As we age, we become slower thinkers and decision makers - this is a widespread assumption in Western societies. Moreover, hundreds of cross-sectional and longitudinal psychological studies conducted in the past decades have seemingly confirmed this idea: Starting from an age as early as 20, the older a person is, the slower we can expect her to respond to a wide range of cognitive tasks. Indeed, response times in cognitive tasks are the variable most previous psychological studies on age differences in mental speed have primarily relied on.

However, when thinking more closely about what exactly is measured in response times, one cannot help but acknowledge that it is in fact not only mental speed that contributes to these measures, but also other factors. For example, when a person is asked to classify words or pictures as belonging to one of two categories by pressing a key, she might exhibit slower response times than another person because her brain is slower in processing information, because she is slower in pressing the required button, or simply because she is very careful and wants to avoid making any mistakes. It might also be a complex combination of these processes. In any way, simple response times (or their averages) are suboptimal proxies of pure mental speed, as they reflect the sum total of disparate processes.

The drift-diffusion model was developed by Roger Ratcliff with the intention of disentangling the processes contributing to raw response time and accuracy data in simple binary decision tasks. By employing the diffusion model, one can estimate a set of parameters tapping into, among others, mental speed, decision cautiousness, and the time taken for non-decisional processes, such as motor response execution. In the past years, the diffusion model (and variants thereof) has found widespread use in experimental and cognitive psychology. More recently, researchers have also started to employ the model as a tool for studying individual differences in cognitive functions. In this way, one can model age differences in mental speed in the context of other contributing factors (e.g., cautiousness), as opposed to simply measuring average response times. However, these initial diffusion model studies were severely limited both regarding sample sizes and the age ranges studied.

To address these shortcomings, we conducted a large-scale cross-sectional study of age differences in drift-diffusion model parameters, utilizing data from a massive online experiment. We used data from a race implicit association test (IAT) openly shared by Project Implicit.  For our purposes, the IAT (typically employed to study implicit racial biases in other research contexts), presents itself as a very convenient example of a cognitive decision making task. Given the sheer magnitude of the data set (our final sample size was close to 1,200,000), it was necessary to apply specialized neural networks within the newly developed BayesFlow framework to arrive at our parameter estimates. Had we used standard estimation methods, it would have taken well over a year to obtain parameter estimates and would have been nearly impossible to validate our results via conventional model checking methods.

Figure 1 depicts our main results. Each dot represents an average (across people) quantity for a specific age. In the top left, we see that our results replicate a well-known finding from previous studies: Average response times tend to become slower from about age 20. However, as already mentioned, average response times do not tell the whole story.  Our analysis yields three further parameters whose relationships with age are depicted in the remaining three panels.

Our model implies that people tend to become more cautious decision makers as they age, starting at about age 20. Differently, the average time spent in non-decisional processes, such as motor response execution, already increases from about age 15 on. Interestingly, and most importantly, the parameter representing speed of information uptake  (“drift rate”), or mental speed, increases (!) until the age of about 30. Average levels of mental speed then stay relatively high throughout the entire age range making up middle adulthood. Only after an age of about 60 do people tend to show signs of notably decreasing mental speed. Another important result is that there is great variability in cognitive parameters within each age group.

Figure 2 shows that our findings on mental speed (“drift rate”) remain essentially unchanged across a variety of robustness checks controlling for demographics and different experimental conditions.

Taken together, our findings imply that the well-documented age-related increase in mean response times is, for most of the adult lifespan, not due to age differences in mental speed. Instead, age differences in decision caution and the time taken for non-decisional (e.g., motoric) processes account for the empirical patterns observed in the raw data. Our results suggest that the interpretation of age differences in response times as indicators of age differences in mental speed might need to be reconsidered. In the end, the notion that mental speed declines from age 20 on might have been overstated in previous research. 

Naturally, our study has several limitations. Most importantly, our results are purely cross-sectional, that is, we did not study longitudinal development, but only age differences between different groups of people. Secondly, our analyses are based only on one specific cognitive task, namely, the IAT. However, it should be noted that our results are in line with previous modeling studies that used a much wider range of cognitive tasks - although with much smaller sample sizes and age ranges. Finally, our study rests on the validity of the basic drift-diffusion model. As with any model, the latter can be questioned, but we note that the model has been extensively validated in the context of both basic psychological and neuroscience research.

Bringing together a theory-grounded, model-based approach, modern deep learning methods, and an unprecedentedly large sample, our research sheds new light on age differences in cognitive processes and mental speed. Our results question the common lore that people generally tend to think slower when moving out of young adulthood - an assumption which influences real-world outcomes, such as hiring policies. Hopefully, future research will further clarify the relationships between age, mental speed, and other cognitive parameters.

Citation:

von Krause, M., Radev, S.T. & Voss, A. Mental speed is high until age 60 as revealed by analysis of over a million participants. Nat Hum Behav (2022). https://doi.org/10.1038/s41562-021-01282-7