1exact failure time

• 2right censored observation

• 3left censored observation

• 4interval or group censored observation

• 5exact failure time recoded as a small interval

• 1no truncation

• 2right truncated observation

• 3left truncated observation

• 4interval truncated observation

iprint = 0 for no debug output, if > 0 dump setup and every iprint iteration, iprint = 1 gives the maximum amount of output, iprint = 10 is a good choice to debug

Value

A list of length 6

• pNumeric vector of lower limits of intervals for the cdf estimate (length = n + 1)

• qNumeric vector of upper limits of intervals for the cdf estimate (length = n + 1)

• probNumeric vector of cdf estimates corresponding to intervals p and q

• mActual length of p, q, and prob (depends on data)

• pchmaxMax change in last iteration (if greater than tol, 0.0 if Kaplan-Meier was used)

• ierError code (see details)

0 no error 1 n <= 0 on input 2 ny not equal to 1 or 2 3 nty not between 0 and 2 4 tol outside range (0, 0.1) 6 censor code out of range 0 to 4 7 y(i,1) != y(i,2) in a type 1, 2, or 3 observation 8 y(i,1) < y(i,2) in type 4 obs 9 ny=1 but type 4 obs found 10 tcode outside range 1 to 4 11 tcode = 1, 2, or 3 but tyl != tyu 12 backwards truncation interval 12 ty(i,1) < ty(i,2) in type 4 obs 13 only 1 col of truncation values, code 4 found 14 not enough data to estimate distribution (e.g., all right censored observations) 15 nstart>0 does not agree with computed m 16 observation not within the truncation interval *21 product-limit estimate could not be computed directly and maxmsd was too small to allow the full information matrix to be computed an approximation was computed under the assumption that the individual hazard estimates are uncorrelated *22 information matrix not positive definite *23 only one non-zero s probability

* signifies warning message only

Auburngrads/SMRD documentation built on Sept. 14, 2020, 2:21 a.m.