Humanitarian Needs Assessment

What is a humanitarian needs assessment?

A humanitarian needs assessment is a structured process for understanding how a crisis affects people, which needs, risks, barriers, capacities and priorities matter most, and what evidence decision-makers require to choose, target or adapt a response.

An assessment is not automatically a survey. It may consist entirely of existing information, discussions with affected people and local organisations, direct observation, service mapping, facility data or a structured joint analysis. When new primary data are required, the method should follow the question, not the other way round.

Good assessment looks beyond deficits. It considers the crisis and its drivers; threats and protection risks; the severity and scale of needs; barriers to rights and services; people’s priorities and preferences; existing capacities and coping strategies; services and response gaps; likely changes; and the consequences of action or inaction. It also asks who is missing from the evidence.

The UNHCR Needs Assessment Handbook provides the leading cross-sector field process. OCHA places coordinated assessment and joint analysis at the foundation of a coherent response, while the Core Humanitarian Standard 2024 requires meaningful participation, safe and ethical data management, disaggregated evidence that minimises demands on people, and the use of feedback and learning in decisions.

Assessment is different from nearby activities

Do you actually need a new assessment?

Before designing a form, write down the decision, the person or group authorised to make it, the deadline and what would change under different findings. If those answers are vague, pause.

The five-question commissioning test

1. What decision will this inform? State it as an action: select priority areas, choose a modality, revise targeting criteria, design services or allocate resources.
2. What is already known? Check assessment registries, national systems, coordination products, community feedback, previous studies, programme monitoring and relevant datasets.
3. What is the smallest important gap? Separate “useful to know” from “must know now”.
4. What could go wrong? Consider protection, political, security, safeguarding, data, reputational and opportunity-cost risks.
5. Who will do what after the findings? Name the owner, deadline and decision route before fieldwork begins.

UNHCR’s current programme guidance explicitly recommends starting with a secondary-data review and avoiding primary collection when existing information is sufficient, when collection could aggravate harm, when known bias would make findings unusable, or when costs outweigh benefits. See UNHCR Programme Handbook for Partners: Assessments.

Choose the assessment type from the decision, not the label

“Rapid”, “multi-sector” and “participatory” describe different features. An assessment can be rapid and sector-specific, joint and qualitative, or comprehensive and mixed-method. Define it across several dimensions so stakeholders understand what it can and cannot answer.

An indicative depth ladder

These levels are not a universal timetable. A good rapid assessment deliberately answers fewer questions with transparent limitations. A poor “comprehensive” assessment may simply collect more weak data. In the earliest days of a sudden-onset crisis, OCHA’s MIRA and coordinated-assessment guidance supports a common initial picture and strategic priorities; it is not a substitute for later sector depth. In refugee emergencies, UNHCR’s NARE provides a coordinated multi-sector starting point.

What good assessment looks like

Quality is not only statistical. A technically sophisticated assessment can still be irrelevant, unsafe, extractive or too late. Use these principles as design criteria and as a final review.

The Sphere Handbook links assessment with rights, protection, participation and technical minimum standards. The CHS 2024 strengthens the requirement to share accessible information, support meaningful participation, manage data safely, minimise demands on people and use evidence for adaptation.

The complete humanitarian assessment workflow

Some steps overlap in a rapid response, but none disappears. A one-day exercise still needs a decision purpose, safe collection, quality checks, analysis and an action route. Agree stage gates so pressure to “start collecting” does not bypass design.

1. Stage 01Name the decisionIdentify users, authority, deadline, options and what evidence would change the choice.
2. Stage 02Understand contextReview crisis drivers, law, power, access, risk, existing systems, actors and likely change.
3. Stage 03CoordinateCheck registries, contact authorities and coordination bodies, and agree joint or harmonised roles.
4. Stage 04Define scopeSet objectives, population, geography, units, analytical framework, deliverables and limitations.
5. Stage 05Review existing evidenceFind, appraise and synthesise secondary sources before proposing new collection.
6. Stage 06Prioritise gapsTurn decision needs into answerable questions, indicators, disaggregation and an analysis plan.
7. Stage 07Design methodsSelect methods, sampling, tools and a realistic field plan that support intended claims.
8. Stage 08Protect people and dataComplete risk, safeguarding, consent, referral, security and data-responsibility planning.
9. Stage 09Prepare and testBuild, translate, cognitively test and pilot instruments; recruit and train the team.
10. Stage 10Collect and superviseMonitor coverage, safety, consent, quality, burden, access and contextual change every day.
11. Stage 11Clean and analyseProtect raw data; document cleaning; apply correct denominators, weights and qualitative coding.
12. Stage 12Triangulate and interpretCompare sources, test alternatives, separate evidence from inference and assess confidence.
13. Stage 13Validate responsiblySense-check conclusions with relevant technical, local and community perspectives without exposing data.
14. Stage 14Communicate for useDeliver concise decision products, methods, limitations and accessible feedback in time.
15. Stage 15Act and recordAssign recommendations, document decisions and explain what will or will not change.
16. Stage 16Review and closeEvaluate usefulness, archive or delete data as planned, record lessons and update the registry.

Suggested stage gates

• Commissioning gate: decision, owner and value are clear.
• Design gate: secondary review, scope, methods, sampling and analysis plan agree.
• Protection gate: ethics, security, safeguarding, referrals and data controls are approved.
• Launch gate: tools are translated, tested, piloted and frozen; teams are trained and assessed.
• Analysis gate: cleaning is documented, denominators checked, limitations agreed and sensitive outputs reviewed.
• Publication gate: claims match evidence; sharing is authorised; actions and community feedback are ready.

Coordinate first and support local leadership

Uncoordinated assessments waste scarce access, produce incompatible results and repeatedly burden the same communities. Coordination is both an efficiency measure and an ethical obligation.

Before commissioning primary data

• Contact the relevant national or local authority, cluster/sector, inter-cluster mechanism, assessment working group or refugee coordination structure.
• Search the country’s assessment registry or survey-of-surveys and ask about planned exercises.
• Review national information systems, censuses, administrative data, early-warning platforms, community feedback and local research.
• Agree common administrative boundaries, place names, population denominators, sector definitions and time periods.
• Decide whether to join, finance, contribute questions, share enumerators, harmonise indicators, coordinate sites or rely on another process.
• Register the assessment plan and later add metadata, methods, products and access conditions.

OCHA describes an assessment registry as a coordination tool for seeing which sectors, groups and areas have been covered, identifying gaps and preventing duplication. UNHCR recommends an inter-agency registry where it coordinates a response.

Joint, harmonised or coordinated?

Local and national actors are not merely data collectors

Local authorities, community organisations, organisations of persons with disabilities, women-led groups, youth groups, professional associations, universities, responders and civil-society networks may hold knowledge, trust, language ability, historical evidence and operational reach that external teams lack. Involve them in governance, question selection, interpretation, authorship, ownership, budgets and decisions, not only enumerator recruitment.

Agree roles and power explicitly: who approves the scope, who can stop collection, who owns raw and derived data, who may publish, whose interpretation is visible, who receives credit, and how disputes are resolved. Resource participation, translation and accessibility rather than expecting unpaid local labour.

Build participation into every stage

People affected by crisis are analysts of their own circumstances, rights-holders and decision-makers, not a source of “beneficiary data”. Participation improves relevance and can reveal barriers, capacities and unintended consequences that external teams miss.

The UNHCR Toolkit for Participatory Assessment provides practical guidance for structured dialogue across age, gender and diversity. CHS 2024 requires information in accessible and contextually appropriate languages and formats, meaningful participation in decisions, safe feedback and complaints, and adaptation based on what people communicate.

Make representation a design question

A “community consultation” is never automatically representative. Map power, exclusion and communication channels. Consider who may be absent because of mobility restrictions, caregiving, work, stigma, language, disability, digital exclusion, fear of authorities, legal status, remoteness, institutionalisation or seasonal movement. Consult people separately when mixed groups would silence them, but do not assume identity groups are internally uniform.

Manage expectations honestly

• Explain who is conducting the assessment and why.
• State clearly that participation does not guarantee assistance or affect eligibility unless that is genuinely the process.
• Explain what participants may decline, how information will be used and who may receive it.
• Provide realistic timing for findings, decisions and follow-up.
• Give a safe contact or feedback route that remains available after the team leaves.

Write the scope before writing questions

A short terms of reference or concept note prevents an assessment from expanding into every topic stakeholders can imagine. It also creates a record against which changes and compromises can be judged.

Minimum terms of reference

Build an analytical framework

An analytical framework organises what the team needs to understand and how the elements relate. It keeps the assessment from becoming a list of disconnected sector questions. Adapt an existing recognised framework where appropriate rather than inventing incompatible categories.

A practical cross-sector frame can move through:

1. Context and shock: what happened, where, when, why and what may happen next?
2. People and exposure: who is affected, how, and which population estimates are sufficiently reliable?
3. Humanitarian conditions: what are the consequences for health, safety, living standards, rights and wellbeing?
4. Threats, vulnerabilities and capacities: which factors increase or reduce risk?
5. Services and systems: what is available, accessible, acceptable, safe, appropriate and of sufficient quality?
6. Priorities and preferences: what do different groups identify as urgent and which response options do they consider workable?
7. Response and gaps: who is doing what, where, for whom, with what reach and limitations?
8. Outlook: how might seasonality, conflict, displacement, markets, disease, policy or funding change the picture?

For crisis-wide planning, the Joint Intersectoral Analysis Framework (JIAF) structures collaborative analysis of the nature, drivers, severity and overlap of needs. Protection teams can use the Protection Analytical Framework. These are analytical systems, not generic questionnaires.

Conduct a rigorous secondary-data review

A secondary-data review is not a quick internet search. It is a documented process for finding, appraising, organising and synthesising information collected for another purpose.

Where to look

OCHA’s Humanitarian Data Exchange (HDX) helps users find shared crisis data, including priority Data Grids and Common Operational Datasets. HDX is a discovery platform, not a guarantee that a dataset is current, complete or safe for your use. Check provenance and metadata.

Use a source log

Appraise evidence in context

Do not rank evidence only by institutional prestige. A current, transparent local source may be more relevant than an older national report. Consider recency, geographic and population coverage, method, sample, definitions, independence, incentives, access constraints and consistency with other evidence. Distinguish an original source from a document that repeats it.

Extract claims into an evidence matrix by analytical topic, location, group and time. Mark convergence, contradiction, missingness and possible explanations. ACAPS recommends beginning with the decision and being explicit about sources, methods, assumptions and bias; its analytical approach emphasises sensemaking rather than simply accumulating data.

Turn information needs into answerable questions

Work backwards from the product. Draft the analysis table or decision brief before the questionnaire. If the team cannot explain how a response will be analysed and used, the question is not ready.

Prioritise each information need

• Decision-critical: the decision cannot be made responsibly without it.
• Interpretive: needed to explain or safely disaggregate a core finding.
• Operationally useful: can improve implementation but is not essential to the immediate decision.
• Interesting: remove from this exercise or address through later research.

Question review test

• Does the question connect to a named analysis and decision?
• Is the selected respondent or source able to know the answer?
• Are the unit, population, place and recall period explicit?
• Does it ask one neutral, clear concept without suggesting a desirable answer?
• Are response options mutually exclusive, sufficiently exhaustive and locally meaningful?
• Are “do not know”, “not applicable” and “prefer not to answer” used appropriately?
• Can the wording be translated without changing the concept?
• Does the requested detail create unnecessary risk or identification?
• Will disaggregation have enough observations and a safe reporting plan?
• Can the same answer be obtained from existing or less burdensome evidence?

Indicators need reference information

For each indicator, document the definition, numerator, denominator, unit, universe, disaggregation, recall period, source, calculation, missing-value treatment, frequency, limitations and responsible person. Words such as “access”, “availability”, “coverage”, “functionality”, “adequacy”, “safety” and “priority” need operational definitions.

11 · Design

Choose methods by the claim you need to make

No method is inherently rigorous. Rigour comes from matching the method, source, unit and selection approach to the question and from stating what the resulting evidence can and cannot support.

Start with the intended claim. A household survey may estimate the prevalence of an experience among a defined household population when a valid probability sample is feasible. It cannot, by itself, explain why the pattern exists, represent people outside households or verify service functionality. A key informant can describe systems, events and informed perceptions; that person should not be treated as a statistical proxy for an entire population. Direct observation can verify visible conditions at a time and place, but not private experiences, motives or events outside the observation window.

Specify four things for every data point

Use mixed methods deliberately

Mixed methods are useful when each component has a defined job. Quantitative evidence may describe how much or how often within its valid population; qualitative evidence may explain mechanisms, variation, meaning and unintended effects; observation or administrative data may verify conditions. Plan how findings will be integrated before collection rather than placing unrelated datasets beside one another at the end.

Method guidance: UNHCR Assessment & Monitoring Resource Centre and ACAPS analytical approach.

12 · Design

Sampling and participant selection

Selection determines the population to which evidence can speak. It is not a technical footnote: it is the boundary between a defensible conclusion and false precision.

Before choosing a design

• Define the target population precisely, including geography, time and eligibility.
• Identify populations missing from the frame: unregistered, mobile, institutionalised, homeless, newly arrived, remote or otherwise less visible people.
• Define the sampling unit at each stage and the final analysis unit.
• Assess frame completeness, duplication, age, access restrictions and security implications.
• Decide the minimum disaggregations and comparisons the decision genuinely requires.
• Estimate likely non-response, clustering and design effects using context evidence, not convenient defaults alone.
• Choose how selection probabilities, replacements, call-backs, refusals and inaccessible units will be recorded.
• Plan weights and variance estimation before fieldwork if probabilities differ.

Sample size is a decision, not a magic number

For probability surveys, size depends on the required precision, confidence level, expected prevalence or variability, design effect, finite population correction where material, anticipated non-response and the estimates required for each domain. Increasing the sample cannot repair a biased frame, unsafe access, poor measurement or non-probability selection. A statistically large convenience sample remains a convenience sample.

Qualitative participant selection

Choose participants for relevance and diversity rather than numerical representativeness. Build a selection matrix covering roles, places, genders, ages, disability, displacement status, livelihood, language and other context-specific dimensions. Use maximum variation to explore differences; critical-case selection to examine high-consequence conditions; and additional recruitment to test emerging explanations or fill clear evidence gaps.

“Saturation” should not be used as a ceremonial claim. Document what became repetitive, for which group and topic, what remained uncertain, and whether access or time, rather than evidential sufficiency, ended collection.

Selection in access-constrained settings

• Map who is excluded by checkpoints, communication coverage, opening hours, digital access, gatekeepers, documentation requirements or fear.
• Use multiple safe recruitment routes and do not publish routes that could expose participants.
• Keep separate what was observed directly, reported by remote sources, modelled or inferred.
• Do not silently substitute accessible populations for inaccessible ones. Describe the coverage boundary prominently.
• Use scenarios or ranges when key denominators are uncertain, and update estimates as access changes.

Practical selection guidance: UNHCR, Selecting participants for qualitative data collection.

13 · Protect

Ethics, protection and safeguarding

An assessment is justified only when its likely value is proportionate to the burden and risk it creates. The duty to avoid harm applies to the questions, selection, setting, staff, technology, analysis, publication and the expectations the process creates.

Consent is a process

Explain who is responsible, why the activity is being conducted, what participation involves, expected time, foreseeable risks and benefits, what will happen to the data, who may receive it, whether quotations or recordings are proposed, the limits of confidentiality, the right to skip or stop, whether withdrawal after submission is possible, complaint routes and the fact that services or assistance do not depend on participation. Check understanding in the participant’s preferred language and communication mode.

A signature can create danger or be inappropriate. The form of consent should match literacy, culture, risk and organisational requirements. Record consent without collecting an unnecessary identifier. Renew consent when the use changes, when a sensitive module begins or when recording is requested.

Power, expectations and compensation

People may perceive aid staff, local leaders or data collectors as controlling access to assistance. Separate assessment from registration and targeting, state this distinction repeatedly, and avoid conducting interviews in a distribution queue where refusal is not realistically free. Compensation or reimbursement should recognise time and access costs without becoming coercive; set a transparent, consistent approach with community and ethics input.

Referrals and urgent risk

• Map verified, functioning and accessible services before asking questions likely to reveal urgent need.
• Agree thresholds, responsibilities and safe pathways for imminent harm, safeguarding, medical crisis and other foreseeable disclosures.
• Explain relevant limits to confidentiality before consent.
• Do not promise a service or report a disclosure without a mandate and protocol.
• Train staff to respond calmly, avoid investigation, record only what is necessary and seek designated support.
• Update referral information during fieldwork; an obsolete list is not a referral system.

Children and adolescents

Use children’s direct participation only when it is necessary, beneficial, age-appropriate and supported by a child-safeguarding and ethical protocol. Determine consent and assent requirements under applicable law and context; assess whether guardian involvement itself could create risk; use trained staff, suitable spaces and accessible methods; avoid repeated or investigative questioning; and pre-position response pathways. Adult proxies cannot fully describe children’s subjective experiences, while children should not be asked to carry responsibility for household-level facts they cannot know.

Staff safety and wellbeing

Assessment personnel are also affected by security threats, distressing material, harassment, discrimination and workload. Conduct role-based security analysis, use check-in and escalation arrangements, limit exposure to traumatic content, provide confidential support and rest, and create a safe route to report misconduct. Never use performance targets that pressure staff to override refusal, privacy or stop rules.

Apply the Core Humanitarian Standard 2024, IASC GBV Guidelines assessment guidance and relevant institutional ethics review. For evidence activities involving people, see UNICEF’s 2026 ethics policy.

14 · Protect

Data responsibility across the lifecycle

Responsible data practice means the safe, ethical and effective management of personal and non-personal data for operational response. It begins before collection and continues through access, analysis, sharing, retention and deletion.

Build a data responsibility plan

Identifiability is cumulative

Removing a name does not make a record anonymous. Exact location, rare disability, occupation, household composition, narrative text, device metadata, time and external datasets may combine to identify a person. Pseudonymised data remain personal data when a key or realistic linkage exists. Assess singling out, linkability and inference risk in the context in which a product will be used.

• Direct identifiers: name, telephone, email, official or assistance ID, photograph, voice and biometrics.
• Indirect identifiers: precise location, age, household structure, employer, rare characteristic, detailed event and free-text narrative.
• Sensitive inferences: protection status, political affiliation, health, ethnicity, religion, sexual orientation, gender identity, legal status or vulnerability score.
• Group risk: aggregate data can expose a community, route, site, minority or resource even when no person is named.

Use a tiered sharing model

Field-device minimums

• Supported devices, strong authentication, encryption and rapid lock; no shared accounts.
• Only the current form and minimum local data; automatic upload and verified removal when connectivity permits.
• Safe update, charging, storage, loss-reporting and remote-management arrangements.
• No respondent identifiers in filenames, notifications, screenshots or casual messaging channels.
• Separate contact or re-contact information from response data using different access permissions.
• Paper forms logged, transported, stored, digitised and destroyed under equivalent controls.

Primary guidance: IASC Operational Guidance on Data Responsibility in Humanitarian Action and the ICRC Handbook on Data Protection in Humanitarian Action, third edition.

15 · Include

Inclusion, accessibility and disaggregation

A technically correct average can conceal systematic exclusion. Inclusive assessment examines who is missing, why participation or services are inaccessible, and how needs, risks, priorities and capacities differ across intersecting identities and circumstances.

Map exclusion before sampling

Use local organisations and representative groups to identify who may be absent from lists, meetings, phone surveys, public spaces or formal leadership. Consider people with disabilities, older people, children and adolescents, women and gender-diverse people, minority language or ethnic groups, people without documentation, people living alone or in institutions, displaced and stateless people, minority religious or caste groups, people with diverse sexual orientation and gender identity, mobile populations, remote settlements and people exposed to stigma or criminalisation. Categories must be locally analysed and never turned into a public map of vulnerable individuals.

Disaggregate with a purpose

At minimum, consider sex, age and disability, then add context-relevant dimensions such as displacement status, location, household composition or livelihood when they inform a decision and can be collected safely. Predefine age bands and explain them. Use the Washington Group question sets or an appropriate validated module to identify functional difficulty when this meets the purpose; do not replace them with “Are you disabled?” or assume a medical diagnosis.

Protect small and stigmatised groups

Detailed cross-tabulation can create small cells and reveal identities, especially when location is precise. Establish suppression, aggregation, rounding and publication rules before analysis. Do not publish a rare characteristic merely because the collection form allowed it. In some contexts, asking about ethnicity, legal status, sexual orientation, gender identity, affiliation or protection incidents may be disproportionate or unsafe.

Intersectionality

People experience multiple systems at once. Do not assume that every older person, woman, person with disability or displaced person has the same experience. Where sample size supports it safely, examine intersections and explain uncertainty. Where it does not, use qualitative evidence and avoid implying that an overall subgroup average captures everyone.

• Representative organisations and local specialists helped define barriers and safe adaptations.
• Budget covers accessibility, interpretation, transport, communication and reasonable accommodation.
• Recruitment reaches beyond formal leaders and service users already known to agencies.
• Team composition and interview matching respond to context and participant preference.
• Questions and consent are tested with people using different languages and communication modes.
• Accessible feedback products and two-way channels are planned from the start.
• Exclusions, refusals, inaccessible areas and missing groups are named in the limitations.

Use the IASC Guidelines on Inclusion of Persons with Disabilities, Washington Group implementation guidance, the IASC Gender Handbook and the Humanitarian Inclusion Standards.

16 · Prepare

Build, translate and test the instrument

A good instrument is a controlled implementation of the analysis plan. Every question has a source, a purpose, a defined unit and a rule for how its answer will be interpreted.

Write from the analysis table, not from a blank form

Start with the decision and analytical question, define the indicator or qualitative theme, choose the source and method, then draft the minimum question or observation item. Reuse an established indicator only after checking that its concept, wording, denominator, recall period and interpretation fit the context. Record adaptations; a familiar label does not make two differently measured indicators comparable.

Question-writing rules

Ordering and flow

• Begin with consent and non-sensitive eligibility; explain why any screening question is necessary.
• Group questions in a natural sequence and use transitions when the topic changes.
• Place sensitive modules only after rapport and renewed information or consent where needed.
• Put classification items only where required; demographics are not automatically harmless.
• Use calculated fields and skip logic to reduce burden, but always provide an auditable path.
• End with participant priorities, anything important that was missed, referral information if relevant, and a clear explanation of next steps.

Translation is part of measurement

Use translators who understand the concept and context. Prepare a terminology sheet, translate for meaning, independently review or back-check difficult items, reconcile differences, and test with intended users. Document approved wording and do not allow field staff to improvise competing translations without recording the issue. Include sign languages and accessible communication where required.

Test in three layers

A pilot is not simply the first interviews. Use a separate version, define acceptance criteria, debrief staff and participants, inspect data, repair the tool and repeat testing when a material change is made. Exclude pilot records from the final dataset unless inclusion was planned, ethically covered and analytically defensible.

Version control

• Unique instrument ID, version, language, date and accountable owner appear in the form and documentation.
• A change log records question, response, logic, translation and calculation changes.
• One approved production version is frozen before launch and deployed to all devices.
• Emergency changes require authorisation, re-test, communication and a documented analysis consequence.
• Exports preserve original variable names, labels, codes and form version.

Testing reference: UNHCR, Key considerations for testing and piloting surveys.

17 · Prepare

Team, training and field operations

Data quality and participant safety are produced by people and management systems. A polished digital form cannot compensate for unsafe recruitment, rushed training or unsupported field staff.

Define roles and separation of duties

Recruit for trust and competence

Recruit language and cultural competence, facilitation skill, digital or paper literacy, reliability, safeguarding awareness and the capacity to work respectfully across difference. Consider gender, age, disability and community acceptance without treating identity matching as a universal solution. Screen conflicts of interest, political or gatekeeper relationships and risks created when collectors are also programme staff, authorities or community leaders.

Training is competency-based

• Purpose, decision use, scope, definitions and the limits of what the assessment promises.
• Humanitarian principles, code of conduct, safeguarding, prevention of sexual exploitation and abuse, and complaint routes.
• Consent, privacy, refusal, distress, urgent risk, referrals and stop rules.
• Inclusive communication, reasonable accommodation, interpreter practice and power dynamics.
• Question intent, neutral probing, response coding, observation standards and avoiding assumptions.
• Sampling or participant-selection procedure, call-backs and strict replacement rules.
• Device, paper, security, synchronisation, version and data-incident procedures.
• Mock interviews, role plays, field simulation and observed certification against explicit criteria.

People who do not demonstrate minimum competence should receive coaching and reassessment or be reassigned. Attendance is not certification. Hold daily debriefs and short refresher exercises when systematic errors appear.

Prepare a field movement and communication plan

18 · Collect

Field quality assurance

Quality assurance should detect risks while they can still be corrected. It combines respectful supervision, process observation, paradata, daily data review and a documented response to each issue.

Monitor a small, interpretable dashboard daily

Supervisor practice

• Observe a rotating selection of introductions, consent processes and interviews with participant agreement.
• Review submitted records and field logs every day; give specific private feedback.
• Verify that recruitment, household selection and call-backs follow the design.
• Maintain a decision log for ambiguities, translation issues, access changes, replacements, incidents and instrument changes.
• Separate supportive quality improvement from investigations of suspected misconduct.
• Escalate patterns early to methods, protection or data leads; do not silently “clean away” field problems.

Back-checks and re-contact

Use only when included in the consent and risk plan. Keep verification brief and non-sensitive, avoid exposing participation to another household member or authority, and separate contact details from responses. A back-check can confirm that contact occurred or a small number of stable items; it should not ask someone to repeat sensitive disclosures.

When quality fails

Pause affected collection, preserve an audit trail, determine the extent and cause, protect participants and staff, and decide whether records can be corrected, excluded, re-collected or reported with a limitation. Do not fabricate replacement data or silently change values. Material departures from the protocol belong in the methods and limitations.

19 · Analyse

Analysis, triangulation and uncertainty

Analysis turns observations into bounded findings and findings into justified implications. The chain must remain visible so decision-makers can distinguish evidence, interpretation, uncertainty and recommendation.

Create a reproducible analysis system

• Preserve a read-only raw dataset and original qualitative records under appropriate access controls.
• Create a documented cleaning script or log; never overwrite original values silently.
• Maintain a data dictionary, value labels, derived-variable definitions, form version and analysis population.
• Record exclusions, duplicates, corrections, imputation, coding decisions and weight construction.
• Use version-controlled analysis code or a clearly auditable procedure and independently review high-consequence outputs.
• Store publication tables separately from identifiable or sensitive source data.

Quantitative analysis essentials

Qualitative analysis essentials

Organise notes or transcripts, familiarise analysts with the material, code against the framework while allowing new themes, compare across groups and places, and record deviant or negative cases. Distinguish frequency from salience: a rarely mentioned risk may still be severe, while a commonly repeated view may reflect recruitment or group dynamics. Use short, safely anonymised quotations only when they add meaning and publication consent covers the use.

Triangulation is structured comparison

Keep the reasoning chain explicit

Write a confidence statement

For every high-consequence conclusion, state the supporting sources, their independence and coverage, important contradictions, missing groups or areas, the role of assumptions and an overall confidence judgement in plain language. Avoid reducing uncertainty to a decorative colour. Decision-makers need to know what additional evidence could change the conclusion.

20 · Judge

Severity, priority and response analysis

Severity describes the seriousness of conditions and consequences. Priority is a decision that also considers rights, urgency, people’s preferences, capacity, feasibility, risk and the likely effects of action or inaction. They are related, but they are not interchangeable.

Keep distinct concepts distinct

Use established intersectoral methods where applicable

In a Humanitarian Programme Cycle context, follow the current Joint and Intersectoral Analysis Framework and country coordination process rather than inventing a parallel severity score. Sector classifications and thresholds should be used only with their required definitions, evidence and technical expertise. Do not average incompatible sector scores merely to produce a ranking.

A transparent prioritisation conversation

Response analysis follows needs analysis

A documented need does not automatically justify a particular modality or programme. Compare options against appropriateness, protection, technical quality, timeliness, coverage, market and service capacity, community preference, cost, operational feasibility, environmental impact, conflict sensitivity and likely unintended effects. Identify what local actors and public systems already do and what support would strengthen rather than displace them.

Intersectoral reference: OCHA, Joint and Intersectoral Analysis Framework.

21 · Act

Turn findings into responsible action

An assessment is complete only when findings reach the right people in time, decisions are recorded, affected communities receive an accessible response, and unresolved uncertainty is monitored.

Design products around decisions and audiences

Release early findings carefully

Rapid findings should be labelled preliminary, dated and bounded by source, coverage and validation status. Separate confirmed observations from unverified reports and analysis. Create a correction process and ensure the final product clearly supersedes earlier versions. Speed is valuable only when users can understand the confidence and limitations.

Validate without transferring ownership of the analysis

Discuss findings with diverse community members, local organisations, technical specialists and operational actors. Ask what is wrong, missing, surprising or unsafe to publish; test explanations and recommendations; and document disagreements. Validation is not a request for participants to endorse an agency’s conclusion, and a powerful stakeholder should not be allowed to veto uncomfortable evidence.

Close the feedback loop

• Tell participants during consent when and how findings and decisions will be shared.
• Use existing trusted, accessible channels and more than one format; do not rely only on a website.
• Explain what was heard, the limits of the evidence, actions agreed, actions declined or delayed, and the reasons.
• Protect small groups and sensitive locations in community-level outputs.
• Provide safe channels to correct, question or complain about findings and decisions.
• Track whether the feedback reached excluded groups and whether it was understandable.

Manage recommendations as decisions

Review whether decisions changed and whether the response reduced barriers or created new harms. Link the one-off assessment to monitoring and community feedback so that changing conditions can trigger re-analysis. Close data access and retention actions at the same time; “published” is not the end of the lifecycle.

22 · Adapt

Sector and cross-cutting lenses

A common analytical frame enables a coherent picture; sector methods supply technical depth. Use current cluster, ministry, Sphere and specialist standards, and preserve the connections between services, access, protection, markets, environment and people’s priorities.

23 · Adapt

Adapt the design to the crisis context

The same quality principles apply everywhere, but pace, access, sources, uncertainty, protection risk and appropriate evidence depth change. State the context assumptions and the trigger for redesign.

For rapid assessment, shorten scope, not safeguards

• Use a smaller set of high-consequence decisions and minimum questions.
• Rely more heavily on coordinated secondary data and direct observation.
• Use explicit uncertainty and update cycles rather than false precision.
• Maintain consent, data minimisation, safe referral, inclusion and stop rules.
• Record what the rapid design omitted and when a deeper assessment is needed.

For sudden-onset coordination, see the UNDAC Handbook and OCHA needs assessment guidance.

24 · Enable

Digital collection, remote methods and AI

Choose technology after defining the method, workflow, risk and users. The best tool is the simplest one that works safely in the operating context and preserves a verifiable record.

Digital collection readiness

Platforms such as KoboToolbox can support offline form-based collection, validation, multiple languages and controlled exports. They do not make a sample representative, obtain meaningful consent, protect unsafe questions or interpret findings. Configure platform and server settings according to the sensitivity and applicable policy.

Platform guidance: UNHCR guidance on KoboToolbox and KoboToolbox data-collection documentation.

Remote methods

Phone, SMS, messaging and web methods can reduce travel and maintain contact, but exclude people by network, device ownership, literacy, language, disability, cost, safety and privacy. Confirm whether a participant can speak privately; avoid leaving sensitive notifications; reimburse costs where appropriate; use multiple call windows; monitor differential non-response; and state that coverage is of the reachable frame, not automatically the whole population.

Responsible uses of AI

AI documentation minimum

• Name the task, tool, provider, model or version where available, date and accountable reviewer.
• Record what source material was processed and confirm it was authorised for that system.
• Test performance across relevant languages and groups, including false positives, omissions and invented content.
• Preserve an audit trail of prompts or settings proportionate to the decision risk.
• Verify all factual outputs against source material; review translations and coded qualitative evidence by qualified people.
• Disclose material AI assistance in the methods and explain its limitations.

25 · Repair

Common assessment failures and how to repair them

Most assessment failures are predictable. Use this diagnostic before launch, at daily review and before publication.

26 · Apply

The field-ready assessment pack

Use these gates at inception, before deployment, during daily review and before release. They are designed to print cleanly, but must be adapted to national requirements, current coordination arrangements and your organisation’s approved procedures.

One-page red-flag check

27 · Communicate

Assessment report and decision-brief template

A useful product lets a busy reader understand the decision, finding, confidence and action quickly, while allowing a technical reviewer to reconstruct how the conclusion was reached.

Two-page decision brief

1. Decision and deadline: the exact choice this brief informs, decision owner and latest useful date.
2. Situation: what changed, affected population, place and relevant time period.
3. Three to five findings: one sentence each with magnitude or pattern, population and confidence.
4. Who is most affected or excluded: disaggregated differences and missing groups, stated safely.
5. Community perspectives: priorities, preferences, capacities, concerns and areas of disagreement.
6. Current response and gap: local, public, market and humanitarian capacity plus access or quality barriers.
7. Options and recommendations: alternatives, trade-offs, risk, feasibility, owner and timeframe.
8. Evidence and limitations: methods, coverage, dates, main uncertainty and what could change the conclusion.

Full report structure

Methods statement model

Limitation statement model

Evidence-to-action table

28 · Define

Humanitarian assessment glossary

Analysis framework

A structured set of questions and concepts that organises evidence around a decision.

Assessment

A process of evidence review, collection where necessary, analysis and communication to inform decisions.

Assessment registry

A shared catalogue of planned, ongoing and completed assessments that supports coordination and reduces duplication.

Bias

A systematic difference between an estimate or account and the condition it is intended to describe.

Cluster sample

A probability design that selects grouped units, often in stages; analysis must account for clustering.

Confidence

A transparent judgement about how strongly available evidence supports a conclusion, given quality, coverage and agreement.

Data minimisation

Collecting and retaining only the data and granularity necessary for a defined purpose.

Denominator

The eligible population or observations to which a count or percentage refers.

Disaggregation

Separating results by meaningful characteristics to identify differences, while managing precision and disclosure risk.

Do no harm

The obligation to anticipate, avoid, mitigate and monitor negative effects created by an activity or decision.

Finding

A bounded statement supported by analysed evidence; distinct from an observation, assumption or recommendation.

Indicator

A defined measure with a numerator or categories, denominator or universe, source, period, calculation and interpretation.

Key informant

A deliberately selected person with role-based or specialist knowledge, not a statistical representative of a population.

Missingness

The absence of a value, which may reflect not asked, not applicable, refusal, do not know, loss or error and must not be conflated.

Non-response

Failure to obtain information from a selected unit or for a selected item; it can introduce bias.

Primary data

Information collected first-hand for the assessment’s defined purpose.

Probability sample

A design in which eligible units have known, non-zero selection probabilities, supporting design-based inference when implemented correctly.

Proxy respondent

A person answering about another person or unit; appropriate only for information they can reasonably know.

Pseudonymisation

Replacing identifiers with a code while retaining a realistic path to re-link; the data remain personal.

Recall period

The defined past interval about which a respondent is asked to remember an event or condition.

Representative

Capable of supporting inference to a defined population under a justified selection and response process; not a synonym for diverse or large.

Secondary data

Information previously collected for another purpose and appraised for relevance to the current decision.

Severity

The seriousness or intensity of conditions, deprivation, risk or consequences affecting life, dignity, rights and wellbeing.

Triangulation

Structured comparison of sources, methods, groups, places or times to examine agreement, explanation and uncertainty.

Unit of analysis

The entity the conclusion describes, such as a person, household, facility, settlement, market or system.

Weight

A value used to reflect unequal selection probabilities or justified adjustments when estimating population results.

30 · Source

Authoritative standards, guidance and tools

This guide prioritises current primary guidance from humanitarian coordination bodies, technical standard-setters and operational organisations. Always confirm the current edition and apply national and country-response protocols.