Line Graphs
We continued our analysis by using line graphs to show the change and evolution of music audio features in more detail.
Line graphs illustrate some basic patterns and general trends through summary statistics such as mean and median which are demonstrated below. All of these statistics are important in determining trends of popular songs and their audio features. For example, they can also help us determine the spread of the data suggesting the existence of many varieties of popular songs and their audio features. Moreover, by comparing different data types, we can evaluate the data more comprehensively (e.g., is mean affected by outliers?).
Rather than group the visualisations by audio features, we decided on grouping by summary statistics (for example, we displayed the mean of all the features in one graph.) Thereafter, you can compare the trends each audio feature followed to make simple observations on whether certain audio features match up in their trends. The interactive graph adjusts its scale for the audio features displayed at that moment. We recommend that you view Danceability, Energy, Acousticness and Liveness together, Speechiness and Instrumentalness together and the rest on their own because of the scales of measurement of each feature discussed in the definitions.
Additionally, to avoid presenting many of graphs on one page, we adopted an interactive graph. In this way, viewers are allowed to select the data they are interested in and see the comparison among factors.
​
The code for the line graphs can be found here.
To view the change for each audio feature, please double-click on the variable names:
The mean is a general average which music producers can take inspiration from when writing a new hit song.
The median helps identify whether the middle value has tended to be higher or lower which can help aid our interpretations of the spread of the data.
The maximum helps identify any upper limits of audio features that popular songs have tended to follow.
The minimum helps identify any lower limits of audio features that popular songs have tended to follow.
The standard deviation helps indicate the spread of the data in relation to the mean.
The variance helps indicate variability of the data.
​
Length:
Shorter songs are becoming more popular although longer popular songs do exist.
The mean peaked at just under 300 000 mili seconds in 1979 but has since undergone a general trend decreasing in size which indicates shorter songs are becoming more and more popular. Having said that, the maximum and minimum have widely varied in size with recent figures have eclipsed the 300 000 marks and the minimum has gone below half this value telling us the longer and shorter songs still have the ability to be popular. The median of the length has followed a similar trend to the mean and so has the standard deviation and the variance. This suggests that whilst longer popular songs do exist, the trend is towards shorter songs and popular songs are tending to follow this trend.
​
​
Danceability:
Danceability follows a trend of increase, reaching high levels.
​
Danceability has been on a general trend of increase as suggested by the graph of the mean and also by the previous polar charts. Notably, the value never falls below 0.5 indicating that high danceability has always been a common theme amongst popular songs throughout the years and is recently increasingly so. The maximum has has risen to above 0.9 but has not followed an overall trend but the minimum has been on a trend of increase since 2004 indicating the lower end of the spectrum are becoming more danceable which has contributed to an increase in the mean. As expected, the median has also followed a trend of increase whereas from 2005, the variance and standard deviation have both followed trends of decrease as more songs become more danceable .
​
​
Energy:
Despite a trend of decrease in the past decade, popular songs have always been energetic.
The mean of energy had overgone a general trend of increase until 2010 where it has decreased over the last 10 years returning to levels reached during the 1970s. It should be noted that like danceability, the mean has almost always been above 0.5 which implies that popular songs are relatively energetic. The maximum has tended to reach high levels consistently meaning highly energetic songs have always been part of the popular song spectrum and the minimum has been volatile but tended to be above 0.3 in the last 20 years. Therefore, few popular songs have low energy. The median is similar to the mean but the variance and standard deviation have both been quite volatile but between smaller scales hence the spread of data can be larger and smaller in different years but both measures are quite low so songs tend to be less variable in terms of energy.
​
​
​
Instrumentalness:
There is low levels of instrumentalness in popular songs.
The mean of instrumentalness was closer to 0.1 during the 1970s but since there, the mean has been close to zero with some years even reaching zero in the last 10 years. The maximum from the 1970s has reached higher levels but in the past 10 years, the value has gotten closer to 0 .1 while the minimum is constantly at 0. The median is similar to the mean. The standard deviation and variance have also tended to be closer to 0 with some outliers in the earlier years during the 1970s. It seems that popular songs have almost always had low levels of instrumentalness especially in the last decade.
​
​
​
Acousticness:
Confidence in acousticness has tended to decrease despite a recent increase.
Over the past 50 years there has been a trend of decrease in mean confidence of acousticness although the last 5 years has shown a slight trend of increase but still at low numbers. The maximum has been volatile with extremely varying figures whereas the minimum has been close to 0 since the 1980’s. The median has also been following a trend of decrease bar the last 2 years but again still low figures under 0.2. The variance and standard deviation have both been volatile suggesting high variability of data and a larger spread although with standard deviation, figures have not climbed above 0.32 thus, the spread of data has changed but is generally not far from the mean.
​
​
Liveness:
Liveness has fluctuated at consistently low levels.
​
The mean of the liveness has not followed an overall trend but has been typically between 0.1 and 0.25 which are low figures. The maximum and minimum have both followed extremely volatile trends indicating liveness in popular songs has varied from year to year. The median has been similar to the mean whereas standard deviation and variance have both also been volatile although at lower values.
​
​
Loudness:
Popular songs are continuing to become louder.
​
Since the mid-1980’s, the mean for the loudness has seen a significant increase and so has the maximum and minimum. The median has also reflected this and naturally, the variance and standard deviation have undergone the opposite trend, telling us that across the data, songs are tending to increase in the loudness and that the variation around this is small.
​
​
Speechiness:
Popular songs have low levels of speechiness.
​
Since, the 1990s, there has been an overall increase followed by an overall decrease in the mean but notably all below 0.15 which indicates a low level of speechiness in popular songs throughout the decades. The maximum has followed this trend with values never climbing above 0.5 except in 2007 where it was 0.576. The minimum has not followed a specific trend but has been notoriously low throughout the years. The standard deviation and variance have fluctuated in the past few decades but at low figures indicating a small spread of data and low variability.
​
​
​
Tempo:
Popular songs seem to have varying levels of tempo.
​
The mean of tempo has not followed a specific trend but has largely followed the moderato category of tempo (Fernández-Sotos et al., 2016). The maximum and minimum both rise and fall to extremely low levels respectively, therefore, leaving the potential for both sides of the spectrum songs to become popular songs. The median is similar to the mean but the standard deviation and variance have both fluctuated but at significant levels which means there is a large spread of data and variability. Therefore, popular songs are not necessarily moderato but they could equally be higher and lower levels consistently.
​
Time Signature:
Time Signature has remained steady at 4 beats per bar.
​
Whilst the time signature has no overall trend in the mean, it has always revolved between 3.9 and 4 except 1991 where it was at 4.05. The maximum and minimum are 3 and 5 and the median is constantly 4. The low variance and standard deviation show little variation and it is clear that popular songs have always been around the 4 beats per bar mark for time signature.
​
​
The above summaries help provide a more detailed picture of the trends audio features have followed, and the levels they have been at in popular songs in the last 50 years. These can be used as the model audio features of a popular song. However, it should be noted that songs may have different combinations of these audio features and using all figures as model numbers may be inaccurate as specific songs may have different audio features at different levels, but when averaged together give a very different value. The variance and standard deviation have helped provide a clearer picture of whether the mean may be representative of all the data. The clustering detailed in the next page will also help shed light on the interaction between different audio features.
​