r - Predicting values and confidence intervals of predictions with the pcse package -
we ran ols regression using standard lm
function. address issues panel data rerun analysis pcse
package calculate panel corrected standard errors. got results , wanted generate graphic display predicted values , confidence intervals (as did normal lm
regression standard ses) instead got error message:
error in usemethod("predict") : no applicable method 'predict' applied object of class "pcse"
any idea how pcse
calculated se lm
object class in order predict?
you find our model , graph function below. thankful suggestions on how solve issue, is, find way come graphic displays want display
greetz
model:
m.2 <- lm(piv~inter_x1+inter_x2+x3+x1+dumx2+x4+x5, data=dataset)) summary(m.2) m.2<- pcse(lm(piv~(x3*x1)+(x3*x2)+x3+x1+x2+x4+x5, data=dataset), groupn = dataset$c1, groupt = dataset$y) pred.val <- predict(m.2, newdata=dataset_2, se.fit=true, interval=c("confidence"), level=0.9) ## error in usemethod("predict") : ## no applicable method 'predict' applied object of class "pcse"
you need along these lines adjusted standard errors of predictions (adapted http://glmm.wikidot.com/faq):
lmfit <- ... form <- formula(lmfit)[-2] ## rhs of formula designmat <- model.matrix(form,data=dataset) ## note model written more compactly ~x3*(x1+x2)+x4+x5 vv <- vcovpc(lmfit,...) pred <- designmat %*% coef(lmfit) ## or predict(lmfit,newdata=dataset) predvar <- diag(designmat %*% vv %*% t(designmat)) se <- sqrt(predvar) ## confidence intervals se2 <- sqrt(predvar+summary(lmfit)$sigma^2) ## prediction intervals qq <- qnorm((1-level)/2) interval <- pred+qq*cbind(se,-se)
a reproducible example nice, don't have time make 1 right ...
Comments
Post a Comment