loglinear（扬亚宁、应志良）.pdf

() Logit models LOG- — Yaning Yang , Zhiliang Ying Yaning Yang , Zhiliang Ying LOG- — () Logit models log-linear model Log-linear models ∞. (saturated model) log() = R + C + R × C log µij = λ + λx + λy + λxy i j ij λxy . ij λx = 0, λy = 0, λxy = 0, λxy = 0. i j ij ij i j λx = λy = 0 I J Yaning Yang , Zhiliang Ying LOG- — () Logit models log-linear model µij λ s , µˆij = nij (perfect fit!) IJ , 0 () H0 : λxy = 0,∀i, j ij Pearson, () χ2 , df = IJ − (I + J − 1) = (I − 1)(J − 1), df / Yaning Yang , Zhiliang Ying LOG- — () Logit models log-linear model : 2 × 2 OR: θ = π11π22 /π12π21 = µ11µ22 /µ12µ21 log θ = [λ + λx + λy + λxy ] + [λ + λx + λy + λxy ] 1 1 11 2 2 22 −[λ + λx + λy + λxy ] − [λ + λx + λy + λxy ] 1 2 12 2 1 21 = λxy + λxy − λxy − λxy 11 22 12 21 ,, log θ = 0 ⇐⇒ θ = 1 ⇐⇒ λxy = 0