Applied ML + human-centered data work.

When analytics ignores relationships, three problems show up:

1. False signals of performance or risk — highly visible users generate more measurable activity, while bridge/peripheral actors can be critical but invisible.
2. Misleading causal conclusions — changes attributed to individuals may actually reflect shifts in their surrounding network.
3. Optimization that harms the ecosystem — optimizing individual metrics can amplify dominant nodes and reduce diversity, resilience, or collaboration.

Better framing: stop asking “who performed?” and start asking “how did the system change?”

I don’t start with fit indices. I start with the decision logic:

• Are we explaining stable individual differences, or dynamic change within people?
• Does the model capture processes unfolding over time—or just static correlations?
• What decision would change if effects reflect within-person dynamics instead of between-person differences?

Traditional CLPM blends variance from:

• Between-person differences (some people are consistently higher/lower)
• Within-person fluctuations (changes across waves/time)

RI-CLPM separates these layers—revealing whether:

• higher-risk individuals differ from others (selection)
• increases within the same person predict future outcomes (process)

Intervention logic changes:

• Between-person effects → target groups
• Within-person effects → time interventions around change

Takeaway: Prediction is estimation. Interpretation is intervention design.