The core difficulty is that behaviours can coincide by pure chance. If one person laughs frequently, any given laugh is likely to be followed shortly by someone else’s laugh simply because laughs are common, not because one triggered the other. Distinguishing genuine transmission from coincidental co-occurrence is therefore the first challenge any contagion measure must meet. A second problem concerns time. Contagion unfolds across a range of latencies, from milliseconds for reflex-like mimicry to minutes for slower, more structured behaviours. Methods that rely on a single, predefined time window risk missing this variability or, worse, conflating rapid responses with extended processes. Third, individuals differ in their baseline tendency to produce a behaviour, and these base rates fluctuate across contexts; without correction, frequent behaviours inflate contagion estimates. Finally, contagion is embedded in social structure. Group size, relational ties and the direction of influence all shape how behaviours spread, and effects in groups of three or more cannot simply be reduced to an aggregation of dyadic links.

SPREAD tackles all four challenges within a single statistical architecture built on permutation testing. The workflow begins with event-based time series: each behavioural occurrence is recorded as a discrete, time-stamped onset within an observation window. For every focal event, the framework computes whether the same behaviour appears in other group members across an expanding range of latency thresholds, evaluated in the demonstrations from one to sixty seconds in one-second increments. Crucially, rather than asking whether a response occurred within one fixed window, SPREAD estimates contagion likelihood as a continuous function of lag, producing a latency-resolved contagion profile that reveals the temporal signature of the effect without imposing arbitrary boundaries.

The second step builds the null model against which the observed data are judged. For each individual, event timestamps are randomly permuted within the original interaction interval, independently for each person. This procedure preserves each individual’s total number of events and the overall interaction duration, but destroys the temporal alignment between people. Because the permutations retain individual base rates while randomising co-occurrence, any excess clustering in the real data cannot be explained by how often someone simply tends to laugh or eat. The researchers repeated the permutation 1,000 times in their analyses and showed that running 10,000 iterations produced nearly identical results, indicating that the null distributions are stable with respect to the number of permutations.

In the third step, observed contagion probabilities at each latency are compared with their corresponding permutation-based null distributions and standardised as Z-scores, yielding latency-resolved measures of both the magnitude and the direction of contagion. Positive values indicate facilitative effects, in which one person’s behaviour raises the likelihood of the behaviour in others; negative values indicate inhibitory effects, a phenomenon that has received far less attention but may be an important dimension of social influence. Alongside the Z-scores, SPREAD reports the Incidence Rate Ratio, a multiplicative effect size expressing how many times more likely co-occurrence is in the observed data than under the null model. An IRR of two, for example, means the observed count is twice the chance expectation, while a value below one signals reduced co-occurrence. Confidence intervals for the ratio are derived directly from the empirical permutation distribution using the percentile method.

Beyond overall contagion, the framework decomposes effects according to the number of responding group members, preserving the relational structure of the interaction. In a triad, a focal event may be followed by a response from one partner, a dyadic effect, or from both partners simultaneously, a triadic effect that cannot be reconstructed from pairwise statistics alone. The same logic generalises to larger groups by counting or proportioning responders. SPREAD also supports analyses conditioned on social roles, such as familiar versus unfamiliar partners or leader-follower structures, and it captures directionality through complementary forward-looking and backward-looking analyses. In forward-looking analyses the focal individual is treated as the emitter, quantifying the influence that person exerts on subsequent behaviour in others; in backward-looking analyses the focal individual is the potential receiver, quantifying how prior behaviour by others shapes the focal person’s own actions.

To validate the approach, the team applied SPREAD to two datasets that place very different demands on temporal resolution and social sensitivity. The first came from a laboratory study in which 84 participants, forming 28 triads, held roughly fifteen-minute conversations over Zoom, discussing either relaxed or tense topics. Laughter events were newly extracted and coded from the audio recordings, yielding 1,032 laughter events with substantial variation across groups and individuals. The observed probability that laughter by other group members followed a focal laugh rose from 17.1 percent within the first second to 38.2 percent within ten seconds and nearly half of all opportunities within thirty seconds. Relative to the permuted null distributions, the observed probabilities were markedly elevated at short latencies, and the effect peaked in the one-to-two-second window with an Incidence Rate Ratio of 3.60, meaning laughter was more than three and a half times more likely than chance. Significant effects persisted with decreasing magnitude until roughly 33 seconds after the focal event.

The decomposition by responder number proved revealing. When laughter spread to only one other person, contagion was detectable within the first second and remained significant up to about 17 seconds. When both other members of the triad laughed, the effect was far stronger at short latencies, with an IRR of 62.68 within the first second, and it persisted longer, until approximately 38 seconds. Contagion strength and temporal persistence thus scaled with the number of responding individuals, consistent with multi-exposure dynamics amplifying transmission. Control analyses strengthened the case that these patterns reflect genuine social influence. The researchers constructed artificial triads by combining individuals from different original interaction groups, eliminating genuine contagion by design, and found no sustained effects at either the dyadic or triadic level. A further check showed that within-individual repetition of laughter was actually lower than chance expectation, ruling out self-triggered carry-over as an alternative explanation.

The second dataset examined eating, a slower behaviour shaped by both biological regulation and social context. Here, 108 participants in 36 triads conversed spontaneously in the laboratory while eating identical slices of cake, with bite onsets coded frame by frame from video recordings, yielding 1,427 bite events. Eating contagion followed a distinctly different temporal profile. Significant deviations from chance appeared within the first second, built up to a maximum around 14 seconds, and gradually declined while remaining significant until 42 seconds, a far more extended timescale than laughter. Dyadic effects peaked at six seconds, whereas triadic effects peaked at 28 seconds and persisted until nearly a minute. Constructed-triad controls produced small positive effects confined to shorter windows, clearly distinguishable from the sustained clustering in real groups, and negative deviations seen in the real data were mirrored in the constructed groups, suggesting those reflected non-social temporal dependencies rather than inhibitory contagion. Together, the two case studies show that SPREAD recovers behaviour-specific temporal signatures: laughter contagion is sharp and fast, eating contagion slow and sustained.

The Konstanz team positions SPREAD as a methodological foundation for studying how behaviours propagate through social systems, in humans and other animals alike. Candidate applications range from yawning and facial expressions to vocalisations and feeding behaviour, and from joint task performance to collective decision-making. The framework can also help distinguish contagion from social facilitation, a related phenomenon in which behavioural rates rise at the group level without structured temporal dependencies between individuals. The authors are candid about limitations: the method assumes behaviours can be represented as discrete event onsets, which may not suit continuously expressed actions, and the permutation procedure treats event times as exchangeable within individuals, so within-individual temporal structure is not explicitly modelled, though additional permutation constraints could account for refractory periods or habituation. Future extensions may incorporate covariates, time-varying base rates and networked interaction contexts that include attention, speaker-listener roles and relationship structure. With the toolbox, the underlying data and the analysis code released openly under a CC BY 4.0 licence, the researchers hope that the study of behavioural contagion will move from heterogeneous, context-specific tests toward a shared, transparent standard for measuring one of the most fundamental dimensions of social interaction.

Subject of Research: A permutation-based statistical framework and open-source R toolbox for quantifying behavioural contagion in time-series social interaction data

Article Title: SPREAD, a framework and open-source R toolbox to quantify behavioural contagion

Article References: Straßheim, J., Bousquet, C., Köchling, J., Renner, B., & Schupp, H. T. (2026). SPREAD, a framework and open-source R toolbox to quantify behavioural contagion. Communications Psychology, 4(1), Article 130. https://doi.org/10.1038/s44271-026-00538-0

Image Credits: AI Generated

DOI: 10.1038/s44271-026-00538-0

Keywords: behavioural contagion, SPREAD, permutation test, R toolbox, social interaction, laughter, eating behaviour, null model, latency, social structure, open-source software, Communications Psychology