Hasse Diagrams for Linear Models 2006 Professional Bowlers Association Qualifying Scores Description • 2006-7 Pro Bowlers Association Tournaments • Bowlers: 37 Bowlers Competing in all Tournaments • Oil Patterns: 5 Patterns Used (Chameleon, Cheetah, Scorpion, Shark, Viper) • Tournaments: 15 Tournaments at Different Venues Across U.S. (3 Tournaments per Oil Pattern) • Replications: 2 Sets of 7 Games/set at each tournament for each bowler • Fixed: Oil Pattern Random: Tournament, Bowler • Nested: Tournament(Oil Pattern) • Crossed: Bowler x Oil, Bowler x Tourney(Oil) • Response: Y = 7 Game Score for each Replication (in 100s) Basic Hasse Diagram M 11 B 37 36 5 4 O T 15 10 (OB )185 144 (TB)555 360 E 1110 555 Obtaining Test Denominators 1. 2. 3. 4. 5. Denominator for Factor U is “leading” random term below U No Random terms between eligible V and U 2 or more leading eligible terms approximate F-test Unrestricted Model All Random Terms below U are eligible Restricted Model All Random terms below U are eligible, EXCEPT those containing a Fixed term not in U • Unrestricted Model Interaction Effect between Fixed and Random factors changes across repetitions of experiment Restricted Model Interaction Effect between Fixed and Random factors Remains constant across repetitions • Unrestricted (Oil x Bowler) Interaction • Suppose Interaction between Bowler and Oil Pattern is not consistent across repetitions of experiment (controlling for alley, etc.). That is, bowlers do not have “consistent preferences” among Oil Patterns • Eligible Random Terms for Oil are Tourney(Oil),(Oil x Bowler),(Bowler x Tourney) since all are directly below Oil. • Eligible Random Terms for Bowler are (Oil x Bowler) and (Bowler x Tourney) since Unrestricted Model allows interaction with Fixed effect (Oil) not included in Random Effect (Bowler) • Eligible Random Term for Tourney is (Tourney x Bowler) Restricted (Oil x Bowler) Interaction • Suppose Interaction between Bowler and Oil Pattern is consistent across repetitions of experiment (controlling for alley, etc.). That is, bowlers do have “consistent preferences” among Oil Patterns • Eligible Random Terms for Oil are Tourney(Oil) & (Oil x Bowler),(Bowler x Tourney) since all are directly below Oil. • Eligible Random Term for Bowler is (Tourney x Bowler) since Restricted Model does not allow for interaction with Fixed effect (Oil) not included in Random Effect (Bowler) • Eligible Random Term for Tourney is (Tourney x Bowler) Obtaining Expected Mean Squares 1. Representative element for each random term is its Variance Component 2. Representative element for fixed terms is Q=effects2/df 3. Contribution of term = (N/#effects)*Rep element where #effects is the superscript for that term 4. E(MS) for U = sum of contributions for U and all eligible random terms below U 5. Unrestricted Model All Random Terms below U are eligible 6. Restricted Model All Random terms below U are eligible, EXCEPT those containing a Fixed term not in U Representative Elements and E(MS) Terms Model : Yijkl i j (i ) k ik kj (i ) l (ijk ) i 1,...,5; j 1,2,3; k 1,...,37; l 1,2 5 Representa tive Elements : Oil : Q o i 1 2 i 4 Bowlers : 2 Tourneys : 2 Oil Bowler : 2 Bowler Tourney : 2 Expected Mean Squares : Unre stricted Restricted Oil : 222Qo 74 2 6 2 2 2 2 222Qo 74 2 6 2 2 2 2 Bowler : 30 2 6 2 2 2 2 30 2 2 2 2 Tourney : 74 2 2 2 2 74 2 2 2 2 Oil Bowler : 6 2 2 2 2 6 2 2 2 2 Bowler Tourney : 2 2 2 2 2 2 F-Tests Oil (Unrestric ted and Restricted ) : MS O MS BT MST MS OB MS O MS BT FO 1 2 MST MS OB MS O 2 MS BT 2 MST 2 MS OB 2 df df BT df OB O dfT Bowler : MS B MS B Unrestrict ed : FB Restricted : FB MS OB MS BT 2 Tourney (Unrestric ted and Restricted ) : MST FT MS BT Oil x Bowler (Unrestric ted and Restricted ) : FOB MS OB MS BT Bowler x Tourney (Unrestric ted and Restricted ) : MS BT FBT MS E 2 Analysis of Variance (Scores Divided by 100) Yijkl i j (i ) k ik kj ( i ) l (ijk ) 5 3 37 2 Oil : SSO Y i Y 2 df O 5 1 4 i 1 j 1 k 1 l 1 5 3 37 2 Tourney(Oi l) : SST (O ) Y ij Y i i 1 j 1 k 1 l 1 5 3 37 2 Bowler : SS B Y k Y i 1 j 1 k 1 l 1 5 3 37 2 2 2 dfT ( O ) 53 1 10 df B 37 1 36 Oil Bowler : SSOB Y ik Y i Y k Y 2 df OB 4(36) 144 i 1 j 1 k 1 l 1 5 3 37 2 Bowler Tourney : SS BT Y ijk Y ij Y ik Y i 2 i 1 j 1 k 1 l 1 Y 5 Error : SS E 3 37 2 ijkl i 1 j 1 k 1 l 1 5 3 37 2 Y ijk 2 Total : SSTotal Yijkl Y i 1 j 1 k 1 l 1 df BT 36(10) 360 df E 5(3)(37)( 2 1) 555 2 dfTotal 5(3)(37)( 2) 1 1109 ANOVA and F-Tests Source Oil Tourney(Oil) Bowler Oil*Bowler Bowler*Tourney Error Total df 4 10 36 144 360 555 1109 SS 337.07 97.65 84.44 129.32 321.05 296.26 1265.80 Unrestricted MS F 84.266 7.986 9.765 10.950 2.346 2.612 0.898 1.007 0.892 1.671 0.534 P-value 0.0028 0.0000 0.0000 0.4721 0.0000 Restricted F P 2.630 0.0000 Oil (Unrestric ted and Restricted ) : MSO MS BT 4.1 MST MSOB 11.8 MS O MS BT FO 1 2 MST MS OB MS O 2 MS BT 2 MST 2 MS OB 2 df df BT df OB O dfT 2 2 Rules for Variances of Means (Fixed Factors) 1. Only Consider Main Effects and Interactions containing only Fixed Factors 2. Identify BASE TERMS and FACTORS a) Main Effects: Base Term=Base Factor b) Interactions: Base Term=Interaction, Base Factor=Main Effects 3. V(Mean) is sum over all contributing terms T of: T2 Product of superscrip ts of all base factors above T(includin g M) Superscrip t of term T 4. Unrestricted Model All random terms contribute to variance of mean of interest 5. Restricted Model All random terms contribute to variance of mean of interest except those containing fixed factor not in main term Rules for Covariances of Means (Fixed Factors) 1. Identify BASE TERMS and FACTORS 2. Determine whether subscripts agree or disagree for each base factor 3. COV(Means) is sum over all contributing terms T of: T2 Product of superscrip ts of all base factors above T(includin g M) Superscrip t of term T 4. Unrestricted Model All random terms contribute to covariance of means of interest except those below a base factor with disagreeing subscripts 5. Restricted Model Same as Unrestricted but also excludes Random terms containing Fixed Factors not in the Base factor Variances and Covariances • Fixed Factor: Oil Pattern • Base Factor: O • Variances: All Random terms contribute since there are no other fixed factors • Covariances: All Random Terms are included except those below a base factor with disagreeing subscripts (Tourney(Oil), OilxBowler, BowlerxTourney(Oil)). 5 5 1 2 5 2 5 2 V Y i 2 2 15 37 185 555 1110 1 COV Y i , Y i ' 2 37 1 2 1 2 1 2 1 V Y i Y i ' 2V Y i 2COV Y i , Y i ' 2 2 3 37 111 222 2E MST E MS OB E MS BT 222 Comparing All 10 Pairs of Oil Patterns 2E MST E MS OB E MS BT E MST E MS OB E MS BT 222 111 ^ MST MSOB MS BT 9.765 0.898 0.892 0.0880 SE Y i Y i ' 0.2967 111 111 V Y i Y i ' ^ V Y i Y i ' MST MS OB MS BT ^ 111 10.00 MST 1112 MS OB 1112 MS BT 1112 df df df BT T OB 2 ^ Bonferron' s MSD : t.05 2(10),10 SE Y i Y i ' 3.58(0.2967) 1.063 Oil Type 2 4 1 3 5 Row-Col 2 4 1 3 5 Mean 15.9045 14.35261 15.43563 15.80703 15.32086 4 5 2 0 -1.55189 -0.46887 -0.09748 -0.58365 1 3 4 1.551892 0 1.083018 1.454414 0.968243 2 1 0.468873874 -1.08301802 0 0.371396396 -0.11477477 3 0.097477 -1.45441 -0.3714 0 -0.48617 5 0.583648649 -0.96824324 0.114774775 0.486171171 0