DINAMICS-2 · Amsterdam UMC & University of Amsterdam

Precarity–Depression Explorer

A simulated neighbourhood of 300 residents, starting from data on more than 21,000 Amsterdam adults (HELIUS). Plan support, tweak how the system works, and the graph redraws instantly.

How to use this tool

What happens over time

Redraws the moment you change anything. Pause any time, or drag across the graph to move through time.
Residents with PHQ-9 ≥ 10 (%) Social precarity (change, SD) Pinned scenario
Residents with PHQ-9 ≥ 10moderate symptoms or worse
Average PHQ-9 score
Social precaritychange, standard deviations
Financial stressof usual level (above the HELIUS minimum)

The neighbourhood

One dot per simulated resident, at the time of the playhead. Hover to meet them.

What drives what

Thicker, faster arrows = stronger push. Click an arrow to change it.

Leverage points from the research

Entry point

Financial stress

Across all methods, recent financial stress was the part of precarity most consistently linked to depression. Lowering it is the most direct way into the system.

Feedback

Depression and social precarity reinforce each other

Depressive symptoms (low mood, guilt, loss of interest) were often linked to social precarity. Once both are high, each can keep the other going.

Duration

Short support fades fast

With a strong loop, improvement comes slowly but lasts longer. With a weak loop, it comes fast and disappears soon after support ends.

Combination

Combine structural and psychological help

Because the domains are linked, support in one spills over into the other. Financial plus mental health support works well whatever the loop strength.

How the model works

Each resident has a level of financial stress (S), drawn from its distribution in HELIUS. Their depressive symptoms (D) and social precarity (P) keep moving toward a target set by S and by each other, at a given recovery speed λ, plus random ups and downs:

dD = λ(α_SD·S + α_PD·P − D − mental health support) dt + σ_D dW dP = λ(α_SP·S + α_DP·D − P − social support) dt + σ_P dW

The HELIUS presets reproduce the variances and covariances in the data, including about 14% of people with PHQ-9 ≥ 10. The model works on standardized scores; to show PHQ-9 points, each resident's score is matched to the same percentile of the HELIUS PHQ-9 distribution. Everything you change in "Tweak the system" goes beyond the data and is meant for thinking through what-ifs. Time is in model units: 1 unit is roughly how long the system takes to adjust.

Keep in mind

  • The residents are simulated, not real people.
  • HELIUS is cross-sectional. Directions of influence are hypotheses that fit the data, not proven causal effects.
  • Mental health support and social support (marked "extra") were not part of the published simulation.
  • Use the numbers for direction and relative size, not as forecasts.

Full guide (English and Dutch)

Sources: Park et al. (2026), SSM – Mental Health · causalnet tutorial (2026), Behavior Research Methods · causalnet on CRAN