Multiple timescales
In this article it is explained how multiple timescales may play a role in the processes you are interested in. This is important to consider when setting up your study, because it has implications for how to best measure the process of interest, and how to analyze the data that are thus obtained.
Many psychological processes can be considered to operate at a range of different timescales. For instance, affect regulation may consist of: taking a deep breath when confronted with a negative remark from a friend; working out multiple times per week to reduce stress hormones and increase the release of endorphins; and going to therapy once every two weeks for several months to talk about current stressors and how these trigger memories of aversive childhood experiences. While a process may be characterized by similar patterns at various timescales, there may also be marked differences in the temporal patterns when zooming in or out. It is therefore important to think about which timescale(s) you are most interested in with respect to your research question, and to think about how to attune your temporal lens such that you can focus on the relevant timescale(s).
In this article, you will find: 1) a presentation of a non-psychological example that helps to illustrate the importance of considering timescales when studying processes; 2) a discussion of how different timescales may play a role in psychological processes; and 3) suggestions on how to study multiple timescales and their contribution to the process of interest.
1 Fluctuations and changes in temperature
Suppose you are interested in fluctuations and changes in outside temperature. There are many ways in which you can obtain data for this, which allow you to focus on different aspects of the process that results in the temperature you measure.
For instance, you can decide to place a thermometer outside your window, and watch the temperature constantly for half an hour. You may see that whenever a cloud moves in front of the sun, the temperature goes down a little, and when the sun comes out again, the temperature rises a little again. Similarly, when there is a bit of wind, you may observe a slight decrease in temperature, while an episode of calm results in a slight increase again.
Instead of these very dense measurements, you can also decide to measure temperature every hour, for a few days. Taking this approach, you probably get to see the ebb and flow of temperature associated with the rhythm of day and night.
Alternatively, you may decide that you want to obtain the average temperature of every day, or the maximum or minimum, for several days. Taking this approach, you can see that this summary of the day varies from day to day, but that there is also some pattern in these fluctuations, in that days that are warm tend to be followed by other warm days, and vice versa.
If you obtain the latter kind of data for the period of an entire year, you should be able to see how temperature fluctuates across the seasons. Especially if you have this for multiple years, the seasonal pattern should be clearly repetitive.
You may also choose to zoom out further, and consider monthly averages instead of day averages, over a period of five years. This will also allow you to see the repetitive pattern of higher temperatures in summer and lower temperatures in winter, but with less detail than when this is based on daily measurements. The degree of detail from daily measures may actually not be of interest to you.
If you are able to obtain this kind of data for enough years, you can also investigate whether temperature tends to fluctuate around a constant in the long run, with some years being a bit warmer than others on average, or if there is a trend that would be indicative of climate change.
These different ways of measuring temperature allow you to learn about different processes, all of which contribute to the temperature that is measured. The wind outside your window that affects momentary temperature is not less real than the influences of diurnal cycles, seasonal patterns, or changes in climate.
What is critical is that you determine what information is relevant for the research question that you want answer. For instance, if you want to make weather forecasts for the next few days, neither annual averages over a period of decades, nor momentary measures over a period of half an hour are going to be very helpful. Similarly, when you want to study climate change, daily averages over one or two years are not as useful as annual averages (or maxima) over a period of decades.
Hence, you have to select a temporal design that ensures that the temporal lens of your study is attuned to the specific aspects of a process that you want to study. If you decide to zoom in or out you may obtain a more microscoping or satellite view (Klonek et al., 2018). Additionally, you can decide to obtain more or less granularity, depending on the degree of detail that is needed to be able to answer your research question (Zaheer et al., 1999).
2 Psychological processes at different timescales
Like the temperature example described above, psychological processes—such as coping, affect regulation, goal pursuit, or social interactions—tend to manifest themselves at a range of different timescales. This implies that you can study the behavior of processes at different timescales using different temporal designs.
To fully understand such processes, it is probably necessary to gain at least some insight into how the process operates at various timescales. Moreover, it may be of interest to obtain an understanding of how different timescales interact with each other.
2.1 Different patterns at different timescales
Processes manifest themselves at various timescales. When you are studying a process, the temporal patterns you get to see likely depend on the timescale you focus on.
Loes and Eeske study how specific actions from the parents affect the behavior of their children. When focusing on how to best handle the tantrums of a toddler who demands a cookie, it becomes clear that in the short-run, it is beneficial to give the toddler a cookie: They calm down and the tantrum seems to dissipate.
But in the longer run, when the parent keeps offering cookies whenever the toddler is upset, the number of tantrums is likely to increase, and the toddler will start to demand cookies more frequently and more fiercely. Thus, while providing a cookie results in an instant gratification and a happy child at the shorter timescale, repeatedly providing a cookie (in the moment) results in a child who seems to be less happy on average (as they display unhappy behavior more frequently and more intensely).
Loes and Eeske consider that, if they want to use this research to formulate parenting advice, they have to take these different timescales into account.
This example is based on (Keijsers & Roekel, 2018).
It is important that you think about this, and that you make conscious decisions when designing an ILD study to ensure you are observing the relevant process variation for the particular research question you are tackling.
2.2 Interacting processes at different timescales
Processes that operate at various timescales may also interact with each other. For instance, in order to reach a more global goal, a person may break it down in intermediate goals that need to be reached first, in order to move towards the more global goal.
Charles and Michael are interested in the way goals shape our behavior. They believe that some goals are rather abstract and may exist over extended periods of time, such as “be a good parent”, whereas other goals are quite concrete and immediate, such as “help my son understand this math exercise”. In between these extremes, a range of other goals may be identified, such as “help my son get his homework done this week”, and “support my son in becoming an independent and confident young adult”.
Charles and Michael believe that the best way to understand goal pursuit is to understand how the more concrete goals at lower levels form intermediate steps towards the more abstract goals at higher levels.
Charles Carver and Michael Scheier wrote about this hierarchy of goals repeatedly in their work (e.g., Carver & Scheier, 1982).
The goals that exist at different timescales (and at various levels of abstraction) interact with each other in multiple ways. On the one hand, the more abstract goal is broken down into smaller, more concrete goals that form stepping stones towards reaching the more global goal. But when a person notices that the intermediate goals are not successfully attained, they can decide to rephrase their global goals.
Hence, the goals at higher levels result in goals at lower levels; and failure to successfully reach lower level goals, may result in a reappraisal and adaption of higher level goals (Carver & Scheier, 1982). This implies that higher level timescale processes and lower level timescale processes affect each other in both directions, and you may be interested in studying their reciprocal effects.
2.3 Various timescales in your measurements
For many of the processes that are measured using ILD, it is likely that there are various underlying processes that operate at different timescales, and that all contribute to your measurements.
Steve is interested in studying the relation between accuracy and speed in a typing task. When considering the relation between accuracy and speed at a short, fast timescale (i.e., minute-to-minute), it becomes clear that these variables are negatively correlated: Tasks on which a participant types faster are characterized by more mistakes.
Yet, when allowing the participant to train a lot, it becomes clear that speed and accuracy both increase over time: This shows the effect of a learning process, which takes place at a longer, slower timescale.
Boker & Martin (2018) describe how we can think about the accuracy-speed trade-off and developmental growth within a single theoretical framework.
To disentangle these various processes, it may be possible to measure the process in such a way that the observed within-person variations are (primarily) determined by the dynamics that characterize the process at one particular timescale. For instance, when you want to study the process of speed-accuracy trade-off, you may isolate this from developmental changes by focusing on within-person fluctuations over a very short period of time: This may be achieved by using a temporal design that combines short time frames of your measurements, short time intervals between your measurements, and a number of measurement occasions that is not too large.
Similarly, when your interest is in the developmental process, you can opt for a temporal design with longer time frames (such that you average over fluctuations that are not relevant for the developmental trajectory), and longer time intervals between your measurements. This ensures that you obtain a much longer total time span for your study, in which development can take place.
3 Studying processes at different timescales
Processes may behave differently at different timescales. Hence, when contradictory results are found across different studies, one aspect you should consider is the timescale to which the results pertain. When you are planning how to measure and analyze ILD to study a particular process, it is critical to think about the timescale at which you believe the process of interest unfolds.
There are various ways in which you can observe multiple timescales and their contribution to the process of interest.
3.1 Measurement-burst design
The measurement-burst design was proposed as a way to combine a focus on short-term intra-individual variation with a focus on long-term intra-individual change (Nesselroade, 1991). The design is based on having multiple waves or bursts of ILD measurements.
Pia is interested in the degree of variability in affect at different timescales. She gets measurement burst data, consisting of ESM measurements obtained in three bursts; these bursts were each four months apart.
Pia decomposes the variance into three sources: within-person variability within a burst, within-person variability across the bursts, and variability across people. She finds that for most affective variables the amount of variability within a burst is largest, followed by the variability across people. The variability across waves tends to be quite small.
This example is based on (Andresen et al., 2025).
To analyze data from a measurement-burst design, it can be useful to think about ways in which analysis techniques developed for ESM or daily diary data can be combined with analysis techniques that exist for panel data (Alessandri et al., 2021; Hamaker et al., 2023). Moreover, the combination allows you to study whether specific dynamic features, such as an autoregressive parameter, change themselves dynamically over time (Andresen et al., 2025).
3.2 Multiple studies combined
Instead of combining different timescales into a single measurement design, you can also consider the combination and comparison of results from studies with different temporal designs.
Savannah wants to study the reciprocal relations between parental support and adolescents’ depressive symptoms at various timescales. To this end, she performs a multi-sample study in which she analyzes data that were obtained using different temporal designs.
Specifically, she compares the results obtained with daily diary data, bi-weekly data, three-monthly data, annual data, and biennial data. Savannah finds that the strongest relation from depressive symptoms to parental support seems to be at an interval of 2 weeks or 3 months; for shorter and longer intervals, Savannah finds no relations in the data.
This example is based on the study by Boele et al. (2023).
Alternatively, if you have a data set that consists of a large number of repeated measures, you can also decide to use different degrees of aggregation to tap into the different timescales that are present in your data.
Anna wants to see how the reciprocal relations between parent–adolescent conflict and ill-being vary across different timescales. Rather than having data from different studies, which tend to have somewhat different samples and measurement instruments, she decides to conduct a 100-day daily diary study and aggregate the data in various ways so she can study the dynamics at daily, weekly, biweekly, monthly, bi-monthly and three-monthly intervals.
In analyzing these data, Anna finds that the relation from conflict to ill-being were strongest at 1 and 3 month intervals, whereas the relation from ill-being to conflict was strongest at 1 or 2 week intervals.
This example is based on the study by Bülow et al. (2025).
While these strategies allow you to see at what time interval the relations seem to be the strongest, it is important to realize that there are multiple factors that may affect your results. When using different data sets, the samples may actually come from somewhat different populations, and the measurement instrument may be actually different (Boele et al., 2023). When you use different degrees of aggregation and analyze these (Bülow et al., 2025), you may need to think about these aggregates as a different variable than the original measurements: A person’s daily happiness score is not the same variable as a person’s weekly average happiness score. While this is to some extent exactly the point of comparing results at different timescales, the effects of such changes can be rather unexpected and require further study (e.g., Mulder et al., 2025).
3.3 Continuous time perspective
The continuous time perspective implies that instead of focusing on the dynamics that can be observed for the particular time interval between the repeated measures (e.g., an hour in ESM data, or a day in daily diary data), you consider the time interval between measurements as somewhat arbitrary: Observations are made in discrete time, but the underlying process does not unfold in discrete steps over time, but rather, it operates continuously over time (Ryan et al., 2018).
This perspective fits very well with ESM and similar forms of data collection, where the observations are obtained at semi-random time intervals (Haan-Rietdijk et al., 2017). To study dynamic relations in such data, it can be useful to take a continuous time approach in your analysis (Driver et al., 2017). The continuous time perspective also fits well in general with the idea that processes do not seize to exist when we do not measure them, and that there is no one particular interval at which the process varies.
However, the continuous time perspective is not the final solution to having multiple timescales: In general, a continuous time approach in the analysis helps to stretch the focus from one particular time interval to a continuum of time intervals, but it remains within a particular range of timescales. For instance, when you have ESM data obtained at intervals ranging from 10 minutes up to a few hours, it is not really possible to determine what happened from minute to minute or second to second; and similarly, when these data were obtained over a period of five days, it is not possible to determine what happens from week to week, or beyond.
Thus, while the continuous time perspective is valuable, it is not a solution to the problem sketched in this article: Even when you plan to take a continuous time modeling approach, you still need to consider the (range of) timescales at which the process you are interested in operates, fluctuates, and changes, and obtain your observations accordingly.
4 Think more about
The fact that processes may operate differently at different timescales, is actually ancient wisdom. This is reflected by a saying like “winning the battle but losing the war”: Successes that are obtained in the short-run, may not translate into victory over much longer periods of time. Another example is the saying “trust comes on foot and leaves on horseback”, which implies that gaining someone’s trust is a slow and gradual process, whereas losing it can happen quite quickly and suddenly. Another wisdom that parents of young children share is “the days are long, but the years are short”.
It is therefore important to realize that results from a study always should be interpreted in light of the timescale to which the study pertains. This is determined by the choices in your temporal design, such as the time frame of measurements, the time interval between measurements, and the number of measurement occasions; this determines the time span of a study (i.e., the degree to which we zoom in or out), but also the granularity of your view on the process (i.e., the degree of detail of the temporal pattern). It also depends on the particular analysis approach that you choose. Together, such choices determine what kind of fluctuations you get to see—and which ones will remain out of sight for you.
In general, you should expect to find different patterns at different timescales and when using different degrees of granularity (Hamaker, 2023). Similarities across timescales should thus not a be an a priori assumption, but could be a research question of your study.
Moreover, multiple timescales may also be important in your study when you are interested in connecting constructs that unfold at different timescales. For instance, you may be interested in the way momentary affect and sleep quality are related to each other, where the latter is measured once every 24 hours upon awakening, while the other can be measured with ESM multiple times throughout the day. Such scenarios pose particular challenges in the analysis of the data (Berkhout et al., 2025).
5 Takeaway
Psychological processes are likely to unfold at different timescales, and these timescales are also likely to interact. How you measure a process and how you analyze the data, determines to a large extent the (range of) timescale(s) at which you are able to investigate the ongoing process.
It may not be easy to make informed decisions about how to measure and analyze a process exactly, and different researchers are likely to make different choices in this. As a result, they may end up observing different features of processes and drawing different conclusions. Being aware of this, is an important first step in reaching a broader understanding of processes.
The temperature example provided in this article shows that a specific measure—here temperature—can serve as an observation of various processes: You can use it to make forecasts for tomorrow, or next week; you may also use it to study the seasons; or you may use it to investigate climate change.
Studying psychological processes using ILD is very similar, as there may be different patterns that you can observe at different timescales; these may be informative about different processes that are operating underneath, driven by different factors. Thinking about what it is you are interested in, what timescales are relevant for this, and how you can tap into these, is an important challenge to tackle in your study.
6 Further reading
We have collected various topics for you to read more about below.
- Temporal lens
- Temporal design and its consequences
- Temporal pattern
- [Timing of measurements]
- Measurement-burst design
Acknowledgments
This work was supported by the European Research Council (ERC) Consolidator Grant awarded to E. L. Hamaker (ERC-2019-COG-865468).
References
Citation
@article{hamaker2025,
author = {Hamaker, Ellen L. and Berkhout, Sophie W.},
title = {Multiple Timescales},
journal = {MATILDA},
number = {2025-05-23},
date = {2025-05-23},
url = {https://matilda.fss.uu.nl/articles/multiple-timescales.html},
langid = {en}
}