Sampling Procedure
The 1987 National Epidemiology and Family Health Survey (EFHS) employed a multistage probability sample of over 11,000 households. Two departments were excluded from the sample, Gracias a Dios and the Bay Islands, due to difficult access and their small and scattered population. Together they make up only 1.5 percent of the total population of Honduras.
To carry out such a survey, adequate maps must be available to construct small area units to serve as sampling units in the later stages of selection. However, for the 1987 survey, current maps were not readily available for all parts of the country. At the time of sample preparation, the Honduran Bureau of Census and Statistics was updating maps for the 1988 Census and had updated maps of municipios (counties) representing a little less than half of the country's population. Included in the updated maps were those of 16 major cities. These cities had been selected for their importance in terms of employment by the Bureau of Census and Statistics which in September of 1986 carried out the Encuesta de Hogares (Labor Force Survey) to determine employment patterns of the labor force. Thus for the 1987 survey, two groups of recently updated maps were available: urban maps of 16 cities divided into colonias (neighborhoods) which were divided into sectors for the Encuesta de Hogares, and census maps of rural municipios divided into aldeas (villages) which Ministry of Health and Management Sciences for Health staff subsequently divided into sectors.
In the remainder of the country where Census personnel had not updated maps, the 1974 Census sector maps were used for the first stage of selection. Once these area units were selected, teams composed of survey staff and cartographers from the Vector Control Division of the Ministry of Public Health visited the sites and updated the maps.
The partial availability of current maps led to the designation of the following four cells (see Table II B1 of the Final Report) which served as the first level of stratification in the primary sample.
This four cell partitioning of the primary sample is not equivalent to stratification using the official definition of urbanization since Cells 2, 3 and 4 in Table II B1 contained both urban and rural areas. However, it is important to note that this feature of the design did not compromise our ability to produce estimates by official urban or rural designations.
Once the interview was completed and questionnaires were coded for data processing, all sectors were classified according to strata: 1) urban Tegucigalpa and urban San Pedro Sula, 2) other urban areas, and 3) rural areas. This classification of strata maintains the final strata designation used in the 1984 MCH/FP Survey. Urban was defined according to criteria used by the Bureau of Census and Statistics: population greater than 2000 inhabitants with public utilities, water and sewage. Since population estimates are based on 1974 census information, rural areas with less than but close to 2000 inhabitants were assumed to have grown and were reclassified as urban.
The 1986 population of Honduras based on estimates from the Latin American Center for Demography (CELADE) and the Secretary for Planning and Finances (SECPLAN) was about 4.2 million inhabitants. With about 5.5 persons per household, this corresponds to 763,636 households. Thus the overall sampling rate in selecting 11,660 households was 0.0153, or about one out of every 65 households. Our aim was to select a sample in which the sample and Census distributions by cell were approximately equal.
In January of 1987 the Ministry of Public Health implemented the National Nutrition Survey which used the sampling frame originally designed for the 1987 Epidemiology and Family Health Survey. Cell 1 of the Nutrition Survey was a subsample of the Encuesta de Hogares.
We selected 550 primary sampling units (PSUs) for the EFHS, 275 of which had been chosen earlier for the Nutrition Survey. However, new segments or secondary sampling units (SSUs) were selected for the EFHS in the previously used PSUs. An additional 275 PSUs were chosen independently from the four original stratified listings which correspond to the 4 design cells. To facilitate selection of the additional PSUs, we doubled the number of PSUs in each cell. To achieve the desired sample size for the EFHS, we increased the segment size used in the Nutrition Survey by 50 percent. The targeted number of households per segment varied.
Selection Protocol for each Cell
Cell 1
In preparation for the Encuesta de Hogares, cartography staff made a rapid enumeration of blocks and households in the 16 selected cities. Blocks were grouped in units that averaged 50 households. However, the units ranged from 25 to 125 households. These units were called PSUs and were listed and sorted hierarchically by health region, city, socioeconomic status, and population size or geographic proximity. Since probability proportional to size (PPS) systematic sampling was used ultimately to choose the PSU sample for each design cell, ordering according to the above variables implicitly stratified the sample. In the Encuesta de Hogares, the measure of size for PPS selection was the number of HHs. PSUs were selected with PPS. For the EFHS and Nutrition Surveys, a subsample of the Encuesta de Hogares sample was chosen. PSUs for these surveys were selected at random using equal probability.
Once the appropriate number of PSUs was selected, the SSUs or segments were delineated to contain approximately 15 households in Tegucigalpa and San Pedro Sula and about 8 households in the other 14 cities. Survey staff chose one segment at random and identified it on the map. In the field, the supervisor counted the number of households in the segment and when there were more than the predetermined cluster size, she chose at random a house with which to begin. Following a clockwise direction, interviewers visited households until the appropriate number was attained. An interview was attempted with all eligible respondents (women aged 15 to 44) in each household selected.
Cell 2
The updated (as of October 24, 1986) rural and urban maps not included in Cell 1, were used to define Cell 2. Municipios were divided into villages which were designated as the PSUs. PSU selection was done by PPS systematic sampling where the selection probability for each PSU was proportional to its size (total number of households) and an interval of fixed length was applied to the frame after a random start. The list of PSUs was ordered according to health region, department, municipio and geographical proximity which means a serpentine route was followed so that any two consecutive PSUs on the list were neighbors geographically. Once chosen, the PSU was divided into segments of about 33 households and one segment was chosen at random. All households in that chosen segment were contacted.
Cells 3 and 4
These cells used 1974 Census sector maps at the first stage of selection. The PSUs for urban areas were ordered according to health region, municipio, city and SES status when possible or geographic proximity. In rural areas, lists were sorted by health region, department, municipio and geographic proximity. PSUs were chosen separately in each cell by PPS systematic sampling and these "census sector maps" were updated for the EFHS and Nutrition Survey. The updated sectors averaged 70 households but ranged from 25 to 130 households. As in Cell 2, one segment of about 33 households was chosen at random for each PSU.
Summary of the sample for the 1987 Epidemiology and Family Health Survey
General description: Two stage area sample of households with stratification in the first stage and area segments of 8-33 households as the ultimate sampling unit. Expected number of selected households: 11,660 Overall household sampling rate: About 1 in 65 households Expected number of responding households: 9,736*
Expected number of responding eligible women in selected households: 9,713* Expected number of eligible children in selected households: 7,334*
(* Based on sample attrition in the 1984 MCH/FP Survey.)