Written and maintained by CASRAI Editorial Board
Published
On this page
A point-in-time (PIT) count is an unduplicated count of the people experiencing homelessness in a community on a single night. In the United States, HUD requires each Continuum of Care (CoC) to run one at least every two years, during the last 10 days of January, and to cover both sheltered and unsheltered people, either by counting everyone (a census) or by counting a sample of areas and extrapolating.
This page uses the term in its homelessness-enumeration sense. It explains the HUD minimum standards, shows the arithmetic of the sampling-and-extrapolation estimate with an illustrative worked example, and sets out what a PIT total can and cannot support when you use it in research or planning. The wider logic of counting a whole population versus part of it is covered in sampling vs census.
Definition: what a point-in-time count measures
The National Alliance to End Homelessness describes a PIT count as an unduplicated count on a single night of the people in a community who are experiencing homelessness. It is the only measure that enumerates people experiencing unsheltered homelessness as well as those who are sheltered, and the results feed HUD’s Annual Homeless Assessment Report (AHAR) to Congress. The unit of analysis is a person on one designated night, not a person across a year.
HUD’s definitions come from 24 CFR 578.3. A sheltered person lives in a supervised publicly or privately operated shelter designated to provide temporary living arrangements, including congregate shelters, transitional housing, and hotels and motels paid for by charitable organizations or government programs. An unsheltered person has a primary nighttime residence that is a public or private place not designed for or ordinarily used as a regular sleeping accommodation, such as a car, park, abandoned building, bus or train station, airport or camping ground.
The point-in-time idea itself is a general measurement concept, a snapshot at one moment rather than a rate over a period. In epidemiology the analogous quantity is point prevalence; see prevalence vs incidence for how a snapshot differs from a flow.
HUD’s minimum standards for a PIT count
HUD publishes 14 minimum standards, which are summarized from section 1.3 of its PIT Count Methodology Guide. The ones that shape the research design are these.
- Standard 1: each CoC plans and conducts a PIT count of homeless persons in its geographic area at least biennially.
- Standard 2: sheltered and unsheltered counts take place during the last 10 days in January and represent everyone sheltered and unsheltered on a single night in that window.
- Standards 6 and 9: the sheltered count and the unsheltered count are each done by a census (complete coverage) or by one or more sampling and extrapolation methods consistent with HUD standards and guidance. HUD evaluates the basis for any estimation and extrapolation in the annual CoC Program Competition.
- Standard 8: CoCs should use client data already entered in the Homeless Management Information System (HMIS) as the primary source for the sheltered count in emergency shelter, Safe Haven and transitional housing projects that participate in HMIS.
- Standard 10: a CoC may exclude areas where it has determined there are no unsheltered people, such as uninhabitable areas, but must document the criteria and decision process.
- Standards 7 and 11: the CoC must be able to verify that sheltered and unsheltered people were in those situations on the designated night.
- Standard 12: every person is counted once; CoCs must collect enough information to deduplicate the count reliably.
The remaining standards cover CoC approval of the final methodology (3), coordination with Consolidated Plan jurisdictions (4 and 5), privacy and safety in survey administration (13) and training of staff and volunteers (14). The data table below compares the count components side by side.
Sheltered vs unsheltered counts
The two halves of the count use different data sources. The sheltered count draws mainly on administrative records, with HMIS the primary source under Standard 8. The unsheltered count has no administrative frame: people are located by outreach workers and volunteers canvassing the area, which is why sampling and extrapolation questions arise mostly there.
Observation-only and survey-based unsheltered counts
HUD’s sampling guidance describes two main methods for the unsheltered count. Observation-only counts produce a head count with no verbal or physical interaction, so they are conducted late at night, typically between 10pm on the night designated for the count and 6am the next morning. Survey-based counts can be completed over a broader period because respondents can be asked where they slept on the designated night; a survey conducted after that night must ask exactly that. Demographic and subpopulation data require a survey, while observation yields a head count.
Formula: estimating an unsheltered count from sampled areas
When a CoC samples rather than counts every area, HUD’s guidance gives a simplified design-based estimator. The CoC divides its geography into subareas, groups them by the expected number of unsheltered people (high-density, medium-density, rare or unlikely, and uninhabitable), designates a few unusual subareas as certainty subareas, and randomly selects subareas within each group. The estimate is:
N̂ = C + Σg wg × yg, with wg = Ng ÷ ng
- N̂ is the estimated number of unsheltered people in the CoC on the designated night.
- C is the number counted in the certainty subareas, which represent only themselves.
- g indexes a sampling group such as high-density or medium-density.
- Ng is the total number of subareas in group g and ng is the number selected.
- wg is the weight: the inverse of the probability of selection, total subareas in the category divided by the number selected. HUD’s example: sampling 25 of 50 subareas gives a weight of 2.
- yg is the number of unsheltered people counted in the selected subareas of group g.
Weights are calculated and applied separately for each group. HUD’s own example samples 8 of 10 high-density subareas (weight 1.25) and 5 of 10 medium-density subareas (weight 2). If rare or unlikely subareas are not sampled at all, they contribute nothing, which assumes there truly are no people in them on that night or too few to matter. For the general weighting logic see survey weighting and inverse probability weighting; for the grouping step see stratified sampling.
Worked example with real arithmetic
This is an illustrative composite, not a real CoC. The subarea structure follows the mock CoC in HUD’s sampling guidance (5 high-density subareas, one of them a certainty subarea; 24 medium-density subareas; 5 rare subareas; 1 uninhabitable subarea) and the sampling percentages HUD suggests for it: the certainty subarea, 2 of the remaining 4 high-density subareas, 6 of 24 medium-density subareas, and none of the rare subareas. The counts per subarea are invented for the calculation.
| Group | Subareas in group (N) | Selected (n) | Weight (N ÷ n) | Counted in selected subareas | Contribution |
|---|---|---|---|---|---|
| Certainty | 1 | 1 | 1 (counts itself) | 40 | 40 |
| High-density | 4 | 2 | 4 ÷ 2 = 2 | 18 + 12 = 30 | 30 × 2 = 60 |
| Medium-density | 24 | 6 | 24 ÷ 6 = 4 | 3 + 0 + 5 + 2 + 1 + 4 = 15 | 15 × 4 = 60 |
| Rare or unlikely | 5 | 0 | not sampled | none | 0 (assumed none) |
The unsheltered estimate is 40 + 60 + 60 = 160 people, of which 85 (40 + 30 + 15) were counted directly and 75 come from extrapolation. If the sheltered census from HMIS records were 420 people (also invented), the CoC total would be 420 + 160 = 580. Note that the estimate is only as good as the grouping: if the six sampled medium-density subareas happened to contain unusually few or many people, the weight of 4 multiplies that error fourfold, which is why HUD advises sampling as many high-density subareas as possible and not sampling so few medium-density ones that one unexpected count spoils the group estimate.
Interpreting a PIT count
| What the number is | What it supports | What it does not support |
|---|---|---|
| A one-night, unduplicated count of sheltered and unsheltered people | A stock measure for the designated night; comparison of sheltered vs unsheltered shares | The number of different people homeless over a year, which needs a flow measure |
| A census total (complete coverage) | Subarea-level reporting of counts and characteristics | A guarantee that no one was missed; it is a count of those found |
| A sampled and extrapolated total | A CoC-level estimate when full canvassing is not feasible | Reliable subarea breakdowns: HUD notes demographic extrapolation may not be valid for subarea reporting |
| A year-over-year change | A trend, if method, coverage and date are held constant | A real change in homelessness if the methodology changed between years |
The National Alliance to End Homelessness notes that methodology varies by year and community and that counts of youth are often inaccurate. HUD’s sampling paper states that sampling is not as accurate as a full census count and relies on assumptions about where people experience unsheltered homelessness, which are limited by the CoC’s initial or known data.
Assumptions and common mistakes
- Treating the count as annual prevalence. It describes one night. People who experience homelessness for short spells are less likely to be on any given night than people who stay homeless for long periods.
- Non-random selection. HUD’s guidance states that no subarea should be prioritized for selection over another within a group, as that would introduce bias; random selection is the basis for applying the weights.
- Applying one weight to all groups. Weights are calculated and applied separately per group (see the 1.25 and 2 example above).
- Assuming zero in unsampled rare areas without documenting it. Standard 10 requires documentation of the criteria and decision process for excluded areas.
- Pooling extrapolation across jurisdictions. The scale of inference is the CoC’s own geography; extrapolate from that CoC’s counts only.
- Double counting. Standard 12 requires enough information to deduplicate; the sampling guidance notes sampling works best on a single night because people are unlikely to move between subareas in that time.
- Mixing methods when comparing years. A move from census to sampling, or a change in survey timing, changes what the series measures.
How to report a point-in-time count
State the date, geography, method for each half of the count, coverage and the estimator. For a sampled unsheltered count, report counted and extrapolated figures separately. An example sentence in a results section:
On the night of January 27, the Continuum of Care counted 420 sheltered persons from HMIS records (census) and estimated 160 unsheltered persons from an observation-only count of 1 certainty subarea and 8 of 28 randomly selected high- and medium-density subareas (85 counted directly; 75 extrapolated using inverse-probability weights of 2 and 4); 5 rare-density subareas were not sampled.
Cite the HUD documents listed under Sources for the standards, and state the date of the count, since the data are valid for that night only. Adapt the number and date formatting to your journal’s or institution’s style guide; the example uses illustrative figures.
How to calculate it in Excel, R or Python
All three snippets reproduce the worked example: certainty count 40, two groups with N = 4 and 24, n = 2 and 6, and counted totals 30 and 15. They return 160.
Excel. Put group sizes N in B2:B3 (4, 24), selected n in C2:C3 (2, 6) and counted totals in D2:D3 (30, 15), then:
=40 + SUMPRODUCT(D2:D3, B2:B3 / C2:C3)
R.
N = c(high = 4, medium = 24)
n = c(high = 2, medium = 6)
y = c(high = 30, medium = 15)
certainty = 40
estimate = certainty + sum(y * N / n)
estimate # 160
Python.
groups = {"high": (4, 2, 30), "medium": (24, 6, 15)} # N, n, counted
certainty = 40
estimate = certainty + sum(N / n * y for N, n, y in groups.values())
print(estimate) # 160.0
This is a point estimate only. HUD’s sampling paper states that it presents a simplified method of random sampling and that more sophisticated methods can be implemented with a sampling expert; it recommends that CoCs work with local research partners, such as university faculty or students with statistical expertise. For variance of such estimators see sampling variance.
Frequently asked questions
When is the point-in-time count conducted?
HUD Standard 2 requires the sheltered and unsheltered counts to take place during the last 10 days in January, representing everyone sheltered and unsheltered on a single night in that period. Observation-only unsheltered counts typically run from 10pm to 6am.
How often is a point-in-time count required?
HUD Standard 1 requires a PIT count at least biennially. The National Alliance to End Homelessness states that the sheltered count is conducted annually and the unsheltered count is required every other year, though most communities do both annually.
Is a point-in-time count a census or a sample?
Either. Standards 6 and 9 allow a census with complete coverage or one or more sampling and extrapolation methods consistent with HUD standards and guidance, separately for the sheltered and unsheltered counts.
What is a certainty subarea?
A subarea expected to hold a much higher share of unsheltered people than others, so it must be counted in full. It represents only itself in the results and is not weighted.
Does the PIT count include everyone who is homeless in a year?
No. It is an unduplicated count of people on one designated night, so it does not measure how many people are homeless over a year.
Sources
- HUD Exchange, HUD’s PIT Count Minimum Standards (summary of section 1.3 of the PIT Count Methodology Guide), https://files.hudexchange.info/resources/documents/PIT-Count-Standards.pdf, checked 2026-11-14.
- HUD Exchange, How to Use Sampling within a CoC to Conduct an Unsheltered Point-in-Time (PIT) Count, https://files.hudexchange.info/resources/documents/How-to-Use-Sampling-within-a-CoC-to-Conduct-an-Unsheltered-PIT-Count.pdf, checked 2026-11-14.
- National Alliance to End Homelessness, What Is a Point-in-Time Count?, https://endhomelessness.org/resource/what-is-a-point-in-time-count/, checked 2026-11-14.
Data: HUD minimum standards for PIT count components: methods, timing and data sources
Compares the components of a HUD point-in-time count (sheltered, unsheltered observation-only, unsheltered survey-based, excluded areas) by permitted method, timing and data source, compiled from HUD's PIT Count Minimum Standards and HUD's unsheltered-sampling guidance.
Period covered: 2021/2026. Coverage: United States.
| Component | Permitted method | Timing | Primary data source or verification |
|---|---|---|---|
| Sheltered count | Census (complete coverage) or sampling and extrapolation consistent with HUD guidance (Standard 6) | Single night in the last 10 days of January (Standard 2) | HMIS client data for HMIS-participating emergency shelter, Safe Haven and transitional housing (Standard 8); verify sheltered on the night (Standard 7) |
| Unsheltered count, observation-only | Census or sampling and extrapolation (Standard 9); head count without interaction | Typically 10pm on the designated night to 6am next morning | Direct observation; verify unsheltered on the night (Standard 11) |
| Unsheltered count, survey-based | Census or sampling and extrapolation (Standard 9); survey with questions on where the person slept | Concentrated on the designated night; may extend over a few days, with respondents asked where they slept on the designated night | Survey responses; privacy and safety protections required (Standard 13) |
| Excluded areas | Areas with no unsheltered people, such as uninhabitable areas (Standard 10) | Not applicable | Criteria and decision process must be documented |
Sources
- HUD's PIT Count Minimum Standards (accessed 2026-11-14)
- How to Use Sampling within a CoC to Conduct an Unsheltered PIT Count (accessed 2026-11-14)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







