Background and introduction
The Cambodia Socio-Economic Survey (CSES) has been conducted by the National Institute of Statistics (NIS) in 1993/94, 1996, 1997, 1999, 2004 and 2007-14. Since 2007 NIS conducts the CSES annually. In 2014 the CSES was conducted with a nationwide representative sample of 12,096 households. The CSES is a household survey covering many areas relating to poverty and living conditions. Questions are asked for the household and for the household members.
Poverty reduction is a major commitment by the Royal Government of Cambodia. Accurate statistical information about the living standards of the population and the extent of poverty is an essential instrument to assist the Government in diagnosing the problems, in designing effective policies for reducing poverty and in monitoring and evaluating the progress of poverty reduction. The Millennium Development Goals (MDG) has been adopted by the Royal Government of Cambodia and a National Strategic Development Plan (NSDP) has been developed. The MDGs are also incorporated into the “Rectangular Strategy of Cambodia”.
Cambodia is still a predominantly rural and agricultural society. The vast majority of the population get their subsistence in households as self-employed in agriculture. The level of living is determined by the household's command over labour and resources for own-production in terms of land and livestock for agricultural activities, equipments and tools for fishing, forestry and construction activities and income-earning activities in the informal and formal sector. The CSES aims to estimate household income and consumption/expenditure as well as a number of other household and individual characteristics.
The earlier CSES rounds have all made it possible to report sets of indicators on 8 main areas of social concern
• Demographic characteristics
• Health and Nutrition
• Labour Force
• Household Income
• Household Consumption
With the exception of the Vulnerability module which is new in 2014, all these areas were also covered by corresponding modules in earlier CSES years. To ensure comparability between years, the questionnaire is kept as consistent as possible over time. Questions are only updated or changed when there is a compelling reason to do so and if the loss to comparability over time is judged to be minimal. There are some changes though, mostly minor except for the questions on current economic activity which were updated in 2010.
Objective of the survey
The main objective of the survey is to collect statistical information about living conditions of the Cambodian population and the extent of poverty. The survey can be used for identifying problems and making decisions based on statistical data.
The main user is the Royal Government of Cambodia (RGC) as the survey supports monitoring the National Strategic Development Plan (NSDP) by different socio-economic indicators. Other users are university researchers, analysts, international organizations e.g. the World Bank and NGO’s. The National Accounts also uses the information from CSES in its calculations. The World Bank published a report on poverty profile and social indicators using CSES 2007 data and has published updated reports in subsequent years. In this regard, the CSES 2014 also continues to serve to all stakeholders involved as essential instruments in order to assist in diagnosing the problems and designing their most effective policies.
Survey planning and organisation
NIS formed a Project Staff in the Core Group in 2006 for managing the CSES’s which since then has been working every year. The project staff for the CSES 2014 consisted of 25-30 persons including technical staff taking different responsibilities in the running CSES. The project staff was responsible for all survey planning and activities and also engaged in establishing and carrying out monitoring schemes during the fieldwork, in arranging stakeholder meeting/workshop/seminar for questionnaire designing, data analysing, dissemination of the results, and for reporting to the Statistical Advisory Committee (SAC). The project staff was responsible for the allocation and utilisation of funds and in solving logistical problems during the course of the survey.
As the most important part of the organisation of the CSES 2014, 60 enumerators and 20 supervisors were recruited in late 2009 and were subject to training for the fieldwork. Some additional enumerators and supervisors were also trained to be able to replace those who resigned during the field work.
A list of NIS survey staff in the CSES 2014 is provided in the report.
Sampling design in CSES 2014
This text describes the sampling design and sample selection for CSES 2014. In general the decisions about the sampling design has been made with the following in mind: a) Comparability with annual CSES 2007-2013 and the large sample CSES 2004. b) Harmonization with other surveys in Cambodia with respect to the sampling frame of villages and the sampling domains (strata). c) The required accuracy of key estimates under the budget constraints given for the large sample CSES. The latter has been the key decision point which also has been greatly affected of the general survey design, in particular the temporary re-introduction of the diary questionnaire, the development and expansion of some modules and the exclusion and decrease of other modules in the household questionnaire.
The sampling frame of the CSES 2014.
The target populations of the CSES are:
All villages in Cambodia (for the village survey).
All normal households in Cambodia (for the household survey). Normal households are households that are not institutional households, homeless households, boat population households or households of transient population. (Institutional households are boarding houses, military barracks, prisons, student dormitories, etc.).
People living in normal households in Cambodia (for the household survey)
Subpopulations of the above
A sampling frame of villages was constructed by joining, processing and preparing three different data sources containing information about Cambodia’s villages, their location, if they are urban or rural and their sizes in number of households. Documentation about the construction of the sampling frame is found in the methodology note described in Holmberg (2013). The sampling frame of villages were compiled in late November 2013 and includes 14 340 villages. This is a significant update compared to the frames used between 2009 – 2013 who all were more or less based on the 2008 Population Census. By combining data from the Cambodian Intercensal Population Survey (CIPS) and Cambodian Agricultural Census (CAC), the regularly updated administrative information in the Commune Data Base (CDB) and official information from the Ministry of Interior, the frame covers all existing villages in Cambodia in the beginning of 2014. The villages constitute the Primary Sampling Units (PSUs) with a few exceptions of very large villages in Phnom Penh that are represented by more than one PSU.
The sampling frame also has auxiliary information about the villages. It contains variables which are used for the stratification such as the province of location and whether a village is urban or rural. It also includes a derived measure of size of the villages. This is used in the first stage sample using a systematic sampling selection scheme with probabilities proportional to this size measure. The size measure is based on the number of households per village retrieved from information sources in 2013.
The second stage of the CSES is the selection of Enumeration Areas (EAs) from the villages. The information about the EAs in previous CSES rounds has come from the Population Census. Because of the long time that has passed since the census and the poor experiences made from outdated EA information in CSES 2011 – 2013, it was decided not to include the EAs in the sampling frame of CSES 2014. Quality concerns and the lack of control in the actual mapping and selection was considered too high and for CSES 2014 new EA information will be collected from the field to ensure better quality.
Stratification and sample allocation
CSES 2014 allow for estimates on a geographical level below the national level. In the CSES 2009 the villages in the first stage were stratified by provinces crossed with a classification by urban or rural. This is changed in 2014 in order to better harmonize with the Cambodian Demographic and Health survey (CDHS). From the 24 provinces 19 groups are formed and each of these is divided into strata of urban and rural villages. In total this yield 38 strata, and independent samples are selected from each one.
The 19 provincial groups are the following:
Fourteen groups of provinces: Banteay Meanchey, Kampong Cham, Kampong Chhnang, Kampong Speu, Kampong Thom, Kandal, Kratie, Phnom Penh, Prey Veng, Pursat, Siemreap, Svay Rieng, Takeo and Otdar Meanchey.
Five groups with combined provinces: Battambang and Pailin, Kampot and Kep, Preah Sihanouk and Koh Kong, Preah Vihear and Stung Treng and Mondol Kiri and Ratana Kiri.
The allocation of the total sample between strata is done in two steps. First, the sample is allocated between urban and rural villages. A little above 20 % of the Cambodian households live in villages classified as urban. In small sample CSES-years, approximately 40 % of the total village and household sample has been allocated to urban villages. This is to ensure adequate accuracy of estimates related to economic activities and to make other estimates of the urban domain more accurate. In the CSES 2014 the total sample size of villages is bigger and therefore it is not necessary to have the same proportional overrepresentation of urban villages to achieve useful estimates of the urban population. However, some allocation of the sample towards the urban villages is desirable to increase the accuracy of estimates of economic activities. Out of a total sample of 1008 villages about 30 % (or 312) are urban and 696 are rural. This is approximately the same urban/rural allocation as 2009.
In the second allocation step these sample villages are allocated within the urban and rural proportionally among the 19 provincial groups. Furthermore, by the chosen sample selection scheme (described in the next section) there is an implicit stratification by location within the provinces, since before the systematic sampling scheme is applied the sampling frame is ordered by the code of the districts, communes and villages
The CSES 2014 is a three stage sampling design that besides from the stratification has the same general structure as previously conducted CSES.
Stage 1: From the village frame, Primary Sampling Units (PSUs) were defined independently in each stratum and for each one a systematic sample with probabilities proportional-to-size without replacement (a ps-sample) was selected. Both the frame and the sample allocation are summarized in table 1. The size measure used is a variable based on the number of households per village derived from various sources. In most cases it is the average between the villages size in the CDB and the available updates from the Census in the CIPS and CAC frames.
Stage 2: From each selected village/PSU a mapping of EAs is done (Village mapping) and from each PSU one EA is selected by simple random sampling. (In the Phnom Penh urban stratum some villages are so large that they have been split into several PSUs which in turn have been selected with certainty. In each of these, one EA is selected. This is equivalent of selecting more than one EA from these villages.)
Stage 3:In each of the selected EAs from stage 2, a mapping of all households is done. And in the third sampling stage 12 households are selected from each EA by systematic sampling.
The choice of 12 households per EA in the third stage is based on estimates from previous CSES years with the aim to find the best sampling scheme to estimate both poverty rates and economic activities, (Pettersson 2011). The total number of households in the CSES 2014 are 12 096 (3744 urban and 8352 rural).
For each household, all members are recorded in the household questionnaire and depending on demands in the different module of the questionnaire interviews are made and data recorded. With an average household size of 4.63 estimated from CSES 2012 it is expected that the total number of individuals in the CSES was a little more than 50000.
The data collection of the CSES 2014 was done throughout 2014. The annual sample was randomly split into 12 equal parts of 84 PSUs each. The twelve PSU PSU were then randomly been allocated a sampling month with the aim that each province strata should be represented in all months of the year. However, two major constraints applied. First, the rain season in Cambodia sometimes makes it difficult to reach rural villages in some provinces. Second, the fieldwork teams with one supervisor and two groups of two enumerators each have to have reasonable travel distances. In one month every team covered four PSUs and the geographical distances between these could be too far. Because of these two constraints, the distribution of PSUs over the months between the regions was in some cases been manually adjusted. Despite this, the monthly samples ought to be representative and large enough for some national estimates, and in some cases maybe even for urban, rural and Phnom Penh. This also enables quarterly estimates if they are desired.
Analysis from earlier CSES data on key variables such as various household expenditures, household income, employment, poverty and literacy (>7 yrs) reveal that the rate of homogeneity (roh) is in the range of 0.15 – 0.2 for expenditure variables, 0.05 – 0.13 for income variables and in the range 0.04 – 0.08 for labour force variables. With the chosen sampling design it means that the value of an approximate design effect (DEFF) will lie in the interval 1.4 – 3.2 probably slightly lower. Power calculations indicate that with this sample size the chance to detect a true change of 2 percentage points on the national level are relatively high (Pettersson (2011)).
Standard errors or confidence intervals are presented for some important estimates in an appendix in the CSES 2014 report. The standard errors have been calculated by the Taylor linearization method. The software used was SPSS, complex samples module. For more detailed information see the report on CSES 2014.