Four-stage simulation: when negative engagement turns positive
A four-stage simulation applied to the Xiaomi Community tests how price, product quality, service quality and personal needs met shape positive and negative customer engagement, and whether targeted service recovery converts negative engagement into positive engagement.
Results
The model separates four stages, from purchase to later-stage updating, with an explicit non-engagement interval between the negative and positive thresholds; reference scenarios are repeated 1,000 times with recorded random seeds and 95% simulation uncertainty intervals. At period 45, mean potential non-engagement reaches 183.11 users, against 49.37 in potential positive engagement and 50.33 in potential negative engagement.
Price and product quality show strong and opposite effects on the PCE-NCE balance: the mean terminal balance at period 45 moves from 224.69 at price 0.5 to -15.04 at the reference price and to -247.86 at price 1.5.
Product quality runs the other way: the mean terminal balance is -411.86 at quality 1.5 and 259.66 at quality 2.5.
A direct price-by-quality factorial test shows that higher quality attenuates the adverse effect of higher price, with a difference-in-differences of 210.63 on the terminal balance.
Between a continuous-service architecture and one limited to Stage 4, the difference in terminal balance stays small (1.41 users), which does not support a claim that later-stage service investment is superior to earlier service.
Targeted recovery applied to users in negative engagement improves the terminal balance by 149.87 users and raises cumulative conversion to positive engagement by 594.80.
Personal needs met, activated at Stage 4 in the model, shifts the terminal balance from -242.30 to 210.64 when its level rises from 0.5 to 1.5.
Critical reading
The authors present coefficient sign allocation as the model's main structural uncertainty: it can reverse the absolute balance between positive and negative engagement.
The reported quantities are interpreted as scenario outputs and robust directional comparisons, not as forecasts of the actual shares of positive, negative, or non-engagement.
The 0.10 potential-engagement rate is only a scenario anchor built from a descriptive ratio, not a calibrated behavioral conversion probability.
Empirical validation stays modest: a previously coded Xiaomi corpus contains more negative than positive reviews (537 versus 326), directionally consistent with the model but not a point calibration.
The authors themselves note that the coefficient sign-allocation rule has neither an independent theoretical derivation nor an empirical estimate.
Application
As an illustration, the authors suggest treating a price increase as a value-communication program rather than an isolated figure.
For already-manifested negative engagement, they propose a targeted recovery sequence: identify the negative episode, classify it by cause, assign a case owner, then provide a matching remedy.
Research & evidence
The study analysed
When negative customer engagement turns positive: A four-stage simulation model of engagement evolution in online brand communities
Read the original study Full textContinue reading
All analyses →
