The monitoring system: MEAL plan, data quality and remote monitoring

1The MEAL plan: the backbone of monitoring

The MEAL plan describes how each indicator will be measured, by whom, when, and for what decision. It takes the form of a table: indicator, definition, baseline, target, source and method, frequency, disaggregation, person responsible for collection, person responsible for analysis, use (monthly review, donor report, targeting adjustment).

It covers four types of monitoring:

• Activity and output monitoring: what has been done, against the plan (distribution registers, attendance lists, worksite sheets).
• Outcome monitoring: changes among people (baseline, midline and endline surveys; post-distribution monitoring).
• Context monitoring: prices, access, security, population movements, and indicators for the logframe's critical assumptions.
• Accountability monitoring: complaints and feedback received, response time, satisfaction, participation (week 8).

The cost of the MEAL plan must be in the budget: staff, tools, transport, surveys, evaluations. A range of 3 to 10% of the project budget is usual; a pilot project whose effect is to be demonstrated can justify more.

The best test of a MEAL plan: for each indicator, can you name the meeting where it will be discussed and the decision it can trigger? If not, ask yourself whether it needs to be collected.

2From collection tools to the dashboard

The data chain has five links: collection, entry and storage, cleaning, analysis, dissemination. Mobile data collection (KoboToolbox, ODK, CommCare) has replaced paper in most contexts: automatic checks at entry (value ranges, required fields, skip logic), geolocation and timestamps, direct upload to the server.

A good form is designed from the analysis plan: first write the tables you want to produce, then the questions needed and only those. It is tested in the field with real respondents before deployment, then a version is frozen; any later change is documented, because it makes data before and after incomparable.

Cleaning follows written rules: duplicates, outliers (a household of 45 people, food expenditure of 2,000 USD per week), inconsistencies (a married 3-year-old), interviews that are too short (under 10 minutes for a 40-minute questionnaire). A value is never corrected without a source; it is flagged, checked with the enumerator or removed, and a cleaning log is kept.

The dashboard shows few indicators, with the target, the trend and a colour code according to thresholds set in advance. It is updated at the pace of decisions: weekly for distributions, monthly for intermediate outcome indicators.

3Data quality: five dimensions

A data quality assessment (DQA) checks five dimensions:

The most concrete method is verification by recount: choose a sample of sites, recount from primary sources (registers, forms) and compare with the reported figures. A verification ratio of 1.00 is perfect; a tolerance is set, for example between 0.95 and 1.05. Above 1.05, the reported figures are inflated; below 0.95, activities are not being reported.

Example. A health centre reports 1,240 consultations; the recount of the registers finds 1,085. Ratio: 1,240 / 1,085 = 1.14. The centre over-reports by 14%: the cause must be understood (double counting of follow-up visits, pressure on targets) before any sanction.

4Quality checks during data collection

Quality is decided during collection, not afterwards. The standard checks:

• Training and testing of enumerators, with a written test and mock interviews; enumerators who fail do not collect data.
• Direct supervision of part of the interviews (for example 5%).
• Back-checks: 10% of households are re-interviewed on a few stable questions (household size, type of shelter). A low agreement rate for one enumerator signals errors or fabricated interviews.
• Daily remote checks: interview duration, GPS coordinates (interviews all in the same place), time (interviews at 3 a.m.), distribution of answers by enumerator (an enumerator whose households all have exactly 5 members).

Fabricated data ("curbstoning") are more frequent than people think when enumerators are paid per interview, under quota pressure, or sent to dangerous areas. Preventing them requires realistic targets, daily pay rather than piece rates, and checks announced in advance.

A MEAL manager who never finds an error is not looking hard enough. Publishing internally the errors found and corrected strengthens confidence in the figures.

5Remote monitoring and third-party monitoring

In areas inaccessible to staff (insecurity, restrictions), remote monitoring combines several sources to make up for the lack of visits:

• Local partners or local staff who have access, with standardised tools and evidence (geotagged photos, signed lists).
• Third-party monitoring (TPM): an independent company or organisation, often required by donors, which verifies a sample of activities.
• Phone calls to a random sample of beneficiaries (post-distribution monitoring by phone).
• Remote data: satellite imagery (camps, crops, destruction), mobile operators' transaction data, price data.
• Community feedback via a toll-free line and messaging.

Remote monitoring has limits: phone samples biased towards households with phones, often represented by men; risks of pressure on partners; transfer of security risks to local staff. This must be stated in reports, partners must be properly paid for this work, and they must not be asked to take risks the organisation would refuse for its own staff.

Triangulation is the rule: a distribution is considered verified when at least two independent sources agree (for example a signed list and calls to a sample of beneficiaries).

6Making monitoring useful: from data to decision

Monitoring rarely fails for lack of data; it fails because nobody uses them. Three mechanisms make monitoring useful:

• Alert thresholds set in advance: if waiting time at distributions exceeds 3 hours, if the rate of unresolved complaints exceeds 20%, if the price of the basket rises by more than 10%, a predefined action is triggered.
• Regular reviews where the programme, MEAL, logistics and finance teams look at the same figures together and decide. The minutes record the decisions, who is responsible and the deadline.
• Feedback to people and teams: present to communities what has been learned and changed; show enumerators what their data were used for.

The independence of MEAL matters. If the MEAL manager reports only to the project manager whose results they measure, the temptation to smooth over bad figures is strong. A dual reporting line (operational to the project, technical to a country or headquarters MEAL manager) protects the quality of information.