Repeated Measures Example In an exercise therapy study, subjects were assigned to one of three Measurements of strength (Y ) were taken on days 2, 4, 6, 8, 10, weightlifting programs 12 and 14 for each subject. (i=1) The number of repetitions of weightlifting was increased as subjects became stronger (RI) (i=2) The amount of weight was Source: Littel, Freund, and Spector (1991) SAS System for Linear Models increased as subjects became R code: RepeatedMeasures.R stronger (WI) (i=3) Subjects did not participate in weightlifting (XCont) 1 2 Mixed model Yijk = μ + αi + Sij + τk + γik + eijk Yijk strength measurement for prog. i, subj. j , time point k αi Sij τk γik eijk fixed program effect Average strength after 2k days on the i-th program is μik = E(Yijk ) = E(μ + αi + Sij + τk + γik + eijk ) = μ + αi + E(Sij ) + τk + γik + E(eijk ) random subject effect = μ + αi + τk + γik fixed time effect for i = 1, 2, 3 and k = 1, 2, . . . , 7 fixed prog.×time interaction The variance of any single observation is random error where the random effects are all independent and V ar(Yijk ) = V ar(μ + αi + Sij + τk + αik + eijk ) = V ar(Sij + eijk ) = V ar(Sij ) + V ar(eijk ) 2) Sij ∼ N ID(0, σS ijk ∼ N ID(0, σ2) 2 + σ2 = σS e 3 4 Correlation between observations taken on the same subject: Cov(Yijk , Yij) = Cov(μ + αi + Sij + τk + γik + eijk , μ + αi + Sij + τ + γi + eij) The correlation between Yijk and Yij is 2 σS 2 + σ2 σS e = Cov(Sij + eijk , Sij + eij) = Cov(Sij , Sij ) + Cov(Sij , eij) Observations +Cov(eijk , Sij ) + Cov(eijk , eij) = V ar(Sij ) ≡ρ taken on different subjects are uncorrelated. for k = 2. = σS 5 6 For the set of observations taken on a single subject, we have ⎛⎡ V ar ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎜⎢ ⎝⎣ Yij1 Yij3 .. Yij7 Write this model in the form ⎤⎞ Y = Xβ + Zu + e ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎥⎟ ⎦⎠ ⎡ 2 2 σe2 + σS σS 2 2 σS σe2 + σS = .. .. 2 2 σS σS ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ 2 ··· σS 2 ··· σS 2 ... σS 2 σ2 + σ2 σS e S ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ 2J = σe2I + σS ↑ J is a matrix of ones ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ Y111 Y112 ... Y117 Y121 Y122 ... Y127 ... Y211 Y212 ... Y217 ... Y3,n31 Y3,n32 ... Y3,n37 ⎤ ⎡ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ = 1 1 0 1 1 0 ... 1 1 0 1 1 0 1 1 0 ... 1 1 0 ... 1 0 1 1 0 1 ... 1 0 1 ... 1 0 0 1 0 0 ... ... 1 0 0 0 1000000 | 0 0100000 | ... | 0 0000001 | 0 1000000 | 0 0100000 | ... | 0 0000001 | | 0 1000000 | 0 0100000 | ... | 0 0000001 | | 1 1000000 | 1 0100000 | ... | 1 0000001 | ⎤ 21 columns for interaction This covariance structure is called compound symmetry. 7 8 ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ μ α1 α2 α3 τ1 τ2 τ3 τ4 τ5 τ6 τ7 γ11 γ12 ... γ37 ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ In this case: ⎡ 1 1 ... 1 0 0 ... + 0 ... ... ... 0 0 ... 0 ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ 0 0 ... 0 1 1 ... 1 ... ... ... 0 0 ... 0 ··· 0 ··· 0 ... ··· 0 ··· 0 ··· 0 ... ··· 0 ... ... ... ··· 1 ··· 1 ... ··· 1 ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ e111 e112 ... e117 e121 e122 ⎤ ... S11 ⎥⎥ ⎥ e127 S12 ⎥⎥⎥ ... ... ⎥⎥⎥ ... ⎥⎥⎥⎥ + e 211 ... ⎥⎥⎥⎥ e 212 ⎥ ⎦ ... S3,n3 e217 ... e3,n31 e3,n32 ... e3,n37 R = V ar(e) = σe2I(7r)×(7r) 2I G = V ar(u) = σS r×r where r is the number of subjects Σ = V ar(Y) = ZGZ T + R is a block diagonal matrix with one block of the form 2J (σe2I7×7 + σS 7×7) for each subject 9 Mean strength at a 10 Program means particular time (averaging across time) in a particular program LSM EAN = μ̂ + α̂i + τ̂k + γ̂ik = Ȳi.k = V ar(Ȳi.k ) = SȲ i.k = ⎛ ⎜ ⎜ ⎜ ⎝ 1 ni ni j=1 = μ̂ + α̂i + Yijk V ar(Ȳi...) = 2 σe2 + σS ni ⎞ 1 ⎟ 1 M Serror + M SSubj ⎟⎟⎠ 7 7 ni 6 LSM EAN = Ȳi.. ↑ Cochran-Satterthwaite degrees of freedom are 65.8 11 SȲ i.. = 1 7 (τ̂k + γ̂ik ) 7 k=1 2 σe2 + 7σS 7ni M Ssubjects 2ni There are n1 = 16, n2 = 21, n3 = 20 subjects in the three programs. 12 Mean strength at a particular time point Difference between strength (averaging across programs) means at two time points LSM EAN = μ̂ + τ̂k + = 1 3 (α̂i + γ̂ik ) 3 i=1 1 3 (Ȳi.k ) 3 i=1 = (averaging across programs) Ȳ..k ⎛ ⎝ because n1 = 16, n2 = 21, n3 = 20. V ar(LSM EAN ) = SLSM EAN = 3 ⎝ 7 1 ⎞⎛ M Serror + M Ssubj 7 3 ⎟⎜ ⎟⎜ ⎠⎝ ⎛ ⎞ ⎞ ⎞ 1 ⎛⎜⎜ 1 ⎛⎜⎜ 3 3 ⎟ ⎟ = ⎝ Ȳi.k ⎠ − ⎝ Ȳi.⎟⎟⎠ 3 i=1 3 i=1 1 3 1 n i = (Yijk − Yij) i=1 j=1 3 ni 2 2 1 3 σe + σS 9 i=1 ni ⎛ 1 ⎜ 6 ⎜ ⎞ τ̂k + 13 3i=1 γ̂ik ⎠ − ⎝τ̂ − 13 3i=1 γ̂i⎠ 1 i=1 ni ↑ subject effects cancel out ⎞ ⎟ ⎟ ⎠ ↑ Cochran-Satterthwaite degrees of freedom are 65.8 13 14 Difference between strength means for two programs at a specific time point Variance formula: 2σe2 3 1 9 Σi=1 ni (μ̂ + α̂i + τ̂ik + γ̂ik ) − (μ̂ + α̂ + τ̂k + γ̂k ) = Ȳi.k − Ȳ.k Standard error: 2M Serror Σ3 1 = 0.206 i=1 ni 9 2 )( V ar(Ȳi.k − Ȳ.k ) = (σe2 + σS ↑ use degrees of freedom for error = 324 SȲ −Ȳ = i.k .k ⎛ ⎜ ⎜ ⎝ 6 1 ni + 1 n ⎞⎛ ⎞ 1 1 ⎟⎟ ⎟⎜ 1 M Serror + M Ssubj ⎟⎠ ⎜⎝ + ⎠ 7 7 ni n ↑ Use Cochran-Satterthwaite degrees of freedom = 65.8 15 ) 16 Difference between strength Difference in strength means at two time points means for two programs within a particular program (averaging across time points) (μ̂ + α̂i + τ̂k + γ̂ik ) − (μ̂ + α̂i + τ̂ + γ̂i) Ȳi.. − Ȳ.. = (α̂i − = Ȳi.k − Ȳi. V ar(Ȳi.k − Ȳi.) = ⎛ ⎞ n i V ar ⎜⎝ n1 Σj=1 (Yijk − Yij)⎟⎠ = i ⎛ 1 1 7 7 γ̂ik ) − (α̂ + γ̂k ) 7 k=1 7 k=1 2 )( V ar(Ȳi.. − Ȳ..) = (σe2 + 7σS 2σe2 ni ⎞ SȲ −Ȳ = M Serror ⎜⎝ n2 ⎟⎠ i.k i. i = SȲ −Ȳ i.. .. ↑ use degrees of freedom for error=324 1 7ni M Ssubjects( + 1 7ni 1 7n + ) 1 7n ) ↑ 54 d.f. 17 18 Specifying other covariance This model was expressed in the matrices form We began with the model Y = Xβ + Zu + e Yijk = μ + αi + Sij + τk + γik + eijk where where ⎡ ⎤ S11 ⎥⎥⎥ .. ⎥⎥⎥ u= ⎥ ⎥ S3,20 ⎦ contained the random subject effects 2) Sij ∼ N ID(0, σS eijk ∼ N ID(0, σe2) ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ and the {Sij } are distributed independently of the {eijk } 19 20 If you are not interested in predict- Here ing subject effects (random subject 2I G = V ar(u) = σS effects are included only to introduce correlation among repeated R = V ar(e) = σe2I measures on the same subject), you can work with an alternative expres- Σ = V ar(Y) = ZGZ T + R ⎡ 2 = σS ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ = 2J σe2I + σS ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ J sion of the same model ⎤ J ... J ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ Y = Xβ + e∗ + σe2I where ⎤ ... 2J σe2I + σS R = V ar(e∗) ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎡ = where J is a matrix of ones. ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ 2J σe2I + σS ⎤ ... 2J σe2I + σS 21 22 Replace the mixed model Y = Xβ + Zu + e First Order Autoregressive: AR(1) ⎡ with the model Y = Xβ + e∗ W = σ2 where ⎡ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ W 0 ··· 0 0 W ··· 0 V ar(Y) = V ar(e∗) = ............ 0 0 ··· W 23 ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ρ ρ2 ρ3 ρ 1 ρ ρ2 ρ2 ρ 1 ρ ρ3 ρ2 ρ 1 1 ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ where σ 2 ∈ (0, ∞) and ρ ∈ (−1, 1) are unknown parameters. 24 ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦