Product 02
MRP/Forecast: who thinks what, where?
We design and field quota surveys, run MRP against our synthetic population, and deliver estimates at Bundesland, Wahlkreis or Gemeinde level — with full QC documentation.
Interactive demo
Fielding surveys: sample construction and data quality
We continuously field large online surveys to measure current opinion. We recruit through panel aggregators, steer quotas live, and remove weak sources during fieldwork when quality drops.
What we check live
- Quota steering during the field — representative distribution before weighting.
- Duplicate detection (IP, device, fingerprint) and postcode-level response caps.
- Straightliner and AI-generated text detection on open-ended answers.
- Attention checks, speeding filters and realistic answer-variance bounds.
- Per-project transparency: every rule is documented and auditable.
Interactive demo
Model training
We choose modelling strategy by prediction task and data structure. MRP is usually the core, complemented by more flexible methods where needed.
Link features in one framework
We estimate how personal and geographic attributes jointly relate to survey responses.
Match method to decision task
Depending on the use case, we combine interpretable, hierarchical, or more non-linear approaches.
Estimate interactions robustly
The goal is not just single effects but how effects change when attributes interact.
Designed for training, not just for topline numbers
Our surveys are designed so models can learn robust differences across places and target groups, not only aggregate toplines.
Interactive demo
MRP Outputs
Example of the base forecasts we deliver, illustrated here for three states at Gemeinde level.
Loading…
Interactive demo
Deliverables: predictions and trends
Alongside the headline predictions we surface the main trends in the data at the geography you care about — so you receive not just point estimates but the story of how demographic and political structure drive them.
Illustrative flow: past federal vote routes through age group and finally the share saying migration is the most pressing issue. Ribbon thickness = share of the voting-age population in that combination.
How an engagement works
1. Large samples
We field large, quota-steered online surveys and secure data quality systematically throughout the field.
2. Model training
We model the relationship between personal and geographic characteristics and the responses. Our standard method is MRP.
3. Apply to the synthetic population
We apply the trained models to our synthetic population to predict who currently thinks what, where.
4. Predictions and trends delivered
We deliver the headline predictions together with the main structural trends in the data, at the geographic level you specify — maps, tables, a short report and a dashboard.
Have a specific question you need answered?
Send us the topic and the geographic scope. We will reply with methodology, timeline and price.