The BayesMFSurv package

Minnie M. Joo, Nicolas Schmidt, Sergio Bejar, Bumba Mukherjee, Vineeta Yadav

Description

Contains a split population survival estimator that models the misclassification probability of failure versus right-censored events. The split population survival estimator is described in Bagozzi et al. (2019) doi:10.1017/pan.2019.6.

Installation

You can install the released version (0.1.0) of BayesMFSurv from CRAN with:

install.packages("BayesMFSurv")

And the development version (0.2.0) from GitHub with:

if (!require("remotes")) install.packages("remotes")
remotes::install_github("Nicolas-Schmidt/BayesMFSurv")

Functions

Function	Description
mfsurv	fits a parametric Bayesian MF model via Markov Chain Monte Carlo (MCMC) to estimate the probability of misclassification in the first stage and the hazard in the second stage. Slice sampling is employed to draw the posterior sample of the model’s split and survival stage parameters.
mcmcsurv	estimates a Bayesian Exponential or Weibull survival model via Markov Chain Monte Carlo (MCMC). Slice samplig is employed to draw the posterior sample of the model’s survival stage parameters.
stats	a function to calculate the deviance information criterion (DIC) and the log-likelihood for fitted model objects of class mfsurv or mcmcsurv.
summary	returns a summary of a mfsurv or mcmcsurv object via coda::summary.mcmc.

Example

The data used to estimate the following examples come from Reenock, Bernhard and Sobek (2007) -DOI: 10.111/j.1468-2478.2007.00469.x-. The RBS (2007) dataset uses continuous-time event history techniques to code episodes of democratic breakdown in all democracies from 1961 to 1995. In addition, it provides data on a number of economic and political variables.

Variable	Description
calinv	inverse of per capita daily caloric supply
lnlevel	natural log of economic development
calileve	inverse of per capita daily caloric supply*lnlevel
necon	economic performance
presi	presidential regime
tag	effective number of parties
rel	religious fractionalization
ethn	ethnic fractionalization
prevdem	numbers of previous democratic episodes
openc	trade openness
Y	years in current democratic episode
Y0	years in current democratic episode (lagged)
C	breakdown of democratic episode

Misclassified-Failure

mfsurv estimated the probability of misclassification failure in the first (split) stage and hazard in the second (survival) stage.

mfsurv should be used when user suspects that data of survival cases could be right-sensored (i.e. when there is a probability that failure events are misclassified).

Example with N = 100000 is here.

# Baseline Bayesian misclassified failure (MF) model. 
# Misclassification stage only includes the intercept while the survival stage 
# includes all covariates described above.  

library(BayesMFSurv)

set.seed(95)
data(RBS)
RBS <- na.omit(RBS)
Y   <- RBS$Y
X   <- as.matrix(cbind(1, RBS[,1:10]))
C   <- RBS$C
Z1  <- cbind(rep(1,nrow(RBS)))
Y0  <- RBS$Y0
model1 <- mfsurv(Y ~ X | C ~ Z1, Y0 = Y0,
                 N = 100,
                 burn = 50,
                 thin = 5,
                 w = c(0.5, 0.5, 0.5),
                 m = 20,
                 form = 'Weibull')


stats(model1)
#> $DIC
#> [1] 256.8576
#> 
#> $Loglik
#> [1] 478.8454

summary(model1, parameter = c("betas"))
#> 
#> Iterations = 1:10
#> Thinning interval = 1 
#> Number of chains = 1 
#> Sample size per chain = 10 
#> 
#> 1. Empirical mean and standard deviation for each variable,
#>    plus standard error of the mean:
#> 
#>                 Mean    SD Naive SE Time-series SE
#> X.intercept  -5.0477 3.384   1.0700         1.0700
#> X1          -10.6678 3.797   1.2006         1.2006
#> Xcalinv      12.8464 7.331   2.3182         6.9599
#> Xlnlevel     -3.2283 3.595   1.1367         1.1367
#> Xcalileve     0.9831 2.667   0.8435         0.8369
#> Xnecon       -8.9798 4.207   1.3302         2.9719
#> Xpresi       -2.4573 3.032   0.9588         0.9588
#> Xtag         -3.7535 2.600   0.8222         0.8222
#> Xrel         -8.1554 3.174   1.0037         2.0935
#> Xethn         2.2446 3.124   0.9878         1.9027
#> Xprevdem     13.8614 2.904   0.9182         0.9182
#> Xopenc       -2.4537 1.367   0.4324         0.4324
#> 
#> 2. Quantiles for each variable:
#> 
#>                2.5%      25%    50%     75%   97.5%
#> X.intercept  -9.040  -7.8035 -5.141 -3.3208  0.6360
#> X1          -17.594 -12.4190 -9.431 -8.0545 -6.4100
#> Xcalinv       2.862   5.6273 15.445 19.0850 20.6433
#> Xlnlevel     -8.261  -5.2289 -3.487 -1.6446  3.1120
#> Xcalileve    -2.559  -1.2788  1.068  2.5555  5.1182
#> Xnecon      -15.669 -10.7354 -9.695 -5.7098 -3.5264
#> Xpresi       -8.147  -3.5865 -2.260 -0.5025  1.2235
#> Xtag         -8.032  -4.6135 -3.771 -1.3903 -0.7852
#> Xrel        -12.638 -10.7618 -7.601 -6.1422 -3.8367
#> Xethn        -1.735  -0.6924  2.470  4.3050  6.8935
#> Xprevdem      9.677  11.7561 13.544 16.2759 18.0732
#> Xopenc       -4.466  -3.6585 -1.993 -1.5582 -0.9043

Non Misclassified-Failure

mcmcsurv estimates a Bayesian equivalent of standard survival models (i.e. Exponential or Weibull).

Example with N = 15000 can be found here.

set.seed(95)
model2 <- mcmcsurv(Y = Y, Y0 = Y0, C =  C,  X = X, 
                   N = 100, 
                   burn = 50, 
                   thin = 5, 
                   w = c(0.5, 0.5, 0.5),
                   m = 20, 
                   form = 'Weibull')


stats(model2)
#> $DIC
#> [1] -976.9334
#> 
#> $Loglik
#> [1] 444.0505

summary(model2, parameter = c("betas"))
#> 
#> Iterations = 1:10
#> Thinning interval = 1 
#> Number of chains = 1 
#> Sample size per chain = 10 
#> 
#> 1. Empirical mean and standard deviation for each variable,
#>    plus standard error of the mean:
#> 
#>               Mean       SD Naive SE Time-series SE
#> X1        -0.49109 0.746752 0.236144       0.416615
#> Xcalinv    0.08710 2.488545 0.786947       0.786947
#> Xlnlevel  -0.60347 0.090054 0.028477       0.034769
#> Xcalileve -1.17459 1.770004 0.559724       0.426229
#> Xnecon    -2.65799 2.004900 0.634005       0.634005
#> Xpresi     0.45379 0.336383 0.106374       0.106374
#> Xtag       0.05672 0.085752 0.027117       0.027117
#> Xrel       1.02890 0.462122 0.146136       0.146136
#> Xethn      1.93289 0.569031 0.179943       0.432835
#> Xprevdem   0.08207 0.262127 0.082892       0.082892
#> Xopenc    -0.01040 0.005484 0.001734       0.001872
#> 
#> 2. Quantiles for each variable:
#> 
#>               2.5%      25%      50%       75%     97.5%
#> X1        -1.95461 -0.53040 -0.27203 -0.048772  0.269946
#> Xcalinv   -2.88002 -1.91528 -0.09226  0.964488  4.146867
#> Xlnlevel  -0.73460 -0.65810 -0.61260 -0.556346 -0.462481
#> Xcalileve -4.29853 -1.86115 -1.24021 -0.137511  1.121879
#> Xnecon    -6.10899 -4.17259 -1.91485 -1.281245 -0.347827
#> Xpresi     0.19266  0.23482  0.30154  0.629261  1.082617
#> Xtag      -0.07137  0.01431  0.05023  0.089369  0.204853
#> Xrel       0.48865  0.71068  0.96574  1.158598  1.863872
#> Xethn      1.10966  1.42295  2.13381  2.190843  2.733231
#> Xprevdem  -0.30593 -0.10253  0.08783  0.327342  0.387947
#> Xopenc    -0.01962 -0.01124 -0.01010 -0.007111 -0.003705

Citation

To cite packageBayesMFSurv in publications, please use:

citation(package = 'BayesMFSurv')