Read beyond the noise.

Left truncation can hide the early part of a story in a dataset

A research desk with an original timeline chart where observation begins after several events.
AI-generated photographic illustration; not documentary photography.

A headline says that most businesses in a survey survived their first five years. Before admiring the survival rate, ask when the researchers began looking. If they recruited only businesses still trading in year three, early failures could never enter that sample.

This invented example illustrates left truncation: entry into observation depends on surviving long enough to be observed. In time-to-event analysis, delayed entry can be handled explicitly. Stata’s survival-analysis reference manual distinguishes entry time from the underlying start of the period at risk.

One clock starts at the event that defines the question, such as a business opening. Another starts when that business becomes observable in the study. They may begin on different dates.

Imagine ten shops opening together. Four close before a researcher starts recruitment, and six remain. If all six complete the next two years, the researcher has useful information about those surviving shops. That does not establish that all ten survived five years, or that early closure was rare.

A suitable statistical analysis can account for delayed entry under its assumptions. A reader should therefore check the method rather than assume that any study with later recruitment is invalid. The important issue is whether the population, time origin and entry process match the claim being made.

Do not confuse missing beginnings with unknown endings

Right censoring occurs when follow-up ends before the event is observed for a participant. Left truncation concerns who could enter observation in the first place. They can occur in the same study, but they are different problems.

When reading a report, look for three details: the date from which duration is measured, the conditions for joining the dataset, and the treatment of people already lost before recruitment. If those details are absent, describe the evidence narrowly: “among organisations included in this survey.” That small phrase can prevent a statement about the observed survivors from becoming an unsupported statement about everyone who started.