Why a Later-Day Pattern Can Mislead Us
For healthcare professionals making sequential medical judgments, a later-day change in urgency can look like the behavioral signature of decision fatigue. This chapter focuses on the narrower diagnostic question: what does a time-of-day pattern actually establish, what other causes travel with the clock, and what comparison would make cumulative effort easier to distinguish? The goal is not to erase the pattern, but to stop a schedule signal from doing the work of a causal explanation.
What the graph shows—and what it does not
Imagine a graph of medical calls arranged by hour. The vertical axis records how often professionals assign the highest urgency. If the line rises through the afternoon and evening, the picture is immediately legible: people may be making more conservative decisions after hours of demanding work. The pattern fits the intuition that repeated effort eventually changes judgment.
A large field study in Swedish telephone triage reported just such a descriptive pattern. In its target data, 14.8% of calls around 08:00 received the highest urgency rating, compared with 19.3% around 18:00 and 21.1% around midnight. Those percentages show that the distribution of urgency ratings differed across clock times. They do not show that cumulative mental effort caused the difference.
That distinction is not semantic. A descriptive association answers when an outcome is more common. A causal explanation answers why it is more common and what would happen if the proposed cause changed while other relevant conditions stayed similar. A graph organized by time can be accurate at the first level and misleading at the second.
One way to see the difference is to ask a counterfactual question. If the same kinds of patient calls had been handled by the same kinds of workers at the same clock times, but with less cumulative effort, would the urgency pattern have changed? The graph cannot answer that on its own because it contains no direct version of the same situation with the presumed cause altered and the other causes held close. A higher late-day percentage is compatible with the counterfactual answer being yes, no, or something more conditional.
The pattern can still matter. If patients are more likely to receive an urgent recommendation at night, a health service may need to understand the operational consequences: staffing, access, referral capacity, and patient communication may all be affected. But practical importance does not settle mechanism. A pattern may deserve attention because it is real even when the reason for it remains provisional.
Time of day is attractive as a fatigue proxy because it is easy to record and seems to summarize accumulated work. A morning call often occurs earlier in a shift than an evening call. A call after many preceding cases seems more likely to follow prolonged effort than a call at the start of work. Yet the clock is a container for many changing conditions. The same variable that marks possible fatigue also marks the movement of patients, services, staff, and institutional routines.
This is why “later decisions are different” cannot be translated directly into “fatigue changed later decisions.” The first claim belongs to the data as described. The second requires a comparison that deals with the other paths.
The clock carries other causes with it
The most direct alternative is patient composition. People do not call for the same reasons at every hour, and the distribution of symptoms or perceived severity may change as other sources of care open or close. If the calls arriving later are, on average, more difficult or more concerning, higher urgency may reflect the cases rather than a change in the professional’s willingness to process information.
Patient composition can change without anyone deliberately sorting cases. The operation of the healthcare system is enough. A patient who would contact a primary-care clinic during its opening hours may call a telephone service when that clinic is unavailable. A symptom that prompts immediate advice at one hour may be managed through another channel at another hour. The resulting calls can differ in urgency even if every professional applies the same judgment process throughout the day.
A second alternative is the surrounding service system. Opening hours of primary care, clinics, or other advice channels can affect who uses a telephone triage service and when. A call that would have been handled elsewhere during the day may be directed to the service at night. The organization’s own procedures may also differ across shifts. A time-of-day effect can therefore reflect a changing flow of work rather than a depleted decision-maker.
Shift composition creates another route. Morning, afternoon, evening, and night workers may not be interchangeable. They may differ in experience, schedules, training, or the kinds of work they usually do. If a particular group is more likely to handle later calls, an apparent time effect may partly reflect who is working rather than how long an individual has been working. Even when the same people rotate across shifts, the occasions on which they work may not be equivalent.
Call order does not solve the problem. It may seem to be a more direct measure of accumulated effort because it counts what came before, but call order is correlated with the clock. The sixth call of a morning shift and the sixth call of a night shift may arrive in different service conditions. A higher call number can also reflect a longer shift, a busier period, or a different schedule. Treating time of day and call order as independent confirmations would count the same entangled pattern twice.
The possibilities can overlap. Suppose urgency rises late in the day. More severe cases may arrive, some other service may close, a different set of nurses may be working, and the professionals may also have handled more calls. A statistical association cannot reveal which combination produced the outcome unless the design supplies additional leverage. The point is not that fatigue is absent; it is that fatigue is one candidate cause among several plausible causes.
The diagram is a diagnosis of the evidence problem, not a claim that every path operates in every service. It shows why the same later-day pattern can be generated by different combinations of conditions. Cumulative effort is the path relevant to decision fatigue, but patient mix, opening hours, routines, and staffing can reach the same outcome without passing through that mechanism.
There is also a measurement problem. A researcher may infer a break from a long gap between calls, but a nurse could have been completing paperwork, training, or another task during that interval. Conversely, a short gap does not prove that the preceding work was equally demanding. A proxy can be useful while remaining imperfect. The question is whether its errors are likely to obscure or imitate the effect being studied.
A further risk is that the observed outcome may be downstream of several decisions, not just the decision being studied. A nurse’s urgency rating can influence what happens next in the care pathway, while the data may not show whether that choice was appropriate for the patient. The time-of-day graph therefore describes a behavior in a service, not a complete measure of clinical quality. It should not be turned into a claim about harm without patient-relevant outcomes and a design that can support that claim.
The practical consequence is a change in the order of reasoning. First identify the observed pattern. Then list the causes that could produce it. Only after that ask what evidence would distinguish the causes. Starting with “fatigue is intuitive” encourages the graph to confirm the explanation; starting with competing causes makes the explanation earn its place.
What a more credible comparison holds constant
A stronger comparison does not require an impossible experiment in which a professional is tired and not tired at the same moment. It requires a setting where cumulative workload differs while other conditions are brought closer together. Two strategies in the field study illustrate the logic.
The first uses an overlap between morning and afternoon shifts. During part of the same early-afternoon window, morning workers have been on duty for much longer while afternoon workers have just started. Because calls are assigned to nurses in a pseudorandom order from the same pool of patients at that time, the comparison is less exposed to the ordinary change in patient mix across the entire day. The clock is held relatively close while cumulative time at work is separated.
This is “near random” in a limited, design-specific sense, not a randomized assignment of fatigue. Scheduling still determines which workers are present, and the analysis relies on rules for selecting the relevant overlap window. The advantage is that the comparison reduces one major source of confounding: high- and low-workload observations are made in a shared clock window rather than at opposite ends of the day.
The second compares intense call sequences before and after a break. The researchers focus on sequences with many calls, little time between them, and a defined break interval. Decisions before the break are treated as plausibly higher in cumulative effort than decisions after it. This comparison is also imperfect: the break may not be the only thing that changes, and non-call work may be hidden in the observed gap. But it asks a more targeted question than “are evening decisions different from morning decisions?”
The design principle is easy to state: keep the context close, separate the presumed cause, and specify the outcome before looking at the result. In practice, that means paying attention to at least three dimensions.
Comparable cases. If high- and low-effort observations contain different kinds of patients, a judgment difference can be attributed to case composition. A comparison drawn from the same flow of calls or a narrow time window can reduce, though not eliminate, that problem.
Separated cumulative workload. The high- and low-effort conditions must differ meaningfully in the amount of preceding work. If the contrast is too small, a genuine effect may be hard to see. If it is defined only by a clock label, it may be unclear whether the relevant mental effort differed at all.
A specified behavioral signature. The study must say what should move and in which direction. A decision-fatigue claim about greater reliance on personal defaults is different from a claim about higher urgency, consultation, call duration, or errors. A design that measures everything and selects the most compatible result after the fact can produce an impressive pattern without a stable interpretation.
Preregistration helps with the third dimension. It restricts decisions about analysis and interpretation before the new data are available. It does not make the design randomized, remove measurement error, or guarantee that the chosen proxy captures subjective fatigue. It does make it harder to redefine the hypothesis after seeing whether urgency rose, fell, or stayed similar.
A large sample helps in a different way. It can make a small association easier to estimate, but precision around a confounded association is still not causal identification. More observations do not compensate for a comparison that leaves the main alternatives untouched. Design and sample size answer different questions: one concerns where the contrast comes from, the other concerns how precisely it can be measured.
The time-of-day pattern therefore has two legitimate uses. It can identify an operational difference worth investigating, and it can suggest where a stronger comparison might be found. What it cannot do alone is certify cumulative decision effort as the cause. That remains true even when the pattern is large, intuitive, and repeated.
A real schedule effect can be useful while the fatigue diagnosis remains provisional. A healthcare organization might respond by examining patient access, staffing, handoffs, and workload rather than by assuming that a single psychological mechanism explains the result. That response is not indecision. It is a way to preserve the observed problem while keeping the proposed cause open to testing.
The same caution applies when evidence is gathered across many studies. If one study uses time of day, another uses appointment order, and a third uses time since a break, their positive findings may not be interchangeable. They may be pointing to a shared effect, to different effects, or to different confounds. The next task is to look across the healthcare literature without counting all of those signals as one uniform test of decision fatigue.