1. 技术文章
  2. Problems displaying LOESS regression line and confidence interval

Problems displaying LOESS regression line and confidence interval

由 假装没事ソ 提交于 2019-12-01 12:24:30
李哲源

correct code

I searched around and read that this issue could be due to the points needing to be ordered, so I proceeded.

No, no. The ordering issue is not related to the error you see. To overcome the error, You need to replace

lines(animals$X15p5, animals.lo, col="red")

with

lines(animals$Period, animals.lo$fitted, col="red")

Here are reasons:

  1. loess returns a list of objects, not a single vector. See str(animals.lo) or names(animals.lo).
  2. why do you use animals$X15p5 as x-axis? You fit your model: X15p5 ~ Period, so x-axis should be Period.

about reordering

You need to do ordering, because by default, R lines up points in order. Take this as an example:

set.seed(0); x <- runif(100, 0, 10)  ## x is not in order
set.seed(1); y <- sqrt(x)  ## plot curve y = sqrt(x)
par(mfrow = c(1,2))
plot(x, y, type = "l")  ## this is a mess!!
reorder <- order(x)
plot(x[reorder], y[reorder], type = "l")  ## this is nice

Similarly, do:

a <- order(animals$Period)    
lines(animals$Period[a], animals.lo$fitted[a], col="red", lwd=3)

follow-up on confidence interval

Try this:

plot(X15p5 ~ Period, animals)
animals.lo <- loess(X15p5 ~ Period, animals)
pred <- predict(animals.lo, se = TRUE)
a <- order(animals$Period)
lines(animals$Period[a], pred$fit[a], col="red", lwd=3)
lines(animals$Period[a], pred$fit[a] - qt(0.975, pred$df)*pred$se[a],lty=2)
lines(animals$Period[a], pred$fit[a] - qt(0.975, pred$df)*pred$se[a],lty=2)

You forgot about reordering again. You need to reorder both fitted values, as well as standard errors.

Now, the dist ~ speed model for cars data has no need for reordering. Because:

is.unsorted(cars$speed)  ## FALSE

Yes, data are already sorted there.

Note I have made two other changes to your code:

  1. I have separated loess call and predict call; Maybe you don't need to do this, but it is generally a good habit to separate model fitting and model prediction, and keeps a copy of both objects.
  2. I have changed loess(animals$X15p5 ~ animals$Period) to loess(X15p5 ~ Period, animals). It is a bad habit to use $ sign in specifying model formula. I have another answer at https://stackoverflow.com/a/37307270/4891738 showing the draw back of such style. You can read on the "update" section over there. I have used the glm as an example, but for lm, glm, loess, things are the same.
标签
r
regression
loess
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!