Search

Mapping the High Performance Mindset: Predicting Tennis Rankings with Language Psychology

It seems like an obvious statement to say that elite athletes look different to everybody else. Physically they are stronger, faster and are more coordinated than the average human. Mentally they are also more focussed, more resilient and more driven than the rest of us. But how important are psychological attributes to success? Do athletes at the very pinnacle of their sport look mentally any different to their less successful peers?


To try and answer this question we analysed 24,365 post-match interviews with professional tennis players between 2000 and 2022 using Receptiviti’s language psychology platform. We extracted personality and psychological measures from these interviews and matched the interviews up against ATP (the men’s tour) and WTA (the women’s tour) world rankings for each player by date. We then grouped each interview by ranking into three categories: ranked in the top 10, ranked between 11-50 and ranked > 50.


As it has been well established that there are subtle differences in language use between men and women, we split our data set by gender. Our feature set was comprised of all LIWC measures except for the punctuation categories (as our sample consists of spoken language) plus the Receptiviti personality (both Big 5 and DISC), Drives, Needs and Values frameworks. With our training set constructed, we next built two models (for each of the men’s and women’s tours) to predict which ranking category each player falls into at the time of each interview.


Our overall model accuracies were 52.5% for the ATP model and 42.6% for the WTA model, far in excess of what would be achieved through random assignment (i.e. 33.3%). Both models could distinguish between a top 10 player and one outside the top 50 with confidence, though the WTA model struggled to accurately classify players ranked between 11-50 (see Figure 1). Regardless, the ability of the two models to separate players into the three ranking groups shows clearly that there are significant differences between the psychological and personality profiles of players at different levels of success.


Figure 1 – Confusion matrices for the ATP (left) and WTA (right) tennis ranking models. These plots show how well the models perform at sorting players correctly into the different ranking groups. The numbers show the percentage of people in each ranking group that were correctly predicted (i.e. the true positive rate). For example, the ATP model correctly classified 54.1% of the top 10 players, while the WTA model correctly classified 57.8% of the players outside the top 50.


Language related to social dominance (specifically "I", "we," "you," and clout, a composite variable that captures all of those variables) is most decisive for the men’s model. People tend to focus less on themselves and more on others as they climb the social ranks, so this is not particularly surprising. In contrast, the most consequential categories for women are a little more abstract and cognitive, having to do with temporal focus, spatial orientation, and cognitive processes. Together, in the context of post-match interviews, those categories may reflect either rumination or thoughtful planning (reflecting on and drawing conclusions about past matches, strategizing for the future).


I-word usage (i.e. first-person singular pronoun), which is a marker of self-focus, is an important feature for both the men's and women's models. By varying the I-word scores across the training set and re-scoring with the models, we found that I-word values between 4% and 7% lead the models to favour a top 10 ranking while above and below this range the models score athletes as most likely to be lower ranked. People with high I-word scores are likely to be depression-prone, anxious and even in pain or physically sick, while those with low scores are likely to be overly-confident, inauthentic and lacking in self-reflection and social sensitivity. So, it makes sense that the models see moderate I-word usage as a sign of an elite athlete, as they would appear to be more balanced individuals.


As our interview data set spans over two decades, we can also chart the careers of specific players who have risen through the ranks over time (and therefore fall into different categories at different times in their career). We have found that there are subtle differences between the language psychology of high and low ranked players, but do players on their rise through the ranks achieve a top 10 mindset in advance of reaching the pinnacle? Or do they achieve success first, and then develop a top 10 mindset later? With the French Open starting this week, we thought it would be interesting to look at a couple of high-profile players who have risen meteorically through the rankings in recent years: Carlos Alcaraz on the men’s tour (currently ranked #6) and Iga Swiatek on the women’s tour (currently ranked #1).


Carlos Alcaraz has charged through the world rankings, reaching the top 100 in 2021 at the age of 18. His rise since then has mirrored some of the all-time great players such as Roger Federer, Rafael Nadal and Novak Djokovic. He breached the top 50 in September 2021 and reached the top 10 only 7 months later in April 2022. He is currently ranked number 6 in the world, after defeating clay court specialist and world #4 Nadal, world #1 Djokovic and world # 3 Zverev to take out the Madrid Open at the tender age of 19. He is going in to the French Open as one of the favourites (second only behind Djokovic) and will likely challenge for a top 3 position in the near future.


Similarly, Iga Swiatek has also rapidly risen through the ranks over the past few years. She reached the top 100 in mid 2019 at the age of 18, the top 50 in January 2021 and the top 10 only 4 months later. With the retirement of Ash Barty she was elevated to #1 in the world and is the youngest female player ranked in the top 10. Having won the French Open in 2020 she is the hot favourite to win it again in 2022.


When did these two players start to look like the real deal? To answer this, we classified the interview transcripts for each player using our models and plotted the predicted class probabilities for each over time (see Figures 2 & 3).


Figure 2 – Ranking model class probabilities over time for Carlos Alcaraz. The coloured points show the probability that the model assigned to each ranking group for each of the post-match interviews Alcaraz has given over the past 2 years. For example, in early 2021 the model estimates that Alcaraz has a probability of ~49-57% of being ranked outside the top 50, while more recently the model estimates his probability of being ranked either between 11-50 or in the top 10 as being ~48%. The solid lines show the probabilities smoothed over time. The dashed vertical lines show when he moved into the top 50 and top 10, respectively.


Figure 3 – Ranking model class probabilities over time for Iga Swiatek. The coloured points show the probability that the model assigned to each ranking group for each of the post-match interviews Swiatek has given over the past 4 years. For example, in mid 2018 the model estimates that Swiatek has a probability of ~51% of being ranked outside the top 50, while from her most recent interview the model estimates her probability of being ranked in the top 10 as being ~51%. The solid lines show the probabilities smoothed over time. The dashed vertical lines show when she moved into the top 50 and top 10, respectively.


While data for both players is sparse early in their careers (likely a result of lower ranked players not drawing as much attention from the media), it is interesting to note that both started looking like higher ranked players before they reached the top 10. Alcaraz scored highest in the 11-50 rank class well before he breached the top 50, although even when he reached the top 10 he still looked more like a lower ranked player, albeit with an upwards trend in his top 10 rank class probabilities. Swiatek, in contrast, shows clearly stratified clusters in her ranking group probabilities, looking more like a top 10 player when she was in the top 50 than either of the other two groups. As she has been ranked in the top 10 for longer than Alcaraz, it will be interesting to see how his scores change over the next few months.


It's clear from these analyses that not only do the very best athletes stand out from their peers psychologically, they also appear to develop their high performance mindset prior to achieving success. This suggests there could be some very interesting applications for this work in high performance recruitment, both within sport and in the broader commercial world.


Cover image credit: Photo by Moises Alex on Unsplash