The Happy Farmer: Self-Employment and Subjective Well-Being in Rural Vietnam

Many developing countries are undergoing a process of ‘structural transformation’ in which large numbers of workers move from agriculture to other sectors and from self-employment into wage work. This study investigates the consequences of this process for subjective well-being. Using data from Vietnam, we focus on the effects on happiness from being self-employed in farming, as opposed to working for a wage or being self-employed in non-farming. The analysis contributes to the growing, but still relatively small literature on happiness in developing countries.¹ Apart from self-employment, we investigate other, potential determinants of subjective well-being, including income, age, gender, marital status, schooling, migration, social networks, and economic shocks.²

Vietnam is experiencing fast structural transformation and economic development. The share of the labor force in agriculture dropped from 70% in 1996 to 47% in 2012. The share of self-employed dropped from 83 to 65% over the same period and these trends are set to continue over the coming decades.³ While such occupational shifts play an important role in generating increased productivity and income (e.g. McMillan and Rodrik 2011) they may also have important effects on subjective well-being. Benz and Frey (2008a, b) show that self-employment is associated with significantly higher job satisfaction than wage labor in a number of countries and ascribe this effect to higher levels of ‘procedural utility’ for the self-employed. The move from agriculture to non-agriculture may also affect subjective well-being. Most people in rural areas were raised on a farm and may feel most competent in this environment. Also, working on a family farm often allows one to maintain close relations with parents and other family members.

Below, we show that the hypothesis of a positive effect of self-employment in farming on subjective well-being can be derived from self-determination theory, an influential, psychological theory of human motivation (Deci and Ryan 1985, 2000). We then use data from a rural household survey in 12 provinces of Vietnam to test this hypothesis. Our measure of happiness is a one-item survey question asking respondents to state how pleased they are with their lives. It is a measure of ‘life evaluation’, as opposed to ‘emotional well-being’ (Kahnemann and Deaton 2010).

Results show, first, that the level of subjective well-being in rural Vietnam is low. This may be surprising given that rapid economic progress is taking place. Some 48% of respondents report being ‘not very’ or ‘not at all’ pleased with their lives. Second, there is a substantial, positive effect of self-employment on happiness. This effect is largely driven by self-employment in farming, rather than self-employment in other sectors. The difference between self-employment in farming and wage work appears only partly to be transitory and is not driven by particular types of wage work; rather it applies across many different types of jobs. The effect of self-employment in farming is strongest when income and other intermediate outcomes are controlled for, but is also significant when they are not. These findings highlight the psychological cost associated with the transition from working on family farms to wage work and self-employment in non-farming.

These results of digging into what Berry (2008) refers to as a third and fourth level of labor market analysis potentially contribute to explaining earnings gaps between formal and informal sectors and between agriculture and other sectors. The presence of such differences is a key observation underpinning classic models of economic development, such as Lewis (1954), Ranis and Fei (1961), and Todaro (1969). Earnings differentials have been explained as the result of transport frictions, minimum wages, union bargaining, unemployment, and other factors (e.g. Stiglitz 1974; Teal 1996). Differences in ‘procedural utility’ associated with different types of employment are an alternative or complementary explanation, as wage workers in industry and services may demand monetary compensation for the psychological cost of working in these types of jobs.

The study is organized as follows: Sect. 2 discusses the potential effects of self-employment in farming on happiness, in the light of self-determination theory. Section 3 describes the data and Sect. 4 presents the variables used and the analytical model. Section 5 provides descriptive statistics and Sect. 6 presents regression results for the effects of self-employment on happiness. Section 7 discusses the effects of various control variables including income, and Sect. 8 concludes.

Our hypotheses about the effects of self-employment are based on the economic concept of ‘procedural utility’ (Frey et al. 2004), which in turn relies on self-determination theory (SDT) (Deci and Ryan 1985, 2000). People are said to obtain procedural utility when their well-being is affected not only by the final outcomes they achieve but also by the process of reaching those outcomes. For example, different types of employment may yield different (procedural) utility, even if they generate the same income. SDT helps us understand which attributes of employment are important for determining procedural utility. SDT posits the existence of three, fundamental, psychological needs, namely autonomy, competence and relatedness. Autonomy is essentially the freedom to act in accordance with one’s values. Competence is the experience of mastering one’s environment. Relatedness refers to the ‘desire to feel connected to others’ (Deci and Ryan 2000, p. 231). Proponents of SDT stress that these are not culture-specific; they are universal, human needs.⁴

In their influential papers about the effects of self-employment on job satisfaction and subjective well-being, Benz and Frey (2008a, b) stress the importance of autonomy. Positive effects of self-employment are interpreted as resulting from procedural utility generated by higher autonomy on the part of the self-employed, relative to wage workers. Other empirical results can be viewed as supporting this interpretation. Most studies of developed countries find a positive effect of self-employment on happiness and/or job satisfaction. Blanchflower and Oswald (1998), Blanchflower et al. (2001), Blanchflower (2004), Andersson (2008), Fuchs-Schündeln (2009), Alesina et al. (2004) all document such effects. Results on occupation and subjective well-being in developing and in transition countries are more mixed. Graham (2005a) presents results on the effect of self-employment in both Russia and Latin America (using a pooled sample from the Latinobarometro survey in the latter case). She finds that in Russia, the self-employed are on average happier than wage workers, whereas in Latin America the self-employed are less happy. Graham and Petinatto (2001) use data from 17 Latin American countries and show that being self-employed has no significant effect on happiness. Bianchi (2012) shows that the positive effect of self-employment on job satisfaction is conditional on financial sector development. In the countries with the least sophisticated financial markets, there is no significant effect of self-employment. Both Graham (2005b) and Bianchi (2012) interpret their negative or insignificant findings as the result of self-employment being less of a choice in some developing countries than in developed countries. If people are self-employed only because no wage work is offered, their experience of autonomy is likely to be reduced and the effect of self-employment on happiness may vanish.⁵

We do not dispute the claim that increased autonomy is potentially an important effect of self-employment, particularly when self-employment is freely chosen. However, we point out that mode of employment may also impact on the fulfilment of other psychological needs, such as competence and relatedness. Particularly, we hypothesize that in the rural areas of a developing country, residents are likely to feel more competent and more closely related to other people, particularly family members, if they work on a family farm than in most other types of employment. Farmers are likely to feel more competent than other people because most people in rural areas (including the non-farmers) were brought up to be farmers. Even if they have received formal schooling, much of their socialization prepared them to be farmers. Working in agriculture could increase relatedness for two reasons. First, farmers typically work in the same, physical location as where they live. This strengthens ties to spouse, children and possibly other family members. Second, since farming was even more common in previous generations than in the present, most people’s parents are/were farmers. Being a farmer oneself is likely to strengthen bonds with the elder generation of one’s family, regardless of whether they live in the same house as oneself. This might be particularly important in the Vietnamese context, where family relations are considered by most people to be highly important. The 2001 World Values Survey in Vietnam asked respondents about the importance of different ‘life domains’. Some 82% of respondents say that the family is ‘very important’, while 57% regard ‘work’ as very important. Only 22% rank ‘friends’ as very important (Dalton et al. 2002).

Together, these considerations lead to the hypotheses that, ceteris paribus:

To summarize the discussion above, the first effect is theorized as mainly driven by autonomy, while the latter is mediated by competence and/or relatedness. To our knowledge, no previous study of subjective well-being has investigated differential effects of self-employment in farming and non-farming on subjective well-being. Also, while a large literature investigates the three dimensions of self-determination (Deci and Ryan 2000 and the references therein), we do not know of other studies that specifically relate the concepts of competence and relatedness to the discussion about effects of self-employment on happiness.

Our empirical analysis is based on the 2012 wave of the Vietnam Access to Resources Household Survey (VARHS), implemented in the rural areas of 12 provinces in Vietnam between June and August 2012. VARHS re-interviewed rural households sampled for the income and expenditure modules of the 2002 and 2004 Vietnam Household Living Standards Survey (VHLSS) in the 12 provinces.⁶ To ensure proportionate representation of households that have come into existence after 2004 and replace households that could not be re-interviewed, 544 additional households were sampled (drawn from the list of households available from the 2009 Population Census). Provinces were selected to facilitate the use of the survey as an evaluation tool for Danida-supported development programs in Vietnam.⁷ Seven of the 12 provinces are covered by the Danida Business Sector Program Support and five provinces by the Agricultural and Rural Development (ARD) program. The provinces supported by ARD are located in the north-west and central highlands, so these relatively poor and sparsely populated regions are somewhat over-represented in the sample, relative to the Vietnamese population. The survey is statistically representative of rural households at province level.

The measure of happiness collected in VARHS is a single-item measure of life evaluation. In particular, it asks: ‘taking all things together, would you say you are: (1) very pleased with your life; (2) rather pleased with your life; (3) not very pleased with your life; (4) Not at all pleased with your life’; and respondents choose one answer. A strong advantage of the single-item approach, as opposed to multiple-item measures, where respondents state their satisfaction with life in several different domains (work, marriage etc.), is reduced cultural sensitivity (Veenhoven 1984). The importance of being satisfied with life domains such as work or marriage may vary greatly across cultures and over time, implying that the appropriate weights for different items also vary.

The question on subjective well-being was only answered by one respondent in each household, typically the household head. Accordingly, our sample is not representative at the individual level. On the other hand, there are important benefits from using household survey data. In particular, the survey collects detailed data on a number of individual- and household level characteristics, including income, occupation, health, education, social networks, migration, and so on. This information is much more detailed than in surveys such as the Gallup World Poll and the World Values Survey (WVS), which have been used in many studies of happiness (e.g. Helliwell et al. 2012; Deaton 2008).

Our main purpose is to investigate the effect of employment category on subjective well-being. We run regression models of the following type:where H _i is the answer of respondent i to the subjective well-being question described above; S _i is a vector of indicators for employment category; X _i is a vector of other variables that may affect subjective well-being; p _i is a dummy for province of residence; and ε _i is the error term. Errors are allowed to be correlated within communes (the primary sampling unit), but not across. β, γ, and λ are vectors of coefficients to be estimated.

In terms of employment category, we distinguish between self-employment, wage work, and no employment. The vast majority in the last category are either too old or too sick to work. Only very few are ‘unemployed’ in the sense of desiring a job without being able to find one. We distinguish between self-employment in farming, in non-farm enterprises, and in collection of common property resources (CPRs).⁸ Classification is based on respondents’ ‘main’ source of employment, defined as the occupation they spend most time in. Because the main aim of the analysis is to identify the difference in happiness between self-employed and wage workers, respondents with no employment are excluded from the estimation sample.

Figure 1 presents the analytical model used to select the elements of X _i , the control variables in the regressions. We distinguish between two types of variables. ‘Background variables’ are exogenous to employment. These variables may affect employment and could also directly drive happiness. ‘Intermediate variables’, on the other hand, are potentially affected by employment and therefore may mediate a causal effect from employment to happiness. Controlling for background variables allows us to identify the total, causal effect of employment category on happiness, including indirect effects that operate through income, risk exposure, social networks, and so on. However, the hypotheses derived in Sect. 2 concern the direct effect of employment category on autonomy, competence and relatedness and therefore on subjective well-being. Testing these hypotheses requires that intermediate variables such as income are controlled for.

Our main purpose is therefore to estimate models where both background- and intermediate variables are controlled, but estimates of the total (direct plus indirect) effect are also of interest.

Background variables include age, gender, ethnicity, place of birth (commune of current residence or elsewhere), and years schooling. All other control variables are viewed as ‘intermediate’. Among these variables, we first include a measure of income, measured at the household level. While a positive correlation between income and happiness is a standard finding in individual level analyses of happiness, it has been hotly debated whether this correlation is driven by absolute or relative income (e.g. Easterlin 1974; Cummins 2000; Berry 2009).

We also control for landlessness. In agriculture-based societies, such as rural Vietnam, land is a key source of income, risk coping, prestige, and identity. Two different measures of health, namely (a) the number of days in the last year the respondent was unable to perform normal activities due to illness, and (b) an indicator for the respondent’s household being hit by any health shocks that led to income losses in the past 2 years are also included. Controls for the number of children in the household below 15 years of age and indicators for marital status are also used.

We include two measures of migration, in addition to the place of birth-indicator mentioned above, namely (a) an indicator for a member of the household having migrated temporarily (and currently being away), and (b) an indicator for former household members having permanently migrated to another commune, district or province (see De Jong et al. 2002; Knight and Gunatilaka 2010b).

Measures of social networks are also used (see Powdthavee 2008; Berry and Hansen, 1996). We distinguish between membership of the Communist Party, ‘mass organizations’, and other formal groups. Mass organizations are the most important type of formal group in Vietnam and include the Women’s, Farmers’, Youth, and Veterans’ Unions. To proxy the strength of respondents’ informal social networks, we use a measure of the number of weddings the respondent’s household has attended in the past year.

We include dummies indicating whether the respondent’s household experienced any of five different types of shocks in the past 2 years. The five types of shocks are (a) health shocks (already discussed above), (b) natural disasters, (c) pest infections and crop disease, (d) ‘economic shocks’ (price changes, unemployment, failure of an investment, and land loss), and (e) a residual category. Finally we include an indicator for being the household head due to the composition of the sample where household heads are over-represented.

Figure 2 presents the distribution of answers to the subjective well-being question described in Sect. 3. There is significant variation across respondents. A total of 52% are ‘very’ or ‘rather’ pleased, while 48% are ‘not very’ or ‘not at all’ pleased with their lives. The share falling in the last two categories is large and may be viewed as a cause for concern.¹⁰ Our questions and sampling are somewhat different from other surveys, so comparison of the measured level of happiness measured is not trivial. We therefore focus on the explanations for the observed variation.

Table 1 presents descriptive statistics on the explanatory variables discussed above, and on subjective well-being. The first column shows the mean of each variable. The second column shows medians for non-dummy variables. The last two columns, with the headings ‘low value on row variable’ and ‘high value on row variable’ show the share of respondents, who are either ‘very’ or ‘rather’ pleased with their lives, separately for respondents with high and low values on the row variables. For dummy variables, ‘low’ is zero and ‘high’ one. For other variables, ‘low’ is at or below the median value, whereas ‘high’ is above the median. For example, the sixth row (income per capita) shows that 41% of those with below-median income, and 63% of those above the median, are rather or very pleased with life.

The table does not reveal strong effects of employment category. The happiest respondents are those self-employed in non-farm enterprises. The tables in what follows show that results change very significantly when these factors are controlled for.

Table 3 presents regressions for happiness. The dependent variable is the four-category subjective well-being measure described in Sect. 3 and Fig. 2. The measure is an ordinal scale variable. Therefore, we estimate ordered probit regressions (except regression 5). Standard errors are adjusted for commune level clustering. All regressions include province dummies (not shown). Results with commune fixed effects are remarkably similar (see Markussen et al. 2014).

Consider first the respondent’s type of employment. The reference category is wage work. Regressions 1, 3, and 5 include a dummy for self-employment, while regressions 2 and 4 sub-divide the self-employed into the three sectors discussed above: own farm, non-farm enterprises, and CPR collection. In addition to the employment indicators, regressions 1 and 2 include background variables and province dummies, while regressions 3 and 4 include background and intermediate variables as well as province dummies. Regression 5 is an instrumental variables model, discussed below. Pseudo log-likelihood and pseudo R-squared values confirm that control variables increase the explanatory power of the model substantially. Linktests in all cases fail to reject the hypothesis that the model is correctly specified (the table displays the p value for the \(\hat{y}^{2}\)-term, the key element of the linktest).

Regressions 1 and 3 both show a statistically significant, positive effect of self-employment, similar to the results from Western countries already discussed. The effect is stronger in model 3, which controls for intermediate variables, than in model 1, which does not. This reflects the negative correlations between self-employment and some of the intermediate factors with a positive effect on happiness, such as income and Communist Party membership. In other words, the indirect effects of self-employment on happiness are negative.

Regression 2 shows statistically significant, positive effects of self-employment in both farm- and non-farm enterprises. Self-employment in CPR collection is positive, but statistically insignificant. Importantly, regression 4 reveals that the direct effect of self-employment on happiness (i.e. the effect that does not operate through income or other factors) is largely driven by self-employment in farming. The indirect effects of self-employment appear to be negative for farm- and positive for non-farm enterprises. This reflects, among other things, the fact that average income is higher for wage laborers than for those self-employed in farming, and lower than for those operating non-farm enterprises.¹¹ The large difference between the effects of self-employment in farming and in non-farm enterprises suggests that in the context of rural Vietnam, the psychological effect of self-employment is not driven by autonomy per se, as suggested by Benz and Frey (2008a, b), but by autonomy in a particular environment, namely that of the family farm. The results are consistent with the theoretical discussion in Sect. 2, which suggested that working on a family farm may increase feelings of competence and relatedness, as well as autonomy, and therefore enhance subjective well-being.

Regression 5 presents the results of an instrumental variables (IV) analysis where self-employment is instrumented by commune level self-employment, wage rates, and the presence of craft villages (see Sect. 4). We are not aware of methods to deal with endogenous regressors in the ordered probit model and therefore estimate a two-stage least squares regression. The bottom lines of the table present tests for endogeneity of self-employment, and for instrument quality. First, a Chi squared test of endogeneity does not reject the null-hypothesis that self-employment is exogenous. This should increase confidence in the results from regressions 1–4, which treat self-employment as exogenous. While this makes the IV-results less interesting we nevertheless present them, noting that the power of the endogeneity test is limited. The Hansen J test supports the assumption of instrument exogeneity. The Kleibergen-Paap F statistic is useful for judging whether instruments are weak (Kleibergen and Paap 2006). The value of the F-statistic is only 2.2. Comparison with the critical values in Stock and Yogo (2005) suggests that the instruments are quite weak and that point estimates as well as inference may be biased. Consequently, the results in regression 5 should the treated with caution. Nevertheless, the fact that the estimated coefficient on self-employment remains positive and statistically significant does lend further support to the hypothesis of a causal effect from self-employment to happiness.

Tables 4 and 5 explore in more detail what drives the negative effect of working for a wage, relative to working on one’s own family farm. Table 4 presents the distribution of wage workers across sectors, showing that services and construction are the largest sectors. Table 5 presents additional regression analyses. The regressions include the same control variables as in regressions 3–5 in Table 3 (estimates not reported). Note that the reference category for employment is now self-employment on own farm, rather than wage labor. This allows us to investigate the effects of different types of wage labor in detail.

Regression 2 investigates whether the negative effect of wage labor is transitory or permanent. The regression includes indicators for being a wage worker (a) now, (b) 2 years ago, and (c) both now and 2 years ago (similarly for self-employment in non-farm enterprise and CPR collection). The data on employment 2 years ago were collected in the 2010 wave of the VARHS survey. Data are therefore only available in those cases where the individual respondent, who answered the question on happiness in 2012, was also present in the household in 2010.¹² Results show that the coefficient on current wage labor remains negative and statistically significant, while the coefficient on wage employment in both 2010 and 2012 is positive, with approximately half the numerical value of the coefficient on current wage work. In other words, the negative effect of wage labor is twice as high for those who turned to wage work recently (within the last 2 years) as for those who worked for a wage longer. This suggests that the effect of wage work is partly transitory. Yet, it takes more than 2 years to fully eliminate it.

Regressions 3–6 divide the wage-worker category into sub-categories. Regression 3 distinguishes between skilled and unskilled workers. Remarkably, the effects of these two types of wage work (again relative to working on own farm) are almost identical. Therefore, upgrading the educational level of the labor force may not necessarily eliminate the psychological cost of wage labor. Regression 4 splits the wage workers into formal and informal sectors. A ‘formal job’ is defined as one with a written labor contract (32% of those classified as wage workers are in the formal sector). The estimated effects are significantly negative for both sectors.

Regression 5 distinguishes between wage work in the public and private sectors, and in SOEs. The effects are negative for each sector, but the point estimate is higher for private sector work (the correlation between informal jobs and private sector jobs is quite high, r = .70). Regression 6 splits the ‘wage worker’ category into sectors of occupation (agriculture, mining, manufacturing, construction, and services). The results for different sectors are remarkably similar. Accordingly, the negative effects of wage labor do not appear to be tied to agriculture. On the other hand, the statistically significant, negative effect on wage work in agriculture shows that it is not agriculture as such that drives happiness—it is working on one’s own family farm that makes a significant difference.¹³

In general, Table 5 shows that the negative effect of wage labor is found across a broad range of jobs and among both young and old respondents; so the effect is unlikely to be driven by the characteristics of some types of particularly tough or low-status jobs. Rather, the negative effect of wage work appears to be accounted for by features shared by many types of jobs. One interpretation, based in self-determination theory, is that the key factor is a decrease in relatedness, rather than decreases in autonomy or competence. First, we would probably expect skilled workers to experience higher autonomy and competence than unskilled workers, yet the subjective well-being of these two groups is almost identical (regression 3). Second, we argued above that many people in rural Vietnam may feel more competent in agriculture than in other sectors. However, this feeling should not depend on working on one’s own farm, rather than someone else’s. In contrast, regression 6 shows that people working on their own farms are happier than those working on other farms. One thing that does distinguish those employed on their own farm from agricultural laborers and from all other type of wage workers is that they are likely to be physically and mentally closer to their family members, including the elder generation. In this interpretation, the positive effect of self-employment in farming on subjective well-being is simply an effect of maintaining close ties to one’s relatives.

Many of the results on control variables are interesting in themselves, especially given the scarcity of happiness analyses in Vietnam and elsewhere in developing countries. We provide a brief overview and refer to Markussen et al. (2014) for further detail. The discussion refers to Table 3, unless otherwise stated.

Focusing on rural Vietnam, this study documented a significant, positive effect on happiness from self-employment on family farms, relative to self-employment in non-farm enterprises and wage labor. The positive, direct effect of self-employment in farming is partly though not fully offset by negative, indirect effects working through income, social networks, and other variables. The negative effect of wage labor, relative to work on own farm, applies across a wide range of jobs. The effect is present for both younger and older workers. While the negative effect of wage labor appears to be partly transitory, it has not vanished after 2 years of wage employment. The positive effect of self-employment on happiness is consistent with results from Western countries, the main difference being that the effect in rural Vietnam is driven mainly by self-employment in farming, rather than other sectors.

Building on self-determination theory, other researchers have interpreted a positive effect of self-employment as the result of stronger feelings of autonomy among the self-employed (Benz and Frey 2008a, b). We point out that employment category may also affect fulfilment of other psychological needs than autonomy, such as the needs for competence and relatedness. Our results are consistent with interpreting the positive effect of self-employment in agriculture as arising mainly from stronger feelings of relatedness among people working on their own farm.

Labor markets in developing countries are often viewed as ‘segmented’, in the sense that there are exogenous barriers to movement from one sector to another (Fields 2011). The presence of such barriers is often inferred from the observation of large income differences across sectors (Wachter 1974; Cain 1976). The results presented here suggest that this inference is not necessarily valid. Earnings differences may exist to compensate for differences in intrinsic (procedural) utility. If anything, our results suggest that earnings differences are too small to compensate wage workers for loss of procedural utility. The psychological burden of economic development is significant and policy makers and employers should consider how this burden may be addressed. We argued that the loss of subjective well-being incurred by wage workers may stem from weakened ties to family members. This implies that an important policy goal is to enable wage workers to maintain a fulfilling family life, e.g. by limiting work hours and protecting the right to vacations and decent living conditions.

One constraint of our study is that it covers rural areas only. No data are available on urban zones, yet many of the people who shift from farming to other types of employment move from rural to urban areas. Would the inclusion of urban people in the sample change our results? Results from China, which shares many similarities with Vietnam, suggest that it would not. Both Brockmann et al. (2008) and Knight and Gunatilaka (2010a) report that in spite of a large income advantage, urban Chinese report lower levels of happiness than their rural countrymen.

In general, the results presented here are remarkably in line with findings from other countries with completely different cultures and levels of development. For example, the effects of self-employment, relative income, marital status, health, schooling, age, and social capital are very similar to those reported for developed Western countries (Veenhoven 2012).¹⁵ This weakens the view that the happiness of rural dwellers in developing countries are determined by ‘traditional’, culture-specific values, which differ strongly from ‘modern, western’ values.