Survey ID Number
AGO_1996_MICS_v01_M
Title
Multiple Indicator Cluster Survey 1996
Sampling Procedure
Interviews were carried out in 4,337 households during the fieldwork stage, which lasted from August-December 1996.
The sampling plan for MICS was intended to obtain a multiple purpose sample to be applied in 6 extended regions, defined as the research areas, on the basis of, on the one hand a UNICEF interventionist plan in Angola, and on the other hand, taking into consideration their geographical features.
From this sample it was possible to obtain estimates at the national level and at the level of the six geographically defined regions. Estimates at a more disaggregated level were not advisable, otherwise running the risk of losing the representative nature of the results.
Due to the war situation the country was facing factors, such as a displaced population and difficult access to certain localities. The last population census dates from 1983-84. Apart from the fact that these data are out of date, they only refer to a part of the country. For this reason, all the information available was used, and from various sources, to construct the Sampling Frame in order to select Primary Sampling Units (P.S.U.). Sources such as the Electoral Register/Census of 1992, information from the Ministry of Territorial Administration (MAT) the Provincial Governments and the social and economic provincial profiles prepared for the donors' round table in Brussels in 1995 for UNDP were used.
With the exception of the first stratum of the region "Capital C" (constituted by the Province of Luanda) the sampling of selected families for the research was probabilistic with 3 selection stages. In each region a potentially self-weighted sample of households was selected, though this characteristic self-weighting could be lost due to various factors, especially variations in population estimates.
The unit used in the first stage (PSU) was the "comuna" (the smallest administrative area in Angola) whose selection within each region was made independently, systematically and with probability proportional to the estimated size of the population.
The village in the rural areas or the neighbourhood in the urban areas constitutes the unit used in the second stage (SSU) and its selection was made without replacement and from a list of villages, which were accessible and based on information collected by regional co-ordinators. Thus, the selection was generally proportional to the number of inhabitants in the villages. In some cases (absence of population information) the treatment was:
1. When no information was available concerning the people of the village, the selection was made on a simple random basis (enumeration method);
2. When a list of villages did not exist or any information concerning its or their population (inhabitants) selection was made randomly from a point on the map, after it had been divided into 20 parts. (Map method)
Finally, the family constituted the Third Stage Unit (TSU) and its selection was without replacement and with equal probability within each selected village. The method used by PAV (Extended Vaccination Programme) was applied to barrios or neighbourhoods outside Luanda. This method consists in spinning a bottle to select a random direction. Following this the first family surveyed is randomly selected in this direction. The other families are those closest to the first.
In the case of Luanda, the sample was probabilistic with two selection stages. The unit used in the first stage was the census section in the Demographic Census of 1983/84, updated when the Priority Survey on Household Living Conditions was carried out in 1995 (IPCVD) by INE. The selection of primary sampling units was made independently and systematically with probability proportional to the number of dwellings. The secondary sampling unit was the family, whose selection was without replacement and with equal probability within each selection made. The selection of families in Luanda was made using a complete list of families taken from the selected census section.
The final probability of selection for each household is obtained from the product of the probabilities at each selection stage. The analysis of the weighted results was used to facilitate national and regional estimates and in order to correct the information used in the selection of PSUs and SSUs in the next selection stage.
The sample size was defined with a level of confidence of 95% to estimate the proportion of variable keys for the research based on information available to UNICEF. The level of precision was 5%, with the exception of some variables linked to breast-feeding in which more limited age-groups were used. In these cases the level of precision used was 8%.
The estimation of the necessary sample size was made separately for each of these key variables. Quite different sizes were obtained for the sample from each of the variables, having in the end to opt for the largest size. This confers a higher level of accuracy on the other variables than that originally expected. Or, that is to say, estimates can be obtained from the survey data with a maximum error of plus or minus 5%, with the exception of those variables related to breast-feeding where the maximum error was plus or minus 8%.
The "Design Effect" (Deff) is a factor used to adjust the variance obtained from a complex sampling design using clusters with the variance of a simple random sample.
In the definition of the sample, size 2 was assumed as the lowest value and 10 as the highest, using the highest value only in the case of Water and Sanitation.
In the analysis of data the confidence intervals of 95% were calculated for the main indicators using Program Epi Info 6, which calculates the value of DEFF directly from the data,
.
The sample size was fixed at 4,410 families distributed equally among the six regions, resulting in a sample of 735 families, 21 primary units (PSUs) and 21 secondary units (SSUs) for each of the six regions. In this way in each secondary unit selected, 35 families were chosen.
In summary, the size of the national sample was defined in: (21 clusters per region) X (35 families per cluster) X (6 regions) = 4,410 families.