Timescales of massive human entrainment

Abstract

The past two decades have seen an upsurge of interest in the collective behaviors of complex systems composed of many agents entrained to each other and to external events. In this paper, we extend the concept of entrainment to the dynamics of human collective attention. We conducted a detailed investigation of the unfolding of human entrainment--as expressed by the content and patterns of hundreds of thousands of messages on Twitter--during the 2012 US presidential debates. By time-locking these data sources, we quantify the impact of the unfolding debate on human attention at three time scales. We show that collective social behavior covaries second-by-second to the interactional dynamics of the debates: A candidate speaking induces rapid increases in mentions of his name on social media and decreases in mentions of the other candidate. Moreover, interruptions by an interlocutor increase the attention received. We also highlight a distinct time scale for the impact of salient content during the debates: Across well-known remarks in each debate, mentions in social media start within 5-10 seconds after it occurs; peak at approximately one minute; and slowly decay in a consistent fashion across well-known events during the debates. Finally, we show that public attention after an initial burst slowly decays through the course of the debates. Thus we demonstrate that large-scale human entrainment may hold across a number of distinct scales, in an exquisitely time-locked fashion. The methods and results pave the way for careful study of the dynamics and mechanisms of large-scale human entrainment.

Fig 1. Excerpt of the waveform and related transcript from the first presidential debate.

Blue highlighting indicates Obama speech turns, red Romney’s and grey Lehrer’s (the moderator). The transcripts were retrieved from the National Public Radio website, cleaned and edited to better reflect the audio files. Start and end time of each speech turn as well as interruptions are aligned to the debate by combination of careful coding and automated processing.

Fig 2. Tweet rate and turn-taking during the presidential debates.

Light red and blue rectangles are periods of time during which candidates were speaking during the debates. Darker red and blue dots represent per-second tweet rate mentioning the corresponding candidates. Visual inspect reveals relatively periodic patterns of Twitter mentions that seem to be cued by turn onset. Plots include both tweets and retweets in the tweet / s rate.

Fig 3. Effects of taking and holding the ground on Twitter mentions.

Starting from the onset of each turn per candidate, plots show relative proportion of Twitter mention rises during that candidate's turn. While others are speaking, proportion mentions drops. Proportions are based on, for example, dividing mention to "Obama" divided by the sum of mentions to "Obama" and "Romney" together. Importantly, these plots only include original tweets, showing the anticipated effect is independent of retweets.

Fig 4. Effects of interruptions on Twitter mentions.

At the onset of speaking, results show that the volume of tweets increases when that spoken turn is in the form of an interruption. Each panel represents the results from one of the debates. Importantly this figure only shows original tweets, omitting retweets.

Fig 5. The temporal profile of public attention to salient events.

At the onset of a salient event, mention of the word (in the context of either "Obama" or "Romney") rapidly rises within 10 seconds (left panel). Mentions are max scaled to facilitate comparison. Right panel shows retweets separately from original tweets, showing the expected delay. Interestingly, these salient events show distinct temporal signatures in their onset and rise to maximum, both in the profile of tweets and retweets. For original tweets, first mention for Big Bird, binder, and bayonet respectively is 4, 5, and 11 seconds; their maximum is achieved at 42, 23, and 67 seconds. In the retweet data, this is lagged, with first retweets at 31, 14, and 17 seconds; maximum achieved at 99, 80, and 78 seconds, respectively for Big Bird, binder, and bayonet.

Fig 6. A model of public attention to salient events.

The model of public attention reactions to salient events as fit to the three case studies: “Big bird”, “binder” and “bayonet,” from left to right. Note two interlocked timescales: a saliency/novelty followed by the establishment of a meme that sustains a base-level of continued attention. In each panel R ² indicates the fit of the model, s is the point (in seconds) at which tweet rate is increasing maximally for the “meme,” m reflects the slope of that rate, and λ reflects the decay rate. For more details on the equation cf. methods section.

Fig 7. Multiple regression fits for all 3 debates.

Variance accounted for by salient events, a quadratic time term, who is talking, and whether interruption is taking place accounts for between 42% and 53% of the variance in observed tweet rate. See text and Supporting Information for further details on the model.