Maximum likelihood estimation for multivariate skew normal mixture models
|
|
- Marlene Hudson
- 5 years ago
- Views:
Transcription
1 Journal of Multvarate Analyss Contents lsts avalable at ScenceDrect Journal of Multvarate Analyss ournal homepage: wwwelsevercom/locate/mva Maxmum lkelhood estmaton for multvarate skew normal mxture models Tsung I Ln Department of Appled Mathematcs Natonal Chung Hsng Unversty Tachung 402 Tawan a r t c l e n f o a b s t r a c t Artcle hstory: Receved 6 February 2006 Avalable onlne 25 Aprl 2008 AMS 1991 subect classfcatons: 62F10 62H10 62H12 Ths paper provdes a flexble mxture modelng framework usng the multvarate skew normal dstrbuton A feasble EM algorthm s developed for fndng the maxmum lkelhood estmates of parameters n ths context A general nformaton-based method for obtanng the asymptotc covarance matrx of the maxmum lkelhood estmators s also presented The proposed methodology s llustrated wth a real example and results are also compared wth those obtaned from fttng normal mxtures 2008 Elsever Inc All rghts reserved Keywords: EM algorthm Multvarate truncated normal dstrbutons Skew normal mxtures Stochastc representaton 1 Introducton A fnte mxture of dstrbutons n partcular the use of normal components has receved consderable attenton and s known to be a very powerful tool for modelng an extremely wde varety of random phenomena Most mportantly mxture modelng has been a favorable model-based technque n handlng supervsed classfcaton and unsupervsed clusterng problems There are a number of farly comprehensve monographs n ths area; see for example [ ] and the references contaned theren It s well known that there stll exst several problems n statstcal modelng of normal mxture NORMIX models For nstance normalty assumptons for component denstes could be volated when a set of data contans asymmetrc outcomes for each component Moreover the classcal normal mxture model tends to overft the data snce they need to nclude addtonal components to capture possbly excess skewness To overcome aforementoned weaknesses n the fttng of normal mxtures Ln et al [17] recently proposed a skew normal mxture SNMIX model usng the unvarate skew normal SN dstrbuton Azzaln [3] and showed ts great flexblty n modelng data wth asymmetrc behavors To allow for modelng real data as approprately as possble and to remedy unrealstc assumptons n classcal normalbased models a poneerng work on multvarate SN dstrbutons was frst studed by Azzaln and Dalla Valle [5] and subsequently generalzed by Gupta et al [14] and Arellano-Valle and Genton [2] among others Sahu et al [22] developed a more general class of dstrbutons by ntroducng skewness n multvarate ellptcally symmetrc dstrbutons They ponted out that the multvarate SN dstrbuton n ths famly s more flexble n adustng the correlaton structure than that proposed by Azzaln and Dalla Valle [5] The man obectve of ths work s to ntroduce a novel mxture modelng usng a new class of multvarate SN dstrbutons proposed by Sahu et al [22] For computatonal aspects I develop an effectve teratve procedure for obtanng maxmum Correspondng author E-mal address: tln@amathnchuedutw X/$ see front matter 2008 Elsever Inc All rghts reserved do:101016/mva
2 258 TI Ln / Journal of Multvarate Analyss lkelhood ML estmates of model parameters va the EM algorthm [7] Moreover I provde a smple way of obtanng standard errors of estmates by nvertng the observed nformaton matrx nstead of performng the computatonally ntensve bootstrap method As an llustraton I apply the proposed method on a real data set and show the advantage of usng SNMIX models Some concludng remarks are gven at the end and techncal dervatons are collected n Appendx 2 Prelmnares I start by formulatng some dstrbutonal propertes of the multvarate skew normal dstrbuton that was ntroduced by Sahu et al [22] Besdes I gve a stochastc representaton whch s useful for the constructon of complete data framework I next revew the multvarate truncated normal dstrbuton where truncaton s at arbtrary ponts and provde general formulae for computng the correspondng frst two moments These analytcal results are useful for the proposed EM algorthm 21 The multvarate skew normal dstrbuton A random vector Y s sad to follow a p-dmensonal skew normal dstrbuton wth a p 1 locaton vector ξ a p p postve defnte scale covarance matrx Σ and a p p skewness matrx Λ f ts densty functon s fy ξ Σ Λ 2 p φ p y ξ ΩΦ p Λ T Ω y ξ 1 wth Ω Σ + ΛΛ T and I p + Λ T Σ Λ I p Λ T Ω Λ where I p s a p p dentty matrx Moreover φ p µ Σ and Φ p Σ denote the probablty densty functon pdf of N p µ Σ and cumulatve densty functon cdf of N p 0 Σ respectvely I denote ths dstrbuton by Y SN p ξ Σ Λ hereafter and note that 1 belongs to the famly of skew-ellptcal dstrbutons as defned n [22] Typcally f Λ s assumed to be dagonal then the covarance structure of Y s not affected by the ntroducton of skewness Let Φ p denote the cdf of N p 0 I p The moment generatng functon of Y s M Y t 2 p exp t T ξ + 1 } 2 tt Ωt Φ p Λ T t t t 1 t p T R p 2 Expressng Λ T t p k1 λ k1t k p k1 λ kpt k T straghtforward calculatons gve the followng results and Φ p Λ t T t t0 2 p 2 π λ r r1 2 Φ p Λ T 2 } t 2 p λ t t r λ r λ r λ r 4 t0 π r1 r1 r1 Takng the frst two dervatves of 2 and applyng 3 and 4 the mean vector and the covarance matrx can be wrtten as 2 EY ξ + π Λ1 p covy Σ ΛΛ T π where 1 p s a p-dmensonal vector of ones Assumng Z N p 0 I p t follows that Z s dstrbuted as a p-dmensonal standard half-normal dstrbuton denoted by HN p 0 I p By Proposton 1 of Arellano-Valle et al [1] t turns out that 1 has a convenent stochastc representaton Y ξ + Λτ + U where τ and U are ndependently dstrbuted as HN p 0 I p and N p 0 Σ respectvely Note that the expresson 5 provdes a useful tool for random number generaton and for theoretcal purposes Moreover t s easy to see Y τ N p ξ + Λτ Σ Hence the densty of Y n 1 can be obtaned by usng the convoluton of denstes of Y τ and τ and Lemma 21 of [2] The multvarate truncated normal dstrbuton Let TN p µ Σ; A denote a p-varate truncated normal dstrbuton for N p µ Σ lyng wthn a truncated hyperplane regon A x x 1 x p T x 1 > a 1 x p > a p } and use the notaton p 1 a p for the abbrevaton of multple ntegrals Talls [23] has provded the formulae for the frst two moments of a multvarate truncated normal dstrbuton TN p 0 R; A where R s a correlaton matrx Under ths truncaton type I shall generalze Talls results to provde explct formulae for computng the frst and second moments of general multvarate truncated normal dstrbutons a a 1
3 TI Ln / Journal of Multvarate Analyss Consder a random vector X X 1 X p T whch has a p-varate truncated normal densty gven by fx µ Σ; A 1 α φ px µ ΣI A x 6 where α p 1 a φ p x µ Σdx wth a s are arbtrary real numbers and I A x s the ndcator functon whose value equals one f x A and zero elsewhere I shall use the notaton X TN p µ Σ; A to denote that X has densty 6 The moment generatng functon of X s of X M X t α exp t T µ + 1 } p 2 tt Σt 1 a φ p x µ + Σt Σdx 7 Dfferentatng 7 wth respect to t then evaluatng the dervatve wth t 0 one readly obtans the margnal mean EX µ + α σ r f r a r G r r1 where σ denotes the th entry of Σ f r a r φa r µ r σ rr denotes a normal densty wth mean µ r and varance σ rr for the rth varable evaluated at a r and G r r a φ p x r µ r 2 1 Σ r 22 1 dx r wth φ p x r µ r 2 1 Σ r 22 1 beng the condtonal densty of the remanng p 1 varables gven X r a r Smlarly we can verfy that σ EX X µ EX + µ EX µ µ + σ + α r σ r a r µ r f r a r G r r1 σ rr + σ r σ s σ } rsσ r f rs a r a s G rs 9 r1 s r σ rr where f rs a r a s s a bvarate normal densty of the r sth varables of N p µ Σ evaluated at a r a s and G rs rs a φ p 2 x rs µ rs 2 1 Σ rs 22 1 dx rs wth φ p 2 x rs µ rs 2 1 Σ rs 22 1 beng the condtonal densty of the remanng p 2 varables gven X r a r and X s a s Throughout ths paper I wll use the followng notatons: [A] rs denotes the r sth entry of a gven matrx A; Dag denotes a dagonal matrx created by extractng the man dagonal elements of a square matrx or the dagonalzaton of a vector; and dag denotes a vector contanng the dagonal elements of a square matrx After some algebrac manpulatons expressons 8 and 9 can be wrtten n matrx notatons as follows: 8 EX η µ + α Σq 10 where q q 1 q p T s a p 1 vector whose rth element s f r a r G r and EXX T µη T + ηµ T µµ T + Σ + α ΣH + DΣ 11 where H s a p p matrx wth all dagonal entres beng zero and f rs a r a s G rs on the r sth off-dagonal entry and D s a p p dagonal matrx whose rth dagonal entry s σ rr ar µ r f r a r G r [ΣH] rr Note that the computaton of truncated moments s hghly depended on the numercal method for Φ p whch can be swftly evaluated by the fast algorthms descrbed n Genz [1213] The procedures proposed n Genz s papers can be mplemented by usng the package mvtnorm avalable at A computer code for the computaton of the mean and covarance matrx of multvarate truncated normal dstrbutons wrtten n R language s avalable from the author upon request 3 ML estmaton for multvarate skew normal mxtures I consder the ML estmaton for a g-component mxture model n whch a set of random sample Y 1 Y n follows a mxture of multvarate skew normal dstrbutons Its probablty densty functon can be wrtten as Y w fy ξ Σ Λ w 0 1 w where Θ θ 1 θ g wth θ w ξ Σ Λ beng the unknown parameters of component and w s beng the mxng probabltes The ML estmates ˆΘ based on a set of ndependent observatons y y T 1 y T n T s ˆΘ argmax lθ y Θ
4 260 TI Ln / Journal of Multvarate Analyss where lθ y log w fy ξ Σ Λ 1 13 s the observed log-lkelhood functon Generally there s no explct analytcal soluton of ˆΘ but t can be acheved teratvely by usng the EM algorthm under the complete data framework 14 dscussed later In the context of herarchcal mxture modelng for each Y t s convenent to ntroduce a set of zero-one ndcator varables Z Z 1 Z g T for 1 n whch s a multnomal random vector wth 1 tral and cell probabltes w 1 w g denoted as Z M1; w 1 w g Note that the rth element Z r 1 f Y arses from component r From 5 wth the ncluson of ndcator varables Z s a herarchcal representaton of 12 s gven by Y τ Z 1 N p ξ + Λ τ Σ τ Z 1 HN p 0 I p Z M1; w 1 w g 1 n 14 By Bayes Theorem t can be shown that τ Y y Z 1 TN p Λ T Ω y ξ ; R p + 15 where Ω Σ + Λ Λ T I p + Λ T Σ Λ I p Λ T Ω Λ and R p + x x 1 x p T R p x > 0 1 p} In what follows denote Eτ y Z 1 η and Eτ τ T y Z 1 Ψ 16 where η and Ψ are both mplct functons of parameters ξ Σ Λ It s crucal to emphasze that evaluatons of 16 rely heavly on the results of 10 and 11 For notatonal smplcty let Z Z T 1 ZT n T and τ τ T 1 τ T n T From 14 the complete-data log-lkelhood functon of Θ gnorng addtve constants s l c Θ y Z τ 1 Z logw 1 2 log Σ 1 2 y ξ Λ τ T Σ y ξ Λ τ 1 } 2 τ T τ 17 I adopt the EM algorthm for fndng ML estmates Formally the E-step of the EM algorthm requres to calculate the socalled Q-functon QΘ ˆΘk E lc Θ y τ Z y ˆΘk whch s the condtonal expectaton of 17 gven observed data y and the current estmated parameters ˆΘk To calculate the Q-functon t can be observed from 17 that the condtonal expectaton of the term 1 2 Z τ Tτ can be omtted because t does not nclude any parameters thereby the necessary condtonal expectatons nvolved n the Q-functon are EZ y ˆΘk EZ τ y ˆΘk and EZ τ τ T y ˆΘk The mplementaton of the EM algorthm proceeds as follows: E-step: At the kth teraton compute and EZ y ˆΘk ŵ k m1 fy ˆξ k k ˆΣ ˆΛk ŵ k m fy ˆξ k ẑ k m ˆΣ m ˆΛk m k EZ τ y ˆΘk EZ y ˆΘk Eτ Z 1 y ˆΘk k ẑ ˆη k EZ τ τ T y ˆΘk EZ y ˆΘk Eτ τ T Z 1 y ˆΘk ẑ k ˆΨ k where ˆη k k and ˆΨ are η and Ψ n 16 wth ξ Σ and Λ replaced by ˆξ k k ˆΣ Therefore the Q-functon can be wrtten by QΘ ˆΘk 1 ẑ k logw log Σ y ξ Λ ˆη k 1 2 tr Σ 1 2 y ξ Λ ˆη k T Σ Λ ˆΨ k ˆη k and ˆΛk respectvely } ˆη kt Λ T 18 M-step: 1 Update ŵ k by ŵ k+1 n n ẑk
5 2 Update ˆξ k ˆξ k+1 3 Fx ξ ˆξ k+1 TI Ln / Journal of Multvarate Analyss by maxmzng 18 over ξ whch leads to / n ẑ k y ˆΛk z k ˆη k n z k ˆΛ k+1 update ˆΛk by maxmzng 18 over Λ whch leads to n ẑ k y ˆξ k+1 n ˆη kt ẑ k ˆΨ k 3 In the case wth Λ assumed to be dagonal say Λ Dagλ where λ s a p-dmensonal vector then update ˆλ k by ˆλ k+1 ˆΣ k ẑ k ˆΨ k ˆΣ k ẑ k ˆη k y ˆξ k+1 where the operator denotes the Hadamard elementwse product [15] of two matrces of the same dmenson 4 Fx ξ ˆξ k+1 and Λ ˆΛk+1 k update ˆΣ by maxmzng 18 over Σ whch leads to ˆΣ k+1 1 ẑ k + ˆΛk+1 n ẑ k n y ˆξ k+1 ẑ k ˆΨ k ˆη k ˆΛk+1 ˆη kt ˆη k y ˆξ k+1 } ˆΛk+1 T T ˆΛk+1 4 k In the case where the scale covarance matrces are homoscedastc say Σ 1 Σ g Σ then update ˆΣ by ˆΣ k+1 1 n ẑ k y ˆξ k+1 ˆΛk+1 ˆη k y ˆξ k+1 ˆΛk+1 ˆη k T n 1 n + ˆΛk+1 ẑ k k ˆΨ ˆη k } ˆη kt ˆΛk+1 T I further offer some remarks on the mplementaton of the proposed EM algorthm Remark 1 To montor the convergence by usng the lkelhood ncreasng property of the EM algorthm [725] a smple way s to repeat teratons after a certan number of teratons or untl the dfference between two successve log-lkelhood evaluatons s small enough Remark 2 As analogous to other teratve optmzaton procedures such as the Newton Raphson or Fsher scorng algorthms one needs to search for approprate ntal values to avod dvergence or tme-consumng computatons I offer a smple way of automatcally generatng a selecton of ntal values The technque proceeds as follows: Perform a K- means clusterng [11] ntalzed wth respect to a random start Specfy the zero-one component membershp ndcator Ẑ 0 Ẑ 0 } g 1 accordng to the the K-means clusterng results The ntal values of mxng probabltes component locatons and scale covarance matrces are then explctly chosen as ŵ 0 Ẑ 0 n ˆξ 0 Ẑ 0 y Ẑ 0 ˆΣ 0 Ẑ 0 y ˆξ 0 y ˆξ 0 0 Meanwhle f Σ s are assumed to be dentcal say Σ 1 Σ g Σ then ˆΣ s taken as the sample covarance of the whole sample Furthermore the ntal skewness matrces can be chosen as dagonal say ˆΛ0 Dag ˆλ 0 1 ˆλ 0 p } wth the value of each entry chosen slghtly devated from zero eg ˆλ 0 r s taken as 3 or 3 whose sgn s measured by the sgn of the sample skewness of the K-means clusterng observatons Ẑ 0 T 1 p ˆη k T Remark 3 The man dffculty n dealng wth mxture models s to fnd the global maxmzer of Θ for nstance the lkelhood functon LΘ y mght be unbounded n certan stuatons Another oft-voced crtcsm s that the EM-type procedure tends to get stuck n local modes One convenent way to crcumvent such lmtatons s to try several EM teratons
6 262 TI Ln / Journal of Multvarate Analyss under a varety of startng values If there exst several modes one can fnd the global mode by comparng ther log-lkelhood values In partcular the algorthm runnng wth dfferent startng values can also be used to assess the stablty of the resultng estmates Under certan boundedness condtons stated n Render and Walker [21] the ML estmate ˆΘ s consstent and converges n dstrbuton to a zero-mean normal random vector whose covarance matrx s the nverse Fsher nformaton matrx That s n ˆΘ Θ 0 d N q 0 I Θ 0 where q DmΘ Θ 0 s the true value of Θ and IΘ E 2 lθ y/ ΘΘ T s the Fsher nformaton matrx For a more detaled dscusson on the asymptotc theores of ML estmators for mxture models nterested readers are referred to [2021] 4 Provson of standard errors A smple way of obtanng the standard errors of ML estmates of mxture model parameters s to approxmate the asymptotc covarance matrx of ˆΘ by the nverse of the observed nformaton matrx see eg Basford et al [6] Let I o Θ y 2 lθ y/ Θ Θ T be the observed nformaton matrx where lθ y s the observed log-lkelhood functon as n 13 The estmated observed nformaton matrx can be reduced to I o ˆΘ y ŝ ŝ T 19 where ŝ E ˆΘ l c Θ y Z τ / Θ y wth l c Θ y Z τ beng the complete-data log-lkelhood formed from the sngle observaton y for 1 n Let vec be the matrx operator whch stacks all columns of a matrx nto a vector and vech the matrx operator whch arranges the supradagonal elements of a symmetrc matrx Let ŝ be a vector contanng ŝ ŝ w1 ŝ wg ŝ ξ 1 ŝ ξ g ŝ λ1 ŝ λg ŝ σ 1 ŝ σ g T where λ vecλ and σ vechσ Expressons for the elements of ŝ w ŝ ξ ŝ λ and ŝ σ are gven by where ŝ w ẑ ŵ ẑg ŵ g ŝ ξ ẑ ˆΣ y ˆξ ˆΛ ˆη ŝ λ vec ẑ ˆΣ y ξ ˆη T ˆΛ ˆΨ 1 ŝ σ vech 2 ẑ 2 Dag  ˆΣ y ˆξ ˆΛ ˆη y ˆξ ˆΛ ˆη T + ˆΛT ˆΨ ˆη ˆη T ˆΛ ˆΣ ˆΣ 20 ẑ ŵ fy ˆξ ˆΣ ˆΛ / g m1 ŵmfy ˆξ m ˆΣ m ˆΛm s the posteror probablty that the observaton y belongs to component and ˆη and ˆΨ are obtaned by substtutng the ML estmates ˆξ ˆΣ ˆΛ nto 16 If the skewness matrces are assumed to be dagonal e λ dagλ then ŝ λ ẑ ˆΣ ˆη y ˆξ T } 1 p ˆΣ ˆΨ ˆλ 21 Furthermore f one assumes that the scale covarance matrces are homoscedastc e σ 1 σ g σ then 1 ŝ σ vech ẑ 2 2 Dag 22 where 1  ˆΣ y ˆξ ˆΛ ˆη y ˆξ ˆΛ ˆη T + ˆΛT ˆΨ ˆη ˆη T ˆΛ ˆΣ ˆΣ The detaled proofs of are gven n Appendx The nformaton-based approxmaton 19 s asymptotcally applcable However t s less relable unless the sample sze s suffcently large Alternatvely t s common practce to perform the parametrc bootstrap approach Efron and Tbshran [8] to obtan more accurate standard error estmates whle t requres enormous amounts of computng power
7 TI Ln / Journal of Multvarate Analyss Table 1 ML estmates and the assocated standard errors for the ftted two-component SNMIX model for the bank data Parameter w ξ 11 ξ 12 σ 111 σ 112 σ 122 λ 111 λ 122 Estmate SE Parameter 1 w ξ 21 ξ 22 σ 211 σ 212 σ 222 λ 211 λ 222 Estmate SE Table 2 Model selecton crtera for the bank data Model m l ˆΘ y LRT P-value AIC BIC NORMIX SNMIX A practcal example As an llustraton I apply the methods descrbed n prevous sectons to the famous bank data set whch was orgnally reported Tables 11 and 12 n Flury and Redwyl [9] and subsequently analyzed by Ma and Genton [18] wth a flexble skewsymmetrc dstrbuton The data consst of sx measurements made on 100 genune and 100 counterfet old Swss 1000 franc blls In ths example the goal s to verfy the developed estmatng devce and assess the relatve performances of the ftted SNMIX and NORMIX models To smplfy the analyss attenton s focused on the sample of X 1 : the wdth of the rght edge and X 2 : the length of the mage dagonal Margnally each of the two varables exhbts a bmodal dstrbuton wth asymmetrc components I now carry out the EM procedure for fndng the parameter estmates of a two-component model 12 To avod the correlaton structure affected by the ncluson of skewness parameters emphaszed by Sahu et al [22] the skewness matrces Λ for 1 2 are chosen as dagonal More specfcally the model to be ftted can be wrtten as fy Θ wfy ξ 1 Σ 1 Λ wfy ξ 2 Σ 2 Λ where [ ] ξ ξ 1 ξ 2 T σ11 σ Σ 12 σ 12 σ 22 [ ] λ11 0 and Λ 0 λ To get several dfferent sets of startng values ths can be done frst by randomly generatng a set of B bootstrap resamplng samples y 1 y from the orgn data y then computng B ˆΘ0 for each bootstrap sample usng the method descrbed n Remark 2 The EM algorthm was run under B 30 dfferent sets of startng values and was termnated when an ncrease n the log-lkelhood s less than 10 4 For ths data set these EM roots computed under dfferent startng values converge to smlar statonary ponts wth the largest log-lkelhood The resultng ML estmates and the assocated standard errors are reported n Table 1 From the reported nformatonbased standard errors all the parameters are statstcally sgnfcant except for σ 212 and λ 222 The estmates of skewness parameters reveal the two varables are both sgnfcantly skewed to the left n component 1 Wth regard to component 2 only X 1 s sgnfcantly skewed to the rght For comparson purposes I also ft a NORMIX model whch can be treated as a reduced model of SNMIX wth parameters n skewness matrces specfed by zeros For testng the null hypothess H 0 : Λ 1 Λ 2 0 NORMIX versus the alternatve hypothess H 1 : at least Λ SNMIX the lkelhood rato test LRT statstc whch s a comparson of lkelhood scores between two compettve models s used to udge whch of the two models s more approprate for ths data set The LRT statstc for testng the exstence of skewness n component denstes gves a value 2418 whch s hghly sgnfcant compared to a χ 2 4 dstrbuton ndcatng that the null hypothess s not acceptable for the bank data Furthermore the fts of two models are also compared based on the Akake nformaton crteron AIC and Bayesan nformaton crteron BIC whch are defned as AIC 2l ˆΘ y m and BIC 2 l ˆΘ y 05 m logn } respectvely where l ˆΘ y s the maxmzed log-lkelhood m s the number of parameters and n s the sample sze The comparson results are lsted n Table 2 It s readly seen from the table that both AIC and BIC values as well as the LRT statstc consstently favor the SNMIX model The contours of the ML-ftted SNMIX and NORMIX denstes are depcted n Fg 1 As antcpated the ftted SNMIX densty has better ablty to capture the asymmetry and tracks the data more closely than does the ftted NORMIX densty
8 264 TI Ln / Journal of Multvarate Analyss Fg 1 Scatter plot of X 1 X 2 overlad on the contours of ftted two-component a SNMIX b NORMIX models The genune old Swss 1000 franc blls are ndcated by the sold crcles and the pluses + denote the counterfet ones 6 Concludng remarks In ths paper I have presented an ML approach to estmatng the parameters as well as ther nformaton-based standard errors for a multvarate settng of SNMIX models I have descrbed a stochastc normal-truncated normal-multnomal herarchcal representaton of SNMIX and presented an effectve EM algorthm for dealng wth ML estmaton n a flexble complete data framework The formulae for computng the frst two moments of the multvarate truncated normal dstrbuton and ther usefulness n computng condtonal expectatons are also shown The proposed EM algorthm appears to be easly mplemented and coded wth exstng statstcal software such as R package Numercal results llustrated n Secton 5 ndcate that the SNMIX model for the bank data s evdently more adequate than the conventonal NORMIX model Whle the SNMIX model consdered n ths paper has proved ts great flexblty n regulatng skewness among components ts robustness aganst outlers could be serously affected by thck taled observatons Ln et al [16] have recently proposed a remedy to accommodate skewness and heavy-taledness smultaneously usng the mxture of skew t dstrbutons Azzaln and Captano [4] However ther approach s restrcted to data wth unvarate outcomes I conecture that the methodology presented n ths paper can be undertaken under a multvarate settng of skew t mxtures and should yeld satsfactory results n certan stuatons at the expense of addtonal complexty of mplementaton Nevertheless a deeper nvestgaton of those modfcatons s beyond the scope of the present paper but provdes nterestng topcs for further research Acknowledgments The author would lke to express hs deepest thanks to the Chef Edtor the Assocate Edtor and two anonymous referees for ther nsghtful comments and valuable suggestons whch led to substantal mprovements n the presentaton of ths work I am also grateful to Ms Chang-Lng Chen for her ntal smulaton study and to Prof Jack C Lee for hs kndness and patence n proofreadng the earler verson of ths paper Ths research was partly supported by the Natonal Scence Councl of Tawan Grant NO NSC M MY2 Appendx Proofs of Eqs Let l c l c Θ y Z τ denote the complete-data log-lkelhood formed from the sngle observaton y Thus l c Z log w 1 2 log Σ 1 2 y ξ Λ τ T Σ y ξ Λ τ 1 } 2 τ T τ 1 Now recall the formulae for matrx dervatves log Σ 2Σ DagΣ Σ trσ A 2Σ AΣ + DagΣ AΣ A1 Σ f Σ and A are symmetrc and Σ s nonsngular
9 TI Ln / Journal of Multvarate Analyss By applyng A1 the frst dervatves of l c wth respect to w ξ Λ and Σ are l c w Z w Z g w g l c ξ Z Σ y ξ Λ τ l c Λ l c Σ Z Σ y ξ τ T Λ τ τ T 1 2 Z + Dag 2Σ Σ 1 2 Z 2A DagA DagΣ 2Σ y ξ Λ τ y ξ Λ τ T Σ } y ξ Λ τ y ξ Λ τ T Σ A2 where A Σ y ξ Λ τ y ξ Λ τ T Σ Σ Now f Λ s a dagonal matrx e λ dagλ then l c λ 1 2 Z Z 2dag Σ Σ y ξ τ T In the case of Σ 1 Σ g Σ one obtans + 2dag Σ Λ τ τ T } τ y ξ T 1 p Σ τ τ T λ } A3 l c Σ 1 Z 2Σ DagΣ 2Σ y ξ Λ τ y ξ Λ τ T Σ Dag Σ y ξ Λ τ y ξ Λ τ T Σ } 1 2 Z 2A DagA where A Σ y ξ Λ τ y ξ Λ τ T Σ Σ On evaluaton at Θ ˆΘ takng the condtonal expectatons of A2 A4 yelds the score estmates References [1] RB Arellano-Valle H Bolfarne VH Lachos Bayesan nference for skew-normal lnear mxed models J Appl Stat [2] RB Arellano-Valle MG Genton On fundamental skew dstrbutons J Multvarate Anal [3] A Azzaln A class of dstrbutons whch ncludes the normal ones Scand J Statst [4] A Azzaln A Captano Dstrbutons generated by perturbaton of symmetry wth emphass on a multvarate skew t-dstrbuton Roy Statst Soc Ser B [5] A Azzaln A Dalla Valle The multvarate skew-normal dstrbuton Bometrka [6] KE Basford DR Greenway GJ McLachlan D Peel Standard errors of ftted means under normal mxture Comp Statst [7] AP Dempster NM Lard DB Rubn Maxmum lkelhood from ncomplete data va the EM algorthm wth dscusson J Roy Statst Soc Ser B [8] B Efron R Tbshran Bootstrap method for standard errors confdence ntervals and other measures of statstcal accuracy Statst Sc [9] B Flury H Redwyl Multvarate Statstcs a Practcal Approach Cambrdge Unversty Press Cambrdge 1988 [10] S Frühwrth-Schnatter Fnte Mxture and Markov Swtchng Models Sprnger New York 2006 [11] JA Hartgan MA Wong Algorthm AS 136: A K-means clusterng algorthm Appl Stat [12] A Genz Numercal computaton of multvarate normal probabltes J Comput Graph Statst [13] A Genz Comparson of methods for the computaton of multvarate normal probabltes Comp Sc Statst [14] AF Gupta G González-Farías JA Domínguez-Monla A multvarate skew normal dstrbuton J Multvarate Anal [15] GPH Styan Hadamard products and multvarate statstcal analyss Lnear Algebra Appl [16] TI Ln JC Lee WJ Hseh Robust mxture modelng usng the skew t dstrbuton Statst Comp [17] TI Ln JC Lee SY Yen Fnte mxture modellng usng the skew normal dstrbuton Statst Snca [18] Y Ma MG Genton Flexble class of skew-symmetrc dstrbtons Scand J Statst [19] GJ McLachlan KE Basord Mxture Models: Inference and Applcaton to Clusterng Marcel Dekker New York 1988 [20] GJ McLachlan D Peel Fnte Mxture Models Wely New York 2000 [21] RA Redner HF Walker Mxture denstes maxmum lkelhood and the EM algorthm SIAM Rev [22] SK Sahu DK Dey MD Branco A new class of multvarate skew dstrbutons wth applcaton to Bayesan regresson models Canad J Statst [23] GM Talls The moment generatng functon of the truncated mult-normal dstrbuton J Roy Statst Soc Ser B [24] DM Ttterngton AFM Smth UE Markov Statstcal Analyss of Fnte Mxture Dstrbutons Wely New York 1985 [25] CFJ Wu On the convergence propertes of the EM algorthm Ann Statst A4
Robust mixture modeling using multivariate skew t distributions
Robust mxture modelng usng multvarate skew t dstrbutons Tsung-I Ln Department of Appled Mathematcs and Insttute of Statstcs Natonal Chung Hsng Unversty, Tawan August, 1 T.I. Ln (NCHU Natonal Chung Hsng
More informationParametric fractional imputation for missing data analysis. Jae Kwang Kim Survey Working Group Seminar March 29, 2010
Parametrc fractonal mputaton for mssng data analyss Jae Kwang Km Survey Workng Group Semnar March 29, 2010 1 Outlne Introducton Proposed method Fractonal mputaton Approxmaton Varance estmaton Multple mputaton
More informationLecture Notes on Linear Regression
Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume
More informationComposite Hypotheses testing
Composte ypotheses testng In many hypothess testng problems there are many possble dstrbutons that can occur under each of the hypotheses. The output of the source s a set of parameters (ponts n a parameter
More informationMATH 829: Introduction to Data Mining and Analysis The EM algorithm (part 2)
1/16 MATH 829: Introducton to Data Mnng and Analyss The EM algorthm (part 2) Domnque Gullot Departments of Mathematcal Scences Unversty of Delaware Aprl 20, 2016 Recall 2/16 We are gven ndependent observatons
More informationComputation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models
Computaton of Hgher Order Moments from Two Multnomal Overdsperson Lkelhood Models BY J. T. NEWCOMER, N. K. NEERCHAL Department of Mathematcs and Statstcs, Unversty of Maryland, Baltmore County, Baltmore,
More informationANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)
Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of
More informationChapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems
Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons
More information2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification
E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationOn an Extension of Stochastic Approximation EM Algorithm for Incomplete Data Problems. Vahid Tadayon 1
On an Extenson of Stochastc Approxmaton EM Algorthm for Incomplete Data Problems Vahd Tadayon Abstract: The Stochastc Approxmaton EM (SAEM algorthm, a varant stochastc approxmaton of EM, s a versatle tool
More information1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands
Content. Inference on Regresson Parameters a. Fndng Mean, s.d and covarance amongst estmates.. Confdence Intervals and Workng Hotellng Bands 3. Cochran s Theorem 4. General Lnear Testng 5. Measures of
More informationEcon107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)
I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes
More informationMaximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models
ECO 452 -- OE 4: Probt and Logt Models ECO 452 -- OE 4 Maxmum Lkelhood Estmaton of Bnary Dependent Varables Models: Probt and Logt hs note demonstrates how to formulate bnary dependent varables models
More informationDurban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications
Durban Watson for Testng the Lack-of-Ft of Polynomal Regresson Models wthout Replcatons Ruba A. Alyaf, Maha A. Omar, Abdullah A. Al-Shha ralyaf@ksu.edu.sa, maomar@ksu.edu.sa, aalshha@ksu.edu.sa Department
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationFinite Mixture Models and Expectation Maximization. Most slides are from: Dr. Mario Figueiredo, Dr. Anil Jain and Dr. Rong Jin
Fnte Mxture Models and Expectaton Maxmzaton Most sldes are from: Dr. Maro Fgueredo, Dr. Anl Jan and Dr. Rong Jn Recall: The Supervsed Learnng Problem Gven a set of n samples X {(x, y )},,,n Chapter 3 of
More informationUNIVERSITY OF TORONTO Faculty of Arts and Science. December 2005 Examinations STA437H1F/STA1005HF. Duration - 3 hours
UNIVERSITY OF TORONTO Faculty of Arts and Scence December 005 Examnatons STA47HF/STA005HF Duraton - hours AIDS ALLOWED: (to be suppled by the student) Non-programmable calculator One handwrtten 8.5'' x
More informationAppendix B. The Finite Difference Scheme
140 APPENDIXES Appendx B. The Fnte Dfference Scheme In ths appendx we present numercal technques whch are used to approxmate solutons of system 3.1 3.3. A comprehensve treatment of theoretcal and mplementaton
More information4.3 Poisson Regression
of teratvely reweghted least squares regressons (the IRLS algorthm). We do wthout gvng further detals, but nstead focus on the practcal applcaton. > glm(survval~log(weght)+age, famly="bnomal", data=baby)
More information4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA
4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected
More informationA Robust Method for Calculating the Correlation Coefficient
A Robust Method for Calculatng the Correlaton Coeffcent E.B. Nven and C. V. Deutsch Relatonshps between prmary and secondary data are frequently quantfed usng the correlaton coeffcent; however, the tradtonal
More informationThe Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction
ECONOMICS 5* -- NOTE (Summary) ECON 5* -- NOTE The Multple Classcal Lnear Regresson Model (CLRM): Specfcaton and Assumptons. Introducton CLRM stands for the Classcal Lnear Regresson Model. The CLRM s also
More informationChapter 5 Multilevel Models
Chapter 5 Multlevel Models 5.1 Cross-sectonal multlevel models 5.1.1 Two-level models 5.1.2 Multple level models 5.1.3 Multple level modelng n other felds 5.2 Longtudnal multlevel models 5.2.1 Two-level
More informationHidden Markov Models & The Multivariate Gaussian (10/26/04)
CS281A/Stat241A: Statstcal Learnng Theory Hdden Markov Models & The Multvarate Gaussan (10/26/04) Lecturer: Mchael I. Jordan Scrbes: Jonathan W. Hu 1 Hdden Markov Models As a bref revew, hdden Markov models
More informationLinear Approximation with Regularization and Moving Least Squares
Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...
More informationChapter 11: Simple Linear Regression and Correlation
Chapter 11: Smple Lnear Regresson and Correlaton 11-1 Emprcal Models 11-2 Smple Lnear Regresson 11-3 Propertes of the Least Squares Estmators 11-4 Hypothess Test n Smple Lnear Regresson 11-4.1 Use of t-tests
More informationParametric fractional imputation for missing data analysis
Secton on Survey Research Methods JSM 2008 Parametrc fractonal mputaton for mssng data analyss Jae Kwang Km Wayne Fuller Abstract Under a parametrc model for mssng data, the EM algorthm s a popular tool
More informationLecture 3: Probability Distributions
Lecture 3: Probablty Dstrbutons Random Varables Let us begn by defnng a sample space as a set of outcomes from an experment. We denote ths by S. A random varable s a functon whch maps outcomes nto the
More informationNegative Binomial Regression
STATGRAPHICS Rev. 9/16/2013 Negatve Bnomal Regresson Summary... 1 Data Input... 3 Statstcal Model... 3 Analyss Summary... 4 Analyss Optons... 7 Plot of Ftted Model... 8 Observed Versus Predcted... 10 Predctons...
More informationWhy Bayesian? 3. Bayes and Normal Models. State of nature: class. Decision rule. Rev. Thomas Bayes ( ) Bayes Theorem (yes, the famous one)
Why Bayesan? 3. Bayes and Normal Models Alex M. Martnez alex@ece.osu.edu Handouts Handoutsfor forece ECE874 874Sp Sp007 If all our research (n PR was to dsappear and you could only save one theory, whch
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationMarkov Chain Monte Carlo (MCMC), Gibbs Sampling, Metropolis Algorithms, and Simulated Annealing Bioinformatics Course Supplement
Markov Chan Monte Carlo MCMC, Gbbs Samplng, Metropols Algorthms, and Smulated Annealng 2001 Bonformatcs Course Supplement SNU Bontellgence Lab http://bsnuackr/ Outlne! Markov Chan Monte Carlo MCMC! Metropols-Hastngs
More informationDr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur
Analyss of Varance and Desgn of Experment-I MODULE VII LECTURE - 3 ANALYSIS OF COVARIANCE Dr Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur Any scentfc experment s performed
More informationCSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography
CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve
More informationGaussian Mixture Models
Lab Gaussan Mxture Models Lab Objectve: Understand the formulaton of Gaussan Mxture Models (GMMs) and how to estmate GMM parameters. You ve already seen GMMs as the observaton dstrbuton n certan contnuous
More informationLimited Dependent Variables
Lmted Dependent Varables. What f the left-hand sde varable s not a contnuous thng spread from mnus nfnty to plus nfnty? That s, gven a model = f (, β, ε, where a. s bounded below at zero, such as wages
More informationStatistics Chapter 4
Statstcs Chapter 4 "There are three knds of les: les, damned les, and statstcs." Benjamn Dsrael, 1895 (Brtsh statesman) Gaussan Dstrbuton, 4-1 If a measurement s repeated many tmes a statstcal treatment
More informationx = , so that calculated
Stat 4, secton Sngle Factor ANOVA notes by Tm Plachowsk n chapter 8 we conducted hypothess tests n whch we compared a sngle sample s mean or proporton to some hypotheszed value Chapter 9 expanded ths to
More informationKernel Methods and SVMs Extension
Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general
More informationSTAT 3008 Applied Regression Analysis
STAT 3008 Appled Regresson Analyss Tutoral : Smple Lnear Regresson LAI Chun He Department of Statstcs, The Chnese Unversty of Hong Kong 1 Model Assumpton To quantfy the relatonshp between two factors,
More informationSIO 224. m(r) =(ρ(r),k s (r),µ(r))
SIO 224 1. A bref look at resoluton analyss Here s some background for the Masters and Gubbns resoluton paper. Global Earth models are usually found teratvely by assumng a startng model and fndng small
More informationComparison of Regression Lines
STATGRAPHICS Rev. 9/13/2013 Comparson of Regresson Lnes Summary... 1 Data Input... 3 Analyss Summary... 4 Plot of Ftted Model... 6 Condtonal Sums of Squares... 6 Analyss Optons... 7 Forecasts... 8 Confdence
More informationMaximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models
ECO 452 -- OE 4: Probt and Logt Models ECO 452 -- OE 4 Mamum Lkelhood Estmaton of Bnary Dependent Varables Models: Probt and Logt hs note demonstrates how to formulate bnary dependent varables models for
More informationSTAT 405 BIOSTATISTICS (Fall 2016) Handout 15 Introduction to Logistic Regression
STAT 45 BIOSTATISTICS (Fall 26) Handout 5 Introducton to Logstc Regresson Ths handout covers materal found n Secton 3.7 of your text. You may also want to revew regresson technques n Chapter. In ths handout,
More informationConjugacy and the Exponential Family
CS281B/Stat241B: Advanced Topcs n Learnng & Decson Makng Conjugacy and the Exponental Famly Lecturer: Mchael I. Jordan Scrbes: Bran Mlch 1 Conjugacy In the prevous lecture, we saw conjugate prors for the
More informationGlobal Sensitivity. Tuesday 20 th February, 2018
Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values
More informationDepartment of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution
Department of Statstcs Unversty of Toronto STA35HS / HS Desgn and Analyss of Experments Term Test - Wnter - Soluton February, Last Name: Frst Name: Student Number: Instructons: Tme: hours. Ads: a non-programmable
More informationComputing MLE Bias Empirically
Computng MLE Bas Emprcally Kar Wa Lm Australan atonal Unversty January 3, 27 Abstract Ths note studes the bas arses from the MLE estmate of the rate parameter and the mean parameter of an exponental dstrbuton.
More informationJoint Statistical Meetings - Biopharmaceutical Section
Iteratve Ch-Square Test for Equvalence of Multple Treatment Groups Te-Hua Ng*, U.S. Food and Drug Admnstraton 1401 Rockvlle Pke, #200S, HFM-217, Rockvlle, MD 20852-1448 Key Words: Equvalence Testng; Actve
More informationStat260: Bayesian Modeling and Inference Lecture Date: February 22, Reference Priors
Stat60: Bayesan Modelng and Inference Lecture Date: February, 00 Reference Prors Lecturer: Mchael I. Jordan Scrbe: Steven Troxler and Wayne Lee In ths lecture, we assume that θ R; n hgher-dmensons, reference
More informationCOMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
More informationBOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS. M. Krishna Reddy, B. Naveen Kumar and Y. Ramu
BOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS M. Krshna Reddy, B. Naveen Kumar and Y. Ramu Department of Statstcs, Osmana Unversty, Hyderabad -500 007, Inda. nanbyrozu@gmal.com, ramu0@gmal.com
More informationA Hybrid Variational Iteration Method for Blasius Equation
Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method
More informationBayesian predictive Configural Frequency Analysis
Psychologcal Test and Assessment Modelng, Volume 54, 2012 (3), 285-292 Bayesan predctve Confgural Frequency Analyss Eduardo Gutérrez-Peña 1 Abstract Confgural Frequency Analyss s a method for cell-wse
More informationTHE SUMMATION NOTATION Ʃ
Sngle Subscrpt otaton THE SUMMATIO OTATIO Ʃ Most of the calculatons we perform n statstcs are repettve operatons on lsts of numbers. For example, we compute the sum of a set of numbers, or the sum of the
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationMMA and GCMMA two methods for nonlinear optimization
MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons
More informationOn Outlier Robust Small Area Mean Estimate Based on Prediction of Empirical Distribution Function
On Outler Robust Small Area Mean Estmate Based on Predcton of Emprcal Dstrbuton Functon Payam Mokhtaran Natonal Insttute of Appled Statstcs Research Australa Unversty of Wollongong Small Area Estmaton
More informationA Comparative Study for Estimation Parameters in Panel Data Model
A Comparatve Study for Estmaton Parameters n Panel Data Model Ahmed H. Youssef and Mohamed R. Abonazel hs paper examnes the panel data models when the regresson coeffcents are fxed random and mxed and
More informationSee Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)
Count Data Models See Book Chapter 11 2 nd Edton (Chapter 10 1 st Edton) Count data consst of non-negatve nteger values Examples: number of drver route changes per week, the number of trp departure changes
More informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationSTAT 511 FINAL EXAM NAME Spring 2001
STAT 5 FINAL EXAM NAME Sprng Instructons: Ths s a closed book exam. No notes or books are allowed. ou may use a calculator but you are not allowed to store notes or formulas n the calculator. Please wrte
More informationChapter 13: Multiple Regression
Chapter 13: Multple Regresson 13.1 Developng the multple-regresson Model The general model can be descrbed as: It smplfes for two ndependent varables: The sample ft parameter b 0, b 1, and b are used to
More information2016 Wiley. Study Session 2: Ethical and Professional Standards Application
6 Wley Study Sesson : Ethcal and Professonal Standards Applcaton LESSON : CORRECTION ANALYSIS Readng 9: Correlaton and Regresson LOS 9a: Calculate and nterpret a sample covarance and a sample correlaton
More informationA new construction of 3-separable matrices via an improved decoding of Macula s construction
Dscrete Optmzaton 5 008 700 704 Contents lsts avalable at ScenceDrect Dscrete Optmzaton journal homepage: wwwelsevercom/locate/dsopt A new constructon of 3-separable matrces va an mproved decodng of Macula
More informationFirst Year Examination Department of Statistics, University of Florida
Frst Year Examnaton Department of Statstcs, Unversty of Florda May 7, 010, 8:00 am - 1:00 noon Instructons: 1. You have four hours to answer questons n ths examnaton.. You must show your work to receve
More informationEM and Structure Learning
EM and Structure Learnng Le Song Machne Learnng II: Advanced Topcs CSE 8803ML, Sprng 2012 Partally observed graphcal models Mxture Models N(μ 1, Σ 1 ) Z X N N(μ 2, Σ 2 ) 2 Gaussan mxture model Consder
More informationAn (almost) unbiased estimator for the S-Gini index
An (almost unbased estmator for the S-Gn ndex Thomas Demuynck February 25, 2009 Abstract Ths note provdes an unbased estmator for the absolute S-Gn and an almost unbased estmator for the relatve S-Gn for
More informationStructure and Drive Paul A. Jensen Copyright July 20, 2003
Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.
More informationLinear Regression Analysis: Terminology and Notation
ECON 35* -- Secton : Basc Concepts of Regresson Analyss (Page ) Lnear Regresson Analyss: Termnology and Notaton Consder the generc verson of the smple (two-varable) lnear regresson model. It s represented
More informationSimulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests
Smulated of the Cramér-von Mses Goodness-of-Ft Tests Steele, M., Chaselng, J. and 3 Hurst, C. School of Mathematcal and Physcal Scences, James Cook Unversty, Australan School of Envronmental Studes, Grffth
More informationOn mutual information estimation for mixed-pair random variables
On mutual nformaton estmaton for mxed-par random varables November 3, 218 Aleksandr Beknazaryan, Xn Dang and Haln Sang 1 Department of Mathematcs, The Unversty of Msssspp, Unversty, MS 38677, USA. E-mal:
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have
More informationAdvances in Longitudinal Methods in the Social and Behavioral Sciences. Finite Mixtures of Nonlinear Mixed-Effects Models.
Advances n Longtudnal Methods n the Socal and Behavoral Scences Fnte Mxtures of Nonlnear Mxed-Effects Models Jeff Harrng Department of Measurement, Statstcs and Evaluaton The Center for Integrated Latent
More informationLecture 12: Classification
Lecture : Classfcaton g Dscrmnant functons g The optmal Bayes classfer g Quadratc classfers g Eucldean and Mahalanobs metrcs g K Nearest Neghbor Classfers Intellgent Sensor Systems Rcardo Guterrez-Osuna
More informationLinear Algebra and its Applications
Lnear Algebra and ts Applcatons 4 (00) 5 56 Contents lsts avalable at ScenceDrect Lnear Algebra and ts Applcatons journal homepage: wwwelsevercom/locate/laa Notes on Hlbert and Cauchy matrces Mroslav Fedler
More informationEstimation of the Mean of Truncated Exponential Distribution
Journal of Mathematcs and Statstcs 4 (4): 84-88, 008 ISSN 549-644 008 Scence Publcatons Estmaton of the Mean of Truncated Exponental Dstrbuton Fars Muslm Al-Athar Department of Mathematcs, Faculty of Scence,
More informationLecture 4 Hypothesis Testing
Lecture 4 Hypothess Testng We may wsh to test pror hypotheses about the coeffcents we estmate. We can use the estmates to test whether the data rejects our hypothess. An example mght be that we wsh to
More informationAppendix B: Resampling Algorithms
407 Appendx B: Resamplng Algorthms A common problem of all partcle flters s the degeneracy of weghts, whch conssts of the unbounded ncrease of the varance of the mportance weghts ω [ ] of the partcles
More informationFor now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.
Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson
More informationChat eld, C. and A.J.Collins, Introduction to multivariate analysis. Chapman & Hall, 1980
MT07: Multvarate Statstcal Methods Mke Tso: emal mke.tso@manchester.ac.uk Webpage for notes: http://www.maths.manchester.ac.uk/~mkt/new_teachng.htm. Introducton to multvarate data. Books Chat eld, C. and
More informationMaximum Likelihood Estimation (MLE)
Maxmum Lkelhood Estmaton (MLE) Ken Kreutz-Delgado (Nuno Vasconcelos) ECE 175A Wnter 01 UCSD Statstcal Learnng Goal: Gven a relatonshp between a feature vector x and a vector y, and d data samples (x,y
More information4DVAR, according to the name, is a four-dimensional variational method.
4D-Varatonal Data Assmlaton (4D-Var) 4DVAR, accordng to the name, s a four-dmensonal varatonal method. 4D-Var s actually a drect generalzaton of 3D-Var to handle observatons that are dstrbuted n tme. The
More informationSTATS 306B: Unsupervised Learning Spring Lecture 10 April 30
STATS 306B: Unsupervsed Learnng Sprng 2014 Lecture 10 Aprl 30 Lecturer: Lester Mackey Scrbe: Joey Arthur, Rakesh Achanta 10.1 Factor Analyss 10.1.1 Recap Recall the factor analyss (FA) model for lnear
More informationCS 2750 Machine Learning. Lecture 5. Density estimation. CS 2750 Machine Learning. Announcements
CS 750 Machne Learnng Lecture 5 Densty estmaton Mlos Hauskrecht mlos@cs.ptt.edu 539 Sennott Square CS 750 Machne Learnng Announcements Homework Due on Wednesday before the class Reports: hand n before
More informationDifference Equations
Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1
More informationChapter 12. Ordinary Differential Equation Boundary Value (BV) Problems
Chapter. Ordnar Dfferental Equaton Boundar Value (BV) Problems In ths chapter we wll learn how to solve ODE boundar value problem. BV ODE s usuall gven wth x beng the ndependent space varable. p( x) q(
More informationComparison of the Population Variance Estimators. of 2-Parameter Exponential Distribution Based on. Multiple Criteria Decision Making Method
Appled Mathematcal Scences, Vol. 7, 0, no. 47, 07-0 HIARI Ltd, www.m-hkar.com Comparson of the Populaton Varance Estmators of -Parameter Exponental Dstrbuton Based on Multple Crtera Decson Makng Method
More informationClassification as a Regression Problem
Target varable y C C, C,, ; Classfcaton as a Regresson Problem { }, 3 L C K To treat classfcaton as a regresson problem we should transform the target y nto numercal values; The choce of numercal class
More informationP R. Lecture 4. Theory and Applications of Pattern Recognition. Dept. of Electrical and Computer Engineering /
Theory and Applcatons of Pattern Recognton 003, Rob Polkar, Rowan Unversty, Glassboro, NJ Lecture 4 Bayes Classfcaton Rule Dept. of Electrcal and Computer Engneerng 0909.40.0 / 0909.504.04 Theory & Applcatons
More information10-701/ Machine Learning, Fall 2005 Homework 3
10-701/15-781 Machne Learnng, Fall 2005 Homework 3 Out: 10/20/05 Due: begnnng of the class 11/01/05 Instructons Contact questons-10701@autonlaborg for queston Problem 1 Regresson and Cross-valdaton [40
More informationGoodness of fit and Wilks theorem
DRAFT 0.0 Glen Cowan 3 June, 2013 Goodness of ft and Wlks theorem Suppose we model data y wth a lkelhood L(µ) that depends on a set of N parameters µ = (µ 1,...,µ N ). Defne the statstc t µ ln L(µ) L(ˆµ),
More informationECONOMICS 351*-A Mid-Term Exam -- Fall Term 2000 Page 1 of 13 pages. QUEEN'S UNIVERSITY AT KINGSTON Department of Economics
ECOOMICS 35*-A Md-Term Exam -- Fall Term 000 Page of 3 pages QUEE'S UIVERSITY AT KIGSTO Department of Economcs ECOOMICS 35* - Secton A Introductory Econometrcs Fall Term 000 MID-TERM EAM ASWERS MG Abbott
More informationPredictive Analytics : QM901.1x Prof U Dinesh Kumar, IIMB. All Rights Reserved, Indian Institute of Management Bangalore
Sesson Outlne Introducton to classfcaton problems and dscrete choce models. Introducton to Logstcs Regresson. Logstc functon and Logt functon. Maxmum Lkelhood Estmator (MLE) for estmaton of LR parameters.
More informationTHE ROYAL STATISTICAL SOCIETY 2006 EXAMINATIONS SOLUTIONS HIGHER CERTIFICATE
THE ROYAL STATISTICAL SOCIETY 6 EXAMINATIONS SOLUTIONS HIGHER CERTIFICATE PAPER I STATISTICAL THEORY The Socety provdes these solutons to assst canddates preparng for the eamnatons n future years and for
More informationLecture 3 Stat102, Spring 2007
Lecture 3 Stat0, Sprng 007 Chapter 3. 3.: Introducton to regresson analyss Lnear regresson as a descrptve technque The least-squares equatons Chapter 3.3 Samplng dstrbuton of b 0, b. Contnued n net lecture
More informationOn the Interval Zoro Symmetric Single-step Procedure for Simultaneous Finding of Polynomial Zeros
Appled Mathematcal Scences, Vol. 5, 2011, no. 75, 3693-3706 On the Interval Zoro Symmetrc Sngle-step Procedure for Smultaneous Fndng of Polynomal Zeros S. F. M. Rusl, M. Mons, M. A. Hassan and W. J. Leong
More informationChapter 12 Analysis of Covariance
Chapter Analyss of Covarance Any scentfc experment s performed to know somethng that s unknown about a group of treatments and to test certan hypothess about the correspondng treatment effect When varablty
More informationOne-sided finite-difference approximations suitable for use with Richardson extrapolation
Journal of Computatonal Physcs 219 (2006) 13 20 Short note One-sded fnte-dfference approxmatons sutable for use wth Rchardson extrapolaton Kumar Rahul, S.N. Bhattacharyya * Department of Mechancal Engneerng,
More information