1. Discuss the results presented in table HH.1 and describe the importance of this table in any survey.

Answer: Nigeria Immunization Cluster Survey, 2016/2017 was designed to assess routine immunization among 12 -23 months children. Six geopolitical regions were taken as six strata and further each region was dichotomized into urban and rural.

Out of 42,981 households visited (to receive required number of respondent to make the survey representative), 41,059 household were found to be occupied (i.e., at least one member was present during the survey). 40,518 household participated in the survey. Household level response rate was 98.7% (40518/41059*100). Among 40,518 household interviewed, 6360 children were found to be illegible for the surveys. The survey could gather information of 6268 children, response rate 98.6 %( 6268/6360*100). However, the overall response rate among 12-23 months children (taking account of household and children non response rate) was 97.3 %( (98.7*98.6)/100).

Household response rate for rural and urban was 99% and 98%. The overall response rate among 12-23 months children for rural and urban was 97.5 % and 96.8%. Rural response rate is slightly higher than urban response rate. Overall response rate in six geological regions of Nigeria ranged from 96% to 99.2%

The table with sample implementation compares estimated number with ground reality (i.e., estimated number of household/children versus household/children found/interviewed in the survey). In survey design, estimated number of children is derived based on parameters such as confidence level, desired coverage, effective sample size etc. in order to get desired precision in survey result with an assumption that sample will be representative of the population. During the field work, actual number of children found for survey will be less than estimated number of children As a result, survey result should be interpreted considering uncertainty introduced due to non-response, especially for interpreting unweighted results. The sample implementation table will provide information and has become an essential table for any survey report

2. From tables HH.3 and HH.5, (a) describe the key differences between weighted and unweighted household composition findings. (b) What factors would influence the differences observed between the weighted and unweighted results.

Answer: Table 1 describes characteristics of surveyed household. Similarly, table 2 describes surveyed 12-23 months children. To describe sample population, weighted analysis is not necessary. However, weight is required to draw inferences on population based on the survey.

Among surveyed households, 27982 (76.5%) household were headed by male and 5919 (14.6%) households were headed by female. In Nigeria, majority of household (86.4%) household are headed by male and 13.6% of household is headed by female. Unweighted N for male is less than weighed N. Unweighted N for Female is more than weighted N. This clearly shows that male are under-sampled and female were over-sampled.

Among surveyed population, 7395 (18.25%) households were from North-West zone. 7244 (17.9%) are from South -South Region. After applying weight, 10944 (27%) households were found to be North-West Region. Though the same number of household sampled in each of the two regions, a sampled household from North-West region represented larger number of household compared to that of South-South region.

Similarly, 3111(49.6%) boys and 3157 (50.3%) girls were surveyed in NICS. 3121 (49.8%) boys and 3147 (50.2%) of age 12-23 months were given by weighted analysis. In this survey, boys were under sampled compared to girls. Hence, N has increased in case of boys and decreased in case of girls after applying weight.

Factors that influences the weighted and unweighted number is weight. Survey data is adjusted for effect for design, non-respondent and post-stratification. Design effect is reciprocal of product of probability of selection of an enumeration area and probability selecting a household in a cluster and probability of selecting a segment (if an EA required segmentation). This weight is required to be adjusted for household level non response and for respondent level nonresponse; these weights are calculated at cluster level.

3.The results of table IM.1 are based on a weighted analysis (a) Discuss the findings of table IM.1 (b) How does sample size influence the findings in table IM.1? (c) Describe what would be the key differences in results of table IM.1 if they were computed on the unweighted sample.

The table provides weighted crude coverage for first does, second dose and third dose of Penta (DPT-Hep B-Hib 3) disaggregated by background characteristics. Confidence interval is provided along with point estimate. National DPT1 coverage for Nigeria is estimated to 49% with 95% Confidence interval between 46% to 51% and that of DPT2 and DPT 3 is found to be 40% with 95% CI between 38 and 35 and 33% with 95% CI between 31% and 35% respectively. While looking at coverage by geopolitical region, it is evident that the lowest in North-West has the lowest coverage for DPT3, followed by North-East region. In North-East region, three states Gombe, Yobe and Taraba have coverage lower than national figure. Similarly, all states of North-West Zone fall below the national figure in all three doses of DPT3.

All three doses of DPT coverage have positive linear relationship with household economic status. Urban coverage is twice high as rural coverage. Hausa ethic group has the lowest coverage amongst the four groups. A total of 29% have HBR card (CI: 27-31) and rest is based on recall.

With increase in sample size, the CI will be narrower. This allows the survey result to be more precise.

As described in answer to earlier question, the result is only to derive for surveyed population, unweighted analysis will be valid. Once the survey result inferences to total population, weighed analysis is essential. In the survey, probability of selecting a child will vary from cluster to cluster as well as on number of completed questionnaire taking account complex sample design.

4. If a survey has “low” card availability, would you do the analyses that require dates (timelines, simultaneity of vaccines recommended at the same age, missed opportunities) weighted or unweighted? Why? Explain.

It is not advisable to calculate indicators which require date if a survey shows low card availability.

I will choose an unweighted analysis which means deriving respondent level inferences rather than estimate for the population (i.e., coverage in our case). If population level inferences are derived from small number of observation (cards), the coverage will be skewed more towards upper bound. In this case, we will be assuming that all the children 12-23 months have similar coverage as those of with proper vaccination card. In contrary, there is high chance that children with proper home-based record would get vaccinated than those who did not produce/retain a card (response based on recall) and the population level interpretation will be misleading (biased upward).

5. (a) With reference to table IM.14 describe key differences between crude and valid coverage. (b) How does card availability (and with dates) influence computation of valid coverage?

Crude coverage is proportion of children received vaccination (based on recall and card) for an antigen divided by 12-23 months children participated in the survey, Whereas, valid coverage takes into account of vaccination received timeframe and interval between two doses (i.e., 28 days between 2 doses).

A home-based record is essential to calculate a valid dose. This means a valid coverage can never exceed the percentage of card availability. As there will be no evidence for children without proper documentation, are excluded from calculation of valid coverage. However, the survey will not be able to state that the second group children did not get valid dose.

.