Explainer Society & Social Sciences · Social Research and Impact Measurement · Issue 2026-W46

Who is missing from the count? Reading hidden homelessness statistics

A recorded caseload may rise because observation improves, or fall because a service stops receiving people. Reading a homelessness indicator therefore begins with its boundaries. A new German measurement report offers a starting point for separating changes in coverage from changes in the phenomenon. Our worked example explains why overlapping records cannot simply be added together.

By ARPS Editorial Team Last reviewed 11 November 2026 About 5 min read (estimate)

Source date 2026-11-05

A cutaway building with a person on a temporary sofa and a transparent frame symbolising the limits of observation; a generated illustration.

A recorded caseload may rise because observation improves, or fall because a service stops receiving people. Reading a homelessness indicator therefore begins with its boundaries. A new German measurement report offers a starting point for separating changes in coverage from changes in the phenomenon. Our worked example explains why overlapping records cannot simply be added together.

What is new in the report?

The Organisation for Economic Co-operation and Development published Measuring Unsheltered and Hidden Homelessness in Germany on 5 November 2026. It reviews methods for identifying people without shelter and people temporarily staying with relatives or friends because they lack housing. Its approach combines documentary research, a government questionnaire and consultations; it is not a new comprehensive population count conducted by the report. [1]

The introduction explains how definitions and collection methods affect who appears in an estimate. The chapter on lessons also concludes that no single approach provides a complete, exact count. [1] This does not make numbers dispensable. It makes their scope essential to interpretation. The following analysis concerns measurement decisions; it does not estimate homelessness in Saudi Arabia or compare national prevalence rates.

Examine coverage before explaining an increase

We propose attaching a short boundary statement to each indicator: who qualifies under its definition, where information is collected, what period it covers and who may be missing. If the source is applications to a service, the figure describes people who reached it under its rules. It cannot automatically become a description of everyone needing housing or support in the surrounding area.

When reviewing annual change, ask whether coverage stayed constant alongside asking whether the number changed. Did more organisations begin submitting records? Did eligibility rules or access hours change? Was a question made clearer? These are possible explanations to investigate, not established causes of every increase. If collection changes were not documented, describe interpretation of the trend as limited instead of selecting the explanation most convenient to the institution.

This distinction matters when evaluating a programme. Discovering people previously outside observation does not, by itself, show that the programme made the problem worse. Equally, disappearing from a register does not establish that a person obtained stable housing. The organisation needs to know what happened after exit, or state that it does not know. Missing information should not be converted into a success outcome.

Why two lists do not automatically produce one count

Consider a hypothetical example. A social service records forty people and another organisation records twenty-five. A lawful, appropriately authorised check establishes that ten people appear in both lists. There are fifty-five distinct people observed: forty plus twenty-five minus ten. This is the number captured by those lists, rather than an estimate of everyone absent from administrative records.

If overlap is unknown, reporting sixty-five as a count of unique individuals is unjustified. The organisation can instead present each source separately and explain why duplication cannot be removed. Improving an indicator does not justify sharing personal information without an appropriate legal basis and safeguards. A limited aggregate analysis may be preferable to an unlawful person-level linkage.

Nor does the example estimate people missed by both organisations using an advanced statistical method. Such estimation would require assumptions, study design and specialist review that the illustration does not provide. Its practical purpose is more basic: define the denominator, distinguish a service record from a person and keep the boundary of knowledge visible to the reader. A more precise label can be more useful than a larger number presented with false confidence.

Choose a method around a decision

Before commissioning collection, a social research team should write the decision the information will support. Is the aim to improve access to a particular service, describe the size of a phenomenon or follow transitions into stable housing? These purposes require different information. Staff time spent recording a case does not automatically buy the ability to compare places with different opportunities to reach services.

We suggest starting with a coverage map rather than a prevalence map. List participating organisations, the populations they can reach and known gaps. Then select one gap relevant to the intended decision and examine an appropriate way of understanding it with the staff involved. Do not expand collection simply because a digital tool makes it possible. The map is a planning product; it is not a map of the number of affected people.

The first signal of usefulness is whether the team can explain more accurately whom its indicator includes and excludes. Stop when no lawful, safe way of collecting the information is available, or when there is no clear use for the additional data. A good questionnaire cannot replace fieldwork training, safeguarding arrangements, consent requirements or privacy protections. Those conditions belong in the design before collection begins.

Limits to taking the lesson elsewhere

The report addresses arrangements and definitions in Germany alongside other countries' practices. Living with relatives is not sufficient, on its own, to classify someone as homeless. The reason for the arrangement, its security and the applicable definition matter. Moving a label into another social setting without examining those conditions could stigmatise people or combine different living situations.

Start within your organisation by reviewing one existing definition and source, rather than announcing a new local prevalence rate. Ask the next report to describe changes in observational coverage alongside changes in the count. Which population might your registration rules exclude even though its needs fall within the programme's purpose?

To continue professional learning and save useful reading, explore ARPS communities. The public page describes free ordinary membership and article saving.

References

  1. Measuring Unsheltered and Hidden Homelessness in Germany: Lessons from OECD Countries — OECD · 2026-11-05 · DOI 10.1787/92488f27-en

References open in a new tab.

Discuss this with practitioners

Create a free account to save articles and join the Specialists community discussion.

Create a free account

Discussion 0 published comments

No published comments yet. Be the first to share your experience.