
oi pessoal alguém já usou a funcao geeglm do geepack? estou tentando usar e apesar de nao receber um erro, algo está errado pois os clusters nao sao reconhecidos: (o banco temp tem diversos clusters na var patient que tem tamanho 1-9)
geeglm(se ~ 1+offset(log(dfu/365.25)), data = temp, family = poisson(), id = patient, corstr = "independence")
Call: geeglm(formula = se ~ 1 + offset(log(dfu/365.25)), family = poisson(), data = temp, id = patient, corstr = "independence") Coefficients: (Intercept) -1.671535 Degrees of Freedom: 100 Total (i.e. Null); 99 Residual Scale Link: identity Estimated Scale Parameters: [1] 4.447461 Correlation: Structure = independence Number of clusters: 1 Maximum cluster size: 100 Warning message: In diff(as.numeric(id)) : NAs introduced by coercion a maneira de especificar o modelo é igual ao glm e usando o glm o modelo estima corretamente:
fit2 <- glm(se ~ 1+offset(log(dfu/365.25)), data = bpal, family = poisson()) summary(fit2)
Call: glm(formula = se ~ 1 + offset(log(dfu/365.25)), family = poisson(), data = bpal) Deviance Residuals: Min 1Q Median 3Q Max -0.689 -0.688 -0.688 -0.688 13.198 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -1.44036 0.01003 -143.6 <2e-16 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Dispersion parameter for poisson family taken to be 1) Null deviance: 61241 on 45028 degrees of freedom Residual deviance: 61241 on 45028 degrees of freedom AIC: 70587 Number of Fisher Scoring iterations: 7 entretanto, eu preciso corrigir pelas medidas repetidas nos pacientes... logo, preciso usar geeglm... se alguém tiver uma dica ou souber de alguma outra forma de rodar um GEE, favor orientar. obrigada, paula.