The 2003 Turkmenistan LSS is the second survey of its kind conducted in Turkmenistan by Institute of State Statistics and Information of Turkmenistan. The first LSS survey was conducted in 1998.
The main objective of the survey (TLSS-03) was to measure the level of living of the people of Turkmenistan with respect to various social and economic indicators and produce comparable statistics to the TLSS-98. The survey results formed an important database for building a system of monitoring of the living standards in the country.
The survey will focus on income level and expenditure pattern of households along with their social opportunity and access to public services. The survey will integrate the social and economic aspects of living standards and reveal the social strata that need more attention and protection from state. The survey will analyse the different factors affecting the living standards and will produce valuable information required in development planning and policy making.
A wide range of information collected from the survey was analysed to reveal the major socio-economic factors affecting the level of living. The basic survey approach and the questionnaire was designed to ensure the comparability of statistics with TLSS-98, so that data analysis can be made in cross-statistics as well as in time series.
Kind of Data
Sample survey data [ssd]
Unit of Analysis
The survey covered the following area of interest related to the living standards of the people of Turkmenistan.
o Demographic characteristics
o Housing Condition
o Income and Expenditure
o Public health
Producers and sponsors
Institute of State Statistics and Information
Government of Turkmenistan
Asian Development Bank
Like in 1998, the survey was designed as a two-stage stratified cluster sampling. The principle of stratification into urban and rural for each 5 regions (Velayats) also remains unchanged. It created 11 independent strata (10 from 5 regions plus one stratum of Ashgabad). Primary sampling units (psu) were clusters formed of enumeration area units as described above. Households were listed in the selected clusters and sub-sampled by field staffs from the listing sheets.
TLSS-03 had a self-weighting design and samples were spread out over the wide area of the country. For this purpose, psu's were arranged in the order of geographical location across the different Etraps. Selection of PSU's was made systematically probability proportional to the number of households in clusters.
A fixed sample of 20 households was selected from each cluster using simple random sampling method. Selection of psu's by pps method at first stage and inversely proportional to the number of households at second stage resulted in a self-weighting sample, which was very important for this survey, especially because a large number of indicators are means and proportions. In a self-weighting design, sample means and sample proportions are unbiased estimators of population means and population proportions.
See detail sampling information in "Turkmenistan Living Standards Survey 2003 Technical Report" document.
Dates of Data Collection
Data Collection Mode
Data Collection Notes
Data collection of TLSS was carried out in December 2003 by administering two types of questionnaires namely: the community and the household questionnaire. Data for the community questionnaire was collected at the level of settlements and the information was provided by the local heads. The questionnaire implemented in TLSS was developed on the basis of the standard LSMS questionnaire recommended by the World Bank and the questionnaire implemented in TLSS-98. The questionnaire was pretested in a small pilot study covering 60 households.
The survey was conducted by a team of 3 central supervisors 12 field supervisors and 109 interviewers. Each of 6 Velayats had one supervisor from the TMH central office and one from the regional office. The ratio between the field supervisors and enumerators was 1:9 against the initially recommended ratio of 1:4. Failure of recruiting required number of supervisors due to the manpower constraints of TMH resulted in the heavy workload of supervisors and thereby insufficient control of errors inthe field. This weakness was somehow compensated through prolonged data cleaning in the office, but still contributed to a higher margin of the non-sampling errors.
The six supervisors of the central office participated in thorough discussions of the questionnaire and survey plan. These supervisors trained other supervisors and enumerators recruited from the regional offices. The main weakness of the training was that the supervisors were trained separately in the regional offices, so they did not have sufficient interaction among themselves on various concepts used in the survey.
The survey was collected using two type of questionnaires:
- Household Questionnaire
- Community Questionnaire
Prior to the data entry, questionnaires filled and returned from the field were checked and edited especially with regard to household identification numbers and data items. The questionnaire included, household listing form, household questionnaire and the community questionnaire. To facilitate the smooth data entry, the community questionnaires were folioed by Oblast, while the household questionnaires were folioed by the survey block. Each folio was provided with appropriate folio cover, which included the household identification and indicators to determine the status of every folio during machine processing. The total folios produced were as follows.
- Community Questionnaire, 6 folios
- Household Questionnare, 120 folios
The data entry programme was developed in CS Pro 2.3. The screen format for data entry was designed to make its look as similar as possible to the questionnaire. The form labels were made in both English and Russian versions. The programme also included the necessary control mechanism to ensure validity of entries. As mentioned above, there were two levels of questionnaires, so programme files were developed separately for community and household questionnaires.
Several department of TMH housed the data entry process. However, it was not felt necessary to install a network due to the relatively smaller size of the data load. An additional computer was designated for batch editing, form receipts and control and the monitoring purposes. The data entry was conducted from 4 January to 7 February 2004.
CSPro 2.3 was also used for editing. A batch edit program was developed to control the quality of data. Range checks were done on every data item. Additional consistency checks between data items were included in the edit programme. The program generated a list of errors for all questionnaires belonging to a particular household. The data items with error were manually compared with the corresponding questionnaire for verification. All necessary corrections were recorded in the error list and were later used for data correction. Since this is a sample based survey, automatic imputations were not done to preserve reliability of data.
Data files were merged and converted to SPSS format for statistical analysis. Frequency tables and cross tabulations were done to further check the quality of data. Based on the pre-designed output table format, more than 500 tables were generated for initial data analysis. Main part of the tabulation process was computing the incomes and expenditures for each household. It required creating additional variables such as to total and per-capita income. Estimation weights were incorporated with the relevant SPSS data files for raising the sample data to total.
Computation of sampling errors and design effects of major variables was done in WesVar 32. For this purpose, SPSS data files with the major variables of interest were converted back to ASCII data files and imported to WesVar. Minor adjustment of analytical tables and graphs were done in MS Excel.
Estimates of Sampling Error
Estimation of the standard error was made based on the Balanced Repeated Replicates (BRR method). The method required exactly two psu’s per stratum. It takes half sample from each stratum and as many complements. The squared differences of two estimates provide an unbiased estimate of variance.
See detail estimation of the standard error and design effect information in "Turkmenistan Living Standards Survey 2003 Technical Report" document.
Limitations of the survey
Although, the utmost attention was paid to ensure the quality of survey results, TLSS had some limitations. Users are strongly recommended to take these limitations into considerations while using the data of this survey. The limitations of the survey are broadly described below.
The survey frame
1. The main limitation of the survey was the quality of the frame used in the survey design. The last population census in Turkmenistan was conducted in 1995. Since then, a lot of demographic changes were observed mainly due the emigration of the Russian speaking population and internal replacement caused by massive housing reconstruction. Despite of all possible attempts directed to improve the frame, it must be recognised that the baseline data still came from the last census.
2. While the last population census results are no more a valid database for any kind of plausible statistical investigations, it is unfortunate that the upcoming Population census in 2005 has now been cancelled, which will be replaced by a “Mini-census of 5%”. Such census may produce the population figures, however, it will not provide so acutely required data for household surveys. Therefore, the problem of the frame is most likely to affect adversely also the quality of other household surveys to be conducted in future.
3. The problem of the frame is related also to the lack of maps of enumeration blocks used in the survey. The size of the earlier blocks in terms of the number of households has significantly changed, so new boundaries were fixed for this survey. However, there was no map available to show the recent changes. Field staffs prepared a new map by themselves for the selected blocks based on the list of households. However, the quality of such map could affect the accuracy of the size of blocks due to the omission or duplication that could occur in the absence of good map. In the absence of the decennial census, maps throughout the country are not updated in terms of the boundaries of enumeration blocks and the number of households. Again, it could also create difficulties in conducting other surveys in future.
Training and the fieldwork
4. During the data editing and consistency checking, several mistakes of field staffs were found in filling the questionnaire. These mistakes actually were the result of insufficient training of the field staffs. The supervisor’s training in the centre was limited only to those from TMH. Field staffs recruited from the centre and from the regional offices did not get the sufficient time of interaction on the various conceptual issues of the questionnaire, so could not sufficiently address much of the expected problems of the survey.
5. The effect of the poor training could have been minimised by an intensive and close supervision of the survey staffs. However, the number of supervisors deployed in the field was often below the initially planned number due to the constraints of time and manpower. There was no coordinated supervision of the fieldwork because the core survey staffs themselves were involved in data collection.
Total survey error
6. Although, sampling error of major variables of interest were at the accepted level, non-sampling errors of the survey were relatively high due to the poor quality of the frame, lack of sufficient training of the field staffs and weak supervision of data collection. Non-sampling error was also caused by measurement and non-response problem as mentioned in the earlier chapter. Therefore, the total margin of error of major estimates was higher, often substantially, than the estimated value of sampling error.
Profile of the living standard
7. The analysis of the living standards requires a statistically viable baseline that allows the results of the survey for comparison over time and territory. In international practice, such baseline is the subsistence minimum, which serves as an objective criterion of measuring the level of living of population. In Turkmenistan, the subsistence minimum is not used for living standard analysis and data on the consumer’s basket is confined to the office use. Therefore, some alternative approaches were used, which were subjective and, to some extent, biased.
8. There is a measurement problem of the actual value of the local currency. The main suppliers of goods and services, the state and the market, use the different exchange rates. The state provides basic food items such as bread and selected clothing items and utilities such as electricity, gas and water at a highly subsidised rate, while other items of consumption are exchanged at the current market prices. The market value of the Turkmen Manat is 4 times lower than the official exchange rate and the official exchange rate is 4 times lower than the official PPP rate. The exchange of goods and services provided by the state corresponds to the official rate and those purchased in the market – to the market rate. The difference between exchange rates is so high that the international comparison of value indicators becomes extremely difficult.
These limitations however, do not undermine the significance of statistics produced by the survey. In general, survey results are consistent and reliable for important socio-economic analysis. The weaknesses of the current survey are brought up to this report so that these problems are taken into due consideration while planning the similar statistical inquiries in future.