Analysis, dashboards, maps and data protection

1Analysing and presenting

• Choose the indicator that answers the question, and the right denominator (a percentage of what?).
• Disaggregate by sex, age, disability and area where relevant.
• One chart per message: bars to compare, lines for a trend, no pie charts with ten slices, no 3D effects.
• Always state the source, date, coverage and limitations of the data.

A useful dashboard fits on one page: a few key figures, the trend, the gaps, and what requires a decision. Tools: Excel, Power BI, Looker Studio, depending on the team's resources.

2Humanitarian mapping

• Tools: QGIS (free) for maps, OpenStreetMap data (roads, buildings), COD boundaries.
• Common types of maps: 4W presence map, needs severity map, access map (cut-off roads, high-risk areas), service location map.
• A good map has: a title that states the message, a legend, a scale, a north arrow, the source, the date, and the boundary disclaimer ("the boundaries shown do not imply official endorsement or acceptance").
• Legible colours (including for people with colour blindness), consistent value classes, little information but the right information.

3Data protection and data responsibility

• Personal data: name, phone number, precise location, biometric data; sensitive data: health, violence, ethnicity, legal status.
• Principles: collect the minimum, with consent, for a specific purpose; restricted access; encryption; a defined retention period; sharing governed by an agreement.
• Even aggregated data can expose a group (for example a map that locates members of a minority or survivors of violence): assess the impact before publishing.
• The IASC Operational Guidance on Data Responsibility in Humanitarian Action (2023) and cluster standards serve as references.

4The data analysis plan: weighting, confidence intervals and publication rules

A clean dataset does not by itself produce a sound analysis. Between the two lies the data analysis plan, drafted before data collection ends: for each indicator in the matrix, it specifies the question asked, the calculation, the denominator, the disaggregations, the weighting and how the result will be presented. Coordinated multi-sector assessments require one, and it is the first document an evaluator asks for when checking a figure.

Minimum content of a data analysis plan

Weighting: why and how

When each stratum receives the same number of interviews although the populations are of different sizes, the unweighted average over-represents the small strata. The weight of an interview is the ratio between its stratum's share of the population and its share of the sample: weight = (N stratum / N total) / (n stratum / n total).

If 52%, 30% and 41% of households in these three strata have less than 15 litres, the unweighted average is 41% and the weighted average is 0.48 × 52 + 0.21 × 30 + 0.31 × 41 = 43.9%, i.e. 44%. The gap looks small here; it often reaches 5 to 10 percentage points when stratum sizes are very unequal.

Confidence intervals and small sub-populations

• A result drawn from a sample is published with its margin: “44% (95% confidence interval: 39 to 49%)”. The approximate interval for a proportion is p ± 1.96 × the square root of p(1 − p)/n, to be widened by the design effect for cluster sampling.
• Sub-groups not planned for in the sampling design (for example female-headed households in a given municipality) are indicative, and this is stated. A common rule is not to publish a percentage calculated on fewer than 30 observations, or to flag it clearly.
• Two results whose intervals overlap substantially must not be presented as different: “the situation is worse in Kessa than in Lounda” requires a statistically significant difference, not just two different figures.

Every published figure must be able to answer three questions: a percentage of what (the denominator), measured how (method and date), and with what precision (n, margin of error or indicative status).

From descriptive to explanatory analysis

Humanitarian analysis does not stop at percentages. The Joint Intersectoral Analysis Framework (JIAF) requires humanitarian conditions to be linked to their drivers and consequences, then a severity to be assigned on a scale from 1 (minimal) to 5 (catastrophic) by area and population group, before estimating the number of people in need. The IMO prepares the indicators; interpretation takes place in a joint workshop with the sectors, because that is where the figures are tested against field knowledge.

5The 4W and consolidating people reached without double counting

The “who does what, where and when” matrix (3W, 4W or 5W depending on the country, the fifth dimension being “for whom”) is the product most frequently requested from a cluster IMO, and the one most often wrong. The core problem is double counting: the same family may receive a hygiene kit from one NGO, a water distribution from another and an awareness session from a third, and be counted three times.

A cluster's minimum 4W template

Consolidation rules

1. Within an activity and a municipality: people reached by different organisations are added up only if they are known to serve different populations (different sites, separate lists). Otherwise, the maximum is taken.
2. Across activities within the same sector: no summing. The sector figure for people reached per municipality is generally the maximum of the key activities, or that of a reference activity defined by the cluster.
3. Across months: a distinction is made between new people (first contact) and people assisted repeatedly. The annual cumulative figure is not obtained by adding up twelve months of recurring distributions.
4. Across sectors: the intersectoral figure for people reached per municipality is often the maximum of the sector figures, which slightly underestimates but avoids overestimation. The method is decided at IMWG level and written into the dashboard's methodological note.
5. Logical ceiling: people reached cannot exceed the population of the municipality. Any excess triggers a verification.

Example of consolidation for one municipality

A 4W does not measure needs met, but operational presence. Comparing people reached with people in need gives an order of magnitude of coverage, never proof that needs have been met: the quality, quantity and duration of assistance are not captured.

Timeline of a monthly cycle

• Days 1 to 5 of the following month: partners submit their 4W (online tool, for example ActivityInfo, or a locked Excel template).
• Days 6 to 8: the IMO runs checks (P-codes, inconsistent disaggregations, figures above the ceiling) and sends queries back to partner focal points.
• Days 9 to 10: consolidation, update of the dashboard and the presence map, submission to OCHA for response plan monitoring.
• Cluster meeting: presentation of coverage gaps, not a mere reading of figures.

6Producing dashboards and maps that support decision-making

Designing a dashboard in five steps

1. Identify the user and their decision: a humanitarian coordinator arbitrating priorities across sectors does not need the same product as a base manager monitoring weekly distributions.
2. Choose four to eight key indicators from the matrix, with their target or alert threshold. Beyond that, the reader no longer sees anything.
3. Sketch the mock-up on paper before opening the software, and have the user validate it. Reading usually runs top to bottom and left to right: key figures, trends, geographic distribution, gaps and alerts.
4. Build from a single source: the dashboard reads from the consolidated dataset and contains no figures typed in by hand. This is the condition for an update to take an hour rather than two days.
5. Document: a visible “methodology and sources” box indicates the date of the data, the coverage (number of partners that reported), the definitions and the limitations.

Choosing the right chart

Method for producing a thematic map in QGIS

1. Load the COD-AB boundaries at the required level, in the current version.
2. Prepare the data table with an identical P-code column (same format, same level) and numeric values with no text.
3. Perform the join on the P-code, then check the number of joined features: any unit left unjoined reveals an incorrect code or a different version of the boundaries.
4. Choose the coordinate reference system: WGS 84 (EPSG:4326) for data exchange; a metric projection (for example the country's UTM zone) to measure distances or areas.
5. Choose the representation: a choropleth for a rate or percentage; proportional circles for numbers of people. A choropleth of counts gives a misleading impression, because large units always appear more affected.
6. Choose the classes: fixed thresholds with operational meaning (for example the five severity levels, or sector thresholds), otherwise quantiles or natural breaks, stating which. Five classes at most, a single-hue sequential palette for intensity, readable by people with colour blindness.
7. Compose the layout: a title that states the message, legend, scale bar, north arrow, inset location map, sources, production date and data date, logos, boundary disclaimer.
8. Have it reviewed by a second person: P-codes, empty units, spelling, disclaimer, absence of sensitive data.

An area with no data must never be given the colour of the lowest class. It receives a distinct colour or hatching and the label “no data”, otherwise the map asserts that there are no needs where nothing is known.

Pitfalls in interpreting maps

• Zoning effect: the same data aggregated by district or by municipality can produce very different pictures. Choose the level that matches the decision and avoid drawing conclusions at a finer level than that of the data collection.
• Area effect: large, sparsely populated units dominate visually. Proportional symbols or a grid map correct this effect.
• Data from different dates combined on the same map: avoid, or flag.

Tools and realistic timeframes

Excel is sufficient for a project dashboard. Power BI or Looker Studio are suitable for data updated regularly and viewed online, provided access rights are managed (a public link exposes all the underlying data, including data not displayed). For a standard map based on an existing template, allow half a day; for a new product (mock-up, validation, first production run), one to two weeks.

7Full example: a shelter cluster's monthly product, from the 4W to the map

Fictional context: the Valdorie Shelter and Non-Food Items Cluster has 23 active partners in the Haut-Lisse region. The response plan sets 96,000 people in need and a target of 54,000 people for the year. The cluster coordinator is preparing the meeting with the inter-cluster coordination group, which must reallocate USD 1.2 million from the country-based pooled fund. The cluster IMO has four days.

Day 1: receiving and checking the 4W

19 of the 23 partners have reported (83%). The checks reveal: 2 partners still using P-codes from the old version, 1 partner reporting households rather than people, and 3 rows where people reached exceed the displaced population of the municipality according to mobility tracking. After discussion with the focal points, the codes are corrected, households are converted using the average household size from the assessment (6.2 people), and 2 of the 3 rows are revised downwards (two distributions to the same households had been added together).

Day 2: consolidation and analysis

The key message emerges: Tessil, at severity 4, is the least covered district, while half of the partners are concentrated in Morane, which is more accessible. The IMO tests this hypothesis against the Logistics Cluster access map: the main road to Tessil has been constrained for two months (escorts, damaged bridges), which partly explains the limited presence.

Day 3: production

• One-page dashboard: four key figures (people reached, % of target, active partners, 83% reporting rate); sorted horizontal bars showing the percentage of target reached by district; monthly line chart of people reached; a “gaps” box on Tessil and Hautes-Rives.
• Map: choropleth of severity by district (five-class sequential palette), proportional circles for people reached, hatching for Hautes-Rives labelled “no data this month”, constrained roads drawn. Title: “Haut-Lisse: shelter coverage remains low in Tessil, a severity 4 district”.
• Methodological note of five lines: sources, date, de-duplication rule (maximum per municipality and per activity), reporting rate, conversion of households into people.

Day 4: review and dissemination

Review by the OCHA IMO and the coordinator; the Hautes-Rives partner sends its figures late, and they are integrated into the next version rather than by silently modifying the product already disseminated. The dashboard is presented to the inter-cluster coordination group with an explicit recommendation: prioritise Tessil in the allocation, combined with funding for access logistics. The public product is posted online without any precise list of sites for areas where displaced people are under threat; the detailed site-level version remains restricted to partners.

What the example shows: the value of the IMO's work lies not in the chart, but in the checks that make the figures credible, in cross-referencing with other data (access) that explains the gap, and in framing a message that can be used to make a decision.

8Protecting data day to day: life cycle, anonymisation and statistical disclosure control

Data protection is not settled by a clause in a contract: it is played out in everyday practice, at every stage of the data life cycle. The ICRC and Brussels Privacy Hub Handbook on Data Protection in Humanitarian Action, the IASC Operational Guidance on Data Responsibility and, for organisations established in the European Union, the General Data Protection Regulation (GDPR) set the framework.

Measures by stage of the life cycle

Truly anonymising: statistical disclosure control

Removing names is not enough. A combination of apparently harmless variables (municipality, age, sex, disability, number of children, occupation) can single out one person in a small village. These variables are called quasi-identifiers. OCHA's Centre for Humanitarian Data recommends assessing the risk of re-identification before sharing microdata, for example with the open-source tool sdcMicro.

1. List the quasi-identifiers and the sensitive variables in the dataset.
2. Measure the risk: how many records are unique on the combination of quasi-identifiers? The principle of k-anonymity requires at least k records to share each combination (k = 3 or 5 are common values).
3. Reduce the risk: group (age into bands, municipality replaced by district), locally suppress rare values, remove unnecessary variables, round coordinates or remove them.
4. Check utility: are the planned analyses still possible? The aim is the best documented trade-off.
5. For aggregated tables: suppress or merge cells with small counts (a widespread rule sets the threshold at fewer than 5 people), making sure they cannot be recalculated from the totals.

The risk does not concern individuals alone. A map showing the concentration of an ethnic or religious group, or the location of women living alone in a site, can expose an entire group: this is what the IASC guidance calls demographically identifiable data. They are handled with the same caution as personal data.

Responding to data requests

Requests for lists of names come from partners, local authorities, sometimes donors or armed forces. The response is prepared before the request arrives:

• What is the exact purpose? Can it be achieved with aggregated data? In the vast majority of cases, yes.
• Is there a legitimate basis, and were people informed in advance about this sharing?
• What is the risk to people if the data are misused or leaked?
• Who decides? Sharing personal data is approved at country management level, never by the IMO alone nor by the field team under pressure.

With donors, verification of beneficiary lists is generally negotiated as on-site consultation or a sample, rather than the transfer of complete files; the organisation must know what its agreement says on this point.

In the event of a data breach

Loss of a survey phone, a dataset sent to the wrong recipient, a compromised account: contain the breach (revoke access, remote wipe), assess the risk to people, inform management and, depending on the applicable framework, the supervisory authority and the people concerned. The GDPR, for example, requires notification of the supervisory authority within 72 hours where the breach poses a risk. Document the incident and fix its cause, even when no harm has been observed.

9Common mistakes, donor expectations and adapting to the context

Common mistakes in analysis and dissemination

What donors and evaluators examine

• Consistency of figures between the donor report, the cluster 4W and financial reports (a number of kits distributed must match stock releases).
• Disaggregation by sex, age and disability, which most major humanitarian donors require; ECHO's Gender-Age Marker is checked against the data actually reported.
• Participation in coordination: OCHA-managed pooled funds and many donors regard regular reporting to the cluster 4W as a condition of partnership.
• Management of personal data: data protection policy, impact assessments for high-risk processing, data sharing agreements, particularly for cash transfer programmes.
• Actual use: an evaluator asks how the data changed decisions; a dashboard that is never consulted is a cost, not a result.

Adapting to the context

10Interview questions and key references

Interview questions

1. “People reached in your 4W exceed the population of several municipalities. What do you do?”

A good answer identifies the likely causes (double counting between organisations or between months, confusion between households and people, a denominator that excludes displaced people), describes the verification with partners, the application of written consolidation rules, and stating the method in the product. It does not correct figures silently.

2. “Why weight survey results?”

The candidate explains stratification and the difference between share of the sample and share of the population, gives the weighting formula and a worked example. They mention that results by stratum are not weighted against each other, only the aggregates are.

3. “Walk us through how you would produce a severity map by district.”

Up-to-date data and P-codes, a checked join, a choropleth with classes that have operational meaning, a “no data” category, mandatory layout elements, disclaimer, review, and consideration of how sensitive the information shown is.

4. “A partner asks you for your programme's beneficiary database with names to avoid duplication of assistance. How do you respond?”

A good answer acknowledges the legitimate objective, explores alternatives (matching on a pseudonymised identifier, a shared de-duplication system with access rules, matching by a trusted third party), checks what information was given to people, asks for a data sharing agreement and management approval, and refuses to send the file by email.

5. “How would you know whether a dashboard is useful?”

A good answer cites the decisions it has informed, usage (access statistics, use in meetings) and user feedback, and proposes removing elements that are never used.

6. “What is k-anonymity and when do you apply it?”

Definition, quasi-identifiers, risk-reduction techniques, trade-off with utility, application before any sharing of microdata and for tables with small counts.

Key references

• Operational Guidance on Data Responsibility in Humanitarian Action: IASC (revised 2023 version).
• Data Responsibility Guidelines and guidance notes from the Centre for Humanitarian Data (statistical disclosure control, impact assessments): OCHA.
• Handbook on Data Protection in Humanitarian Action: ICRC and Brussels Privacy Hub.
• The European Union's General Data Protection Regulation (GDPR), for organisations subject to it.
• Joint Intersectoral Analysis Framework (JIAF) and guidance on calculating people in need: IASC.
• OCHA guidance on the 3W/4W and on monitoring humanitarian response plans; OCHA map product standards.
• QGIS and sdcMicro documentation.
• IASC Guidelines on the Inclusion of Persons with Disabilities in Humanitarian Action (2019), for disaggregation.