CODEWITHSUNDEEP
r
Poza
Poza

Reputation: 395

Confidence Interval length in prediction_summary()

I want to use the function prediction_summary() of the prediction package and adjust the confidence interval length. However, it seems as if the level specification does not make any difference. Am I doing something wrong here?

Reproducible example:

# Make data
DF <- data.frame(
    x = rnorm(n= 100, 20,3),
    y = rbinom(n = 100, size = 1, prob=0.4)
)

# Specify model
model <- glm(x ~ y, data = DF)

# Prediction
prediction::prediction_summary(
  model = model, 
  level = 0.2
  )

# Same result
prediction::prediction_summary(
  model = model, 
  level = 0.9
  )

Upvotes: 1

Views: 53

Answers (1)

akrun
akrun

Reputation: 887118

The issue is related to the source code

library(prediction)
prediction_summary
function (model, ..., level = 0.95) 
{
    predictions <- prediction(model, ...)
    summary(predictions, level = 0.95)
}

Note that the level is hard-coded as 0.95 in summary. We can make a correction with by modifying the summary line to take the input as level = level i.e. user specified level

body(prediction_summary)[[3]] <- quote(summary(predictions, level = level))

Now, do the test

> prediction_summary(
+   model = model, 
+   level = 0.9
+   )
 Prediction     SE     z p lower upper
      19.96 0.3178 62.81 0 19.44 20.49
> prediction_summary(
+   model = model, 
+   level = 0.2
+   )
 Prediction     SE     z p lower upper
      19.96 0.3178 62.81 0 19.88 20.04

Upvotes: 4

Related Questions