(some) Tableau Results

All of the visualization settings in this section can be downloaded as a Tableau Packaged Workbook and opened in your desktop version if you have access to the software.  To download the data package, click here (1.26 MB).  If you don't have Tableau and you're interested in exploring the data we've compiled, you can download a full free trial version of the software at: http://www.tableausoftware.com/trial.  Alternatively, you can also use Tableau Reader.

Also, please note that these represent "some" of the results achieved.  The fact of the matter is that this data can be altered in so many different ways that we couldn't find them all!  Please, explore on your own and tell us about your results via email.  Or, if you don't have access to Tableau, tell us a visualization you're interested in viewing, and we'll build it for you!

The "Discovery" Screen
We were really interested in letting users jump right into the data, so we designed this "discovery" screen.  In its initial state, the scatter plot displays each and every song that we collected data for (over 4,200) with axes measuring average tempo in BPM vs. average danceabilty.  This view also displays a number of filters which allow users to sift through the data according to their own interests.  Although the initial state could be interpreted as displaying data overload, we believe that this view actually serves to enhance the user's interaction with the data.  What starts out as a generally useless blob of data begins taking on new meanings as users manipulate the filters and discover interesting relationships among song characteristics and Billboard data.

The level of filtering sophistication one can achieve in this screen is impressive.  For example, a user could filter the data to display songs between 1963 and 1981 that were recorded in the key of C-Major, peaked above number 5, and were released on the Motown record label.  Once the view updates, users can hover over graph marks in order to gather more information about the songs, such as artist, song title, danceability and energy levels, the number of weeks it spent on the charts, and more (see figure 2).

Figure 1: The Discovery Screen

Figure 2: Example of details available upon hovering

Artist + Number of Hits
This visualization allows users to view the number of hits each artist has produced over the decades.  The length of each bar in the graph corresponds to the number of songs performed by that artist which peaked within the top 10.  We also set an average and maximum reference line, so that users can compare artists relative to the majority.

In Figure 3, we have filtered the data so that only artists who have performed 10 or more songs that have peaked in the top 10 are displayed.  As you can see, Madonna has had 36 songs that meet this criteria!  The Beatles are right behind her, with 34 songs.  Other fun facts abound!

Figure 3: Number of hits produced by artist

Danceability Over the Years
Danceability (full explanation here) is measured using a mixture of song features such as beat strength, tempo stability, and overall tempo. The value returned determines the ease with which a person could dance to a song over the course of the whole song.  We were interested to see whether danceability has changed over the decades.  As Figure 4 shows, there is a clear trend which shows that songs peaking in the top 10 are increasing in danceability.

Is this a reflection of changes within the music, though, or a reflection of what we, 21st century citizens, look for in a dance hit?  After all, these song metrics weren't developed until a few years ago.  This trend might even be a result of changes within the recording procedure and/or medium (e.g., does beat strength vary between vinyl and mp3?).

Despite these difficult questions, the visualization is still fun!

Figure 4: Average danceability by year

Song Duration Over the Years
This visualization was designed in order to see whether the average duration of songs has increased between 1960 and 2010.  The trend from the image below is quite clear that duration is increasing.

Figure 5: Average song length by year

We wanted to explore this trend a bit further, in conjunction with an increase in the average amount of time songs spend on the Hot 100 over the last 50 years (see Hit Lasting Power below).  So, we plotted average duration by year per chart peak position vs. average weeks spent on the charts.  Sure enough, this motion scatter plot shows a slow but steady migration of peaks from the middle-left side of the graph to the upper-right, meaning average duration and weeks on are increasing.

Figure 6: Average duration vs. average weeks spent on charts (motion chart)

Energy Trends
Energy (full explanation here) uses a mixture of the loudness and segment duration features in order to determine whether a song makes you want to bop all over the room or fall into a coma.  Interestingly, energy has been trending upwards, though at a very slow, almost unrecognizable rate.  Looks like the 1980s were a particularly hyper decade!  The 90s, on the other hand, were a "green" decade...relatively low energy!

Figure 7: Average energy by year

Record Label + Average Song Tempo
Do some record labels produce songs with a faster tempo than other labels?  This visualization was designed in order to answer this question.

From the filter schema in Figure 7, we can see that the Arista label has produced a relatively high number of singles in the top 10 that contain just below the average beats per minute.  In fact, most of the labels producing relatively high numbers of hits all linger around the average in this view.  Perhaps the major labels know something about song tempo that their competitors are missing.

Figure 8: Average tempo displayed by record label

Loudness (or..."Turn that racket down!")
Although we couldn't find a detailed explanation of the loudness measure, we thought it would be interesting to visualize this trend.  Clearly, songs are becoming "louder," but we are forced to ask questions about changes in production processes and media formats over the decades, yet again.  Most track decibel levels these days are normalized so that peaks and valleys in the track's waveform are largely eradicated, producing that lovely tinny quality that audiophiles and vinyl enthusiasts loath.  So it might not actually be the case that that "gosh danged rock n' roll music" is getting louder, only that volume levels are becoming stabilized over the course of an entire song and, therefore, being interpreted as "louder" by Echo Nest's computer algorithms.

Figure 9: Average loudness by year

Once again, we were interested in analyzing this trend in conjunction with another.  This time, we plotted average loudness by year per peak position vs. average energy.  As expected, graph marks display a slow, steady migration from the middle-left to the upper-right.

Figure 10: Average loudness vs. average energy (motion chart)

Tempo Trends
Have songs been speeding up, or slowing down?  Or is there a golden window of beats per minute that your song should fit through?  Enter, the tempo visualization screen!  Looks like most songs linger right around that optimal figure of 119.80 BPM.  Also, hits between 1976 and 1984 displayed rather mono-rhythmic qualities, with few year averages drifting above or below the golden mark.  Get out your metronomes!

Figure 11: Average tempo by year

Hit Lasting Power
This visualization was designed in order to see how long hits are staying on the charts over the years.  There is a clear trend which shows that songs peaking in the top 10 are definitely staying on the Hot 100 for longer periods of time.  This is not to say that number 1s are staying at number 1 for longer periods of time...only that songs peaking between 1 and 10 are tending to get somewhere into the top 100 and then stay there longer than before.  For example, our data shows that songs in 1995 stayed on the Hot 100 for over double the amount of time (average of 29 weeks) than their counterparts in 1966 (average of 12 weeks).

Findings such as these might interest media analysts, historians, even economists.  Hot 100 chart rankings are based on radio play and sales, and it's no secret that the Clear Channel Corporation started its consolidation of U.S. terrestrial radio stations in the early 1990s.  A visualization such as this clearly shows that average weeks on the charts absolutely exploded in the 90s, suggesting that there may be a connection between media consolidation and song lasting power on the Billboard Hot 100.

Also, notice the ease with which we can see that something fishy is going on in 1984.  Either 1984 songs just weren't as popular, or we've got faulty data.  Turns out the latter is true.  Using visualizations to identify faulty data is discussed further in the Conclusions.

Figure 12: Average weeks on charts per year