In AmberlyDziesinski/ds: Files for Vanderbilt Data Science 1000

library(learnr)
knitr::opts_chunk$set(echo = FALSE)
knitr::opts_chunk$set(error = TRUE)

Practice: Conditional Means

Motivating Example

Suppose you are a high school counselor trying to predict student GPAs as a function of other student characteristics.

We will use the High School Longitudinal Data to answer these questions.

First, load the data and packages. You will need ds, tidyverse, yardstick, and the hsls_clean data.

library(ds)
library(tidyverse)
library(yardstick)
data("hsls_clean")

Unconditional Means

If the only information we had about students was their GPA, our best guess for a given student's GPA would be the school's average GPA.

Calculate the average GPA, use the mutate command to make a new variable for the prediction:

#Fill in the blanks with the functions to make a new variable that is the mean of GPA
hsls_clean<-hsls_clean%>%
  _____(mean_gpa_uncond=_____(gpa,na.rm=TRUE))

#add the ls function to see that the new variable has been created 
ls(hsls_clean)

#Solution
hsls_clean<-hsls_clean%>%
  mutate(mean_gpa_uncond=mean(gpa,na.rm=TRUE))

ls(hsls_clean)

How far off on average is the mean from a given student's actual GPA? (Note: I've already copied the code from the prior exercise, add your new code after)

hsls_clean<-hsls_clean%>%
  mutate(mean_gpa_uncond=mean(gpa,na.rm=TRUE))

#Add code here:

hsls_clean<-hsls_clean%>%
  mutate(mean_gpa_uncond=mean(gpa,na.rm=TRUE))

#Add code here:
#Calculate the RMSE

hsls_clean<-hsls_clean%>%
  mutate(mean_gpa_uncond=mean(gpa,na.rm=TRUE))

#Use the RMSE function
rmse_uncond<-hsls_clean%>%____(gpa,mean_gpa_uncond)
rmse_uncond

#Solution
hsls_clean<-hsls_clean%>%
  mutate(mean_gpa_uncond=mean(gpa,na.rm=TRUE))

rmse_uncond<-hsls_clean%>%rmse(gpa,mean_gpa_uncond)
rmse_uncond

Conditional Mean with One Predictor

We think that there may be a relationship between a student's GPA and if they have taken AP or IB classes. Maybe these students have lower GPAs because these classes are graded more rigorously. Or maybe these students have higher GPAs because students are only allowed to take AP/IB if they have a high GPA to begin with. Generate a table of the conditional mean of the outcome variable by levels of the predictor variable.

#Use the group_by, summarize, and mean functions
#GPA is the outcome, AP/IB is the predictor variable

#Use the group_by, summarize, and mean functions
#GPA is the outcome, AP/IB is the predictor variable 
hsls_clean%>%
  _____(ap_ib_credit)%>%
  _____(mean_gpa_ap=_____(gpa,na.rm=TRUE))%>%
  arrange(ap_ib_credit)

#Solution
hsls_clean%>%
  group_by(ap_ib_credit)%>%
  summarize(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  arrange(ap_ib_credit)

Is this a better predictor?

Is the conditional mean a better predictor of GPA than the unconditional mean? First, make a prediction for GPA at each level of AP/IB and add that to your dataset.

#Group variables by AP/IB credit levels and use the mutate function

#Group variables by AP/IB credit levels and use the mutate function
hsls_clean<-hsls_clean%>%
  _____(ap_ib_credit)%>%
  _____(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()
#add the list function to see that the new variable has been created 
ls(hsls_clean)

#Solution
hsls_clean<-hsls_clean%>%
  group_by(ap_ib_credit)%>%
  mutate(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()

ls(hsls_clean)

Second, compare your RMSEs. (Note: I've already copied the code from the prior exercise, add your new code after)

hsls_clean<-hsls_clean%>%
  group_by(ap_ib_credit)%>%
  mutate(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()

#Add code below

hsls_clean<-hsls_clean%>%
  group_by(ap_ib_credit)%>%
  mutate(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()

#Add code below
#Use the RMSE function

hsls_clean<-hsls_clean%>%
  group_by(ap_ib_credit)%>%
  mutate(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()

#Add code below
#Use the RMSE function
rmse_ap<-hsls_clean%>%____(gpa,mean_gpa_ap)
rmse_ap

#Solution
hsls_clean<-hsls_clean%>%
  group_by(ap_ib_credit)%>%
  mutate(mean_gpa_ap=mean(gpa,na.rm=TRUE))%>%
  ungroup()

rmse_ap<-hsls_clean%>%rmse(gpa,mean_gpa_ap)
rmse_ap

question("Is the conditional or unconditional mean a better predictor of student GPA in this case?",
  answer("conditional", correct = TRUE),
  answer("unconditional")
)