Introduction

Goal: to determine which factors influence the overall danceability of a song over time

• Data comes from Spotify’s Top 100 Tracks 2001-2019
• Inspired by projects done in previous courses analyzing datasets from Spotify playlists

Statement of Problem

• Previous datasets were found on Kaggle
• Only two years worth of data was available
  • More data needed to be extracted for longitudinal analysis
• Knowledge of Python was minimal so research had to be completed to learn Python

Methods

Spotify for Developers

• Explains the process for retrieving data from Spotify
  • Authorization information
  • How to work with playlists
  • Quick start guide to the Spotify Web API
• Most examples are in Java

Spotify Web API

• Server-side application to access data from user profiles
  • Create new playlists
  • Find information about songs and artists

Methods

Spotipy

• Python library for the Spotify Web API
• Site included examples of Python code used to extract track information

Previous Uses

• To extract a list of song IDs to pull audio features for the Million Songs Dataset
• To assess what songs will play more expensive advertisements using machine learning

Coding Process

• Python Code had to be produced to extract data points from Spotify
• Spotify Web API tutorial was followed
  • Produced code where specific tracks were able to be extracted from specific artists
  • Complete playlists could not be extracted
    
• Tamer’s GitHub page became useful
  • Her code was able to extract data from entire playlists

Data Extraction Process

An app needed to be created through the Spotify Dashboard

• Gives access to Spotify Client ID and Client Secret
• Permitted access to various data points in Spotify through Python
• Provided a path to extract the data from the playlists

Data Extraction Process

• A whitelisted redirect URI had to be created with Spotify within the app for re-authentication when running the Python code
• Playlist URI’s could be inserted into the Python code and produced a .csv file with all of the data points needed for analyses

   

Variables

Outcome Variable

• Danceability - how easy it is to dance to a song, from 0 to 1

Predictor Variables

• Energy - intensity of a track, from 0 to 1
• Key - the key the song is in, from 0 to 11 (A to G#)
• Loudness - how loud a track is, in decibels
  
• Mode - indicates whether a song is in a major key or a minor key, 1 or 0
• Speechiness - the portion of the song consisting of spoken word, from 0 to 1
• Acousticness - determines if a song is acoustic or not, from 0 to 1

Variables

Predictor Variables

• Instrumentalness - detects how instrumental a song is, from 0 to 1
• Liveness - detects whether or not there is an audience present, from 0 to 1
• Valence - how uplifting a song is, from 0 to 1
• Tempo - the speed of the song, in beats per minutes
• Duration - how long the song is, in milliseconds
• Time signature - how many beats are in a measure

Previous Research

• In Biostatistics,
  • Top 100 Tracks of 2017
  • Found valance and energy to be significant predictors of danceability
• In Regression Analysis,
  • Top 100 Tracks of 2018
  • Along with valance and energy, speechiness, acousticness, and tempo were also significant predictors of danceability

Statistical Methods

• Multiple Linear Regression
  • Used for datasets with more than one predictor variable

\[\hat{Y} = \beta_0 + \beta_k{X_k} \]

• Assumptions: \[\varepsilon_i \overset{\text{iid}}{\sim} N(0, \sigma^2)\]

Statistical Methods

• Backward Selection
  • Begins with full model and strips away predictors based on some criteria
  • Must have more observations than variables

From previous research, energy, speechiness, acousticness, valence, and tempo are significant predictors of danceability.

Initial Model

The initial regression model including all variables is: \[\hat{Y} = -7.0358 + 0.0039\mbox{ year} -0.2561\mbox{ energy} + 0.0003\mbox{ key}\] \[+ 0.001\mbox{ loudness} -0.0178\mbox{ mode} + 0.1798\mbox{ speechiness} -0.1227\mbox{ acousticness}\] \[+ 0.119\mbox{ instumentalness} -0.0928\mbox{ liveness} + 0.3141\mbox{ valence} -0.0009\mbox{ tempo}\] \[ + 0\mbox{ duration} + 0.0185\mbox{ time signature}\]

Final Model

After removing the variables that are not significant predictors of danceability, the regression model is: \[\hat{Y} = -16.8717 + 0.0088\mbox{ year} -0.2426\mbox{ energy} -0.017\mbox{ mode} + 23.877\mbox{ speechiness}\] \[ -0.1224\mbox{ acousticness} + 0.1145\mbox{ instumentalness} -0.0946\mbox{ liveness} + 13.9302\mbox{ valence}\] \[ -0.0008\mbox{ tempo} -0.0118\mbox{ year} \times \mbox{ speechiness} -0.0068\mbox{ year } \times \mbox{ valence}\]

Interactions between year and speechiness and year and valence.

Testing Assumptions

Graphical Analysis

Interaction between Year and Speechiness

• As year increases, slope of speechiness has decreased
  • Relationship is weakening

Interaction between Year and Valence

• As year increases, slope of valence has decreased
  • Relationship is weakening

Conclusions

• Conducted longitudinal analysis, expanding on prior projects
• Variables used in final model:
  • Year, energy, mode, speechiness, acousticness, instrumentalness, liveness, valence, tempo, year * speechiness, year * valence
  • P-values less than 0.05
• Limitation not accounted for:
  • The instance of one artist having multiple songs in the Top 100 Tracks

Impact

The datasets created for this project have been uploaded to Kaggle for future analysts to utilize.

Suggestions for Further Study

• Look for any distinct differences of top songs in 2020
  • COVID-19, riots, the election
• Generational differences in music choices
  • Public user profiles are accessible by the user’s URI through the Spotify Web API
  • Compare the types of music listened to by different age groups