April 18, 2021 — Open review of preprint by Azhari and colleagues:
Azhari, A., Bizzego, A., & Esposito, G. (2020). Father-child dyads exhibit unique inter-subject synchronisation during co-viewing of animation video stimuli. bioRxiv.
Azhari and colleagues use fNIRS to demonstrate that father-child dyads exhibit greater response synchrony in prefrontal cortex while co-viewing short movies relative to synthetic control dyads. The manuscript is generally well-written; however, there are several places where we need a more detailed description of the analysis. There are also some thorny interpretational limitations with this kind of study that need to be clearly discussed.
There’s a troubling confound in this design where, unlike the synthetic control dyads, the father-child dyads are actually in the same space at the same time. So any external factor could drive greater real-time synchrony than seen in control dyads—for example, if the father tenses up at a particular moment in the movie or the child starts to get uncomfortable and move around at some point. None of these idiosyncrasies will be present in the control dyads as they are computed across separate sessions. These sort of simple same-place-same-time explanations aren’t uninteresting in their own right (e.g. Cheong et al., 2020), but it seems impossible to rule these out in favor of more cognitive explanations that father and child share a common script for processing the stimuli. The authors mention this in Discussion (page 10 line 10, page 11 line 6), but I think this limitation could be discussed more thoroughly (for example, another approach would be to deploy a two-group design where in one group children are paired with an experimenter, and in the other children are paired with their father).
We need more details on how exactly the “control synchrony” dyads were constructed. Is there only one control dyad assigned to each true dyad? Or are all non-true pairs computed? If the latter, with N = 29, this will result in 29 true dyads and N * (N – 1) / 2 = 406 control dyads; is this correct? Are all of these control dyads supplied to the statistical test? Is the paired structure retained in the statistical test (i.e. difference between true dyad and the control dyads that share a father).
I don’t fully understand the implementation of the GLM described on page 6. I would reiterate whatever GLM specifications are inherited from reference 18 (i.e. what “GLM settings”?). Is the “Gaussian Full Width at Half Maximum (FWHM) 4 model” smoothing in time (i.e. across time points) or space (i.e. across channels)? What does the “4” mean here? I would also better explain the motivation for this analysis in the first place. When I started reading this section of the methods, I was struggling to understand why computing a single regression coefficient for an entire 60-second movie stimulus could be useful.
What do metrics such as “video complexity” and “audio fundamentals” actually capture? The authors should make clear that these metrics were computed to demonstrate that the stimuli are relatively similar, but the metrics are not used in further analyses.
I’m not really an fNIRS expert, but I think it would be helpful to provide a bit more detail on how preprocessing was conducted. For example, you say “motion correction of discontinuities and spike artefacts were conducted”—but how is this actually implemented in the NIRSLab software (e.g. curve-fitting for despiking? ICA?)?
The sample size here (N = 29 subjects) is pretty small (and only 3 minutes of stimuli/data per subject). This may be sufficient for the contrast between true and control dyads, but it seems like inter-individual differences (e.g. the age covariate) should be interpreted with caution in a small sample. The authors should be very explicit about this limitation.
The authors should be cautious about reverse inference in the discussion. For example, in the paragraph starting at page 10 line 29, just observing higher activation for the movie stimuli in PFC doesn’t necessarily entail that the “mentalizing” process is taking place.
Seems like the high-pass filter with 128-second cutoff (page 6, line 8) is probably redundant with the prior bandpass filtering step—but I don’t think it should make any difference.
I assume this is a non-overlapping sample of children from the authors prior work with mother-child dyads (reference 18), but this should be noted explicitly.
Table 1: I assume this is monthly household income?
Cheong, J. H., Molani, Z., Sadhukha, S., & Chang, L. J. (2020). Synchronized affect in shared experiences strengthens social connection. bioRxiv.