SEMIPARAMETRIC INFERENTIAL PROCEDURES FOR COMPARING MULTIVARIATE ROC CURVES WITH INTERACTION TERMS
|
|
- Heather Garrison
- 5 years ago
- Views:
Transcription
1 Sttistic Sinic ), SEMIPARAMETRIC INFERENTIAL PROCEDURES FOR COMPARING MULTIVARIATE ROC CURVES WITH INTERACTION TERMS Linsheng Tng 1 nd Xio-Hu Zhou 2,3 1 George Mson University, 2 VA Puget Sound Helth Cre System nd 3 University of Wshington Abstrct: Multivrite ROC curve models tht include n interction term between biomrker type nd flse positive rte re importnt in comprtive biomrker studies, becuse such interction llows ROC curves of different biomrkers to cross ech other. However, there hs been limited work in drwing inference for compring multivrite ROC curves, especilly when interction terms re present. In this rticle we derive the symptotic covrince of three estimtors for multivrite ROC models. These covrince estimtes hve not been redily vilble in the literture, nd bootstrp methods hve been used to obtin them. With the redily vilble vrince estimtes, we cn esily perform hypothesis testing mong ROC curves, while bootstrp tests re not so esily performed. The symptotic results re pplied to compre ROC curves nd their res under ROC curves. Moreover, we derive simultneous confidence bnds for multivrite ROC curves. We evlute nd compre the finite smple performnce of our symptotic covrince estimtors. We lso discuss the dvntge of using our symptotic results over bootstrp procedures. Finlly, we illustrte our pproch through well-known pncretic cncer study. Key words nd phrses: Bootstrp, dignostic ccurcy, simultneous confidence bnd. 1. Introduction Reserch in erly cncer detection involves developing dignostic tools, such s biomrker, to distinguish disesed ptients from non-disesed ptients. Biomrkers often yield continuous mesurements. A populr tool to evlute nd compre the ccurcy of biomrkers is receiver operting chrcteristic ROC) curve Zhou, McClish nd Obuchowski 2002)), which is plot of true positive rtes vs flse positive rtes cross ll thresholds. Estimting single binorml ROC curve of continuous-scle biomrker hs been well studied in the literture Metz, Hermn nd Shen 1998) nd Ci nd Moskowitz 2004)). However, inferentil procedures for compring multivrite ROC curve models
2 LIANSHENG TANG AND XIAO-HUA ZHOU Figure 1. Empiricl ROC curves for CA 19 9 nd CA 125: solid line, CA 19 9; dshed line, CA 125. with interction terms between biomrker type nd flse positive rtes FPRs) hve not been well studied, minly becuse it is sometimes difficult to drw inferences the presence of these interction terms. However, such interctions re importnt becuse they llow ROC curves to cross ech other. For exmple, Wiend, Gil, Jmes nd Jmes 1989) studied pncretic cncer biomrkers, CA 19 9 nd CA 125, which were mesured on 51 pncretitis ptients nd 90 pncretic cncer ptients. The empiricl ROC curves were generted from this dt set nd their plots re shown in Figure 1. It is cler tht two ROC curves cross ech other when FPR gets close to 1. This shows the existence of interction terms between biomrker type nd FPRs. Metz, Wng nd Kronmn 1984) proposed mximum likelihood estimtor MLE) for estimting bivrite binorml models from ordinl dt, but their method requires estimting correltion prmeters, besides the loction nd scle prmeters in mrginl norml distributions. It would be difficult to extend their MLE method to more thn two ROC curves when mny more correltion prmeters re to be estimted. As the number of biomrkers gets lrge, the MLE method becomes inpplicble. For multiple independent ROC curves, Zhng 2004) nd Zhng nd Pepe 2005) proposed n intuitive lest squres LS) method, nd Pepe 2000) presented n elegnt generlized liner model GLM) pproch. Since it is complicted to derive symptotic results, the uthors did not consider lrge smple inference for the LS nd GLM estimtors with clustered d t. Ci nd Pepe 2002) considered n interesting semiprmetric
3 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1205 generlized estimting eqution GEE) method to llow n unknown bseline function when estimting ROC curves from correlted biomrker dt. They derived symptotic results for their estimtors, but they did not consider how to compre multiple ROC curves from clustered dt with the presence of interctions between biomrker type nd FPRs. In this rticle we consider the LS method for clustered ROC curve dt with the presence of interctions between biomrker type nd FPRs. We derive explicit covrince structures between empiricl ROC curves nd use our results to derive the symptotic covrinces of the LS estimtor. We lso dpt the GLM nd GEE methods to this type of biomrker dt nd derive their symptotic sndwich covrince estimtors. These covrinces hve not been redily vilble in the literture, nd bootstrp methods hve been used to obtin them. With redily vilble vrince estimtes we cn esily perform hypothesis testing mong ROC curves, while bootstrp tests re not so esily performed. For exmple, if we wnt to test whether two ROC curves vry by certin mount δ t specified FPR u 0, i.e., H 0 : ROC 1 u 0 ) ROC 2 u 0 ) = δ, it is not cler how to bootstrp from the null distribution, while it is strightforwrd to perform such hypothesis tests using our symptotic results nd the delt method tht is introduced in Section 4. We derive inferentil procedures for compring multivrite ROC curves tht include interction terms, which hve multivrite binorml ROC curves s specil cse. In prticulr, we develop methods for compring ROC curves nd the res under these ROC curves. We derive symptotic simultneous confidence bnds for ROC curves. Such symptotic results of simultneous confidence bnds of ROC curves re rrely studied. Insted, computer intensive methods re often employed to construct confidence bnds Ci nd Pepe 2002)). Our confidence bnds provide n esy-to-use tool to illustrte the smpling vribility of the ROC curve estimtes. In ddition, we develop method for compring multiple ROC curves t some specified FPR. This rticle is orgnized s follows. In Section 2 we consider LS, GEE nd GLM methods for multivrite ROC curves. In Section 3 we derive the symptotic results for the LS method when estimting multivrite ROC models; the symptotic results re lso derived for GLM nd GEE methods. In Section 4 we pply the results to compre ROC curves nd the res under them. In ddition, symptotic simultneous confidence bnds re derived for multivrite ROC curves. We crry out lrge scle simultion studies to evlute nd compre the finite smple performnce of our covrince estimtors from the LS, GLM nd GEE methods. We lso crry out simultion studies to evlute the dvntges of using symptotic results over using bootstrp procedures. The simultion results re summrized in Section 5. A comprtive pncretic cncer dignostic tril serves s n illustrtive exmple in Section 6, nd some discussion is presented in Section 7.
4 1206 LIANSHENG TANG AND XIAO-HUA ZHOU 2. Three Estimtors of Multivrite ROC Curves In this section, we dpt the LS nd GLM methods to clustered ROC dt nd give simplified version of the GEE method for estimting multivrite ROC curves. Let X l = X l,1,..., X l,m ) denote mesurements of the lth biomrker on m disesed subjects, nd Y l = Y l,1,..., Y l,n ) denote mesurements of the lth biomrker on n helthy subjects, where l, l = 1,..., K. For the lth nd lth different biomrkers mesured on the ith disesed subject, i = 1,..., m, the mesurements X l,i nd X l,i follow bivrite survivl function F l, l with the mrginl distributions F l nd F l, respectively, nd the mesurements of the jth helthy subject, Y l,j nd Y l,j with j = 1,..., n, follow bivrite survivl function Ḡl, l with the mrginl distributions Ḡl nd Ḡ l, respectively. The ROC curve of the lth biomrker is then given by Q l u) = F l Ḡ 1 l u)), nd its empiricl form is Q l u) = F l Ḡ 1 l u)), where F l nd Ḡ l re empiricl functions of F l nd Ḡ l, respectively. Let Z k be dummy vrible for the kth biomrker, k = 2,..., K. The multivrite ROC curves re given by nd Q 1 u) = gθ 10 + θ 11 hu)}, [ K ] Q k u) = g θ 10 + θ 11 hu) + θ k0 Z k + θ k1 Z k hu)}, 2.1) for 0 < u b < 1, where g is some specified link function nd h is some specified bseline function. In the regression ROC modeling, θ 10 + θ 11 hu) is the bseline function, usully denoted s h 0 u). In this rticle we tke this bseline function to be known function, up to two unknown prmeters θ 10 nd θ 10, s in Pepe 2000), Zhng 2004) nd Zhng nd Pepe 2005). The importnt components of the model 2.1) re the θ k1 Z k hu), which re the interction terms between FPRs nd dummy vribles indicting biomrker types. Such interction terms ply n importnt role in estimting ROC curves, especilly when estimting the multivrite binorml ROC curves. With these terms, the 2.1) includes the multivrite binorml ROC model s the specil cse tht is commonly seen in the literture Metz, Wng nd Kronmn 1984)). Specificlly, let g be Φ, let h be Φ 1, nd let Z k be n indictor vrible for the kth biomrker, so 2.1) cn be written s k=2 Q 1 u) = Φθ 10 + θ 11 Φ 1 u)} nd Q k u) = Φθ 10 + θ 11 Φ 1 u) + θ k0 + θ k1 Φ 1 u)}, 2.2) for k = 2,..., K. When there is only one biomrker, 2.2) reduces to the commonly used binorml model Zhou, McClish nd Obuchowski 2002)).
5 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1207 Let u l = u l,1,..., u l,pl ) T be some fixed prtition points in [, b] on the lth empiricl ROC curve. Here P l is rbitrrily chosen for the lth ROC curve where 0 < = u l,1 < < u l,pl = b < 1. For exmple, if we choose 50 jump points for the 1st ROC curve, we cn choose u 1 = 1/51,..., 50/51). Also, for simplicity, we write Lu l ) = Lu l,1 ),..., Lu l,pl )) T, where l = 1,..., K, for ny process or function L. Zhng 2004) nd Zhng nd Pepe 2005) proposed lest squres LS) pproch to estimte multiple ROC curves. They let Z k be the indictor vribles for the kth biomrker. In the LS estimting procedure, the ROC curve corresponding to reference biomrker is chosen s the reference ROC curve. For the lth empiricl ROC curve, the prtition points, u l = u l,1,, u l,pl ) T, re chosen within intervl boundries [, b]. When we re interested in the entire ROC curve, nd b cn be chosen to be close to 0 nd 1, respectively. If prtil ROC curve is of interest, nd b cn be chosen ccordingly. By plugging in the empiricl functions F 1 l nd Ḡ l, the lth empiricl ROC curve is Q l u l ) = F l Ḡ 1 l u l )). Let Ỹl = g 1 Q l u l )). We combine Ỹ1,..., ỸK to get the liner regression eqution Ỹ = Mθ + ɛ, 2.3) where Ỹ = Ỹ 1 T,..., Ỹ K T )T is l P l) 1 vector with elements Ỹl = g 1 Q l u l )). The l P l) 2K) design mtrix M is M M 2 M M = M 3 0 M3 0, with its P l 2 submtrices 1 1 M l = hu l,1 ) hu l,pl ) M K 0 0 M K ) T ) T, nd Mk Zk = Z k. Z k hu l,1 ) Z k hu l,pl ) Also, the error term ɛ hs multivrite norml distribution given by ɛ N0, Σ ɛ ). The detiled proof of its multivrite normlity is given in the online version of the pper t Bsed on the regression equtions in 2.3), the LS estimtor ˆθ LS of θ is given by ˆθ LS = M T M) 1 M T Ỹ. There re severl other estimting eqution methods for estimting ROC prmeters. Pepe 2000) observed tht the expected vlue of indictor vribles
6 1208 LIANSHENG TANG AND XIAO-HUA ZHOU 1 IX l,i Ḡ l u l,p )} converges to the true ROC curve of the lth biomrker nd proposed GLM method to estimte the ROC curve model 2.1). If prtil ROC curves on [, b] re of interest, the u l,p re chosen within this rnge. The GLM pproch estimtes prmeters in the model 2.1) by the estimting equtions K P l m i=1 or g M l,p T θ) 1 } } Ml,p g M l,p T θ)1 g M l,p T θ)) I X l,i Ḡ l u l,p ) g M l,p T θ) = 0, K P l w l u l,p ) M l,p Ql u l,p ) g M l,p T θ)} = 0, for l = 1,..., K nd p = 1,..., P l, with θ = θ 10, θ 11, θ 20, θ 21,..., θ K0, θ K1 ) T nd weight function w l u l,p ) = g M l,p T θ)}/g M l,p T θ)1 g M l,p T θ))}. Here M 1,p is 1 2K vector with the first two elements being 1 nd hu 1,p ), respectively, nd the rest of the elements zero. The M k,p hve the first two elements being 1 nd hu k,p ), respectively, the 2k + 1)th nd 2k + 2)th elements being Z k nd Z k hu k,p ), respectively, nd the rest of the elements being zero. The symptotic results of the prmeter estimtor not provided in Pepe 2000) re given in Section 3. The GEE method of Ci nd Pepe 2002) lso used the indictor vribles 1 IX l,i Ḡ l u l,p )}, nd relied on the fct tht points on ROC curves cn be interpreted s conditionl expecttions of these indictor vribles. Their method is flexible enough to llow n unknown bseline function h 0 in the model 2.1). When the bseline function hs the form of h 0 u) = θ 10 + θ 11 hu), the GEE method is similr to the GLM method. Specificlly, the GEE method solves the estimting equtions K P l } M l,p Ql u l,p ) g M l,p T θ)) = 0. It is observed tht if the bseline function hs known form, the GLM method differs from the GEE method by including the weight function w l. 3. Lrge Smple Theory of ROC Estimtors We derive the symptotic covrince structure of multiple empiricl ROC curves for clustered dt. The result is then pplied to derive symptotic covrinces of the LS estimtor. We lso derive symptotic sndwich covrinces of GLM nd GEE estimtors. Asymptotic results of the LS nd GLM estimtors for clustered dt hve not been provided in the literture Pepe 2000),
7 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1209 Zhng 2004) nd Zhng nd Pepe 2005)). Although Ci nd Pepe 2002) hve studied the lrge smple theory of the GEE estimtor, their covrince estimtor hs complicted form. For the multivrite ROC models we consider here, the covrince estimtor of resulting estimtor hs simplified covrince estimtor tht is esy to pply. These covrince results re essentil for drwing inference on compring ROC curves nd constructing simultneous confidence ROC bnds, s discussed in Section 4. We derive the covrince structure between empiricl ROC curves, nd summrize the results in Theorem 1. The proof of Theorem 1 is given in the online version of the pper t Theorem 1 provides bsis for deriving symptotic covrinces of the LS estimtor. Theorem 1. If 1) F l nd Ḡl hve continuous densities F l nd Ḡ l, respectively, nd 2) the first derivtive Q l of Q l is bounded in, b), when m/n λ s m, n, cov[ m Q l s) Q l s)}, m Q lt) Q lt)}] converges in distribution to [ Ḡ 1 Fl, l l for s, t in, b). } ] [ s), Ḡ 1 t) Q l l s)q lt) +λ Q Ḡl, lḡ 1 l s)q lt) l 3.1. Asymptotic covrince of the LS estimtor s), t)} st} ] Ḡ 1, l Let θ l1 = θ 11 when l = 1, nd θ l1 = θ 11 + θ l1 when l 2. Consider the 2K 2K squre mtrix 1 KD Z 2 D Z K D I 2 I 2 I 2 Z 2 D Z2 2 J = D O O I 2 O , Z K D O ZK 2 D O O I 2 where b b D = hu)du, ) b hu)du b h2 u)du for 0 < < b < 1, nd I 2 is 2 2 identity mtrix. Let V l s, t) = where l = 1,..., K, nd Q l s t) Q l s)q l t) g g 1 Q l s)))g g 1 Q l t))) + λ θ 2 l1 h s)h t)s t st), Ṽ l, ls, t) = F l, lḡ 1 l s), Ḡ 1 t)) Q l l s)q lt) g [g 1 Q l s)}])g g 1Q lt)))
8 1210 LIANSHENG TANG AND XIAO-HUA ZHOU +λ θ l1 θ l1 h s)h t)ḡl, l Ḡ 1 l s), Ḡ 1 t)) st}, l for l, l = 1,..., K, nd l l, where λ = lim m,n m/n. Theorem 2 gives results for the LS estimtor. The proof of Theorem 2 is given in the online version of the pper t Theorem 2. Under the regulrity conditions of Theorem 1, when m/n λ s m, n, nd P l, the regression prmeter estimtor ˆθ LS stisfies mˆθls θ) D N0, Σ LS = JΣ y J T ). Here Σ y is the mtrix Σ y 11 Σ y 12 Σ y 1K Σ y Σ y 21 Σ y 22 Σ y 2K =... Σ y K1 Σy K2 Σ y KK whose 2 2 digonl symmetric submtrices, Σ y ll hve elements σ 1,1) ll = b b σ 1,2) ll = σ 2,1) ll = V l s, t)dsdt, σ 2,2) b b b b ll = hs)v l s, t)dsdt, hs)ht)v l s, t)dsdt, nd whose 2 2 off-digonl symmetric submtrices, Σ y, hve elements l, l σ 1,1) l, l σ 1,2) l, l = b b = σ 2,1) l, l = Ṽ l, ls, t)dsdt, σ 2,2) l, l b b = hs)ṽl, l s, t)dsdt. b b hs)ht)ṽl, l s, t)dsdt In prctice, Fl, l, Ḡ l, l, Fl, nd Ḡl re unknown, nd re estimted by their respective empiricl functions. If the ROC dt re unclustered, the off-digonl submtrices, Σ y s, re zero mtrices. If we further let h = l, l g 1 nd Z k be indictor vribles in the model 1), we get the sme symptotic result s in Zhng 2004) Asymptotic covrinces of the GLM nd GEE estimtors The GLM estimtor, ˆθ L, is obtined by solving the estimting equtions Uθ) = K P l } w l u l,p ) M l,p Ql u l,p ) g M l,p T θ) = 0,
9 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1211 where M l,p is defined in Section 2.1. To solve these equtions for ˆθ L, we need the Newton-Rphson method. Then covˆθ L ) = Uθ) ) 1 K vr θ l=1 Uθ) ) 1, θ P ) w l u l,p ) M l,p Q l u l,p ) g M l,p T θ)} where Uθ)/ θ is the prtil derivtive of Uθ) with regrd to θ. Let U 1i = nd U 2j = K P l K P p=1 w l u l,p ) M l,p IX li Ḡ 1 l u l,p )) g M T l,p θ) }, w l u l,p ) M l,p Q l u l,p) IY lj Ḡ 1 l u l,p )) u l,p }. The result of Theorem 1 gives us tht, under mild regulrity conditions, when m/n λ s m, n, the GLM estimtor ˆθ L stisfies mˆθl θ) D N0, Σ L ), where K P l ) 1 Σ L = w l u l,p ) M l,p g M l,p T θ) K P l ) 1 w l u l,p ) M l,p g M l,p T θ). lim m,n m U 1i U1i T + λ i=1 n } U 2j U2j T The symptotic property of the modified GEE estimtor, ˆθ E, cn be similrly derived. Let U 1i = nd U 2j = K P l K P l M l,p IY li Ḡ 1 l u l,p )) g M T l,p θ) }, v=1 M l,p Q l u l,p) IY lj Ḡ 1 l u l,p )) u l,p }. When m/n λ s m, n, the GEE estimtor ˆθ E stisfies mˆθe θ) D N0, Σ E ),
10 1212 LIANSHENG TANG AND XIAO-HUA ZHOU where K P l ) 1 Σ E = M l,p g M l,p T θ) K P l ) 1 M l,p g M l,p T θ). lim m,n m r=1 U 1iU T 1i + λ n v=1 } U2jU 2j T These covrince estimtors re sndwich estimtors. finite smple performnce in our simultion studies. We will evlute their 4. Multivrite ROC Anlysis Estimted ROC curves, denoted by Q l, re obtined by replcing θ with the estimtor ˆθ in the model 2.1). ˆθ is estimted vi either of three forementioned methods. Here ˆθ is generl nottion nd cn be found vi other methods s well. But, since symptotic results re derived for the three methods, it is convenient to utilize these results. Our symptotic results give the corresponding covrince mtrix estimtor ˆΣ of Σ for ˆθ. For further nottionl convenience, let Σ l, l be the 2 2 submtrices of Σ 11 Σ 12 Σ 1K Σ = Σ 21 Σ 22 Σ 2K.... Σ K1 Σ K2 Σ KK 2K 2K The estimtor ˆΣ l, l of Σ l, l is the corresponding submtrix of ˆΣ. In this section we derive methods for pirwise comprison of ROC curves. We lso develop methods for compring more thn two res under ROC curves under multivrite binorml ssumptions. In ddition, we derive inferentil procedures for simultneous confidence ROC bnds, nd for compring multiple ROC curves t some specified FPR Pirwise comprison of ROC curves It is often of interest to compre ROC curves to investigte the ccurcy of biomrkers. The reference ROC curve, Q 1 u), nd the kth ROC curve, Q k u), in [, b] only differ by prmeter vector θ k = θ k0, θ k1 ) T. Consequently, testing the equlity of these two ROC curves is equivlent to testing H 0 : θ k = 0, 0) T. It follows from Section 3 tht the symptotic distribution of the test sttistic, κ k = ˆθ k θ k ) T Σ 1 kk ˆθ k θ k ) is χ 2 2. Similrly, testing the equlity of two ROC curves of the ωth nd νth different biomrkers in [, b], ω, ν = 2,..., K, lso
11 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1213 reduces to χ 2 test. The null hypothesis becomes H 0 : θ ω0 θ ν0, θ ω1 θ ν1 ) = 0. Let θ ω,ν = θ ω0, θ ω1, θ ν0, θ ν1 ) T. The resulting chi-squre sttistic is ˆθω0 ˆθ ) T ν0 ) θ ω0 θ ν0 ) κ ω,ν = ˆθ ω1 ˆθ AΣ ω,νa T ) 1 ˆθω0 ˆθ ) ν0 ) θ ω0 θ ν0 ) ν1 ) θ ω1 θ ν1 ) ˆθ ω1 ˆθ, ν1 ) θ ω1 θ ν1 ) where A = I 2, I 2 ) with 2 2 identity mtrix I 2, nd Σ ω,ν is the covrince mtrix of θ ω,ν Compring the res under multivrite binorml ROC curves Our symptotic results in Section 3 cn be used to compre res under multivrite ROC curves, especilly multivrite binorml ROC curves. Let A = A 1,..., A K ), where A l is the re under the lth ROC curve. It is well known tht under the binorml ssumption, θ 10 A 1 θ 10, θ 11 ) = Φ }, 1 + θ 2 11 θ 10 + θ k0 nd A k θ 10, θ 11, θ k0, θ k1 ) = Φ }. 4.1) 1 + θ11 + θ k1 ) 2 Let q be second-order differentible nd rel-vlued function of A. It follows tht when m/n λ s m, n, mq Â) qa)} D N0, σ 2 q), where σ 2 q = lim m m,n + K K k=2 k=2 A 1 A k A 1 vra 1 ) + 2 A k cova k, A k) K A 1 k=2 }. A k cova 1, A k ) By Tylor expnsions on A l nd our symptotic results, we find nd vra 1 ) = B T 1 Σ 11 B 1, cova 1, A k ) = B T 1 Σ 11 + Σ k1 )B k, cova k, A k) = B T k Σ 11 + Σ k k)b k, where B 1 = φ θ10 ) 1, φ 1 + θ θ 2 11 θ θ 2 11 ) θ θ 2 11 )3/2 } T,
12 1214 LIANSHENG TANG AND XIAO-HUA ZHOU [ } θ 10 + θ k0 1 B k = φ 1 + θ11 + θ k1 ) θ11 + θ k1 ), 2 } ] T θ 10 + θ k0 θ φ 11 + θ k1 1 + θ11 + θ k1 ) 2, 1 + θ 11 + θ k1 ) 2 } 3/2 nd Σ l l, for l, l = 1,..., K, is estimted from symptotic results. DeLong, DeLong nd Clrke-Person 1988) gve similr formul for compring the res under nonprmetric ROC curves. Although their pproch is robust, semiprmetric pproches my be more ppeling for deriving smooth ROC curves for continuous biomrker dt. If q is some liner function, the theoreticl result is simplified to similr formul in their pper. A simple exmple hs E s vector with the lth element 1, lth element 1 nd other elements zero. Then qâ) = E corresponds to Âl  l, whose vrince estimtor follows from the vrince result. Inference is then esily drwn for testing H 0 : A l = A l, nd for constructing confidence intervl Simultneous confidence ROC bnds The vrince of estimted ROC curves t ech FPR cn be derived from the prmeter estimtor ˆθ nd its covrince mtrix estimtor ˆΣ. Let H = 1, hu)), H = H, H). Corollry 2. The vrince of estimted ROC curves t u re nd σ1u) 2 = g [g 1 Q 1 u)}] 2 HΣ 11 H T ) σk 2 u) = g [g 1 Q k u)}] 2 Σ11 Σ 1k H Σ k1 Σ kk H T, respectively, for k = 2,..., K. The 1 α)100% pointwise confidence intervl of Q l u) is then given by Q l u) ± z α/2ˆσ l u), 0 u 1. In Theorem 3 below, we give explicit expressions for simultneous bnds for multivrite ROC curves. The proof of Theorem 3 is given in the online version of the pper t Theorem 3. Under the regulrity conditions of Theorem 1, the 1 α)100% simultneous confidence bnds for multivrite ROC curves in [, b] re g H ˆθ } 1 ± χ 2 2,α HΣ 11H T,
13 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1215 nd ) g Hˆθ 1 T, ˆθ k T )T ± Σ11 Σ χ 2 1k 4,α H Σ k1 Σ kk H T }, respectively, for k = 2,..., K. Note the reson why simultneous confidence bnds for ROC curves hve such simple expressions is tht we ssume tht g nd h re known. The estimted ROC curves re fully determined by two prmeters for the reference biomrker nd four prmeters for other biomrkers. Therefore, χ 2 2 nd χ2 4 distributions rise, nd the derivtion of simultneous bnds is nturlly simplified Compring multiple ROC curves t some specified FPR Let Qu) = Q 1 u),..., Q K u)) T. Similr to Section 4.2, suppose tht q is rel-vlued nd second-order differentible function of the vector Q. We hve tht q Qu 0 )), t specified FPR u 0, converges to norml distribution with men zero nd vrince σ 2 qu 0 ) = lim m m,n + K K k=2 k=2 [ Q 1 Q k Q 1 vrq 1 u 0 )} + 2 Q k K Q 1 ] k=2 covq k u 0 ), Q ku 0 )}. Q k covq 1 u 0 ), Q k u 0 )} Here we hve vrq 1 u 0 )} = C 1 u 0 ) T Σ 11 C 1 u 0 ), covq 1 u 0 ), Q k u 0 )} = C 1 u 0 ) T Σ 11 + Σ k1 )C k u 0 ), covq k u 0 ), Q ku 0 )} = C k u 0 ) T Σ 11 + Σ k k)c k u 0 ), where C 1 u) = g θ 10 + θ 11 hu)}, θ 11 g θ 10 + θ 11 hu)}) T, C k u) = [g θ 10 + θ 11 hu) + θ k0 + θ k1 hu)}, θ 11 + θ k1 )g θ 10 + θ 11 hu) + θ k0 + θ k1 hu)}] T. If q is liner function, the result cn be gretly simplified. 5. Simultion Studies 5.1. Finite smple performnce of hypothesis testing We rn lrge set of simultion studies to evlute nd compre the finite smple performnce of our symptotic covrince estimtors for the LS, GLM nd GEE methods. The bivrite norml dt were N1, 1), Σ 0 ) for the disesed nd N0, 0), Σ 0 ) for the helthy, where Σ 0 hs the vrinces 1 nd 2 with correltion prmeter ρ. True ROC curves of tests 1 nd 2 hve the sme form given by Q 1 u) = Q 2 u) = Φ1/ 2 + 1/ 2Φ 1 u)}, for 0 u 1. Thus, the true vlue of the prmeter vector is θ = 1/ 2, 1/ 2, 0, 0). We fit bivrite binorml model to the dt; tht is, we let K = 2 in the model 2.2). The null hypothesis of equl ROC curves, H 0 : θ 20, θ 21 ) = 0, 0), cn be tested
14 1216 LIANSHENG TANG AND XIAO-HUA ZHOU by the χ 2 test sttistic κ = ˆθ 20, ˆθ 21 )ˆΣ 1 22 ˆθ 20, ˆθ 21 ) T with 2 degrees of freedom, where ˆθ 20, ˆθ 21 ) s covrince mtrix estimte, ˆΣ 22, is clculted using symptotic results in Theorem 2. We simulted 1,000 dt sets under the null hypothesis with m = 50, 100, 200) nd n = 50, 100, 200) under ρ = 0, 0.25, 0.5, 0.75). The nominl rejection rte ws set t 5%. In the simultion, the vrinces of LS estimtors were estimted using symptotic results developed in Theorem 2. The vrinces of GLM nd GEE s estimtors were obtined using symptotic results in Section 3. For smll smple size such s 50, the GEE method sometimes did not converge nd we hd to run more thn 1,000 simultions in order to obtin 1,000 vlid estimtes. Since the LS method does not require itertions, the computtion time is gretly reduced compred to tht of GLM or GEE. Under the sme circumstnces, the computing time of LS is less thn hlf tht of GLM or GEE. In prticulr, we conducted our simultion study on the sme Unix mchine. It took 30 seconds for LS to estimte prmeters from 1,000 simulted dt sets when m = n = 200, while the computtionl times of GLM nd GEE were 870 nd 348 seconds, respectively. When m = n = 50, the computtion time of LS ws 24 seconds, while times of GLM nd GEE were 210 nd 98 seconds, respectively. Tble 1 presents rejection rtes from the three pproches. As shown in Tble 1, the symptotic results for LS work well for ll combintions of smple sizes s the rejection rtes re close to the nominl level of 5%. Even for smll smple size 50 for both disesed nd helthy groups, rejection rtes do vry much from the nominl level. Moreover, rejection rtes of LS re not ffected by vlues of the correltion prmeter ρ. From Tble 1, GEE nd GLM hve much in common. Both hve over-rejection rtes when smple sizes re smll. This is minly due to the vribility of their sndwich vrince estimtors. Reders re referred to Kuermnn nd Crroll 2001) for more detils. It is lso noticeble in Tble 1 tht s smple sizes for the helthy get lrger, rejection rtes of GEE nd GLM get closer to the nominl level even when smple sizes for the disesed re smll. However, rejection rtes for the disesed, given smll smple sizes for the helthy, re not improved with lrge smple sizes. We tried bootstrp methods in these situtions. The bootstrp method performed similrly to the LS method on the rejection rtes regrdless of smple sizes. The bootstrp performed better thn GLM nd GEE when smple sizes were smll, nd similrly to GLM nd GEE when smple sizes were lrge Finite smple performnce of point nd intervl estimtes We used the sme setting s in the previous section to evlute nd compre estimtion precision of the methods in the simultion study. We gin simulted 1,000 dt sets under smple sizes m = n = 50, 200, 400) with ρ = 0.5. The nominl coverge probbility of confidence intervls ws 95%. We pplied LS,
15 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1217 Tble 1. Rejection rtes in %) with the nominl level α = 0.05, from symptotic results. LS GEE GLM m n ρ = ρ = ρ = The rejection rte with 1,000 reliztions of norml model. The 95% prediction intervl of the rejection rte is 5.0% ± 1.4%). GEE nd GLM to get estimtes of the ROC prmeter vector θ 10, θ 11, θ 20, θ 21 ). Confidence intervls for the prmeters were clculted bsed on symptotic results. We then compred these methods bsed on bis, squre root of MSE RMSE), nd the coverge probbilities of confidence intervls. The results re shown in Tble 2. All three methods hve good ccurcy for estimting the prmeters, while coverge probbilities differ mong them. Our simultion results show tht LS hs the nice finite smple property tht the coverge probbilities of ll prmeters re close to the nominl level for smll smple sizes. When smple sizes re smll, confidence intervls computed from sndwich covrince estimtors for GLM nd GEE cover the intercept prmeters properly, but these confidence intervls over-cover slope prmeters. As smple sizes pproch 400, the coverges of GLM nd GEE estimtors get closer to the nominl level for slope prmeters The dvntge of our symptotic results over bootstrp procedures Mny uthors hve pplied bootstrp methods to estimte covrince mtrices for the LS nd GLM estimtors when the symptotic results were not vilble Pepe 2000) nd Zhng nd Pepe 2005)). However, it cn tke much more computtion time to do bootstrpping thn to use the symptotic results. In this simultion study, we compred coverge percentges of bootstrp covrince estimtes with those of symptotic covrince estimtes for the LS pproch. We used the sme setting in Section 5.1 with m = n = 50, 200, 400). Under ech combintion of smple sizes, we simulted 1,000 dt sets with ρ = 0.5. For ech dt set we pplied bootstrp procedures to get covrince estimtes of the LS
16 1218 LIANSHENG TANG AND XIAO-HUA ZHOU Tble 2. Bis, RMSE, coverge probbility CP) of prmeter estimtors. LS GEE GLM m Bis RMSE CP BT Bis RMSE CP Bis RMSE CP n) in %) in %) in %) in %) in %) in %) in %) 50 θ θ θ θ θ θ θ θ θ θ θ θ CP is the coverge percentge for 95% confidence intervls using symptotic stndrd errors with norml quntile. BT is the coverge percentge for 95% confidence intervls using the bootstrp. Results re bsed on 1,000 reliztions of the bivrite norml model. pproch. The number of bootstrp ws 1,000. We then used covrince estimtes to get confidence intervls nd their coverge percentges. The coverge percentges of bootstrp methods were shown in Tble 2. As cn be seen there, our symptotic results were s good s bootstrp results in tht coverge percentges were very close. More importntly, computtion time ws much shorter. For exmple, when using the LS method with m = n = 400, it took 30 seconds to obtin bootstrp covrince estimte for one dt set on PC, it took only 5 seconds to obtin n symptotic covrince estimte for the sme dt set on the sme PC. 6. Appliction to Pncretic Cncer Biomrkers The min interest in the forementioned biomrker exmple is to determine whether ROC curves generted by two biomrkers re equl. If not, one would wish to sy which biomrker cn better distinguish the disesed from the helthy. We pplied the LS estimtion procedure to this dt set. With the probit link, g = Φ, nd h = Φ 1, ROC curves of these biomrkers hve the sme structure s those in 2.2) when K = 2. We got the estimte of the prmeter vector s ˆθ 10 LS, ˆθ LS 11, ˆθ LS 20, ˆθ LS 21 ) = 1.18, 0.47, 0.49, 0.55). Its covrince mtrix estimte ws clculted bsed on the symptotic result in Theorem 2. The GLM nd GEE prmeter estimtes were very close to the LS estimte, nd thus were not listed. The χ 2 ws then clculted to be with the p-vlue , indicting significnt difference in dignostic ccurcy between two biomrkers. We lso
17 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1219 Figure 2. Estimted ROC curves nd their 95% confidence bnds for CA 19-9 nd CA 125: dshed lines, empiricl ROC; solid lines, estimted ROC; shded regions, 95% confidence bnd; dotted lines, confidence bnd boundries. Figure 3. Overlpping 95% confidence bnds for CA 19 9 nd CA 125: solid lines, estimted ROC; shded regions, 95% confidence bnd; dotted lines, confidence bnd boundries. clculted the difference between the two res under estimted ROC curves using the procedure in Section 4.2, nd found significnt difference between them, p-vlue To visulize the smpling vribility of estimted ROC curves, simultneous ROC bnds were constructed using the result in Theorem 3. Figure 2 shows the estimted ROC curves nd their simultneous bnds. These ROC curves were very close to the empiricl curves. Although our results on ROC curves nd their res show tht the two biomrkers re different, it is cler in Figure 3 tht two ROC curves intersect. The simultneous bnds cn help us determine the region of FPRs where two ROC curves re different; Figure 3 shows overlpping confidence bnds. It is obvious tht the ROC curve for CA 19 9 is significntly better thn tht for CA 125 when the flse positive rte is less thn round 0.17, nd tht the two ROC curves do not differ much elsewhere.
18 1220 LIANSHENG TANG AND XIAO-HUA ZHOU 7. Discussion The interction terms in our multivrite ROC model re completely different from the interctions of the K dignostic tests on the mesurement level. In fct, the interctions we refer to re between test types nd FPRs. Multivrite ROC models such s the multivrite binorml ply n importnt role in ROC nlysis of clustered biomrkers. This rticle derived symptotic covrinces of the LS, GLM nd GEE estimtors for the multivrite ROC models with the presence of interction terms between biomrker type nd FPRs. To our knowledge, such symptotic properties with correlted dt hve not been ddressed in empiricl process theory. Our new contributions include procedures to compre AUCs nd ROC curves, especilly when the interction terms between FPR s nd biomrker type re present. We drew inference for pirwise comprison between ROC curves, multiple comprisons of res under ROC curves under binorml ssumptions. The inferentil procedures in Section 4 for compring ROC curves re very generl for multivrite ROC models. Besides the three estimtors discussed, if other estimtors nd their covrinces were vilble, they cn lso be used in these procedures. Acknowledgement The uthors would like to thnk n ssocited editor nd two referees for their constructive comments nd suggestions. This work is supported in prt by NIH grnt R01EB Xio-Hu Zhou, Ph.D., is presently Core Investigtor nd Biosttistics Unit Director t the Northwest HSR&D Center of Excellence, Deprtment of Veterns Affirs Medicl Center, Settle, WA. The views expressed in this rticle re those of the uthors nd do not necessrily represent the views of the Deprtment of Veterns Affirs. References Ci, T. nd Moskowitz, C. 2004). Semiprmetric estimtion of the binorml ROC curve. Biosttistics 5, Ci, T. nd Pepe, M. S. 2002). Semi-prmetric ROC nlysis to evlute biomrkers for disese. J. Amer. Sttist. Assoc. 97, DeLong, E. R., DeLong, D. M. nd Clrke-Person, D. L. 1988). Compring the res under two or more correlted receiver operting chrcteristic curves: nonprmetric pproch. Biometrics 44, Kuermnn, G. nd Crroll, R. J. 2001). A note on the efficiency of sndwich covrince mtrix estimtion. J. Amer. Sttist. Assoc. 96, Metz, C. E., Wng, P. nd Kronmn, H. B. 1984). A new pproch for testing the significnce of differences between ROC curves mesured from correlted dt. In Informtion Processing in Medicl Imging Edited by F. Deconinck),
19 SEMIPARAMETRIC INFERENTIAL ROC PROCEDURES 1221 Metz, C. E., Hermn, B. A. nd Shen, J-H. 1998). Mximum-likelihood estimtion of ROC curves from continuously-distributed dt. Sttist. Medicine 17, Pepe, M. S. 2000). An interprettion for the ROC curve nd inference using GLM procedures. Biometrics 56, Ro, C. R. 2002). Liner Sttisticl Inference nd Its Appliction. John Wiley nd Sons: London. Vn der Vrt, A. nd Wellner, J. 1996). Wek Convergence nd Empiricl Processes: With Applictions to Sttistics. Springer-Verlg, New York. Wiend, S., Gil, M. H., Jmes, B. R. nd Jmes, K. L. 1989). A fmily of non-prmetric sttistics for compring dignostic mrkers with pired or unpired dt. Biometrik 76, Zhng, Z. 2004). Semiprmetric lest squres nlysis of the receiver operting chrcteristic curve. University of Wshington Doctorl Disserttion. Zhng, Z. nd Pepe, M. S. 2005). A liner regression frmework for receiver operting chrcteristic ROC) curve nlysis. Working Pper 253, UW Biosttistics Working Pper Series. Zhou, X. H., McClish, D. K. nd Obuchowski, N. A. 2002). Sttisticl Methods in Dignostic Medicine. Wiley, New York. Deprtment of Sttistics, The Volgenu School of Informtion Technology nd Engineering, George Mson University, Science & Tech II, Room 155, 4400 University Drive, MS 4A7, Firfx, VA , U.S.A. E-mil: ltng1@gmu.edu Deprtment of Biosttistics, School of Public Helth nd Community Medicine, University of Wshington, U.S.A. E-mil: zhou@u.wshington.edu Received July 2007; ccepted Februry 2008)
Semiparametric Inferential Procedures for Comparing Multivariate ROC Curves with Interaction Terms
UW Biostatistics Working Paper Series 4-29-2008 Semiparametric Inferential Procedures for Comparing Multivariate ROC Curves with Interaction Terms Liansheng Tang Xiao-Hua Zhou Suggested Citation Tang,
More informationTesting categorized bivariate normality with two-stage. polychoric correlation estimates
Testing ctegorized bivrite normlity with two-stge polychoric correltion estimtes Albert Mydeu-Olivres Dept. of Psychology University of Brcelon Address correspondence to: Albert Mydeu-Olivres. Fculty of
More informationA NOTE ON ESTIMATION OF THE GLOBAL INTENSITY OF A CYCLIC POISSON PROCESS IN THE PRESENCE OF LINEAR TREND
A NOTE ON ESTIMATION OF THE GLOBAL INTENSITY OF A CYCLIC POISSON PROCESS IN THE PRESENCE OF LINEAR TREND I WAYAN MANGKU Deprtment of Mthemtics, Fculty of Mthemtics nd Nturl Sciences, Bogor Agriculturl
More informationNew Expansion and Infinite Series
Interntionl Mthemticl Forum, Vol. 9, 204, no. 22, 06-073 HIKARI Ltd, www.m-hikri.com http://dx.doi.org/0.2988/imf.204.4502 New Expnsion nd Infinite Series Diyun Zhng College of Computer Nnjing University
More informationSemiparametric Least Squares Based Estimation of the Receiver Operating Characteristic(ROC) Curve. Zheng Zhang Emory University Atlanta, GA 30322
Semiprmetric Lest Squres Bsed Estimtion of the Receiver Operting Chrcteristic(ROC) Curve Zheng Zhng Emory University Atlnt, GA 30322 Mrgret Sullivn Pepe University of Wshington Fred Hutchinson Cncer Reserch
More informationStudent Activity 3: Single Factor ANOVA
MATH 40 Student Activity 3: Single Fctor ANOVA Some Bsic Concepts In designed experiment, two or more tretments, or combintions of tretments, is pplied to experimentl units The number of tretments, whether
More informationTests for the Ratio of Two Poisson Rates
Chpter 437 Tests for the Rtio of Two Poisson Rtes Introduction The Poisson probbility lw gives the probbility distribution of the number of events occurring in specified intervl of time or spce. The Poisson
More informationLecture 21: Order statistics
Lecture : Order sttistics Suppose we hve N mesurements of sclr, x i =, N Tke ll mesurements nd sort them into scending order x x x 3 x N Define the mesured running integrl S N (x) = 0 for x < x = i/n for
More informationNon-Linear & Logistic Regression
Non-Liner & Logistic Regression If the sttistics re boring, then you've got the wrong numbers. Edwrd R. Tufte (Sttistics Professor, Yle University) Regression Anlyses When do we use these? PART 1: find
More informationEstimation on Monotone Partial Functional Linear Regression
A^VÇÚO 1 33 ò 1 4 Ï 217 c 8 Chinese Journl of Applied Probbility nd Sttistics Aug., 217, Vol. 33, No. 4, pp. 433-44 doi: 1.3969/j.issn.11-4268.217.4.8 Estimtion on Monotone Prtil Functionl Liner Regression
More information8 Laplace s Method and Local Limit Theorems
8 Lplce s Method nd Locl Limit Theorems 8. Fourier Anlysis in Higher DImensions Most of the theorems of Fourier nlysis tht we hve proved hve nturl generliztions to higher dimensions, nd these cn be proved
More information7.2 The Definite Integral
7.2 The Definite Integrl the definite integrl In the previous section, it ws found tht if function f is continuous nd nonnegtive, then the re under the grph of f on [, b] is given by F (b) F (), where
More informationThe steps of the hypothesis test
ttisticl Methods I (EXT 7005) Pge 78 Mosquito species Time of dy A B C Mid morning 0.0088 5.4900 5.5000 Mid Afternoon.3400 0.0300 0.8700 Dusk 0.600 5.400 3.000 The Chi squre test sttistic is the sum of
More informationNUMERICAL INTEGRATION. The inverse process to differentiation in calculus is integration. Mathematically, integration is represented by.
NUMERICAL INTEGRATION 1 Introduction The inverse process to differentition in clculus is integrtion. Mthemticlly, integrtion is represented by f(x) dx which stnds for the integrl of the function f(x) with
More informationThe Regulated and Riemann Integrals
Chpter 1 The Regulted nd Riemnn Integrls 1.1 Introduction We will consider severl different pproches to defining the definite integrl f(x) dx of function f(x). These definitions will ll ssign the sme vlue
More informationA REVIEW OF CALCULUS CONCEPTS FOR JDEP 384H. Thomas Shores Department of Mathematics University of Nebraska Spring 2007
A REVIEW OF CALCULUS CONCEPTS FOR JDEP 384H Thoms Shores Deprtment of Mthemtics University of Nebrsk Spring 2007 Contents Rtes of Chnge nd Derivtives 1 Dierentils 4 Are nd Integrls 5 Multivrite Clculus
More informationChapter 4 Contravariance, Covariance, and Spacetime Diagrams
Chpter 4 Contrvrince, Covrince, nd Spcetime Digrms 4. The Components of Vector in Skewed Coordintes We hve seen in Chpter 3; figure 3.9, tht in order to show inertil motion tht is consistent with the Lorentz
More informationChapter 9: Inferences based on Two samples: Confidence intervals and tests of hypotheses
Chpter 9: Inferences bsed on Two smples: Confidence intervls nd tests of hypotheses 9.1 The trget prmeter : difference between two popultion mens : difference between two popultion proportions : rtio of
More information1 Online Learning and Regret Minimization
2.997 Decision-Mking in Lrge-Scle Systems My 10 MIT, Spring 2004 Hndout #29 Lecture Note 24 1 Online Lerning nd Regret Minimiztion In this lecture, we consider the problem of sequentil decision mking in
More informationEstimation of Binomial Distribution in the Light of Future Data
British Journl of Mthemtics & Computer Science 102: 1-7, 2015, Article no.bjmcs.19191 ISSN: 2231-0851 SCIENCEDOMAIN interntionl www.sciencedomin.org Estimtion of Binomil Distribution in the Light of Future
More informationTESTING CATEGORIZED BIVARIATE NORMALITY WITH TWO-STAGE POLYCHORIC CORRELATION ESTIMATES
IE Working Pper WP 0 / 03 08/05/003 TESTING CATEGORIZED BIVARIATE NORMALITY WITH TWO-STAGE POLYCHORIC CORRELATION ESTIMATES Alberto Mydeu Olivres Instituto de Empres Mrketing Dept. C/ Mri de Molin, -5
More information2008 Mathematical Methods (CAS) GA 3: Examination 2
Mthemticl Methods (CAS) GA : Exmintion GENERAL COMMENTS There were 406 students who st the Mthemticl Methods (CAS) exmintion in. Mrks rnged from to 79 out of possible score of 80. Student responses showed
More informationTESTING CATEGORIZED BIVARIATE NORMALITY WITH TWO-STAGE POLYCHORIC CORRELATION ESTIMATES. IE Working Paper MK8-106-I 08 / 05 / 2003
TESTING CATEGORIZED BIVARIATE NORMALITY WITH TWO-STAGE POLYCHORIC CORRELATION ESTIMATES IE Working Pper MK8-06-I 08 / 05 / 2003 Alberto Mydeu Olivres Instituto de Empres Mrketing Dept. C / Mrí de Molin
More informationMultivariate problems and matrix algebra
University of Ferrr Stefno Bonnini Multivrite problems nd mtrix lgebr Multivrite problems Multivrite sttisticl nlysis dels with dt contining observtions on two or more chrcteristics (vribles) ech mesured
More informationApplicable Analysis and Discrete Mathematics available online at
Applicble Anlysis nd Discrete Mthemtics vilble online t http://pefmth.etf.rs Appl. Anl. Discrete Mth. 4 (2010), 23 31. doi:10.2298/aadm100201012k NUMERICAL ANALYSIS MEETS NUMBER THEORY: USING ROOTFINDING
More informationFor the percentage of full time students at RCC the symbols would be:
Mth 17/171 Chpter 7- ypothesis Testing with One Smple This chpter is s simple s the previous one, except it is more interesting In this chpter we will test clims concerning the sme prmeters tht we worked
More informationThe First Fundamental Theorem of Calculus. If f(x) is continuous on [a, b] and F (x) is any antiderivative. f(x) dx = F (b) F (a).
The Fundmentl Theorems of Clculus Mth 4, Section 0, Spring 009 We now know enough bout definite integrls to give precise formultions of the Fundmentl Theorems of Clculus. We will lso look t some bsic emples
More informationMonte Carlo method in solving numerical integration and differential equation
Monte Crlo method in solving numericl integrtion nd differentil eqution Ye Jin Chemistry Deprtment Duke University yj66@duke.edu Abstrct: Monte Crlo method is commonly used in rel physics problem. The
More informationLecture INF4350 October 12008
Biosttistics ti ti Lecture INF4350 October 12008 Anj Bråthen Kristoffersen Biomedicl Reserch Group Deprtment of informtics, UiO Gol Presenttion of dt descriptive tbles nd grphs Sensitivity, specificity,
More informationDecision Science Letters
Decision Science Letters 8 (09) 37 3 Contents lists vilble t GrowingScience Decision Science Letters homepge: www.growingscience.com/dsl The negtive binomil-weighted Lindley distribution Sunthree Denthet
More information1.9 C 2 inner variations
46 CHAPTER 1. INDIRECT METHODS 1.9 C 2 inner vritions So fr, we hve restricted ttention to liner vritions. These re vritions of the form vx; ǫ = ux + ǫφx where φ is in some liner perturbtion clss P, for
More informationMath 1B, lecture 4: Error bounds for numerical methods
Mth B, lecture 4: Error bounds for numericl methods Nthn Pflueger 4 September 0 Introduction The five numericl methods descried in the previous lecture ll operte by the sme principle: they pproximte the
More informationPhysics 116C Solution of inhomogeneous ordinary differential equations using Green s functions
Physics 6C Solution of inhomogeneous ordinry differentil equtions using Green s functions Peter Young November 5, 29 Homogeneous Equtions We hve studied, especilly in long HW problem, second order liner
More informationAcceptance Sampling by Attributes
Introduction Acceptnce Smpling by Attributes Acceptnce smpling is concerned with inspection nd decision mking regrding products. Three spects of smpling re importnt: o Involves rndom smpling of n entire
More informationTHE EXISTENCE OF NEGATIVE MCMENTS OF CONTINUOUS DISTRIBUTIONS WALTER W. PIEGORSCH AND GEORGE CASELLA. Biometrics Unit, Cornell University, Ithaca, NY
THE EXISTENCE OF NEGATIVE MCMENTS OF CONTINUOUS DISTRIBUTIONS WALTER W. PIEGORSCH AND GEORGE CASELLA. Biometrics Unit, Cornell University, Ithc, NY 14853 BU-771-M * December 1982 Abstrct The question of
More informationCredibility Hypothesis Testing of Fuzzy Triangular Distributions
666663 Journl of Uncertin Systems Vol.9, No., pp.6-74, 5 Online t: www.jus.org.uk Credibility Hypothesis Testing of Fuzzy Tringulr Distributions S. Smpth, B. Rmy Received April 3; Revised 4 April 4 Abstrct
More informationTHE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS.
THE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS RADON ROSBOROUGH https://intuitiveexplntionscom/picrd-lindelof-theorem/ This document is proof of the existence-uniqueness theorem
More informationThe Riemann-Lebesgue Lemma
Physics 215 Winter 218 The Riemnn-Lebesgue Lemm The Riemnn Lebesgue Lemm is one of the most importnt results of Fourier nlysis nd symptotic nlysis. It hs mny physics pplictions, especilly in studies of
More informationODE: Existence and Uniqueness of a Solution
Mth 22 Fll 213 Jerry Kzdn ODE: Existence nd Uniqueness of Solution The Fundmentl Theorem of Clculus tells us how to solve the ordinry differentil eqution (ODE) du = f(t) dt with initil condition u() =
More informationP 3 (x) = f(0) + f (0)x + f (0) 2. x 2 + f (0) . In the problem set, you are asked to show, in general, the n th order term is a n = f (n) (0)
1 Tylor polynomils In Section 3.5, we discussed how to pproximte function f(x) round point in terms of its first derivtive f (x) evluted t, tht is using the liner pproximtion f() + f ()(x ). We clled this
More informationUnit #9 : Definite Integral Properties; Fundamental Theorem of Calculus
Unit #9 : Definite Integrl Properties; Fundmentl Theorem of Clculus Gols: Identify properties of definite integrls Define odd nd even functions, nd reltionship to integrl vlues Introduce the Fundmentl
More informationMath 426: Probability Final Exam Practice
Mth 46: Probbility Finl Exm Prctice. Computtionl problems 4. Let T k (n) denote the number of prtitions of the set {,..., n} into k nonempty subsets, where k n. Argue tht T k (n) kt k (n ) + T k (n ) by
More informationCzechoslovak Mathematical Journal, 55 (130) (2005), , Abbotsford. 1. Introduction
Czechoslovk Mthemticl Journl, 55 (130) (2005), 933 940 ESTIMATES OF THE REMAINDER IN TAYLOR S THEOREM USING THE HENSTOCK-KURZWEIL INTEGRAL, Abbotsford (Received Jnury 22, 2003) Abstrct. When rel-vlued
More informationNOTE ON TRACES OF MATRIX PRODUCTS INVOLVING INVERSES OF POSITIVE DEFINITE ONES
Journl of pplied themtics nd Computtionl echnics 208, 7(), 29-36.mcm.pcz.pl p-issn 2299-9965 DOI: 0.752/jmcm.208..03 e-issn 2353-0588 NOE ON RCES OF RIX PRODUCS INVOLVING INVERSES OF POSIIVE DEFINIE ONES
More informationReview of Calculus, cont d
Jim Lmbers MAT 460 Fll Semester 2009-10 Lecture 3 Notes These notes correspond to Section 1.1 in the text. Review of Clculus, cont d Riemnn Sums nd the Definite Integrl There re mny cses in which some
More informationLecture 14: Quadrature
Lecture 14: Qudrture This lecture is concerned with the evlution of integrls fx)dx 1) over finite intervl [, b] The integrnd fx) is ssumed to be rel-vlues nd smooth The pproximtion of n integrl by numericl
More informationMIXED MODELS (Sections ) I) In the unrestricted model, interactions are treated as in the random effects model:
1 2 MIXED MODELS (Sections 17.7 17.8) Exmple: Suppose tht in the fiber breking strength exmple, the four mchines used were the only ones of interest, but the interest ws over wide rnge of opertors, nd
More informationDiscrete Mathematics and Probability Theory Spring 2013 Anant Sahai Lecture 17
EECS 70 Discrete Mthemtics nd Proility Theory Spring 2013 Annt Shi Lecture 17 I.I.D. Rndom Vriles Estimting the is of coin Question: We wnt to estimte the proportion p of Democrts in the US popultion,
More informationSection 11.5 Estimation of difference of two proportions
ection.5 Estimtion of difference of two proportions As seen in estimtion of difference of two mens for nonnorml popultion bsed on lrge smple sizes, one cn use CLT in the pproximtion of the distribution
More informationRecitation 3: More Applications of the Derivative
Mth 1c TA: Pdric Brtlett Recittion 3: More Applictions of the Derivtive Week 3 Cltech 2012 1 Rndom Question Question 1 A grph consists of the following: A set V of vertices. A set E of edges where ech
More informationChapter 5 : Continuous Random Variables
STAT/MATH 395 A - PROBABILITY II UW Winter Qurter 216 Néhémy Lim Chpter 5 : Continuous Rndom Vribles Nottions. N {, 1, 2,...}, set of nturl numbers (i.e. ll nonnegtive integers); N {1, 2,...}, set of ll
More informationComparison Procedures
Comprison Procedures Single Fctor, Between-Subects Cse /8/ Comprison Procedures, One-Fctor ANOVA, Between Subects Two Comprison Strtegies post hoc (fter-the-fct) pproch You re interested in discovering
More informationQuantum Physics II (8.05) Fall 2013 Assignment 2
Quntum Physics II (8.05) Fll 2013 Assignment 2 Msschusetts Institute of Technology Physics Deprtment Due Fridy September 20, 2013 September 13, 2013 3:00 pm Suggested Reding Continued from lst week: 1.
More informationLecture 3. In this lecture, we will discuss algorithms for solving systems of linear equations.
Lecture 3 3 Solving liner equtions In this lecture we will discuss lgorithms for solving systems of liner equtions Multiplictive identity Let us restrict ourselves to considering squre mtrices since one
More informationDiscrete Mathematics and Probability Theory Summer 2014 James Cook Note 17
CS 70 Discrete Mthemtics nd Proility Theory Summer 2014 Jmes Cook Note 17 I.I.D. Rndom Vriles Estimting the is of coin Question: We wnt to estimte the proportion p of Democrts in the US popultion, y tking
More information1B40 Practical Skills
B40 Prcticl Skills Comining uncertinties from severl quntities error propgtion We usully encounter situtions where the result of n experiment is given in terms of two (or more) quntities. We then need
More informationQUADRATURE is an old-fashioned word that refers to
World Acdemy of Science Engineering nd Technology Interntionl Journl of Mthemticl nd Computtionl Sciences Vol:5 No:7 011 A New Qudrture Rule Derived from Spline Interpoltion with Error Anlysis Hdi Tghvfrd
More informationFrobenius numbers of generalized Fibonacci semigroups
Frobenius numbers of generlized Fiboncci semigroups Gretchen L. Mtthews 1 Deprtment of Mthemticl Sciences, Clemson University, Clemson, SC 29634-0975, USA gmtthe@clemson.edu Received:, Accepted:, Published:
More informationGoals: Determine how to calculate the area described by a function. Define the definite integral. Explore the relationship between the definite
Unit #8 : The Integrl Gols: Determine how to clculte the re described by function. Define the definite integrl. Eplore the reltionship between the definite integrl nd re. Eplore wys to estimte the definite
More informationContinuous Random Variables
STAT/MATH 395 A - PROBABILITY II UW Winter Qurter 217 Néhémy Lim Continuous Rndom Vribles Nottion. The indictor function of set S is rel-vlued function defined by : { 1 if x S 1 S (x) if x S Suppose tht
More informationA BRIEF INTRODUCTION TO UNIFORM CONVERGENCE. In the study of Fourier series, several questions arise naturally, such as: c n e int
A BRIEF INTRODUCTION TO UNIFORM CONVERGENCE HANS RINGSTRÖM. Questions nd exmples In the study of Fourier series, severl questions rise nturlly, such s: () (2) re there conditions on c n, n Z, which ensure
More informationResearch Article Moment Inequalities and Complete Moment Convergence
Hindwi Publishing Corportion Journl of Inequlities nd Applictions Volume 2009, Article ID 271265, 14 pges doi:10.1155/2009/271265 Reserch Article Moment Inequlities nd Complete Moment Convergence Soo Hk
More informationRiemann Sums and Riemann Integrals
Riemnn Sums nd Riemnn Integrls Jmes K. Peterson Deprtment of Biologicl Sciences nd Deprtment of Mthemticl Sciences Clemson University August 26, 2013 Outline 1 Riemnn Sums 2 Riemnn Integrls 3 Properties
More informationDesign and Analysis of Single-Factor Experiments: The Analysis of Variance
13 CHAPTER OUTLINE Design nd Anlysis of Single-Fctor Experiments: The Anlysis of Vrince 13-1 DESIGNING ENGINEERING EXPERIMENTS 13-2 THE COMPLETELY RANDOMIZED SINGLE-FACTOR EXPERIMENT 13-2.1 An Exmple 13-2.2
More informationA Matrix Algebra Primer
A Mtrix Algebr Primer Mtrices, Vectors nd Sclr Multipliction he mtrix, D, represents dt orgnized into rows nd columns where the rows represent one vrible, e.g. time, nd the columns represent second vrible,
More informationNumerical integration
2 Numericl integrtion This is pge i Printer: Opque this 2. Introduction Numericl integrtion is problem tht is prt of mny problems in the economics nd econometrics literture. The orgniztion of this chpter
More informationProperties of Integrals, Indefinite Integrals. Goals: Definition of the Definite Integral Integral Calculations using Antiderivatives
Block #6: Properties of Integrls, Indefinite Integrls Gols: Definition of the Definite Integrl Integrl Clcultions using Antiderivtives Properties of Integrls The Indefinite Integrl 1 Riemnn Sums - 1 Riemnn
More informationAbstract inner product spaces
WEEK 4 Abstrct inner product spces Definition An inner product spce is vector spce V over the rel field R equipped with rule for multiplying vectors, such tht the product of two vectors is sclr, nd the
More informationMATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1
MATH34032: Green s Functions, Integrl Equtions nd the Clculus of Vritions 1 Section 1 Function spces nd opertors Here we gives some brief detils nd definitions, prticulrly relting to opertors. For further
More informationRiemann Sums and Riemann Integrals
Riemnn Sums nd Riemnn Integrls Jmes K. Peterson Deprtment of Biologicl Sciences nd Deprtment of Mthemticl Sciences Clemson University August 26, 203 Outline Riemnn Sums Riemnn Integrls Properties Abstrct
More informationStein-Rule Estimation and Generalized Shrinkage Methods for Forecasting Using Many Predictors
Stein-Rule Estimtion nd Generlized Shrinkge Methods for Forecsting Using Mny Predictors Eric Hillebrnd CREATES Arhus University, Denmrk ehillebrnd@cretes.u.dk Te-Hwy Lee University of Cliforni, Riverside
More informationNotes on length and conformal metrics
Notes on length nd conforml metrics We recll how to mesure the Eucliden distnce of n rc in the plne. Let α : [, b] R 2 be smooth (C ) rc. Tht is α(t) (x(t), y(t)) where x(t) nd y(t) re smooth rel vlued
More informationMAA 4212 Improper Integrals
Notes by Dvid Groisser, Copyright c 1995; revised 2002, 2009, 2014 MAA 4212 Improper Integrls The Riemnn integrl, while perfectly well-defined, is too restrictive for mny purposes; there re functions which
More informationBest Approximation. Chapter The General Case
Chpter 4 Best Approximtion 4.1 The Generl Cse In the previous chpter, we hve seen how n interpolting polynomil cn be used s n pproximtion to given function. We now wnt to find the best pproximtion to given
More informationUNIFORM CONVERGENCE. Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3
UNIFORM CONVERGENCE Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3 Suppose f n : Ω R or f n : Ω C is sequence of rel or complex functions, nd f n f s n in some sense. Furthermore,
More informationN 0 completions on partial matrices
N 0 completions on prtil mtrices C. Jordán C. Mendes Arújo Jun R. Torregros Instituto de Mtemátic Multidisciplinr / Centro de Mtemátic Universidd Politécnic de Vlenci / Universidde do Minho Cmino de Ver
More informationLecture 1. Functional series. Pointwise and uniform convergence.
1 Introduction. Lecture 1. Functionl series. Pointwise nd uniform convergence. In this course we study mongst other things Fourier series. The Fourier series for periodic function f(x) with period 2π is
More informationChapter 0. What is the Lebesgue integral about?
Chpter 0. Wht is the Lebesgue integrl bout? The pln is to hve tutoril sheet ech week, most often on Fridy, (to be done during the clss) where you will try to get used to the ides introduced in the previous
More informationChapter 2 Fundamental Concepts
Chpter 2 Fundmentl Concepts This chpter describes the fundmentl concepts in the theory of time series models In prticulr we introduce the concepts of stochstic process, men nd covrince function, sttionry
More informationApplication of Exp-Function Method to. a Huxley Equation with Variable Coefficient *
Interntionl Mthemticl Forum, 4, 9, no., 7-3 Appliction of Exp-Function Method to Huxley Eqution with Vrible Coefficient * Li Yo, Lin Wng nd Xin-Wei Zhou. Deprtment of Mthemtics, Kunming College Kunming,Yunnn,
More informationEuler, Ioachimescu and the trapezium rule. G.J.O. Jameson (Math. Gazette 96 (2012), )
Euler, Iochimescu nd the trpezium rule G.J.O. Jmeson (Mth. Gzette 96 (0), 36 4) The following results were estblished in recent Gzette rticle [, Theorems, 3, 4]. Given > 0 nd 0 < s
More informationW. We shall do so one by one, starting with I 1, and we shall do it greedily, trying
Vitli covers 1 Definition. A Vitli cover of set E R is set V of closed intervls with positive length so tht, for every δ > 0 nd every x E, there is some I V with λ(i ) < δ nd x I. 2 Lemm (Vitli covering)
More informationConvex Sets and Functions
B Convex Sets nd Functions Definition B1 Let L, +, ) be rel liner spce nd let C be subset of L The set C is convex if, for ll x,y C nd ll [, 1], we hve 1 )x+y C In other words, every point on the line
More informationLorenz Curve and Gini Coefficient in Right Truncated Pareto s Income Distribution
EUROPEAN ACADEMIC RESEARCH Vol. VI, Issue 2/ Mrch 29 ISSN 2286-4822 www.eucdemic.org Impct Fctor: 3.4546 (UIF) DRJI Vlue: 5.9 (B+) Lorenz Cure nd Gini Coefficient in Right Truncted Preto s Income Distribution
More informationSolution for Assignment 1 : Intro to Probability and Statistics, PAC learning
Solution for Assignment 1 : Intro to Probbility nd Sttistics, PAC lerning 10-701/15-781: Mchine Lerning (Fll 004) Due: Sept. 30th 004, Thursdy, Strt of clss Question 1. Bsic Probbility ( 18 pts) 1.1 (
More informationMath& 152 Section Integration by Parts
Mth& 5 Section 7. - Integrtion by Prts Integrtion by prts is rule tht trnsforms the integrl of the product of two functions into other (idelly simpler) integrls. Recll from Clculus I tht given two differentible
More informationFundamental Theorem of Calculus and Computations on Some Special Henstock-Kurzweil Integrals
Fundmentl Theorem of Clculus nd Computtions on Some Specil Henstock-Kurzweil Integrls Wei-Chi YANG wyng@rdford.edu Deprtment of Mthemtics nd Sttistics Rdford University Rdford, VA 24142 USA DING, Xiofeng
More informationg i fφdx dx = x i i=1 is a Hilbert space. We shall, henceforth, abuse notation and write g i f(x) = f
1. Appliction of functionl nlysis to PEs 1.1. Introduction. In this section we give little introduction to prtil differentil equtions. In prticulr we consider the problem u(x) = f(x) x, u(x) = x (1) where
More informationASYMPTOTIC BEHAVIOR OF INTERMEDIATE POINTS IN CERTAIN MEAN VALUE THEOREMS. II
STUDIA UNIV. BABEŞ BOLYAI, MATHEMATICA, Volume LV, Number 3, September 2010 ASYMPTOTIC BEHAVIOR OF INTERMEDIATE POINTS IN CERTAIN MEAN VALUE THEOREMS. II TIBERIU TRIF Dedicted to Professor Grigore Ştefn
More informationPredict Global Earth Temperature using Linier Regression
Predict Globl Erth Temperture using Linier Regression Edwin Swndi Sijbt (23516012) Progrm Studi Mgister Informtik Sekolh Teknik Elektro dn Informtik ITB Jl. Gnesh 10 Bndung 40132, Indonesi 23516012@std.stei.itb.c.id
More informationThe Shortest Confidence Interval for the Mean of a Normal Distribution
Interntionl Journl of Sttistics nd Proility; Vol. 7, No. 2; Mrch 208 ISSN 927-7032 E-ISSN 927-7040 Pulished y Cndin Center of Science nd Eduction The Shortest Confidence Intervl for the Men of Norml Distriution
More informationNUMERICAL ANALYSIS MEETS NUMBER THEORY: USINGROOTFINDINGMETHODSTOCALCULATE INVERSES MOD p n
Applicble Anlysis nd Discrete Mthemtics vilble online t http://pefmth.etf.bg.c.yu Appl. Anl. Discrete Mth. x (xxxx), xxx xxx. doi:10.2298/aadmxxxxxxxx NUMERICAL ANALYSIS MEETS NUMBER THEORY: USINGROOTFINDINGMETHODSTOCALCULATE
More information5.7 Improper Integrals
458 pplictions of definite integrls 5.7 Improper Integrls In Section 5.4, we computed the work required to lift pylod of mss m from the surfce of moon of mss nd rdius R to height H bove the surfce of the
More informationNormal Distribution. Lecture 6: More Binomial Distribution. Properties of the Unit Normal Distribution. Unit Normal Distribution
Norml Distribution Lecture 6: More Binomil Distribution If X is rndom vrible with norml distribution with men µ nd vrince σ 2, X N (µ, σ 2, then P(X = x = f (x = 1 e 1 (x µ 2 2 σ 2 σ Sttistics 104 Colin
More informationJournal of Inequalities in Pure and Applied Mathematics
Journl of Inequlities in Pure nd Applied Mthemtics MOMENTS INEQUALITIES OF A RANDOM VARIABLE DEFINED OVER A FINITE INTERVAL PRANESH KUMAR Deprtment of Mthemtics & Computer Science University of Northern
More informationImproved Results on Stability of Time-delay Systems using Wirtinger-based Inequality
Preprints of the 9th World Congress he Interntionl Federtion of Automtic Control Improved Results on Stbility of ime-dely Systems using Wirtinger-bsed Inequlity e H. Lee Ju H. Prk H.Y. Jung O.M. Kwon S.M.
More informationProbability Distributions for Gradient Directions in Uncertain 3D Scalar Fields
Technicl Report 7.8. Technische Universität München Probbility Distributions for Grdient Directions in Uncertin 3D Sclr Fields Tobis Pfffelmoser, Mihel Mihi, nd Rüdiger Westermnn Computer Grphics nd Visuliztion
More informationSOME NEW INSIGHTS ABOUT THE ACCELERATED FAILURE TIME MODEL
SOME NEW INSIGHTS ABOUT THE ACCELERATED FAILURE TIME MODEL by Ying Ding A disserttion submitted in prtil fulfillment of the requirements for the degree of Doctor of Philosophy (Biosttistics) in The University
More informationESTIMATION OF AGE AT DEATH DISTRIBUTION FOR A SPECIFIC CAUSE
ESTIMATION OF AGE AT DEATH DISTRIBUTION FOR A SPECIFIC CAUSE by Regin C. Elndt~Johnson Deprtment of Biosttistics University of North Crolin t Chpel Hill Institute of Sttistics Mimeo Series No, 1120 My
More informationData Assimilation. Alan O Neill Data Assimilation Research Centre University of Reading
Dt Assimiltion Aln O Neill Dt Assimiltion Reserch Centre University of Reding Contents Motivtion Univrite sclr dt ssimiltion Multivrite vector dt ssimiltion Optiml Interpoltion BLUE 3d-Vritionl Method
More information