Extended Data Fig. 4: The effect of virtual communication on forward flow across the progression of idea generation.

There was a significant interaction between modality and the position of an idea in the pair’s idea sequence on forward flow score across all studies (linear mixed-effect regression, n = 9966 idea scores, interaction term: b = −0.01, s.e. = 0.01, t₃₅₈ = −2.09, P = 0.038). At the beginning of the idea generation task, ideas generated by in-person and virtual pairs were similarly connected to past ideas generated by each pair. However, by the eleventh idea, ideas generated by in-person pairs began to exhibit significantly more forward flow (that is, the ideas were less semantically associated) compared to those of virtual pairs (linear mixed-effect regression, n = 9966 idea scores, simple effect of modality on forward flow at the 11^th idea: b = −0.12, s.e. = 0.06, t₆₂₁ = −2.00, P = 0.047). Thus, in-person pairs generate progressively more disconnected ideas relative to virtual pairs. See Supplementary Information D for model assumption tests of normality and heteroskedasticity. We truncated the graph at 30 ideas to provide the most accurate representation of the majority of the data. All tests are two-tailed and there were no adjustments made for multiple comparisons (for a discussion of our rationale, see Supplementary Information S).