For three months, I tried to fit our longitudinal survey data into a neat, three-step mediation model linking communication frequency to retention.
It simply refused to hold. The statistical significance vanished every time we controlled for negative feedback cycles. It was a frustrating dead end, but it forced us to look at the quality of communication — specifically reciprocity — rather than just raw volume. The mess was telling us the truth.