Leave the screen: The influence of everyday behaviors on self-reported interoception

Study at a glance

• Authors: Amie Wallman-Jones, Carina Nigg, Valentin Benzing, Mirko Schmidt
• Year: 2023
• Journal: Biological Psychology
• DOI: 10.1016/j.biopsycho.2023.108600
• N: 70
• Design: 7-day ambulatory assessment (accelerometry-triggered smartphone e-diary), multilevel within/between-person modelling; laboratory baseline heartbeat-counting task

Key findings

• Half the variance in interoception is within-person. ICC = 0.51 for state self-reported interoception — the first datum in the wiki bearing directly on whether the field’s constructs are traits at all.
• Momentary physical activity raises self-reported interoception (B = 0.0025 per millig, p = .013; ≈0.88 points on a 0–35 scale at walking intensity), momentary sedentary behaviour lowers it (B = −0.06 per minute, p = .009; ≈0.90 points over a full sedentary quarter-hour).
• The same measure reverses sign between timescales. Habitual weekly physical activity predicted lower self-reported interoception (B = −0.05, β = −.29, p < .001) — the largest continuous effect in the model, and opposite to the momentary effect in the same 70 people.
• Activity type matters more than intensity. Against a screen-time reference: exercise +4.48, daily-life physical activity +1.21, non-screen social interaction +1.13, non-screen sedentary behaviour +0.67, sleeping −1.73 (all p ≤ .004). Two sedentary categories beat screen time, so the effect is not sitting per se.
• Baseline heartbeat-counting accuracy positively predicted daily self-reported interoception (B = 4.50, β = 0.21, p = .007) — an accuracy↔sensibility association, which most of the wiki’s evidence says should be absent.
• And it moderated the activity effect in the opposite direction. Low- and medium-IAcc tertiles showed activity-related increases; the high tertile showed decreases (interaction B = −0.01, p = .004), replicating the authors’ own laboratory result (Wallman-Jones et al. 2022).

Limitations

• The interoception measure was not built to measure interoception. The 6-item body subscale of the State Mindfulness Scale (Tanay & Bernstein 2013); the authors concede one item (‘sensations caused by my surroundings — heat, coolness, wind on my face’) is frankly exteroceptive, and that the subscale straddles self-reported accuracy and attention, so what the dependent variable indexes is undetermined.
• No non-interoceptive control questionnaire, so nothing distinguishes a rise in bodily attention from a rise in attentiveness generally — which is what a heatmap of ‘exercise scores highest’ would look like either way.
• The sedentary result is cut-off-dependent: significant with the paper’s 187-millig criterion and with a 100-millig criterion, non-significant with body-position-only or a 350-millig criterion (direction stable in all three).
• Intensity was never measured, so the paper’s own central mechanistic question — attention to the body vs salience of the signal — is left explicitly unresolved (‘we can only speculate’).
• The baseline accuracy measure is the heartbeat-counting task, whose validity the authors themselves flag via Corneille et al. (2020), Desmedt et al. (2018, 2022) and Adams et al. (2022).
• N = 70 young, healthy, normal-BMI adults (M age 21.7); no a priori power analysis (the authors say so), and they concede the sample may be too small for the cross-level interaction the tertile analysis rests on.
• Observational throughout. Nothing was assigned; people who exercise differ from people who scroll in every other way too.

Connections

• Methods: Ecological Momentary Assessment (EMA), Heartbeat detection task
• Concepts: Physical activity, sedentary behaviour and interoception, State vs trait interoception, Interoceptive taxonomy (Farb et al. Box 1), Interoceptive sensitivity, Interoception, Insular cortex
• Debates: Is more interoceptive awareness better?, Is the heartbeat-counting task a valid measure of cardiac perception?
• Related studies: Banellis et al. (2026), MacCormack et al. (2021), Tong & Jia (2017), Harrison et al. (2021), Khalsa et al. (2018)

Open access (CC BY). Seventy University of Bern students wore thigh-mounted accelerometers for seven days; the sensor triggered a smartphone prompt after ten minutes of light activity or thirty minutes of uninterrupted sitting (random prompts otherwise, capped at ten per day, 07:30–21:30). At each prompt they rated the last fifteen minutes on the body subscale of the State Mindfulness Scale and named the predominant activity. Baseline session: height, weight, questionnaires, and a three-trial Schandry counting task.

The wiki’s first source on exercise, sedentary behaviour or screen time, and its first attempt to measure interoception repeatedly in ordinary life.

The finding the paper is named for, and the finding that matters more

The title result is the screen-time contrast, and it is clean: against screen time as reference, every other category scored higher except sleeping, including two sedentary categories. Sitting is not the problem; sitting at a screen is. The authors read this through Pennebaker & Lightner’s (1980) competition-of-cues hypothesis — attention to one source depletes it for others — and note that non-screen social interaction scored higher than solitary non-screen behaviour, which they connect to interoception’s social origins via Oldroyd et al. and Palmer & Tsakiris (2018).

But the result that reorganizes things here is one the authors treat as a puzzle in their discussion: the within-person and between-person effects of physical activity have opposite signs. Move now, and self-reported interoception goes up. Be someone who moves a lot, and it is lower. Same measure, same participants, same model. See State vs trait interoception for what that does to the wiki’s construct pages, and Physical activity, sedentary behaviour and interoception for the mechanisms on offer.

The exploratory interaction, and why it is the most interesting thing in the paper

Baseline heartbeat-counting accuracy did two things at once.

As a main effect it predicted higher daily self-reported interoception (B = 4.50, β = 0.21, p = .007). This wiki has recorded the accuracy/sensibility dissociation as one of its most robust patterns — meditators attend without detecting (contemplative training increase objective interoceptive accuracy?), confidence generalizes where performance does not, the Interoceptive taxonomy (Farb et al. Box 1) exists because the two come apart. Here they correlate. Not a contradiction (different sensibility instrument; a state measure aggregated over a week rather than a trait questionnaire; N = 70 with a wide CI, 1.31–7.69), but it is friction worth holding, and it is recorded as such on Interoceptive taxonomy (Farb et al. Box 1) rather than smoothed away.

As a moderator it reversed the activity effect. Split into tertiles, the low and medium groups’ self-reported interoception rose with acute activity; the high group’s fell (Fig. 1). The authors offer two readings — habituation (familiar signals are less perceptually salient, so a trained perceiver notices less at the same intensity) and a ceiling effect — and cannot separate them. Either way this is the same shape as their between-person result one level down, which is the strongest internal support the paper has for the habituation account: the person who reads their body well, and the person who moves a lot, both report noticing it less.

That gives Is more interoceptive awareness better? an unusual row. Most rows there read one verdict off one association. This one, like Banellis et al., produces opposite verdicts inside a single sample — and it is now the second such row, which makes the timescale problem structural rather than an oddity of one study.

What the mechanism section can and cannot support

Two mechanisms are proposed throughout and never separated:

• Salience. Sympathetic activation during activity makes bodily signals louder (Craig 2006); Easterbrook’s (1959) cue-utilization narrowing lets them dominate consciousness.
• Attention. Goal-directed exercise directs attention to the body (pacing breath, placing limbs; Toner et al. 2016) where automatized daily activity does not.

The exercise > daily-life-PA gap is offered as evidence for attention — but the authors concede immediately that exercise is also more intense, and intensity was not recorded, so the gap is equally a salience effect. The paper’s own conclusion (“we can only speculate”) is the right one, and the wiki should not resolve it on their behalf. This is the same undecidability Physical activity, sedentary behaviour and interoception carries as its central open question.

The sedentary story leans on an “untrained pathway” argument from the group’s own theory paper (Wallman-Jones et al. 2021): absent exercise-induced perturbation, brain–body communication degrades toward a mismatch between actual and perceived bodily state. The supporting evidence is all held at one remove and none of it is longitudinal — reduced heartbeat-evoked potentials in sedentary vs trained young adults (Perakakis et al. 2017, a bioRxiv preprint), attenuated anterior insula response to inspiratory loading in elite athletes (Paulus et al. 2012), BMI–interoception correlations (Robinson et al. 2021). Note that the Paulus finding is cited twice, for opposite purposes: as evidence that sedentariness is bad, and later as the explanation for why habitually active people report less interoception. It cannot be both without an account of when attenuation is efficiency and when it is decay, and the paper does not supply one.

For screen time specifically the evidence is thinner still: two resting-state connectivity studies (Paik et al. 2019 on bedtime smartphone use and the insula; Kwon et al. 2022 on problematic smartphone use and cingulate connectivity), plus a theoretical piece from outdoor-education (Puk 2021). The screen-time claim in this paper’s title rests on this paper.

What it contributes as a method

It is the wiki’s first use of accelerometry-triggered ambulatory assessment — prompts fired by measured behaviour rather than by clock or randomness — which is what makes the within-person activity contrast estimable at all. See Ecological Momentary Assessment (EMA), where it now sits alongside Tong & Jia’s beeped EMA, MacCormack’s Day Reconstruction Method and MDES as a fourth point on the same spectrum.

Its instrument problem is worth stating plainly, because it is the field-level gap rather than this paper’s failing. There is no validated state measure of interoception. The authors reached for a mindfulness scale, checked its multilevel reliability properly (ω_within = 0.92, ω_between = 0.97 by MLCFA), and then said in their limitations that the field needs the instrument they did not have. That is the same bottleneck Chen et al. name for non-verbal populations and Interoceptive taxonomy (Farb et al. Box 1) records for regulation: a construct with four competing partitions and, in this case, no clock.

Provenance notes

The corresponding address is Bern but the corresponding e-mail is UCSF — Wallman-Jones had moved by publication. No funding declared, no competing interests, and an explicit statement that no generative AI was used in preparation. Data available on request; six supplementary files (activity-category mapping, MLCFA loadings, interaction models, tertile descriptives, sedentary-cutoff sensitivity analyses) are referenced but not in raw/, which matters for the cut-off sensitivity result above — it is known here only through the authors’ summary of it.

One typographical error in the results-to-discussion transition (“The results of the present study vuild on findings from previous laboratory-based studies”), recorded in the same spirit as the Pollatos statistics and the Lyons transposed table: production errors are common in this literature and worth noticing when they appear near load-bearing sentences. This one is not — it is a verb.