Farmland loss, nonfarm diversification and inequality among households in Hanoi s peri-urban areas, Vietnam

Transcription

1

2

3

4

5 IDPR, 36 (3) 2014 doi: /idpr Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong Farmland loss, nonfarm diversification and inequality among households in Hanoi s peri-urban areas, Vietnam Using a novel dataset from a 2010 household survey involving 477 households, this study provides the first econometric evidence for the impacts of farmland loss (due to urbanisation) on nonfarm diversification among households in Hanoi s peri-urban areas in Vietnam. The results from fractional logit and fractional multinomial logit models indicate that farmland loss has a negative effect on the share of farm income but a positive effect on the share of various nonfarm incomes, notably informal wage income. We also investigate the relationship between various income sources and income inequality using a Gini decomposition analysis. While income from informal wage work and farm work are inequalitydecreasing, other income sources are inequality-increasing. Thus, this suggests that farmland loss has indirect mixed effects on income inequality. Keywords: farmland loss, informal wage income, formal wage income, Gini decomposition, Vietnam International experience indicates that rapid urbanisation and economic growth coincide with the conversion of land from the agricultural sector to industry, infrastructure and residential uses (Ramankutty, Foley and Olejniczak, 2002). In developing countries, land beyond the urban fringe is in huge demand for various purposes, including the construction of public infrastructure, factories, commercial centres and housing. These demands for peri-urban land can bring about considerable changes in peri-urban livelihoods, for better or worse (Mattingly, 2009). According to Gregory and Mattingly (2009), urbanisation on the one hand leads to intense competition for land, deterioration and loss of access to natural resources, and these in turn have a detrimental effect on natural resource-based livelihoods. On the other hand, urbanisation offers a greate choice of jobs, better transport availabilty to markets, an expansion of services and trade, and the competitive advantage of proximity for fruit and vegetable products. These factors can help peri-urban households diversify their livelihoods and mitigate their dependence on natural resources (Gregory and Mattingly, 2009). Over the past two decades in Vietnam, a large area of farmland has been taken to provide space for urbanisation and industrialisation. As calculated by Le (2007), at Tran Quang Tuyen (corresponding author) is affiliated with Faculty of Political Economy, University of Economics and Business, Vietnam National University, Room 100, Building E4, 144 Xuan Thuy Road, Cau Giay District, Hanoi, Vietnam; Steven Lim, Michael P. Cameron and Vu Van Huong are affiliated with the Department of Economics, University of Waikato, Hamilton Campus, Gate 1, Knighton Road, Private Bag 3105, Hamilton 3240, New Zealand; qtt1@waikato.ac.nz; tuyentq@vnu.edu.vn; slim1@waikato.ac.nz; mcam@waikato.ac.nz; vhv1@waikato.ac.nz Paper submitted August 2013; revised paper accepted October 2013.

6 358 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong a national scale from 1990 to 2003, 697,417 ha of land were compulsorily acquired by the state for the construction of industrial zones, urban areas and infrastructure, and other national use purposes. Furthermore, in 2000 to 2007 it was estimated that approximately 500,000 ha of agricultural land were converted to nonfarm use, accounting for 5 per cent of the country s land (VietNamNet/TN, 2009). Increasing urban population and rapid economic growth, particularly in the urban areas of Vietnam s large cities, have resulted in a great demand for urban land. This has led to an intensive conversion of agricultural land into higher-value nonagricultural land, particularly within the urban fringe. In order to satisfy this demand for land in the northern key economic region, the state has conducted many farmland acquisitions in the Red River Delta, which has a large area of fertile agricultural land, a prime location and high population density (Hoang, 2008). 1 Such farmland acquisitions have major effects on poor households in Vietnam s rural and peri-urban areas (ADB, 2007). In the context of accelerating loss of farmland for urbanisation and industrialisation in the urban fringe of Vietnam s large cities, a number of studies have examined the impacts of farmland loss on households livelihood adaptation (Do, 2006; Le, 2007; Nguyen, Vu and Philippe, 2011; Nguyen, 2009). The studies indicate that, while farmland loss causes the loss of traditional agricultural livelihoods and food insecurity, it also expands the space for urbanisation and industrialisation, which in turn result in improvements in local infrastructure, new industrial zones and urban areas. Such changes offer a wide range of nonfarm livelihood opportunities for local people. As in Vietnam, negative impacts of farmland loss have been found in China (Deng et al., 2006) and India (Fazal, 2000; 2001). In contrast, other studies show positive effects of farmland loss on rural livelihoods in China (Parish, Zhe and Li, 1995; Chen, 1998) and Bangladesh (Toufique and Turton, 2002). In addition, varying results from farmland loss on peri-urban livelihoods have been reported in Ghana and India (Mattingly and Gregory, 2006). Although much has been discussed about the mixed effects of farmland loss on household livelihoods, to date no econometric evidence of these impacts exists. Thus, this study applies econometric methods to answer the key research question: how and to what extent has farmland loss affected household nonfarm diversification, as measured by household income shares by source? Another important contribution of this study is that we examine whether farmland loss has any impact on income inequality. Income sources have been found to be closely associated with income inequality in Vietnam (Adger, 1999; Cam and Akita, 2008; 1 Compulsory land acquisition is applied to cases in which land is acquired for national or public projects; for projects with 100 per cent contribution from foreign funds (including FDI (Foreign Direct Investment) and ODA (Official Development Assistance)); and for the implementation of projects with special economic investment such as building infrastructure for industrial and services zones, hi-tech parks, urban and residential areas and projects in the highest investment fund group (World Bank, 2011).

7 Farmland loss, nonfarm diversification and inequality among households in Hanoi 359 Gallup, 2002). If farmland loss has a major impact on household income sources, then it may cause changes in income inequality. Our study confirms this hypothesis: farmland loss has a significant impact on household income sources, particularly through nonfarm income diversification, and it also has indirect mixed impacts on inequality. Background of the case study Research site The research was carried out in Hoai Duc, a peri-urban district located on the northwest side of Hanoi, 19 km from the Central Business District. Of the districts of Hanoi, Hoai Duc holds the largest number of farmland-acquisition projects (Huu Hoa, 2011). Over the period 2006 to 2010, around 1,560 ha of farmland were compulsorily acquired by the state for 85 projects (Ha Noi Moi, 2010). The district covers an area of 8,247 ha of land, of which agricultural land accounts for 4,272 ha, and 91 per cent of this area is used by households and individuals (Hoai Duc District People s Committee, 2010). Hoai Duc has 20 administrative units, including 19 communes and one town. There are around 50,400 households with a population of 193,600 people living in the district. In the whole district, the share of agricultural employment dropped around 23 per cent over the past decade. However, a considerable share of employment has still remained in agriculture, making up around 40 per cent of the total employment in 2009 (Statistics Department of Hoai Duc District, 2010). Compensation for land-losing households According to our household survey, each household on average received a total compensation of VND 98,412,000. The minimum and maximum amounts were VND 4,000,000 and VND 326,000,000, respectively. 2 An adequate compensation for land loss was proposed as a possibility that might help households switch to an alternative livelihood in the peri-urban areas of Kumasi, Ghana (Mattingly, 2009). Unfortunately for Vietnamese households, there has been a large gap between the compensation level defined by the government guidelines, and the real value of the land determined by market principles (Han and Vu, 2008). Although the compensation has been well below the fair market value of the land, it would however have provided households with a significant amount of capital with which they can initiate a new income earning activity or invest more in existing activities. However, most households have used this valuable source for non-production purposes rather than production purposes. 3 2 USD 1 equated to about VND 18,000 in According to the surveyed data, about 60 per cent of land-losing households used the compensation for daily

8 360 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong This trend is also evident in other districts of Hanoi as described by Do (2006) and Nguyen (2009). Therefore, this suggests that compensation might have little impact on nonfarm diversification in our sample. Also, Ha Tay Province People s Committee issued the Decision 1098/2007/ QĐ-UB and Decision 371/2008/QĐ-UB, which state that a plot of commercial land (đất dịch vụ) will be granted to households which lose more than 30 per cent of their agricultural land. Each household receives an area of đất dịch vụ equivalent to 10 per cent of the area of farmland taken for each project (Hop Nhan, 2008). Đất dịch vụ is located close to industrial zones or residential land in urban areas (World Bank, 2009). Thus, it can be used as a business premises for nonfarm activities such as opening a shop or a workshop, or for renting to others. While this compensation policy with land for land has been successfully implemented in some provinces, this solution is believed to be unsuitable for other provinces due to insufficient land for this purpose (World Bank, 2009). Data and methods Data Adapted from the General Statistical Office (GSO) (2006), a household questionnaire was constructed to collect quantitative data on household characteristic and assets, income-earning activities (working time allocation) and household economic welfare (income and consumption expenditure). A disproportionate stratified sampling method was employed with two steps as follows. First, 12 communes that lost their farmland (due to the state s compulsory land acquisition) were divided into three groups based on their employment structure. The first group consisted of three agriculture-based communes; the second group was represented by five communes based on both agricultural and non-agricultural production; the third group included four non-agriculture-based communes. From each group, two communes were randomly chosen. Second, from each of these six communes, 80 households, including 40 households with farmland loss and 40 households without farmland loss, were randomly chosen for a target sample size of 480. The survey was implemented from April to June 2010; 477 households were successfully interviewed, of which 237 households lost some or all of their farmland. Of the 237 households with farmland loss, 113 households had farmland acquired in early 2009 and 124 households had farmland acquired in the first half of In the remainder of this paper, households whose farmland was lost partly or totally by the state s compulsory acquisition of land will be referred to as land-losing households. living expenses, and about a quarter of them purchased furniture and appliances, while a similar proportion of land-losing households spent this money in repairing or building houses. By contrast, only 4 per cent among them used this resource for investing in nonfarm production.

9 Farmland loss, nonfarm diversification and inequality among households in Hanoi 361 Methods Classification of livelihood strategies Partition cluster analysis was used to group households into distinct livelihood categories. Proportions of time allocated for different economic activities (before farmland acquisition) were used as variables for clustering past livelihood categories (the livelihood strategies that households pursued before farmland acquisition). Similarly, proportions of income by various sources were used as variables for clustering current livelihood categories (the livelihood strategies after farmland acquisition). The two-stage procedure suggested by Punj and Stewart (1983) was applied for cluster analysis, which identified various livelihood strategies that households pursued before and after farmland acquisition. Specification of econometric models Econometric methods were then to quantify the impact of farmland loss on household income shares by source. Because the share of farm income is a proportion, the determinants of farm income share were modeled using a fractional logit model (FLM), which was proposed by Papke and Wooldridge (1996). FLM has similarities with the standard logit model, with the difference that the response variable is a continuous variable bounded between zero and one instead of being a binomial variable. This model is estimated using a quasi-maximum likelihood procedure (Jonasson, 2011). As demonstrated by Wagner (2001), the fractional logit approach is the most appropriate approach because this model overcomes many difficulties related to other more commonly used estimators such as ordinary least squares (OLS) and TOBIT. To quantify factors affecting the share of nonfarm incomes, a set of simultaneous equations was estimated with the share of farm, informal wage, formal wage, nonfarm self-employment and other income as dependent variables. Because each of these dependent variables is a fraction and the shares from this set of dependent variables for each observation add up to one, a fractional multinomial logit model (FMLM), as proposed by Buis (2008), was employed. As Buis (2008) notes, the FMLM is a multivariate generalisation of the FLM developed by Papke and Wooldridge (1996) to deal with the case where the shares add up to one. Similar to the FLM, the FMLM is estimated by using a quasi-maximum likelihood method, which includes robust standard errors (Buis, 2008). There have been a growing number of studies applying the FMLM to handle models containing a set of fractional response variables with shares that add up to one (Kala, Kurukulasuriya and Mendelsohn, 2012; Winters et al., 2010). Following the framework for micro-policy analysis of rural livelihoods proposed by Ellis (2000), income shares by source were assumed to be determined by household livelihood assets (including natural, physical, human, financial and social capital). In addition, other factors, in this case past livelihood strategies, farmland loss and

10 362 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong commune dummy variables, were included as regressors in the models. Summary statistics for the included variables are available in Appendix 1. In the present study, the loss of farmland of households is an exogenous variable, resulting from the state s compulsory land acquisition. 4 The farmland acquisition by the state took place at different times; therefore, land-losing households were divided into two groups, namely (i) those that lost their farmland in 2008 and (ii) those lost their farmland in The reason for this division is that the length of time since farmland acquisition was expected to be highly associated with the changes in income sources. In addition, the level of farmland loss was quite different among households. Some lost little, while others lost all their land. As a consequence, the level of farmland loss, as measured by the proportion of farmland acquired by the state in 2008 and in 2009, was used as the variable of interest. In general, households with a higher level of land loss were hypothesised to have a lower share of farm income after land loss and, conversely, were expected to raise the proportion of nonfarm income sources. Household size and dependency ratio (calculated by the number of household members under 15 and over 59, divided by the total members aged 15 to 59) were included in the models as measures of human capital, along with the number of male working household members, gender and age of the household head, and average education of the working members of the household. In rural Vietnam, men are more likely than women to participate in non-agricultural wage work (Pham, Bui and Dao, 2010), so having more male working members was expected be associated with a higher wage income share. Households with better human capital, as measured by the average years of formal schooling of household working members, were expected to receive a higher percentage of formal wage income. Older working members tend to be more involved in farming as their main income-earning activity. Therefore, the age of household heads and of working members (those who worked in the last 12 months) was also expected to be positively linked with the share of farm income. Owning more farmland per adult (100 m 2 ) is indicative of households that specialise in farming and thus households with more farmland were hypothesised to have a greater share of farm income. Residential land can be used as collateral for credit. Therefore, households with a larger size of residential land were expected to have greater financial resources for productive activity. Consequently, a larger size of residential land was hypothesised to be associated with a higher share of farm and nonfarm self-employment income. Furthermore, a higher percentage of income from nonfarm self-employment was also expected for households owning a house or a plot of residential land in a prime location. 5 4 According to Wooldridge (2013), an exogenous event is often a change in the state s policy that affects the environment in which individuals and households operate. 5 A prime location is defined as: the location of house or the location of a plot of residential land situated on the main road of a village or at the crossroads or very close to local markets or to industrial zones, and to a highway

11 Farmland loss, nonfarm diversification and inequality among households in Hanoi 363 Households with a higher number of group memberships (a proxy for social capital) may benefit from access to information, technology and credit for production. Therefore, social capital was expected to be associated with income shares by source. Financial capital is represented by two dummy variables, namely access to formal and informal credit, and was hypothesised to be positively linked with the proportion of farm and nonfarm self-employment income. In addition, higher shares of these income sources were also expected for households with higher physical capital as measured by the natural log of the value of all productive assets per working member. Livelihood strategies may change year to year, but they generally change slowly because of irreversible investments in human and social capital that are requirements for switching to a new income-generating strategy. Due to this path dependence, past livelihood choices are thought to considerably determine the present livelihood choices (Pender and Gebremedhin, 2007). This implies that households current income shares by source might be largely determined by their past livelihood strategies. Hence, we included the past livelihood strategy variable as an important explanatory predictor. Finally, commune dummy variables were also included to control for unobserved differences between communes in terms of farmland fertility, educational tradition, local infrastructure development and geographic attributes, and other unobserved community level factors that may affect households income sources. Measuring income inequality The Gini coefficient is popularly used to measure the disparity in the distribution of income, consumption and other welfare indicators (López-Feldman, 2006). Following Lanjouw, Murgai and Stern (2013), we examine the relationship between income sources and income inequality using Gini decomposition analysis by income source (Lerman and Yitzhaki, 1985; Shorrocks, 1982). According to Lerman and Yitzhaki (1985), the Gini coefficient of total income inequality (G) can be denoted as: where represents for the share of income source in total income, is the Gini coefficient of the income distribution from source, and is the correlation coefficient between income from source and with total income Y. Babatunde (2008) shows the share or contribution of income source to total income inequality can be expressed as: (1) or new urban areas. Such locations enable households to use their house for opening a shop, a workshop or for renting.

12 364 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong As shown by Stark, Taylor and Yitzhaki (1986), the income source elasticity of inequality indicates the percentage change in the overall Gini coefficient resulting from a 1 per cent change in income from source, and can be expressed as: (2) where is the overall Gini coefficient prior to the income change. As noted by Van Den Berg and Kumbi (2006), Equation (3) is the difference between the share of source in the overall Gini coefficient and its share of total income (Y). It should be noted that the sum of income source elasticities of inequality should be zero, which means that if all the income sources changed by same percentage, the overall Gini coefficient ( ) would remain unchanged. (3) Results and discussion Household income-generating activities and income composition Based on our own fieldwork experience and survey data, and combined with the definition of the Vietnamese informal sector introduced by Cling et al. (2010), five types of income sources are identified at the household level: (1) farm income (income from household agriculture, including crop and livestock production and other related activities); (2) nonfarm self-employment income (income earned from own household businesses in nonfarm activities); (3) informal wage income (income from wage work that is often casual, low paid and requires little or no education, often involving manual labour without formal labour contracts); (4) formal wage income (wage work that is regular and relatively stable in factories, enterprises, state offices and other organisations with formal labour contracts, and often requires skills and higher levels of education); and (5) other income (such as remittances, rental and pensions). Table 1 summarises the income shares by source for the sample. The overwhelming majority of surveyed households (83 per cent) derived some income from farming, but this was shown to account for only about 28 per cent of total income on average. This suggests that farming has remained important in terms of food security and cash income to some extent in Hanoi s peri-urban areas. A similar trend was also observed in the peri-urban areas of India and Ghana by Mattingly and Gregory (2006). Almost all surveyed households (90 per cent) participated in at least one nonfarm activity,

13 Farmland loss, nonfarm diversification and inequality among households in Hanoi 365 and income from nonfarm activities contributed about two-thirds of total income on average. Formal wage work and nonfarm self-employment offer much higher levels of income per hour compared to those of farm work and informal wage work. Table 1 Composition of household income and participation in and returns from different activities Income and its components Income per working hour Annual income per household Annual income per capita Share of total income (%) Participation rate ( %) Total income ,642 13,513 SD ,034 7,091 Nonfarm income ,801 9, SD ,571 7,140 A. Informal wage income ,559 2, SD ,703 3,973 B. Formal wage income ,431 3, SD ,762 6,232 C. Nonfarm self-employment ,811 3, SD ,803 6,231 D. Farm income ,432 3, SD ,169 3,621 Non-labour income (E) 3, SD 8,676 2,410 Note: SD (standard deviations). Estimates in columns 3 6 are adjusted for sampling weights. N= 477. Income and its components measured in VND 1,000. USD 1 equated to about VND 18,000 in Nonfarm income = (A+B+C). Table 2 presents the four main types of labour income-based strategies (livelihoods A to D) that households pursued before and after farmland acquisition, which were classified using cluster analysis. Cluster analysis also identified 21 households that pursued the non-labour income-based strategy (livelihood E) after farmland loss, as compared to 10 households that followed this strategy before farmland loss. Household livelihood strategies have dramatically changed after farmland loss. Prior to farmland loss, the proportion of households pursuing livelihood D used to be predominant, accounting for nearly half of the total households. This share, however, almost halved to around one-fifth of total households after farmland loss. Simultaneously, an increase is observed in all other types of livelihoods. This suggests that the loss of farmland has had a considerable effect on the choice of household livelihood strategy.

14 366 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong Table 2 Households past and current livelihood strategies Changes in livelihood strategies of households Livelihood strategy Whole sample Land-losing households Non-land-losing households Past Current Past Current Past Current Informal wage work (A) Formal wage work (B) Nonfarm self-employment (C) Farm work (D) Non-labour income (E) Total Note: Ten households that depended largely or totally on non-labour income were excluded from cluster analysis of the past livelihood strategy because they had no or little time allocation to labour activities. Determinants of household income shares by source Table 3 and Table 4 report the estimation results from the fractional logit and fractional multinomial logit models. Note that RPRs (Relative Proportion Ratios) are the exponentials of coefficients to measure the change in the relative proportion of income shares due to a unit increase in the explanatory variable, while keeping all other variables constant. Both sets of the results show that many coefficients are statistically significant, with the pattern of signs as expected. As shown in Table 3, the coefficients on the land loss variables in both years are highly statistically significant and negative, suggesting that a higher level of land loss is closely linked with a lower proportion of farm income. Holding all other variables constant, if the land loss in 2009 and land loss in 2008 rises by 10 percentage points the relative proportion of farm income share decreases 12 per cent and 18 per cent, respectively. As indicated in Table 4, the coefficients on the land loss variables in both years are statistically significant and positive, suggesting that land loss is positively associated with the share of all nonfarm income sources except for nonfarm self-employment income, where the coefficient on land loss in 2009 is not significant. Among nonfarm income sources, land loss is found to be most positively related to the share of informal wage income. Holding all other variables constant, a 10 percentage-point increase in land loss in 2009 and in 2008 corresponds with around a 17 per cent and a 32 per cent increase respectively in the relative proportion of the informal wage income share. The corresponding figures for the increases in the share of formal wage income are 16 and 18 per cent. For the case of the share of nonfarm self-employment income, only land loss in 2008 is statistically significant with a 14 per cent increase in the relative proportion. This implies that there may be some potentially high entry barriers to

15 Farmland loss, nonfarm diversification and inequality among households in Hanoi 367 adopting nonfarm self-employment, and simultaneously easier access to informal wage work, which makes this type of employment the most popular choice among land-losing households. A similar trend was also observed in a peri-urban village of Hanoi by Do (2006), and in some urbanising communes in Hung Yen, a neighboring province of Hanoi by Nguyen et al. (2011). Table 3 Fractional logit estimates for determinants of farm income share Farm income share Explanatory variables RPRs SE Coefficients SE Land loss ** (0.147) ** (0.530) Land loss *** (0.055) *** (0.419) Household size 1.172*** (0.067) 0.159*** (0.058) Dependency ratio (0.108) (0.132) Number of male working members (0.101) (0.108) Household head s gender 1.580** (0.309) 0.457** (0.195) Household head s age (0.008) (0.008) Age of working members 1.036*** (0.012) 0.035*** (0.012) Education of working members 0.876*** (0.031) *** (0.035) Social capital (0.050) (0.052) Farmland per adult 1.149*** (0.047) 0.139*** (0.041) Residential land size (0.005) (0.005) House location 0.627*** (0.100) *** (0.160) Formal credit (0.163) (0.173) Informal credit 1.470** (0.286) 0.385** (0.195) Productive assets/working members (Ln) 1.180** (0.084) 0.165** (0.071) Past informal wage work 0.303*** (0.069) *** (0.227) Past formal wage work 0.283*** (0.072) *** (0.254) Past nonfarm self-employment 0.174*** (0.042) *** (0.243) Commune dummy (included) Intercept 0.053*** (0.050) *** (0.942) Observations 457 Log pseudo likelihood Note: Estimates are adjusted for sampling weights. RPRs are relative proportion ratios. SE: robust standard errors. *, **, *** mean statistically significant at 10%, 5% and 1%, respectively.

16 368 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong Table 4 Fractional multinomial logit estimates for determinants of nonfarm income shares Explanatory variables Informal wage income share Formal wage income share RPRs Coefficients RPRs Coefficients Land loss ** 1.606** 4.309* 1.461* (3.177) (0.638) (3.365) (0.781) Land loss *** 2.769*** 5.400*** 1.686*** (8.778) (0.551) (3.299) (0.611) Household size 0.788*** *** (0.059) (0.075) (0.087) (0.095) Dependency ratio (0.194) (0.171) (0.302) (0.300) Number of male working members 1.486*** 0.396*** (0.214) (0.144) (0.264) (0.210) Household head s gender (0.251) (0.301) (0.266) (0.372) Household head s age (0.011) (0.011) (0.015) (0.015) Age of working members 0.948*** *** 0.949*** *** (0.016) (0.017) (0.017) (0.018) Education of working members *** 0.292*** (0.064) (0.063) (0.090) (0.067) Social capital * 0.138* (0.081) (0.078) (0.092) (0.080) Farmland/adult 0.866*** *** 0.879*** *** (0.046) (0.053) (0.043) (0.049) Residential land size (0.006) (0.006) (0.011) (0.011) House location (0.198) (0.246) (0.373) (0.326) Formal credit (0.214) (0.236) (0.211) (0.306) Informal credit (0.215) (0.270) (0.197) (0.330) Productive assets/working members (Ln) 0.697*** *** 0.711*** *** (0.063) (0.091) (0.084) (0.118) Past informal wage work 6.605*** 1.888*** 2.812** 1.034** (1.819) (0.275) (1.360) (0.483) Past formal wage work *** 2.590*** (0.499) (0.582) (4.959) (0.372) Past nonfarm self- employment (0.301) (0.460) (1.105) (0.554) Commune dummy (included) Intercept *** 5.574*** ( ) (1.328) (6.578) (1.757) Observations Wald chi2(96) Prob> chi Note: Robust standard errors in parentheses. RPRs are relative proportion ratios. Estimates are adjusted for sampling weights. *, **, *** mean statistically significant at 10%, 5% and 1% respectively. The farm income share is the excluded category.

17 Farmland loss, nonfarm diversification and inequality among households in Hanoi 369 Table 4 (continued) Explanatory variables Nonfarm self-employment income share Other income share RPRs Coefficients RPRs Coefficients Land loss *** 2.114*** (1.251) (0.662) (6.688) (0.807) Land loss *** 1.354*** 6.776** 1.913** (2.025) (0.523) (5.391) (0.796) Household size *** *** (0.086) (0.092) (0.075) (0.107) Dependency ratio *** 0.655*** (0.201) (0.159) (0.365) (0.190) Number of male working members 0.671** ** 0.416*** *** (0.123) (0.183) (0.122) (0.293) Household head s gender 0.510** ** 0.592* * (0.140) (0.274) (0.179) (0.303) Household head s age *** 0.036*** (0.012) (0.012) (0.012) (0.011) Age of working members (0.015) (0.015) (0.021) (0.021) Education of working members 1.110** 0.104** 1.332*** 0.287*** (0.056) (0.050) (0.087) (0.065) Social capital (0.075) (0.078) (0.108) (0.102) Farmland/adult 0.839*** *** (0.050) (0.060) (0.109) (0.118) Residential land size (0.009) (0.009) (0.007) (0.007) House location 2.936*** 1.077*** (0.649) (0.221) (0.281) (0.287) Formal credit 1.524* 0.421* (0.372) (0.244) (0.381) (0.315) Informal credit 0.542** ** (0.131) (0.241) (0.232) (0.395) Productive assets/working members (Ln) ** ** (0.114) (0.103) (0.094) (0.118) Past informal wage work * 0.765* (0.221) (0.346) (0.939) (0.437) Past formal wage work 0.443** ** 5.965*** 1.786*** (0.179) (0.403) (2.624) (0.440) Past nonfarm self- employment 7.408*** 2.002*** 5.741*** 1.748*** (2.088) (0.282) (2.372) (0.413) Commune dummy (included) Intercept * * (1.006) (1.329) (0.076) (1.962) Observations Wald chi2(96) Prob> chi Note: Robust standard errors in parentheses. RPRs are relative proportion ratios. Estimates are adjusted for sampling weights. *, **, *** mean statistically significant at 10%, 5% and 1% respectively. The farm income share is the excluded category.

18 370 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong As expected, age of working members is positively linked to the share of farm income but negatively related to the share of informal and formal wage income. Schooling of working members is negatively associated with the share of farm income, but positively correlated with that of nonfarm self-employment income and formal wage income. The findings were also similar in Shandong Province, China, where younger and more educated working members are more likely to participate in off-farm activities (Huang, Wu and Rozelle, 2009). Male-headed households were more likely to have a higher share of farm income than female-headed households. Having more working members who are male is associated with a higher proportion of informal wage income, but with a lower proportion of nonfarm self-employment income and other income. This may be because the majority of nonfarm self-employment activities are small trades and because of the provision of local services, which may be relatively more suitable for women. This finding is consistent with that of Pham, Bui and Dao (2010), who found that in rural Vietnam women are more likely than men to engage in nonfarm self-employed jobs, while men are more likely to be wage earners in nonfarm activities. These findings are also partly in line with Mattingly and Gregory (2006), who found that men have more opportunities to take up paid jobs in nonfarm activities in Kumasi, Ghana, Kolkata and Hubli Dharwad, India. However, in contrast to Mattingly and Gregory (2006), we find that lucrative nonfarm self-employment activities are not more restricted for women. Farmland per adult has a negative association with every share of nonfarm labour income. While the size of residential land is not related to any of the income shares by source, house location is positively associated with the percentage of nonfarm selfemployment income. The relative proportion of the share of nonfarm self-employment income is around three times higher for households with a conveniently situated house than those without it, holding all other variables constant. This implies that having a house in a prime location might allow many households to actively seize new nonfarm opportunities. A similar phenomenon was also observed in a peri-urban Hanoi village by Nguyen (2009) and in some rapidly urbanising areas of Hung Yen Province by Nguyen et al. (2011), where houses with a suitable location were utilised for nonfarm businesses such as restaurants, small shops, bars, coffee shops or beauty salons. Access to financial capital is related to shares of farm income and nonfarm selfemployment income, whereas each share of other income sources is not significantly related to financial capital. However, there are some interesting points to note. Access to formal credit has a positive association with the proportion of nonfarm self-employment income, but a similar relationship it is not observed for the case of farm income share. In addition, while access to informal credit is positively linked with the farm income share, it is negatively related to the nonfarm self-employment

19 Farmland loss, nonfarm diversification and inequality among households in Hanoi 371 income share. Formal loans may be used for nonfarm production rather than farm production, whereas informal loans may be used more often for farm production than nonfarm production. 6 Physical capital has a positive relationship with farm income share, but that is not the case for nonfarm self-employment income share. This may be because the majority of nonfarm self-employment activities are small-scale units, specialising in small trades and the provision of local services, which may not require a large amount of productive assets. Finally, social capital, as measured by the number of group memberships, is positively associated with the formal wage income share, but a similar association is not found for other income shares. Gini decomposition by income sources Figure 1 presents the distribution of income sources by income quintile. As compared to households in the higher-income quintiles (4 and 5), the lower-income quintile households (1 and 2) have a higher share of farm income and lower shares of nonfarm self-employment and formal wage income. This suggests that income shares by source are closely associated with the income distribution; specifically there is a positive association between the nonfarm self-employment income share, formal wage income share and per capita income, but a negative correlation between the farm and informal wage income shares and per capita income. Figure 2 shows the distribution of income sources by the size of farmland holdings. As revealed in this figure, households in the higher landholding quintiles have a much higher percentage of farm income but a lower share of nonfarm selfemployment, formal wage income and other income. By contrast, households in the lower landholding quintiles receive more income from nonfarm self-employment and informal wages, which implies that households with limited farmland might be pushed into these activities as a way to complement their income. Finally, the share of formal wage income appears not to be associated with the distribution of farmland, This suggests that this income source may be associated with other factors, such as education, and the availability of formal employment provided by proximity to industrial zones, commercial centres and new urban areas. 6 As revealed by the surveyed households, about 45 per cent of borrowing households said that one of their purposes of borrowing formal loans was for nonfarm production, while the corresponding figure for farm production was only about 10 per cent. By contrast, 40 per cent answered that one of the purposes of borrowing informal loans was for farm production, while the corresponding figure for nonfarm production was only around 12 per cent.

20 372 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong Figure 1 Income shares by source and income quintiles Figure 2 Income shares by source and farmland holding quintiles Table 5 presents the Gini decomposition of income inequality by income source. The overall Gini coefficient for the sample households was 0.267, which is much lower than the Gini coefficient of for the whole country and for the Red River Delta reported by GSO (2008). This indicates a quite low degree of income inequality among the sample households. Lower measures of inequality can be expected for smaller geographical areas, due to the fact that households in a small region are likely to have more similarities than households across the whole country or region (Minot, Baulch and Epprecht, 2006).

21 Farmland loss, nonfarm diversification and inequality among households in Hanoi 373 In previous studies on the decomposition of income inequality in Vietnam, household income has been often disaggregated into various sources, including wage income, nonfarm self-employment income, agricultural income and other income (Adger, 1999; Cam and Akita, 2008; Gallup, 2002). Our paper is the first to further break down wage income into two sub-categories, namely informal wage income and formal wage income. The results reveal that nonfarm self-employment, formal wage income and other income are the major contributors to overall income inequality among the sample households. Taken together, they account for 93 per cent of the total income inequality. By contrast, farm and informal wage income are inequalityreducing; the pseudo-gini coefficients of these income sources are much lower than the total Gini coefficient, whereas the pseudo-gini coefficients for nonfarm selfemployment income, formal wage income and other income are much higher than the total Gini coefficient. Specifically, 10 per cent increases in income from farm and informal wage activities are associated with 1.7 per cent and 1.9 per cent declines in the overall income inequality, respectively. In contrast, the same increase in nonfarm self-employment, formal wage income and other income is associated with a 1.4 per cent, 1.6 per cent and 0.57 per cent increase in the overall income inequality, respectively. Table 5 Gini decomposition of income inequality by income source Income source Income share Gini Correlation with the distribution of total income Pseudo-Gini Share to total income inequality Source elasticity of total inequality Sk Gk Rk GkRk (RkGkSk)/G (RkGkSk)/G-Sk Farm Nonfarm self-employment Informal wage Formal wage Other income Total Note: Estimates are based on annual per capita incomes. N=477. Looking at the third and fourth column in Table 5, the results show that the inequality of farm and informal wage incomes among households is lower than the inequality of nonfarm self-employment, formal wage income and other income among households. In addition, as compared with nonfarm self-employment income,

22 374 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong formal wage income and other income, farm and informal wage incomes each have a much lower correlation with the distribution of total income. Consequently, the incomes from farm and informal wage work have had an equalising effect on the income distribution. This finding is partly in accordance with Gallup (2002) and Cam and Akita (2008), who found that while agricultural income reduced the inequality of income distribution, it was nonfarm self-employment income and other income sources that mainly contributed to inequality in Vietnam. Conclusion and policy implications Under the impact of farmland loss due to urbanisation and industrialisation, landlosing households have diversified into nonfarm activities. Among the sources of nonfarm income, the income share from informal wage jobs is found to be most positively associated with land loss, which suggests that such low-skilled, paid jobs have been emerging as the most common choice of land-losing households in or near Hanoi s peri-urban areas. Possibly, this is also indicative of a high availability of manual labour jobs in Hanoi s peri-urban areas. According to Cling et al. (2010), the informal sector in Hanoi offers the greatest job opportunities for unskilled workers. Such job opportunities are also often found in Hanoi s rural and peri-urban areas, and those working in this sector have much a lower level of education than those in other sectors (Cling, Razafindrakoto and Roubaud, 2011). Consequently, such job opportunities might allow many land-losing households to supplement a shortfall of income with informal wage income, which in turn might mitigate the negative effects of land loss and improve household welfare. The results suggest an important role for natural capital in shaping peri-urban livelihoods. Having more farmland is associated positively with farming, but negatively with nonfarm activities. A house or plot of residential land in a prime location is emerging as a crucial asset that is closely linked with nonfarm household businesses. In addition, the results indicate that there are also other important asset-related variables that are positively related to diversifying into lucrative nonfarm activities. Access to formal credit has a positive relationship to the share of nonfarm self-employment income. As a result, government assistance in access to formal credit may help households diversify into nonfarm household businesses. Better education is found to be positively linked with shifting away from farming and diversifying toward highly remunerative jobs. This implies that investment in children s education may be a way to take advantage of opportunities for well-paid jobs for the next generation. The results indicate that farmland loss has a negative effect on the share of farm income, which is one of two income sources that had a reducing effect on income inequality. Given the context of shrinking farmland due to rapid urbanisation in Hanoi s peri-urban areas, a declining share of farm income will be inevitable. Conse-

23 Farmland loss, nonfarm diversification and inequality among households in Hanoi 375 quently, increasing inequality might seemingly be difficult to avoid without restricting farmland conversion for industrialisation and urbanisation. Nevertheless, farmland loss has a positive effect on the share of informal wage income, which is the only source among nonfarm income sources that had an equalising effect on the income distribution. Thus, land loss seems to have indirect mixed impacts on the income distribution. This study has contributed to the understanding of the impacts of farmland loss on nonfarm diversification and inequality, but still has several limitations that offer possibilities for future work. First, given the loss of land due to urbanisation, periurban households have adapted by intensively farming small plots of land and moving towards high value agricultural products with a ready urban market (Mattingly, 2009). This suggests that future studies should examine the impact of land loss on agricultural intensification and transition towards highly profitable farming. Second, although the compensation with land for land provides households with a plot of commercial land they can use to change or diversify their livelihoods towards nonfarm activities (ADB, 2007), this policy might increase inequality because some might receive plots in a prime location (corner plots, for example) whereas many others might be allocated plots in a non-prime location. Therefore, this interesting issue should be investigated in future research. Finally, another interesting question for future investigation is that, while compensation money for land loss might provide the means to help households diversify their livelihood towards lucrative nonfarm activities, why have only a few households used the compensation for investing in nonfarm production? Acknowledgements The authors thank the Vietnam Ministry of Education and Training and the University of Waikato, New Zealand, for funding this research. The authors would like to thank Dr Maarten L. Buis for helpful feedback regarding the STATA command for and the interpretation of the fractional multinomial logit model authored by him.

24 376 Tran Quang Tuyen, Steven Lim, Michael P. Cameron and Vu Van Huong Appendix 1 Summary statistics of explanatory variables included in the models Explanatory variables M SD Mean SD Min Max Farmland acquisition Land loss 2009 (%) Land loss 2008 (%) Human capital Household size Dependency ratio Number of male working members Gender of household head* Age of household head Age of working members Education of working members Natural capital Owned farmland size per adult Residential land size House location* Physical capital Social capital Financial capital Formal credit* Informal credit* Past livelihood Informal wage work* Formal wage work* Nonfarm self-employment * Note: Estimates in the second and third columns, including Mean (M) and standard errors (SD) are adjusted for sampling weights; * means dummy variables. References adb (asian development bank) (2007), Agricultural Land Conversion for Industrial and Commercial Use: Competing Interests of the Poor, in Asian Development Bank (ed.), Markets and Development Bulletin, Hanoi, Vietnam, Asian Development Bank, pp adger, w. n. (1999), Exploring Income Inequality in Rural, Coastal Viet Nam, The Journal of Development Studies, 35, babatunde, r. o. (2008), Income Inequality in Rural Nigeria: Evidence from Farming Households Survey Data, Australian Journal of Basic and Applied Sciences, 2, buis, m. l. (2008), FMLOGIT: Stata Module Fitting a Fractional Multinomial Logit Model by Quasi Maximum Likelihood [Statistical Software Components ], Boston College Depart-