Logistic Regression Sara Vyrostek Senior Exercise November 16, 2001

Size: px
Start display at page:

Download "Logistic Regression Sara Vyrostek Senior Exercise November 16, 2001"

Transcription

1 ogstc Rgrsso Sara Vrostk Sor Ercs Novmbr 6,

2 Itroducto: I th modlg of data, aalsts dvlop rlatoshps basd upo th obsrvd valus of a st of prdctor varabls ordr to dtrm th pctd valu of th rspos varabl of trst, a tchqu kow as rgrsso. A vr commo form of rgrsso aalss s lar rgrsso, whch ams th ffct of a ut chag a prdctor varabl o th rspos varabl. owvr, thr ar ma cass, whthr t s du to pattrs th data or th dsg of th data tslf, whr a lar aalss s approprat. O partcular cas whr th lar aalss dos ot provd a adquat ft occurs wh th rspos varabl s bar, mag t has ol two possbl valus. For ths scaro, a rlatd tchqu kow as logstc rgrsso s usd to modl th data. Th logstc modl provds a curvlar ft wth asmptots at both zro ad o, th two possbl valus for th rspos varabl. Fgur dsplas a graph of a logstc modl ft to a tpcal bar data st, vsuall dsplag ths charactrstcs. Fgur : A Tpcal ogstc Modl Ntr, Joh, Wllam Wassrma, Chrstophr J. Nachtshm, ad Mchal. Kutr, Appld ar Rgrsso Modls, 3 rd d. Chcago: McGraw-ll, 996 p. 57.

3 Th logstc rgrsso tchqu s wdl usd a ara whch th rspos varabl tds to tak o th bar form. For ampl, halth rsarch t s commo for th rspos varabl to hav th form s or o or rspodd to mdcato vrsus dd ot rspod to mdcato, stuatos whch logstc rgrsso would b mplod. Not ol wll ths papr srv as a troducto to rgrsso th logstc sttg, but t wll also stp through th procsss of paramtr stmato ad trprtato, th buldg of a logstc rgrsso modl, ad th aalss of ft. Throughout th papr, a ogog aalss of a data st b mas of both had calculatos as wll as th trprtato of SAS output wll provd th opportut to work through all of th topcs dscussd. Th data st, a subst of varabls ad obsrvatos tak from a largr stud do at th Uvrst of Massachustts Ads Rsarch Ut ordr to am drug us assocato wth IV, was foud osmr ad mshow s Appld ogstc Rgrsso. Th varabls th data st ar as follows: Fgur : Dscrpto of Varabls Varabl Dscrpto Cods/Valus Nam Idtfcato Cod -575 ID Ag at Erollmt Yars ag 3 Bck Dprsso Scor at.-54. bck Admsso 4 IV Drug Us stor at Nvr, Prvous, IV Admsso 3Rct 5 Numbr of Pror Drug -4 prors Tratmts 6 Subjct s Rac Wht, Othr rac 7 Tratmt Radomzato Short, og tratmt Assgmt 8 Tratmt St A, B st 9 Rmad Drug Fr for Moths Rmad Drug Fr, Othrws drugfr osmr, Davd W., ad Stal mshow, Appld ogstc Rgrsso Nw York: Joh Wl & Sos, pp

4 Smpl ogstc Rgrsso Modl: Workg th bar sttg, Y ca tak o ol o of two valus, ad thus ca b codd as follows:, Y, f f th vt occurs th vt dos ot occur I th smpl sttg wth ol o prdctor varabl, th logstc rgrsso modl wll tak o th followg gral form: whr, Y ε, E[ Y X ] ad ε s a radom rror trm. 3 Altratvl, ca b wrtt as Bcaus rprsts th probablt that Y gv th valu of X, ths mpls that rprsts th probablt that Y, gv th valu of X. Y s dstrbutd such that t has two possbl outcoms wth a probablt of succss, mag that th Y 's follow th Broull dstrbuto. Ulk th lar modl whr th rror trms ar ormall dstrbutd, th rror trms ths modl hav ol two possbl valus. If Y, th ε wth probablt, corrspodg to th abov formato. O th othr had, f Y whch occurs wth probablt, th ε. Th ma of th rror trms ca b foud b calculatg thr pctd valu. Sc ε ca ol tak o two possbl valus, th pctd valu s obtad through th followg formula: 3 Chattrj, Samprt, Al S. ad, ad Brtram Prc, Rgrsso Aalss b Eampl Nw York: Joh Wl & Sos, p. 3. 4

5 5 ] ][ [ ] [ P P E ε ε ε ε ε ε ε Th varac of th rror trms ca b dtrmd a smlar mar: ] [ } ] ]{[ [ ] [ ] [ ] [ ] [ ] [ P P E E Var ε ε ε ε ε ε ε ε ε Ths formato provds a basc udrstadg of th dstrbutoal sttg of ths aalss. 4 avg dfd ad dscrbd th logstc rgrsso fucto, th t task volvs stmatg th paramtrs ad for th modl wh fttg t to a actual data st. B th vr atur of th probablt fucto dscrbd abov, th stmato procss ca b smplfd b formg a lar quato volvg ths two paramtrs. Wth dotg th probablt that Y, th odds of Y occurrg ca b dfd as th probablt that Y dvdd b th probablt that Y, or Y. I othr words, odds. Th odds ar mportat for a bar data st bcaus th offr a rato for th chacs of a vt occurrg as opposd to a vt ot occurrg. For ampl, f th probablt of th su comg out o a gv da s two-thrds, mag that th probablt of ot sg th su o a gv da s o-thrd, th th odds of a su da ar 3 / 3 /, or a two to o rato. Ths 4 Chattrj, p. 3.

6 mpls that a su da s twc as lkl as a cloud da. Bcaus th quato for th odds rsults a dffcult fucto to ft, h ca b dfd as th atural log of th odds, also kow as th logt fucto: h l l l l Ths fucto, h, s of th lar form, provdg a much asr modl to ft. 5 Th paramtrs ad ca b stmatd through th mthod of mamum lklhood stmato. Ths mthod solats th valus of a paramtr that mamz th lklhood fucto, whr th lklhood fucto, s dfd as th jot probablt dstrbuto for all of th data pots. 6 Sc th Y 's hav a Broull dstrbuto, th probablt dst fucto ca b dfd as P Y f [ ], whr or, ad,,. Bcaus th Y 's ar dpdt, th lklhood fucto ca b dfd as follows:, g,..., f [ ] 5 osmr, Davd W., ad Stal mshow, Appld ogstc Rgrsso Nw York: Joh Wl & Sos, 989 p osmr, p. 8. 6

7 7 I ordr to mamz ths fucto, th drvatv must b tak wth rspct to ach of th paramtrs. Th, th rsultg quatos would b st qual to zro ad solvd smultaousl. Ths procss ca b smplfd b prformg th sam aalss o th atural log of th lklhood fucto, bg that mamzg th atural log of th fucto would rsult th sam valu as mamzg th lklhood fucto tslf. Obtag th log-lklhood fucto: [ ] { } { } { } [ ] [ ] { } { } l l l l l ] l[ ] l[ ] l[ ] l[ l ] [ l, l Now takg th drvatv, frst wth rspct to ad th wth rspct to ad sttg ach qual to zro, th followg lklhood quatos ar formd: ] [, l ] [, l Bcaus th lklhood quatos ar ot lar, solvg ths quatos smultaousl rqurs a tratv procdur that s ormall lft to a softwar packag. 7 7 Ntr, pp

8 t us cosdr th data st o drug us. Suppos a smpl modl s dsrd usg ol o prdctor varabl, t, whr t rprsts th tp of tratmt admstrd. W wll dsgat th rspos varabl Y as a dcator for whthr or ot th subjct rmad drug fr for twlv moths. Th basc logstc modl s: Y ε. t t ε t Th accompag logt fucto s: l t t. h t t Usg SAS to obta th approprat paramtr stmats, th followg output s producd: Dspla : Smpl ogstc Rgrsso--Paramtr Estmats Th OGISTIC Procdur Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. tratmt Odds Rato Estmats Pot 95% Wald Effct Estmat Cofdc mts tratmt Dspla rports th valu for b, th paramtr stmat for, as.978, ad th valu for b, th paramtr stmat for, as.437. Thus, th stmatd logt fucto would rad h t t, wth th paramtr stmats for th modl havg b dtrmd through th mthod of mamum lklhood stmato. Th valu for th log-lklhood fucto s also producd th SAS output: Dspla : Smpl ogstc Rgrsso--Th og-klhood Fucto Modl Ft Statstcs Itrcpt Itrcpt ad 8

9 Crtro Ol Covarats AIC SC og B trprtg th output Dspla, t s vdt that th log-lklhood fucto s qual to / Th output also rports th valu of th log-lklhood fucto for th modl wth just th trcpt, whch s qual to / Sc th logt fucto s wrtt a lar form wth rprstg th slop, b rprsts th chag h for a o-ut chag. Thrfor, b h h l l l odds l odds odds l odds. Takg th potal of both sds, w gt: odds odds b, whch ca b dfd as th odds rato. Thrfor, aftr fdg th paramtr stmats, th b valu wll rprst th prctag cras th probablt that Y for ach ut cras X. 8 Ths valu for th drug data s s Dspla, wth a pot stmat for th odds rato of.548. Ths dcats that b chagg th tratmt prod from short to log, th probablt of a subjct rmag drug fr for a twlv-moth prod aftr th tratmt crass b 54.8%. Dspt havg ft th modl ad trprtd th coffcts, ths formato s uslss ulss th prdctor varabl s sgfcat to th modl. O mthod for dtrmg whthr or 8 Ntr, p

10 ot th coffct s sgfcat s through th lklhood rato tst. Assumg that w hav obsrvatos o ach varabl, th dvac of th modl, D, wll b dfd as follows: lklhood of th currt Modl D l lklhood of th saturatd Modl Th saturatd modl rprsts th modl cotag paramtrs, such that th modl prfctl prdcts th obsrvd data st. I othr words, th prdctd valus for ths modl ar qual to th obsrvd valus th data st. Th dvac provds a mas of comparg th lklhood of th modl that has b ft, or th probablt of obtag th obsrvd data st gv th currt modl, to that of th saturatd modl. 9 B mapulatg th abov dfto, w fd: [ l lklhood of th currt Modl l lklhood of th saturatd ] D Modl Rcall that th gral modl s of th form Y ε. For th currt modl, ˆ srvs as th stmator for. owvr, th saturatd modl grats th complt data st such that srvs as th stmator for. Also rcallg th drvato of th lklhood fucto, w kow that: l, l [ ] { l[ ] l[ ]}. Ths quato ca b substtutd to th formula for th dvac ad th mapulatd ordr to gt th followg workg quato:. 9 osmr, p. 3.

11 D l[ ˆ ] l[ ˆ ] l l ˆ ˆ l l. l[ ˆ ] l l[ ˆ ] l Esstall, th dvac taks th lklhood of th currt modl whr a lmt of rror s prst ad subtracts th lklhood of th saturatd modl whch thr s o rror trm prst, ad th sums ovr ths dffrc. Thus, th dvac th logstc sttg srvs th sam purpos as th rsdual sums of squars th lar sttg. I ordr to dtrm whthr or ot th paramtr s sgfcat to th modl, th dvac of th modl cotag th paramtr must b compard wth th dvac of th modl wthout th paramtr. Thrfor, th tst statstc, G, s: G Dfor th modl wthout th varabl - Dfor th modl wth th varabl lklhood of currt Modl wthout G l lklhood of saturatd Modl lklhood of currt Modl wthout G l lklhood of currt Modl wth lklhood of currt Modl wth l lklhood of saturatd Modl G l lklhood of currt Modl wthout l lklhood of currt Modl wth W hav alrad foud th lklhood fucto for th modl wth th sgl prdctor varabl. For th lklhood fucto for th modl wthout th sgl prdctor varabl, th probablt that Y wll b qual to th avrag umbr of zros th sampl, whl th probablt that Y wll b qual to th avrag umbr of os th sampl. Thrfor, osmr, p. 3. osmr, pp. 3-4.

12 Gv that lklhood fucto, w ca dtrm th approprat quato for th tst statstc G: G ˆ ˆ l. G follows th ch-squard dstrbuto wth o dgr of frdom. I chckg th sgfcac of th coffct, th followg ull ad altratv hpothss ar to b tstd: : : a. For ths tst, th dcso rul rqurs that f ; χ α > G, s to b rjctd, mag that th coffct would b dmd sgfcat. O th othr had f ; χ α G, th o must fal to rjct, cocludg that th coffct s sgfcat. SAS rports th tst statstc ad th corrspodg p-valu, whr th p-valu s qual to th probablt of obsrvg a tst statstc at last as trm as th obsrvd valu assumg that th ull hpothss s tru. For ths lklhood rato tst, th p-valu PG th obsrvd G-valu, whr G ~ ; α χ.

13 Bcaus th α-lvl rprsts th probablt of rjctg th ull hpothss wh th ull hpothss s tru, as log as th p-valu s lss tha th chos α-lvl, th ull hpothss ca b safl rjctd. For all of th futur tsts ths papr, w wll us a α-lvl of.5. For th drug data st, G ca b calculatd from th SAS output two mars. Frst, kowg th formula for G, w s that: G l lklhood of currt Modl wthout l lklhood of currt Modl wth. Th valus for *log-lklhood trs from Dspla ca b srtd to calculat th tst statstc. Usg ths formato, G Thrfor wth a alpha lvl of.5 ad χ.5; , th ull hpothss ca b rjctd, mag that th paramtr stmat for th tp of tratmt admstrd s sgfcat to th modl. Th p- valu for ths tst rprsts th probablt of obsrvg a valu for G of at last 5.78 such that th p-valu P G Bcaus ths p-valu s lss tha th dsrd alpha lvl of.5, th ull hpothss would b rjctd hr, ladg to th sam cocluso. Aothr scto of th SAS output actuall computs G, or th lklhood rato tst statstc, ad calculats th approprat p-valu for th tst of th abov hpothss: Dspla 3: Smpl ogstc Rgrsso--Th klhood Rato Tst Tstg Global Null pothss: BETA Tst Ch-Squar DF Pr > ChSq klhood Rato Scor Wald Th row radg lklhood rato rports th tst statstc G 5.78, th sam as that calculatd abov, wth o dgr of frdom, ad a p-valu of.9. osmr, pp

14 Aothr mthod for computg th sgfcac of a coffct s through th Wald tst, bk whr ordr to tst th sam hpothss as abov, th tst statstc W s usd. Sc s b ths tst statstc s appromatl ormal, f W z α / th o must fal to rjct th ull hpothss, whl f W > z α /, th th ull hpothss ca b rjctd at th gv alpha k lvl. 3 For ths tst, th p-valu ca b dfd as follows: p valu P W > th obsrvd tst statstc P W > th obsrvd tst statstc,.437 whr W ~ z α /. For th drug data st, W.636. Usg a alpha lvl of.93.5, th crtcal valu s z , ad th corrspodg p-valu s P W >.636 * Sc W > z ad th corrspodg p-valu s lss tha.5, th ull hpothss wll stll b rjctd ad sam cocluso of varabl sgfcac ca b rachd. Ths tst also ca b wrtt a altratv mar. Bcaus th squarg a ormal radom varabl wll rsult a ch-squar radom varabl wth o dgr of frdom 4, th bk Wald tst statstc ca b wrtt as W whr W ~ χ α;. I accordac s bk wth ths chag, th dcso rul must b adjustd such that th ull hpothss s rjctd wh W > χ α;. kws, th p-valu wll b rdfd so that th p valu P W > th obsrvd tst statstc. ookg back at Dspla, w ca s that SAS rports ths tst statstc, rathr tha thos prvousl dscrbd. For th sgl prdctor 3 Ntr, pp

15 varabl tratmt, th Wald ch-squar tst statstc s 5.66 wth a corrspodg p-valu of.36, aga dcatg that ths prdctor varabl s sgfcat to th modl. Multpl ogstc Rgrsso Modl: W ca td th abov aalss for th smpl logstc rgrsso modl whr w hav ol o prdctor varabl to a sttg wth mor tha o prdctor varabl. I ths sttg, th vctor... rprsts th collcto of p prdctor varabls for ths modl. p Th formulas for th probablt that Y,, as wll as for th logt trasformato, h, ca b tdd whr:... p p.. p p ad h... p p. Th ol tm that th modl wll dffr from ths gral formula s f o of th prdctor varabls s a dscrt, catgorcal varabl wth mor tha two lvls. If o or mor of ths s prst ad f th umbr of varabl catgors s qual to k, th k- dsg varabls must b cratd. 5 Ths dsg varabls ar just bar varabls mat to srv as a dcator for o of th lvls of th assocatd varabl. Usg our drug data varabl IV as a ampl, ths varabl would hav to b rcodd bcaus t s a dscrt varabl wth thr lvls: rct, prvous, or o hstor of IV drug us. Bcaus IV currtl has thr dscrt lvls, two dcator varabls would hav to b cratd ad substtutd to th modl. Ths two w varabls mght b IV ad IV, whr, IV 3 IV, ad, IV 3, IV IV., IV 4 ogg, Robrt V., ad All T. Crag, Itroducto to Mathmatcal Statstcs odo: Th MacMlla Compa, 97 p osmr, pp

16 W would ol b rqurd to clud two dcator varabls bcaus th thrd valu of th prdctor varabl s mplctl cludd th modl. I ths cas, wh IV whch mas that IV, ths dcats that th subjct has o hstor of drug us. kws, wh IV whch mas that ow IV, ths mas that th subjct has a hstor of prvous drug us. owvr, f both of ths dcator varabls qual zro, ths would mpl that th subjct has a hstor of thr o drug us or prvous drug us. Istad, ths would dcat that th subjct has a hstor of rct drug us. Th paramtrs th multpl sttg ar oc aga dtrmd through mamum lklhood stmato. Bcaus Y stll rmas a Broull varabl wth th sam probablt dstrbuto, th drvato of th mamum lklhood stmators rmas th sam, wth th cpto of th cluso of mor paramtrs. Thus, th log-lklhood quato would tak th form: l... p p {... p p l },,..., p I th sam mar as bfor, th quatos rsultg from takg th drvatv of th loglklhood quato wth rspct to ach of th paramtrs ad th sttg ach drvatv qual to zro would b solvd smultaousl ordr to obta th stmats. Bcaus ths procdur s v mor computatoall tsv wth multpl paramtrs, th stmato oc aga s lft to computr softwar. 6 Eamg multpl logstc rgrsso b mas of th drug us data st, w ca dvlop a modl cludg thr chos prdctor varabls: ag, rac, ad tratmt. Th modl wll tak th form:. 6 Ntr, p. 7. 6

17 Y ε, whr 33 33, wth rprstg ag, rac, ad 3 tratmt. Th logt wll tak th form: h 33 Usg SAS to stmat th approprat paramtrs, th followg formato s prtd th output: Dspla 4: Multpl ogstc Rgrsso--Paramtr Estmats Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt ag rac tratmt Odds Rato Estmats Pot 95% Wald Effct Estmat Cofdc mts ag rac tratmt Usg ths valus, th stmatd logt quato s: h Th trprtato of th paramtrs s th sam hr as for th smpl modl. Th odds rato stmat, or b, rprsts th prctag cras or dcras th probablt that a drug usr rmas drug fr moths aftr th cocluso of th tratmt program du to a o-ut chag th varabl, holdg all ls costat. That prctag chag probablt s qual to th odds rato stmat mus o. Accordg to th pot stmats for th odds ratos gv th output abov, th lklhood of a subjct rmag drug fr upo th cocluso of tratmt crass b.% for ach addtoal ar ag. Also, th odds that a o-wht prso rmas drug fr ar 5.7% gratr tha th odds of a wht prso rmag drug fr.. 7

18 Fall, a prso rcvg a log tratmt as opposd to a short tratmt would b about.5 tms mor lkl to rma drug fr as wll. Aga, bfor drawg a coclusos from th stmatd logstc rgrsso modl, th complt modl as wll as th dvdual coffcts should b tstd for sgfcac. Frst, usg th dvac prvousl dscrbd, th sgfcac of th modl ca b tstd. Th hpothss of trst ar: : 3 I ordr to tst ths, w wat to comput G whr A : at last oof th abov ' s G Dfor th modl cotag, or th rducd modl - Dfor th full modl lklhood of th rducd Modl lklhood l l lklhood of th saturatd Modl lklhood of of th full Modl th saturatd Modl lklhood of th rducd Modl l lklhood of th full Modl l lklhood of th rducd Modl l lklhood of th full Modl From th SAS output, Dspla 5: Multpl ogstc Rgrsso Sgfcac of Modl Modl Ft Statstcs Itrcpt Itrcpt ad Crtro Ol Covarats AIC SC og Thus, G Notc that th valus th trcpt ol colum ar dtcal to thos valus Dspla. Bcaus ths colum s dpdt of th umbr or tp of prdctor varabls put to th 7 osmr, p. 3. 8

19 modl, ths colum wll rma th sam for whatvr st of prdctor varabls ar cludd. Th computd G-statstc s chod Dspla 6 blow, aothr porto of th SAS output that cluds th lklhood rato tst: Dspla 6: Multpl ogstc Rgrsso--Th klhood Rato Tst Tstg Global Null pothss: BETA Tst Ch-Squar DF Pr > ChSq klhood Rato Scor Wald G wll oc aga follow th ch-squar dstrbuto, but ths cas wth thr dgrs of frdom bcaus th sgfcac of thr paramtrs s bg tstd. Th sam dcso ruls appl to ths sttg as to th smpl sttg abov. Thrfor, th ull hpothss wll b rjctd f G > χ α;3 ad th ull hpothss wll ot b rjctd f G χ α;3. Wth χ.95; ad th assocatd p-valu.34, thr s ough vdc to rjct th ull hpothss ad coclud that th paramtrs ar sgfcat. Kowg that th tr modl s sgfcat dos ot guarat th sgfcac of all of th paramtrs cludd th modl. Thrfor, th paramtrs should b tstd dvduall or small groups usg thr th lklhood-rato tst or th Wald tst, alrad dscrbd, ordr to chck for thr sgfcac. I usg th lklhood rato to tst th sgfcac of a group of varabls, th G statstc would b obtad b comparg th dvac of th full modl, or that cotag all of th paramtrs, to th dvac of th rducd modl, or that wth th paramtrs qusto havg b lmatd. Th rsultg tst statstc would th b compard wth th ch-squar crtcal valu basd upo dgrs of frdom qualg th umbr of paramtrs bg tstd. Th sam dcso ruls alrad dscrbd would b usd. For ampl, th thr varabl modl ft abov, th sgfcac of th dvdual paramtrs ca b tstd usg th Wald p-valus gv th SAS output from Dspla 4. Ol o paramtr 9

20 ag wth a p-valu of.57 appars to b sgfcat at th α.5 lvl. Kowg ths formato, ag could b lmatd from th st of prdctor varabls ad th modl could b rfttd. I tstg th sgfcac of a dscrt catgorcal varabl dvdd to varous sublvls, a clar dcso ca b mad ol wh th sam tst rsult occurs for all lvls of th varabl. Othrws, a larg lmt of ucrtat accompas a cocluso. 8 Buldg th Rgrsso Modl: Thr ar two ma statstcal mthods that ar oft mplod dtrmg whch varabls to clud a modl, th stpws rgrsso mthod ad th bst subst mthod. owvr, addto to th statstcal dvlopmt of th modl, t s mportat to hav a substatal amout of kowldg about th topc bg studd. I som stacs, a varabl that mght b cosdrd vr mportat trms of practcalt would b lmatd statstcall ad would hav to b forcd to th modl. Thrfor whl ths procdurs provd a good tool for th dvlopmt of a workg modl, t s a good da to cosdr th tr of all possbl varabls basd upo a workg kowldg of th data. 9 Th frst tchqu, stpws rgrsso, bgs wth a bas modl cotag ol th trcpt paramtr. It th adds varabls sgfcat to th modl utl thr ar o rmag sgfcat varabls lft to b addd. Th c fatur of ths procdur s that at ach stp aftr a varabl has b addd to th modl, all of th varabls cludd prvous stps ar rtstd ordr to s f th ar stll sgfcat. Th cluso ad tracto of varabls from th modl th stpws procdur s basd upo th lklhood-rato tst. Normall wh dog th lklhood-rato tst, a commol accptd alpha lvl such as. or.5 s chos as th crtcal valu for th tr of varabls to th modl. For th modl buldg procss, ths 8 osmr, p osmr, p. 83.

21 cutoff valu for th tr or rmoval of a varabl should b crasd to aroud.. Ths wll hlp avodg possbl sgfcat varabls from bg ovrlookd or rmovd ucssarl from th modl. Usg th stpws procdur to buld a modl for our drug us data st, th procss ad rsults wll b dscussd wth th ad of SAS output. Th rspos varabl for ths modl wll b Y drugfr, whr Y s a dcator varabl for whthr or ot a subjct rmas drug fr for twlv moths. Th possbl prdctor varabls wll clud th complt pool of varabls avalabl, whr ag, bck, 3 IV, 4 IV, 5 prors, 6 rac, 7 tratmt, ad 8 st. Itall a modl s ft wth just th trcpt trm ad th th corrspodg lklhood,, s computd. Dspla 7: Stp -- Th trcpt s trd. Modl Covrgc Status Covrgc crtro GCONVE-8 satsfd. Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. Rsdual Ch-Squar Tst Ch-Squar DF Pr > ChSq <. SAS th fts ght dvdual rgrsso modls, ach cotag just o of th ght avalabl prdctor varabls, ad computs th lklhood, for ach of ths modls. Usg ths lklhood valus, a Scor ch-squar tst statstc s foud b mas of a farl volvd matr computato. It volvs a comparso of th full modl vrsus th rducd modl, whr th full modl cotas th paramtrs whos sgfcac ou ar tstg, whl th rducd modl dos ot clud ths paramtrs. Th dfto for th Scor statstc follows for a full modl cotag s paramtrs ad a rducd modl cotag t paramtrs: osmr, pp. 6-9.

22 ˆ ˆ ˆ U I U Scor whr t t t t t t t t t t t t I U,..., l,..., l,..., l,..., l,,..., l,...,,..., l,,...,, Th Scor statstc follows a ch-squar dstrbuto wth s - t dgrs of frdom. Ths statstc for all of th possbl varabls, as wll as th assocatd p-valus, ca b s Dspla 8: Dspla 8: Th Stpws Procdur--Aalss of th Effcts ot th Modl Scor Effct DF Ch-Squar Pr > ChSq ag bck IV IV prors rac tratmt st I gral, a hghr Scor statstc s bttr tha a lowr o. Wth prors havg th largst sgfcat tst statstc, ths varabl wll b trd to th modl. SAS outputs th paramtr stmats as wll as th lklhood valus for th w modl cotag o prdctor varabl, prors. Ths w formato appars th followg dspla. Dspla 9: Stp --Prors s trd to th modl Modl Covrgc Status Covrgc crtro GCONVE-8 satsfd. Modl Ft Statstcs Itrcpt Itrcpt ad Crtro Ol Covarats AIC SC og SAS, STAT Usr s Gud, Vrso 8, Volum Car, NC: SAS Publshg, 999, p. 948.

23 Tstg Global Null pothss: BETA Tst Ch-Squar DF Pr > ChSq klhood Rato Scor Wald Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. prors ookg at Dspla 9, t s vdt from th modl ft statstcs that l ad l, I addto, th rsults for th lklhood rato tst dcat that th w modl s sgfcat wth G.839 ad th p-valu.6. Th w paramtr stmats ar also gv, wth th Wald p-valu dcatg that th varabl prors s dd sgfcat to th modl. Bcaus th varabl prors has b trd, SAS rchcks th sgfcac of all trd paramtrs, ths cas ol th varabl prors, to mak sur that all of th cludd varabls ar stll sgfcat to th w modl. Nt, th lklhood of th sv modls cotag th trcpt, prors, ad o of th rmag varabls s amd. Aothr varabl s trd basd upo th calculato of th w Scor ch-squar statstcs. 3

24 Dspla : Aalss of Prors ad th Effcts ot th Modl Aalss of Effcts Modl Wald Effct DF Ch-Squar Pr > ChSq prors Aalss of Effcts Not th Modl Scor Effct DF Ch-Squar Pr > ChSq ag bck IV IV rac tratmt st B th formato gv Dspla, IV s ow trd to th modl, havg th largst Scor ch-squar statstc. Th rsultg formato rgardg lklhood valus, th lklhood rato tst, ad th paramtr stmats for th w modl s show Dspla. Dspla : Stp --IV s Etrd to th Modl Modl Covrgc Status Covrgc crtro GCONVE-8 satsfd. Modl Ft Statstcs Itrcpt Itrcpt ad Crtro Ol Covarats AIC SC og Tstg Global Null pothss: BETA Tst Ch-Squar DF Pr > ChSq klhood Rato 7.5. Scor Wald Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. IV prors Ths procss cotus, at ach stp addg aothr varabl ad th usg th dvac tst to mak sur that o varabls should b rmovd from th w modl. Oc SAS has fshd th stpws procdur, mag that o addtoal sgfcat varabls rma, a fal modl s outputtd. Th fal modl for our data st s show Dspla. Dspla : Th Stpws Fal Modl Effct Numbr Scor Wald Stp Etrd Rmovd DF I Ch-Squar Ch-Squar prors IV ag tratmt IV

25 Summar of Stpws Slcto Stp Pr > ChSq Varabl abl.8 Pror Drug Tratmts.4 Idcator for Rct IV Drug Us 3. Ag at Erollmt 4.96 Tratmt: Short, og 5.75 Idcator for Prvous IV Drug Us I ths ampl, th fv varabls foud to b sgfcat to th modl ar prors, IV, ag, tratmt, ad IV. ookg at th paramtr stmats ad p-valus gv for th modl cotag ths fv varabls Dspla 3, a quato ca b stmatd for th bst modl as dtrmd b th stpws procdur. Dspla 3: Stpws Fal Paramtr Estmats Aalss of Mamum klhood Estmats Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. ag IV IV prors tratmt Thus, h Aga, f othr varabls ar dmd practcall mportat to th modl, ths could b cludd as wll. Aothr popular procdur for fttg ths logstc modls s th bst subst procdur. Thr ar svral dffrt crtro usd to dtrm bst for ths computatos. SAS uss th Scor ch-squar statstc, whch was dscussd a prvous scto. I gral, whchvr modl has th hghst Scor statstc s cosdrd th bst modl. owvr, bcaus a dgr of frdom s lost for vr addtoal varabl addd to a modl, th bst modl would ot ol hav a rlatvl hgh Scor ch-squar statstc, but also a small umbr of prdctor varabls. Callg up th bst substs procdur SAS, th two bst modls for vr lvl of q s show th followg output, wth q rprstg th umbr of prdctor varabls: 5

26 Dspla 4: Bst Subst b th Scor Crtro Numbr of Scor Varabls Ch-Squar Varabls Icludd Modl prors IV IV prors prors tratmt 3.75 ag IV prors ag IV IV ag IV IV prors ag IV prors tratmt ag IV IV prors tratmt ag IV prors rac tratmt ag IV IV prors rac tratmt ag IV IV prors tratmt st ag IV IV prors rac tratmt st ag bck IV IV prors rac tratmt ag bck IV IV prors rac tratmt st ookg at th abov output, w s that th modl wth all ght varabls provds th hghst Scor statstc, as would b pctd. owvr, thr ar svral othr modls wth fwr prdctor varabls that hav Scor statstcs almost as hgh as ths modl. ookg at th substs for th modls wth fv prdctor varabls, w s that th frst modl has a Scor statstc of 3.565, ol slghtl smallr tha th mamum valu. I addto, t ol has fv varabls. Thrfor ths modl, whch s th sam modl slctd b our stpws mthod, would b a good choc for th bst subst of prdctor varabls. I ths cas, th modl dcso basd upo th bst subst mthod s quvalt to th modl dcso th stpws procdur, but t should b otd that ths procdurs wll ot alwas produc th sam rsults. Adquac of th Modl: B mas of both mthods, th strogst modl wth whch to ft th data s th followg logt modl: h Aftr fttg ths modl, a fw dagostc masurs ca b usd to look at th ft of th modl. Th stmatd modl s as follows: 6

27 Dspla 5: Fal Modl Stadard Wald Paramtr DF Estmat Error Ch-Squar Pr > ChSq Itrcpt <. ag IV IV prors tratmt Thrfor, w aga fd that Odds Rato Estmats Pot 95% Wald Effct Estmat Cofdc mts ag IV IV prors tratmt h B lookg at th odds rato stmats Dspla 5, w s that a addtoal ar ag crass th probablt of rmag drug fr b 5.4%. Rct IV drug us dcrass that probablt b 55.3% whl prvous IV drug us dcrass that probablt b 46.4%. W also ca ot that for vr addtoal pror drug tratmt to whch th subjct was posd, th probablt of rmag drug fr falls b 6.%. Fall, udrgog th log tratmt procss as opposd to th short tratmt procss crass th chacs of rmag drug fr b 57%. A graphc mag of ths ft s dpctd blow: Fgur : Fal ogstc Ft. 7

28 O wa of assssg th ft of ths modl s b mas of th osmr-mshow Goodss of Ft tst. I ordr to prform ths tst, th data must b dvdd to g groups, whr t s ormall rcommdd that g s qual to t. For all of th goodss-of-ft tsts, th hpothss ar: A : E{ Y} : E{ Y} Th assocatd tst statstc s: Cˆ, or th modl ft s approprat, or th modl ft s approprat. g [ ] k k k k k k [ k ] whr Ĉ s a ch-squar radom varabl wth g dgrs of frdom. Thrfor, th ull hpothss wll b rjctd f Cˆ > χ α; g whl th ull hpothss wll ot b rjctd ˆ f C χ α; g. SAS prforms ths goodss of ft tst usg g. Accordg to SAS, Dspla 6: osmr ad mshow Goodss of Ft Tst drugfr drugfr Group Total Obsrvd Epctd Obsrvd Epctd osmr ad mshow Goodss-of-Ft Tst Ch-Squar DF Pr > ChSq Ra, pp

29 Wth Ĉ 3.3 ad a p-valu of.935, th ull hpothss wll ot b rjctd, rsultg th cocluso that, accordg to th osmr ad mshow Goodss-of-Ft tst, th modl dos fact provd a good ft to th data. A smlar goodss-of-ft tst ca b prformd o th sam st of hpothss as abov, whr ths cas th tst would b basd upo th modl dvacs. Th tst statstc hr would b th dvac of th modl that has b ft, as dfd abov, or whr D l l ˆ ˆ b b... bp p b b... b ˆ. p p Th ull hpothss should b rjctd f DEV > χ α; p ad th ull hpothss should ot b rjctd f DEV χ α; p. 3 Gv th drug data, th dvac s lstd as whl χ.5; Modl Ft Statstcs Itrcpt Itrcpt ad Crtro Ol Covarats AIC SC og Bcaus th calculatd dvac s lss tha th ch-squard crtcal valu, oc aga th ull hpothss wll ot b rjctd, furthr supportg th adquac of th ft of ths modl to th data. Cocluso 3 Ntr, p

30 I th modlg of a bar rspos varabl, th logstc rgrsso tchqu srvs as a k lmt to a aalss. B th vr atur of th rspos varabl, th rsults of th fttd modl ca asl b trprtd trms of th probablt that th vt of trst occurs. Th modl fttg procdur of mamum lklhood stmato s farl computatoall tsv, but wth so ma statstcal programs avalabl, th rsults ca b asl computd. Ths tchqu s vr mportat for statstcal aalss, partcularl aras such as halth scc whr utl th dvlopmt of logstc rgrsso, thr was o wa of accuratl fttg a modl to th data. Wth th currt focus mdc lg toward fdg tratmts ad v curs for ma of th halth ssus plagug our soct, such as cacr, AIDS, ad v th commo cold, logstc rgrsso wll srv as a good tool for dtrmg whthr or ot ths proposd tratmts ar ffctv. Th growg dsr of th gral huma populato to ga mor kowldg mdcal ad pschologcal aras, as wll as ma othr aras, wll caus th cotud us of ths rgrsso tchqu. 3

31 SAS Cod Usd for Aalss OPTIONS S75; FIENAME data ':\Comps\drugData.dat'; DATA drugus; INFIE data; INPUT ID ag bck IV prors rac tratmt st drugfr; obsum_n_; ABE IDIdtfcato Cod agag at Erollmt bckbck Dprsso Scor IVIV Drug Us stor prorspror Drug Tratmts rac'rac: Wht, Othr' tratmt'tratmt: Short, og' st'st: A, B' drugfr'rmad Drug Fr, Othrws' ; DATA drugus; SET drugus; IF IV3 TEN IV; ESE IV; IF IV TEN IV; ESE IV; ABE IVIdcator for Rct IV Drug Us IVIdcator for Prvous IV Drug Us ; /*Smpl ogstc Rgrsso Modl*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrtratmt/lklogt; OUTPUT OUTsmpl Phat RESDEVdvRsd; RUN; PROC GPOT DATAsmpl; POT drugfr*tratmt hat*tratmt/ovrla; smbol vcrcl lo cblack; smbol vdot lspl cblack; /*Multpl ogstc Rgrsso Modl*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrag rac tratmt/ lklogt Rsquar; OUTPUT OUTmultpl Phat RESDEVdvRsd; RUN; /*Modl Buldg*/ /*Stpws*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrag bck IV IV prors rac tratmt st/ lklogt slctostpws dtals sltr. slsta.; OUTPUT OUTstpws Phat; RUN; /*Modl Buldg*/ /*Bst Subst*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrag bck IV IV prors rac tratmt st/ slctoscor dtals; OUTPUT OUTbst Phat; RUN; /*Modl Buldg*/ /*mtd Bst Subst*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrag bck IV IV prors rac tratmt st/ slctoscor bst dtals; OUTPUT OUTtopbst Phat; RUN; /*Tmporar Modl*/ PROC OGISTIC DATAdrugUs dscdg; MODE drugfrag IV IV prors tratmt/rsquar Ifluc ackft Iplots; OUTPUT OUTtmp Phat RESDEVdvRsd RESCIchRsd hat DIFDEVdvChg DIFCISQchChg DFBETAS_A_; RUN; PROC GPOT DATAtmp; POT drugfr*obsum hat*obsum/ovrla; smbol vcrcl o cblack; smbol vdot o cblack; RUN; 3

32 Sourcs Chattrj, Samprt, Al S. ad, ad Brtram Prc, Rgrsso Aalss b Eampl Nw York: Joh Wl & Sos, ogg, Robrt V., ad All T. Crag, Itroducto to Mathmatcal Statstcs odo: Th MacMlla Compa, 97 osmr, Davd W., ad Stal mshow, Appld ogstc Rgrsso Nw York: Joh Wl & Sos, 989 osmr, Davd W., ad Stal mshow, Appld ogstc Rgrsso Nw York: Joh Wl & Sos, 989 Mard, Scott, Appld ogstc Rgrsso Aalss Thousad Oaks, Calfora: Sag Publcatos, 995 Ntr, Joh, Wllam Wassrma, Chrstophr J. Nachtshm, ad Mchal. Kutr, Appld ar Rgrsso Modls, 3 rd d. Chcago: McGraw-ll, 996 Ra, Thomas P., Modr Rgrsso Mthods Nw York: Joh Wl & Sos, Ic., 997 SAS, STAT Usr s Gud, Vrso 8, Volum Car, NC: SAS Publshg, 999. SAS vrso 8. 3

Introduction to logistic regression

Introduction to logistic regression Itroducto to logstc rgrsso Gv: datast D { 2 2... } whr s a k-dmsoal vctor of ral-valud faturs or attrbuts ad s a bar class labl or targt. hus w ca sa that R k ad {0 }. For ampl f k 4 a datast of 3 data

More information

Unbalanced Panel Data Models

Unbalanced Panel Data Models Ubalacd Pal Data odls Chaptr 9 from Baltag: Ecoomtrc Aalyss of Pal Data 5 by Adrás alascs 4448 troducto balacd or complt pals: a pal data st whr data/obsrvatos ar avalabl for all crosssctoal uts th tr

More information

Introduction to logistic regression

Introduction to logistic regression Itroducto to logstc rgrsso Gv: datast D {... } whr s a k-dmsoal vctor of ral-valud faturs or attrbuts ad s a bar class labl or targt. hus w ca sa that R k ad {0 }. For ampl f k 4 a datast of 3 data pots

More information

Note on the Computation of Sample Size for Ratio Sampling

Note on the Computation of Sample Size for Ratio Sampling Not o th Computato of Sampl Sz for ato Samplg alr LMa, Ph.D., PF Forst sourcs Maagmt Uvrst of B.C. acouvr, BC, CANADA Sptmbr, 999 Backgroud ato samplg s commol usd to rduc cofdc trvals for a varabl of

More information

LECTURE 6 TRANSFORMATION OF RANDOM VARIABLES

LECTURE 6 TRANSFORMATION OF RANDOM VARIABLES LECTURE 6 TRANSFORMATION OF RANDOM VARIABLES TRANSFORMATION OF FUNCTION OF A RANDOM VARIABLE UNIVARIATE TRANSFORMATIONS TRANSFORMATION OF RANDOM VARIABLES If s a rv wth cdf F th Y=g s also a rv. If w wrt

More information

Math Tricks. Basic Probability. x k. (Combination - number of ways to group r of n objects, order not important) (a is constant, 0 < r < 1)

Math Tricks. Basic Probability. x k. (Combination - number of ways to group r of n objects, order not important) (a is constant, 0 < r < 1) Math Trcks r! Combato - umbr o was to group r o objcts, ordr ot mportat r! r! ar 0 a r a s costat, 0 < r < k k! k 0 EX E[XX-] + EX Basc Probablt 0 or d Pr[X > ] - Pr[X ] Pr[ X ] Pr[X ] - Pr[X ] Proprts

More information

Bayesian Shrinkage Estimator for the Scale Parameter of Exponential Distribution under Improper Prior Distribution

Bayesian Shrinkage Estimator for the Scale Parameter of Exponential Distribution under Improper Prior Distribution Itratoal Joural of Statstcs ad Applcatos, (3): 35-3 DOI:.593/j.statstcs.3. Baysa Shrkag Estmator for th Scal Paramtr of Expotal Dstrbuto udr Impropr Pror Dstrbuto Abbas Najm Salma *, Rada Al Sharf Dpartmt

More information

A Stochastic Approximation Iterative Least Squares Estimation Procedure

A Stochastic Approximation Iterative Least Squares Estimation Procedure Joural of Al Azhar Uvrst-Gaza Natural Sccs, 00, : 35-54 A Stochastc Appromato Itratv Last Squars Estmato Procdur Shahaz Ezald Abu- Qamar Dpartmt of Appld Statstcs Facult of Ecoomcs ad Admstrato Sccs Al-Azhar

More information

Binary Choice. Multiple Choice. LPM logit logistic regresion probit. Multinomial Logit

Binary Choice. Multiple Choice. LPM logit logistic regresion probit. Multinomial Logit (c Pogsa Porchawssul, Faculty of Ecoomcs, Chulalogor Uvrsty (c Pogsa Porchawssul, Faculty of Ecoomcs, Chulalogor Uvrsty 3 Bary Choc LPM logt logstc rgrso probt Multpl Choc Multomal Logt (c Pogsa Porchawssul,

More information

Reliability of time dependent stress-strength system for various distributions

Reliability of time dependent stress-strength system for various distributions IOS Joural of Mathmatcs (IOS-JM ISSN: 78-578. Volum 3, Issu 6 (Sp-Oct., PP -7 www.osrjourals.org lablty of tm dpdt strss-strgth systm for varous dstrbutos N.Swath, T.S.Uma Mahswar,, Dpartmt of Mathmatcs,

More information

Counting the compositions of a positive integer n using Generating Functions Start with, 1. x = 3 ), the number of compositions of 4.

Counting the compositions of a positive integer n using Generating Functions Start with, 1. x = 3 ), the number of compositions of 4. Coutg th compostos of a postv tgr usg Gratg Fuctos Start wth,... - Whr, for ampl, th co-ff of s, for o summad composto of aml,. To obta umbr of compostos of, w d th co-ff of (...) ( ) ( ) Hr for stac w

More information

Lecture 1: Empirical economic relations

Lecture 1: Empirical economic relations Ecoomcs 53 Lctur : Emprcal coomc rlatos What s coomtrcs? Ecoomtrcs s masurmt of coomc rlatos. W d to kow What s a coomc rlato? How do w masur such a rlato? Dfto: A coomc rlato s a rlato btw coomc varabls.

More information

3.4 Properties of the Stress Tensor

3.4 Properties of the Stress Tensor cto.4.4 Proprts of th trss sor.4. trss rasformato Lt th compots of th Cauchy strss tsor a coordat systm wth bas vctors b. h compots a scod coordat systm wth bas vctors j,, ar gv by th tsor trasformato

More information

On Estimation of Unknown Parameters of Exponential- Logarithmic Distribution by Censored Data

On Estimation of Unknown Parameters of Exponential- Logarithmic Distribution by Censored Data saqartvlos mcrbata rovul akadms moamb, t 9, #2, 2015 BULLETIN OF THE GEORGIAN NATIONAL ACADEMY OF SCIENCES, vol 9, o 2, 2015 Mathmatcs O Estmato of Ukow Paramtrs of Epotal- Logarthmc Dstrbuto by Csord

More information

HANDY REFERENCE SHEET HRP/STATS 261, Discrete Data

HANDY REFERENCE SHEET HRP/STATS 261, Discrete Data Bary prdctor Bary outcom HANDY REFERENCE SHEE HRP/SAS 6, Dscrt Data x Cotgcy abls Dsas (D No Dsas (~D Exposd (E a b Uxposd (~E c d Masurs of Assocato a /( a + b Rs Rato = c /( c + d RR * xp a /( a+ b c

More information

A COMPARISON OF SEVERAL TESTS FOR EQUALITY OF COEFFICIENTS IN QUADRATIC REGRESSION MODELS UNDER HETEROSCEDASTICITY

A COMPARISON OF SEVERAL TESTS FOR EQUALITY OF COEFFICIENTS IN QUADRATIC REGRESSION MODELS UNDER HETEROSCEDASTICITY Colloquum Bomtrcum 44 04 09 7 COMPISON OF SEVEL ESS FO EQULIY OF COEFFICIENS IN QUDIC EGESSION MODELS UNDE HEEOSCEDSICIY Małgorzata Szczpa Dorota Domagała Dpartmt of ppld Mathmatcs ad Computr Scc Uvrsty

More information

MODEL QUESTION. Statistics (Theory) (New Syllabus) dx OR, If M is the mode of a discrete probability distribution with mass function f

MODEL QUESTION. Statistics (Theory) (New Syllabus) dx OR, If M is the mode of a discrete probability distribution with mass function f MODEL QUESTION Statstcs (Thory) (Nw Syllabus) GROUP A d θ. ) Wrt dow th rsult of ( ) ) d OR, If M s th mod of a dscrt robablty dstrbuto wth mass fucto f th f().. at M. d d ( θ ) θ θ OR, f() mamum valu

More information

Logistic Regression I. HRP 261 2/10/ am

Logistic Regression I. HRP 261 2/10/ am Logstc Rgrsson I HRP 26 2/0/03 0- am Outln Introducton/rvw Th smplst logstc rgrsson from a 2x2 tabl llustrats how th math works Stp-by-stp xampls to b contnud nxt tm Dummy varabls Confoundng and ntracton

More information

Numerical Method: Finite difference scheme

Numerical Method: Finite difference scheme Numrcal Mthod: Ft dffrc schm Taylor s srs f(x 3 f(x f '(x f ''(x f '''(x...(1! 3! f(x 3 f(x f '(x f ''(x f '''(x...(! 3! whr > 0 from (1, f(x f(x f '(x R Droppg R, f(x f(x f '(x Forward dffrcg O ( x from

More information

Second Handout: The Measurement of Income Inequality: Basic Concepts

Second Handout: The Measurement of Income Inequality: Basic Concepts Scod Hadout: Th Masurmt of Icom Iqualty: Basc Cocpts O th ormatv approach to qualty masurmt ad th cocpt of "qually dstrbutd quvalt lvl of com" Suppos that that thr ar oly two dvduals socty, Rachl ad Mart

More information

Suzan Mahmoud Mohammed Faculty of science, Helwan University

Suzan Mahmoud Mohammed Faculty of science, Helwan University Europa Joural of Statstcs ad Probablty Vol.3, No., pp.4-37, Ju 015 Publshd by Europa Ctr for Rsarch Trag ad Dvlopmt UK (www.ajourals.org ESTIMATION OF PARAMETERS OF THE MARSHALL-OLKIN WEIBULL DISTRIBUTION

More information

Notation for Mixed Models for Finite Populations

Notation for Mixed Models for Finite Populations 30- otato for d odl for Ft Populato Smpl Populato Ut ad Rpo,..., Ut Labl for,..., Epctd Rpo (ovr rplcatd maurmt for,..., Rgro varabl (Luz r for,...,,,..., p Aular varabl for ut (Wu z μ for,...,,,..., p

More information

Ordinary Least Squares at advanced level

Ordinary Least Squares at advanced level Ordary Last Squars at advacd lvl. Rvw of th two-varat cas wth algbra OLS s th fudamtal tchqu for lar rgrssos. You should by ow b awar of th two-varat cas ad th usual drvatos. I ths txt w ar gog to rvw

More information

The R Package PK for Basic Pharmacokinetics

The R Package PK for Basic Pharmacokinetics Wolfsggr, h R Pacag PK St 6 h R Pacag PK for Basc Pharmacotcs Mart J. Wolfsggr Dpartmt of Bostatstcs, Baxtr AG, Va, Austra Addrss of th author: Mart J. Wolfsggr Dpartmt of Bostatstcs Baxtr AG Wagramr Straß

More information

IAEA-CN-184/61 Y. GOTO, T. KATO, K.NIDAIRA. Nuclear Material Control Center, Tokai-mura Japan.

IAEA-CN-184/61 Y. GOTO, T. KATO, K.NIDAIRA. Nuclear Material Control Center, Tokai-mura Japan. IAEA-CN-84/6 Establshmt of accurat calbrato curv for atoal vrfcato at a larg scal ut accoutablt tak RRP - For strgthg stat sstm for mtg safguards oblgato. GOO. KAO K.NIDAIRA Nuclar Matral Cotrol Ctr oka-mura

More information

APPENDIX: STATISTICAL TOOLS

APPENDIX: STATISTICAL TOOLS I. Nots o radom samplig Why do you d to sampl radomly? APPENDI: STATISTICAL TOOLS I ordr to masur som valu o a populatio of orgaisms, you usually caot masur all orgaisms, so you sampl a subst of th populatio.

More information

Channel Capacity Course - Information Theory - Tetsuo Asano and Tad matsumoto {t-asano,

Channel Capacity Course - Information Theory - Tetsuo Asano and Tad matsumoto   {t-asano, School of Iformato Scc Chal Capacty 009 - Cours - Iformato Thory - Ttsuo Asao ad Tad matsumoto Emal: {t-asao matumoto}@jast.ac.jp Japa Advacd Isttut of Scc ad Tchology Asahda - Nom Ishkawa 93-9 Japa http://www.jast.ac.jp

More information

Correlation in tree The (ferromagnetic) Ising model

Correlation in tree The (ferromagnetic) Ising model 5/3/00 :\ jh\slf\nots.oc\7 Chaptr 7 Blf propagato corrlato tr Corrlato tr Th (frromagtc) Isg mol Th Isg mol s a graphcal mol or par ws raom Markov fl cosstg of a urct graph wth varabls assocat wth th vrtcs.

More information

Chapter 6 Student Lecture Notes 6-1

Chapter 6 Student Lecture Notes 6-1 Chaptr 6 Studnt Lctur Nots 6-1 Chaptr Goals QM353: Busnss Statstcs Chaptr 6 Goodnss-of-Ft Tsts and Contngncy Analyss Aftr compltng ths chaptr, you should b abl to: Us th ch-squar goodnss-of-ft tst to dtrmn

More information

Machine Learning. Principle Component Analysis. Prof. Dr. Volker Sperschneider

Machine Learning. Principle Component Analysis. Prof. Dr. Volker Sperschneider Mach Larg Prcpl Compot Aalyss Prof. Dr. Volkr Sprschdr AG Maschlls Lr ud Natürlchsprachlch Systm Isttut für Iformatk chsch Fakultät Albrt-Ludgs-Uvrstät Frburg sprschdr@formatk.u-frburg.d I. Archtctur II.

More information

Consistency of the Maximum Likelihood Estimator in Logistic Regression Model: A Different Approach

Consistency of the Maximum Likelihood Estimator in Logistic Regression Model: A Different Approach ISSN 168-8 Joural of Statstcs Volum 16, 9,. 1-11 Cosstcy of th Mamum Lklhood Estmator Logstc Rgrsso Modl: A Dffrt Aroach Abstract Mamuur Rashd 1 ad Nama Shfa hs artcl vstgats th cosstcy of mamum lklhood

More information

Estimation Theory. Chapter 4

Estimation Theory. Chapter 4 Estmato ory aptr 4 LIEAR MOELS W - I matrx form Estmat slop B ad trcpt A,,.. - WG W B A l fttg Rcall W W W B A W ~ calld vctor I gral, ormal or Gaussa ata obsrvato paramtr Ma, ovarac KOW p matrx to b stmatd,

More information

Total Prime Graph. Abstract: We introduce a new type of labeling known as Total Prime Labeling. Graphs which admit a Total Prime labeling are

Total Prime Graph. Abstract: We introduce a new type of labeling known as Total Prime Labeling. Graphs which admit a Total Prime labeling are Itratoal Joural Of Computatoal Egrg Rsarch (crol.com) Vol. Issu. 5 Total Prm Graph M.Rav (a) Ramasubramaa 1, R.Kala 1 Dpt.of Mathmatcs, Sr Shakth Isttut of Egrg & Tchology, Combator 641 06. Dpt. of Mathmatcs,

More information

Entropy Equation for a Control Volume

Entropy Equation for a Control Volume Fudamtals of Thrmodyamcs Chaptr 7 Etropy Equato for a Cotrol Volum Prof. Syoug Jog Thrmodyamcs I MEE2022-02 Thrmal Egrg Lab. 2 Q ds Srr T Q S2 S1 1 Q S S2 S1 Srr T t t T t S S s m 1 2 t S S s m tt S S

More information

COMPLEX NUMBERS AND ELEMENTARY FUNCTIONS OF COMPLEX VARIABLES

COMPLEX NUMBERS AND ELEMENTARY FUNCTIONS OF COMPLEX VARIABLES COMPLEX NUMBERS AND ELEMENTARY FUNCTIONS OF COMPLEX VARIABLES DEFINITION OF A COMPLEX NUMBER: A umbr of th form, whr = (, ad & ar ral umbrs s calld a compl umbr Th ral umbr, s calld ral part of whl s calld

More information

Weights Interpreting W and lnw What is β? Some Endnotes = n!ω if we neglect the zero point energy then ( )

Weights Interpreting W and lnw What is β? Some Endnotes = n!ω if we neglect the zero point energy then ( ) Sprg Ch 35: Statstcal chacs ad Chcal Ktcs Wghts... 9 Itrprtg W ad lw... 3 What s?... 33 Lt s loo at... 34 So Edots... 35 Chaptr 3: Fudatal Prcpls of Stat ch fro a spl odl (drvato of oltza dstrbuto, also

More information

signal amplification; design of digital logic; memory circuits

signal amplification; design of digital logic; memory circuits hatr Th lctroc dvc that s caabl of currt ad voltag amlfcato, or ga, cojucto wth othr crcut lmts, s th trasstor, whch s a thr-trmal dvc. Th dvlomt of th slco trasstor by Bard, Bratta, ad chockly at Bll

More information

Estimation of Population Variance Using a Generalized Double Sampling Estimator

Estimation of Population Variance Using a Generalized Double Sampling Estimator r Laka Joural o Appl tatstcs Vol 5-3 stmato o Populato Varac Us a Gralz Doubl ampl stmator Push Msra * a R. Kara h Dpartmt o tatstcs D.A.V.P.G. Coll Dhrau- 8 Uttarakha Ia. Dpartmt o tatstcs Luckow Uvrst

More information

Chapter 6. pn-junction diode: I-V characteristics

Chapter 6. pn-junction diode: I-V characteristics Chatr 6. -jucto dod: -V charactrstcs Tocs: stady stat rsos of th jucto dod udr ald d.c. voltag. ucto udr bas qualtatv dscusso dal dod quato Dvatos from th dal dod Charg-cotrol aroach Prof. Yo-S M Elctroc

More information

Tolerance Interval for Exponentiated Exponential Distribution Based on Grouped Data

Tolerance Interval for Exponentiated Exponential Distribution Based on Grouped Data Itratoal Rfrd Joural of Egrg ad Scc (IRJES) ISSN (Ol) 319-183X, (Prt) 319-181 Volum, Issu 10 (Octobr 013), PP. 6-30 Tolrac Itrval for Expotatd Expotal Dstrbuto Basd o Groupd Data C. S. Kaad 1, D. T. Shr

More information

Analyzing Frequencies

Analyzing Frequencies Frquncy (# ndvduals) Frquncy (# ndvduals) /3/16 H o : No dffrnc n obsrvd sz frquncs and that prdctd by growth modl How would you analyz ths data? 15 Obsrvd Numbr 15 Expctd Numbr from growth modl 1 1 5

More information

Lecture Notes Types of economic variables

Lecture Notes Types of economic variables Lecture Notes 3 1. Types of ecoomc varables () Cotuous varable takes o a cotuum the sample space, such as all pots o a le or all real umbers Example: GDP, Polluto cocetrato, etc. () Dscrete varables fte

More information

Independent Domination in Line Graphs

Independent Domination in Line Graphs Itratoal Joural of Sctfc & Egrg Rsarch Volum 3 Issu 6 Ju-1 1 ISSN 9-5518 Iddt Domato L Grahs M H Muddbhal ad D Basavarajaa Abstract - For ay grah G th l grah L G H s th trscto grah Thus th vrtcs of LG

More information

More Statistics tutorial at 1. Introduction to mathematical Statistics

More Statistics tutorial at   1. Introduction to mathematical Statistics Mor Sttstcs tutorl t wwwdumblttldoctorcom Itroducto to mthmtcl Sttstcs Fl Soluto A Gllup survy portrys US trprurs s " th mvrcks, drmrs, d lors whos rough dgs d ucompromsg d to do t thr ow wy st thm shrp

More information

ST 524 NCSU - Fall 2008 One way Analysis of variance Variances not homogeneous

ST 524 NCSU - Fall 2008 One way Analysis of variance Variances not homogeneous ST 54 NCSU - Fall 008 On way Analyss of varanc Varancs not homognous On way Analyss of varanc Exampl (Yandll, 997) A plant scntst masurd th concntraton of a partcular vrus n plant sap usng ELISA (nzym-lnkd

More information

Estimating the Variance in a Simulation Study of Balanced Two Stage Predictors of Realized Random Cluster Means Ed Stanek

Estimating the Variance in a Simulation Study of Balanced Two Stage Predictors of Realized Random Cluster Means Ed Stanek Etatg th Varac a Sulato Study of Balacd Two Stag Prdctor of Ralzd Rado Clutr Ma Ed Stak Itroducto W dcrb a pla to tat th varac copot a ulato tudy N ( µ µ W df th varac of th clutr paratr a ug th N ulatd

More information

ON ESTIMATION OF STRESS STRENGTH MODEL FOR GENERALIZED INVERTED EXPONENTIAL DISTRIBUTION

ON ESTIMATION OF STRESS STRENGTH MODEL FOR GENERALIZED INVERTED EXPONENTIAL DISTRIBUTION Joural of Rlablt ad Statstcal Studs; ISSN Prt: 974-84, Ol:9-5666 Vol. 6, Issu 3: 55-63 ON ESTIMATION OF STRESS STRENGTH MODEL FOR GENERALIZED INVERTED EXPONENTIAL DISTRIBUTION Mohad A. Hussa Dpartt of

More information

Repeated Trials: We perform both experiments. Our space now is: Hence: We now can define a Cartesian Product Space.

Repeated Trials: We perform both experiments. Our space now is: Hence: We now can define a Cartesian Product Space. Rpatd Trals: As w hav lood at t, th thory of probablty dals wth outcoms of sgl xprmts. I th applcatos o s usually trstd two or mor xprmts or rpatd prformac or th sam xprmt. I ordr to aalyz such problms

More information

Irregular Boundary Area Computation. by Quantic Hermite Polynomial

Irregular Boundary Area Computation. by Quantic Hermite Polynomial It. J. Cotmp. Mat. Sccs, Vol. 6,, o., - Irrgular Boudar Ara Computato b Quatc Hrmt Polomal J. Karwa Hama Faraj, H.-S. Faradu Kadr ad A. Jamal Muamad Uvrst of Sulama-Collg of Scc Dpartmt of Matmatcs, Sualma,

More information

Almost all Cayley Graphs Are Hamiltonian

Almost all Cayley Graphs Are Hamiltonian Acta Mathmatca Sca, Nw Srs 199, Vol1, No, pp 151 155 Almost all Cayly Graphs Ar Hamltoa Mg Jxag & Huag Qogxag Abstract It has b cocturd that thr s a hamltoa cycl vry ft coctd Cayly graph I spt of th dffculty

More information

Department of Mathematics and Statistics Indian Institute of Technology Kanpur MSO202A/MSO202 Assignment 3 Solutions Introduction To Complex Analysis

Department of Mathematics and Statistics Indian Institute of Technology Kanpur MSO202A/MSO202 Assignment 3 Solutions Introduction To Complex Analysis Dpartmt of Mathmatcs ad Statstcs Ida Isttut of Tchology Kapur MSOA/MSO Assgmt 3 Solutos Itroducto To omplx Aalyss Th problms markd (T) d a xplct dscusso th tutoral class. Othr problms ar for hacd practc..

More information

Chapter (8) Estimation and Confedence Intervals Examples

Chapter (8) Estimation and Confedence Intervals Examples Chaptr (8) Estimatio ad Cofdc Itrvals Exampls Typs of stimatio: i. Poit stimatio: Exampl (1): Cosidr th sampl obsrvatios, 17,3,5,1,18,6,16,10 8 X i i1 17 3 5 118 6 16 10 116 X 14.5 8 8 8 14.5 is a poit

More information

BAYESIAN ANALYSIS OF THE SIMPLE LINEAR REGRESSION WITH MEASUREMENT ERRORS

BAYESIAN ANALYSIS OF THE SIMPLE LINEAR REGRESSION WITH MEASUREMENT ERRORS BAYESIAN ANALYSIS OF THE SIMPLE LINEAR REGRESSION WITH MEASUREMENT ERRORS Marta Yuk BABA Frado Atoo MOALA ABSTRACT: Usually th classcal approach to mak frc lar rgrsso modl assums that th dpdt varabl dos

More information

The real E-k diagram of Si is more complicated (indirect semiconductor). The bottom of E C and top of E V appear for different values of k.

The real E-k diagram of Si is more complicated (indirect semiconductor). The bottom of E C and top of E V appear for different values of k. Modr Smcoductor Dvcs for Itgratd rcuts haptr. lctros ad Hols Smcoductors or a bad ctrd at k=0, th -k rlatoshp ar th mmum s usually parabolc: m = k * m* d / dk d / dk gatv gatv ffctv mass Wdr small d /

More information

CHAPTER: 3 INVERSE EXPONENTIAL DISTRIBUTION: DIFFERENT METHOD OF ESTIMATIONS

CHAPTER: 3 INVERSE EXPONENTIAL DISTRIBUTION: DIFFERENT METHOD OF ESTIMATIONS CHAPTER: 3 INVERSE EXPONENTIAL DISTRIBUTION: DIFFERENT METHOD OF ESTIMATIONS 3. INTRODUCTION Th Ivrs Expoal dsrbuo was roducd by Kllr ad Kamah (98) ad has b sudd ad dscussd as a lfm modl. If a radom varabl

More information

Integral points on hyperbolas over Z: A special case

Integral points on hyperbolas over Z: A special case Itgral pots o hprbolas ovr Z: A spcal cas `Pag of 7 Kostat Zlator Dpartmt of Mathmatcs ad Computr Scc Rhod Islad Collg 600 Mout Plasat Avu Provdc, R.I. 0908-99, U.S.A. -mal addrss: ) Kzlator@rc.du ) Kostat_zlator@ahoo.com

More information

Today s logistic regression topics. Lecture 15: Effect modification, and confounding in logistic regression. Variables. Example

Today s logistic regression topics. Lecture 15: Effect modification, and confounding in logistic regression. Variables. Example Today s stc rgrsson tocs Lctur 15: Effct modfcaton, and confoundng n stc rgrsson Sandy Eckl sckl@jhsh.du 16 May 28 Includng catgorcal rdctor crat dummy/ndcator varabls just lk for lnar rgrsson Comarng

More information

10/7/14. Mixture Models. Comp 135 Introduction to Machine Learning and Data Mining. Maximum likelihood estimation. Mixture of Normals in 1D

10/7/14. Mixture Models. Comp 135 Introduction to Machine Learning and Data Mining. Maximum likelihood estimation. Mixture of Normals in 1D Comp 35 Introducton to Machn Larnng and Data Mnng Fall 204 rofssor: Ron Khardon Mxtur Modls Motvatd by soft k-mans w dvlopd a gnratv modl for clustrng. Assum thr ar k clustrs Clustrs ar not rqurd to hav

More information

Review - Probabilistic Classification

Review - Probabilistic Classification Mmoral Unvrsty of wfoundland Pattrn Rcognton Lctur 8 May 5, 6 http://www.ngr.mun.ca/~charlsr Offc Hours: Tusdays Thursdays 8:3-9:3 PM E- (untl furthr notc) Gvn lablld sampls { ɛc,,,..., } {. Estmat Rvw

More information

ME 501A Seminar in Engineering Analysis Page 1

ME 501A Seminar in Engineering Analysis Page 1 St Ssts o Ordar Drtal Equatos Novbr 7 St Ssts o Ordar Drtal Equatos Larr Cartto Mcacal Er 5A Sar Er Aalss Novbr 7 Outl Mr Rsults Rvw last class Stablt o urcal solutos Stp sz varato or rror cotrol Multstp

More information

Chapter 14 Logistic Regression Models

Chapter 14 Logistic Regression Models Chapter 4 Logstc Regresso Models I the lear regresso model X β + ε, there are two types of varables explaatory varables X, X,, X k ad study varable y These varables ca be measured o a cotuous scale as

More information

Chiang Mai J. Sci. 2014; 41(2) 457 ( 2) ( ) ( ) forms a simply periodic Proof. Let q be a positive integer. Since

Chiang Mai J. Sci. 2014; 41(2) 457 ( 2) ( ) ( ) forms a simply periodic Proof. Let q be a positive integer. Since 56 Chag Ma J Sc 0; () Chag Ma J Sc 0; () : 56-6 http://pgscccmuacth/joural/ Cotrbutd Papr Th Padova Sucs Ft Groups Sat Taș* ad Erdal Karaduma Dpartmt of Mathmatcs, Faculty of Scc, Atatürk Uvrsty, 50 Erzurum,

More information

On Three-Way Unbalance Nested Analysis of Variance

On Three-Way Unbalance Nested Analysis of Variance Joural o Mathmatcs ad Statstcs 8 : -4 0 ISS 549-3644 0 Scc Publcatos O Thr-Wa Ubalac std alss o Varac Smala S. Sa ad ug. Uagbu Dpartmt o Statstcs Facult o Phscal Sccs Uvrst o gra sua gra bstract: Problm

More information

On the Possible Coding Principles of DNA & I Ching

On the Possible Coding Principles of DNA & I Ching Sctfc GOD Joural May 015 Volum 6 Issu 4 pp. 161-166 Hu, H. & Wu, M., O th Possbl Codg Prcpls of DNA & I Chg 161 O th Possbl Codg Prcpls of DNA & I Chg Hupg Hu * & Maox Wu Rvw Artcl ABSTRACT I ths rvw artcl,

More information

Bayesian Test for Lifetime Performance Index of Ailamujia Distribution Under Squared Error Loss Function

Bayesian Test for Lifetime Performance Index of Ailamujia Distribution Under Squared Error Loss Function Pur ad Appld Mathmatcs Joural 6; 5(6): 8-85 http://www.sccpublshggroup.com/j/pamj do:.648/j.pamj.656. ISSN: 36-979 (Prt); ISSN: 36-98 (Ol) Baysa Tst for ftm Prformac Idx of Alamuja Dstrbuto Udr Squard

More information

A Measure of Inaccuracy between Two Fuzzy Sets

A Measure of Inaccuracy between Two Fuzzy Sets LGRN DEMY OF SENES YERNETS ND NFORMTON TEHNOLOGES Volum No 2 Sofa 20 Masur of accuracy btw Two Fuzzy Sts Rajkumar Vrma hu Dv Sharma Dpartmt of Mathmatcs Jayp sttut of formato Tchoy (Dmd vrsty) Noda (.P.)

More information

Outlier-tolerant parameter estimation

Outlier-tolerant parameter estimation Outlr-tolrant paramtr stmaton Baysan thods n physcs statstcs machn larnng and sgnal procssng (SS 003 Frdrch Fraundorfr fraunfr@cg.tu-graz.ac.at Computr Graphcs and Vson Graz Unvrsty of Tchnology Outln

More information

Complex Numbers. Prepared by: Prof. Sunil Department of Mathematics NIT Hamirpur (HP)

Complex Numbers. Prepared by: Prof. Sunil Department of Mathematics NIT Hamirpur (HP) th Topc Compl Nmbrs Hyprbolc fctos ad Ivrs hyprbolc fctos, Rlato btw hyprbolc ad crclar fctos, Formla of hyprbolc fctos, Ivrs hyprbolc fctos Prpard by: Prof Sl Dpartmt of Mathmatcs NIT Hamrpr (HP) Hyprbolc

More information

Solution of Assignment #2

Solution of Assignment #2 olution of Assignmnt #2 Instructor: Alirza imchi Qustion #: For simplicity, assum that th distribution function of T is continuous. Th distribution function of R is: F R ( r = P( R r = P( log ( T r = P(log

More information

8-node quadrilateral element. Numerical integration

8-node quadrilateral element. Numerical integration Fnt Elmnt Mthod lctur nots _nod quadrlatral lmnt Pag of 0 -nod quadrlatral lmnt. Numrcal ntgraton h tchnqu usd for th formulaton of th lnar trangl can b formall tndd to construct quadrlatral lmnts as wll

More information

Aotomorphic Functions And Fermat s Last Theorem(4)

Aotomorphic Functions And Fermat s Last Theorem(4) otomorphc Fuctos d Frmat s Last Thorm(4) Chu-Xua Jag P. O. Box 94 Bg 00854 P. R. Cha agchuxua@sohu.com bsract 67 Frmat wrot: It s mpossbl to sparat a cub to two cubs or a bquadrat to two bquadrats or gral

More information

Worksheet: Taylor Series, Lagrange Error Bound ilearnmath.net

Worksheet: Taylor Series, Lagrange Error Bound ilearnmath.net Taylor s Thorm & Lagrag Error Bouds Actual Error This is th ral amout o rror, ot th rror boud (worst cas scario). It is th dirc btw th actual () ad th polyomial. Stps:. Plug -valu ito () to gt a valu.

More information

( ) = ( ) ( ) Chapter 13 Asymptotic Theory and Stochastic Regressors. Stochastic regressors model

( ) = ( ) ( ) Chapter 13 Asymptotic Theory and Stochastic Regressors. Stochastic regressors model Chapter 3 Asmptotc Theor ad Stochastc Regressors The ature of eplaator varable s assumed to be o-stochastc or fed repeated samples a regresso aalss Such a assumpto s approprate for those epermets whch

More information

Power Spectrum Estimation of Stochastic Stationary Signals

Power Spectrum Estimation of Stochastic Stationary Signals ag of 6 or Spctru stato of Stochastc Statoary Sgas Lt s cosr a obsrvato of a stochastc procss (). Ay obsrvato s a ft rcor of th ra procss. Thrfor, ca say:

More information

Applied Statistics II - Categorical Data Analysis Data analysis using Genstat - Exercise 2 Logistic regression

Applied Statistics II - Categorical Data Analysis Data analysis using Genstat - Exercise 2 Logistic regression Applid Statistics II - Catgorical Data Analysis Data analysis using Gnstat - Exrcis 2 Logistic rgrssion Analysis 2. Logistic rgrssion for a 2 x k tabl. Th tabl blow shows th numbr of aphids aliv and dad

More information

Dr. Junchao Xia Center of Biophysics and Computational Biology. Fall /21/2016 1/23

Dr. Junchao Xia Center of Biophysics and Computational Biology. Fall /21/2016 1/23 BIO53 Bosascs Lcur 04: Cral Lm Thorm ad Thr Dsrbuos Drvd from h Normal Dsrbuo Dr. Juchao a Cr of Bophyscs ad Compuaoal Bology Fall 06 906 3 Iroduco I hs lcur w wll alk abou ma cocps as lsd blow, pcd valu

More information

Course 10 Shading. 1. Basic Concepts: Radiance: the light energy. Light Source:

Course 10 Shading. 1. Basic Concepts: Radiance: the light energy. Light Source: Cour 0 Shadg Cour 0 Shadg. Bac Coct: Lght Sourc: adac: th lght rg radatd from a ut ara of lght ourc or urfac a ut old agl. Sold agl: $ # r f lght ourc a ot ourc th ut ara omttd abov dfto. llumato: lght

More information

Odd Generalized Exponential Flexible Weibull Extension Distribution

Odd Generalized Exponential Flexible Weibull Extension Distribution Odd Gralzd Epotal Flbl Wbull Etso Dstrbuto Abdlfattah Mustafa Mathmatcs Dpartmt Faculty of Scc Masoura Uvrsty Masoura Egypt abdlfatah mustafa@yahoo.com Bh S. El-Dsouy Mathmatcs Dpartmt Faculty of Scc Masoura

More information

MODELING TRIVARIATE CORRELATED BINARY DATA

MODELING TRIVARIATE CORRELATED BINARY DATA Al Azhar Bult of S Vol.6 No. Dmbr - 5. MODELING TRIVARIATE CORRELATED BINAR DATA Ahmd Mohamd Mohamd El-Sad Dartmt of Hgh Isttut for Sf Studs Maagmt Iformato Sstms Nazlt Al-Batra Gza Egt. ABSTRACT Ths ar

More information

Chapter 5 Special Discrete Distributions. Wen-Guey Tzeng Computer Science Department National Chiao University

Chapter 5 Special Discrete Distributions. Wen-Guey Tzeng Computer Science Department National Chiao University Chatr 5 Scal Dscrt Dstrbutos W-Guy Tzg Comutr Scc Dartmt Natoal Chao Uvrsty Why study scal radom varabls Thy aar frqutly thory, alcatos, statstcs, scc, grg, fac, tc. For aml, Th umbr of customrs a rod

More information

Chapter 4 NUMERICAL METHODS FOR SOLVING BOUNDARY-VALUE PROBLEMS

Chapter 4 NUMERICAL METHODS FOR SOLVING BOUNDARY-VALUE PROBLEMS Chaptr 4 NUMERICL METHODS FOR SOLVING BOUNDRY-VLUE PROBLEMS 00 4. Varatoal formulato two-msoal magtostatcs Lt th followg magtostatc bouar-valu problm b cosr ( ) J (4..) 0 alog ΓD (4..) 0 alog ΓN (4..)

More information

Search sequence databases 3 10/25/2016

Search sequence databases 3 10/25/2016 Sarch squnc databass 3 10/25/2016 Etrm valu distribution Ø Suppos X is a random variabl with probability dnsity function p(, w sampl a larg numbr S of indpndnt valus of X from this distribution for an

More information

On the Beta Mekaham Distribution and Its Applications. Chukwu A. U., Ogunde A. A. *

On the Beta Mekaham Distribution and Its Applications. Chukwu A. U., Ogunde A. A. * Amrca Joural of Mathmatcs ad Statstcs 25, 5(3: 37-43 DOI:.5923/j.ajms.2553.5 O th Bta Mkaham Dstruto ad Its Applcatos Chukwu A. U., Ogud A. A. * Dpartmt of Statstcs, Uvrsty Of Iada, Dpartmt of Mathmatcs

More information

Mathematical Statistics. Chapter VIII Sampling Distributions and the Central Limit Theorem

Mathematical Statistics. Chapter VIII Sampling Distributions and the Central Limit Theorem Mahmacal ascs 8 Chapr VIII amplg Dsrbos ad h Cral Lm Thorm Fcos of radom arabls ar sall of rs sascal applcao Cosdr a s of obsrabl radom arabls L For ampl sppos h arabls ar a radom sampl of s from a poplao

More information

Soft k-means Clustering. Comp 135 Machine Learning Computer Science Tufts University. Mixture Models. Mixture of Normals in 1D

Soft k-means Clustering. Comp 135 Machine Learning Computer Science Tufts University. Mixture Models. Mixture of Normals in 1D Comp 35 Machn Larnng Computr Scnc Tufts Unvrsty Fall 207 Ron Khardon Th EM Algorthm Mxtur Modls Sm-Suprvsd Larnng Soft k-mans Clustrng ck k clustr cntrs : Assocat xampls wth cntrs p,j ~~ smlarty b/w cntr

More information

Objectives of Multiple Regression

Objectives of Multiple Regression Obectves of Multple Regresso Establsh the lear equato that best predcts values of a depedet varable Y usg more tha oe eplaator varable from a large set of potetal predctors {,,... k }. Fd that subset of

More information

Chapter Two. An Introduction to Regression ( )

Chapter Two. An Introduction to Regression ( ) ubject: A Itroducto to Regresso Frst tage Chapter Two A Itroducto to Regresso (018-019) 1 pg. ubject: A Itroducto to Regresso Frst tage A Itroducto to Regresso Regresso aalss s a statstcal tool for the

More information

STA 108 Applied Linear Models: Regression Analysis Spring Solution for Homework #1

STA 108 Applied Linear Models: Regression Analysis Spring Solution for Homework #1 STA 08 Appled Lear Models: Regresso Aalyss Sprg 0 Soluto for Homework #. Let Y the dollar cost per year, X the umber of vsts per year. The the mathematcal relato betwee X ad Y s: Y 300 + X. Ths s a fuctoal

More information

Almost unbiased exponential estimator for the finite population mean

Almost unbiased exponential estimator for the finite population mean Almos ubasd poal smaor for f populao ma Rajs Sg, Pakaj aua, ad rmala Saa, Scool of Sascs, DAVV, Idor (M.P., Ida (rsgsa@aoo.com Flor Smaradac ar of Dparm of Mamacs, Uvrs of Mco, Gallup, USA (smarad@um.du

More information

4. Standard Regression Model and Spatial Dependence Tests

4. Standard Regression Model and Spatial Dependence Tests 4. Stadard Regresso Model ad Spatal Depedece Tests Stadard regresso aalss fals the presece of spatal effects. I case of spatal depedeces ad/or spatal heterogeet a stadard regresso model wll be msspecfed.

More information

On Approximation Lower Bounds for TSP with Bounded Metrics

On Approximation Lower Bounds for TSP with Bounded Metrics O Approxmato Lowr Bouds for TSP wth Boudd Mtrcs Mark Karpsk Rchard Schmd Abstract W dvlop a w mthod for provg xplct approxmato lowr bouds for TSP problms wth boudd mtrcs mprovg o th bst up to ow kow bouds.

More information

T and V be the total kinetic energy and potential energy stored in the dynamic system. The Lagrangian L, can be defined by

T and V be the total kinetic energy and potential energy stored in the dynamic system. The Lagrangian L, can be defined by From MEC '05 Itrgratg Prosthtcs ad Mdc, Procdgs of th 005 MyoElctrc Cotrols/Powrd Prosthtcs Symposum, hld Frdrcto, Nw Bruswc, Caada, ugust 7-9, 005. EECROMECHNIC NYSIS OF COMPEE RM PROSHESIS (EMS) Prmary

More information

Nuclear Chemistry -- ANSWERS

Nuclear Chemistry -- ANSWERS Hoor Chstry Mr. Motro 5-6 Probl St Nuclar Chstry -- ANSWERS Clarly wrt aswrs o sparat shts. Show all work ad uts.. Wrt all th uclar quatos or th radoactv dcay srs o Urau-38 all th way to Lad-6. Th dcay

More information

Option 3. b) xe dx = and therefore the series is convergent. 12 a) Divergent b) Convergent Proof 15 For. p = 1 1so the series diverges.

Option 3. b) xe dx = and therefore the series is convergent. 12 a) Divergent b) Convergent Proof 15 For. p = 1 1so the series diverges. Optio Chaptr Ercis. Covrgs to Covrgs to Covrgs to Divrgs Covrgs to Covrgs to Divrgs 8 Divrgs Covrgs to Covrgs to Divrgs Covrgs to Covrgs to Covrgs to Covrgs to 8 Proof Covrgs to π l 8 l a b Divrgt π Divrgt

More information

Exploring Human Mobility Patterns Based on Location Information of US Flights. Bin Jiang and Tao Jia

Exploring Human Mobility Patterns Based on Location Information of US Flights. Bin Jiang and Tao Jia Eplorg Huma Moblty Pattrs Basd o Locato Iformato of US Flghts B Jag ad Tao Ja Dpartmt of Tchology ad Bult Evromt, Dvso of Gomatcs Uvrsty of Gävl, SE-80 76 Gävl, Swd Emal: b.jag@hg.s, jatao8@6.com (Draft:

More information

Washington State University

Washington State University he 3 Ktics ad Ractor Dsig Sprg, 00 Washgto Stat Uivrsity Dpartmt of hmical Egrg Richard L. Zollars Exam # You will hav o hour (60 muts) to complt this xam which cosists of four (4) problms. You may us

More information

Learning from Data with Information Theoretic Criteria II

Learning from Data with Information Theoretic Criteria II Larg from Data th Iformato Thortc Crtra II Jos C. Prcp, Ph.D. Dstgushd Profssor of Elctrcal ad Bomdcal Egrg ad BllSouth Profssor Computatoal uroegrg Laborator Uvrst of Florda http://.cl.ufl.du prcp@cl.ufl.du

More information

ERDOS-SMARANDACHE NUMBERS. Sabin Tabirca* Tatiana Tabirca**

ERDOS-SMARANDACHE NUMBERS. Sabin Tabirca* Tatiana Tabirca** ERDO-MARANDACHE NUMBER b Tbrc* Tt Tbrc** *Trslv Uvrsty of Brsov, Computr cc Dprtmt **Uvrsty of Mchstr, Computr cc Dprtmt Th strtg pot of ths rtcl s rprstd by rct work of Fch []. Bsd o two symptotc rsults

More information

Lecture Notes 2. The ability to manipulate matrices is critical in economics.

Lecture Notes 2. The ability to manipulate matrices is critical in economics. Lecture Notes. Revew of Matrces he ablt to mapulate matrces s crtcal ecoomcs.. Matr a rectagular arra of umbers, parameters, or varables placed rows ad colums. Matrces are assocated wth lear equatos. lemets

More information

Chp6. pn Junction Diode: I-V Characteristics II

Chp6. pn Junction Diode: I-V Characteristics II Ch6. Jucto od: -V Charactrstcs 147 6. 1. 3 rvato Pror 163 Hols o th quas utral -sd For covc s sak, df coordat as, - Th, d h d' ' B.C. 164 1 ) ' ( ' / qv L P qv P P P P L q d d q J '/ / 1) ( ' ' 같은방법으로

More information