Vietnam Household Living Standard Surveys (VHLSS)
Data for this study are drawn from the Vietnam Household Living Standard Surveys (VHLSS). These nationally representative surveys have been conducted every two years since 2000 by the General Statistics Office (GSO) of Vietnam, with technical assistance from the World Bank. These surveys consist of two parts: a household survey and a commune survey. The household survey collects very detailed infor- mation on households, such as demography, education, employment and labor force participation, income, expenditure, health, housing, durable goods, fixed assets and participation in poverty programs. The commune survey collects basic information on demography, socioeconomic characteristics, and the infrastructure of communes. They are cross-sectional data but it is possible to build a panel dataset due to the overlap of samples.
Procedure for collecting data
In these surveys, a method of stratified random cluster sampling is applied to ensure the household samples are representative for national, rural and urban, and regional levels. The sample of households in the series of VHLSS from 2002 to 2006 is selected from a master sample which was randomly chosen from the enumeration areas (EA) of the 1999 Population Census. The sampling procedure can be briefly described in three stages. First, 2300 rural communes and 700 urban wards were selected as the
primary sampling units (PSU). Second, three EAs of the 1999 Population Census were selected from each PSUs. Finally, in each EA, 20 households in rural EAs and 10 households in urban EAs were selected.
In the VHLSS series, the sample is revolved from year to year. For two consecutive surveys, 50 per cent of the households selected from the EAs in a half of the PSUs from the previous survey are re-surveyed in the next survey. The other 50 per cent of households are chosen from new EAs in the remaining half of the PSUs. The 50 per cent overlap between the two surveys allows for a household panel.
In this study, we first use three surveys from 2002 to 2006 as cross-sectional data. Then we build two separate panels 2002-2004 and 2004-2006. We also construct a panel data set including information from all three surveys from 2002 to 2006. The household sample in the 2000 and 2010 surveys came from other master samples; therefore, we cannot use them to construct a panel data set.
Panel data with VHLSS
The survey of VHLSS 2002 was conducted between May and November of 2002. In the final release of this survey, 29,530 households were surveyed for both income and expenditure. Similar to VHLSS 2002, the survey of VHLSS 2004 was imple- mented between May and November of 2004. Both income and expenditure of 9,189 households were collected. There were two additional new modules in the 2004 questionnaires but this does not affect the core modules.
In VHLSS 2006, two expanded modules in the 2004 questionnaires were dropped and two other commune sections were added in order to collect information on schools and healthcare services. The number of households in the final release remains at 9,189.
Due to the inconsistency in household identification across surveys, a panel of only 3931 households is constructed between VHLSS 2002 and VHLSS 2004. Similarly, a panel of 4,193 households is built across the VHLSS 2004 and VHLSS 2006. Ulti- mately, three waves of VHLSS from 2002 to 2006 allow for a panel of 1844 households. The data structure of VHLSS 2002, 2004 and 2006 is presented in Table A.1 and the statistical summary of variables is presented in Table A.2, A.3 and A.4 of Appendix A.
As can be seen in Table A.2, A.3 and A.4, during 2002-2006, household consumption increased since the log of expenditure slightly increases. The age of head continuously increases while the female share fluctuates and the dependent share decreases. The mean of the samples shows that levels of education attainment fluctuates over three surveys, except that the mean of the technical school category has consecutively risen. It is important to notice that landholding rises during 2002-2004 but falls during 2004-2006. This reflects changes in economic policies and a shift in the role of land in household living. Households in the samples tend to leave rural areas, as evidenced by the gradual increase in the value of the urban variable across surveys. Last, but not least, household access to electricity and market improves considerably.
Data issues
The use of the household panel data might raises concerns about the sample attri- tion. When households with certain characteristics leave the panel, the panel data are incomplete and biases emerge. If the attrition rate of vulnerable households and non-vulnerable households is considerately different, attrition becomes particularly problematic. Either households who leave the survey may be among the most vul- nerable, or they may be less vulnerable as migration can be an effective way to cope with risks. Without follow-up surveys, it is impractical to identify the exact nature of biases. However, in the case of Vietnam, several previous studies using VLSS and VHLSS have proven that the attrition rate is rather low and random (e.g. Baulch & Masset (2003) and G¨unther & Harttgen (2009) for VLSS; Roelen (2010) for VHLSS). Another potential problem with VHLSS is the changing household composition and size (Kamanou & Morduch 2002). The changes are generally less sharp for the non- poor. The changes are partly attributed to family splits and to migration, some of which was motivated by the economic forces. Births and the arrival of relatives and others both contribute to change in household composition. When household size changes, then per capita income or consumption may vary. However, this is not considered a shock, as it is the result of a deliberate household choice. This problem is also addressed in Pincus & Sender (2008). Thus there is a caveat on interpreting our result and there might be some resultant underestimation of vulnerability in our analysis.