Investigating the determinants of household welfare in the Central Highland, Vietnam

To provide policy implication for improving household welfare, one should understand which characteristics of households in a specific location they live, enable them to raise their welfare levels. This paper uses micro-econometric models for investigating the determinants of household welfare in the Central Highland, Vietnam using the recent data from the Vietnam Household Living Standard Survey in 2016. Despite the relative simplicity, rich information is obtained from its use on crosssectional survey data. Both descriptive statistics and regression analysis were employed in the study. Ordinary Least Square (OLS) and a Logit estimators were used to examine factors affecting household income and poverty incidence, respectively. Results specific to the region include: a substantial contribution of nonfarm selfemployment and education to household income and poverty eradication; wage employment is positively associated with poverty alleviation but not per capita income; and only some types of land were positively related to income and poverty reduction. The findings suggest that policies for poverty reduction should aim at improving the access of the poor to education and nonfarm employment. Subjects: Development Studies; Regional Development; Rural Development; Economics and Development; Economics; Political Economy

ABOUT THE AUTHORS Dr Hung Manh Nguyen is a senior lecturer at the Ho Chi Minh National Academy of Politics. He teaches several courses for Master and PhD Students, including political economy and international relationship.
Dr Tuan Anh Nguyen is a senior lecturer in financial economics at the University of Economics and Business, Vietnam National University, Hanoi. He teaches several courses at both under and graduate levels, including microeconomics and micro-economics of finance and banking.

PUBLIC INTEREST STATEMENT
This paper uses micro-econometric models for investigating the determinants of household welfare in the Central Highland, Vietnam using the recent data from the Vietnam Household Living Standard Survey in 2016. Despite the relative simplicity, rich information is obtained from its use on cross-sectional survey data. Results specific to the region include: a substantial contribution of nonfarm self-employment; the ethnicity and education of household heads to household income and poverty reduction; the gender of household heads and wage employment strongly affect poverty but not per capita income; and only some types of land were positively related to income and poverty reduction. The findings suggest that policies for poverty reduction should aim at improving the access of ethnic minorities to education and nonfarm employment.

Introduction
Vietnam has attained great achievements poverty reduction over the past two decades. The proportion of population living below the poverty line reduced significantly from 55% in 1993 (Haughton, Haughton, & Phong, 2001) to 18.1% in 200418.1% in and 5.8% in 201618.1% in (GSO, 2018. Despite prominent progress in the overall poverty alleviation, including a steady reduction in ethnic minority poverty, there exists a large and increasing gap in living standards and poverty rates across geographical regions within the country. In 2016, there was still about 13.8% and 9.1% of the population living below the poverty line in the Northwest and Central Highland regions, respectively. By contrast, the corresponding figures for the Red river and Mekong river deltas were only about 2.4% and 5.2% (GSO. 2018).
One of the main objectives of economic development is the alleviation of poverty and more generally, the improvement of household welfare. To obtain this end it is vital to understand what factors affect household welfare. The results of such an investigation would serve as a guide to policy implications in developing countries (Glewwe, 1991). An increasing number of studies have examined factors affecting income or poverty among rural households (Nguyen & Tran, 2013, 2018Nguyen, Van den Berg, & Lensink, 2011;Viet, 2008); peri-urban or urban households (Doan & Tran, 2015;Tran, Lim, Cameron, & Van, 2014) or ethnic minorities (Tran, 2015b;Tran, Nguyen, Vu, & Nguyen, 2015) in Vietnam.
The literature in Vietnam shows that while many studies have focused on rural areas or some geographical regions of Vietnam, to the best of our knowledge, limited evidence exits on the factors determining income and poverty among households in the Central Highland region. The Central Highland region of Vietnam has faced some specific challenges in terms of the overall welfare of the population and in particular among ethnic minorities groups. The region is one of the two poorest and predominantly ethnic minority regions in Vietnam (World Bank [WB], 2013). Thus, a better understanding of what factors influencing household welfare in this poorest region is much of importance, when designing or implementing policy interventions to improve their economic welfare. The current study was conducted to fill in this gap in the literature.
The main purpose of this study was to investigate the socio-economic determinants of household welfare among households in the Central Highland, Vietnam. This is the first study to analyze factors affecting both (i) household income; and (ii) poverty status using a microdata set from a recent survey of Vietnam Household Living Standard Survey in 2016. Therefore, the study contributed to the extant literature by providing the first econometric evidence for factors affecting household welfare in one of the poorest regions of Vietnam. We find a substantial contribution of nonfarm self-employment and education to household income and poverty eradication; wage employment is positively associated with poverty alleviation but not per capita income; and only some types of land were positively related to income and poverty reduction. The findings suggest that policies for poverty reduction should aim at improving the access of the poor to education and nonfarm employment. The findings suggest that policies for poverty reduction should aim at improving the access of the poor to education and nonfarm employment.
The paper is structured into five sections. The next section provides a brief literature review on determinants of household welfare. The third section describes data source and econometric models used in this study. The fourth section presents the determinants of household welfare while the conclusion and policy implications are given in the final section.

Data source
The secondary data from the Vietnam Household Living Standard Survey (VHLSS) 2016 were utilized for the current study. The 2016 VHLSS was conducted by General Statistical Office of Vietnam (GSO). The VHLSS was conducted by the General Statistics Office of Vietnam (GSO) with technical assistance from the World Bank. The 2016 survey covers around 46,000 households in 10,339 communes/wards. The sample of the VHLSS is selected in the way to represent the entire country at national, regional, urban, rural and provincial levels.
Data on households and individuals consist of basic demography, employment, education, health, income, housing, fixed assets and durable goods, and various types of land. In this study, we used data for the Central Highland region, including about 3,229 households that were surveyed in five provinces, namely Kon Tum, Gia Lai, Dac Lac, Dac Nong and Lam Dong. Household income data are calculated from various sources (both cash and in kind), namely crops, forestry, animal husbandry, aquaculture, wage work, non-farm self-employment, pensions, rentals, interests, transfers, remittances and other sources. It should be noted that both incomes were measured accounting for own consumption of products produced by households.

Data analysis
The main statistical analyses applied in this study were descriptive statistics and regression analyses. First, households were classified as poor and non-poor households using the poverty line of 630 thousand dongs per capita per month for the rural area and 780 thousand dongs per capita per month for the urban area (GSO, 2018). Then, we compare the means of household characteristics and assets between the two groups using Analysis of Variance (ANOVA) models. In addition, a chi-square test was utilized to analyze whether a statistically significant relationship existed between two categorical variables such as the type of households (poor and non-poor households) and the types of employment.
Because household income per capita is a continuous variable, econometric analysis using ordinary least squares (OLS) regression models was used to analyze relationships between per capita household income and various explanatory variables (Model 1). Following several studies in Vietnam Nguyen et al., 2011;Tran, 2015b), a number of explanatory variables were selected as being important to household income in our regression analysis. These were (i) household size, dependency ratio, gender, age and education of household head; (ii) owned farmland size per capita; the log of total values of all fixed assets; (iii) participation in non-farm self-employment and wage employment activities; (iv) the presence of means of transportation, paved roads, post offices, electricity, local markets, and (v) the provincial dummy variables which control for provincial fixed effects. The definitions and measurements of included variables are given in Table 1.
We also examine factors affecting the likelihood of a household falling in poverty using a logit model (Model 2), where the dependent variable (poverty status) is a binary variable that has a value of one if a household was classified as poor and a value of zero otherwise. Model 2 uses the same explanatory variables as Model 1.

Model2 : Household income per capita
We addressed the heteroscedasticity by transforming income per capita and the size of land into their natural logarithms. In addition, the option "robust" in STATA was used to produce robust standard errors in both models. In order to identify possible indications of multicollinearity, a correlation matrix analysis and an analysis of VIF (variance inflation factor) have been performed. The results confirm that the models do not suffer from the multicollinearity problems. Table 2 shows that there are considerable differences in the mean values of almost household characteristics between poor and non-poor households. The King people (the major ethnicity) accounted for about 70% of the non-poor, whereas the corresponding for the poor was only 17%. This suggests that ethnicity was closely linked with poverty status. The non-poor had a smaller household size and lower dependency ratio than that of the poor. The differences in the age and education of heads between the two groups were highly statistically significant. On average, the heads of non-poor households were approximately 2 years older and had two more years of schooling than did those of non-poor households. Interestingly, the participation rate in wage employment was found to be slightly higher for the poor than that for the non-poor (70% vs 64%), while the participation rate in nonfarm self-employment was much higher for the non-poor than that for the poor (25% vs 5%). Also, the data in Table 2 indicate that the percentage of households who had at least one member being a member of communist party or farmer association or women's union was higher among the non-poor that are among their counterparts.

Background on household characteristics and income
Furthermore, Table 2 reveals that the poor earned a very low level of per capita income, which is just equivalent to one-fifth of that earned by the non-poor. The differences in all types of land and the total value of durable assets between the two groups are found to be highly statistically significant. The area of annual cropland and forestland owned by poor households was quite bigger than that owned by non-poor households. This might be explained by the fact that there were several policies that allocated forestland to the ethnic minority poor in this region . However, the non-poor had much more perennial cropland than did the poor. The non-poor also had a larger size of aquaculture land than that of the poor. The non-poor also had a total value of durable assets that was nearly five times as much as that of the poor. Substantial  Note: SD: standard deviations. *, **, *** mean statistically significant at 10%, 5% and 1%, respectively. Income and durable assets measured in 1,000 VND (Vietnam Dong). 1 USD was equal to about 21,000 VND in 2016.
differences in some household characteristics and assets between the two groups were expected to be closely associated with variations in household income and poverty status. Figure 1 shows that agriculture income contributed the largest share of total household income for the whole sample as well as for each of the income group. Combined together, the income from crop, livestock, forestry, and aquaculture accounted for nearly 44% of total income, followed by wage income, nonfarm self-employment income and other incomes. A closer look at the income structure of income groups indicates that the agricultural income share of the poor is larger than that of the non-poor, while the share of nonfarm self-employment income and other incomes is much greater for the non-poor than that of the poor. Interestingly, the wage income share is almost the same for both income groups. The data suggest that differences in income sources between the two groups might account for the differences in income per capita between them. Table 3 reports the results from the income model. The model explains roughly 50% of the variation in household income. In addition, many coefficients are highly statistically significant (p < 0.05) with their signs as expected. As shown in Table 3, both household size and dependency ratio are negatively associated with income per capita. Similar results were reported by Jansen, Pender, Damon, Wielemaker, and Schipper (2006) and Tran, Tran, Tran, and Nguyen (2018) who found that having more dependent members and more family members in general tends to reduce per capita income. For instance, an additional family member would be associated with a decrease in income per capita of about 9%. Not as expected, the age and gender of household head were found not to be linked with household income. However, we found that the ethnicity of household heads has a strong impact on household income. Specifically, the income per capita would be 33% higher for a household with the head being in the ethnic majority groups (Kinh/Hoa) than for a household with the head belonging to ethnic minorities. Interestingly, our study reveals that having a communist party membership is positively associated with per capita income but this is not the case for a membership of women's union or farmer association.

Determinants of household welfare
As expected, we find that education has a positive effect on household income per capita. An additional year of formal schooling year would increase per capita income by 5.8%. The finding supports previous evidence that the education of household heads plays an important role in improving household income in Vietnam's Northcentral region    holding all other variables constant, a household with nonfarm self-employment, on average, would have an income per capita level approximately 27% higher than that of those without nonfarm self-employment. The finding is consistent with several studies in Vietnam (Tran, 2015a(Tran, , 2015bTran et al., 2018) and other developing countries (Rigg, 2006). However, our study found no link between wage employment and household income, which does not support other studies showing that wage employment is positively associated with household income in the Northwest region (Tran, 2015b) and the Northcentral region of Vietnam . The difference might be explained by the fact that wage employments might bring lower returns compared to other economic activities in the study region. Also, the difference might result from the difference in the study context, time and coverage of the research sample.
Regarding the role of production land in household income, the study found that not all types of land are associated with household income. While both aquaculture and perennial croplands have a positive effect on household income, this effect was not found for the case of forestland and annual cropland. The finding is in line with that in the Northwest region  but contrast with that in Nguyen and Tran (2018)'s study which found that forestland has a significantly positive effect on household income in the Northcentral region. Notes: *, **, *** mean statistically significant at 10%, 5% and 1%, respectively. SE: robust standard errors We also used the standardized or beta coefficient to show which factors are the major contributor to household income. The Beta coefficients in Table 3 show that the association between education and income was 0.297 standard deviations which is higher than that for the rest of the factors in the model. This indicates that education is the biggest contributor to household income. Ethnicity and nonfarm self-employment are also major contributors to income, with the beta coefficient is 0.193 and 0.153 standard deviations, respectively. However, there are two other factors, household size and dependency ratio, with the corresponding beta coefficient is −0.185 and −0.163, suggesting that an increase in these factors would result in substantial decreases in household income. Table 4 reports the estimation results from the poverty model. The results indicate that many explanatory variables are statistically significant at the 10% or lower level, with their signs as expected. In addition, in addition, the Pseudo-R 2 = 0.254 and is highly significant, suggesting that this model has strong explanatory power (Louviere, Hensher, & Swait, 2000). Interestingly, a comparison of the results in Table 3 and those Table 4 reveals that some household-related factors determine both household income while some others affect only either per capita income or poverty status. This suggests that previous poverty studies using only a household income model or a poverty model might not adequately evaluate or even ignored important impacts of some factors on household welfare. For example, both household size and dependency ratio were found to reduce per capita income and increase the chance of falling in poverty, while the gender Notes: *, **, *** mean statistically significant at 10%, 5% and 1%, respectively. SE: robust standard errors of household heads affects poverty status but not per capita income. An additional member increases the odds of a household remaining in poverty by around 33%, holding all other things constant. The gender evidence in Table 4 indicates that the odd of being poor is about 40% lower for a male-led household than for a female-led household, holding the other variables in the model constant in the model.
The results 4 confirm that both the ethnicity and education of household heads have a strong reducing effect on poverty. The odd of a household being poor was about 73% lower for a household with the head belonging to the Kinh/Hoa ethnicity group than those with the head being in the ethnic minority population. For a one-year increase in the average years of formal schooling of household head, it is expected to see about a 11% decrease in the odds of a household being poor, holding all other factors constant. The finding is consistent with previous studies Tran et al., 2015) which showed that education and ethnicity were major factors affecting poverty in some poorest regions of Vietnam.
While the household income model showed that only nonfarm self-employment was positively associated with per capita income in Table 3, both wage and nonfarm self-employment were found to reduce the probability of households falling in poverty. For example, holding all else constant, the odds of being poor would be around 48% lower for a household taking up wage employment than another household without wage employment. A similar but stronger impact was also recorded for a household participating in nonfarm self-employment, with the corresponding odds are 74% lower than another household without nonfarm self-employment. These are consistent with the findings in Vietnam (Pham, Anh Tuan, & Thanh, 2010;Tran, 2015a) and other developing countries (Haggblade, Hazell, & Reardon, 2010;Rigg, 2006). However, our study found no relationship between participation in any formal group and poverty status. The result differs from previous findings which indicated that holding a membership of co-operation reduced the risk of falling into in Armenia (Bezemer & Lerman, 2004) or a household holding a membership of farmer association was less likely to be poor in the Northwest region, Vietnam .
Regarding the role of production land in poverty reduction, the results show that holding more perennial cropland would reduce the likelihood of being poor while a similar impact was not found for the case of annual, aquaculture and crop croplands. The finding is not in line with that in the Northwest region which households having more annual cropland, forestland and aquaculture land were less likely to be poor . The coefficients of provincial dummy variables are not statistically significant at the conventional level (10%), suggesting that the probability of households falling in poverty is not statistically different across provinces, even after controlling for other factors in the model.

Conclusion and policy implications
The objective of the current study was to examine the socio-economic determinants of household welfare among households in the Central Highland, Vietnam. Using an updated dataset from the 2016 VHLSS, this study provides the first evidence of factors affecting per capita and poverty status of households in the second poorest region of Vietnam. Interestingly, we find that some factors affect both household income and poverty, while some others affect either income or poverty status. We also used the standardized coefficients to identify which factors are major contributors to household income in the study region.
First, our micro-econometric analysis confirms that education and nonfarm self-employment were major contributors to household income and poverty reduction. In addition, while wage employment is found to have no relationship with household income, it is negatively linked with poverty incidence. Second, the current study finds that different types of land play different roles in household income and poverty status. While the income effect is positive for both perennial cropland and aquaculture land, this effect is found to be negative for annual cropland and not statistically significant for forestland. Poverty status is negatively related to perennial cropland but not associated with other types of land. Our research finding on the income and poverty effect of forestland is in line with that in the Northwest region but in contrast with that in the Northcentral region which found that forestland has a positive effect on income and poverty reduction. This difference may stem from differences in the quality of forestland or differences in the efficiency of forest management between geographic regions in Vietnam. This is an interesting topic offers for future research.
Our research findings provide some useful policy implications. While only perennial cropland is found to increase income and reduce poverty, land distribution policy should not be considered as the main approach to combat rural poverty because land is fixed in supply. Instead of this, improving the access of local households to nonfarm activities, either wage employment or nonfarm self-employment, should be viewed as a very practical policy for poverty eradication in the study area. The finding on the role of household size in income and poverty implies that reducing larger family sizes would help alleviate poverty in this region. International experience shows that family planning measures, among others, have been used as a powerful tool in reducing poverty in many developing countries (United Nations Population Fund, 2006). A policy implication here is that improving the National Target Program on Population and Family Planning is expected to be an effective way of combating poverty in the Central Highland region. Finally, the finding on the major role of education in improving household welfare suggests that the National Target Program on Education and Training should aim at ensuring sustained and improved access for the poor's children to education and training. This would help their next generation have more chance to take up lucrative jobs and improve their income.