Skip to main content
Erschienen in:

Open Access 01.12.2024 | Research Paper

Income Fluctuations and Subjective Well-being: The Mediating Effects of Occupational Switching and Remittances

verfasst von: Azizbek Tokhirov

Erschienen in: Journal of Happiness Studies | Ausgabe 8/2024

Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.

search-config
loading …

Abstract

Does money bring happiness? To answer this question, I study the consequences of income fluctuations caused by commodity price changes on well-being patterns in regions specializing in export agriculture. Using nationally representative survey data in a difference-in-differences framework, I investigate the effects of the 2010/11 short-term increase in the global price of cotton. I demonstrate that it can be viewed as a positive income shock for the cotton-producing communities of Tajikistan. The main results indicate that the net subjective well-being effects of the cotton price increase are negative: exposure to the shock at the aggregate level is associated with a notable decrease in the reported levels of financial and life satisfaction. To explain this paradox, I consider split sample analyses, which suggest that the shock led to within-community occupational sorting and that its well-being effects are negative among households that were in the agriculture sector before the shock and barely positive for newly become farmers. Observing the increasing volume of remittances in the world and their significance to the economy of Tajikistan, I also study how remittances affect the relationship between income volatility and happiness. Further estimations reveal that family remittances are not significantly affected by and can partially mediate the negative effects of short-term income changes. The mediating effects of remittances only affect financial satisfaction, suggesting that a mere compensation of losses does not fully restore the quality of life.
Hinweise

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1 Introduction

Climate risks have been associated with economic downturns both historically (Gupta et al., 2023) and in contemporary times (Zhao et al., 2022). One of the sectors that significantly depends on climate is agriculture. Unexpected and even expected environmental changes might lead to uncertainty and production disruptions. Thus, it is not surprising that there is a consensus in the literature on the negative effects of climate change on agriculture (Nelson et al., 2014). However, this was not always the case, and the views of academics were more optimistic in the early 1990s (Reilly et al., 1994; Tobey et al., 1992). It was argued that the effects of climate change can be positive, while its negative effects might be mitigated by inter-regional adjustments. Despite a notable increase in the usage of natural resources from 1992 to 2005, which was much larger than population growth (UN News, 2011), the paradigm shift was mainly driven by improvements in data availability and modeling (Nelson et al., 2014).
Methodological advancements also made it possible to capture the effects of climate change on agricultural workers, even within directly unaffected countries. One such mechanism operates via global commodity prices. Environmental incidents might create deficits and encourage producers to increase production via higher prices, confirming the predictions of the King’s Law (Mpabe Bodjongo, 2022). Moreover, these indirect effects could also be amplified by spillovers because agribusinesses become more connected with agricultural commodities during turmoil periods (Cagli et al., 2023). The inverse relationship between climate change and commodity prices is well-documented (Nelson et al., 2014). In this study, using econometric robust tools, I investigate a less researched topic: the well-being consequences of climate-induced price changes that happened in the cotton industry in the 2010/11 agricultural season.
The cotton commodity market is typically characterized by stable prices. However, due to floods and droughts in the major cotton-producing countries, including China, India, and Pakistan, cotton more than doubled in global price between July 2010 and March 2011. Observing this sudden price increase, agricultural workers in cotton-suitable areas in non-affected countries saw an opportunity to improve their incomes by switching from other crops to cotton. One of these countries is Tajikistan, where cotton accounted for 30% of export revenue back in 2011 (FAO et al., 2011). Figure 1 illustrates a link between cotton production in Tajikistan and global cotton prices, suggesting that Tajik agricultural workers did respond to the cotton price increase in the expected way. The figure also indicates the short-term nature of the 2010/11 spike. Tajikistan was possibly not the only country affected by this shock. In Fig. 2, I present a map depicting export product diversification across the globe to show that the export sector of Tajikistan is representative of developing countries. Since the map is based on Theil indices, higher values indicate lower export diversification. Thus, the cotton production response in Tajikistan can be attributed to the country’s low export diversification.
Typically, economic volatility should motivate internal and external migration (Kroeger & Anderson, 2014). As shown in Fig. 3, this is especially relevant for Tajikistan as this country is significantly dependent on migrant transfers. In certain years, the value of official remittance flows accounted for more than 40% of Tajikistan’s GDP. Moreover, according to Ivlevs et al. (2019), the relationship between remittances and subjective well-being is significant and positive. Thus, given the significance of remittances to the economy of Tajikistan, I also investigate whether remittances can mediate the well-being effects of export price shocks.
Based on the statistics on remittances, it might appear that Tajikistan’s economy is significantly different from that of other developing countries. In Fig. 4, I present two variables that might be important for the analysis of well-being and agricultural shocks to test the representatives of Tajikistan’s economy. I show that the income per capita in Tajikistan has been converging with that of its counterparts in other developing countries. More importantly, the share of agriculture-related activities in the economy of Tajikistan and developing countries are largely similar. If viewed together, Figs. 2 and 4 indicate that the analysis based on data from Tajikistan is not country-specific and should be generalizable to other contexts.
In a recent study, Danzer and Grundke (2020) investigate the labor market consequences of the 2010/11 cotton price increase in Tajikistan and demonstrate how the wages of Tajik cotton workers changed after this price shock. In the baseline estimations, the authors report that the average hourly wage of female agriculture workers in the cotton-producing regions increased by 62%. Since cotton picking is mainly performed by female workers in Tajikistan, they also find the effects on male workers statistically insignificant. Moreover, according to their estimations, the wages of non-agriculture workers in the cotton-producing regions were not affected.
Based on the results of Danzer and Grundke (2020), I empirically investigate the consumption patterns of Tajik households and show that the 2010/11 change in global cotton price can be viewed as a positive income shock for households living in cotton areas, without repercussions on households from other parts of the country. By comparing the well-being levels among these households before and after the shock, I demonstrate that, on average, even positive income fluctuations might lead to happiness losses. The results of further split sample analyses suggest that the effects of the cotton price increase are negative among households that were in agriculture before the shock and barely positive for newly become farmers. I also find that remittances positively alter the negative well-being effects of income fluctuations, but only within the financial domain. Finally, I demonstrate that age, gender, and education cannot be considered a significant mediating factor in the relationship between income and well-being. The magnitude of changes these variables induce is smaller than that of occupation and remittances.
The results of this study contribute to several strands of subjective well-being literature. The main result, which is that the net effects of the cotton price increase were negative, challenges the standard economic theory that happiness increases with higher average income (Cuong, 2021). This result also contributes to the scarce empirical literature devoted to identifying the effects of income on subjective well-being in developing economies. Both historically (see a review by Cummins, 2000) and more recently (D’Ambrosio et al., 2020; Distante, 2013; Kushlev et al., 2015), previous studies tend to use data from higher-income countries. However, on rarer occasions, there were attempts to examine the relationship between income and subjective well-being in the setting of lower-income countries (Cuong, 2021; Fuentes & Rojas, 2001; Knight & Gunatilaka, 2010; Kollamparambil, 2020). Based on the search results, this list appears to be short and does not include Tajikistan. Another strand of literature to be mentioned is the one based on cross-national data (Jebb et al., 2018; Sacks et al., 2012; Selezneva, 2011; Stevenson & Wolfers, 2013). These studies might be able to identify general trends but possibly fail to detect the mediating factors. Tajikistan and Central Asia in general should not be disregarded since it is one of the vulnerable regions, as identified by Collier (2007), where development assistance is needed.1 Thus, by robustly determining the negative consequences of export price fluctuations by occupational status in Tajikistan, this study provides new evidence for why less developed countries lag (Acemoglu & Robinson, 2012). Although the significance of occupation to subjective well-being has been previously demonstrated by several authors (see a review of 22 studies by Law et al., 1998), this study brings a new household-level perspective from a less researched setting. More generally, since I show that household decisions can mediate the effects of aggregate income shocks, I reinforce and partially explain the observation that the happiness effects of income vary across contexts (Graham, 2011). Finally, I contribute to the literature on remittances and subjective well-being by detecting the positive mediating effects of remittances on financial satisfaction (Ivlevs et al., 2019).
The rest of the paper proceeds as follows. Section 2 reviews the literature on well-being and income. Section 3 provides a brief description of the data and setting. Section 4 summarizes the empirical methods used. Section 5 presents the results of baseline estimations, several robustness tests, and heterogeneity analyses. Section 6 discusses possible transmission mechanisms. Section 7 concludes.
There have been numerous empirical attempts to investigate the well-being effects of income in general and income shocks in particular (Krueger et al., 2024; Stutzer, 2004). For a long time, the area of well-being research remained limited to psychology (Frey & Stutzer, 2002). Although psychological studies emphasize the role of personality traits in affecting the level of happiness, they might fail to robustly quantify the effects of external well-being determinants, such as income (Powdthavee, 2010). For instance, “personality bias” highlighted by psychologists might be driven by unobserved heterogeneity.
Starting from the seminal work of Easterlin (1974) on the non-trivial relationship between national income and well-being, economists have contributed to the field of happiness research by means of large-scale empirical analyses (Frey & Stutzer, 2002). However, economists typically have a bias toward a quantitative approach. Previous economic studies tend to concentrate on observable behavior and infer the well-being implications of choices made at individual and household levels. In addition to the well-known consumption patterns (Christelis et al., 2019), other utility determinants of this type, for example, comprise savings decisions (Berloffa & Modena, 2013) and labor-leisure trade-offs (Frey & Stutzer, 2002).
The standard “objectivist” position is not a sole theoretical foundation that can be used for an empirical specification. Another less conventional approach relates to self-reported judgments. In this case, well-being indicators are typically based on the survey answers of individuals (or a group of individuals living together) to general and specific questions about their lives and family matters. According to Durand and Smith (2013), it is possible to further decompose measures of subjective well-being across three dimensions: evaluative, hedonic, and eudaimonic. The former captures a reflective assessment of one’s life. In the second case, subjective well-being is viewed from the standpoint of affective states and emotional experiences. The latter is the most abstract dimension and encompasses opinions about life’s purpose, challenges, and growth.
Since it might be difficult to operationalize and estimate eudaimonic well-being, applied researchers are typically left with evaluative and hedonic well-being. Based on the empirical results of Graham and Nikolova (2015), it is possible to conclude that life evaluation is an appropriate measure of happiness as well as economic freedom. This well-being category should represent not only actual capabilities and means that individuals have but also perceived opportunities, allowing for better identification of poor and financially deprived people. Indeed, Kahneman and Deaton (2010) empirically demonstrate that income matters more for life evaluation than for emotional well-being (of US residents). Similarly, Tay and Diener (2011) demonstrate that global life evaluation patterns correlate mostly with fulfilling basic needs (for food and shelter), while positive and negative feelings around the world are associated more with non-economic needs.
Extensive literature has also emerged on the relationship between subjective well-being and income (Clark et al., 2008; Frey & Stutzer, 2002). Although several studies have found the positive effects of income on life satisfaction (Cuong, 2021; Powdthavee, 2010), this relationship is potentially more complex. For instance, a certain number of studies suggest that individuals tend to value income only in comparison to other people’s incomes (Ferrer-i-Carbonell, 2005; Powdthavee, 2010). This observation is nothing but new: even more than a hundred years ago, John Stuart Mill highlighted the significance of relative income over absolute income (Syrovátka, 2007). More recent studies demonstrate that the relative well-being effects of income might be caused by income aspirations (Stutzer, 2004) and/or depend on the individual’s ranked position within a reference group (Boyce et al., 2010). Alternatively, individuals might view income only as a tool to achieve a desired level of status and autonomy (Gardner & Oswald, 2007). In this case, they are concerned whether the income change is persistent or transitory (Bayer & Juessen, 2015).
More generally, the subjective well-being effects of aggregate income shocks might be heterogenous across endogenous and exogenous factors. In addition to intrinsic responses based on age and gender (Inglehart, 2002), they could also affect individuals by changing their economic environment. For instance, positive agricultural income shocks might lead to labor emigration flows (Bazzi, 2017). Reallocation might be not only geographic but also happen across occupations (Lee, 2020). Both these types of mobility might mitigate the initial effects of income shocks leading to differential economic outcomes. Finally, aggregate income shocks could also affect individuals via changing employment patterns in affected regions. Although Böckerman and Ilmakunnas (2009) show that the effects of becoming unemployed on self-assessed health are not significant in Finland, in the context of developing countries, where institutional support for the unemployed is absent, their effects might be significant and negative.
On a technical note, due to endogeneity problems caused by, for example, the absence of sufficient controls or measurement error, existing non-experimental studies might fail to investigate the relationship between income and subjective well-being beyond statistical correlation (Powdthavee, 2010). Conversely, it might be infeasible or ethical to conduct a large-scale experimental intervention where a significant number of individuals are given money on a continuous or even lump sum basis, and a comparable number of individuals are randomly allocated to a control group. Exogenous variation in the variables of interest can still be found by using natural and quasi experiments.
In this study, I modify the methodology of Danzer and Grundke (2020) and quantify the effects of income fluctuations using global commodity price spikes and geographic variation within a country where it is possible to grow this commodity. More specifically, I estimate the consequences of the 2010/11 cotton price increase in the classical (see Cunningham, 2021 for an overview) and triple (see Olden & Møen, 2022 for an overview) difference-in-differences (DD) frameworks. Since standard econometric methods are designed for continuous outcome variables, they might fail to capture the non-linear nature of subjective well-being measures. To account for this, I also consider the binary conditional logit estimator, based on the model of Chamberlain (1980), and the blowup and cluster estimator, introduced by Baetschmann et al. (2015) specifically for ordinal variables in the fixed effects (FE) model.

3 Data and Setting

I use data from the Tajikistan Living Standards Survey (TLSS) conducted by the World Bank and UNICEF in 2007 and 2009 and the Tajikistan Household Panel Survey (THPS) conducted by the Institute for East- and Southeast European Studies in 2011. Initially, 4,860 households were randomly selected to participate in a country-representative study. After 2 and 4 years, respectively, a random subsample of 1,503 and 1,392 households from the first survey were reinterviewed within subsequent studies. After merging survey waves, inspecting, modifying, and deleting data records, the final sample comprises a balanced panel of 1,305 households. The number of households is comparable with the recent study that uses the same data sources, where the investigation is based on a panel of 1,257 households (Gang et al., 2018).
Table 12 provides summary statistics for the sample under consideration. Overall, the selected variables are comparable across the years and progressed naturally. Household consumption notably increased from 2007 to 2011 in local currency, but the changes are modest if measured in USD.2 There was also a significant increase in the number of households receiving remittances, possibly due to the economic recovery after the 2007–2008 global economic crisis; concurrently, agricultural employment at the aggregate level remained relatively stable over the observed years despite the cotton price shock.
The territory of Tajikistan is covered by mountains, and it is possible to produce cotton only in certain areas. To construct a treatment variable, as in Danzer and Grundke (2020), I identify those areas at the community level prior to the cotton price shock. I concentrate on the 2007 TLSS community questionnaire and define a primary sampling unit as cotton-producing if cotton was reported to be the first or second major agricultural crop grown in this unit. Table 1 and Fig. 5 illustrate the current setting, which comprises 810 households in 104 cotton communities and 495 households in 63 non-cotton communities in each survey. In line with the descriptive statistics, there were indeed no changes in the sample composition of communities. More generally, cotton and non-cotton communities should not significantly differ in non-geographic characteristics as they are located close to each other.
Table 1
Distribution of the treatment variable
 
2007
2009
2011
Households living in non-cotton communities
495
495
495
Households living in cotton communities
810
810
810
For outcome variables, I consider two measures of subjective well-being based on the following questions: “How satisfied are you with your current financial situation?” and “Overall how satisfied are you with your life?”. The survey questions were asked at the household level based on the opinion of the most informed household member. Although not ideal, it is plausible to assume that these indicators should be highly correlated with the actual levels of household well-being. As shown in Table 2, the answers were recorded on a Likert scale but with slightly different formats across the waves. To enable comparison, I generate binary variables indicating households’ general and financial well-being. To do so, I first combine “rather satisfied/satisfied” and “very/fully satisfied” options for satisfaction. Then, I combine the rest of the options to capture dissatisfaction.3
Table 2
Distribution of outcome variables
2007
2009
2011
“Overall how satisfied are you with your life?”
Very unsatisfied
97
Very unsatisfied
59
Not at all satisfied
32
Unsatisfied
190
Unsatisfied
119
Unsatisfied
93
Neither unsatisfied nor satisfied
373
Neither unsatisfied nor satisfied
322
Neither unsatisfied nor satisfied
244
Satisfied
645
Satisfied
777
Satisfied
814
  
Very satisfied
28
Fully satisfied
122
“How satisfied are you with your current financial situation?”
Not at all satisfied
85
Not at all satisfied
83
Not at all satisfied
38
Less than satisfied
652
Less than satisfied
651
Unsatisfied
130
Rather satisfied
450
Rather satisfied
398
Neither unsatisfied nor satisfied
275
Fully satisfied
118
Fully satisfied
173
Satisfied
781
    
Fully satisfied
81
In Fig. 6, I plot the evolution of observed means for the newly created well-being variables from 2007 to 2011 by the treatment status. Although households both from cotton and non-cotton communities saw an increase in the average levels of subjective well-being, the growth curve for the treated group was much steeper. More generally, the upward trend in the selected well-being indicators (especially in the case of financial satisfaction) is in line with the findings of Guriev and Melnikov (2018) on the “happiness recovery” in post-communist countries after the 2007–2008 global economic crisis. To verify that the binary transformation of data does not distort actual subjective well-being patterns, I plot the evolution of the original ordinal variables in Fig. 7. Figures 6 and 7 resemble each other, and the conclusions regarding the well-being differences between communities before and after the 2010/11 cotton price increase remain qualitatively unchanged after binarization.
In Fig. 8, I additionally plot the evolution of agricultural employment (whether a household has at least one agricultural worker) and cross-country remittances (whether a household receives cash or in-kind international transfers over the year preceding the survey). In line with the nature of the shock, agricultural employment increased only in cotton areas. Concurrently, the patterns of remittances were similar across the country, possibly due to the short duration of the shock.

4 Empirical Strategy

I view households as being affected by income fluctuations if they live in a community that specializes in producing cotton. Identification at the community level should address, at least partially, the possibility of spillovers within communities. As the treatment is assumed to affect all households in sampling units either directly or indirectly, I cluster standard errors in the main specification at the community level as well. I define the timing of the treatment based on the world cotton price growth that occurred between July 2010 and March 2011. Since the 2011 THPS data were collected from October to December, this survey wave provides post-treatment data. Consequently, the 2007 TLSS and the 2009 TLSS surveys provide pre-treatment data. Based on the treatment definition and timing, I initially consider a standard 2 × 2 DD specification:
$${Y}_{ict}=\alpha +{\gamma }_{t}+{\beta Cotton}_{c}+\sigma \left({Cotton}_{c}\times {Post}_{t}\right)+{X}_{ict}{\prime}\theta +{\varepsilon }_{ict},$$
(1)
where \({Y}_{ict}\) is the well-being outcome for household \(i\) in community \(c\) at survey year \(t\). \({Cotton}_{c}\) indicates communities specializing at producing cotton and \({Post}_{t}\) captures the post-shock (2011) period. \(\alpha\) is the intercept, \({\gamma }_{t}\) is the set of survey-wave FE, and \({\varepsilon }_{ict}\) is the error term. \({X}_{ict}\) is the vector of time-varying controls: I account for the effects of household composition, location, and economic situation. I also include controls for the education of household members and the general characteristics of household heads.
As suggested by Borusyak et al. (2024), conventional DD designs without staggered rollout can be implemented with a two-way FE model. One of the benefits of this specification is that it can be estimated “dynamically” by replacing the single post-treatment variable with a set of period indicators. Below, I present static and dynamic versions of the DD specification based on two-way FE regressions:
$$\begin{aligned} Y_{{ict}} = & \alpha _{i} + \gamma _{t} + \sigma \left( {Cotton_{c} \times Post_{t} } \right) + X^{\prime } _{{ict}} \theta + \varepsilon _{{ict}} , \\ Y_{{ict}} = & \alpha _{i} + \gamma _{t} + \mathop \sum \limits_{{\begin{array}{*{20}c} {\tau = 2007} \\ {\tau \ne 2009} \\ \end{array} }}^{{2011}} \sigma _{\tau } Cotton_{c} + X^{\prime } _{{ict}} \theta + \varepsilon _{{ict}} , \\ \end{aligned}$$
(2)
where in addition to the previously defined variables, \({\alpha }_{i}\) captures the household FE. The indicator for the 2009 period is excluded as a normalization.
The identification in the DD framework relies on the parallel trends assumption, which I assume holds since the cotton price changes were caused by external environmental factors and the production capacities of Tajikistan are limited to affect the global cotton market. With the available data, this assumption can be tested (albeit not perfectly) by comparing the behavior of estimates for \(\tau\) = 2007 in the dynamic specification in Eq. 2. Given the short-term nature of the shock and the absence of inter-communal migration, endogenous selection can be ruled out. I also consider splitting the sample by agricultural employment and comparing a more similar group of households. In this case, estimations on the non-agricultural sample can be viewed as placebo tests. Furthermore, I include a variety of controls such that after conditioning on them, the DD estimator should retrieve the treatment effects of income fluctuations.
To investigate heterogeneity in the treatment effects of the export price shock, I consider a split sample analysis. As the shock happened in the agriculture sector, the first splitting variable is agriculture employment. To account for the possibility of occupational switching, I also divide the whole sample into those households that did not have agriculture workers during the observed period, those that had at least one agriculture worker, and those who moved between the sectors. To further determine the variables for heterogeneity analysis, I conduct balancing tests in Table 14. In addition to agriculture employment, the differences between treated and control samples changed in terms of average education and the number of adult female household members. Thus, I also estimate split sample regressions based on the age and gender of the household head and the average education of household members. To avoid confounding, I consider the values of these variables before the shock.
Given the significance of remittances to Tajik households, I also aim to determine whether remittances affect the relationship between income shock and subjective well-being. To do so, I exploit a triple difference-in-difference (DDD) specification by augmenting the communal treatment exposure with a household remittance status. Although remittances can be modeled as an income shock, the probability of receiving transfers from migrants is typically endogenously determined within households. Thus, I aim to quantify only the mediating effects of remittances induced by exogenous shock:
$$\begin{aligned} Y_{{ict}} = & \alpha _{i} + \gamma _{t} + \beta _{1} Remit_{{ict}} + \beta _{2} \left( {2007 \times Cotton_{c} } \right) + \beta _{3} \left( {2011 \times Cotton_{c} } \right) \\ & + \beta _{4} \left( {2007 \times Remit_{{ict}} } \right) + \beta _{5} \left( {2011 \times Remit_{{ict}} } \right) \\ & + \beta _{6} \left( {Remit_{{ict}} \times Cotton_{c} } \right) + \beta _{7} \left( {2007 \times Remit_{{ict}} \times Cotton_{c} } \right) \\ & + \beta _{8} \left( {2011 \times Remit_{{ict}} \times Cotton_{c} } \right) + X^{\prime } _{{ict}} \beta + \varepsilon _{{ict}} , \\ \end{aligned}$$
(3)
where additionally, \(2007\) and \(2011\) are the indicator variables for observations from the 2007 TLSS and the 2011 THPS, respectively, and \({Remit}_{ict}\) is the indicator variable for remittances at the household level. The parameters of interest are \({\beta }_{7}\) and \({\beta }_{8}\). They capture the well-being differences between households that receive remittances and live in cotton communities and their counterparts from non-cotton communities before and after the income shock. The estimates for \({\beta }_{2}\) and \({\beta }_{3}\) provide further insight into the underlying relationship by performing the same comparisons but for households without remittances.

5 Results

5.1 Descriptive Findings

Before proceeding to DD estimations, I first report the results of year-by-year Ordinary Least Squares (OLS) regressions. Figure 9 depicts unconditional estimates with 95% confidence intervals.4 The results indicate that cotton and non-cotton communities of Tajikistan were comparable in terms of subjective well-being in 2009 and 2011. The regional differences in well-being patterns became statistically significant only in 2011 when cotton-producing households became less satisfied with life and financial situation than their counterparts from non-cotton areas of the country. The net happiness losses are large in magnitude in both financial and general well-being domains. On a technical note, the confidence intervals of the estimates are comparable across subjective well-being measures and survey years, highlighting the stability of the estimations.
Table 15 presents the results of the OLS estimations with an extensive set of control variables, including total household consumption and location indicators. Baseline conclusions regarding the effects of income fluctuations remain unchanged after adding controls. In line with the standard economic theory, I also find that consumption is positively related to utility (Frey & Stutzer, 2002). The remaining estimates provide new correlational evidence for the U-shaped happiness effects of age (Frijters & Beatton, 2012) and positive subjective well-being effects of education (Yakovlev & Leguizamon, 2012) and remittances (Ivlevs et al., 2019). More generally, the results reinforce the importance of socio-demographic and economic variables in determining the quality of life (Isaeva & Salahodjaev, 2021; Salahodjaev & Ibragimova, 2020).

5.2 Main Results

Table 3 presents the results of baseline DD estimations. I first estimate a standard, 2 × 2 design with period FE. Then, I add time-varying controls and household FE. The results are stable between specifications and highlight the negative treatment effects of income fluctuations on subjective well-being. In the absence of the 2010/11 global cotton price change, general and financial satisfaction levels in cotton areas of Tajikistan would have been higher by nearly 11 and 13 percentage points, respectively. Based on average subjective well-being in non-cotton areas of the country in 2011, these values translate into happiness losses of about 14–17%. The treatment coefficients for life satisfaction are both qualitatively and quantitatively similar to the ones estimated for financial satisfaction. This implies that the income shock was sufficiently large to affect financial and non-financial domains of subjective well-being.
Table 3
Baseline DD estimations
 
(1)
(2)
(3)
(4)
(5)
(6)
Cotton × 2011
− 0.120**
− 0.117**
− 0.114**
− 0.130**
− 0.127**
− 0.129**
(0.048)
(0.049)
(0.047)
(0.051)
(0.053)
(0.051)
Survey-year FE
Yes
Yes
Yes
Yes
Yes
Yes
Household FE
No
No
Yes
No
No
Yes
Controls
No
Yes
Yes
No
Yes
Yes
Observations
3,915
3,915
3,915
3,915
3,915
3,915
Households
1,305
1,305
1,305
1,305
1,305
1,305
(Within) R2
0.042
0.093
0.077
0.051
0.122
0.094
Note: the outcome variable in Columns (1)-(3) is life satisfaction, and in Columns (4)-(6), it is financial satisfaction. R2 is reported for Columns (1), (2) and (4), (5); these estimations also include the treatment indicator. Within R2 is reported for Columns (3) and (6). Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
In Table 4, I present the results of additional DD estimations for the main outcome variables. The regression results are statistically significant and robust with respect to matching on control variables and the choice of a baseline period (the maximum difference between non-matched treatment regressions is less than 0.5 percentage points). The magnitude change in the matched estimations can be partially attributed to sample shrinkage. To achieve better matching (the results of which are presented in Table 16 and Figure 12), in addition to the inclusion of economic variables, I also trimmed the data by deleting 100 of the highest- and lowest-spending households in 2009. Some of the observations were further deleted in the process of matching iterations. Table 4 also indicates that there were no notable well-being differences between cotton and non-cotton sampling units before the income shock, supporting the identification assumption of DD estimations. Moreover, from 2007 to 2009, the growth rates of well-being indicators in cotton and non-cotton regions were statistically similar.
Table 4
Further DD estimations
 
(1)
(2)
(3)
(4)
(5)
(6)
Life satisfaction:
Cotton × 2007
− 0.002
− 0.014
    
(0.045)
(0.046)
    
Cotton × 2009
     
0.002
     
(0.045)
Cotton × 2011
− 0.121**
− 0.120**
− 0.119**
− 0.121**
− 0.177**
 
(0.054)
(0.054)
(0.052)
(0.054)
(0.072)
 
Household FE
Yes
Yes
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Yes
Yes
Controls
No
Yes
No
No
No
No
Matching
No
No
No
No
Yes
No
Survey years
2007–2011
2007–2011
2007, 2011
2009, 2011
2009, 2011
2007, 2009
Observations
3,915
3,915
2,610
2,610
2,148
2,610
Households
1,305
1,305
1,305
1,305
1,074
1,305
Within R2
0.060
0.077
0.112
0.032
0.032
0.033
Financial satisfaction:
      
Cotton × 2007
0.003
     
(0.042)
     
Cotton × 2009
     
− 0.003
     
(0.042)
Cotton × 2011
− 0.129**
− 0.132**
− 0.132**
− 0.129**
− 0.217***
 
(0.054)
(0.055)
(0.056)
(0.054)
(0.073)
 
Household FE
Yes
Yes
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Yes
Yes
Controls
No
Yes
No
No
No
No
Matching
No
No
No
No
Yes
No
Survey years
2007–2011
2007–2011
2007, 2011
2009, 2011
2009, 2011
2007, 2009
Observations
3,915
3,915
2,610
2,610
2,148
2,610
Households
1,305
1,305
1,305
1,305
1,074
1,305
Within R2
0.083
0.094
0.117
0.119
0.085
0.000
Note: clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
To further test the robustness of findings, I perform the Granger causality tests based on modified regressions without household FE for the 2007–2009 series. The results of joint tests for the pre-shock effects of living in cotton communities are insignificant: Prob. > F (2) = 0.71 for life satisfaction and Prob. > F (2) = 0.39 for financial satisfaction. It is thus possible to assume that sampled households did not change their behavior in anticipation of the treatment. However, the results of the Granger tests should not be interpreted as if the causal effects of living in cotton communities were absent (see a review by Shojaie & Fox, 2022). Because this test is based on many assumptions, including linearity and perfect observability, I instead can conclude that interactions between subjective well-being and producing cotton were minimal before the export shock. Concurrently, as noted by Clark and Granato (2005), Granger causality tests are tests for strong exogeneity. Therefore, the results can be interpreted as if living in cotton communities would not have predicted well-being patterns in the absence of the shock.
As a next robustness test, in Table 5, I consider a DD specification with region-specific linear time trends. The inclusion of trends instead of period FE should capture regional dynamics in subjective well-being measures by using the variation between cotton and non-cotton communities over time within each region of Tajikistan. As this speciation eliminates region-specific persistent unobserved factors, it should also address estimation bias across different contexts and over time. The results on the absence of pre-trends and the net negative effects of income fluctuations on both life and financial satisfaction are robust to the inclusion of region-specific linear time trends.
Table 5
Dynamic DD estimations with regional trends
 
(1)
(2)
(3)
(4)
Cotton × 2007
− 0.004
− 0.019
− 0.037
− 0.052
 
(0.051)
(0.052)
(0.044)
(0.044)
Cotton × 2011
− 0.141**
− 0.141**
− 0.126**
− 0.128**
 
(0.064)
(0.064)
(0.058)
(0.059)
Household FE
Yes
Yes
Yes
Yes
Region × Survey-year FE
Yes
Yes
Yes
Yes
Controls
No
Yes
No
Yes
Observations
3,915
3,915
3,915
3,915
Within R2
0.075
0.093
0.104
0.116
Note: the outcome variable in Columns (1) and (2) is life satisfaction, and in Columns (3) and (4), it is financial satisfaction. Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Finally, in Table 6, I demonstrate that the conclusions regarding the net negative effects of income fluctuations on subjective well-being are unaltered when non-linear estimation regressions with and without controls are considered. The odds of having a positive well-being status are 1 to 2 after being exposed to the income shock. A decrease in the probability of life and financial satisfaction in response to the shock are statistically significant and similar across the binary and ordinal logit estimations. In the pre-shock period, the odds ratios are statistically insignificant and nearly equal to 1. As in the linear regressions, this finding highlights the null effect of living in cotton communities on family well-being before the shock.
Table 6
Non-linear DD estimations
 
(1)
(2)
(3)
(4)
Cotton × 2007
1.007
1.014
0.972
0.969
 
(0.206)
(0.209)
(0.193)
(0.180)
Cotton × 2011
0.511**
0.495**
0.507**
0.484**
 
(0.146)
(0.150)
(0.146)
(0.164)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls, matched
No
No
No
No
Observations
2,667
2,565
5,073
6,294
Pseudo R2
0.084
0.119
0.108
0.460
Note: the results in Columns (1) and (2) are based on the binary logistic regression, and in Columns (3) and (4), they are based on the ordered logistic regression. The outcome variable in Columns (1) and (3) is life satisfaction, and in Columns (2) and (4), it is financial satisfaction. The changes in the number of observations are due to the omission of all positive and negative outcomes and the creation of clone observations. Coefficients are reported as odds ratios with clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1

5.3 Heterogeneity Analyses

To further explore the consequences of the cotton price increase, I perform several heterogeneity analyses. I first consider plausibly exogenous household characteristics. Figure 10 illustrates the results of DD regressions without controls disaggregated by household head’s gender and age, household members’ education level, and the number of female household members.5 The choice of variables for baseline heterogeneity analyses are driven by the results of balancing tests. To ensure that these variables are exogenous to the shock, I restrict the sample to the 2009–2011 period and consider the values of the variables as of 2009. Except for the household head’s gender, split-sample estimations are based on average values. Specifically, I consider a dichotomous division below and above the mean values. According to Fig. 10, the adverse effects of the cotton price increase do not considerably differ by household characteristics as the confidence intervals of the estimates overlap across subsamples. However, it should be noted that the treatment effects become statistically insignificant, even at a 10% level, for households with more female members and less educated households. Moreover, in the case of households with older heads, the effect is significant for life satisfaction and insignificant for financial satisfaction. The smaller sample size in these disaggregated estimations possibly causes the drop in statistical significance.
Given the nature of the shock, it should mainly affect households working in the agriculture sector. Tables 7 and 21 present the results of DD regressions disaggregated by household occupational status without and with controls, respectively. To show that the effects emerged after the shock, I consider the full sample for the 2007–2011 period. According to the estimations, the well-being differences between non-agricultural households in cotton and non-cotton regions of Tajikistan were absent both before and after the 2010/11 income shock. Concurrently, in the sample of agricultural households, as in the baseline estimations, the well-being differences were statistically insignificant before the shock and became significant and skewed in favor of households from non-cotton communities after the shock.
Table 7
Split-sample DD estimations: agriculture vs. non-agriculture sectors
 
(1)
(2)
(3)
(4)
Cotton × 2007
− 0.061
− 0.170
− 0.007
0.033
(0.138)
(0.146)
(0.049)
(0.046)
Cotton × 2011
− 0.247**
− 0.271*
− 0.086
− 0.090
(0.120)
(0.147)
(0.058)
(0.062)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls, matched
No
No
No
No
Observations
713
713
3,202
3,202
Households
470
470
1,257
1,257
Within R2
0.091
0.135
0.052
0.065
Note: the results in Columns (1) and (2) are based on the agriculture sample, and in Columns (3) and (4), they are based on the non-agriculture sample. The outcome variable in Columns (1) and (3) is life satisfaction, and in Columns (2) and (4), it is financial satisfaction. Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
I present the results of DDD estimations with and without controls for the mediating effects of remittances in Table 8. Additional terms included to capture the remittances-receiving status of households do not significantly alter the effects of income fluctuations on life satisfaction. However, in the case of financial well-being, the treatment effects of income fluctuations are negative only on those households that live in cotton communities and do not receive remittances, but households in the same communities with remittances are well-off after the shock. Moreover, in 2007, the effects of remittances on both general and financial satisfaction in cotton communities were absent. Thus, it is possible to hypothesize that the meditating effects of remittances are driven by the shock and not caused by omitted variable bias. These findings possibly indicate that remittances can provide financial protection against agriculture-related income shocks, but their mediating effects are limited to the financial domain. It is possible to further hypothesize that happiness losses from income fluctuations could be driven by empathy for neighbors or anticipation of negative socio-economic transformations in respective communities (e.g., inequalities or unemployment) that could not be solved at the household level with extra monetary or in-kind transfers.
Table 8
DDD estimations: export shock and remittances
 
(1)
(2)
(3)
(4)
Cotton × 2007
− 0.005
− 0.017
− 0.010
− 0.019
(0.048)
(0.047)
(0.046)
(0.045)
Cotton × 2011
− 0.132**
− 0.132**
− 0.187***
− 0.189***
(0.063)
(0.062)
(0.058)
(0.057)
Remittances
0.065
0.087
0.143**
0.152**
(0.073)
(0.072)
(0.066)
(0.065)
Remittances × 2007
− 0.059
− 0.052
− 0.047
− 0.034
(0.093)
(0.092)
(0.096)
(0.096)
Remittances × 2011
− 0.047
− 0.037
− 0.195**
− 0.180**
(0.074)
(0.077)
(0.089)
(0.089)
Cotton × Remittances
− 0.031
− 0.044
− 0.168*
− 0.176*
(0.101)
(0.100)
(0.096)
(0.095)
Cotton × Remittances × 2007
− 0.005
0.008
0.111
0.114
(0.136)
(0.135)
(0.125)
(0.125)
Cotton × Remittances × 2011
0.049
0.069
0.303**
0.306**
(0.112)
(0.113)
(0.125)
(0.124)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
No
Yes
No
Yes
Observations
2,610
2,610
2,610
2,610
Households
1,305
1,305
1,305
1,305
Within R2
0.061
0.077
0.087
0.097
Note: the outcome variable in Columns (1) and (2) is life satisfaction, and in Columns (3) and (4), it is financial satisfaction. Clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1

6 Mechanisms

In this section, I explore why the net effects of the 2010 cotton shock on both life and financial satisfaction are negative. To do so, I consider the changes at a more granular household level. Consequently, I cluster standard errors at the household level as well. In Table 9, I restrict the sample to the 2009–2011 period for more meaningful comparisons and use consumption, agricultural employment, and remittances as outcome variables.
Table 9
Baseline analysis of mechanisms
 
(1)
(2)
(3)
(4)
(5)
(6)
Cotton × 2011
41.893**
52.736***
0.048*
0.047*
0.001
-0.003
(17.646)
(19.149)
(0.027)
(0.027)
(0.028)
(0.025)
Household FE
Yes
Yes
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Yes
Yes
Controls
No
Yes
No
Yes
No
Yes
Observations
2,610
2,610
2,610
2,610
2,610
2,610
Households
1,305
1,305
1,305
1,305
1,305
1,305
Within R2
0.099
0.133
0.003
0.027
0.064
0.285
Note: the outcome variable in Columns (1) and (2) is expenditures per household member; in Columns (3) and (4), it is agricultural employment; and in Columns (5) and (6), it is receiving remittances. Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
It could be the case that the 2010/11 export price fluctuations did not affect the material well-being in cotton-producing communities. The estimations in Columns (1) and (2) of Table 9, with and without time-varying controls, indicate that monthly spending among households living in cotton communities increased relatively to households living in non-cotton communities by approximately TJS 50 (USD 10.5) per household member after the shock. Considering the mean value of TJS 262 for this variable, the change is economically notable. Given the statistically significant increase in per capita consumption in 2011 and the results of the Granger causality tests, it is possible to view the 2010/11 export price increase as a positive income shock. Moreover, in Table 22, I show that before the shock, the effect of consumption per capita on subjective well-being was positive, and after the shock, it became statistically insignificant. This result suggests that the net negative effects of the shock are not driven by consumption patterns or reverse causality between consumption and subjective well-being.
The observed wealth differences could motivate households living in cotton communities to change their occupations and specialize in agriculture-related jobs. Indeed, Columns (3) and (4) of Table 9 indicate that the export price fluctuations led to occupational sorting in cotton communities. The probability of having at least one agriculture worker in the household increased by nearly 5 percentage points after the shock. Similarly, migrants who were in contact with their households could have decided to stop sending remittances in response to increased income. According to Columns (5) and (6), this was not the case: the probability of receiving remittances in cotton and non-cotton communities was comparable after the shock. Overall, in line with Fig. 8, the results in Columns (3)–(6) indicate that receiving remittances is possibly exogenous with respect to export shocks, while agriculture-based occupational sorting is possibly endogenous. Thus, split-sample regressions based on involvement in agriculture might be biased due to sample selection. To address the possibility of bias, I propose to explicitly consider the possibility of occupational switching in the DD estimations.
Since the effects of the cotton price increase are observed in the agricultural subsample and the shock affected the probability of working in the agricultural sector, I disaggregate the effects of the export price fluctuations by agriculture-related occupational switching in Table 10. I interact the main treatment variable with 3 indicator variables describing households that were in agriculture in 2009 and 2011, that switched to agriculture in 2011, and that switched from agriculture in 2011. The reference group in these regressions are households that were in non-agriculture occupations both in 2009 and 2011. The results indicate that the negative well-being effects of income fluctuations are driven by non-switching agricultural workers who faced increased competition within cotton communities. Those who switched to the agriculture sector could benefit from the shock, but mainly in financial terms. The limited effects of the export price increase among switchers could be explained by its short-term nature. Finally, the well-being changes among those who switched away from agriculture are statistically insignificant, possibly because this switching was not related to the shock.
Table 10
Further analysis of mechanisms: agriculture vs. non-agriculture occupations
 
(1)
(2)
(3)
(4)
Cotton × Non-switching farmers × 2011
− 0.188*
− 0.196*
− 0.176*
− 0.193**
(0.103)
(0.104)
(0.093)
(0.091)
Cotton × Switched to agriculture × 2011
0.044
0.075
0.078
0.104*
(0.066)
(0.066)
(0.063)
(0.062)
Cotton × Switched from agriculture × 2011
− 0.109
− 0.094
− 0.107
− 0.096
(0.088)
(0.088)
(0.074)
(0.073)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
No
Yes
No
Yes
Observations
2,610
2,610
2,610
2,610
Households
1,305
1,305
1,305
1,305
Within R2
0.028
0.053
0.115
0.136
Note: the outcome variable in Columns (1) and (2) is life satisfaction, and in Columns (3) and (4), it is financial satisfaction. Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
Finally, I investigate whether the mediating effects of remittances on financial satisfaction are driven by the income shock or omitted variable bias. To do this, I year-by-year regress financial satisfaction on remittances separately in cotton and non-cotton communities in Table 11. As these are OLS regressions, I include controls to capture general differences between communities. The estimations indicate that in the non-cotton sample, the effects of remittances were positive and statistically significant before and after the shock. The magnitude of the effects even decreased after the shock. However, in the cotton sample, the effects of remittances became positive and statistically significant only in 2011, highlighting the robustness of DDD estimations. In 2009, the point estimate for the remittance variable in the cotton subsample was even negative.
Table 11
Further analysis of mechanisms: remittances and financial satisfaction
 
(1)
(2)
(3)
(4)
(5)
(6)
 
2007
2009
2011
2007
2009
2011
Receive remittances
0.006
− 0.026
0.097**
0.159**
0.157**
0.081*
(0.055)
(0.059)
(0.047)
(0.062)
(0.062)
(0.044)
Controls
Yes
Yes
Yes
Yes
Yes
Yes
Observations
810
810
810
495
495
495
R2
0.085
0.129
0.099
0.121
0.152
0.212
Note: the results in Columns (1)-(3) are based on the cotton sample, and in Columns (4)-(6), they are based on the non-cotton sample. The outcome variable in Columns (1)-(6) is financial satisfaction. Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1

7 Concluding Remarks

In this study, I investigate how income fluctuations caused by environmental factors affect agricultural regions. More specifically, I empirically examine the consequences of the unexpected 2010/11 global cotton price increase on the welfare of households from Tajikistan. This shock resulted in an increase in the global price of cotton and affected the income of its exporters and importers. Tajikistan is a small, landlocked country that specializes in exporting cotton. Tajikistan’s economy has recently advanced from low-income to lower-middle-income status. In addition to income level, the export and agricultural sectors of Tajikistan are representative of developing countries. Considering the low levels of export diversification in and market power of developing countries and Tajikistan, the results of this study should be externally valid for other export-related agricultural shocks because the type of income shock and the well-being changes investigated in this study can and possibly did happen in other developing countries.
Since the relationship between income and quality of life has been previously investigated without reaching a consensus, the contribution of this study is that it brings to the subjective well-being literature new household-level evidence on the effects of a less researched type of income shock. Specifically, I show that the net effects of exposure to agriculture-related income fluctuations on the subjective well-being of families are negative. This result is in contradiction with the view that transitory shocks do not have a significant impact on happiness. On the contrary, I demonstrate that households living in the communities affected by export price volatility, on average, are likely to become significantly dissatisfied with their lives and financial situation. Based on consumption patterns, I further demonstrate that the 2010/11 increase in the global price of cotton can be viewed as a positive income shock for Tajik cotton producers at the community level. Therefore, it is important to monitor the subjective well-being of households after aggregate income shocks, including development assistance programs, even if objective well-being measures indicate positive dynamics. Although this policy recommendation applies to different types of non-export shocks, it is more relevant for other countries with lower income levels and where the variety of consumption goods and services is limited.
To explore possible reasons behind the net negative effects of the cotton price increase, I estimate split sample regressions. I first consider plausibly exogenous mediators. Age, gender, and completed education do not significantly affect the treatment effects of the shock. Since the effects of the cotton price increase should be mainly present within the agricultural sector, I then consider occupational choice. I first demonstrate that the shock made the agriculture sector more attractive and induced households in cotton communities to start working in the agriculture sector. After determining that the cotton price increase mainly affected agricultural households, I show that the effects of the cotton price increase are negative among households that were in agriculture before the shock and barely positive for newly become farmers. Based on these results, it is possible to hypothesize that export price fluctuations change the labor market and lead to occupational sorting. Thus, one of the ways commodity-related income fluctuations affect subjective well-being in the regions specializing in export agriculture could be increased competition. As I document well-being patterns when the price of cotton returned to its pre-shock level, the farmers in cotton communities might have been dissatisfied because they faced increased competition even after the shock. To sum up, when examining heterogeneity in the effects of income shocks, it is important to consider not only standard variables but also the nature of the income shock.
Finally, I find that households can protect themselves against the negative effects of income volatility via remittances; however, the mediating effects of migration-based income loss compensation are limited to financial satisfaction and do not cover life satisfaction. Therefore, policy interventions aimed at improving the well-being of families should be comprehensive and go beyond mere income compensation. Moreover, this finding highlights structural differences between subjective well-being domains. A conventional separation between evaluative, hedonic, and eudaimonic well-being might be insufficient, and even within these disaggregated categories, a distinction between general and financial satisfaction should ideally be made. In practical terms, a better categorization of well-being measures could result in the policy evaluations of post-aggerate-shock interventions being more comprehensive. This is especially relevant in light of recent monetary compensation schemes that were adopted to mitigate the consequences of the COVID-19 pandemic.

8 Appendix

See Tables 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, and 22.
See Figs. 11 and 12.
Table 12
Year-by-year descriptive statistics
  
2007
2009
2011
Obs.
Mean (St. dev.)
Mean (St. dev.)
Mean (St. dev.)
Household (HH) composition:
    
Number of children below the age of 6
1,305
0.76 (1.02)
0.88 (1.14)
0.89 (1.20)
Number of children aged between 6 and 15
1,305
1.57 (1.36)
1.47 (1.33)
1.32 (1.33)
Number of adults aged between 16 and 65
1,305
3.77 (1.93)
4.16 (2.07)
3.93 (2.01)
Number of elderly over the age of 65
1,305
0.26 (0.54)
0.26 (0.53)
0.27 (0.54)
Number of female adults aged between 16 and 65
1,305
2.02 (1.15)
2.17 (1.20)
2.10 (1.17)
Educational attainment of HH members:
  
Average education (0 if None, …, 7 if tertiary education)
1,305
2.55 (0.90)
2.62 (0.90)
2.67 (0.95)
Number of HH members with tertiary education
1,305
0.39 (0.76)
0.42 (0.80)
0.44 (0.83)
HH head characteristics:
Age
1,305
51.69 (13.32)
52.92 (12.83)
54.52 (12.85)
Ethnicity (1 if Tajik, 0 otherwise)
1,305
0.77 (0.42)
0.76 (0.43)
0.78 (0.42)
Gender (1 is male, 0 otherwise)
1,305
0.81 (0.39)
0.83 (0.38)
0.74 (0.44)
Marital status (1 is married, 0 otherwise)
1,305
0.81 (0.40)
0.80 (0.40)
0.77 (0.42)
HH location:
Urban (1 if yes, 0 otherwise)
1,305
0.34 (0.47)
0.34 (0.47)
0.33 (0.47)
Districts of Republican Subordination (1 if yes, 0 otherwise)
1,305
0.21 (0.41)
0.21 (0.41)
0.21 (0.41)
Dushanbe (1 if yes, 0 otherwise)
1,305
0.16 (0.37)
0.16 (0.37)
0.16 (0.37)
Gorno-Badakhshan region (1 if yes, 0 otherwise)
1,305
0.10 (0.30)
0.10 (0.30)
0.10 (0.30)
Khatlon region (1 if yes, 0 otherwise)
1,305
0.26 (0.44)
0.26 (0.44)
0.26 (0.44)
Sughd region (1 if yes, 0 otherwise)
1,305
0.26 (0.44)
0.26 (0.44)
0.26 (0.44)
Economic situation of HH:
In agriculture (1 if a HH member works in agriculture, 0 otherwise)
1,305
0.20 (0.40)
0.17 (0.37)
0.18 (0.38)
Remittances (1 if receive, 0 otherwise)
1,305
0.14 (0.35)
0.12 (0.32)
0.24 (0.43)
Total HH expenditure (in TJS)
1,305
996.42 (901.03)
1479.94 (935.04)
2105.03 (2184.76)
Total HH expenditure (in USD)
1,305
291.88 (263.03)
371.30 (234.59)
442.61 (459.37)
Treatment and outcome variables:
Cotton community (1 if yes, 0 otherwise)
1,305
0.62 (0.49)
0.62 (0.49)
0.62 (0.49)
Financial satisfaction (1 if satisfied)
1,305
0.44 (0.50)
0.44 (0.50)
0.66 (0.47)
Life satisfaction (1 if satisfied)
1,305
0.49 (0.50)
0.62(0.49)
0.72 (0.45)
Table 13
Binary outcome variables
2007
2009
2011
Life satisfaction
 
Life satisfaction
 
Life satisfaction
 
Unsatisfied
660
Unsatisfied
500
Unsatisfied
369
Satisfied
645
Satisfied
805
Satisfied
936
Financial satisfaction
 
Financial satisfaction
 
Financial satisfaction
 
Unsatisfied
737
Unsatisfied
734
Unsatisfied
443
Satisfied
568
Satisfied
571
Satisfied
862
Table 14
Baseline balancing tests
 
Non-cotton
Cotton
Diff.
Non-cotton
Cotton
Diff.
 
Before the shock
After the shock
Household composition:
      
Number of children (< 6)
0.809
0.825
− 0.016
0.901
0.888
0.013
Number of children (6–15)
1.544
1.502
0.042
1.333
1.310
0.023
Number of adults (16–65)
4.189
3.833
0.356***
4.071
3.841
0.230**
Number of elderly (> 65)
0.279
0.253
0.026
0.293
0.259
0.034
Number of female adults
2.178
2.046
0.132***
2.119
2.093
0.027
Education of household members:
   
Average education (0 = None, … 7 = Higher education)
2.473
2.648
− 0.174***
2.648
2.691
− 0.043
Number of members with tertiary education
0.358
0.433
− 0.075**
0.380
0.482
− 0.102**
Household head characteristics:
   
Age
53.048
51.846
1.201**
55.449
53.948
1.500**
Ethnicity (1 = Tajik)
0.809
0.736
0.073***
0.808
0.756
0.053**
Gender (1 = Male)
0.850
0.801
0.048***
0.780
0.721
0.059
Marital status (1 = Married)
0.844
0.773
0.072***
0.806
0.749
0.057**
Household location:
      
1 = Urban
0.143
0.457
− 0.313***
0.129
0.447
− 0.318***
1 = DRS
0.325
0.136
0.190***
0.325
0.136
0.190***
1 = Dushanbe
0
0.263
0.263***
0
0.263
0.263***
1 = Gorno-Badakhshan
0.234
0.020
0.215***
0.234
0.020
0.215***
1 = Khatlon
0.162
0.327
− 0.166
0.162
0.327
− 0.166
1 = Sughd
0.279
0.254
0.024
0.279
0.254
0.024
Household economic characteristics:
   
Number of employed members
1.662
1.817
− 0.155***
1.434
1.759
− 0.325***
1 = agricultural worker
0.226
0.157
0.069***
0.198
0.168
0.030
Consumption per member, in Somoni
212.408
211.632
0.777
344.021
371.605
− 27.584
1 = receive remittances
0.171
0.103
0.068***
0.289
0.212
0.077***
Note: the estimations are based on two-way mean comparisons; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 15
Year-by-year OLS estimations with controls
 
(1)
(2)
(3)
(4)
(5)
(6)
 
Life satisfaction
Financial satisfaction
 
2007
2009
2011
2007
2009
2011
Cotton community
0.002
0.027
− 0.133***
− 0.058
0.003
− 0.126**
(0.034)
(0.040)
(0.047)
(0.036)
(0.040)
(0.049)
Household composition:
Number of children
0.015
− 0.013
− 0.009
0.012
− 0.030**
− 0.011
(0.013)
(0.013)
(0.015)
(0.011)
(0.013)
(0.016)
Number of adults
− 0.010
0.027**
− 0.036***
0.001
− 0.023*
− 0.034***
(0.016)
(0.013)
(0.013)
(0.016)
(0.013)
(0.013)
Number of elderly
− 0.052
− 0.031
− 0.005
− 0.007
0.068
0.005
(0.035)
(0.041)
(0.035)
(0.034)
(0.045)
(0.035)
Number of female adults
− 0.024
− 0.068***
0.007
− 0.041*
− 0.012
0.010
(0.024)
(0.021)
(0.021)
(0.021)
(0.019)
(0.021)
Educational attainment of household members:
Average education
0.057**
0.001
0.055**
0.049**
0.032
0.032
(0.023)
(0.022)
(0.025)
(0.019)
(0.024)
(0.026)
Number of people with tertiary education
0.004
0.023
0.013
0.012
0.043**
0.038*
(0.026)
(0.020)
(0.021)
(0.024)
(0.020)
(0.020)
Household location:
Urban
− 0.104**
− 0.006
− 0.013
− 0.021
0.039
− 0.040
(0.043)
(0.038)
(0.069)
(0.041)
(0.041)
(0.075)
Districts of Republican Subordination
0.003
− 0.080
− 0.036
− 0.163***
− 0.039
− 0.084
(0.062)
(0.059)
(0.093)
(0.059)
(0.057)
(0.090)
GBA Region
− 0.037
0.120*
0.163*
− 0.166**
0.076
0.216**
(0.077)
(0.065)
(0.093)
(0.073)
(0.072)
(0.108)
Khatlon Region
− 0.051
− 0.152***
0.094
− 0.088
− 0.032
0.185*
(0.060)
(0.057)
(0.098)
(0.054)
(0.058)
(0.095)
Sughd Region
0.135***
− 0.047
0.057
− 0.043
0.124**
0.158*
(0.046)
(0.055)
(0.080)
(0.058)
(0.052)
(0.082)
Household head characteristics:
Age
− 0.015*
− 0.016**
− 0.012*
− 0.005
0.003
− 0.007
(0.008)
(0.008)
(0.006)
(0.008)
(0.009)
(0.007)
Age2
0.000*
0.000*
0.000
0.000
− 0.000
0.000
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
Male
0.055
− 0.010
0.038
0.023
− 0.007
0.029
(0.064)
(0.060)
(0.043)
(0.061)
(0.061)
(0.045)
Married
0.019
0.050
− 0.008
0.092
0.051
− 0.008
(0.065)
(0.057)
(0.049)
(0.062)
(0.063)
(0.054)
Tajik
0.083**
0.019
0.081*
− 0.000
− 0.038
0.055
(0.035)
(0.038)
(0.045)
(0.032)
(0.040)
(0.048)
Household economic characteristics:
Number of employed people
0.007
0.012
0.022*
0.017
0.035***
0.012
(0.012)
(0.012)
(0.012)
(0.014)
(0.013)
(0.013)
In agriculture
− 0.022
0.022
0.019
− 0.085**
0.011
− 0.008
(0.042)
(0.039)
(0.037)
(0.033)
(0.040)
(0.041)
Receive remittances
0.044
0.041
0.106***
0.080**
0.074*
0.084*
(0.044)
(0.044)
(0.039)
(0.037)
(0.044)
(0.043)
ln(Total household expenditure)
0.128***
0.115***
0.114***
0.153***
0.216***
0.104***
(0.025)
(0.027)
(0.035)
(0.031)
(0.026)
(0.034)
Observations
1,305
1,305
1,305
1,305
1,305
1,305
R2
0.072
0.064
0.111
0.084
0.120
0.116
Note: the outcome variable in Columns (1)-(3) is life satisfaction, and in Columns (4)-(6), it is financial satisfaction. The reference categories for household composition and location are the number of children below the age of 6 and Dushanbe (capital city), respectively. Clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 16
Balancing tests before and after Kernel Propensity Score Matching
 
Non-cotton
Cotton
Diff.
Non-cotton
Cotton
Diff.
 
Unmatched
Matched
Household composition:
Number of children (< 6)
Reference category
Reference category
Number of children (6–15)
1.600
1.471
− 0.130
1.454
1.471
0.017
Number of adults (16–65)
4.442
4.078
− 0.364**
4.250
4.078
− 0.173
Number of elderly (> 65)
0.277
0.227
− 0.049
0.231
0.227
− 0.004
Number of female adults
2.296
2.158
− 0.137
2.268
2.158
− 0.110
Education of household members:
   
Average education (0 = None, … 7 = Higher education)
2.482
2.658
0.176**
2.591
2.658
0.067
Number of members with tertiary education
0.326
0.462
0.136**
0.417
0.462
0.044
Household head characteristics:
   
Age
53.277
51.641
− 1.636*
52.505
51.641
− 0.864
Ethnicity (1 = Tajik)
0.809
0.748
− 0.061
0.689
0.748
0.059
Gender (1 = Male)
0.872
0.830
− 0.042**
0.837
0.830
− 0.007
Marital status (1 = Married)
0.849
0.795
− 0.054**
0.801
0.795
− 0.006
Household location:
1 = Urban
0.135
0.434
0.299***
0.294
0.434
0.140
1 = DRS
0.314
0.141
− 0.174**
0.220
0.141
− 0.080
1 = Dushanbe
Reference category
Reference category
1 = Gorno-Badakhshan
0.246
0.019
− 0.227***
0.011
0.019
0.008
1 = Khatlon
0.156
0.340
0.184***
0.417
0.340
− 0.076
1 = Sughd
0.284
0.248
− 0.036
0.352
0.248
− 0.104
Household economic characteristics:
Number of employed members
1.586
1.864
0.277**
2.005
1.864
− 0.142
ln(Consumption per HH member, in Somoni)
5.284
5.304
0.020
5.301
5.304
0.002
Own land used for farming, in 100m2
23.728
10.877
− 12.851***
17.981
10.877
− 7.104
Rented land used for farming, in 100m2
9.402
4.443
− 4.959
8.523
4.443
− 4.080
Note: the estimations are based on two-way mean comparisons and propensity scores are derived from Logit estimation; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 17
Split-sample DD estimations with controls: life satisfaction I
 
(1)
(2)
(3)
(4)
Household head:
Older
Younger
Female
Male
Cotton × 2011
− 0.144**
− 0.099
− 0.219**
− 0.104*
(0.066)
(0.063)
(0.092)
(0.058)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
Yes
Yes
Yes
Yes
Observations
1,194
1,416
448
2,162
Households
597
708
224
1,081
Within R2
0.050
0.089
0.165
0.058
Note: clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Table 18
Split-sample DD estimations with controls: life satisfaction II
 
(1)
(2)
(3)
(4)
Household members:
More women
Fewer women
More educated
Less educated
Cotton × 2011
− 0.060
− 0.164***
− 0.154**
− 0.104
(0.077)
(0.053)
(0.060)
(0.073)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
Yes
Yes
Yes
Yes
Observations
912
1,628
1,725
1,440
Households
456
849
585
720
Within R2
0.079
0.084
0.069
0.063
Note: clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 19
Split-sample DD estimations with controls: financial satisfaction I
 
(1)
(2)
(3)
(4)
Household head:
Older
Younger
Female
Male
Cotton × 2011
− 0.116*
− 0.157**
− 0.233**
− 0.129**
(0.065)
(0.064)
(0.099)
(0.055)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
Yes
Yes
Yes
Yes
Observations
1,194
1,416
448
2,162
Households
597
708
224
1,081
Within R2
0.150
0.158
0.225
0.138
Note: clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 20
Split-sample DD estimations with controls: financial satisfaction II
 
(1)
(2)
(3)
(4)
Household members:
More women
Fewer women
More educated
Less educated
Cotton × 2011
− 0.107
− 0.158***
− 0.198***
− 0.096
(0.074)
(0.058)
(0.061)
(0.073)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
Yes
Yes
Yes
Yes
Observations
912
1,698
1,170
1,440
Households
456
849
585
720
Within R2
0.120
0.182
0.132
0.166
Note: clustered standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1
Table 21
Agriculture-based split-sample DD estimations with controls
 
(1)
(2)
(3)
(4)
Cotton × 2007
− 0.098
− 0.179
− 0.016
0.023
(0.131)
(0.148)
(0.049)
(0.045)
Cotton × 2011
− 0.224*
− 0.267*
− 0.085
− 0.095
(0.118)
(0.140)
(0.059)
(0.063)
Household FE
Yes
Yes
Yes
Yes
Survey-year FE
Yes
Yes
Yes
Yes
Controls
Yes
Yes
Yes
Yes
Observations
713
713
3,202
3,202
Households
470
470
1,257
1,257
Within R2
0.135
0.159
0.072
0.077
Note: the results in Columns (1) and (2) are based on the agriculture sample, and in Columns (3) and (4), they are based on the non-agriculture sample. The outcome variable in Columns (1) and (3) is life satisfaction, and in Columns (2) and (4), it is financial satisfaction. Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Table 22
Further analysis of mechanisms: consumption and well-being
 
(1)
(2)
(3)
(4)
(5)
(6)
2007
2009
2011
2007
2009
2011
Expenditures per member
0.128***
0.073**
0.030
0.129***
0.174***
0.029
(0.026)
(0.033)
(0.053)
(0.044)
(0.033)
(0.046)
Controls
Yes
Yes
Yes
Yes
Yes
Yes
Observations
1,305
1,305
1,305
1,305
1,305
1,305
R2
0.063
0.055
0.084
0.064
0.101
0.094
Note: the outcome variable in Columns (1)-(3) is life satisfaction, and in Columns (4)-(6), it is financial satisfaction. The reported estimates are based on standardized outcome and treatment variables for a better presentation of results. Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1

Acknowledgements

I would like to thank the editor, Stephanie Rossouw, the associate editor, Fengyu Wu, and two anonymous referees for helpful comments and suggestions. I am grateful to Dušan Drbohlav and Vasily Korovkin for their guidance and support. I would also like to thank the participants of the 33rd Annual Conference of the International Trade and Finance Association and the 15th Joint IOS/APB/EACES Summer Academy. The parts of this study were developed at the time of my research stay at the Center on Global Economy and Governance of Columbia University, hosted by Jan Švejnar, and the Griswold Center for Economic Policy Studies of Princeton University, hosted by Dana Molina. An earlier version of this article is circulated as IOS Working Papers No. 400, handled by Vladimir Kozlov.

Declarations

Conflict of interest

I state that no conflict of interest exists in this study.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creativecommons.​org/​licenses/​by/​4.​0/​.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Fußnoten
1
As of 2021, GDP per capita in Tajikistan was (current) USD 897.1. Concurrently, 26.3% of the country’s population lived below the national poverty line in 2019.
 
2
In both cases, the consumption variables are expressed in current prices.
 
3
I present the distribution of the newly created binary variables in Table 13.
 
4
To show that the results are not driven by sample selection, I explore subjective well-being dynamics between cotton and non-cotton communities using the original non-panel samples in Fig. 11. For the year 2007, the sample comprises 4,860 households. For the years 2009 and 2011, the number of households is 1,503. The conclusions regarding the effects of exposure to the shock remain unchanged even in the larger samples. Moreover, these estimations also show that the well-being differences between communities were absent before the shock.
 
5
I report the results of regressions with controls in Tables 17, 18, 19, and 20.
 
Literatur
Zurück zum Zitat Acemoglu, D., & Robinson, J. A. (2012). Why nations fail: The origins of power, prosperity, and poverty. Crown Business. Acemoglu, D., & Robinson, J. A. (2012). Why nations fail: The origins of power, prosperity, and poverty. Crown Business.
Zurück zum Zitat Clark, H. D., & Granato, J. (2005). Time series analysis in political science. In K. Kempf-Leonard (Ed.), Encyclopedia of Social Measurement (pp. 829–837). Elsevier.CrossRef Clark, H. D., & Granato, J. (2005). Time series analysis in political science. In K. Kempf-Leonard (Ed.), Encyclopedia of Social Measurement (pp. 829–837). Elsevier.CrossRef
Zurück zum Zitat Collier, P. (2007). The bottom billion: Why the poorest countries are failing and what can be done about it. Oxford University Press. Collier, P. (2007). The bottom billion: Why the poorest countries are failing and what can be done about it. Oxford University Press.
Zurück zum Zitat Cunningham, S. (2021). Causal inference: The mixtape. Yale University Press. Cunningham, S. (2021). Causal inference: The mixtape. Yale University Press.
Zurück zum Zitat Durand, M., & Smith, C. (2013). The OECD approach to measuring subjective well-being. In J. F. Helliwell, R. Layard, & J. Sachs (Eds.), World Happiness Report 2013 (pp. 112–137). UN Sustainable Development Solutions Network. Durand, M., & Smith, C. (2013). The OECD approach to measuring subjective well-being. In J. F. Helliwell, R. Layard, & J. Sachs (Eds.), World Happiness Report 2013 (pp. 112–137). UN Sustainable Development Solutions Network.
Zurück zum Zitat Easterlin, R. A. (1974). Does economic growth improve the human lot? Some empirical evidence. In P. A. David & M. W. Reder (Eds.), Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz (pp. 89–125). Academic Press. Easterlin, R. A. (1974). Does economic growth improve the human lot? Some empirical evidence. In P. A. David & M. W. Reder (Eds.), Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz (pp. 89–125). Academic Press.
Zurück zum Zitat Nelson, G. C., van der Mensbrugghe, D., Ahammad, H., Blanc, E., Calvin, K., Hasegawa, T., Havlik, P., Heyhoe, E., Kyle, P., Lotze-Campen, H., von Lampe, M., Mason d’Croz, D., van Meijl, H., Müller, C., Reilly, J., Robertson, R., Sands, R. D., Schmitz, C., Tabeau, A., … Willenbockel, D. (2014). Agriculture and climate change in global scenarios: Why don’t the models agree. Agricultural Economics, 45(1), 85–101. https://doi.org/10.1111/agec.12091CrossRef Nelson, G. C., van der Mensbrugghe, D., Ahammad, H., Blanc, E., Calvin, K., Hasegawa, T., Havlik, P., Heyhoe, E., Kyle, P., Lotze-Campen, H., von Lampe, M., Mason d’Croz, D., van Meijl, H., Müller, C., Reilly, J., Robertson, R., Sands, R. D., Schmitz, C., Tabeau, A., … Willenbockel, D. (2014). Agriculture and climate change in global scenarios: Why don’t the models agree. Agricultural Economics, 45(1), 85–101. https://​doi.​org/​10.​1111/​agec.​12091CrossRef
Zurück zum Zitat Syrovátka, M. (2007). Aggregate measures of welfare based on personal consumption. In P. Šauer (Ed.), Environmental Economics, Policy and International Relations (pp. 154–163). Litomyšl Seminar Publishing. Syrovátka, M. (2007). Aggregate measures of welfare based on personal consumption. In P. Šauer (Ed.), Environmental Economics, Policy and International Relations (pp. 154–163). Litomyšl Seminar Publishing.
Metadaten
Titel
Income Fluctuations and Subjective Well-being: The Mediating Effects of Occupational Switching and Remittances
verfasst von
Azizbek Tokhirov
Publikationsdatum
01.12.2024
Verlag
Springer Netherlands
Erschienen in
Journal of Happiness Studies / Ausgabe 8/2024
Print ISSN: 1389-4978
Elektronische ISSN: 1573-7780
DOI
https://doi.org/10.1007/s10902-024-00814-y