More activity, better welfare? Read the missing steps

Does more activity mean better welfare?

A step count, a distance traveled, an hourly activity score: none of these tell you whether an animal is doing better. They tell you that movement happened, and how much of it, within whatever window the study measured.

Turning that number into a welfare claim requires an extra layer of reasoning that the data alone can’t supply.

Consider what “more activity” could mean in practice. It might reflect curiosity, restlessness, pain-driven pacing, or a response to a novel object placed in an enclosure an hour earlier.

Two animals with identical activity scores could be in very different internal states. A tracker or observer logging movement has no built-in way to tell those states apart.

This is why the gap between measurement and inference deserves attention before a number gets repeated as a finding. An activity count describes behavior recorded during a set period, under set conditions, using a specific method.

Whether that behavior maps onto comfort, distress, boredom, or simple opportunity depends on the rest of the study design. That means how the groups were formed, how the behavior was recorded, and what controls were in place.

Design choices shape what a number can support. If animals were not randomly assigned to conditions, an observed difference in activity between groups might trace back to how the animals were selected, separate from any real treatment effect.

Guidance on evaluating environmental enrichment describes exactly this risk, and its example is worth having in mind. Fill the first group with the animals that are easier to catch and handle, it says. The difference you later measure may then only apply to the animals you picked. The easy-to-catch animals could be bolder and more confident than their cage mates, and they might respond to new enrichment differently. The resource on improving the scientific quality of enrichment evaluation walks through several of these bias points. They include how animals are selected, how cages are allocated, and how behavior gets recorded.

Activity data still has real value. A single figure, read in isolation, can’t carry the weight of a welfare conclusion on its own.

The interpretation has to come from somewhere: the method section, where the definitions and design controls actually live.

Read the study’s definition

The words “activity” and “welfare” rarely mean what a casual reading assumes. Each study operationalizes its own terms, and those operational definitions determine what the resulting number can and cannot support.

Activity might be defined as distance traveled per hour, frequency of a specific posture change, or time spent away from a resting area. Each definition captures something different.

A study measuring locomotion won’t necessarily capture vigilance behavior, and a study built around one enclosure type may not generalize to another setup entirely.

The same caution applies to behavioral recording itself. Coding decisions about what counts as a given behavior can be inconsistent. Preconceived ideas about what a behavior “should” look like can shape how observers categorize it, without anyone noticing.

The nc3rs guidance on recording behavior with minimized bias names the consequence directly. An inconsistent dataset is unreliable, and it could lead to unsuitable enrichment being introduced to the animal facility. That means the actual equipment placed in the room can be the wrong choice for that animal.

Reading past the abstract matters here. A results section often reports the measure precisely, while a summary or press description compresses it into something looser and more welfare-flavored.

A paper might report “activity” without stating what behaviors were counted, over what time period, or with what recording method. When that happens, there’s no way to know what claim the number is actually able to make.

A term like “activity” only supports a welfare conclusion when its definition, recording method, and design controls are stated clearly enough to check. Absent that detail, a reader is left inferring what the study itself never confirmed.

Practical questions worth asking of any activity-and-welfare claim:

• What behaviors counted toward the activity measure, and which were excluded?
• Over what time window was activity recorded, and was that window representative of the animal’s day?
• Were observers aware of which animals belonged to which group when they scored behavior?
• Was the allocation of animals to groups randomized, or based on convenience?

Each of these questions traces back to a specific design choice someone made when the study was built. The NC3Rs resource matches each stage with its own remedy. Randomisation covers how animals are selected and allocated. Testing observer consistency covers how behaviour gets recorded. Masking applies at the data analysis step. Actively seeking out alternative explanations belongs at the interpretation step. Knowing which remedy belongs where is what lets a reader tell a well-controlled study from one that mentions the word blinding somewhere.

Those steps don’t guarantee a correct welfare conclusion, but their absence is a reason to hold the conclusion loosely.

Look for more than one kind of evidence

A single activity measure, however carefully defined, is one data stream among several that welfare researchers might use. Triangulating across different types of evidence gives a more complete picture than any one number can offer on its own.

Physiological markers, physical condition, and behavioral diversity all speak to different aspects of an animal’s state, and they don’t always move in the same direction. An animal could show elevated activity while also showing signs of stress on another measure entirely.

Relying on activity alone, without checking whether other indicators tell a consistent story, risks mistaking one slice of behavior for the whole picture.

Study design plays a direct role in whether multiple evidence streams can even be compared fairly. If cages or animals were not allocated to groups through a randomized process, differences across measures become harder to trust. They might trace back to the allocation method itself, separate from any real effect of the enrichment or condition being tested.

The NC3Rs guidance addresses this with practical steps. It describes how to randomize the allocation of animals or cages to experimental groups using simple tools, including spreadsheet functions or free list-randomizing services. That way, group differences are less likely to trace back to selection bias.

Observation scheduling matters too, and the guidance is specific about why. Randomising the order of observations, the guidance says, can keep the animals from anticipating the observation period. It also controls for other factors that come from observing in the same order every time. Those factors include the circadian rhythms of the animals and the energy levels, or fatigue, of the observer.

The same page walks through building one. Ten cages, five control and five with a new enrichment item, get randomised into groups and then randomised again into morning and afternoon slots. That example gives a reader a concrete allocation procedure to compare with a methods section. If a paper leaves the procedure out, note the gap without guessing why.

And there’s a stage after analysis that the guidance counts as a bias point too, which readers almost never get to see. Under sharing results, it flags a common pattern: publishing findings where enrichment had a positive effect, but not where it had a negative effect or none at all. The consequence is that studies get repeated, because nobody knows they were already done. Its instruction is blunt: share your results regardless of the outcome. There is really no such thing as a negative result if the study is robust and can be replicated. A reader weighing one published activity finding has no way to count the unpublished ones sitting behind it.

A behavioral count paired with attention to allocation method, observation order, and at least one other type of measure gives a reader real grounds for confidence. A single number reported without any of that scaffolding gives much less.

A study might report only one measure and skip any description of its randomization approach. When that happens, the honest response is to treat the finding as preliminary, open to revision as more evidence comes in.

Keep household decisions separate

A research finding about activity and a decision about how to care for an animal at home are two different kinds of judgment. Blending them risks overstating what either one supports. Published measures describe what was observed in a specific study population, under specific conditions. A household choice about enrichment, exercise, or routine is a practical decision made under different constraints entirely.

Household ideas such as rotating toys are practical choices to try, not findings an activity study has proven. A household might choose to rotate toys, vary walking routes, or introduce new objects on a schedule. These are reasonable organizational choices a caregiver can make. None of them is validated by an activity study the way a controlled experiment validates its own specific finding.

The nc3rs framework for reducing bias in enrichment evaluation was built for laboratory and facility settings, with defined groups, randomized allocation, and structured observation schedules. A household doesn’t have a control group, and it doesn’t need one to make sensible day-to-day choices.

That distinction matters when a study’s language starts to resemble care advice. A paper might report that a particular enrichment item was associated with a change in a defined activity measure. That association holds under the conditions researchers set up and controlled for bias. Promising the same outcome in a different setting, with a different animal, observed a different way, goes beyond what that study can support.

Keeping these two categories apart protects against a common slide: reading a specific, bias-controlled research measurement as if it directly authorizes a general household rule. A study like that describes one outcome, under one design, with one definition of the behavior in question. Whatever a caregiver decides to try at home can be reasonable and still sit outside what the cited research actually established.