How to find the confidence and prediction intervals when using broom

[Rsky]

How to find the confidence and prediction intervals when using broom.

I am trying to understand prediction intervals and confidence intervals.

broom::augment() has a .sigma column, which has one value per data.
I understood this to be a confidence interval.

mod <- lm(dist ~ speed - 1,data=cars)
mod

#augment(mod)
augment(mod) %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=.fitted))+
  geom_ribbon(aes(ymin=.fitted-.sigma*1.96,ymax=.fitted+.sigma*1.96),alpha=0.7)

glance() has a sigma column, which is one whole data.
I understood that the range obtained by this is the prediction interval.

glance(mod)

augment(mod) %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=.fitted))+
  geom_ribbon(aes(ymin=.fitted- (16.3*1.96),ymax=.fitted+(16.3*1.96)),alpha=0.7)

The result of the predict("predict") function matches the figure.

predict(mod,cars,interval = "predict") %>% 
  as_tibble() %>% 
  cbind(cars) %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=fit))+
  geom_ribbon(aes(ymax=upr,ymin=lwr),color="red",alpha=0.7)
  

However, it does not match predict("confidence").
Am I making a mistake here?

predict(mod,cars,interval = "confidence") %>% 
  as_tibble() %>% 
  cbind(cars) %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=fit))+
  geom_ribbon(aes(ymax=upr,ymin=lwr),color="blue",alpha=0.7)

how to make confidence interval with predict() and broom::??

[fcas80]

Will this suffice?

mod <- lm(dist ~ speed - 1,data=cars)
predict(mod, newdata = cars, interval = "predict")
predict(mod, newdata = cars, interval = "confidence")

[nirgrahamuk]

@Rsky

broom::augment() has a .sigma column, which has one value per data.
I understood this to be a confidence interval.

the documentation says
.sigma
Estimated residual standard deviation when corresponding observation is dropped from model.
it also shows you can get a
.lower
Lower bound on interval for fitted values.
and
.upper
Upper bound on interval for fitted values.

but to do this you need to choose (just as you did with predict() the interval type you want.
i.e. you can do

#augment(mod)
augment(mod,interval = "prediction") %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=.fitted))+
  geom_ribbon(aes(ymin=.lower,ymax=.upper),alpha=0.7)

augment(mod,interval = "confidence") %>% 
  ggplot(aes(x=speed,y=dist))+
  geom_point()+
  geom_line(aes(x=speed,y=.fitted))+
  geom_ribbon(aes(ymin=.lower,ymax=.upper),alpha=0.7)