MPEMathematical Problems in Engineering1563-51471024-123XHindawi Publishing Corporation39309510.1155/2010/393095393095Research ArticleHypothesis Designs for Three-Hypothesis Test ProblemsLiYanPuXiaolongLiMingSchool of Finance and StatisticsEast China Normal UniversityNo. 500 Dongchuan Road Shanghai 200241Chinaecnu.edu.cn201027042010201025012010180320102010Copyright © 2010This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

As a helpful guide for applications, the alternative hypotheses of the three-hypothesis test problems are designed under the required error probabilities and average sample number in this paper. The asymptotic formulas and the proposed numerical quadrature formulas are adopted, respectively, to obtain the hypothesis designs and the corresponding sequential test schemes under the Koopman-Darmois distributions. The example of the normal mean test shows that our methods are quite efficient and satisfactory for practical uses.

1. Introduction

In practice, the multihypothesis test problems are of considerable interest in the areas of engineering, agriculture, clinical medicine, psychology, and so on. For instance, the multihypothesis tests are involved in pattern recognition , multiple-resolution radar detection , products' comparisons [8, 9], and information detection . Before the inspections, the hypotheses must be determined according to such practical needs as the balance of risks and costs. As Wetherill and Glazebrook  pointed out, combinations of hypotheses, risks, and costs may need to be tried iteratively until an acceptable design is attained. This bothers and burdens the practitioners.

To avoid too many troublesome trials and to produce the hypotheses directly, we discuss the hypothesis designs under the controlled risks and expected costs in this paper. As an initial exploration, only the three-hypothesis test problems are considered here. Indeed, our methods may extend to the multihypothesis cases.

In practice, test costs are mainly determined by sample sizes. Therefore, the sample size becomes an issue relating to the statistical analysis of problems in many aspects; see for example, Chen et al. , Oliveira et al. , Li and Zhao , Li et al. , Bakhoum and Toma , Cattani , as well as Cattani and Kudreyko . Accordingly, we consider the Average Sample Number (ASN), which is one of the most important values in evaluating the expected costs of sequential test schemes.

In the three-hypothesis test problem, the null hypothesis is always set as a standard and medium status. For example, Anderson  discussed the three-hypothesis test problem to decide whether the difference of two yarns' strength is zero (the null hypothesis), positive or negative. Realistically, the standard and medium status (denoted as θ0) is definite, while the two alternatives beside it need to be designed to balance the risks and costs. Thus, in this paper, we try to design the alternatives θ-1 and θ1(θ-1<θ0<θ1) under the required error probabilities and ASN for testing the parameter θ of the Koopman-Darmois distributionfθ(x)=exp{l(x)+θx-b(θ)},where  b(θ)  is  a  convex  function,θΘ.

To simplify the discussion, we only consider the designs of the two alternative hypotheses symmetric with the null hypothesis, that is, θ1-θ0=θ0-θ-1=k(>0). Actually, the asymmetric designs may be obtained by extending our methods slightly.

Then, the test problem here isH-1:θ=θ-1=θ0-kvs.    H0:θ=θ0vs.  H1:θ=θ1=θ0+k.

For the multihypothesis test problems, Armitage  provided a classical test scheme by simultaneously applying the method of Sequential Probability Ratio Test (SPRT) on each pair of the hypotheses. This test scheme pattern is simple and easy to implement. When testing the three hypotheses for the Koopman-Darmois distribution (1.1), Armitage's scheme may be illustrated as in Figure 1, where AL//CM are boundaries for “θ=θ1 versus θ=θ0" and CP//DQ are for “θ=θ0 versus θ=θ-1" when the boundaries for “θ=θ1 versus θ=θ-1" are encircled by AL and DQ and thus are neglected. According to Figure 1, the decision rule should beAccept  H1if  Tna+ntanψ,Accept  H0if  c+(n-n0)tanφTnc+(n-n0)tanψ,Accept  H-1if  Tnd+ntanφ,Continue  sampling  without  any  decision,otherwise, where Tn=i=1nXi and X1,X2, are independent sequential observations from a Koopman-Darmois distribution.

For the given θ-1,θ0, and θ1, the test scheme in Figure 1 is decided by 6 parameters (n0,a,c,d,ψ,φ). ψ and φ may be determined according to Armitage , that is, tanψ=[b(θ1)-b(θ0)]/(θ1-θ0), tanφ=[b(θ0)-b(θ-1)]/(θ0-θ-1) under the Koopman-Darmois distribution (1.1), then the remaining 4 parameters (n0,a,c,d) form the scheme. Altogether with the hypothesis design value k in the test problem (1.2), the 5 underdetermined values are (k,n0,a,c,d).

In the three-hypothesis test problems, the error probabilities α and β should be assigned to the error probabilities (γ1,γ2,γ3,γ4) in correspondence with the requirementsP(Accept  H0H1)γ1,P(Accept  H0H-1)γ2,P(Accept  H1H0)γ3,P(Accept  H-1H0)γ4. Commonly, we set γ1=γ2=β, γ3=γ4=α/2, as Payton and Young [20, 21] indicated.

And the request on the ASN should beASN(θASN)N, where N(>0) is a provided integer and θASN(Θ) is the point at which the ASN needs to be controlled. θASN may take values of θ-1, θ0, θ1, and so on according to practical needs.

Then, under the constraints (1.4) and (1.5), we may find the proper (k,n0,a,c,d) by virtue of their relationships with the error probabilities and ASN.

Unfortunately, however, to the best knowledge of the authors, the accurate formulas for the performances of the three-hypothesis test scheme are still unavailable possibly because of its sequential feature and anomalistic continuing sampling area. In the following, the hypothesis designs and the test scheme parameters are determined under the required error probabilities and ASN in terms of some approximate expressions, that is, the asymptotic formulas and the proposed numerical quadrature formulas.

2. Designs under Asymptotic Formulas

In this section, we try to find the hypothesis designs and test schemes under the required error probabilities and ASN by virtue of the asymptotic formulas of the multihypothesis sequential test scheme by Dragalin et al. in [22, 23].

Firstly, we discuss how to control the error probabilities. Let Ci be the critical value of the logarithmic likelihood ratio function for accepting θi, and let Ri be the probability limit of incorrectly accepting θi, i=-1,0,1. According to Dragalin et al. , under the condition of equal prior probabilities for the three hypotheses, the probability of wrongly accepting θi for the Armitage  scheme may be controlled by Ri if the critical value Ci is set asCi=ln{23Ri},i=-1,0,1. Thus, the error probabilities (γ1,γ2,γ3,γ4) are in control if we follow the critical values in (2.1), where R-1=γ4, R0=min{γ1,γ2}, and R1=γ3. Setting the critical values (C-1,C0,C1) equal to the corresponding logarithmic likelihood ratio functions, we have the following expressions for the test scheme parameters under the Koopman-Darmois distribution (1.1):n0=C0(1/(θ0-θ-1)-1/(θ0-θ1))(b(θ0)-b(θ1))/(θ0-θ1)-(b(θ0)-b(θ-1))/(θ0-θ-1)=2C0b(θ0+k)+b(θ0-k)-2b(θ0),a=C1θ1-θ0=C1k,c=C0θ0-θ1+n0tanψ=-C0k+n0b(θ0+k)-b(θ0)k,d=C-1θ-1-θ0=-C-1k.

Note that the expressions in (2.2) define the relations between the hypothesis design parameter k and the test scheme parameters (n0,a,c,d), while k has not been determined so far.

In the following, the hypothesis design parameter k is found with the help of Dragalin et al.'s asymptotic ASN formulas .

Based on the nonlinear renewal theory, Dragalin et al.  summarized and developed the asymptotic ASN formulas under max{α,β}0. Specifically, when θ1-θ0=θ0-θ-1, the asymptotic ASN formulas under the two alternatives θ-1 and θ1 areASN(θi)Ci+OθiDθi,i=-1,1, where Dθi=minjiEθi(ln{fθi(x)/fθj(x)}) and Oθi is the expected limiting overshoot under θi, i=-1,0,1.

And for the null hypothesis θ0 under θ1-θ0=θ0-θ-1, the asymptotic ASN formula isASN(θ0)F2(C0,Dθ0,v)+Oθ0Dθ0, where F2(x,q,u)=x+uh2*x/q+u2(h2*)2/(4q2)+u2(h2*)2/(2q) (h2*=0.5641895835 here), and v is the value related to the covariance of the logarithmic likelihood ratio functions.

Notice that the approximate ASN formulas (2.3) and (2.4) only depend on the hypothesis design parameter k when θ0 is given. Therefore, to find the proper hypothesis design under the desired number N, we set up an equation about k to meet the ASN requirement on one of the three hypothesis values, that is,ASN(θASN)=N, where θASN may be θ-1,θ0, or θ1.

Then, the hypothesis design parameter k is the solution to (2.5) and the test scheme with (n0,a,c,d) may be obtained correspondingly according to (2.2). Illustrations are provided in Example 1 for testing the normal mean with the variance known.

Example 1.

Suppose that the sequential observations X1,X2, are independent and identically distributed (i.i.d.) with N(μ,1). Let μ0=0, γ1=γ2=β, and γ3=γ4=α/2. Small values (30) are set on N as practical sequential inspections always require.

Accordingly, we have C0=ln{2/(3γ1)}, C1=C-1=ln{2/(3γ3)}. In this example, the test scheme parameters should be n0=2C0k2,a=C1k,c=0,d=-a,tanψ=k2,tanφ=-tanψ. And for the normal distribution N(μ,1), there are Dμi=minji(μi-μj)22=k22,i=-1,0,1,Oμi=1+k24-kl=11l[ϕ(k2l)-k2lΦ(-k2l)],i=-1,0,1,v=2k, where ϕ(·) and Φ(·) are the probability density function (p.d.f.) and cumulative distribution function (c.d.f.) of the standard normal distribution, respectively.

Consider the following 4 cases, respectively: μASN=μ0,α=β=0.002,μASN=μ0,α=β=0.05,μASN=μ1,α=β=0.002,μASN=μ1,α=β=0.05. Then, solving (2.5), we obtain the hypothesis designs k as shown in Column 2 of Tables 1, 2, 3, and 4. The corresponding test scheme parameters (n0,a,tanψ) from (2.6) are listed in Columns 3–5 of Tables 14. To evaluate the method's efficiency, we record the Monte Carlo simulation study results with 1,000,000 replicates in Tables 5, 6, 7, and 8, where ASN(μASN) is the simulated value of ASN(μASN) and ɛ is the relative difference between ASN(μASN) and N. Note that the simulated probabilities under μ-1 are neglected here since they are nearly equivalent to their counterparts under μ1 in terms of the schemes' symmetry.

Hypothesis designs and test schemes for μASN=μ0, α=β=0.002.

 N Under Asymptotic formulas Under Gaussian quadrature formulas ɛk  (%) k n0 a tanψ k n0 a tanψ 5 2.1090 2.6121 3.0831 1.0545 1.8953 2.6876 3.0725 0.9477 11.28 10 1.4400 5.6030 4.5155 0.7200 1.3193 5.8833 4.6576 0.6597 9.15 15 1.1604 8.6279 5.6034 0.5802 1.0674 9.0266 5.8910 0.5337 8.71 20 0.9977 11.6709 6.5170 0.4989 0.9228 12.3354 6.9039 0.4614 8.12 25 0.8882 14.7257 7.3204 0.4441 0.8246 15.6592 7.7944 0.4123 7.71 30 0.8082 17.7888 8.0458 0.4041 0.7523 18.9949 8.5984 0.3762 7.43

Hypothesis designs and test schemes for μASN=μ0, α=β=0.05.

 N Under Asymptotic formulas Under Gaussian quadrature formulas εk  (%) k n0 a tan ψ k n0 a tan ψ 5 1.6066 2.0072 2.0438 0.8033 1.3025 2.1953 2.2175 0.6513 23.35 10 1.0908 4.3542 3.0102 0.5454 0.9087 5.0028 3.4233 0.4544 20.04 15 0.8767 6.7395 3.7450 0.4384 0.7401 8.0001 4.2242 0.3701 18.46 20 0.7527 9.1448 4.3624 0.3764 0.6392 11.0000 5.2525 0.3196 17.76 25 0.6693 11.5630 4.9054 0.3347 0.5708 14.0000 5.9086 0.2854 17.26 30 0.6085 13.9903 5.3957 0.3043 0.5204 17.0000 6.5244 0.2602 16.93

Hypothesis designs and test schemes for μASN=μ1, α=β=0.002.

 N Under Asymptotic formulas Under Gaussian quadrature formulas εk  (%) k n0 a tan ψ k n0 a tan ψ 5 1.7893 3.6290 3.6340 0.8947 1.7105 3.2864 3.4644 0.8553 4.61 10 1.2179 7.8334 5.3391 0.6090 1.1900 7.1875 5.2255 0.5950 2.34 15 0.9800 12.0973 6.6350 0.4900 0.9667 11.1560 6.5639 0.4834 1.38 20 0.8419 16.3926 7.7236 0.4210 0.8352 15.1594 7.6882 0.4176 0.80 25 0.7490 20.7080 8.6809 0.3745 0.7460 19.1851 8.6770 0.3730 0.40 30 0.6812 25.0379 9.5454 0.3406 0.6803 23.2266 9.5698 0.3402 0.13

Hypothesis designs and test schemes for μASN=μ1, α=β=0.05.

 N Under Asymptotic formulas Under Gaussian quadrature formulas ɛk  (%) k n0 a tan ψ k n0 a tan ψ 5 1.3089 3.0237 2.5085 0.6545 1.1875 2.9124 2.4858 0.5938 10.22 10 0.8835 6.6369 3.7164 0.4418 0.8264 6.2545 3.8213 0.4132 6.91 15 0.7083 10.3268 4.6358 0.3542 0.6715 10.0000 4.8351 0.3358 5.48 20 0.6071 14.0566 5.4085 0.3036 0.5803 13.4993 5.6872 0.2902 4.62 25 0.5393 17.8118 6.0883 0.2697 0.5183 17.0061 6.4368 0.2592 4.05 30 0.4899 21.5853 6.7022 0.2450 0.4727 20.8842 7.1131 0.2364 3.64

Simulated performances for the schemes under asymptotic formulas in Table 1.

 N When μ1 is true When μ0 is true ASN′(μ0) ε  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0007 0.9993 0.0005 0.9990 0.0005 4.7053 6.26 10 0 0.0009 0.9991 0.0007 0.9986 0.0007 9.3601 6.84 15 0 0.0011 0.9989 0.0007 0.9985 0.0008 13.9845 7.26 20 0 0.0011 0.9989 0.0008 0.9984 0.0008 18.5910 7.58 25 0 0.0012 0.9988 0.0009 0.9982 0.0009 23.1885 7.81 30 0 0.0013 0.9987 0.0010 0.9980 0.0010 27.8046 7.90

Simulated performances for the schemes under asymptotic formulas in Table 2.

 N When μ1 is true When μ0 is true ASN′(μ0) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0186 0.9814 0.0148 0.9703 0.0149 4.4409 12.59 10 0 0.0273 0.9727 0.0195 0.9610 0.0195 8.4229 18.72 15 0 0.0317 0.9682 0.0219 0.9562 0.0219 12.3687 21.27 20 0 0.0323 0.9677 0.0236 0.9529 0.0235 16.4767 21.38 25 0 0.0344 0.9656 0.0246 0.9508 0.0245 20.3535 18.59 30 0 0.0361 0.9638 0.0252 0.9494 0.0255 24.2233 19.26

Simulated performances for the schemes under asymptotic formulas in Table 3.

 N When μ1 is true When μ0 is true ASN′(μ1) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0008 0.9992 0.0005 0.9990 0.0005 4.9907 0.19 10 0 0.0010 0.9990 0.0007 0.9985 0.0008 9.9818 0.18 15 0 0.0010 0.9990 0.0009 0.9983 0.0008 14.9705 0.20 20 0 0.0012 0.9988 0.0010 0.9981 0.0009 19.9525 0.24 25 0 0.0013 0.9987 0.0009 0.9981 0.0010 24.9494 0.20 30 0 0.0012 0.9988 0.0010 0.9980 0.0010 29.9059 0.31

Simulated performances for the schemes under asymptotic formulas in Table 4.

 N When μ1 is true When μ0 is true ASN′(μ1) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0226 0.9774 0.0173 0.9655 0.0172 4.7976 4.05 10 0 0.0313 0.9687 0.0221 0.9560 0.0219 9.4322 5.68 15 0 0.0336 0.9664 0.0241 0.9518 0.0241 14.1221 5.85 20 0 0.0346 0.9654 0.0255 0.9492 0.0253 18.7665 6.17 25 0 0.0367 0.9632 0.0265 0.9471 0.0264 23.3316 6.67 30 0 0.0371 0.9628 0.0270 0.9459 0.0270 27.9989 6.67

Obviously, the accuracy of the solution k to (2.5) is decided by the efficiency of the ASN formulas (2.3) and (2.4). On one hand, from Dragalin et al.  and the ε’s in Tables 58, we conclude that the formulas in (2.3) for ASN(θ-1) and ASN(θ1) are more efficient than the one in (2.4) for ASN(θ0) when testing the normal mean. On the other hand, the asymptotic ASN formulas perform better under smaller error probabilities since the asymptotic limit is taken as max{α,β}0. For applications, with such a simple computation, the efficiency of the design is quite satisfactory for small error probabilities conditions.

However, this method may only serve to control the ASN on the three hypothesis values since the asymptotic ASN formulas out of these points are absent so far. And the quantities Dθi,Oθi(i=-1,0,1), and v should be deduced according to specific distributions (see ). Besides, the discrepancies between the real performances and the required ones show the method's conservativeness. In the next section, an improved method is proposed and more efficient formulas are developed through the numerical quadrature.

3. Designs under Numerical Quadrature Formulas

This section proposes a method to obtain more efficient hypothesis designs and test schemes through a system of equations based on the numerical quadrature formulas of the error probabilities and ASN.

In studies by Payton and Young in [20, 21], for the provided hypotheses, the error probabilities (γ1,γ2,γ3,γ4) are approximately attained by solving a system of equations about the 4 scheme parameters (n0,a,c,d). This method is hoped to fully make use of the required error probabilities and to obtain efficient designs. Enlightened by Payton and Young, we propose to find the hypothesis design and test scheme by solving the following system of equations:P(Accept  H1H0)=γ3,P(Accept  H-1H0)=γ4,P(Accept  H0H1)=γ1,P(Accept  H0H-1)=γ2,ASN(θASN)=N.

Obviously, the key is to find the formulas of the error probabilities and ASN on the left side of the equations in (3.1). Unfortunately, the available approximate formulas cannot meet applicable needs well. For example, Payton and Young [20, 21] adopted the formulas under the continuous-time process and the required minimum sample size before decisions, and obtained some inefficient results. Also, as mentioned in Section 2, Dragalin et al.'s results are restricted to the conditions of small error probabilities and θASN=θi(i=-1,0,1) [22, 23].

To find efficient and applicable designs, we develop the approximate formulas through the numerical quadrature for the three-hypothesis test scheme's performances on the error probabilities and ASN.

To deduce the formulas for the realistic discrete-time situation, we denote nt as the minimum integer that is not less than n0. Let Lj and Uj be the values on the two boundaries DQ and AL in Figure 1 at n=j, that is, Lj=d+jtanφ, Uj=a+jtanψ, j=1,,nt. Denote cL=c+(nt-n0)tanφ, cU=c+(nt-n0)tanψ, a=a+nttanψ, and d=d+nttanφ. With the decision rule (1.3), we rewrite the system of (3.1) asR1(θ0)+S1(θ0)-L1(θ0)=γ3,R-1(θ0)+S-1(θ0)-L-1(θ0)=γ4,L1(θ1)+L-1(θ1)+L0(θ1)=γ1,L1(θ-1)+L-1(θ-1)+L0(θ-1)=γ2,N0(θASN)+N1(θASN)+N-1(θASN)+ntJ(θASN)=N, where R1(θ)=Pθ(Accept  H1  through  AL  when  nnt);R-1(θ) = Pθ(Accept  H-1 through DQ  when  nnt); S1(θ)=Pθ(cU<Tnt<a, Lj<Tj<Uj, j=1,,nt-1); S-1(θ)=Pθ(d<Tnt<cL, Lj<Tj<Uj, j=1,,nt-1); L1(θ)=Pθ(Accept  H0  through  CM  when  n>nt); L-1(θ) = Pθ(Accept  H0  through  CP  when  n>nt); L0(θ)=Pθ(Accept  H0  at  n=nt); J(θ)=S1(θ)+S-1(θ)+L0(θ);N0(θ) is the average sample number from a point in (d,a) at n=0 to the point of accepting H1 or H-1 when nnt. N1(θ) is the average sample number from a point in (cU,a) at n=nt to the point of making a decision when n>nt. And N-1(θ) is the average sample number from a point in (d,cL) at n=nt to the point of making a decision when n>nt.

The following theorem provides the approximate formulas through the numerical quadrature for the quantities in (3.2). In fact, these formulas are developed by a stepwise dealing for the continuing sampling area before n0 and the results by Li and Pu in  for the parallel lines areas inside AL//CM and inside CP//DQ, respectively. With such an idea, the proof of Theorem 3.1 is trivial and is neglected here.

Theorem 3.1.

Assume that X1,X2, are i.i.d. observations. Let fθ(x) and Fθ(x) be the p.d.f. and c.d.f. of X, respectively. Assume that Fθ-(x)=Pθ(X<x). Denote g̃1θ(x)=fθ(x), and g̃j+1θ(x) = i=1mω(ui(j))g̃jθ(ui(j))fθ(x-ui(j)), where ui(j) is the ith numerical quadrature root for [Lj,Uj], i=1,,m, j=1,,nt-1, and ω(u) is the corresponding weight for the numerical quadrature root u. Let ui(nt) and ui(nt)' be the ith numerical quadrature root for [cU,a] and for [d,cL], respectively, i=1,,m.

Then, the approximate values R̃1(θ), R̃-1(θ), S̃1(θ), S̃-1(θ), L̃1(θ), L̃-1(θ), L̃0(θ), Ñ0(θ), Ñ1(θ), and Ñ-1(θ) for the quantities in (3.2) are the following.

R̃1(θ)=1-Fθ-(U1)+j=2nti=1mω(ui(j-1))g̃j-1θ(ui(j-1))[1-Fθ-(Uj-ui(j-1))].

R̃-1(θ)=Fθ(L1)+j=2nti=1mω(ui(j-1))g̃j-1θ(ui(j-1))Fθ(Lj-ui(j-1)).

S̃1(θ)=i=1mω(ui(nt-1))g̃nt-1θ(ui(nt-1))[Fθ-(a-ui(nt-1))-Fθ(cU-ui(nt-1))].

S̃-1(θ)=i=1mω(ui(nt-1))g̃nt-1θ(ui(nt-1))[Fθ-(cL-ui(nt-1))-Fθ(d-ui(nt-1))].

Denote qθ(1)(x)=[qθj(1)(x)]1×m, where qθj(1)(x)=ω(uj(nt))fθ(uj(nt)+tanψ-x),   j=1,,m; pθ(1)=[pθi(1)]m×1, where pθi(1)=Fθ(cU+tanψ-ui(nt)),   i=1,,m; Qθ(1)=[qθij(1)]m×m, where qθij(1)=ω(uj(nt))fθ(uj(nt)+tanψ-ui(nt)), i,j=1,,m. Let I be the m×m identity matrix. Then, there is L̃1(θ)=i=1mω(ui(nt))g̃ntθ(ui(nt))OC̃1θ(ui(nt)), where OC̃1θ(x)=Fθ(cU+tanψ-x)+qθ(1)(x)(I-Qθ(1))-1pθ(1).

Denote qθ(2)(x)=[qθj(2)(x)]1×m, where qθj(2)(x)=ω(uj(nt)')fθ(uj(nt)'+tanφ-x), j=1,,m; pθ(2)=[pθi(2)]m×1, where pθi(2)=Fθ(d+tanφ-ui(nt)'), i=1,,m; Qθ(2)=[qθij(2)]m×m, where qθij(2)=ω(uj(nt)')fθ(uj(nt)'+tanφ-ui(nt)'), i,j=1,,m. Then, we have L̃-1(θ)=S̃-1(θ)-i=1mω(ui(nt))g̃ntθ(ui(nt))OC̃-1θ(ui(nt)), where OC̃-1θ(x)=Fθ(d+tanφ-x)+qθ(2)(x)(I-Qθ(2))-1pθ(2).

L̃0(θ)=i=1mω(ui(nt-1))g̃nt-1θ(ui(nt-1))[Fθ(cU-ui(nt-1))-Fθ-(cL-ui(nt-1))].

Ñ0(θ)=1+j=2ntji=1mω(ui(j-1))g̃j-1θ(ui(j-1))[1-Fθ-(Uj-ui(j-1))+Fθ(Lj-ui(j-1))].

Ñ1(θ)=i=1mω(ui(nt))g̃ntθ(ui(nt))ñ1θ(ui(nt)), where ñ1θ(x)=1+qθ(1)(x)(I-Qθ(1))-11 with 1 being the m×1 vector of 1's.

Ñ-1(θ)=i=1mω(ui(nt))g̃ntθ(ui(nt))ñ-1θ(ui(nt)), where ñ-1θ(x)=1+qθ(2)(x)(I-Qθ(2))-11.

Notice that the values on the left side of the equations in (3.2) must be obtained through a computer program with much iterative work, which reveals the method's complexity in computation and impairs the speed of solving the system of (3.2). Nevertheless, the time it costs is tolerable when the accuracy of solving the equations is not too demanded.

Example 1 (Continued).

Consider the same problems as those in Example 1 in Section 2. By applying the formulas (3.3)–(3.12) and the 64 Gaussian quadrature roots, we solve the system of (3.2) in a computer program. The hypothesis designs and the corresponding test schemes are listed in Columns 6–9 of Tables 14. As a comparison with the method under the asymptotic formulas in Section 2, ɛk in Column 10 of Tables 14 records the relative difference between the two hypothesis designs of the two methods. The Monte Carlo simulation study with 1,000,000 replicates in Tables 9, 10, 11, and 12 reveal the schemes' real performances.

Simulated performances for the schemes under Gaussian quadrature formulas in Table 1.

 N When μ1 is true When μ0 is true ASN′(μ0) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0021 0.9979 0.0010 0.9980 0.0010 5.0003 0.01 10 0 0.0020 0.9980 0.0010 0.9980 0.0010 10.0078 0.08 15 0 0.0019 0.9981 0.0010 0.9980 0.0009 14.9949 0.03 20 0 0.0020 0.9980 0.0010 0.9980 0.0010 19.9888 0.06 25 0 0.0020 0.9980 0.0011 0.9979 0.0010 24.9980 0.01 30 0 0.0020 0.9980 0.0010 0.9980 0.0010 29.9972 0.01

Simulated performances for the schemes under Gaussian quadrature formulas in Table 2.

 N When μ1 is true When μ0 is true ASN′(μ0) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0500 0.9500 0.0250 0.9501 0.0249 4.9998 0.00 10 0 0.0499 0.9501 0.0249 0.9499 0.0251 10.0032 0.03 15 0 0.0486 0.9514 0.0271 0.9456 0.0274 15.0038 0.03 20 0 0.0520 0.9480 0.0228 0.9543 0.0229 20.0061 0.03 25 0 0.0510 0.9490 0.0236 0.9530 0.0234 24.9795 0.08 30 0 0.0512 0.9488 0.0235 0.9532 0.0233 29.9926 0.02

Simulated performances for the schemes under Gaussian quadrature formulas in Table 3.

 N When μ1 is true When μ0 is true ASN′(μ1) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0019 0.9981 0.0011 0.9979 0.0010 4.9994 0.01 10 0 0.0020 0.9980 0.0010 0.9980 0.0010 9.9917 0.08 15 0 0.0020 0.9980 0.0010 0.9980 0.0010 15.0154 0.10 20 0 0.0020 0.9980 0.0010 0.9979 0.0010 20.0129 0.06 25 0 0.0020 0.9980 0.0010 0.9980 0.0010 24.9880 0.05 30 0 0.0020 0.9980 0.0010 0.9980 0.0010 30.0151 0.05

Simulated performances for the schemes under Gaussian quadrature formulas in Table 4.

 N When μ1 is true When μ0 is true ASN′(μ1) ɛ  (%) Probability of accepting Probability of accepting μ-1 μ0 μ1 μ-1 μ0 μ1 5 0 0.0499 0.9501 0.0252 0.9500 0.0248 4.9973 0.05 10 0 0.0499 0.9501 0.0254 0.9496 0.0250 9.9876 0.12 15 0 0.0505 0.9495 0.0250 0.9499 0.0252 15.0181 0.12 20 0 0.0502 0.9498 0.0249 0.9502 0.0249 20.0105 0.05 25 0 0.0498 0.9502 0.0251 0.9500 0.0249 24.9896 0.04 30 0 0.0499 0.9500 0.0250 0.9502 0.0248 30.0180 0.06

The real performances in Tables 912 show that the requirements on controlling the error probabilities and ASN may be fully made use of under this method and the numerical quadrature formulas are almost accurate. Therefore, the hypothesis designs and test schemes are highly efficient in terms of, for instance, more efficient designs with smaller k in Tables 14 under this method.

To further explain the methods, an example of the airbag quality inspection is provided in the appendix.

4. Conclusions and Remarks

For the three-hypothesis test problems, the methods of designing the hypotheses, together with obtaining the corresponding test schemes, are proposed by adopting asymptotic formulas or numerical quadrature formulas in this paper. As a helpful guide for practitioners, they aid to directly find proper hypotheses under controlled risks and costs in preventing from too many iterative trials on combinations of hypotheses to meet practical needs.

The asymptotic formulas and the numerical quadrature formulas are both alternative tools for the hypothesis designs. Several aspects should be considered when choosing between them in applications.

The method with numerical quadrature formulas outperforms the one under the asymptotic formulas especially when the error probabilities are not very small, as the example shows. In reality, the required error probabilities always range from 0.05 to 0.30 in sequential inspections, which seems to suggest choosing the numerical quadrature formulas to obtain efficient designs.

In computation, the asymptotic formulas provide great convenience for applications, while the numerical quadrature formulas demand much iterative computational work especially when the number of numerical quadrature roots is large. But from the computation with the 64 Gaussian quadrature roots in the example, the time it costs in a common computer is tolerable if the start values for the system of equations are proper. We recommend finding the designs under the asymptotic formulas first, and then apply them as starts to obtain more efficient hypotheses from the numerical quadrature formulas when needed.

When adopting the asymptotic formulas, the expressions for the quantities Dθi,Oθi(i=-1,0,1), and v should be developed for a specific distribution (see ). For the use of numerical quadrature formulas, the quadrature roots may be particularly arranged to fit the support points in the discrete distributions (e.g, see Reynolds and Stoumbos ). And for the θASN out of (θ-1,θ0,θ1), only the method with numerical quadrature formulas may take effect.

Actually, the two methods may apply to any distribution out of the Koopman-Darmois family. However, the test schemes under these distributions may be different from that in Figure 1, and the numerical quadrature formulas should be changed according to the test scheme patterns.

For the hypothesis designs asymmetric with the null hypothesis or the multihypothesis test problems, the methods proposed in this paper are still applicable by some extensions of adding more constraints on the designs. The hypothesis design problems under other requests, for example, under the desire of stopping sampling before a limit guaranteed by a provided probability, are still open to scholars and practitioners.

Appendix Illustration of Airbag Quality Inspection

According to Li et al. , the airbag deployment pressure rate per unit of time, which is always assumed to conform with a standard normal distribution after some standardized transformation, is a key index of the airbag quality. The concerned problem here is whether the quality index is zero, positive, or negative. This quality index is measured in a 100 cubic feet testing air tank with sensors and the inspection is destructive. Since the airbag is expensive, the three-hypothesis sequential test scheme is needed to reduce the average inspection costs.

Suppose that the two error probabilities are α=β=0.05 and the required ASN(μ0) is no more than 5. Then, the hypothesis designs and test schemes of “N=5" in Table 2 should be taken, that is, (k,n0,a,c,d,tanψ,tanφ) = (1.6066,2.0072,2.0438,0, -2.0438,0.8033,-0.8033) under the method with asymptotic formulas and (k,n0,a,c,d,tanψ,tanφ) = (1.3025,2.1953,2.2175,0,-2.2175,0.6513,-0.6513) under the method with Gaussian quadrature formulas.

Under the method with asymptotic formulas, the hypothesis test problem should be H-1:μ=-1.6066vs.H0:μ=0vs.H1:μ=1.6066. Taking the simulated observations from N(0,1) by Li et al. in , we may reach a decision of accepting H0 when T3=-0.0642 falls in [c+(3-n0)tanφ,c+(3-n0)tanψ]=[-0.7975,0.7975] according to the test process in Table 13.

Test process under the test scheme from asymptotic formulas.

 j Xj Tj a+jtan ψ c+(j-n0)tan ψ c+(j-n0)tan φ d+jtan φ Decision when j≥n0 when j≥n0 1 0.5689 0.5689 2.8471 * * -2.8471 * 2 -0.2556 0.3133 3.6504 * * -3.6504 * 3 -0.3775 -0.0642 4.4537 0.7975 -0.7975 -4.4537 Accept H0

Under the method with Gaussian quadrature formulas, the hypothesis test problem should be H-1:μ=-1.3025vs.H0:μ=0vs.H1:μ=1.3025. Also taking the simulated observations from N(0,1) by Li et al. in , we may accept H0 after inspecting the third airbag according to the test process in Table 14.

Test process under the test scheme from Gaussian quadrature formulas.

 j Xj Tj a+jtan ψ c+(j-n0)tan ψ c+(j-n0)tan φ d+jtan φ Decision when j≥n0 when j≥n0 1 0.5689 0.5689 2.8688 * * -2.8688 * 2 -0.2556 0.3133 3.5201 * * -3.5201 * 3 -0.3775 -0.0642 4.1714 0.5241 -0.5241 -4.1714 Accept H0
Acknowledgments

This work was partly supported by the National Natural Science Foundation of China (NSFC) under the project Grants numbers 60573125 and 60873264. The authors cordially thank the editor and anonymous referees for their helpful reviews.

FuK. S.Sequential Methods in Pattern Recognition and Learning1968New York, NY, USAAcademic PressWatersW. E.Sequential sampling in forest insect surveysForest Science195516879LyeB.StoryR. N.Spatial dispersion and sequential sampling plan of the southern green stink bug on fresh market tomatoesEnvironmental Entomology1989181139144McMillenT.HolmesP.The dynamics of choice among multiple alternativesJournal of Mathematical Psychology2006501305710.1016/j.jmp.2005.10.003MR2208063ZBL1139.91028BussgangJ. J.Sequential methods in radar detectionProceedings of the IEEE1970585731743GrossiE.e.grossi@unicas.itLopsM.lops@unicas.itSequential along-track integration for early detection of moving targetsIEEE Transactions on Signal Processing2008568396939822-s2.0-002470251810.1109/TSP.2008.924870GoodmanN. A.goodman@ece.arizona.eduVenkataP. R.NeifeldM. A.neifeld@ece.arizona.eduAdaptive waveform design and sequential hypothesis testing for target recognition with active sensorsIEEE Journal on Selected Topics in Signal Processing2007111051132-s2.0-2294443968610.1109/JSTSP.2007.897053AndersonS. L.A simple method of comparing the breaking loads of two yarnsTextile Institute195445472479LiteanuC.RicaI.Statistical Theory and Methodology of Trace Analysis1980New York, NY, USAHalstedTartakovskyA. G.tartakov@math.usc.eduRozovskiiB. L.rozovski@math.usc.eduBlažekR. B.blazek@math.usc.eduKimH.hongjoong@korea.ac.krDetection of intrusions in information systems by sequential change-point methodsStatistical Methodology2006332522932-s2.0-034595579910.1016/j.stamet.2005.05.003WetherillG. B.GlazebrookK. D.Sequential Methods in Statistics19863rdLondon, UKChapman & Hallxii+264Monographs on Statistics and Applied ProbabilityMR865649ChenT.-H.chen.tsuhao@msa.hinet.netChenC.-Y.chency@mail.ytit.edu.twYangH.-C. P.hcyang@ccms.nkfust.edu.twChenC.-W.cwchen@mail.stu.edu.twA mathematical tool for inference in logistic regression with small-sized data sets: a practical application on ISW-ridge relationshipsMathematical Problems in Engineering20082008121863722-s2.0-002902296710.1155/2008/186372OliveiraT. F.MiserdaR. B.CunhaF. R.Dynamical simulation and statistical analysis of velocity fluctuations of a turbulent flow behind a cubeMathematical Problems in Engineering200720072810.1155/2007/2462724627MR2417224LiM.ZhaoW.Variance bound of ACF estimation of one block of fGn with LRDMathematical Problems in Engineering201020101456042910.1155/2010/560429LiM.ChenW.-S.HanL.Correlation matching method for the weak stationarity test of&nbsp;LRD trafficTelecommunication Systems2010433-418119510.1007/s11235-009-9206-52-s2.0-70349893215BakhoumE. G.TomaC.Relativistic short range phenomena and space-time aspects of pulse measurementsMathematical Problems in Engineering200820082010.1155/2008/410156410156MR2471419ZBL1163.83319CattaniC.Harmonic wavelet approximation of random, fractal and high frequency signalsTelecommunication Systems2010433-4207217CattaniC.KudreykoA.Application of periodized harmonic wavelets towards solution of eigenvalue problems for integral equationsMathematical Problems in Engineering20102010857013610.1155/2010/570136ArmitageP.Sequential analysis with more than two alternative hypotheses, and its relation to discriminant function analysisJournal of the Royal Statistical Society. Series B195012137144MR0038618ZBL0038.09502PaytonM. E.YoungL. J.A sequential procedure for deciding among three hypothesesSequential Analysis199413427730010.1080/07474949408836310MR1308964ZBL0816.62065PaytonM. E.YoungL. J.A sequential procedure to test three values of a binomial parameterMetrika1999491415210.1007/s001840050024MR1713648ZBL1093.62566DragalinV. P.TartakovskyA. G.VeeravalliV. V.Multihypothesis sequential probability ratio tests. I. Asymptotic optimalityIEEE Transactions on Information Theory19994572448246110.1109/18.796383MR1725130ZBL1131.62313DragalinV. P.TartakovskyA. G.VeeravalliV. V.Multihypothesis sequential probability ratio tests. II. Accurate asymptotic expansions for the expected sample sizeIEEE Transactions on Information Theory20004641366138310.1109/18.850677MR1768555ZBL1059.62581LiY.PuX. L.A method on designing three-hypothesis test problemsto appear in Communications in Statistics—Simulation and ComputationReynoldsM. R.Jr.mrr@vt.eduStoumbosZ. G.stoumbos@andromeda.rutgers.eduThe SPRT chart for monitoring a proportionIIE Transactions19983065455612-s2.0-0000723639LiY.PuX. L.xlpu@stat.ecnu.edu.cnTsungF.Adaptive charting schemes based on double sequential probability ratio testsQuality and Reliability Engineering International200925121392-s2.0-2114444074610.1002/qre.938