Two Factor ANOVA with 1 Unit per Treatment KNNL – Chapter 20 One Observation per Cell of 2-Way Table • When n=1, the overall sample size is nT = ab(1)=ab • For the interaction model, each observation is equal to its fitted value (each observation is its cell mean) • No problems when the additive (no interaction) model is fit • The “error sum of squares” is equivalent to interaction sum of squares for the additive model • We can test whether there is an interaction only if it takes on some specific form such as: (ab)ij = Daibj • We can test whether D=0 based on Tukey’s One Degree of Freedom for Non-Additivity Test (ODOFNA) Additive Model Model: Yij a i b j ij ^ ^ ^ Y a i Y i Y ^ ^ ^ i 1,..., a; j 1,..., b b j Y j Y ^ Y ij a i b j Y Y i Y Y j Y Y i Y j Y 2 ^ Yij Y ij Yij Y i Y j Y i 1 j 1 i 1 j 1 Error Sum of Squares is SSAB a b a b Y Y i Y j Y df MS 2 a b i 1 j 1 ij 2 SSAB Analysis of Variance: Source SS E{MS} a a A SSA b Y i Y i 1 2 a 1 MSA SSA a 1 b a i2 2 i 1 a 1 b b B j 1 a Error SSB a Y j Y b SSAB Yij Y i Y j Y i 1 j 1 a Total 2 b SSTO Yij Y i 1 j 1 SSB b 1 b 1 MSB (a 1)(b 1) MSAB 2 SSAB (a 1)(b 1) 2 a b j2 j 1 b 1 2 2 ab -1 MSA RR : FA* F 1 a ; a 1, (a 1)(b 1) MSAB MSB H 0B : b1 ... bb 0 TS : FB* RR : FB* F 1 a ; b 1, (a 1)(b 1) MSAB Testing for Main Effects: H 0A : a1 ... a a 0 TS : FA* Tukey’s Test for Additivity Assumed form of interaction: ab ij Da i b j where D is a constant Yij a i b j Da i b j ij R eplacing , a i , b j with their estimates: ^ ^ ^ Y a i Y i Y b j Y j Y Yij Y Y i Y Y j Y Yij Y i Y j Y D Y i Y Y j Y ij Yij* DX ij* ij where: Yij* Yij Y i Y j Y a ^ Fitting a regression through the origin: D X ij* Y i Y Y j Y b X ij*Yij* i 1 j 1 a b X i 1 j 1 * 2 ij Y a b i i 1 j 1 a Y Y j Y Yij Y i Y i 1 2 b Y j Y j 1 2 Sum of Squares for Interaction: 2 2 a b ^ ^ SSAB ab ij D Y i Y i 1 j 1 i 1 j 1 a b * Y 2 j Y 2 a b Y Y Y Y Y i j ij i 1 j 1 a 2 b 2 Y i Y Y j Y i 1 j 1 2 The "Remainder" Sum of Squares is: SS Re m* SSTO SSA SSB SSAB* To test for no interaction (of this form): H 0 : D 0 H A : D 0 SSAB* /1 Test Statistic: F SS Re m* (a 1)(b 1) 1 * Rejection Region: F * F 1 a ;1, (a 1)(b 1) 1