Measuring accurately: indicators and surveys

1The Food Consumption Score (FCS)

The FCS asks on how many of the last 7 days the household ate eight food groups. Each frequency (0 to 7) is multiplied by a weight reflecting nutritional value:

• Cereals and tubers: 2; pulses: 3; vegetables: 1; fruit: 1.
• Meat, fish, eggs: 4; milk and dairy: 4; sugar: 0.5; oil: 0.5; condiments: 0.

The maximum score is 112. Usual thresholds: poor from 0 to 21, borderline from 21.5 to 35, acceptable above 35. Where sugar and oil are eaten almost daily, 28 and 42 are used.

Example: cereals 7 days, pulses 3, vegetables 4, oil 7, sugar 5, nothing else. FCS = 14 + 9 + 4 + 3.5 + 2.5 = 33, so borderline consumption.

2Coping strategies: rCSI and LCS

The rCSI (reduced Coping Strategies Index) counts, over 7 days, the days on which the household had to: rely on less preferred or less expensive food (weight 1), borrow food or rely on relatives (weight 2), limit portion size (weight 1), restrict adult consumption so children can eat (weight 3), reduce the number of meals (weight 1). The score ranges from 0 to 56; from about 19, the situation often corresponds to a crisis.

Livelihood coping strategies (LCS) look at the last 30 days and fall into three levels:

• Stress: selling non-productive assets, spending savings, borrowing money.
• Crisis: selling productive assets or tools, cutting health or education spending, eating seed stock.
• Emergency: selling the last female animals, the house or land, begging, turning to illegal or dangerous activities.

The household is classified by the most severe strategy used. A strategy not used because it has already been exhausted counts as used.

3Diversity, hunger and expenditure

• HDDS (Household Dietary Diversity Score): number of food groups, out of 12, eaten in the last 24 hours. It reflects economic access.
• HHS (Household Hunger Scale): three questions over 30 days (no food at home, going to sleep hungry, a whole day and night without eating), score from 0 to 6. From 4 to 6, hunger is severe; the IPC uses it to identify catastrophe situations.
• Food expenditure share: below 50% is reassuring; 50 to 65% indicates vulnerability; 65 to 75% high vulnerability; above 75% very high vulnerability.
• Minimum Dietary Diversity for Women (MDD-W): at least 5 groups out of 10, useful for a nutrition-sensitive response.

4A reliable household survey and CARI

For an estimate per stratum with 95% confidence and a ±10-point margin, you need about 96 households (n = 1.96² × 0.5 × 0.5 ÷ 0.1²). With cluster sampling, multiply by the design effect (often 2), i.e. about 192, plus 10% for non-response.

Good practice: a tested and translated questionnaire, trained male and female enumerators, respected recall periods, phone-based collection with consistency checks, informed consent, no promise of aid. Data are disaggregated by sex of the household head, status (displaced, residents) and area.

WFP's CARI combines current status (FCS) and coping capacity (food expenditure share or economic capacity, and LCS) to classify households into four categories: food secure, marginally food secure, moderately and severely food insecure.

5Worked example: from the sample to the CARI console

This fictional case follows, step by step, the survey of the Sarakoro agropastoral zone (Haut-Kéla district), carried out in October by "Terres Solidaires" for the Cadre Harmonisé and for its programme baseline.

1. Sample size and selection

Population: 11,800 households spread across 23 villages. Expected prevalence of food insecurity: 50% (a conservative assumption that maximises the sample). Desired precision: ±10 percentage points. Confidence level: 95%. This gives 96 households under simple random sampling.

Cluster sampling (villages selected, then households within each village) increases variance, because households in the same village tend to be similar. The design effect is calculated as DEFF = 1 + (m − 1) × ICC, where m is the number of households per cluster and ICC the intra-cluster correlation coefficient. With 10 households per cluster and an ICC of 0.1 (a value often observed for consumption indicators in previous surveys), DEFF = 1 + 9 × 0.1 = 1.9, rounded up to 2. Sample: 96 × 2 = 192, plus 10% non-response, i.e. 211, rounded up to 22 clusters of 10 households (220 households).

The 22 clusters are selected with probability proportional to village size (a large village may be selected twice). In each village, the team uses the household list drawn up with the neighbourhood chief and selects 10 households by systematic sampling; where there is no list, the village segmentation method is used. An absent household is replaced according to a written rule (next household on the list), never at the enumerator's discretion.

2. Household-level calculation

CARI combines two domains. Current status is measured by the FCS, supplemented by the rCSI: an acceptable FCS gives a score of 1 if the rCSI is below 4, and 2 if the rCSI is 4 or above; a borderline FCS gives 3; a poor FCS gives 4. Coping capacity is the average of economic vulnerability (food expenditure share: less than 50% = 1; 50 to 65% = 2; 65 to 75% = 3; more than 75% = 4) and asset depletion (LCS: no strategy = 1; stress = 2; crisis = 3; emergency = 4). The final score is the average of the two domains, rounded to the nearest whole number.

Look at H4: good consumption, but coping capacity already eroded (high food expenditure, crisis strategies). CARI classifies it as marginally food secure, whereas the FCS alone would have judged it food secure. This is exactly what the method is designed to capture. Recent WFP guidance also allows the food expenditure share to be replaced by the economic capacity to meet essential needs (ECMEN), which compares household expenditure with the MEB; the report must state which version was used.

3. The CARI console for the zone

Result: 60% of households in the zone are food insecure (moderately or severely), i.e. around 7,100 households, with a confidence interval of roughly ±9 percentage points given the observed design effect. Results are weighted if clusters were not selected strictly in proportion to size.

4. Disaggregation and interpretation

Caution: with 51 or 38 households, confidence intervals are wide (in the order of ±13 to ±15 percentage points). A trend can be reported, but a precise difference cannot be asserted. The report shows these sample sizes next to each percentage.

What the example shows: a calculated and justified sample, documented selection, a reproducible household-by-household calculation, a console that shows where the final result comes from, and disaggregations presented with their limitations. This level of transparency is what makes data acceptable in an IPC or Cadre Harmonisé workshop.

6Tools: analysis plan, consistency checks and enumerator monitoring

Analysis plan template

Consistency check rules

Raw data are never modified: work is done on a copy, and every correction is recorded in a cleaning log (identifier, variable, original value, corrected value, justification, author). This log is annexed to the report or available on request.

Daily supervision sheet

• Number of interviews planned, completed and refused, absent households replaced.
• Average duration per enumerator.
• Provisional FCS and rCSI means per enumerator.
• Comprehension problems reported (for example: "reduced portions" understood as "skipped meals").
• Security or protection incidents; referrals made.
• Decisions for the following day.

Enumerator training checklist

1. Survey objectives and ethics: consent, confidentiality, no promise of assistance, code of conduct and PSEA.
2. Reading through the questionnaire question by question in the language of the interview, with the exact meaning of each term.
3. Exercises on the FCS food groups using local dishes: in which group should a dish of rice with a sauce containing a little dried fish be counted? (Usual answer: very small quantities used as a condiment do not count in the animal protein group, according to the WFP technical guidance note.)
4. Role plays in pairs, with an observer noting deviations.
5. Field test in a village not selected for the sample, followed by a debriefing.
6. Use of the phone, saving and sending data, procedure in case of loss.

7Common mistakes and what analysts, evaluators and donors look for

The most costly mistakes

• Inflated FCS: counting a food consumed in a tiny quantity (a pinch of powdered milk in tea, a stock cube) as a day of consumption of that group. The FCS becomes artificially "acceptable". The remedy lies in training, with examples of local dishes, and in consistency checks.
• Poorly asked rCSI: asking "did you reduce your meals?" instead of "in the past 7 days, on how many days did you have to…". The question concerns a number of days, from 0 to 7, for each strategy.
• LCS list not adapted: using a generic list that includes strategies that are impossible in the context (selling a house in a camp), or classifying a strategy at the wrong level. The list and its classification into stress, crisis and emergency are validated at the level of the cluster or the analysis group.
• Miscoded "not applicable": a household that no longer sells animals because it has none left must be coded "no, because already exhausted", which counts as a strategy used. Coding it "no" underestimates severity.
• Incomplete expenditure: forgetting purchases on credit or annual non-food expenditure converted to a monthly figure, or mixing different recall periods without conversion.
• Thresholds changed along the way: switching from the 21 and 35 thresholds to 28 and 42 between the baseline and the endline survey, which creates a false deterioration or improvement.
• Missing confidence intervals: claiming a 5-point improvement when the confidence interval is ±9 points.
• Comparing different seasons: comparing a lean-season baseline with a post-harvest endline survey and attributing the improvement to the project.

What IPC and Cadre Harmonisé analysts look for

• The date and period covered: data more than a few months old, or pre-dating a major shock, carry less weight.
• Representativeness: does the sample cover the unit of analysis (district, livelihood zone)?
• Method: standard modules, sample size, selection, recall period.
• Consistency between indicators and with other sources.
• Availability of raw data or at least detailed tables.

What evaluators and donors look for

• A baseline conducted before start-up, in the same season as the endline survey or with an explanation of the seasonal effect.
• Standard indicators matching those in the donor's results framework (for BHA and ECHO, typically FCS and rCSI at outcome level, sometimes LCS and HDDS).
• Disaggregation by sex, age and disability as required by the donor and by the organisation's commitments.
• Data protection: consent, anonymisation of shared datasets, retention period.
• Separation between the team that collects data and the team that implements, or at least independent supervision.

A survey is not judged on its figures but on its ability to defend them. If an analyst asks you in a workshop "how many households, what selection, what design effect, what thresholds, what date?", you must be able to answer in one minute, with documents to back it up.

8Adapting surveys to the context

The particular case of telephone surveys

WFP has scaled up remote surveys (mVAM) since the mid-2010s. They allow frequent, low-cost monitoring, but respondents are on average better off, more urban and more often male. Good practice: set quotas, weight according to the population, compare regularly with a face-to-face survey, focus on trends rather than absolute levels, and do not change the data collection mode between the baseline and the endline survey without reporting it.

9Interview questions and key references

Interview questions

1. "Calculate the sample size for two strata using cluster sampling."

A good answer sets out the formula, justifies the 50% prevalence, the precision and the confidence level, explains the design effect (and its formula based on the ICC), adds non-response, multiplies by the number of strata and translates the result into a number of clusters and households per cluster.

2. "Explain CARI to a non-specialist country director."

The answer explains in a few sentences that we look at what households eat today and what they sacrifice to achieve it, that each dimension is scored from 1 to 4, and that the average classifies households into four groups; it gives the example of a household that eats well but sells its productive assets.

3. "Your results show a 6-point improvement in FCS between the baseline and the endline survey. Can you attribute it to the project?"

The answer checks the confidence intervals, the season, the comparability of methods, trends in areas not covered or in a comparison group, and other interventions in the area. It speaks of contribution rather than attribution if there is no comparison design.

4. "One enumerator obtains much higher FCS values than the others. What do you do?"

The answer describes detection, verification without accusation (observation, re-interviewing a sample), retraining, a documented decision on the data concerned, and prevention through daily supervision.

5. "When should the FCS thresholds of 28 and 42 be used?"

The answer: when sugar and oil are consumed almost every day by most households, which inflates scores; the decision is taken before the survey, consistent with the cluster and with past practice, and kept for the whole series.

Key references

• Technical guidance note on the Food Consumption Score and guidelines for calculating and using food security indicators: World Food Programme.
• The Coping Strategies Index: Field Methods Manual (second edition): CARE and WFP.
• Technical guidance on Livelihood Coping Strategies (LCS-FS and LCS-EN): WFP.
• CARI guidance, latest edition, and guidance on the ECMEN indicator: WFP.
• Household Hunger Scale (HHS) and Household Dietary Diversity Score (HDDS) guides: FANTA.
• Minimum Dietary Diversity for Women (MDD-W) guide: FAO and FHI 360.
• Washington Group Short Set of Questions on Disability.
• IPC Technical Manual 3.1: reference tables for outcome indicators.