Today's total may still be unknown
If recent hospitalizations have not all been reported, a nowcast can estimate the eventual count using observed reporting delays. It is estimating events that may already have occurred, not necessarily predicting new hospitalizations next week.
A forecast extends beyond the current period and needs assumptions about future dynamics.
| Product | Main question |
|---|---|
| Reported count | What has arrived so far? |
| Nowcast | What is likely already happening? |
| Forecast | What may happen next? |
Model assumptions can become outdated
A nowcast based on stable past delays may struggle when reporting suddenly changes. A laboratory backlog or system outage can produce a delay pattern the model has not learned.
Uncertainty intervals should accompany the estimate, and revision is expected as more information arrives.
Do not judge only one published point
Evaluate performance across archived releases and later, more complete data. A model that looks accurate after its final revision may have been less useful in real time.
Label the product clearly. Readers should not have to guess whether a line is a direct observation, a delay-adjusted estimate, or a future projection. The distinction changes what confidence and action the result can reasonably support.
