Sampling plan and post-distribution monitoring questionnaire

Step 1: n0 = z² × p(1 - p) / e² = 1.96² × 0.5 × 0.5 / 0.05² = 384.2 (p = 0.5 because the proportion is unknown: this is the most conservative case).

Step 2, finite population correction: n = n0 / (1 + (n0 - 1) / N) = 384.2 / (1 + 383.2 / 4,800) = 355.8.

(a) Simple random: 355.8 / 0.9 = 395.3, i.e. 396 households.

(b) Clusters: 355.8 × 2 = 711.5; / 0.9 = 790.6, i.e. 791 households, rounded to 80 clusters of 10 (800).

Recommendation: the complete list exists, so random sampling (stratified by commune) is possible and requires half as many interviews. But 396 randomly selected households will be scattered across almost all the villages, which is costly in travel in Androy. The real cost must be compared: if the team can cover the selected villages in the time available, choose (a); otherwise (b), bearing in mind that the design effect of 2 is an assumption to be checked after the survey.

(a) Proportional stratification (396 × households in the commune / 4,800, rounded up): Ambovombe 90, Ambanisarika 35, Ambazoa 45, Antanimora Atsimo 55, Maroalopoty 30, Sihanamaro 50, Erada 40, Jafaro 55; total 400. Systematic selection from each commune's list: interval = households / sample (Ambovombe: 1,080 / 90 = 12), random starting point between 1 and 12, then every 12th household.

(b) Clusters (80 × households / 4,800): 18, 7, 9, 11, 6, 10, 8, 11 clusters, i.e. 80. Villages selected with probability proportional to their number of beneficiaries, then 10 households randomly selected from the village list.

Replacements: reserve list drawn at the start (for example 10%), used in order and only after two unsuccessful visits; each replacement is recorded with its reason.

1. Double-barrelled question (amount and times): split into two questions with a satisfaction scale, and ask why in case of dissatisfaction.
2. No reference or period: "At the [month] distribution, what amount did you receive in person?" then compare with the planned amount.
3. Leading question: "What did you spend this money on?" (multiple choice, not read out, with approximate amounts).
4. No recall period and not standard: use the food consumption score module (food groups, number of days of consumption over the last 7 days).
5. Sensitive question asked in public: ask it in private, without the village chief; neutral wording ("Has anyone asked you for money or a favour in connection with the assistance?"); plan for referral to the complaints mechanism.
6. Yes/no invites a yes: "If you had a problem with the assistance, how would you report it?", answer checked by the enumerator.
7. Excessive collection: replace with the household ID; no unnecessary identity data; obtain informed consent at the start.

General flaw: distribution staff acting as enumerators create a courtesy bias and prevent households from complaining about those same staff.

Quality: enumerators independent of the distribution, trained for two days with a pilot test of the questionnaire; electronic form with consistency checks, mandatory fields and GPS; back-checking of about 5% of interviews by phone call or visit; daily monitoring dashboard (interview duration, outliers, non-response rate per enumerator); post-distribution monitoring 2 to 4 weeks after the distribution.

Allegation: the enumerator does not ask detailed questions and does not investigate; he or she informs the person of the services available and, with their agreement, of how the allegation will be passed on; he or she reports the information the same day, confidentially, to the PSEA focal point, without recording it in the survey database. All enumerators are trained in this procedure and sign the code of conduct before the survey.