Prezados(as) Ajustei o seguinte modelo de GEE para dados binários, longitudinais: load(https://dl.dropboxusercontent.com/u/73337918/d6.rda) attach(d6) d6$REF_P <- as.factor(d6$REF_P) d6$TCM_PRE <- as.factor(d6$TCM_PRE) d6$TCM_REL <- as.factor(d6$TCM_REL) d6$MAN_CON <- as.factor(d6$MAN_CON) d6$MAN_PER <- as.factor(d6$MAN_PER) d6$EST_CIV <- as.factor(d6$EST_CIV) d6$GR_ESC <- as.factor(d6$GR_ESC) d6$GR_OCUP <- as.factor(d6$GR_OCUP) d6$GR_ESCP <- as.factor(d6$GR_ESCP) d6$INT_PGOV <- as.factor(d6$INT_PGOV) d6$INT_PPRE <- as.factor(d6$INT_PPRE) d6$INT_PART <- as.factor(d6$INT_PART) d6$GR_DCAP <- as.factor(d6$GR_DCAP) d6$RESC_ANT <- as.factor(d6$RESC_ANT) # d6$GR_ESC <- relevel(d6$GR_ESC, ref="LE_FI") d6$REG_MES <- relevel(d6$REG_MES, ref="Ms_Mt") library(geepack) mod6c <- geeglm(RES_CONT ~ GR_ESC+GR_OCUP+SEXO+MAN_PER+REF_P+ TCM_PRE+TCM_REL+INT_PGOV+INT_PPRE+REG_MES+GR_DCAP+IDHM_PNUD+RESC_ANT,id = MUNIC, data = d6, family = binomial(link=logit), corstr = "ar1") summary(mod6c) Coefficients: Estimate Std.err Wald Pr(>|W|) (Intercept) 1.54944 0.58054 7.123 0.00761 ** GR_ESCFC_MI 0.05587 0.15362 0.132 0.71610 GR_ESCMC_SI 0.25302 0.12323 4.215 0.04006 * GR_ESCSC_PG 0.47851 0.15347 9.722 0.00182 ** GR_OCUP2 -0.01318 0.19240 0.005 0.94540 GR_OCUP3 -0.49723 0.20344 5.974 0.01452 * GR_OCUP4 0.20693 0.18915 1.197 0.27397 GR_OCUP5 0.05465 0.13705 0.159 0.69005 GR_OCUP6 -0.16555 0.20084 0.679 0.40979 GR_OCUP7 0.25329 0.22578 1.259 0.26192 GR_OCUP8 -0.56514 0.20535 7.574 0.00592 ** GR_OCUP9 0.11389 0.23092 0.243 0.62187 GR_OCUP10 -0.32457 0.16519 3.860 0.04944 * SEXOM -0.38449 0.15291 6.322 0.01192 * MAN_PER2 -0.02760 0.08483 0.106 0.74493 MAN_PER3 0.89274 0.22867 15.241 9.46e-05 *** REF_P2 -0.90280 0.16081 31.516 1.98e-08 *** TCM_PRE2 0.32531 0.17539 3.440 0.06362 . TCM_PRE3 1.10453 0.38724 8.136 0.00434 ** TCM_PRE4 0.07393 0.16962 0.190 0.66292 TCM_PRE5 0.45656 0.17748 6.618 0.01010 * TCM_PRE6 -1.07012 0.43169 6.145 0.01318 * TCM_REL2 -0.88695 0.22036 16.202 5.69e-05 *** TCM_REL3 -0.51005 0.23513 4.706 0.03006 * TCM_REL4 -0.20505 0.21699 0.893 0.34466 TCM_REL5 0.05124 0.26072 0.039 0.84418 TCM_REL6 -0.53652 0.27731 3.743 0.05303 . TCM_REL7 -0.49148 0.21464 5.243 0.02204 * TCM_REL8 0.21678 0.21973 0.973 0.32385 TCM_REL9 0.32581 0.31959 1.039 0.30799 TCM_REL10 0.59204 0.25881 5.233 0.02216 * TCM_REL11 -0.06144 0.22802 0.073 0.78759 TCM_REL12 -0.51748 0.28703 3.250 0.07140 . INT_PGOV1 -0.86442 0.37774 5.237 0.02211 * INT_PGOV2 -0.62852 0.39466 2.536 0.11126 INT_PPRE1 1.07552 0.37019 8.441 0.00367 ** INT_PPRE2 0.94990 0.43197 4.836 0.02788 * REG_MESMs_CN -0.29830 0.19545 2.329 0.12696 REG_MESMs_CS -0.25052 0.22930 1.194 0.27460 REG_MESMs_EO 0.01757 0.31662 0.003 0.95575 REG_MESMs_Nd -0.41248 0.20274 4.139 0.04190 * REG_MESMs_S -1.09423 0.22003 24.732 6.59e-07 *** REG_MESMs_VSF -0.68058 0.26449 6.621 0.01008 * GR_DCAP2 0.05559 0.15406 0.130 0.71821 GR_DCAP3 -0.02412 0.18018 0.018 0.89351 GR_DCAP4 0.21081 0.21921 0.925 0.33622 GR_DCAP5 0.62488 0.22953 7.412 0.00648 ** GR_DCAP6 0.60297 0.74007 0.664 0.41522 IDHM_PNUD -2.48469 0.77384 10.310 0.00132 ** RESC_ANT1 1.60297 0.08113 390.413 < 2e-16 *** Não consegui entender os resultados seguintes: INT_PGOV1 -0.86442 0.37774 5.237 0.02211 * INT_PGOV2 -0.62852 0.39466 2.536 0.11126 ..............................................
tapply(fitted(mod6c), INT_PGOV, mean) 0 1 2 0.6868 0.7403 0.7717 tapply(predict(mod6c), INT_PGOV, mean) 0 1 2 0.9619 1.2828 1.4899
A variável categórica INT_PGOV tem três níveis: 0, 1, e 2. O R coloca o 0 como baseline. Se o preditor linear e o valor ajustado da resposta são maiores nos níveis 1 e 2 de INT_PGOV, as estimativas dos parâmetros correspondentes a (INT_PGOV1:-0.86442) e (INT_PGOV2: -0.62852) não deviam ser positivas ????. Alguém tem alguma ideia? Desde já agradeço qualquer comentário. Att. -- Gilênio Borges Fernandes Professor Associado IV (Aposentado) Professor Adjunto A (Substituto) Universidade Federal da Bahia Instituto de Matemática Departamento de Estatística Av. Adhemar de Barros, s/n – Ondina. 40.170-110 - Salvador - BA, Brasil Tel.: (071)3283-6340/6341/6337 Fax: (071)3283-6336 URL: http://lattes.cnpq.br/6764860618464860