Sports fans love to obsess over statistics. Name the sport, there are numbers to match. It’s safe to say, nearly every measurable facet of game play is noted. Taken together, the data allows players, teams, and fans to quantify the performances involved.
Namanja Vaci approached data from the National Basketball Association from a different and creative angle. He used the detailed analytics available to teams as a way to study age-related motor and cognitive degradation.
SCIENTIFIC INQUIRER: Why did you study the effects of aging motor and cognitive skills? Why are the changes important in psychology?
NAMANJA VACI: The main reasons are the age shifts in the population, where the median age of the population is increasing across the globe. These changes in combination with relatively unknown causes of old-age mental disease paint quite a grim future for our generations.
In the overall ageing literature, there is a rich history of age-related changes investigation, but primarily in the case of the speed-dependent cognitive and motor tasks. Due to the preciseness of reaction time measurements, the change in the processing speed and executive functions across the age is one of the most thoroughly investigated effects. There are also few studies that investigate changes of knowledge as people age, e.g. size of the mental lexicon.
Contrary to this, studies that investigate the most prevalent type of behaviours in our daily lives, that is, skills which depend on executive functions and knowledge structures across the lifetime, are few and far between. This is the main reason why we focused on the games, such as NBA or chess performance, which allow us precise quantification of the performance measures.
The NBA or chess play depend on the executive functions, eye to hand coordination or estimation of a situation for a few moves in the future, but also on a vast amount of knowledge patterns and chunks of experience which players collect throughout their career. We hope that developing and testing models, as well as, understanding how people age in their performance in the case of artificial systems, we will be able to apply same procedures in everyday behaviours and generalize our findings.
SI: Prior to your experiment, why was it so hard to study changes in motor and cognitive skills among aging individuals?
NV: The main problem behind any ageing question investigation is the data size. Most of the studies that collect performance level data have small sample sizes and often come from experimental settings. The large data usually comes from environments that are difficult to operationalize and derive measures of performance (such as Facebook, Twitter and similar). The size and breadth of the data are directly connected to the type of statistical and machine learning models that we can build. Subsequently, the common practice in the literature was to build relatively simple models on a small amount of data.
In the case of our studies, we used the archival approach where we utilize collected performance scores throughout the last few decades. The size of the data allows us to build more flexible models, which can reveal new insights into the ageing process. However, even the archival approach brings similar issues, as it mainly focuses on the performance in games and gamified environments.
SI: How did you come to the realization that NBA data could be used?
NV: My first analysis and models were developed on the chess data and the main result was that experts experience different ageing patterns in comparison to the average players. Namely, they tend to preserve their performance for much longer and always outperform average players. From these first findings, my plan was to test the same hypothesis on the performance data in physical sports, where the ageing process is much stronger. The NBA is one of the environments where we tested this hypothesis, besides chess data.
SI: You developed a new model to measure age-related changes. First can you explain why you did this? How were previous models lacking?
NV: I cannot say that other models are lacking, all of them are investigating and quantifying different parts of the problem. The main problem with the age-related changes in performance is the nonlinear behaviour of the function. Different statistical models can relatively well adapt and be used to investigate these changes.
The most common type of modelling in the field is to use polynomial or exponential functions, where these models quantify the changes across the lifetime and give us a good indication on the shape of the function, that is when performance increases or decreases. However, they tend to have problems with overfitting in more complex systems, especially, when we have interactions of nonlinear functions. The splines and generalized additive modelling results in optimal functions for nonlinear changes and can fit nonlinear interactions, but they do not quantify the effects and interactions in a standard statistical way.
The idea of our model was to combine these two approaches, to allow the nonlinear form of the function (exponential, polynomial, etc.) for the age-related changes in performance, but also to interact parameters of the ageing function with other variables. This way we can investigate how the strength of the skill acquisition phase or age-related decline, for example, modelled with power-law function, correlates with activity, intelligence, or any other variable of interest.
SI: How is the model you designed, B-Ianus, designed different?
NV: We use the cognitive latent variable modelling method to combine three levels of the model. The first level represents the nonlinear function of age-related changes in performance, separately for skill acquisition phase (development) and skill decline phase (ageing).
In the second level, we use factorial structure to combine the parameters from these functions with other variables, e.g. minutes per game in the case of NBA data. This parameter gives us the strength of the relation between activity (MPG) and development or ageing. In the final layer, we correlate the development and ageing latent factor to quantify the interaction between the strength of the increase (development) and strength of decrease (ageing) when controlling for activity.
SI: How did you quantify basketballs motor and cognitive skills? How did you insert the values into B-Ianus? What did you discover?
NV: In the case of our studies, we use standard measures of performance, such as Elo performance in chess or Win Shares, Player Efficiency Rating and Value over replacement player in NBA. As already mentioned, we hope that these measures are a joint product of multiple subspecific cognitive systems and processes that are going to differently change over the lifetime, that is, as people age.
The performance measures are time-dependent and we often have multiple measures per player, thus, we adjust the intercept and the slope of the functions. The measures of other explanatory variables, e.g. minutes per game, are transformed (e.g. sum, mean) to have one value for each function (development or ageing) per person.
We find that people who develop more and gain more skills, preserve their performance better as they age, confirming that age is kinder to the abler (or more knowledgeable in our case) hypothesis. It seems that players who gain more knowledge and reach a higher level of expertise, use this knowledge later to dampen the ageing effects.
SI: WS, VORP, and PER are not exactly hard statistics. Much like Wins Above Replacement (WAR) in baseball, they’re more a hybrid of ideas and statistics. How can this “squishiness” be compensated for?
NV: In our studies, we are mainly focused on the measures that aim to quantify a player’s value and their contributions. As I mentioned previously, we hope that these metrics measure players overall expertise and knowledge that they have about the system and how good they are in those systems. Going to the more fine-grained measures will probably result in conclusions that can be related to subspecific cognitive systems and processes, but will also lose the complexity of behaviours that we use in everyday life.
SI: What is next for your research?
NV: The next studies are focused on underlying factors of skill preservation as people age. We are examining interactions between general ability factors (e.g. different types of intelligence) and activity (a.g. the number of played games or MPG) on the preservation of skills. Finally, I am also moving towards the clinical aspect of the situation (maladaptive ageing), where I hope to investigate potential moderating effects of expertise and knowledge on the development of age-related mental diseases.
For more information on Nemanja Vaci and his research visit his university page.
IMAGE SOURCE: Creative Commons
The Scientific Inquirer needs your support. Please visit our Patreon page and discover ways that you can make a difference. http://bit.ly/2jjiagi. Alternatively, to make a one time $10 contribution visit our Support page.