How Communication Format Shapes Informational and Social Decisions

The same words may not create the same interaction.

Communication channels differ in the number and type of cues they make available. Live video can convey a speaker’s face, gaze, gestures, and contingent nonverbal responses, whereas voice chat preserves the spoken message but removes many visual cues. Media Richness Theory describes this variation as a channel’s capacity to carry multiple cues, support rapid feedback, and personalize communication (Daft & Lengel, 1986; Daft, Lengel, & Trevino, 1987).

Research on children’s selective trust has typically emphasized evidential information: whether a speaker has been accurate, knowledgeable, or able to provide a good explanation. Yet children’s testimonial decisions are also sensitive to the immediate context in which information is delivered. For example, direct address can affect whether children spontaneous protest and later endorse counterintuitive testimony even when it does not provide new evidence that the claim is true (Corriveau & Kurkul, 2014; Lane & Harris, 2014; Li & Koenig, 2020).

This created two related gaps. First, relatively little developmental work had treated the communication medium itself as an ecological constraint on which cues children can use to interpret a speaker’s intent and commitment. Second, epistemic trust—relying on someone for information—and social trust— relying on someone to follow through—were often examined in separate studies or reduced to explicit ratings.

The purpose of my study was therefore not simply to ask whether video was “better.” It was to test whether changing the channel, while holding the speaker and verbal information constant, altered distinct decisions about the same person: whether children spontaneous protestd her, revised a prior belief, and persisted while relying on her stated intention to return.

Changing the channel while holding the interaction constant

Children were randomly assigned to live video chat or live voice chat. Thirty children participated in the video condition and 32 in the voice condition. Within each age group, 15 children were assigned to video and 16 to voice.

The approximately 25–30 minute Zoom session included an introduction, a testimony phase, a delay-of-gratification phase, and a post-testimony phase. The focal researcher, E1, followed the same sequence and used the same verbal materials in both conditions. The central manipulation was whether E1’s camera was on or off. A second researcher, E2, appeared later with her camera on in both conditions.

During the categorization task, children first labeled ambiguous hybrid animals and objects. E1 then provided a counterintuitive label that conflicted with each child’s initial response—for example, calling a rabbit-squirrel hybrid the category the child had not selected. The task adapted established counterintuitive-testimony paradigms (Jaswal, 2004; Li & Koenig, 2020). The hybrid stimuli were created with DALL·E for use in the remote task.

E2 later reminded each child of both answers and asked for a final categorization. This design separated the moment of exposure from the later belief-revision decision and reduced the likelihood that children changed their answer merely because E1 asked them again.

Between the two testimony phases, E1 offered a choice between an immediate plain certificate and a delayed certificate that the child could customize with colors and stickers. She stated that she would return, and children could wait for up to eight minutes or end the wait early. The researcher did not use the literal phrase “I promise,” so this was an implicit, future-oriented commitment rather than an explicit promise.

Waiting was treated as a behavioral index of social trust in E1’s return, while recognizing that delay behavior also depends on attention and self-regulation (Kidd et al., 2013; Pesch & Koenig, 2018; Ma et al., 2018). Regardless of how long a child waited, all participants were ultimately allowed to customize their certificate before the post-testimony phase.

Matching each model to the structure of the decision

Belief revision and protest were repeated binary decisions rather than one-time participant scores. I therefore used trial-level generalized linear mixed-effects models with a random intercept for child. This preserved within-child variation and accounted for the fact that responses from the same participant were not independent.

Before testing the main effects, I examined whether counterbalancing sequence, child sex, or the identity of the researcher serving as E1 improved model fit. They did not, so the final analyses were collapsed across those procedural factors. I also tested Age Group × Condition interactions and trial type. Neither improved the belief-revision or protest models, so I retained the more interpretable main-effects models.

I then tested whether protesting during the testimony phase predicted later endorsement of the same counterintuitive claim. This analysis was central to distinguishing protest as outright rejection from protest as a possible form of active epistemic engagement.

Waiting required a different approach. All 31 seven-year-olds completed the full 480-second interval, whereas only 14 of 31 five-year-olds did so. The absence of any stopping events among seven-year-olds created a ceiling effect and prevented a valid age-by-condition moderation analysis.

I therefore analyzed the five-year-old subgroup using Firth-penalized Cox regression. Ending the wait before 480 seconds was the event; completing the interval was treated as right-censored. Firth penalization reduced bias from the small subgroup and uneven event distribution. The model tested condition while controlling for age in months.

What the pattern may mean—and what it does not prove