Interactive
Counterfactual Scenarios
Apply pre-defined scenarios to a building and compare the Bayesian posterior mean prediction under the baseline vs. the counterfactual. All scenarios isolate one variable at a time, holding all others constant.
A counterfactual asks: what would this building's GHG emissions be if one specific characteristic were different, holding everything else constant? This is distinct from prediction, which asks what a building with a given profile is expected to emit. Counterfactual analysis isolates the causal contribution of a single variable by constructing a hypothetical twin that differs in only that dimension.
The estimates come from the IPW Hierarchical Bayesian Model — the inference model chosen in Section 7. Because the posterior is a full probability distribution over every parameter, the counterfactual inherits genuine uncertainty: both the baseline and the scenario prediction are drawn from the posterior, and the difference is itself a posterior quantity with a credible interval.
How a scenario works
Each pre-defined scenario modifies exactly one input (e.g., reporting year, floor area, or property type) and feeds the modified building through the same posterior predictive distribution. The model uses the full posterior over βA, βT, the property-type intercept αj, and the residual SD σ to generate both predictions. The counterfactual gap is the posterior mean difference; the credible interval propagates uncertainty through all parameters simultaneously.
What the numbers mean
Baseline GHG — posterior mean predicted emissions (metric tonnes CO₂e/yr) for the building as specified. Scenario GHG — the same quantity after the scenario change. Δ% — the percentage change from baseline to scenario; negative means the scenario reduces emissions. 89% CrI — the credible interval: there is an 89% posterior probability that the true value falls within this range.
Green bar / negative Δ%
The scenario reduces predicted emissions relative to the baseline. The magnitude tells you how much the changed characteristic matters in absolute and relative terms.
Amber bar / positive Δ%
The scenario increases predicted emissions. Useful for stress-testing: how much worse would this building perform under a different profile?
Overlapping credible intervals
If the 89% CrIs of the baseline and scenario overlap substantially, the difference is uncertain — the model does not confidently distinguish the two outcomes for this specific building profile.
Interpretation caution
Counterfactuals from an observational model are associational, not causal, unless the identification assumptions hold. The IPW correction addresses selection bias on observable covariates, but unobserved confounding (management quality, tenant behaviour) could still bias the implied effect of changing a variable. Read the Δ% as the model's best estimate of the average association between the changed characteristic and GHG, not as a guaranteed engineering outcome. Scenarios that change property type are especially sensitive to this caution — the type intercepts capture average differences between building categories, not what would happen if a specific building were physically converted.
Select a scenario and click Run.