(2025) Enquête à haute fréquence Haïti 2025 : plan de sondage et pondération

Haiti High Frequency Survey (HFS) - 2025 Sampling Design and Weighting June 23th 2025 Haiti High Frequency Survey (HFS) - 2025 Sampling Design and Weighting The sample for the Haiti HFS-2025 was generated using a Random Digit Dialing (RDD) methodology, which covered all active cell phone numbers at the time of sample selection (May 2025). The survey estimates represent households with at least one cell phone, and individuals aged 18 or older residing in those households. 1. Sampling design The RDD methodology generates all possible phone numbers under the national numbering plan and draws a random sample. This approach ensures full coverage of the population with a mobile phone.1 A large first-phase sample was drawn from the complete number frame. An automated screening process identified active numbers. These active numbers were matched against business registries (e.g., yellow pages and websites) to remove business numbers, which are ineligible for this survey. A smaller second-phase sample2 was then selected from the list of active numbers identified in the firstphase sample and handed over to the field team for contact and interviews.3 2. Weighting This survey includes two sample units: households and individuals. Sampling weights were computed for each unit and should be used according to the estimate of interest. The weighting process follows five steps: 1. 2. 3. 4. 5. Estimation of cell phone inclusion probabilities. Computation of design weights for households and individuals. Adjustment for nonresponse. Calibration using external population data (adjusted for national phone coverage). Trimming and recalibration of weights. 1 Given that the survey used a sampling frame of telephone numbers, results represent the population with at least one active phone and exclude the population with no phone. 2 Note that the selection of phone numbers involves two sampling phases, and not two sampling stages. The survey involves only one sampling stage. 3 Furthermore, the second-phase sample was delivered in batches to the country teams during fieldwork. Delivering large lists of numbers could have facilitated the “misuse” of the sample by easily replacing non-answering numbers, raising nonresponse rates and potentially increasing nonresponse biases. 2 Step 1: Cell Phone Inclusion Probabilities A first-phase sample was selected using simple random sampling without replacement. The selected numbers were then screened and classified into active and inactive. The first-phase inclusion probabilities of cell phone numbers are4  𝐶(1)𝑖 = 𝐶 𝑛(1) 𝐶 𝑁(1) = 𝐶 𝐶 𝑛(1)𝐴 + 𝑛(1)𝐼𝑁 𝐶 𝑁(1) where  𝐶(1)𝑖 is the first-phase inclusion probability of the i-th cell phone number; 𝐶 𝐶 𝑛(1) is the size of the first-phase sample of cell phones, composed of 𝑛(1)𝐴 active cell phones and 𝐶 𝑛(1)𝐼𝑁 inactive cell phones; 𝐶 𝑁(1) Total number of possible cell numbers under the national plan.; Next, a second-phase sample was selected systematically out of the first-phase samples of active cell telephone numbers. The second-phase inclusion probabilities of cell phones are  𝐶(2)𝑖|(1)𝑖 = 𝐶 𝑛(2)𝐴 𝐶 𝑛(1)𝐴 where  𝐶(2)𝑖|(1)𝑖 is the second-phase inclusion probability of the i-th active cell phone number conditional on being selected in the first phase; 𝐶 𝑛(2)𝐴 is the size of the second-phase sample of active cell phones; The unconditional inclusion probabilities of the second-phase active cell phones are  𝐶𝑖 =  𝐶(1)𝑖  𝐶(2)𝑖|(1)𝑖 = 𝐶 𝐶 𝐶 𝑛(1)𝐴 + 𝑛(1)𝐼𝑁 𝑛(2)𝐴 𝐶 𝑁(1) 𝐶 𝑛(1)𝐴 = 𝐶 𝐶 𝐶 𝑛(1)𝐴 + 𝑛(1)𝐼𝑁 𝑛(2)𝐴 𝐶 𝑛(1)𝐴 𝐶 𝐶 𝑛(2)𝐴 𝑛(2)𝐴 = 𝐶 𝐶 = ̂𝐶 𝐶 𝑁(1) 𝑅𝐴(1) 𝑁(1) 𝐴̂(1) 4 Inclusion probabilities of cell phones do not show a stratum index since most cell phone samples were not stratified for the reasons stated above. 3 where ̂ 𝑅𝐴(1) is the rate of active phones estimated in the first phase.5 Hence, the unconditional inclusion probabilities of the second-phase active numbers  𝐶𝑖 can be expressed as the ratio between the active numbers selected in the second phase and an estimate of the total active numbers in the frame 𝐴̂(1) . Step 2: Design weights for households and individuals The selection probabilities for households and individuals aged 18 and over are derived from the inclusion probabilities of the cell phone numbers through which they are reachable. Therefore, the computation of household and individual weights should account for multiple chances of selection. This multiplicity weighting adjusts estimates to eliminate the over-representation of households and individuals in the sample that can be reached through more telephone numbers than other households and individuals, thereby reducing bias associated with unequal probabilities of selection. Multiplicity adjustment There is multiplicity probability when a household has a larger selection probability because it can be selected through different sample elements (telephone numbers). Households with more than one cell phone number are over-represented in sample designs like this. As a result, their selection probabilities need to be adjusted to account for this increased chance of selection. The multiplicity-adjusted household selection probabilities are computed as  𝐶𝑚𝑗 = 𝑚𝑐𝑗  𝐶𝑖 where  𝐶𝑚𝑗 is selection probability of the j-th household when contacted through a cell phone, adjusted for multiplicity of working cell phones in the household; 𝑚𝑐𝑗 is the number of working cell phones in the j-th household; Therefore, if a household has mc cell phones, its chance of being selected through a cell phone is mc higher than a household where there is only one cell phone. Since the number of cell phones in a household is unknown at the time of the sample design, it needs to be asked during the interview in the questionnaire. For this purpose, the survey collected information about the number of cell phones in the respondent households through the following question: How many working cell phones in total are owned by the persons in your household, including you? The probability of an individual being selected through a cell phone equals the inclusion probability of his or her cell phone number.  𝐶𝑘 =  𝐶𝑖 where 5 ̂ 𝑅𝐴(1) estimates are highly precise due to the very large size of the first-phase samples. 4  𝐶𝑘 is the selection probability of the k-th individual when contacted through a cell phone. Household and individual design weights, w0j and w0k respectively, are the inverse of the above selection probabilities 𝑤0𝑗 = 𝜋𝑗−1 𝑤0𝑘 = 𝜋𝑘−1 Weighting of data on children & adolescents HFS collected specific data about a randomly selected child (0 – 5) or adolescent 6 through 17 years of age in each interviewed household. To implement this, the questionnaire first collected a roster of all children and adolescents living in each respondent household and selected one at random. The child/adolescent weight is based on his/her probability of selection within the household, conditional on his/her household being selected in the sample. Hence  𝐶𝑛𝑗 =  𝐶𝑗 1/ ∑𝑗 𝑛 where  𝐶𝑛𝑗 is the selection probability of the n-th child/adolescent in the j-th household when the household is contacted through a cell phone,  𝐶𝑗 is selection probability of the j-th household when contacted through a cell phone, adjusted for multiplicity of working cell phones in the household; ∑𝑗 𝑛 is the number of eligible children and adolescents (6-17 years old) in the j-th household; Children and adolescents’ design weight w0n is the inverse of the above selection probabilities 𝑤0𝑛 = 1/ 𝐶𝑛𝑗 Step 3: Nonresponse adjustment When a phone number is called, it is not always possible to carry out an interview. Nonresponse occurs because of a number of constraints. Most common are that nobody answers the call (no contact), the respondent is unwilling to cooperate (refusal), or language barriers exist. The design weights of responding households and individuals were adjusted for nonresponse. This adjustment is based on the inverse of the weighted response rate estimate. This is the ratio of the sum of the design weights of all units (respondents and nonrespondents) to the sum of the design weights of respondents. 5 𝑎𝑗 = ∑𝑗,𝑅 𝑤0𝑗 + ∑𝑗,𝑁𝑅 𝑤0𝑗 ∑𝑗,𝑅 𝑤0𝑗 ; 𝑎𝑘 = ∑𝑘,𝑅 𝑤0𝑘 + ∑𝑘,𝑁𝑅 𝑤0𝑘 ∑𝑘,𝑅 𝑤0𝑘 where aj is the nonresponse adjustment factor that should be applied to responding households and ak is the nonresponse adjustment factor for responding individuals. R and NR indicate the responding and nonresponding units, respectively. Thus, the nonresponse adjusted weights for responding households and individuals are 𝑤′𝑗 = 𝑤0𝑗 𝑎𝑗 ; 𝑤′𝑘 = 𝑤0𝑘 𝑎𝑘 Step 4: Calibration Finally, the weights for responding households and individuals were calibrated to align with the distribution of the phone-owning population by sex, age, and region, based on external data from official national sources. Calibration was performed by minimizing a measure of the distance between the input weights (nonresponse adjusted weights in this case) and the calibrated weights, under the constraint that the sum of the calibrated weights equals the sum of the totals of the auxiliaries from the external source. Unlike the nonresponse adjustment, weights calibration requires auxiliary variables for respondents only. Among available calibration methods, this survey employed raking (iterative proportional fitting) using a logit distance function, which ensures that calibrated weights remain within acceptable bounds while satisfying marginal control totals. The final weights for responding households and individuals can then be expressed as 𝑤𝑗 = 𝑤𝑗′ 𝑔𝑗 = 𝑤0𝑗 𝑎𝑗 𝑔𝑗 𝑤𝑘 = 𝑤𝑘′ 𝑔𝑘 = 𝑤0𝑘 𝑎𝑘 𝑔𝑘 where 𝑤0𝑗 is the design weight for the j-th household; 𝑎𝑗 is the nonresponse adjustment factor for households; 𝑔𝑗 is the calibration factor for the j-th household; 𝑤0𝑘 is the design weight for the k-th individual; 𝑎𝑘 is the nonresponse adjustment factor for individuals; and 𝑔𝑘 is the calibration factor for the k-th individual. 6