Lab 3D: Are You Sure about That?

Lab 3D - Are you sure about that?

Directions: Follow along with the slides and answer the questions in bold font in your journal.

Confidence and intervals

  • Throughout the year, we've seen that:

    – Means are used for describing the typical value in a sample or population, but we usually don't know what they are, because we can't see the entire population.

    – Means of samples can be used to estimate means of populations.

    – By including a margin of error with our estimate, we create an interval that increases
    our confidence that we've located the correct value of the population mean.

  • Today, we'll learn how we can calculate margins of error by using a method called the bootstrap.

    – Which comes from the phrase, Picking yourself up by your own bootstraps.

In this lab

  • Load the built-in atus (American Time Use Survey) data set, which is a survey of how a sample of Americans spent their day.

    The United States has an estimated population of 327,350,075. How many people were surveyed for this particular data set?

  • The statistical question we wish to investigate is:

    What is the mean age of people older than 15 living in the United States?

  • Why is it important that the ATUS is a random sample?

  • Use our atus data to calculate an estimate for the average age of people older than 15 living in the U.S.

One bootstrap

  • A bootstrapped sample is when we take a random sample() of our original data (atus) WITH replacement.

    – The size of the sample should be the same size as the original data.

  • We can create a single bootstrapped sample for the mean in 3 steps:

    `1. Sample the number of the rows to use in our bootstrap.

    `2. slice those rows from our original data into our bootstrap data.

    `3. Calculate the mean of our bootstrapped data.

Our first bootstrap

  • Fill in the blanks to sample the row numbers we'll use in our bootstrapped sample.

    – Be sure to re-read what a bootstrapped sample is from the previous slide to help you fill in the blanks.

    – Use set.seed(123) before taking the sample.

    bs_rows <- ____(1:____, size = ____, replace = ____)
  • We can use the slice function to create a new data set that includes each row from our sample

    bs_atus <- slice(atus, bs_rows)

Take a look

  • Look at the values of bs_rows and bs_atus.

    Write a paragraph that explains to someone that's not familiar with R how you created bs_rows and bs_atus. Be sure to include an explanation of what the values of bs_rows mean and how those values are used to create bs_atus. Also, be sure to explain what each argument of each function does.

One strap, two strap

  • Calculate the mean of the age variable in your bootstrapped data, then use a different value of set.seed() to create your own, personal bootstrapped sample. Then calculate its mean.

    – Compare this second bootstrapped sample with three other classmates and write a sentence about how similar or different the bootstrapped sample means were.

Many bootstraps

  • To use bootstrapped samples to create confidence intervals, we need to create many bootstrapped samples.

    – Normally, the more bootstrapped samples we use, the better the confidence interval.

    – In this lab, we'll do() 500 bootstrapped samples.

  • To make do()-ing 500 bootstraps easier, we'll code our 3-step bootstrap method into a function.

    – Open a new R script (File -> New File -> R Script) to write your function into.

Bootstrap function

  • Fill in the blank space below with the 3-steps needed to create a bootstrapped sample mean for our atus data.

    – Each step should be written on its own line between the curly braces.

    bs_func <- function() {
  • Highlight and Run the code you write.

Visualizing our bootstraps

  • Once your function is created, fill in the blanks to create 500 bootstrapped sample means:

    bs_means <- do(____) * bs_func()
  • Create a histogram for your bootstrapped samples and describe the center, shape and spread of its distribution.

    – These bootstrapped estimates no longer estimate the average age of people in the U.S.

    – Instead, they estimate how much the estimate of the average age of people in the U.S. varies.

  • In the next slide, we'll look at how we can use these bootstrapped means to create 90% confidence intervals.

Bootstrapped confidence intervals

  • To create a 90% confidence interval, we need to decide between which two ages the middle 90% of our bootstrapped estimates are contained.

  • Using your histogram, fill in the statement below:

    The lowest 5% of our estimates are below                  years and the highest 5% of our estimates are above                  years.

  • Use the quantile() function to check your estimates.

  • Based on your bootstrapped estimates, between which two ages are we 90% confident the actual mean age of people living in the U.S. is contained?

On your own

  • Using your bootstrapped sample means, create a 95% confidence interval for the mean age of people living in the U.S.

    Why is the 95% confidence interval wider than the 90% interval?

    Write down how you would explain what a 95% confidence interval means to someone not taking Introduction to Data Science.