Studies of within-person variability can be used as a first step to constrain the search space for later experimental research to test mechanistic accounts. If it turns out that two measures are correlated negatively within participants (i.e., trade-off), this suggests that processes or representations overlap and are used competitively. An example that most experimental cognitive psychologists will be familiar with is the speed-accuracy trade-off (SAT), which is the phenomenon that for decision making systems the speed with which a decision is made negatively correlates with the accuracy with which the decision is made. Discovered relatively early in the history of modern psychology (Henmon, 1911, for a broader overview see Heitz, 2014), the SAT can, for cognitive psychologists, sometimes be more of a problem to be dealt with (e.g., Vandierendonck, 2018; Liesefeld and Janczyk, 2019), but it illustrates our point about trade-offs well: There seems to be some sort of limit on decision quality gained per time invested, which in accumulator models would be the velocity of evidence accumulation (see, e.g., Bogacz et al., 2010). Shifting toward speed at the cost of accuracy or vice versa results in a negative within-person correlation (Bakdash and Marusich, 2017). As such, the SAT is not only a confound adding noise to be controlled but also an epistemological signal to be used, as it indicates the presence of a cognitive resource.

As an example on the level of task representations, Schuck et al. (2015) used fMRT to track how redundant variants of representing a task as a color- vs. a spatial task were represented and found a negative coupling (space or color) rather than redundant coding in task-set relevant brain areas. In the applied domain this might mean that on some days people may approach a traffic light with a task-set strongly weighing color and on other days focusing on light position instead (cf. Overton and Brown, 1957, for a between-person difference perspective on this issue). Follow-up work can target active and passive mechanisms clearing redundant parts from task representations. Apart from task sets, trade-offs are also documented on the level of features. A central aspect of the Theory of Event Coding (Hommel et al., 2001) is that features (such as the code “left”) already used for an event file (such as planning a left arm movement) are less available when concurrently needed in a different task (i.e., recognizing a left-pointing arrow, Wühr and Müsseler, 2001).

In case of a positive correlation, candidates for a common factor can be tested. For instance, Brose et al. (2012) documented that days with lower working memory performance, were days with more negative affect and reduced control of attention, suggesting, to follow-up on the role of working memory in emotion regulation.

As mentioned, studies of within-person variability can not only be used to constrain first steps in constructing mechanistic theories, but also to test them. Some theories strongly linked to experimental work can make predictions on the covariance structure of task performance that can be tested in multi-session datasets that allow for day-to-day fluctuation. For instance, bottleneck-theories in dual-tasking (cf. Pashler, 1994; Tombu and Jolicœur, 2005) can be taken to suggest that measures of performance in the two tasks should correlate negatively within subjects.

Lastly, this approach can also be turned on its head to inform how to design environments in which there are multiple related concurrent tasks where good performance in one task is desired and in the other irrelevant. If it is established that a cognitive resource is divided among the tasks, then minimizing the amount of resources used by the irrelevant tasks is a design goal, while if a common factor pattern is established, no such precautions need to be taken. Cognitive load theory in the context of instructional design is an example of this logic: If a student is to learn a subject from teaching materials and needs to use a cognitive resource on both parsing the material and then processing the material, lowering the resource draw of the parsing component will free up resources for the processing component (Sweller et al., 1998).

Correlative studies can help evaluate to what extent mechanisms implying a trade-off structure or mechanisms implying a common factor structure are relevant. Potential outcomes can be that there is evidence only for one or the other case, or that both mechanisms contributed (potentially with different weight). Importantly, to fully harvest the potential of correlative studies for constraining candidate mechanisms, the studies should not be limited to cross-sectional assessment of correlation.

An example involving typing might illustrate that correlations across persons are logically independent from correlations within person (Hamaker, 2012). In cross-sectional studies, the between-person correlation between the time needed for a typing task and error rate might be positive: Some people are good typists. They type quickly and accurately. In contrast, the within-person correlation obtained in a longitudinal study across many typing sessions might be negative: On occasions a person types faster, the error rate will be higher.

A small but growing body of research is now using intensive repeated measures, wherein participants complete tasks on many sessions, to examine within-person coupling of indicators of different cognitive processes. For example, Brose et al. (2012) studied within-person and between-person differences in working memory, control of attention, and affect in 101 young adults across 100 sessions. They obtained evidence for the common factor view (positive correlations), within- and between persons. They found that the same variables that predict between-person differences in working memory performance (cross-sectional correlations) also predict within-person (session-to-session) differences in working memory performance (longitudinal correlations). Given the logical independence of the within- and between-person variability, these results suggest that findings from prior studies of between-person differences that were at risk for ecological fallacy may be at least partially informative about within-person cognitive processes (Molenaar, 2004; Hamaker, 2012).

In summary, we think that the concept of a cognitive resource can be made epistemologically useful and empirically tractable by contrasting it with the concept of a common factor and identifying the two concepts with negative and positive within-person correlations on theoretically related measures. The epistemological use is primarily one of restricting the theoretical space within which mechanistic explanations are to be searched for, but also includes testing hypotheses.

CN wrote the first draft of the manuscript and created the figure. All authors wrote and revised sections of the manuscript and read and approved the submitted version.

This research was funded within the priority program on multitasking SPP1772 of the German Research Foundation (DFG) awarded to Robert Gaschler (GA 2246/1-2) and supported by the FernUniversität in Hagen.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.