Survival curve
A survival curve shows what proportion of a population remains unresolved at each point in time. It is the standard tool for questions of the form how long until something happens, and it exists to handle a problem that averages cannot.
The problem it solves#
Most tracked windows have not finished.
Of Govire's 1,671 redemption windows, 1,258 are still running. Computing an average time-to-outcome from the 326 that resolved would measure only the fastest cases, because the slow ones have not finished yet and are excluded by construction.
That is a systematic bias, and it makes every outcome look faster than it is.
Censoring#
A window observed for some period without the event occurring is censored.
It is not missing data. It carries real information: the event had not happened by that point. A window open for 400 days without a foreclosure sale tells you something specific about the first 400 days.
Survival analysis uses that partial information rather than discarding it. Of 1,671 windows in the feature set, 1,453 are censored and 218 uncensored — which is to say the great majority of what is known is partial, and a method that ignored it would be throwing away nearly everything.
What the Minnesota curves show#
Across 1,336 sheriff's sale windows, tracking two competing outcomes:
| Days from sale | Still unresolved by foreclosure sale | Still unresolved by owner exit |
|---|---|---|
| 90 | 98.9% | 95.6% |
| 180 | 96.8% | 88.4% |
| 270 | 89.1% | 87.2% |
| 365 | 67.4% | 85.2% |
| 540 | 50.9% | 85.2% |
The two shapes are completely different, and the difference is the finding.
Owner exits happen early, then stop#
The owner exit curve drops steeply through the first 180 days and then flattens almost completely. Between day 270 and day 540 it barely moves.
That matches the median of 106 days: owners who act do so inside the redemption period, and once the period closes, they stop. The opportunity ends with the deadline, and the curve records exactly that.
Foreclosure outcomes keep accumulating#
The foreclosure sale curve behaves the opposite way. It barely moves through the first 180 days — because the redemption period is still running and title has not vested — then falls sharply between 270 and 540 days.
By 540 days roughly half the windows have resolved that way.
That lag is the statutory structure showing up in the data: nothing can happen until the period expires, and then a resale process begins.
Competing risks#
The two curves describe competing outcomes. A window that ends in redemption cannot later end in a foreclosure sale.
That is why the percentages do not sum to 100 in any simple way, and why the two are modelled separately rather than as a single time-to-resolution.
What the curves do not do#
They describe the tracked population over the tracked period. They are not a prediction for an individual property, and Govire deliberately does not ship a model that claims to be — a curve fitted to these windows did not reproduce consistently across counties, and shipping it would have implied a precision the data does not support.
The curve describes what happened. That is a different claim from what will happen, and the distinction is worth keeping.