Social signalling explains almost everything and predicts little.

In this case, the social signaling interpretation predicts a discrepancy between peoples' expressed preferences in distant situations, and peoples' felt responses in situations where they can act.

We can acquire evidence for or against the social signaling interpretation by e.g. taking an "underdog" scene, where a popular kid fights with a lone unpopular kid, and having two randomized groups of kids (both strangers to the fighters): (a) actually see the fight, as if by accident, nearby where they can in principle intercede; or (b) watch video footage of the fight, as a distant event that happened long ago and that they are being asked to comment on. Watch the Eckman expressions of the kids in each group, and see if the tendency to empathize with the underdog is stronger when signaling is the only issue (for group (b)) than when action is also a possibility (for group (a)). A single experiment of this sort wouldn't be decisive, but with enough variations it might.