Competing Risks Model with Partially Step-Stress Accelerate Life Tests in Analyses Lifetime Chen Data under Type-II Censoring Scheme

Hanaa H. Abu-Zinadah; Neveen Sayed-Ahmed

doi:10.1515/phys-2019-0019

Article Open Access

Competing Risks Model with Partially Step-Stress Accelerate Life Tests in Analyses Lifetime Chen Data under Type-II Censoring Scheme

Hanaa H. Abu-Zinadah and Neveen Sayed-Ahmed

Published/Copyright: April 28, 2019

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information Explore this Subject

From the journal Open Physics Volume 17 Issue 1

Abstract

The experiment design may need a stress level higher than use condition which is called accelerate life tests (ALTs). One of the most ALTs appears in different applications in the life testes experiment is partially step stress ALTs. Also, the experiment items is failure with several fatal risk factors, the only one is caused to failure which called competing risk model. In this paper, the partially step-stress ALTs based on Type-II censoring scheme is adopted under the different risk factors belong to Chen lifetime distributions. Under this assumption, we will estimate the model parameters of the different causes with the maximum likelihood method. The two, asymptotic distributions and the parametric bootstrap will be used to build each confidence interval of the model parameters. The precision results will be assessed through Monte Carlo simulation study.

Keywords: Accelerates life tests; Chen distribution; Computing risk model; Type-II censoring scheme; Maximum likelihood estimation; Bootstrap confidence interval

PACS: 02.50.Ga; 02.50.Ng

1 Introduction

Under modern technology, products are becoming highly reliable, this makes life-testing under use conditions is difficult and the process of collecting sufficient information about the lifetime of the products is more expensive. The ALTs is the effective one solution to this problem, in which the test units are subjected to different stress levels than the use stress to cause rapid failures. The ALTs is applied to collect enough failure information in a shorter period of time as well as to discuss the effect of lifetime and the external stress variables. According to [1, 2, 3, 4] there are different type of ALTs, the first one called constant stress ALTs, in this type the stress is still in a constant level with the life test product. Secondly, is a progressively stress ALTs, in which the stress applied to the product items in the test still increasing in time see Bai and Chung [5]. The final one is the step stress ALTs, in this test condition change for a given time or a specified number of failures, studied by some authors see Miller and Nelson [6] and Bai and Chung [7]. For more recent research on constant stress partially ALTs see Tahani and Soliman [8] and Abd-Elmougod and Emad [9]. Partially ALTs, the experiment run at use and stress conditions, such as constant and step-stress partially ALTs. In partially constant-stress ALTs, the experiment run simultaneously at use and stress condition, but in partially step-stress ALTs, the experiment run at use condition and stress change at a prefixed time or number through the experiment.

In a life testing experiment, commonly that the failure time of a product under consideration is record due to more than one causes. Our problem in such a situation, is assessed the effect of one cause in the presence of other causes. This problem known as the competing risks problem. Different work is discussed the problem of the analysis of competing risks model, see, Cox [10], David and Moeschberger [11], Crowder [12], Balakrishnan and Han [13] and Ganguly and Kundu [14]. In this paper, the model of the partially step-stress ALTs is applied when the units are fails due to two risk factor. The data will be collected from such experiment is used to estimate the parameters of the Chen failure time model under use stress level. For more detial of this problem see, Guan and Tang [15], Balakrishnan et al. [16], Lin and Chou [17], David and Kundu [18] and Abd-Elmougod and Abu-Zinadah [19].

Censoring is the common phenomenon in the life time experiments, it is applied for consideration of time and cost. The most common censoring scheme is Type-I and Type-II censoring schemes in life test experiments. In Type-I censoring terminate the experiment at a prior pre-fixed time point, but in Type-II censoring at a prior pre-fixed number of failure. The model of competing risks under consideration only two risk factors and Type-II censoring scheme is presented as the follows.

Let n independent items are put in life test experiment, and the prior m number of failure will be observe. The time X_i and cause of failure δ_i,where i = 1, 2, ..., m, δ_i ∈ {1, 2} and m ≤ n has record. The likelihood function of lifetime sample (X_i , δ_i), is presented by

(1) L θ _ | x _ = Q ∏ i = 1 m h 1 ( x i ) ρ ( δ i = 1 ) h 2 ( x i ) ρ ( δ i = 2 ) × S 1 ( x m ) S 2 ( x m ) ( n − m ) ,

where Q = n ! ( n − m ) ! , S ( x ) = 1 − F ( x ) , h ( x ) = f ( x ) F ( x ) , ρ ( δ i = j ) = 1 , δ i = j 0 , δ i ≠ j

0 < x 1 < x 2 < … < x m < ∞ .

The problem of analyzing the Type-II competing risks sample under partially step-stress ALTs model from Chen lifetime distribution is the main objective in this paper. The maximum likelihood estimation of the model parameters and accelerated factor is developed. Interval estimation with the two, approximate information matrix and bootstrap techniques are discussed. The performances of estimates are measured with average and mean squared error (MSE) for point estimation and mean length and probability coverage for interval estimation through Monte Carlo simulation.

The paper is organized as follows, in Section 2, the complete description of model formulation and the likelihood function for the partially step-stress ALTs of Type-II competing risks Chen lifetime sample. In Section 3, the MLEs of parameters and accelerated factor with the asymptotic confidence intervals. In Section 4, we discussed bootstrap confidence intervals. The quality points and interval estimates are assessed via Monte Carlo study in Section 5. Finally, some comment in Section 6.

2 Notation and Model Description

Some notations that used in this work under consideration failure items under independent two cause of failure and partially step-stress ALTs model, as given in notation table

Notations

δ_i The cause of failure for i-th item.
F (.) : The cumulative distribution function of X_i.
f (.) : The probability density function function of X _i.
F _j(.) : The cumulative distribution function of X _ji .
f_j(.) : The probability density function of X _ji .
S_j(.) : The survival density function of X _ji .
X _i : The lifetime random variable presented by i-th item.
X_ji : The lifetime random variable presented by i-th item and the cause j, j = 1, 2.

For given n identical items and the prior fixed number m and fixed time τ, the failure times X_i and cause of failure δ_i are recorded until fixed time τ is reached the items are tested under stress condition. The experiment is running to fixed number of failures m is observed. Under consideration that, the failure time has an independent Chen distribution and two cause of failure the model considered in the paper satisfies the following assumptions.

The random variable X_ji is Chen distribution with the parameter a, b_j, j = 1, 2 has the pdf

(2) f j 1 ( x ) = a b j x a − 1 exp x α exp ⁡ ( b j 1 − exp ⁡ ( x a ) ) , x > 0 , a , b j > 0 ,

and cdf, given by

(3) F j 1 ( x ) = 1 − exp ⁡ ( b j 1 − exp ⁡ ( x a ) ) .

Also, reliability S_j1(t) and failure rate functions h_j1(t) of Chen distribution for given time t, respectively presented by

(4) S j 1 ( t ) = exp ⁡ ( b j 1 − exp ⁡ ( t a ) ) , t > 0 ,

(5) h j 1 ( t ) = a b j x a − 1 exp t a , t > 0.

Item has the lifetime denoted as X_i, i = 1, 2, · · · , m and the time at which the item i fails due to cause j is X_ji, and X_i = min{X_1i , X_2i}.
Under consideration the total lifetime of items is multiply of inverse of the accelerated factor which shorten the lifetime of test items. The total lifetime of a test items, denoted by Z, defined under, use and accelerated conditions. So, the lifetime of the item in partially step-stress ALTs, is presented by

(6) Z = X , X < τ τ + β − 1 ( X − τ ) , X > τ ,

where β is the accelerated factor, τ, is the chang time from use to higher stress level and the lifetime of the unit X computed at use condition. Under Chen lifetime distribution with parameters a, b _j, the lifetime distribution of Z has the pdf, given by.

(7) f j ( z ) = f j 2 ( z ) , z > τ , f j 1 ( z ) , 0 < z ≤ τ 0 , z < 0 ,

where

(8) f j 2 ( z ) = a b j β exp ⁡ ( b j 1 − exp ( τ + β ( z − τ ) ) a ) × ( τ + β ( z − τ ) ) a − 1 exp ( τ + β ( z − τ ) ) α ,

and f_j1(z), is given by (2). The cdf, S_j2(z), and hazard rate function h_j2(z), is given by

(9) F j 2 ( z ) = 1 − exp ⁡ ( b j 1 − exp ( τ + β ( z − τ ) ) a ) ,

(10) S j 2 ( z ) = exp ⁡ ( b j 1 − exp ( τ + β ( z − τ ) ) a ) ,

and

(11) h j 2 ( z ) = a b j β ( τ + β ( z − τ ) ) a − 1 exp ( τ + β ( z − τ ) ) a .

The life test experiment is terminated with Type-II censoring scheme when the numbers of failure is reached to m < n. Then the random sample of the total lifetime Z is presented by (Z₁, δ₁) < (Z₂, δ₂) < . . . < (Z_J , δ_J) < τ < (Z_J₊₁, δ_J₊₁) < . . . < (Z_m, δ_m) where J are the number of items failed under use conditions and m − J at accelerated conditions.

The likelihood function of observed values is reduced to (1) if τ > z_m but, under consideration the observed values (z₁, δ₁)< (z₂, δ₂)< . . . < (z_J , δ_J) < τ < (z_J₊₁, δ_J₊₁) < . . . < (z_m, δ_m) and the time τ < z_m, can be presented by

(12) L θ _ | z _ = Q S 21 ( z m ) S 22 ( z m ) ( n − m ) × ∏ i = 1 J h 11 ( z i ) ρ ( δ i = 1 ) h 21 ( z i ) ρ ( δ i = 2 ) × ∏ i = J + 1 m h 12 ( z i ) ρ ( δ i = 1 ) h 22 ( z i ) ρ ( δ i = 2 )

where 0 < (Z₁, δ₁) < (Z₂, δ₂) < . . . < (ZJ , δJ) < τ < (ZJ₊₁, δJ₊₁) < . . . < (Zm, δm) < ∞.

3 Maximum Likelihood Estimation

In this section, we adopted the point and interval maximum likelihood estimates of model parameters and accelerate factor under consideration that two cause of failure and items is failure under only one cause of failure.

3.1 The point MLEs

With the consideration that, observed random sample (Z₁, δ₁) < (Z₂, δ₂) < . . . < (Z_J , δ_J) < τ < (Z_J₊₁, δ_J₊₁) < . . . < (Z_m, δ_m) has life time Chen distribution with parameters a, bj, j = 1, 2. The likelihood function in (12) without normalized conatant is reduced to

(13) L a , b 1 , b 2 , β | z _ = a m b 1 s 1 b 2 s 2 β m − J exp ( n − m ) b 1 + b 2 1 − exp ( τ + β ( z m − τ ) ) a × ∏ i = 1 J z i a − 1 ∏ i = J + 1 m ( τ + β ( z i − τ ) ) a − 1 × exp ∑ i = 1 J z i a + ∑ i = J + 1 m ( τ + β ( z i − τ ) ) a ,

where J is the number of items failure at the use condition, s₁ and s₁ are the number of items failure under causes (δ₁, δ₂) and m is the total failure time. Then after taken the natural likelihood function, we obtain

(14) ℓ a , b 1 , b 2 , β | z _ = m log ⁡ a + s 1 log ⁡ b 1 + s 2 log ⁡ b 2 + ( m − J ) log β + ( n − m ) ( b 1 + b 2 ) × 1 − exp ( τ + β ( z m − τ ) ) a + ( a − 1 ) × ∑ i = 1 J log ⁡ z i + ∑ i = J + 1 m log ⁡ ( τ + β ( z i − τ ) ) + ∑ i = 1 J z i a + ∑ i = J + 1 m ( τ + β ( z i − τ ) ) a .

The likelihood equations can be obtain from (14) by taken the first partial derivatives to a, b₁, b₂ and β as follows

(15) ∂ ℓ a , b 1 , b 2 , β | z _ ∂ b 1 = s 1 b 1 + ( n − m ) × 1 − exp ( τ + β ( z m − τ ) ) a = 0 ,

(16) ∂ ℓ a , b 1 , b 2 , β | z _ ∂ b 2 = s 2 b 2 + ( n − m ) × 1 − exp ( τ + β ( z m − τ ) ) a = 0.

Equations (15) and (16) are reduce to the MLE of b₁ and b ₂ as follows

(17) b ^ 1 ( a , β ) = s 1 ( n − m ) exp ( τ + β ( z m − τ ) ) a − 1 .

and

(18) b ^ 2 ( a , β ) = s 2 ( n − m ) exp ( τ + β ( z m − τ ) ) a − 1 .

The likelihood equations respected to a and β, are given by

(19) ∂ ℓ a , b 1 , b 2 , β | z _ ∂ a = m a − ( n − m ) ( b 1 + b 2 ) exp ( τ + β × ( z m − τ ) ) a log τ + β ( z m − τ ) + ∑ i = 1 J log ⁡ z i + ∑ i = 1 J z i a log ⁡ z i + ∑ i = J + 1 m log ⁡ ( τ + β ( z i − τ ) ) + ∑ i = J + 1 m ( τ + β ( z i − τ ) ) a × log ⁡ ( τ + β ( z i − τ ) ) = 0.

and

(20) ∂ ℓ a , b 1 , b 2 , β | z _ ∂ β = ( m − J ) β − a ( n − m ) ( b 1 + b 2 ) × ( z m − τ ) ( τ + β ( z m − τ ) ) a − 1 × exp ( τ + β ( z m − τ ) ) a + ( a − 1 ) ∑ i = J + 1 m ( z i − τ ) τ + β ( z i − τ ) + a ∑ i = J + 1 m ( z i − τ ) × ( τ + β ( z i − τ ) ) a − 1 = 0.

Equations (19) and (20) presented the two non-linear equations of a and β, solved numerical with the iteration method such as Newton Raphson method hence the MLE â and β ^ of a and β are obtained. The MLE of b₁ and b ₂ are obtained from (17) and (18) after replace a and β by a ^ and β ^ .

3.2 Approximate information matrix and interval estimation

The second partial derivatives of (14) with respect to a, β, b₁ and b ₂ presented by

(21) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ b 1 2 = − s 1 b 1 2

(22) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ b 2 2 = − s 2 b 2 2 .

(23) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ a 2 = − m a 2 − ( n − m ) ( b 1 + b 2 ) exp ( τ + β ( z m − τ ) ) a log 2 τ + β ( z m − τ ) + ∑ i = 1 J z i a log ⁡ z i 2 + ∑ i = J + 1 m ( τ + β × ( z i − τ ) ) a log 2 ⁡ ( τ + β ( z i − τ ) ) ,

(24) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ β 2 = − ( m − J ) β 2 − a ( n − m ) ( b 1 + b 2 ) × ( z m − τ ) 2 exp ( τ + β ( z m − τ ) ) a × ( τ + β ( z m − τ ) ) a − 2 a − 1 + a ( τ + β ( z m − τ ) ) a + ( a − 1 ) × ∑ i = J + 1 m − ( z i − τ ) 2 τ + β ( z i − τ ) 2 + a a − 1 ∑ i = J + 1 m ( z i − τ ) 2 × ( τ + β ( z i − τ ) ) a − 2

(25) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ b 1 ∂ b 2 = ∂ 2 ℓ a , b 1 , b 2 , β | y _ ∂ b 2 ∂ b 1 = 0

(26) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ b 1 ∂ a = ∂ 2 ℓ a , b 1 , b 2 , β | y _ ∂ a ∂ b 1 = ∂ 2 ℓ a , b 1 , b 2 , β | y _ ∂ b 2 ∂ a = ∂ 2 ℓ a , b 1 , b 2 , β | y _ ∂ a ∂ b 2 = − ( n − m ) ( τ + β ( z m − τ ) ) a × log ⁡ ( τ + β ( z m − τ ) ) × exp ( τ + β ( z m − τ ) ) a ,

(27) ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ a ∂ β = ∂ 2 ℓ a , b 1 , b 2 , β | z _ ∂ β ∂ a = − ( n − m ) ( b 1 + b 2 ) ( z m − τ ) τ + β ( z m − τ ) × exp ( τ + β ( z m − τ ) ) a × a ( τ + β ( z m − τ ) ) a × log τ + β ( z m − τ ) + 1 + ∑ i = J + 1 m ( z i − τ ) τ + β ( z i − τ ) + ∑ i = J + 1 m ( z i − τ ) ( τ + β ( z i − τ ) ) a − 1 × a log ⁡ ( τ + β ( z i − τ ) ) + 1

The expected information matrix I (a, β, b₁, b₂) of parameters a, β, b₁ and b₂ is negative expectation of second derivative of log likelihood function. Practice, I⁻¹ (a, β, b₁, b₂) is estimated by I⁻¹ a ^ , β , b ^ 1 , b ^ 2 . Hence, the normal approximation is used as follows

(28) ( a ^ , β , b ^ 1 , b ^ 2 ) → N ( a , β , b 1 , b 2 ) , I 0 − 1 ( a ^ , β , b ^ 1 , b ^ 2 ) ,

where I 0 − 1 (a, β, b₁, b₂) is the inverse of observed information matrix, presented by

distribution with mean a, β, b₁, b₂) and variance covariance matrix I 0 − 1 a ^ , β , b ^ 1 , b ^ 2 is used to present the aproximate confidence intervals of a, β, b₁ and b₂. Hence, the 100(1-2_ϒ)% approximate confidence intervals of a, β, b₁ and b₂ presented by

(30) a ^ ∓ z γ V 11 , β ^ ∓ z γ V 22 , b ^ 1 ∓ z γ V 33 and b ^ 2 ∓ z γ V 44

respectively, where value V₁₁, V₂₂ , V₃₃ and V₄₄ are the elements of the diagonal of I 0 − 1 a ^ , β , b ^ 1 , b ^ 2 and zϒ is the percentile right-tail with probable of ϒ standard normal distribution.

4 Bootstrap Confidence Intervals

Bootstrap technique is the important methods not only in estimations of confidence intervals but can be used to measures of accuracy (defined in terms of bias, variance, prediction error or some other such measure) to sample estimates, can be easily obtained with the bootstrap technique. Parametric and nonparametric bootstrap technique are the important two types of bootstrap technique are available see Davison and Hinkley [20] and Efron and Tibshirani [21]. In the following, we adopted to the parametric bootstrap technique to obtaining interval bootstrap estimations follows.

The original sample, (Z₁, δ₁) < (Z₂, δ₂) < . . . < (Z_J , δ_J) < < (Z_J₊₁, δ_J₊₁) < . . . < (Z_m, δ_m) are used to estimates a ^ , β ^ , b ^ 1 and b ^ 2 .
The independent bootstrap sample ( z 1 ∗ , δ 1 ∗ ) < ( z 2 ∗ , δ 2 ∗ ) < … < ( z J ∗ , δ J ∗ ) < τ < ( z J + 1 ∗ , δ J + 1 ∗ ) < … < ( z m ∗ , δ m ∗ ). (z*_m, _δ*m). generate from Chen distribution with parameters values given by estimates a ^ , β ^ , b ^ 1 and b ^ 2 with prior value m and τ.
The bootstrap sample ^estimates a ^ ∗ , β ^ ∗ , b ^ 1 ∗ and b ^ 2 ∗ are computed of â, β, b ^ 1 and b ^ 2 as in step 1.
Steps 2 and 3 are repeted N times, then N different bootstrap samples are represented.
Let the bootstrap sample estimates vector Σ ∗ = ( a ^ ∗ , β ^ ∗ , b ^ 1 ∗ , b ^ 2 ∗ ) ,,,put in assiding order ( Σ k ∗ [ 1 ] , Σ k ∗ [ 2 ] , … , Σ k ∗ [ N ] ) , k = 1, 2, 3, 4.
The cdf of Σ^*_k is given by G ( x ) = P ( Σ ^ k ∗ ⩽ x ) for any given x. Then Σ ^ k − b o o t ∗ = G − 1 ( x ) for given z. Then percentile bootstrap confidence intervals 100(1 − 2ϒ) confidence interval of Σ ^ k ∗ is given by

(31) Σ ^ k − b o o t ∗ ( γ ) , Σ ^ k − b o o t ∗ ( 1 − γ ) .

Let us define statistic Ω k ∗ [ j ] by

(32) Ω k ∗ ( j ) = Σ ^ k ∗ ( i ) − Σ ^ k var Σ ^ k ∗ ( i ) ,

the order values of statstics Ω k ∗ ( j ) defined by Ω k ∗ [ j ] , we define the cmulative distribution G ( x ) = P ( Ω k ∗ < x ) for any given x, define

(33) Σ ^ k − boot-t ∗ = Σ ^ k + Var ( Σ ^ k ) G − 1 ( z ) .

Then bootstrap-t confidence intervals (BTCI) of 100(1 − 2_ϒ) approximate confidence intervals of Σ ^ k is given by

(34) Σ ^ k − boot-t ∗ ( γ , Σ ^ k − boot-t ∗ ( 1 − γ ) .

5 Monte Carlo Simulations

In this section, we adopted the simulation study to assess and compare the developed results in this paper. Some numerical experiments performed for sample of sizes n, the effective sample sizes m, accelerated time τ and model parameters (a, β, b₁, b₂). In the simulation stuides,we adopted the case, the model parameters (a, β, b₁, b₁) = (1, 1.5, 1.5, 2.0) and accelerate time τ = (0.5, 1.0, 1.5). Essentially, the simulation results are computed to compare the MLEs and bootstrap estimators, mainly are compared in terms of their average (AV) and MSE. The confidence intervals are compared in terms of their average lengths (AL) and the probability coverage (CP). The results in this paper is computed with Mathematica version 9 and reported in Tables 1-2.

Table 1

The mean estimates and MSEs for the parameters (a, β, b₁, b₂) at (1, 1.5, 1.5, 2.0)

		MLE
τ	(n, m))	AV				MSE
		a	β	b₁	b₁	a	β	b₁	b₁
0.5	(30,15)	0.81	1.307	1.34	1.72	0.231	0.336	0.421	0.721
	(30,25)	0.81	1.319	1.34	1.74	0.221	0.161	0.321	0.621
	(50,25)	0.84	1.33	1.37	1.79	0.204	0.131	0.231	0.654
	(50,40)	0.84	1.36	1.40	1.81	0.199	0.153	0.203	0.521
	(75,50)	0.87	1.38	1.38	1.82	0.136	0.136	0.188	0.521

1.0	(30,15)	0.85	1.33	1.40	1.78	0.230	0.161	0.175	0.421
	(30,25)	0.80	1.38	1.41	1.79	0.205	0.134	0.124	0.521
	(50,25)	0.92	1.43	1.57	1.80	0.174	0.161	0.133	0.621
	(50,40)	0.93	1.43	1.42	1.99	0.170	0.131	0.109	0.421
	(75,50)	0.93	1.64	1.54	1.92	0.102	0.153	0.111	0.321

1.5	(30,15)	0.94	1.64	1.54	1.92	0.201	0.137	0.100	0.754
	(30,25)	0.99	1.65	1.58	1.93	0.105	0.163	0.090	0.451
	(50,25)	0.98	1.69	1.61	1.95	0.097	0.120	0.101	0.452
	(50,40)	0.97	1.69	1.62	1.96	0.097	0.140	0.107	0.421
	(75,50)	0.99	1.74	1.40	1.82	0.084	0.114	0.094	0.399

		Bootstrap
τ	(n, m)	AV				MSE
		a	β	b₁	b₁	a	β	b₁	b₁
0.5	(30,15)	0.91	1.88	1.88	1.99	0.421	0.542	0.621	0.942
	(30,25)	0.89	1.45	1.82	1.75	0.401	0.512	0.584	0.7541
	(50,25)	0.84	1.354	1.81	1.94	0.365	0.499	0.542	0.771
	(50,40)	0.78	1.42	1.84	1.89	0.321	0.475	0.511	0.612
	(75,50)	0.84	1.30	1.38	1.93	0.213	0.421	0.421	0.600

1.0	(30,15)	0.75	1.17	1.60	1.96	0.421	0.411	0.521	0.599
	(30,25)	0.82	1.75	1.77	1.85	0.321	0.385	0.478	0.583
	(50,25)	0.94	1.40	1.64	1.81	0.210	0.375	0.402	0.531
	(50,40)	0.99	1.43	1.69	1.90	0.199	0.321	0.339	0.466
	(75,50)	0.92	1.77	1.91	1.92	0.189	0.321	0.310	0.399

1.5	(30,15)	1.04	1.64	1.88	1.99	0.210	0.199	0.254	0.623
	(30,25)	100	1.62	1.78	1.88	0.23	0.189	0.213	0.509
	(50,25)	0.92	1.77	1.62	1.91	0.120	0.179	0.213	0.520
	(50,40)	0.94	1.69	1.69	1.92	0.110	0.160	0.188	0.499
	(75,50)	0.92	1.85	1.74	1.87	0.109	0.158	0.142	0.421

6 Conclusions

Type-II censoring competing risks model is discussed in this paper in the presence of partially step-stress ALTs under consideration failure times of the competing risks follow independent Chen distributions. The MLEs of the unknown model parameters are derived. The asymptotic distribution of MLEs and bootstrap method for constructing CIs. Results from Tables 1-2 of the simulation study reported as follows

Table 2

The AL and (CP), respectivly of MLEPBCIs and PTCIs of 95% CIs for the parameters (a, β, b₁, b₂) at (1, 1.5, 1.5, 2.0)

τ	(n, m)	MLE				PBCIs				PTCIs
		a	β	b₁	b₂	a	β	b₁	b₁	a	β	b₁	b₁
0.5	(30,15)	2.323	3.543	3.421	4.410	3.355	4.223	4.001	5.433	2.122	3.421	3.339	4.254
		(0.89)	(0.89)	(0.88)	(0.89)	(0.87)	(0.88)	(0.89)	(0.89)	(0.89)	(0.90)	(0.89)	(0.90)
	(30,25)	2.311	3.42	3.326	4.214	3.322	4.111	4.421	5.338	2.218	3.321	3.214	4.051
		(0.89)	(0.90)	(0.91)	(0.90)	(0.89)	(0.89)	(0.90)	(0.90)	(0.91)	(0.91)	(0.92)	(0.91)
	(50,25)	2.302	3.214	3.300	4.124	3.344	4.252	4.305	5.121	2.270	3.154	3.221	4.021
		(0.88)	(0.91)	(0.91)	(0.92)	(0.90)	(0.90)	(0.90)	(0.90)	(0.91)	(0.93)	(0.94)	(0.97)
	(50,40)	2.198	3.109	3.002	3.950	3.125	3.999	3.998	5.050	2.100	3.025	2.902	3.741
		(0.90)	(0.92)	(0.90)	(0.93)	(0.91)	(0.91)	(0.89)	(0.91)	(0.93)	(0.95)	(0.94)	(0.93)
	(75,50)	2.082	2.998	2.977	3.258	3.074	3.991	3.901	5.008	2.005	2.742	2.900	3.133
		(0.91)	(0.93)	(0.92)	(0.94)	(0.92)	(0.91)	(0.90)	(0.91)	(0.91)	(0.93)	(0.95)	(0.94)

1.0	(30,15)	2.301	3.520	3.400	4.3854	3.332	4.223	3.932	5.404	2.002	3.362	3.329	4.226
		(0.87)	(0.88)	(0.89)	(0.90)	(0.88)	(0.88)	(0.89)	(0.88)	(0.89)	(0.91)	(0.89)	(0.91)
	(30,25)	2.295	3.385	3.311	4.212	3.309	4.101	4.414	5.322	2.189	3.285	3.197	4.031
		(0.90)	(0.89)	(0.90)	(0.91)	(0.889)	(0.88)	(0.91)	(0.91)	(0.91)	(0.93)	(0.92)	(0.93)
	(50,25)	2.288	3.190	3.277	4.101	3.322	4.253	4.312	5.100	2.257	3.142	3.207	4.018
		(0.91)	(0.90)	(0.91)	(0.93)	(0.90)	(0.91)	(0.91)	(0.90)	(0.92)	(0.94)	(0.92)	(0.93)
	(50,40)	2.171	3.111	3.011	3.938	3.114	3.987	3.975	5.041	2.099	3.019	2.910	3.738
		(0.92)	(0.90)	(0.92)	(0.91)	(0.97)	(0.90)	(0.91)	(0.97)	(0.94)	(0.95)	(0.95)	(0.96)
	(75,50)	2.100	2.892	2.962	3.241	3.064	3.977	3.842	5.011	2.011	2.726	2.904	3.121
		(0.921)	(0.92)	(0.94)	(0.93)	(0.91)	(0.98)	(0.93)	(0.92)	(0.93)	(0.94)	(0.95)	(0.93)

1.5	(30,15)	2.338	3.550	3.445	4.432	3.355	4.247	4.031	5.451	2.144	3.445	3.350	4.260
		(0.89)	(0.88)	(0.89)	(0.87)	(0.86)	(0.88)	(0.89)	(0.88)	(0.89)	(0.91)	(0.89)	(0.91)
	(30,25)	2.325	3.436	3.341	4.225	3.339	4.124	4.442	5.339	2.241	3.335	3.227	4.063
		(0.88)	(0.91)	(0.90)	(0.92)	(0.89)	(0.89)	(0.910)	(0.91)	(0.91)	(0.93)	(0.91)	(0.91)
	(50,25)	2.314	3.227	3.313	4.137	3.350	4.211	4.314	5.128	2.269	3.166	3.240	4.017
		(0.92)	(0.93)	(0.92)	(0.93)	(0.91)	(0.91)	(0.90)	(0.91)	(0.96)	(0.95)	(0.95)	(0.93)
	(50,40)	2.204	3.121	3.019	3.960	3.141	3.989	4.091	5.062	2.111	3.035	2.918	3.752
		(0.92)	(0.92)	(0.91)	(0.93)	(0.91)	(0.90)	(0.90)	(0.92)	(0.93)	(0.94)	(0.93)	(0.95)
	(75,50)	2.097	3.018	2.987	3.270	3.079	3.999	3.923	5.011	2.014	2.763	2.911	3.108
		(0.92)	(0.93)	(0.92)	(0.92)	(0.93)	(0.91)	(0.91)	(0.92)	(0.94)	(0.93)	(0.94)	(0.96)

From the Table 2,we observe that PTCIs performs the best as its CIs has a small length and coverage probabilities to be much closer to the nominal levels than ACs and PBCIs.
From the Table 1, we observe the point estimate of MLE performs the best than bootstrap estimates.
From all tables, we observe that results for the value of accelerate change time τ performs the best for the vale of τ narrow of distribution mean
For the effective m sample increases, the MSEs and the AL of different estimators are reduced.

Statistics Department, Faculty of Science, University of Jeddah, Jeddah, Saudi Arabia

(Girls’ Branch); Mathematics Department, Faculty of Science, Taif University, Taif City 888, Saudi Arabia

Acknowledgement

This project was funded by the Deanship of Scientific Research (DSR) at King Abdulaziz University, Jeddah under Grant No. G-144-363-1439. The authors, therefore, acknowledge with thanks DSR for technical and financial support.

References

[1] Nelson W., Accelerated Testing: Statistical Models, Test Plans and Data Analysis.Wiley, New York,(1990)10.1002/9780470316795Search in Google Scholar

[2] Kim C. M., Bai D.S., Analysis of accelerated life test data under two failure modes. Int. J. Reliability, Quality and Safety Engineering, (2002), 9, 111-125.10.1142/S0218539302000706Search in Google Scholar

[3] AL-Hussaini E.K., Abdel-Hamid A. H., Bayesian estimation of the parameters, reliability and hazard rate functions of mixtures under accelerated life tests, Communication in Statistics - Simulation and Computation,(2004), 33 963-982.10.1081/SAC-200040703Search in Google Scholar

[4] AL-Hussaini E.K., Abdel-Hamid A.H., Accelerated life tests under finite mixture models. J. Statist, Comput. Simul., (2006),76 673-690.10.1080/10629360500108087Search in Google Scholar

[5] Bai D.S., Chung S.W., Optimal design of partially accelerated life tests for the exponential distribution under Type-I censoring, IEEE Trans. Reliab,(1992) 41, 400-406.10.1109/24.159807Search in Google Scholar

[6] Miller R., Nelson W.B., Optimum simple step-stress plans for accelerated life testing, IEEE Trans. Reliab., (1983) 32, 59-65.10.1109/TR.1983.5221475Search in Google Scholar

[7] Bai D.S., Chung S.W., Optimal design of partially accelerated life tests for the exponential distribution under Type-I censoring, IEEE Trans. Reliab., (1992) 41, 400-406 .10.1109/24.159807Search in Google Scholar

[8] Tahani A. Abushala, Ahmed A. Soliman, Estimating the Pareto parameters under progressive censoring data for constantpartially accelerated life tests, Journal of Statistical Computation and Simulation,(2013), 37-41.10.1080/00949655.2013.853768Search in Google Scholar

[9] Abd-Elmougod G. A., Emad E. M., Paramters estimation of compound Rayleigh distribution under an adaptive Type-II Progressively hybrid censored data for constant partially accelerated life tests, Global Journal of Pure and Applied Mathematics, 1(2016)2, 3253-3273.Search in Google Scholar

[10] Cox D. R., The analysis of exponentially distributed lifetimes with two types of failures", Journal of the Royal Statistical Society, (1959) 21, 411-421.10.1111/j.2517-6161.1959.tb00349.xSearch in Google Scholar

[11] David H. A., Moeschberger M. L., The theory of competing risks, Grin, London, (1978).Search in Google Scholar

[12] Crowder M. J., Classical competing risks, Chapman and Hall, London, (2001).10.1201/9781420035902Search in Google Scholar

[13] Balakrishnan N., Han D., Exact inference for a simple step-stress model with competing risks for failure from exponential distribution under Type-II censoring", Journal of Statistical Planning and Inference, (2008), 138, 4172-4186.10.1016/j.jspi.2008.03.036Search in Google Scholar

[14] Ganguly A., Kundu D., Analysis of simple step-stress model in presence of competing risks", Journal of Statistical Computation and Simulation, (2016)86, 1989 - 2006.10.1080/00949655.2015.1096362Search in Google Scholar

[15] Guan Q., Tang Y., Optimal step-stress test under Type-I censoring for multivariate exponential distribution, Journal of Statistical Planning and Inference,(2012), 142, 1908-1923.10.1016/j.jspi.2012.02.029Search in Google Scholar

[16] Balakrishnan N.Z., hang L., Xie Q., Inference for a simple step-stress model with Type-I censoring and lognormally distributed lifetimes, Communication in Statistics - Theory and Methods, (2009)38, 1690-1709.10.1080/03610920902866966Search in Google Scholar

[17] Lin C. T., Chou C. C., Statistical inference for a lognormal step-stress model with Type-I censoring, IEEE Transactions on Reliability, (2012), 61, 361-377.10.1109/TR.2012.2194178Search in Google Scholar

[18] Han D., Kundu D., Inference for a step-stress model with competing risks for failure from the generalized exponential distribution under type-I censoring, IEEE Transactions on Reliability,(2015), 64, 31-43.10.1109/TR.2014.2336392Search in Google Scholar

[19] Abd-Elmougod, G. A ., Abu-Zinadah H. H., Estimation of P(Y < X) for two-parameter bathtub Shape distribution using records: Bayesian and Non-Bayesian approaches, Journal of Computational and Theoretical Nanoscience, (2017), 14, 1127-1135.10.1166/jctn.2017.6416Search in Google Scholar

[20] Davison A. C., Hinkley D. V.„ Bootstrap Methods and their Applications 2^nd Cambridge University Press, Cambridge United Kingdom, (1997).10.1017/CBO9780511802843Search in Google Scholar

[21] Efron B., Tibshirani R. J., An introduction to the bootstrap. New York Chapman and Hall, (1993).10.1007/978-1-4899-4541-9Search in Google Scholar

Received: 2018-10-23

Accepted: 2019-03-01

Published Online: 2019-04-28

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

https://doi.org/10.1515/phys-2019-0019

Keywords for this article

Accelerates life tests; Chen distribution; Computing risk model; Type-II censoring scheme; Maximum likelihood estimation; Bootstrap confidence interval

Creative Commons

BY 4.0