PART 2: Quantitative History of a Top Dog Single
PART 2: DO NUMBER ONES LAST LONGER ONCE THEY REACH THE TOP?
An analysis of Longest Number Ones of All Time
READING TIME: 20 MINUTES
INTRODUCTION
1.1 Background on the Project and Billboard Methodology
In Part 1 of this three part series, the three titles which held the longest number one spot on the Billboard Hot 100 charts were analysed to see what kept them atop for so long. Part 1 was looking at snapshot moments in the chart publication’s history. This next part will be a mini analysis looking at patterns and trends that have formed by collecting data on the longest running number ones every year for the entirety of the chart publication’s history. In August 1958 Billboard merged jukebox plays, singles sales and radio airplay into one metric system rather than keeping them as separate entities, forming the basis for the Billboard Hot 100 we know today. Increasing consensus is that mainstream musicianship has been lost and the general public are sick of being exposed to the same songs being stuck on repeat or hogging the top of charts and that there is a lack of success for genre diversity. This has been an ongoing discussion ever since music fans on social media have been able to more easily discuss the threats of the long hit single reigns in the 2010s that Mariah Carey and Boyz II Men’s One Sweet Day had to withstand against, including Blurred Lines, Uptown Funk, Shape of You, Despacito and finally Old Town Road. Hence I decided to set up an investigation to see if these opinions have any justifications. These observations can seem subjective therefore an objective way of looking at this is by using quantitative data which can be measured objectively.
Billboard publishes its year-end chart statistics on a December-to-December basis because the company aims to publish content summarising the year’s music before the end of the year, or the decade, in order to keep its audience’s interest at bay. However, the below analysis will be looked at from the perspective of a January-to-January calendar. This means that regardless of if a single hits number one in the first week or the last week in the year, that number one is categorised to belonging to the year it reached the top spot. For example, ’No One’ hit number one on December 1st 2007. Billboard would list this song under their 2008 calendar, whereas I have listed it under the 2007 calendar.
RESEARCH METHODS
2.1 Data Collection
The data for this investigation was collected by going through Billboard Magazine archives available on Google Scholar. For every year, the longest, second, third, fourth and fifth longest numbers one were recorded and tabulated into a spreadsheet. The latter two variables were only used in limited analyses.
2.2 Statistics
The data underwent various basic statistical analyses such as the mean, median and mode to see if there were any trends. Then a smoothed line chart with each variable was created to produce a graphical visualisation of the data from 1958 until 2019. After the long-term trend chart was created, a series of regression analyses were conducted decade-by-decade to find out if there were any relationships between the data points and to see if any relationships were being formed between different decades. Correlation co-efficients were then calculated to measure the strength of the data points for every decade.
2.3 Hypothesis Testing
Then a hypothesis test was conducted by using the first, second and third longest number ones' overall trend data to see if one data group was more statistically different than the others for every decade. This would help determine which data sets were more or less influential on influencing the overall chart trend data per decade. The most appropriate hypothesis test was a Kruskall Wallis test because it met all the test conditions from the collated data. It is a non-parametric test, alternative to the one-way ANOVA, which is usually used to compare three separate variables (the first, second and third longest number ones). Then because it is a non-parametric test it does not assume that the data comes from a particular distribution and as the data was discovered to be skewed in different directions and the variances varied, it was more fitting to use this test. An alternative option could have been to have transformed the data values into normalised ones and use ANOVA instead. Other assumptions include that the observations recorded are independent of each other and each independent variable is measured with two or more other independent variables. So when using a hypothesis test when assumption for ANOVA are not met, mostly the assumption of normality (which is a spread occurring naturally in a population) the Kruskall Wallis test was the best fitting test to apply.
The null hypothesis (H0) for each set of data was that the medians/populations are all equal or distributed the same way. The alternative hypothesis (H1) was that at least one of the data groups were distributed differently or were unequal from the others, and this was determined by a statistical significance level set at P < 0.05.
RESULTS
3.1 Overall Trend Data and the Influence of Data Sets
Figure 1. A line chart showing the longest (blue), second longest (red) and third longest (red) reigning number one singles every year. The symbols (pink/purple dots and red cross) are when the record for the longest running number one was broken at the time of its release: Debby Boone You Light Up My Life (1977), Boyz II Men End of the Road (1992), Whitney Houston I Will Always Love You (1992) and then the three singles mentioned in Part 1. The dotted line at 1991 is when Billboard incorporated Nielsen Soundscan.
Figure 1 shows three distinct trends that occurred through Billboard Hot 100 number one history. The first is that from 1958 and through the 1960s, reigns at number one were getting shorter year-by-year shown by a negative, decreasing relationship. However, by the end of the decade reigns at the top were increasing shown by an upward trend. Then there is a short sudden drop in the early 1970s. The second trend begins throughout the mid-to-late 1970s and into the late 1980s. In the early 80s top song reigns were increasingly longer and the first, second and third longest reigns began becoming closer together than before. There was a descending trend after the peak in the early 80s and into the mid-to-late 80s. After the introduction of Nielsen Soundscan song reigns drastically rose mostly seen by the predominantly long reigns of the longest running number one song of each year. After reaching a peak point in 1995, the 1990s followed a decreasing trend of reigns. Then moving into the 2000s there appears to be no overall trend or relationship, shown by the trend lines oscillating with great magnitude year upon year. However, in the 2010s the trend lines show a more clear relationship showing an increasing trend of reigns at number one. There is overall oscillating year by year relationship where every two years the longest number one single would increase by 2 weeks until the longest running number one single in 2019 was achieved.
Figure 2. A line chart showing the top five longest reigning number one singles every year (year) and the average reign length using the top 3 longest reigning number ones (purple) and then the top 5 longest reigning number ones (dotted blue).
The purpose of Figure 2 is to see to what extent do some of the reigns influence the ‘average’ trend line. While the two trend lines follow an almost identical pattern, the purple line, is much higher than the dotted line. The dotted line which is affected by the fourth and fifth longest reigns, which can be about 3 to 5 weeks long, pull down each year’s data point. Using these average trend lines there is some variation in the length of reign firstly in the mid 70s and the early 80s. Rod Stewart’s Tonight’s the Night and Debby Boone’s You Light Up My Life’s 9 and 10 week stints, respectively, while were outliers in the 1970s as they pull up the purple line, were early indicators that reigns were beginning to get longer moving into the 1980s. After introducing Nielsen Soundscan, the average data point is pulled up in the purple line significantly particularly the extreme number one reigns shown in 1992, 1995, 2009 and the latter half of the 2010s.
3.2 Decade Analysis
Figure 3. Scatter graphs showing the data points of the number one reigns every decade. Longest (blue), second (red), third (yellow).
Figure 3 shows various scatter graphs to show the correlation and strength of the reigns at number ones for every decade. By appearance the 1960s was relatively neutral with no determinable relationship formed between the data points. The 1970s then formed a weak, positive correlation with hit songs staying longer at the top of the charts as the decade progressed most noticeably in 1976 and 1977. The data points are closest together in the 1980s with a strong, negative correlation formed throughout the decade most noticeably around 1981. The disparity between the decades is more apparent in the 1990s with the longest rulers distancing themselves off from the runners up and the second and third longest hit songs forming no overall relationship. Moving into the 2000s there is a weak but oscillating relationship every couple of years. Finally the 2010s show a weak, positive correlation with the longest running number one song of the year heavily influencing the trend line and even the third longest running one number one single data set is much higher than those in the previous five decades seeing that all their single reigns lasted longer than five weeks.
3.3 Statistical Analysis
Figure 4. A table showing the mean, median and mode for the longest, second longest and third longest reigns at number one every decade.
It can be seen that overall the mean length of number ones per decade has been increasing, with a slight drop in the 2000s and recovery (though not to its peak) in the 2010s so songs are reigning at the top longer. A similar relationship is seen with the median lengths and modes also. Interestingly there are multi modalities in the latter decades, particularly in the 2010s than there has even been in the past decades with record 5 modes for the second longest number one reigns in the 2010s. There is also a disparity between reign lengths using these averages between the 1960s-1980s and the 1990s-2000s.
Figure 5. A table showing results of statistical analyses for each decade. The H-statistics and p-values are figures calculated by the Kruskall Wallis test, with p-values set at p < 0.05 and then the correlation co-efficient values are shown on the right hand side.
Part of the Kruskall Wallis hypothesis test calculation is to find the H-statistic which is in Figure 5. It is another test to discover if there is any statistical significance in a hypothesis test. The H-statistic was compared against the critical-chi squared value (CCSQV) which was found to be 5.99. If the CCSQV is less than the H-statistic, then one should reject the null hypothesis that all the medians equal. If the CCSQV is not less than the H-statistic, then there is not enough evidence to suggest that the medians are all unequal. In this study, only the 1980s H-statistic was the CCSQV not less than the H-statistic so this means that the null hypothesis is not to be rejected. Then regarding the Kruskall Wallis p-values, again only the 1980s was the only decade where there was no statistical significance therefore the null hypothesis was not rejected. The p-value significance increased in the order of 1970s, 1960s, 1990s, 2000s and finally the 2010s. Then the correlation co-efficient values which show how closely the data points are with one another showed that the strongest relationship was between the data points was the 1980s at -0.7441 which is a relatively strong, negative relationship, followed by a weak, positive relationship in the 2010s. The 1970s, 1990s and 2000s show weak, positive relationships or displayed no relationship, respectively, between the data points. Finally the 1960s showed a negative, very weak relationship between the data points.
DISCUSSION
Nielson Soundscan is a system still used today which collects information electronically on data around radio airplay spins (measured by audience impressions) across the country, record sales (both physical and digital) and, now, on-demand streaming services information. Nelson Soundscan’s introduction into calculating the Billboard Hot 100 is probably the most significant factor into why hit songs stay longer at the top when recorded in music publications. High single turnover from the 1980s was reducing, as seen in Figure 1, and it appeared that hit songs were staying around longer than labels and retailers anticipated, when they originally would cut a single’s performance short in order to move onto the next single. It meant that songs were popular even past the peak of their chart performance in sales or radio. Before 1991 only two songs had reached a peak of 10 weeks at the top spot whereas a multitude of songs have hit that stride or even longer in the past 28 years. This description is best seen in Figure 1 where there is a drastic increase in the overall song reign length before and after 1991 and so this could suggest that a reduction in singles turnover in a charting period, and the longest number one reigning songs of the year, could lead to repetitiveness of the same songs and/or song genres. Figure 2 also best shows that it was the longest running songs which were in fact seemingly influencing the increase in average song reign length at the top spots. This could be a driving factor into what leads the second and third longest running number one singles to be sonically similar to the longest running hit single to try and emulate its success.
Increased Weighting to Streaming
The increasing weighting given to streaming services in the Billboard Hot 100 formula in the 2010s such as Spotify and Apple Music, to combat Napster and illegal downloads in the 2000s, in the past decade seems to have also had a correlated effect with the increase in hit songs staying atop for longer too, as seen in Figure 1 with its upwards trend in all three variables in the 2010s. Since most streaming services can now be accessed and utilised with no initial purchase price, consumers are effectively consuming their music for free (though some are indirectly paying for it through advertising). The nature of streaming suggests that the majority of music consumers could be replaying the same songs on repeat and for each repeat is a contribution to a song's points to chart. Whereas in the past one sale of a record, be it a digital sale on iTunes or a physical copy, would only count as one sale despite however many times one could replay it on their own music player or device. This could be the reason why Figure 4 showed a revival in the mean, median and mode for chart hit reign from the 2000s to the 2010s when streaming allowed song repeats to count as charting numbers and why songs may have lasted along the top much longer ever since. Also as discussed in part 1, part of Generation Z’s social media world is meme culture, which is hosted through platforms such as Vine, Instagram and Twitter challenges and Tik Tok. Songs (or song remixes) played on here, even if only featured for 30 seconds, can contribute to a single stream on the platform. 2ith social media’s constant expansion, viral memes and the music attached to it can help contribute to a song’s ever lasting success for an extended period of time. As a result of finding similarity through online globalisation is a similarity in music choices which means the same song could be spread around for an extended period of time too. A great example of this was discussed in Part 1 with meme culture tactics Lil Nas X used to keep Old Town Road atop.
Wider Radio Listenership
Also American radio broadcasting has reached wider audiences with iHeartRadio for example, the largest American radio broadcaster, expanding its studios across the country. As a result labels can reach a much wider and diverse audience on radio then they ever could in the past. Once songs begin climbing in radio play, the songs’ infectious nature means that by human nature we will be subjected to enjoy what is being forced or recommended to us. In turn this means there is a high chance that these songs will continue to be requested. The increase in requests, and thus spins, contributes to higher chart points in the radio portion of the Hot 100 formula and this could be why songs are also staying longer at the top for longer too. Nelson Soundscan in 1991 could help keep track of radio plays more accurately and the shift from singles sales to 50% radio, 50% sales is probably why hit songs were also dominating longer on the charts due to the contribution of a song’s radio dominance as well as great performing sales as seen in Figure 1 and 4. An example of this is One Sweet Day’s 13 week stint at number one on radio.
Using figures 3 and 5, they show that in the 1980s hit song reigns between all three variables were more than likely to be equal or close to equal as possible both using the p-value and H-statistic and CCSQV. Whereas in all the other decades there was a higher chance that at least one of the variables would be more dissimilar from the rest. From a visual standpoint using Figures 1 and 3 it appears that the longest number one songs of the year seem to be the variable more dissimilar than the other two more noticeably post-1991. This is most evident using the statistical significance 2010s p-value of 0.007 where the longest running number one singles on Billboard seem to have a large disparity visually from the trend lines of the other two variables. Furthermore as this is a low p-value it suggests that this is a strong enough finding to reject the null hypothesis (but not determinable to prove the alternative hypothesis yet).
CONCLUSION
Based on this study hit songs are remaining atop for longer than they have been in the past. The most significant reason for this was the incorporation of Nielson Soundscan in 1991 which showed that there was immediate and extended popularity with a lot of hit music. So songs debut at the higher half of charts faster and they continuously grow on multiple formats via more radio stations, more plays, more music outlets and more streams. While the results of the hypothesis tests report that there is a significant difference between the groups for most of the decades, the test does not determine which of the data groups are dissimilar from the other groups despite any inferences that could be made visually. Therefore post-hoc tests would need to be carried out because one, we want to find out which group(s) this is and two, because there is also the probability that at least one Type I error had arisen in the data sets when comparing the variables in the Kruskal Wallis test. These post-hoc tests potentially include the Rodger’s Method, Tukey’s test or Dunnett’s correction. The former for example, would detect if there is any difference among the groups but by protecting against the loss of any statistical power and any increases in degrees of freedom in the non-parametric test calculations.
The limitations of this investigation is that outliers of trends, particularly in the longest running number one record breaking singles, can be detected very easily. However, they cannot really be removed even when identified because the sample data is too small, being another issue. A removal would create such a large knock-on effect on the ‘real’ representation of the data and it might remove the relevance and significance of any long reigns in this investigation. Also an investigation with a lack of repeated measurements and only using 10 data items means that inferences of the data must be taken under consideration; a higher number of data points and/or measurements if ever possible would have increase the chance of inferences being made more accurately.
For future investigations it would be useful to look at incorporating predictive analytics such as creating a model using Azure or R. The model could use the current data, recognise any obvious and underlying patterns and potentially predict a general short term trend line continuing on from Figure 1. However, it is unknowing whether artificial intelligence would be able to adapt to the quickly evolving nature of the Billboard formula reshuffle because of the quickly advancing and evolving consumption of music. It would also be interesting to see why there is a lack of stabilisation year-upon-year on song reign length, especially the longest running number one, and more of an oscillating relationship instead; it would also be an interesting finding to know if hit songs will just keep increasing in the future. If so, perhaps Billboard should regulate their formula more closely and regularly and put control measures in place to reflect a more fairer system of judging the most popular organically-grown songs of the week. Solutions otherwise could be to introduce a separate remix chart or to reduce the weight remixes can make to a song’s charting potential on the Billboard Hot 100 to give a more accurate view of which original songs last atop the longest and at what peak.
Part 3 of this series will be looking at seeing if there are any common trends in the longest running number one singles post-1991. This will be to determine if there are any ingredients or similar elements a long reigning hit needs.
