A few weeks ago saw the final of the tenth series of The Voice UK – a reality TV singing competition. The first stage of The Voice consists of blind auditions, where contestants sing, unseen by the four coaches whose chairs face away from the stage. If a coach likes what they hear, they press a button to turn around. If more than one coach turns, the contestant chooses which team to join. The blind auditions finish when each coach has a team of ten. Subsequent rounds reduce the field until four remain for the final, with the winner chosen by public vote. This article looks at the blind audition rounds over the ten series.
Although the basic format has remained constant, there have been a few changes. The first five series were broadcast by the BBC, but series 6 and later have been on ITV. In series 2 to 5, each team comprised twelve acts rather than ten. There have always been four coaches, but not always the same four: will.i.am has been in every series, and Tom Jones has been in each except series 5, when he was replaced by Boy George. There has always been just one female coach – a seat occupied by Jessie J (series 1 and 2), Kylie Minogue (3), Rita Ora (4), Paloma Faith (5), Jennifer Hudson (6-8), Meghan Trainor (9) and Anne-Marie (10). The fourth seat has been occupied by Danny O’Donoghue (series 1 and 2), Ricky Wilson (3-5), Gavin Rossdale (6), and Olly Murs (7-10). Series 10 introduced a “block” during the blind auditions, where a coach could block another coach from turning – this could only be done once by each coach.1
Information on the blind auditions can be found on each series’ Wikipedia page.2 Extracting the data is slightly fiddly due to the way the tables are formatted, and having to take account of shading as well as the text.
The final dataset covered 715 acts, with information on the series number, week number, act number, act name, their age, the song performed, which coaches turned (if any), and which team the act joined. I also derived the gender for most of the acts, using the
gender package in
R (which makes a guess based on first names), with a few done manually based on google searches.
It is unclear how complete the data is. The shows are edited and it is possible that some “no turn” acts are omitted from the broadcast shows. It is possible that the actual order of the acts differs from the edited broadcast version.
The following chart shows the distribution of the number of turns by series, as a proportion of the successful acts. For every successful act, around 50% more receive no turns and are rejected – although this proportion was more like 75% for series 2-5. The number of four-chair turns was highest in series 1 and 2, which also had a very low number of three-chair turns. The most recent series (10) was unusually low on both three and four-chair turns, with over half of the successful acts receiving just one turn.
Across the 715 acts there were a total of 863 turns, giving an average probability of just over 30% that a particular coach will turn for an act.3 If the coaches acted independently of each other, we would expect the number of turns per act to follow a binomial distribution, which we can compare with the actual distribution, as shown in the following chart.
If the coaches were independent we would expect significantly more one and two-chair turns, fewer no-chair turns,4 and many fewer four-chair turns. In practice, there is quite a lot of agreement among the coaches as to which are the good acts, so four-chair turns are actually received by about 8% of acts rather than the 1% that we would expect.
Another way of looking at this lack of independence is to calculate the correlation between coaches – i.e. the extent to which they tend to turn, or not turn, together for the same acts. The following chart shows the correlation coefficients between each pair of coaches that have appeared together. The average correlation is about 0.3, and there is only one pair (will.i.am and Paloma) with a (small) negative correlation. Anne-Marie (in series 10) has the lowest average correlations, suggesting that she has been the most independently-minded coach. Of the two long-standing coaches, will.i.am has generally below-average correlations, whereas Tom is mainly above-average – suggesting that Will is more likely to go out on a limb, whereas Tom is more likely to copy (or be copied by) the other coaches. The two highest correlation scores of about 0.6 indicate an unusual amount of agreement between Tom, Kylie and Ricky (in series 3), with, again, Will thinking rather differently.
The average rate of turning (i.e. the chance that a particular coach will turn for an act) declines steadily as the auditions progress and the number of vacancies on each team declines. The average rate, as noted above, is just over 30%, but, as shown below, the turn rate is about 45% during the recruitment of the first fifth of acts (the first two on each team), but falls steadily to around 20% for the final fifth.5 For participants, there is therefore a significant advantage in singing early in the competition.
The following chart illustrates this in more detail, showing the distribution of the number of turns depending on the proportion of acts already chosen. For the first fifth of the competition (i.e. the recruitment of the first two deciles of acts) there are plenty of 2, 3, and 4-chair turns, but by the end of the series 3 and 4-chair turns are very rare, and the proportion of no-chair turns and one-chair turns is over 80%.
The turn rate varies not only by how full the coaches’ team are, but also by the difference between the size of the teams. If the four teams were to fill up randomly, then (apart from near the start and end of the competition, when they are all of the same size) we would expect the median difference between the sizes of the largest and smallest teams to be three, and for differences of five or more to be relatively common.6 Instead, as the chart below shows, the actual median difference between the largest and smallest teams during the middle part of the competition (i.e. between successful acts 10 and 30, in the team-size-10 format of the show) is actually two, and there are very few occasions when the difference exceeds three. So we can conclude that coaches with larger teams tend to reduce their turn rate (and conversely, those with smaller teams perhaps increase their turn rate) in order to keep the size of the teams within about three of each other.
We can also look at which sorts of act are favoured by different coaches. In the chart below I have kept Will and Tom separate but have taken the liberty of grouping all of the female coaches, and all of the other coaches together. The acts are classified by gender (red and blue) and by age band (in five groups of roughly equal sizes). The highest turn rate is by the female coaches for male acts in the 20-23 age group, and it is interesting that, on the whole, the female coaches are more likely to turn for the male acts. Tom, on the other hand, prefers female acts at all age bands, and also has a preference for older performers. Will also tends to prefer the female acts, especially those in the 24-27 age group. The “other” coach is less keen on the older acts, and, among the under-24s, has a preference for females.
If you are over 27, then Tom is most likely to turn for you, although the female coach will also be tempted by males in the 28-33 group. If you are a woman under 20, the female coach is most likely to turn. Men under 20 are less likely than average to get a turn, but if they do, Will and the “other” coach are the most likely.
A coach turning their chair is only the first part of the process. If more than one coach turns, then the contestant gets to choose which team to join.
The following chart shows the “win rate” of each coach over the ten series. This is the number of acts choosing that coach’s team, as a proportion of the number that would have chosen them if acts selected a team at random (among those that had turned for them). A high score therefore indicates a coach’s popularity relative to the other coaches, among those acts that have a choice.
Interestingly the female coaches (in red) and will.i.am (in purple) are generally above the dotted line, indicating that they have a better than average chance of being chosen by acts with multiple turns.7 Tom (in light blue) and the “other” coach (in green) have been generally below average. Interestingly in the most recent series there was one coach a long way above the dotted line – will.i.am stood a good chance of winning acts against any of the other three coaches. The same was true of Jessie J in series 1, and (less strongly) with Will in series 4. In series 7, 8, and 9, the opposite was the case (albeit weakly), with Tom and then Olly (twice) being likely to lose out against any of the other coaches.
Finally, let’s consider the acts by age and gender and see where they are likely to end up. The chart below shows the likely fate of different types of act. Among the no-turns group (darkest blue), young females are least likely to have no turns (under 25%), and older acts are most likely (over 50%). Older acts – and older females in particular – are likely to end up in Tom’s team. The youngest acts (especially the men) are least likely to join team Tom, but are well spread among the other teams.
Some of these findings will be unsurprising to those familiar with the programme: the preferences of individual coaches (such as the tendency for Team Tom to have a high proportion of women); the reduction in turn rates as the competition progresses; and a reasonable degree of correlation between coaches. Other things are perhaps less expected – the different strategies pursued by Will and Tom; the high success rate of female coaches in being chosen by multi-turn acts; or the extent to which coaches (perhaps unconsciously) try to keep the teams of similar size throughout the blind auditions.
The latest series, produced under Covid restrictions without a live audience, was unusual in having very few three or four-chair turns; the most independently-minded coach (Anne-Marie, who went on to win the competition with her act Craig Eddie); and a very high success rate for one coach (Will) in winning multi-turn acts.
Despite all this, it is interesting to note that, of the ten series, after the blind auditions and the subsequent elimination rounds and public vote, Will’s team has won the competition just once, Tom twice, the female coach twice (Jennifer and Anne-Marie), and the “other” coach five times (Danny once, and twice each for Ricky and Olly).
- There have also been some changes in the later rounds, which do not concern us here. See this Wikipedia page for details.
- They have URLs like this – https://en.wikipedia.org/wiki/The_Voice_UK_(series_1) – with the number replaced as appropriate, and are all consistently formatted.
- i.e. (863 turns) / (715 x 4 possible turns) = 0.302
- This perhaps suggests that not many of the no-chair turns are edited out of the broadcast programme.
- The horizontal axis in this chart (and the next) indicates “deciles” (tenths) of the way through recruiting all of the teams. So the value for “3” includes all acts performing after the 12th successful act (if the team size is 10), up to and including the 16th successful act. This allows the series with teams of 12 to be included alongside those with teams of ten.
- These figures were derived by simulation. They are true for team sizes of both ten and twelve.
- The consistent success of the female coach, despite many changes over the years, is perhaps an argument for having a more even gender balance among the coaches!