靼nc regr:_一一_ j Q2. Explain the similar and dissimilar properties of pairs of multivariate methods below: (15 points) l (l) Multivariate linear regression vs. Simultaneous equations (2) Discrimination analysis vs. one-way MANOVA (3) Discrimination analysis vs. Logistic regression L,J 疝 j I:,, 改勺 乂 遜 』L Q3 . Table I shows the numbers of passengers choosing public or private transportation modes - the effects of gender, weat}i'er and age on mode across gender, weather and age. How to test choice? (20 points) 霹」#:=) K- 伊 .;;j 囧心 ck。 ia 之 駟 糾叭 严\玉 巧 Sunny Rainy Age .S 20 Age> 20 严玉] 歸 丑] fr \ !..J ..l -, ;.:,,-:: . 丶 I i I . (/ 1' T-Tr.,M tr, invP<::tia 研 t证 fr,llr,~na hvnnthP.<::P.<:: h 囧rl r.n thP. rl 而 aivP.n in T~hlP. ?·(?" nnint<:: 丶 、- / \ 1紅翌 ,` : 「' / - - --- - - (2) Public transportation {PT) supply has a unifo岡磾ct on PT ridership a~ross Taiwan -(Le., 白二 ,· , , 4 國¢ `. J) i f passengers in choosmg public and private mod Male Female Public Private Private Public 5 15 20 10 12 18 12 8 22 10 8 10 4 16 15 15 \ 笈L Q4 Cr6}) - V的la t.讠vi._ Table 1 Numb I/ 泣}}.、'『\ f, `,'· ` ;^ 洎 J 二~:,< '.,J t c . 弋 、 y . r . , 、, ~ 5 」- I~ 三 Definition Bus ridershi Variables I ;; .\ -f. r;:.. '.'. ,, . ,'.、 ; . fl . . D: ---·- .. - ·· 、 l)nit ;'.+,1 4、 w~\ 辻\ QS. Table 3 shows th9/estimation results of 竺!NOVA analysis of c,ustomer satisfaction (xi 9), likelihood of future purchase (x2 l~~and curr nt purchase level (x22忻 across~1:_1s拉~~rt刃J_e (~I, 1: less than one year; 2: I to 5 years; 3: more than 5 years) and industry type (x2, 0: wholesaler; - -1: retailer). Interpret the estimation results accordingly. (20 points) 、L /:、 Tt c 1/4 丶, 翡i...Jr.~ 砹叮,\ 1, ::;,-,, ! 0· ,\ ` ^, ) .- .'.; `、 、 ~—~.:-;;:.;; Source DF Type I SS Mean Square F Value Pr> F xi 2 30. 18262082 15.0913104 1 25.78 <.0001 x2 I 0.44146997 0.44146997 0.75 0.3874 xl*x2 2 0.33969426 0.16984713 0.29 0.7488 vK V Dependent Variable: x22 OF Sum of Squares Mean Square F Value Pr> F Source Model 5 5596.306975 1119.26 1395 Error 94 2176.693025 23.156309 Corrected Total 99 7773.000000 R-Square CoeffVar RootMSE x22 Mean 4.812100 58.40000 0.719961'- 8.239898 Source DF Type I SS Mean Square F Value Pr> F 2664.922294 115.08 <.0001 238.711401 238.711401 10.31 0.0018 27.750985 13.875492 0.60 0.5513 xl 2 5329.844589 x2 I xl*x2 2 Source DF 48.34 <.0001 vv/ Type II SS Mean Square F Value Pr>F I 11.86 <.0001 238.711401 238.711401 10.31 0.0018 27.750985 13.875492 0.60 0.5513 x2 I xl*x2 2 ✓ 2 5180.632913 VJ 2590.316457 xl Source DF Type III SS Mean Square F Value Pr> F x1 2 5092. 78952 l x2 I xl*x2 2 2546.394761 109.97 <.0001 244.294518 244.294518 10.55 0.0016 27.750985 13.875492 0.60 0.5513 3/4 J VV V .. MULTIVARIATE ANALYSIS (ITT5515) 岬) TERM EXAMINATION Why蕡龘謚鷗崮』二肆囑f這諡謚己編af江龘齿靠霄'氕严 `甲釕囯山對[ Ql. -t. 11V麩扣1國ivo.祁 一 11三严严 t1·n I;,, 回 (L ., `沼這衍勺、如;(, If it is not hold, what problems will be for the methods of MANOVA, Multivariate regression, Discrimination analysis, Logistic 7 ression, and Factor analysis, respyctively? (10 points) 選和漪 N-0, 丫园 F, I-(仞l)\Jlr 戶汔翊巴 庄 Q2. Explain the similar and dissiinilar properties of following pairs of multivariate methods: 伍 (1) Principal麟onents vs. Factor analysis 10 points) 叩打)銍攻姜 凪k飆 (2) Multivariate linear regression vs. (10 points) 辶食~ (3) Discrimination analysis vs. Logi tic regression (10 points) 扭易 Simult諡涵 equations 丰\虐 邛午 Q3. To identify the potential factors- influencing p苧匣唧9血匣卫些呾age in Taipei City, a dataset of ten-year monthly information (a total of 120 time senes observations) is collected, which contains: y: public transportation patronage percentage(%), XI: number 『雪王 召戶 (in millions), X2: average income (in thousands), X3: bus production (in million us? X4: number of registered cars and motorcycles (in thousands), XS: length of road (in thousand km). The estimated regression model is shown in Table 1. (1) Diagnose the estimated regression model and pro ose corresponding correction methods for each of problems you identify. (10 points) PJ) 羞濁莖 (2) If you want to analyze the seasonal changes in the public transportation patronage, what will you do? (off-peak seasons are Jw 癮山 Q4. 丶全 己 凸 ~~ 手 T丶 彧 邕古面 or~三辶 L 切 . f 三钇、 ·~ Q芝严莖陴皿呻sis, LQ罪瞬陴竺io~閂二~al 」子 componen~了dFac『『alys旦 (JO points) l 屯 傳 't) I 曰 紅 尸 _owing_multivari~te methods are~ MANOVA, ~ultJ.vanate regression, i \v___. 縻 靠 巧料 o:/.'.l( 邲5 必 (0 三蔻fr- 才f~ ( 科 lJ.I,.炸)( d._o,邸```丶 -e ,r.-, (j J rC 丶 5 3,,, : -1,' i 自~---:; 叱十 」 示位 · . 巴, 0 詞与 〈 •~~ :,~• ·r. ' ; 「.. L'3-`-sA` Yl. O L,{ 彞二 吟 丕 '姦 r able I Estimated results of regression an~b7 The REG Procedure Dependent Variable: y Number of Observations Read 120 Number of Observations Used 120 Analysis of Variance Sum of Mean Source DF Pr> F FValue Squares Square Model <.0001 49.32 8953.56507 1790.71301 Error 19/ 7044.43493 36.31152 Corrected Total 199 15998 RootMSE R-Squarc / 0:5~97 fi: 6.02590 Dependent Mean 58.20000 ( 0.5~?:f. CocffVar 10.35379 Parameter Estimates Parameter Standard Variance D1IIIII Variable Estimate Error t Value Pr> I~Inflation Intercept 37.48072 3.12938 11.98 <.0001 0 XI 1.16527 0.64773 1.80 X2 -0.17502 0.26244 -0.67 X3 3.00494 0.41871 7.18 < 0001 L4066-3.... VrF:> I X4 -3.94840 0.69607 -5.67 . X5 0.72109 0.09476 7.61 # <.0001~ 咋 \ 18 .98618/ 'i. /\ Durbin-Watson D 尹韶 1st Order Autocorrelatio Heteroscedasticity Test Equation Statistic DF Pr ·· q Variables Test y White's Test Pk 悔. 頃乒 怀 碼 R-Sq 勾 .07 14 rable 2 Results of ANOVAIMAONVA analyses ~I / Sum of Mean Square Squares 11274. 47222 2254.89444 24.34808 4723.52778 199 15998. 00000 Coeff Var Root MSE HC Mean 8.478312 4.934378 58.20000 DF Type I SS Mean Square F Value Pr> F <.0001 之 2 10997. 53309 5498. 76654 225. 84 1 233. 57554 233. 57554 9. 59 0.0022 、/ o.4121 2 43.36359 21.68180 0 . 89 OF Type III ss Mean Square F Value Pr> F 2 11161 • 32808 5580. 66404 229 . 20 丶 1 228. 03846 228. 03846 9. 37 2 43.36359 21.68180 0.89 F墨] :二 DF 5 194 Tota 三 鬥 t> Source Model Error Corrected Total A-Square p . 286077, Source ---ES AGE ES*AGE Source ES AGE ES*AGE -- v 二 0.704743 Source ES AGE ES*AGE Source ES · AGE ES*AGE Dependent Variable: CO 丨咒'!"三 圧严 · 科孑 - -- The GLM Procedure Class Level Information Class · Levels Values ES 3 1 2 3 AGE 2 0 1 Dependent Variable: HC Source Model Error Corrected ~ Sum of DF Squares Mean Square F Value __Er- >-f、 v 5 14.02347222 2.80469444 15.55/ <.0001 ' 194 34. 99652778 0.18039447 199 49. 02000000 Coeff Var Root MSE co Mean ' . 98.77412 · 0.424729 0.430000 DF Type I SS Mean Square F Value Pr > F 2 13.93544118 6.96772059 38.62 <.0001 1 0. 06243063 0. 06243063 0. 35 0. 5570 -;t. 2 0.02560041 0.01280021 0.07 0.9315 。 k DF Type III ss Mean Square F Value P > F 2 13.82804214 6.91402107 1 0.06416934 0.06416934 0.36 0污 沃、 2 0.02560041 0,01280021 0.07 0,9315 V 丶 I/ 38.33 汽pJ f 2 .,...., ,' Tukey's Studentized Range (HSD) Test for HC Comparisons significant at the 0.05 level are indicated by *** . Difference ES Between Simultaneous 95% Comparison Means Confidence Lim 拉 S 3 - 2 >)_:, I 8 . 3511 6.3214 10.3808 *** 3 - 1 17 . 9706 15.9719 19. 9693 *** 2 - 3 -8.3511 -6.3214 ••• -10.3808 2·1 9 . 6195 11. 6492 ••• 7 . 5898 1 - 3 -17.9706 - 19.9693 -15.9719 *** 1 - 2 -9.6195 -11 .6492 ·7. 5898 *會* Tukey's Studentized Range (HSD) Test for CO Difference ES Between Simultaneous 95% Comparison Means Confidence Limits 3 - 2 0 .15993 3 戸>->I 3·1 0 . 61765 0.. 44561 -0 01478 0. 0 .78968 33463 2 · 3 -0.15993 -0.33463 0.01478 2·1 0. 28302 0. 63243 0.45772 1 - 3 - 0.61765 -0.78968 -0.44561 -0.45772 -0.63243 -0.28302 1 - 2 5 ..:已 ..刀 J ] Multivariate Analysis of Variance ~國O勺已 o 砭L鍀 Character--:詛注-i-c-Roo-t-s-and-V.ecto r-s - of-:一 E一r-nvers·e• H, where H = Type III SSCP Matrix for S E = Error SSCP Matrix Characteristic Vector V'EV=1 Characteristic HC CO Root Percent 0 . 03088269 2.43963509 99.17 0.01372317 0. 02032487 0 . 83 -0. 00587571 0 . 17065909 三ests for the Hypothesis of No Overall ES Effect 之 H = Type III SSCP Matrix for 邸^,鬥 、 I , E -=Er繹 SSCP Matrix S=2 M= -0.5 N=95 . 5 舄 3 寸 H C: 名 0~ Co Statistic Value P-Value Wilks'Lambda O. 28493720 L籬 Pillai's Trace 0.72919148 <. 0001 Hotelling-Lawley Trace 2 . 45995996 <. 0001 Roy's Greatest Root 2. 43963509 <. 0001 Characteristic · Roots and Vectors of: E Inverse* H, where H = Type III SSCP Matrix for GE E = Error SSCP Matrix Characteristic Characteristic Vec~or Root Percent HC CO 0. 05717754 100. 00 0. 01468683 -0. 06842552 0. 00000000 0. 00 0. 00267328 0. 15936189 MANOVA Tests for the Hypothesis of No Overall AGE Effect Ri 6 t 嶧玉 卟囧 H = Type III SSCP Matrix for@ /戸 E = Error SSCP Matrix S=1 M=O N=95 . 5 0 ,- . J <Q I ;:;;;~t~:mbda o. 94~~1~891 V'EV=1 )C:;:,__~」 寸) P之繻o J Pillai ' s Trace O. 05408509 o. 0047 Hotelling - Lawley Trace 0 . 05717754 0.0047 Roy's Greatest Root O. 05717754 O. 0047 MANOVA .Tesfs .for the Hypothesis of No Ov l ' ES*AGE Effect H = Type III SSCP Matrix for Esi"~G Z, ,, 主 逗 Stat1st1c .~;';.'.':,。::" :::;~ C Value~ , Va J 菩旦:[言;-~,:;"' ~言言;;; 言差 3 J 覃~\( " 和 ) li ·