I had the opportunity to interview Fred Hasselman, the main architect of casnet: An R toolbox for studying Complex Adaptive Systems and NETworks. We spoke of how compatible the complex systems perspective is with some methods widely used in social sciences.
A few notes:
- Multilevel models (and what you put in those) come in many varieties – and some are useful
- See this blog post for an overview of how multi-level survival model can be used to study early warning signals for clinical change.
- Same strategy used in physical activity research in a recent preprint of ours (includes data and code): Characterizing and predicting person-specific, day-to-day, fluctuations in walking behavior
- Interaction is not interaction
- Interaction (1): Two variables are intertwined – or “coupled” – in such a way, that they cannot be separated without severing the phenomena arising from their interplay.
- Interaction (2): A multiplicative, instead of additive, relationship in a linear regression model, where you can partial out variance and get nice beta weights for each variable to determine their individual impacts.
- The two meanings presented above are logically inconsistent: See #36 in Scott Lilienfield’s “Fifty psychological and psychiatric terms to avoid“
- See also this piece: On the distinction between interaction and effect modification
- Interdependence means you can’t use the regular statistics which social scientists know and love.
- … because you lose additivity.
- “Don’t infer causality, observe it.”
- When the system you’re looking at is an individual instead of e.g. the society, you’re in the quite happy position, that lab studies are possible (if you’re smart about them).
- An excellent paper from Merljin Olthof: Complexity in psychological self-ratings: implications for research and practice
- Additional resources: