Overview

The questions and exercises below all include examples using a dataset put together in the Tottenham Lab of many Billboard Hot 100 songs from 1946-1983. Today in this challenge you’ll want to work to answer several questions:

1. Did pop and rock songs on the Billboard charts tend to get louder on averaage from 1946-1983?
2. From the 1946-1983 Billboard charts, which were louder on average – pop or rock songs?
3. Did changes in average loudness from 1946-1983 differ for pop and rock songs?

• To answer these questions, please use a combination of Bayesian regression models with rstanarm::stan_glm(), and plotting with ggplot()
• If anything is confusing or you get stuck, please use the previous lesson plans and challenge answer keys as a reference, or reach out to an instructor!
• Remember, we do not expect you to memorize syntax, nor is this a crucial part of coding or stats. Use google, stack overflow, documentation, previous lessons and challenges, colleagues, and instructors as a reference – this is what ‘real-life’ coding is like!

0. Setup

You don’t have to write code in this section, but it will be helpful to read through to understand the data a little better.

Load the needed packages (install if necessary and ASK AN INSTRUCTOR if you’re getting stuck here

library(tidyverse)
library(car)
library(rstanarm)
library(tidybayes)
library(bayesplot)

Now, we can load in the data from Github using read_csv()

music = read_csv('https://raw.githubusercontent.com/pab2163/danl_code_workshop/main/data/familiar_music_database.csv')

## Parsed with column specification:
## cols(
##   .default = col_double(),
##   title = col_character(),
##   artist = col_character(),
##   id = col_character(),
##   uri = col_character(),
##   genre = col_character(),
##   cover = col_logical(),
##   start_time = col_time(format = "")
## )

## See spec(...) for full column specifications.

names(music)

##  [1] "year"             "title"            "artist"           "id"              
##  [5] "uri"              "rank"             "genre"            "cover"           
##  [9] "danceability"     "energy"           "key"              "loudness"        
## [13] "speechiness"      "acousticness"     "instrumentalness" "liveness"        
## [17] "valence"          "tempo"            "mode"             "time_signature"  
## [21] "duration_ms"      "start_time"

We have a lot of columns here we could look at, but for today’s challenge just focus on genre, loudnesss, and year. We’ll also only look at the Pop and Rock genres for today

# select just the relevant columns
music = dplyr::select(music, title, artist, year, genre, loudness) %>%
  # filter the data to only include Pop/Rock genre
  dplyr::filter(genre %in% c('Pop', 'Rock')) %>%
  # scale/center loudness
  dplyr::mutate(loudness = as.vector(scale(loudness, center = TRUE, scale = TRUE)))

head(music)

## # A tibble: 6 x 5
##   title                                         artist       year genre loudness
##   <chr>                                         <chr>       <dbl> <chr>    <dbl>
## 1 I Can't Begin To Tell You - Single Version    Bing Crosby  1946 Pop     -2.96 
## 2 Let It Snow! Let It Snow! Let It Snow!        Vaughn Mon…  1946 Pop      1.24 
## 3 Doctor, Lawyer, Indian Chief                  Betty Hutt…  1946 Pop     -0.461
## 4 Personality                                   Johnny Mer…  1946 Pop     -1.68 
## 5 Oh! What It Seemed to Be                      Frankie Ca…  1946 Pop     -1.13 
## 6 Prisoner of Love (with Russ Case & His Orche… Perry Como   1946 Pop     -0.495

1. Did pop and rock songs on the Billboard charts tend to get louder on averaage from 1946-1983?

• Build a Bayesian regression model using rstanarm::stan_glm() with 1 continuous predictor to understand relationships betweenyear and loudness in the music dataset!
• Interpret your model parameters: what can you conclude about average changes in loudness between 1946-1983? how strong is your evidence?
• Graph your model predictions using tidybayes and ggplot()
  • Bonus: graph model predictions and the raw data on the same plot!

# here's 1 open code chunk for this question, but feel free to make more!

2. From the 1946-1983 Billboard charts, which were louder on average – pop or rock songs?

• Build a Bayesian regression model using rstanarm::stan_glm() with 1 categorical predictor to understand relationships between genre and loudness in the music dataset!
• Interpret your model parameters: what can you conclude about average difference in loudness between rock versus pop songs? how strong is your evidence?
• Graph your model predictions using tidybayes and ggplot()
  • Bonus: graph model predictions and the raw data on the same plot!

# here's 1 open code chunk for this question, but feel free to make more!

3. Did changes in average loudness from 1946-1983 differ for pop and rock songs?

• Build a Bayesian regression model using rstanarm::stan_glm() with 1 categorical predictor, 1 continuous predictor, and an interaction term between them, to understand loudness as a function of genre, year, and their interaction.
• Interpret your model parameters: how do changes in loudness over time from 1946-1983 differ for pop versus rock songs? how strong is your evidence?
  • And – how do the results you are seeing here add more nuance to your interpretations from parts 1 & 2.
• Graph your model predictions using tidybayes and ggplot()
  • Bonus: graph model predictions and the raw data on the same plot!

# here's 1 open code chunk for this question, but feel free to make more!

4. Wrap-up

Congrats! You’ve finished the challenge!

• Before you move on, we recommend that you take 5 minutes to go back and comment your code – as a gift to yourself when you come back later to run similar analyses! Commenting and documenting right away is almost always going to ultimately save you time and energy.
• For any concepts or pieces of code that don’t make sense yet, write down your questions! Be sure to either ask a friend or an instructor right away, or ask in slack so you can learn.