Bayesian Estimation of Generalized Process Capability Indices for Hybrid Type-II Censored Data via Importance Sampling

Introduction

The gpcihybridIIImpSam package provides a comprehensive, generalized Bayesian framework for evaluating Generalized Process Capability Indices (GPCIs) under Hybrid Type-II Censored Lifetime Data using Importance Sampling (Sampling Importance Resampling, SIR).

Under Hybrid Type-II censoring, \(n\) identical units are placed on life test. The experiment terminates at: \[T^* = \max(x_r, T_c)\] where \(r \le n\) is the target number of failures and \(T_c > 0\) is the pre-fixed censoring time. The likelihood function is: \[L(\theta \mid \mathbf{x}, r, T_c, n) = \left[ \prod_{i=1}^d f(x_i; \theta) \right] [S(T^*; \theta)]^{n - d}\] where \(d \ge r\) is the number of observed failures up to time \(T^*\).

Getting Started

library(gpcihybridIIImpSam)

# 1. User-supplied custom probability density, CDF, and survival functions
my_pdf  <- function(x, rate) stats::dexp(x, rate = rate)
my_cdf  <- function(q, rate) stats::pexp(q, rate = rate)
my_surv <- function(q, rate) stats::pexp(q, rate = rate, lower.tail = FALSE)

# 2. Observed Hybrid Type-II Censored Data
# n = 10 units placed on test, target r = 3, fixed censoring time Tc = 2.5
x_data <- c(0.4, 0.9, 1.5, 2.3)

# 3. Fit Importance Sampling Model
fit <- gpci_hybrid2_impsam(
  x = x_data,
  r = 3,
  tc = 2.5,
  n = 10,
  pdf = my_pdf,
  cdf = my_cdf,
  surv = my_surv,
  param_names = "rate",
  start = c(rate = 0.5),
  chain_length = 500,
  burn_in = 100,
  thinning = 1,
  USL = 8,
  LSL = 0,
  target = 4,
  indices = c("Cpy", "Cp", "Cpk", "Cpm", "CNpmc")
)

# 4. Print Summary
print(fit)
#> --- Bayesian Importance Sampling of GPCIs under Hybrid Type-II Censoring ---
#> Distribution:  Custom 
#> Observed Failures (d):  4  | Target (r):  3  | Total (n):  10 
#> Censoring Time (Tc):  2.5  | Effective Termination (T*):  2.5 
#> Specification Limits: LSL = 0 | Target = 4 | USL = 8 
#> 
#> Initial Parameter MLEs (under Hybrid Type-II Censoring):
#>  rate 
#> 0.199 
#> 
#> Point Estimates of GPCIs (at Initial Estimates):
#>    Cpy     Cp    Cpk    Cpm  CNpmc 
#> 0.7987 0.2654 0.1974 0.2600 0.2362 
#> 
#> Posterior GPCI Summary (Importance Sampling Chain):
#>   Variable Estimate    Bias    MSE Bayes_Risk HPD_95_Lower HPD_95_Upper
#> 1      Cpy   0.7170 -0.0817 0.0339     0.0273       0.4111       0.9556
#> 2       Cp   0.2410 -0.0243 0.0126     0.0121       0.0855       0.4961
#> 3      Cpk   0.1166 -0.0808 0.0343     0.0279      -0.1574       0.3333
#> 4      Cpm   0.2280 -0.0321 0.0130     0.0120       0.0686       0.4458
#> 5    CNpmc   0.2045 -0.0317 0.0078     0.0068       0.0721       0.3523
#>   HW_Passed
#> 1      TRUE
#> 2      TRUE
#> 3      TRUE
#> 4      TRUE
#> 5      TRUE

Statistical Summary and Diagnostics

The package automatically reports: - Posterior point estimate (posterior mean) - Bias and Mean Squared Error (MSE) - Bayes risk under squared error loss (posterior variance) - Highest Posterior Density (HPD) credible intervals at 90%, 95%, and 99% levels - Heidelberger and Welch’s MCMC convergence diagnostics (stationarity test and half-width test) - Convergence probability and empirical coverage probability

# Full posterior summary
summary(fit)
#> ======================================================================
#>  Bayesian Importance Sampling of GPCIs under Hybrid Type-II Censoring 
#> ======================================================================
#> 
#> --- Parameter Posterior Summary ---
#>   Variable  Estimate        Bias         MSE  Bayes_Risk HPD_90_Lower
#> 1     rate 0.1807836 -0.01823986 0.007115577 0.006796477   0.06281003
#>   HPD_90_Upper HPD_95_Lower HPD_95_Upper HPD_99_Lower HPD_99_Upper    HW_Stat
#> 1    0.3242011   0.06413243     0.372092   0.03093259     0.372092 0.03926216
#>   HW_PValue HW_Passed Convergence_Prob Coverage_Prob
#> 1       0.5      TRUE              0.5             1
#> 
#> --- GPCI Posterior Summary ---
#>   Variable  Estimate        Bias        MSE  Bayes_Risk HPD_90_Lower
#> 1      Cpy 0.7170128 -0.08166349 0.03389204 0.027277675   0.48025647
#> 2       Cp 0.2410448 -0.02431982 0.01264991 0.012082626   0.08374671
#> 3      Cpk 0.1165777 -0.08081813 0.03433746 0.027861612  -0.11602393
#> 4      Cpm 0.2279546 -0.03205961 0.01303366 0.012029900   0.06703725
#> 5    CNpmc 0.2045042 -0.03166537 0.00778791 0.006798812   0.09338460
#>   HPD_90_Upper HPD_95_Lower HPD_95_Upper HPD_99_Lower HPD_99_Upper    HW_Stat
#> 1    0.9555643   0.41112084    0.9555643   0.23131056    0.9555643 0.02623310
#> 2    0.4322682   0.08550991    0.4961227   0.04124345    0.4961227 0.03926216
#> 3    0.3333333  -0.15744960    0.3333333  -0.24591462    0.3333333 0.02946424
#> 4    0.4144005   0.06862241    0.4458004   0.03101934    0.4458004 0.03716081
#> 5    0.3553280   0.07208756    0.3522998   0.03326190    0.3522998 0.03285368
#>   HW_PValue HW_Passed Convergence_Prob Coverage_Prob
#> 1       0.5      TRUE              0.5             1
#> 2       0.5      TRUE              0.5             1
#> 3       0.5      TRUE              0.5             1
#> 4       0.5      TRUE              0.5             1
#> 5       0.5      TRUE              0.5             1
#> 
#> --- Effective Sample Ratio / Acceptance Rates ---
#>   rate 
#> 0.4416

# Extract 95% HPD credible intervals for capability indices
confint(fit, what = "indices", level = 0.95)
#>              2.5%     97.5%
#> Cpy    0.41112084 0.9555643
#> Cp     0.08550991 0.4961227
#> Cpk   -0.15744960 0.3333333
#> Cpm    0.06862241 0.4458004
#> CNpmc  0.07208756 0.3522998

Goodness-of-Fit Testing

Goodness-of-fit testing for Hybrid Type-II censored data is supported via gofPHCS:

dist_exp <- dist_exponential(rate = 0.5)
gof_res <- gof_test_hybrid2(
  fit = fit,
  statistic = "auto",
  p.method = "montecarlo",
  nsim = 50
)
print(gof_res)
#> --- Goodness-of-Fit Test (Hybrid Type-II Censored Data) ---
#> Distribution:  Custom 
#> Test Method:   Monte Carlo Test (50 replicates) (KSII for hybridII scheme) 
#> Statistic:     KSII = 1.0361 
#> p-value:       0.70588

References

  1. Alotaibi, N., Elbatal, I., Shrahili, M., et al. (2022). Estimation and Confidence Intervals of a New PCI \(C_{Npmc}\) for Logistic-Exponential. Journal of Mathematics, 2022:3135264. doi:10.1155/2022/3135264.
  2. Childs, A., Chandrasekar, B., Balakrishnan, N., & Kundu, D. (2003). Exact likelihood inference based on Type-I and Type-II hybrid censored samples from the exponential distribution. Annals of the Institute of Statistical Mathematics, 55(2), 319-330. doi:10.1080/0266476032000053637.
  3. Dey, S., & Saha, M. (2019). The generalized process capability index \(S_{pmk}\) for Gompertz distribution. Quality and Reliability Engineering International, 35(8), 2603-2615. doi:10.1007/s41872-019-00081-4.
  4. Kundu, D., & Pradhan, B. (2009). Estimating the parameters of the generalized exponential distribution in presence of hybrid censoring. Communications in Statistics - Simulation and Computation, 38(10), 2030-2041. doi:10.1016/j.spl.2008.09.006.
  5. Maiti, S. S., Saha, M., & Nanda, A. K. (2010). On generalizing process capability indices. Quality Technology & Quantitative Management, 7(3), 279-300. doi:10.1080/16843703.2010.11673233.
  6. Saha, M., Smarandache, F., & Dey, S. (2022). Process capability index \(C_{pc}\) for lifetime data. Journal of Applied Statistics, 49(8), 2011-2032. doi:10.1080/02664763.2021.1971632.
  7. Saha, M., Tyagi, S., et al. (2024). Generalized process capability index \(C_{Npmkc}\) for quality evaluation. International Journal of Quality & Reliability Management. doi:10.1142/S021853932450013X.