A weighted, multistage, stratified cluster sampling approach was used in the selection of the TDHS-2003 sample. The sample was designed in this fashion because of the need to provide estimates for a variety of characteristics for various domains. These domains, which are frequently employed in the tabulation of major indicators from the survey, are:
- Turkey as a whole;
- Urban and rural areas (each as a separate domain);
- Each of the conventional major five regions of the country, namely the West, South, Central, North, and East regions
- The 12 NUTS 13 regions, for selected indicators which are based on sufficient number of observations
The major objective of the TDHS-2003 sample design was to ensure that the survey would provide estimates with acceptable precision for these domains for most of the important demographic characteristics, such as fertility, infant and child mortality, and contraceptive prevalence, as well as for the health indicators.
Different criteria have been used to describe "urban" and "rural" settlements in Turkey. In the demographic surveys of the 1970s, a population size of 2,000 was used to differentiate between urban and rural settlements. In the 1980s, the cut-off point was increased to 10,000 and, in some surveys in the 1990s, to 20,000. A number of surveys used information on the administrative status of settlements in combination with population size for the purpose of differentiation. The urban frame of the TDHS-2003 consisted of a list of provincial centers, district centers, and other settlements with populations larger than 10,000, regardless of administrative status. The rural frame consisted of all district centers, sub-districts and villages not included in the urban frame. The urban-rural definitions of the TDHS-2003 are identical with those in the TDHS-1998.
Initial information on all settlements in Turkey was obtained from the 2000 General Population Census. The results of 2000 General Population Census provided a computerized list of all settlements (provincial and district centers, sub-districts and villages), their populations and the numbers of households.
Currently Turkey is divided administratively into 81 provinces. For purposes of selection in prior surveys in Turkey, these provinces have been grouped into five regions. This regional breakdown has been popularized as a powerful variable for understanding the demographic, social, cultural, and economic differences between different parts of the country. The five regions, West, South, Central, North, and East regions, include varying numbers of provinces.
In addition to the conventional five geographic regions, a new system of regional breakdown was adopted in late 2002. In accordance with the accession process of Turkey to the European Union, the State Planning Office and the State Institute of Statistics constructed three levels of NUTS regions, which have since become official (Law No. 2002/4720). "NUTS" stands for "The Nomenclature of Territorial Units for Statistics". NUTS is a statistical region classification that is used by member countries of European Union (EU). The 81 provinces were designated as regions of NUTS 3 level; these were further aggregated into 26 regions to form the NUTS 2 regions. NUTS 1 regions were formed by aggregating NUTS 2 regions into 12 regions. Two of the NUTS 1 regions, Istanbul and the Southeastern Anatolia, were given special attention in the sample design process and a comparatively larger share of the total sample was allocated to these regions to ensure that statistically sound estimates for a larger number of indicators would be obtained than would be the case for the remaining 10 NUTS 1 regions. Policymakers, researchers and other concerned circles had voiced interest in information on demographic and health indicators for Istanbul and the Southeastern Anatolian regions in the past. Furthermore, as an add-on study, the Istanbul metropolitan area was designated by UN-Habitat as one of the mega-cities in their International Slum Survey series. In co-operation with UN-Habitat, HUIPS wished to be able to produce estimates for slum4 and non-slum areas within Istanbul; for this reason, the total sample size for Istanbul was kept at a relatively high magnitude.
One of the priorities of the TDHS-2003 was to produce a sample design that was methodologically and conceptually consistent with the designs of previous demographic surveys carried out by the Hacettepe Institute of Population Studies. In surveys prior to the TDHS-1993, the five-region breakdown of the country was used for stratification. In TDHS-1993, a more detailed stratification taking into account subregions was employed to obtain a better dispersion of the sample. The criteria for subdividing the five major regions into subregions were the infant mortality rates of each province, estimated from the 1990 Population Census using indirect techniques.5 Using the infant mortality estimates as well as geographic proximity, the provinces in each region were grouped into 14 subregions at the time of the TDHS-1993. The sub-regional division developed during the TDHS-1993 was used in TDHS-1998.
However, the new NUTS regions necessitated further steps for sample design, namely that the sample design of the TDHS-2003 would allow using the conventional five regions as well as the NUTS 1 regions as sample domains. The conventional five regions cannot be obtained by aggregating the 12 NUTS 1 regions. To ensure both regional breakdowns were served by the sample design, 20 mutually exclusive strata had to be created, which, when appropriately aggregated, would produce the five conventional regions or the NUTS 1 regions. It became clear during this exercise, however, that if slight modifications were made to the boundaries of the 5 regions a smaller number of strata would be sufficient for reflecting both breakdowns in the sample design. More specifically, changing the regions to which only 6 provinces out of 81 were included would make it possible to construct 15 strata and serve the same purpose. This exercise was undertaken; also, a series of statistical tests were carried out to make sure that the modification to the regional boundaries would not make any difference in terms of regional indicators.
As a result of these considerations and exercises, 40 separate strata were created for the sample design of the TDHS-2003. This included the designation of 15 "divisions" by urban and rural stratum, the two strata within Istanbul (slum and non-slum), and metropolitan cities as mutually exclusive strata. The stratification also makes possible to combine provinces, which were affected by the earthquake in 1999.
The target sample size of the TDHS-2003 was set at 13,160 households, some 30 percent larger than that of the TDHS-1998. This increase is mainly related with the designation of new strata, the special attention given to Istanbul and Southeast Anatolia region, and with the adjustment of optimum allocation among the NUTS 1 regions. The targeted number was allocated among the five major regions as similar as possible to the TDHS-1998 (Table B.2). However, since Istanbul and Southeast Anatolia regions are over-sampled, the number of observations is higher for West and East regions relative to the previous survey. It was also aimed to target not less than 740 households for each NUTS 1 region. Based on previous experience in sample surveys, the target number of 13,160 households was expected to yield about 11,000 completed household interviews. To have an adequate representation of clusters within each of the five major regions, it was decided to select 25 households per standard urban segment (under the assumption of each cluster consisting of 100 households) and 15 households per standard rural segment. One exception to this was the selection of 12 households from the two urban segments in Istanbul (slum and non-slum). It was also determined that any of the strata should consist of at least 4 clusters, in order to make easier the sampling error calculations
For the first-stage sample selection, settlements were grouped within each of the 40 strata, and a systematic random sample of settlements with probability proportional to size (PPS) based on the 2000 General Population Census was selected from the settlement lists. The output from this first stage of the selection was a list of all of the settlements included in the TDHS-2003 sample along with the number of clusters to be drawn from each settlement.
The first stage selection for the two strata of Istanbul metropolitan area was performed by using a more detailed settlement list due to the need for stratification of the city into slum-and non-slum strata. Quarters of Istanbul were classified as slum or non-slum using expert opinion, simply to create probabilistic stratification and taking care of selection probabilities. Similar to settlement selection in other strata in the sample, quarters were selected systematically from these two strata in the first stage. In Turkey, settlements are not divided into small area units with well-defined boundaries (e.g., census enumeration areas) that can be used for conducting surveys. For some settlements, however, household lists were available from the Structure Schedules that were prepared in 2000 by municipalities in collaboration with the State Institute of Statistics (SIS). Some of these lists were updated in 2002. For 563 clusters, SIS was able to provide household lists from the Structure Schedules. For those settlements, the household lists were subdivided into segments of approximately 100 households with the exception of the two Istanbul metropolitan strata, where the segments included approximately 50 households. The list of these segments constituted the frame for the selection of the 563 clusters. For each of the selected clusters, SIS provided a list of the dwellings units with their full addresses (quarter, area, avenue/street, building and door number).
SIS was not able to provide household lists from the Structure Schedules for settlements without municipalities from which 137 clusters were to be drawn for the TDHS-2003. For these settlements, the list of households had to be prepared in the field. In the case of small settlements (less than 250 households), the entire settlement was listed. In the case of the small number of settlements in which there were more than 250 households, 250 households were listed and an estimate of the remaining number of households in the settlement was obtained through a quick count.
LISTING AND MAPPING ACTIVITIES
Although the SIS had dwelling lists for many clusters, they did not have the corresponding maps. For this reason, the selected clusters had to be formed with streets that were not always adjacent to each other. Moreover, the lists provided by the SIS did not reflect changes that may have occurred during the period from the 2000 or 2002 to the survey date. Two types of changes were possible: those that could be updated during listing, such as the construction of a new building on the street, a change in the use of a building (e.g., a flat can be used as an office instead of a dwelling), or changes in the names of streets, and those that were more problematic, e.g., the appearance of new quarters in urban centers.
In an effort to develop strategies for dealing with these as well as other possible problems that might arise, a pilot listing activity was undertaken in the capital, Ankara, before the actual listing activity began. The final listing forms, sketch map formats, and listing and mapping manuals were developed based on this pilot activity.
Forty university students/graduates were trained for the main listing activity. Listing teams were formed following a three-day training program in the beginning of November 2003. Each team was provided with maps describing the location of the settlements they were expected to visit, as well as other materials needed for the listing. Sixteen listing teams were formed, each including one mapper and one lister. The listing operation started on 5th of November and it was carried under the supervision of the research assistants and regional coordinators from the Hacettepe Institute of Population Studies.
The cluster (standard segment) size was around 100 households (50 households for Istanbul metropolitan) for most of the clusters in urban areas. Only five urban clusters had extremely low numbers of households; in order to obtain 100 households in these clusters, adjacent streets were added to the original cluster. In some of the selected villages, the total populations were too small, and the original cluster did not include 100 households. In these cases, the village that was nearest to the selected village was included in the sample, and the names of these villages were provided to the listing teams; the lists of 100 households were completed from the two neighboring villages.
The listing operation was implemented in three stages due to seasonal conditions and completed in April 2004. Overall, the quality of the listing work produced by the listers was good although it varied somewhat largely in response to problems the listing teams experienced in working in some geographic areas.