Rainfall variability and labor allocation in Uzbekistan: the role of women’s empowerment

ABSTRACT

Employing a novel georeferenced household survey enriched with data on precipitation and temperature, this paper examines how rainfall variability affects individual labor supply in Uzbekistan, a highly traditional lower-middle-income country in Central Asia. The findings suggest that rainfall variability induces the reallocation of labor supply: (1) out of agriculture to unemployment, (2) from unemployment to business activities and irregular remunerated activities, and (3) from being out of labor force to unemployment. These effects differ in rural and urban areas and by gender. In addition, active women’s involvement in the labor market and household decision-making mediates the impact of rainfall variability on employment choices, especially in rural areas. This implies that traditional gender roles may make households in developing countries more vulnerable to adverse consequences of climate change, while women’s empowerment may mitigate such consequences.

KEYWORDS:

• Rainfall variability
• labor market
• agriculture
• women’s empowerment
• Uzbekistan
• Central Asia

Introduction

Despite their lower emission of greenhouse gases, developing countries remain more vulnerable to climate variability than developed countries (Diffenbaugh and Burke 2019; Fankhauser and McDermott 2014; Tol et al. 2004). At the country and regional levels, this vulnerability of developing countries is reflected in their slower economic growth, adverse health outcomes, and lower productivity as a result of an increasing number of extreme weather events such as floods, droughts, and extreme temperatures (Burgess et al. 2017; Park et al. 2018; Tol et al. 2004). At the individual and household levels, the impact of extreme weather and climate variability is more nuanced. Economically disadvantaged groups such as low-income individuals, women-headed households, and migrants often suffer more from the consequences of climate variability than their more prosperous counterparts (Flatø, Muttarak, and Pelser 2017; IPCC 2014). For instance, low-income earners who often work in sectors with more exposure to extreme weather lose (part of) their income (Park et al. 2018), female farmers are less capable of adopting drought-tolerant technologies and bear more risks of losing agricultural income (Fisher and Carr 2015), while women-headed households left behind by migrant partners may also have fewer contacts with social networks and lower possibilities for coping with adverse consequences of extreme weather (Flatø, Muttarak, and Pelser 2017).

Household employment choices and allocation of labor to agricultural and non-agricultural activities may be especially affected by climate variability. Many studies document that extreme weather events increase mortality (Burgess et al. 2017; Deschênes and Greenstone 2011; Deschênes and Moretti 2009; Otrachshenko, Popova, and Solomin 2017, 2018), reduce self-rated health (Yang et al. 2022), reduce cognitive performance (Chang, Kajackaite, and Capraro 2019; Cho 2017; Groppo and Kraehnert 2017; Park 2022; Park, Goodman, and Behrer 2021; Park et al. 2020; Randell and Gray 2016), retard birth weight and early-childhood development (Banerjee and Maharaj 2020; Greenstone, Guryan, and Deschênes 2013; Lohmann and Lechtenfeld 2015; Rudi and Soares 2015; Skoufias and Vinha 2012), increase the likelihood of conflicts (Otrachshenko, Popova, and Tavares 2021; Ranson 2014; Vestby 2019), and reduce labor productivity (Maccini and Yang 2009; Letta and Tol 2019; Park et al. 2018; Zhang et al. 2018). These impacts have direct implications for labor market activities. However, relatively few studies investigate how climate change and extreme weather events affect labor allocation, and their results are quite mixed. For instance, in a pioneering study on this topic, Graff Zivin and Neidell (2014) suggest that extreme heat in the US induces reallocation of time from work to non-work. In line with this finding, Otrachshenko and Popova (2022) show that extreme heat affects regional-level income distribution in Russia by increasing the unemployment rate, while extreme rainfall has no effects on employment. Recent studies on China also suggest that extreme heat leads to a reduction in work hours and the reallocation of labor from agricultural to non-agricultural sectors, but does not drive individuals out of the labor force or into unemployment (Huang et al. 2020; Jessoe, Manning, and Taylor 2018; Jiao, Li, and Liu 2021; Li and Pan 2021).

This paper examines how climate variability affects individual labor supply in Uzbekistan, a highly traditional lower-middle-income country in Central Asia. Employing novel georeferenced household survey data enriched with data on precipitation and temperature, we first study the impact of rainfall variability and mean temperature on employment in agricultural and non-agricultural sectors. Then, we focus on more specific employment choices in rural and urban areas, such as having own business activities, salaried employment, self-employment in a farm or a croft, having irregular remunerated activities, or being out of the labor force, holding unemployment as a default choice. In addition, we analyze whether greater women’s empowerment at the household level helps a household to be more resilient to climate variability. Women’s empowerment is measured as the degree of women’s involvement in household spending decisions, labor market participation, and various social activities. We hypothesize that women’s empowerment brings additional adaptive capacity to households in the face of climate change, for instance, by bringing additional household income, directing employment choices toward less risky activities, and/or encouraging other household members to do so.

Our findings suggest that both mean temperature and rainfall variability affect occupational choices, especially the likelihood of having one’s own business, of being involved in irregular remunerated activities, and of being out of the labor force. Climate variability also affects the probabilities of being in agricultural and non-agricultural sectors. The results differ between rural and urban areas, and women’s empowerment serves as a channel for the relationship between climate variability and employment choices, especially in rural areas.

Our paper contributes to the literature in several ways. First, we provide evidence that women’s empowerment helps to make households in lower-middle-income and developing countries more resilient to risks associated with climate change. This has the important policy implication that strengthening the role of women in household decision-making creates additional adaptive capacity in the face of climate change. Second, we provide a comprehensive analysis of individual labor supply decisions in response to climate variability. For this, we focus on both the individual involvement in agricultural and non-agricultural activities and on several specific employment choices, including employment, self-employment, business activities, irregular activities, staying out of labor force, and being unemployed. Such an analysis is important for understanding the employment dynamics and factors linking individuals in rural and urban areas to the labor market. Finally, we focus on Uzbekistan, a country prone to climate risks and dependent on agriculture that also maintains traditional gender roles, but has a legacy of the Soviet past, with its focus on gender equality.

Mechanisms and hypotheses

Both exposure to extreme heat and rainfall scarcity may affect labor market decisions (Huang et al. 2020). The impact of temperature on labor market decisions may be attributed to several interrelated channels. The first is biological. The exposure to extreme heat is a thermal stress to the human body that induces thermoregulation and physiological adjustment (Basu and Samet 2002; Dell, Jones, and Olken 2014). This reduces human performance in cognitive and physical tasks (Park, Goodman, and Behrer 2021; Seppänen, Fisk, and Lei 2006; Graff Zivin, Hsiang, and Neidell 2018), leading to lower labor productivity and the reallocation of labor from employment to leisure and unemployment (Zhang et al. 2018; Graff Zivin and Neidell 2014).

Another channel that may explain the effect of temperature on household employment decisions is the difference in relative exposure to heat between economic sectors. For instance, extremely hot temperature leads to reallocation of labor away from sectors with a relatively high exposure—e.g. agriculture and farming – to sectors with a relatively low exposure—e.g. non-agricultural activities (Li and Pan 2021; Huang et al. 2020; Jessoe, Manning, and Taylor 2018; Park et al. 2018). Such reallocation may be triggered by health-related reasons. Due to detrimental effects on their health, individuals would wish to reduce their exposure to extreme temperatures and choose employment in a sector with a lower exposure. Another reason for such reallocation is economic opportunity. Due to lower labor and capital productivity, exposure to heat may reduce relative returns in sectors with greater exposure, such that sectors with a lower exposure become more attractive (Huang et al. 2020; Jessoe, Manning, and Taylor 2018; Zhang et al. 2018).

While most literature on extreme weather events and labor market outcomes focuses on the impact of extreme heat, the impact of rainfall (scarcity) on labor market choices has received less attention. This is unfortunate because rainfall variability may also have important effects on human health and labor market decisions, especially in low-income economies. Specifically, in low-income economies a large share of the labor force is devoted to and dependent on agriculture, and often has no means for drought-tolerant agricultural production technologies. Given that water is an important input in agricultural production, rainfall shortage and unpredictable rainfall amounts lead to agricultural income loss, increased food prices, and consumption shocks, forcing individuals to seek ways to adjust to such adverse shocks (Barrios, Bertinelli, and Strobl 2010; Chuang 2019; Flatø, Muttarak, and Pelser 2017; Hirvonen 2016). One such adjustment is the diversification of household income sources by relocating labor supply of household members from farming-related self-employment and business activities to non-farming activities and wage-earning employment (Bandyopadhyay and Skoufias 2015; Chuang 2019; Huang et al. 2020). Low precipitation also reduces demand for agricultural workers, leading to higher unemployment (Huang et al. 2020; Jessoe, Manning, and Taylor 2018; Mueller, Gray, and Hopping 2020).

Given these mechanisms, we therefore outline a first set of testable hypotheses.

H1a:

Climate variability leads to reallocation of labor from agricultural to non-agricultural activities and unemployment.

H1b:

Climate variability leads to reallocation of labor from business and employment to unemployment.

A second set of our hypotheses is related to women’s empowerment and differences in labor market activities between men and women. A recent study documents that facing heat stress, women perform on math and verbal tasks better than without heat stress, while men perform better at lower temperatures (Chang, Kajackaite, and Capraro 2019). Moreover, women are more risk averse, prefer less competitive situations, are more cooperative, more sensitive to social signals, and more emotional in uncertain situations (Ortmann and Tichy 1999; Croson and Gneezy 2009; Frank, Gilovich, and Regan 1993; Loewenstein et al. 2001). Given these differences between men and women, we hypothesize that women’s empowerment may affect employment decisions of household members and increase household resilience to climate shocks. This may occur through several possible mechanisms.

First, female labor force participation brings additional income to a household, reducing its liquidity constraints and smoothing its consumption when faced with climate shocks (Cattaneo and Peri 2016; Flatø, Muttarak, and Pelser 2017; Hirvonen 2016; Park et al. 2018).¹ In households with a higher degree of women’s empowerment, women are more likely to be in the labor market. Having an additional earner in a family makes such households less vulnerable to temperature and rainfall variability.

Households with more members participating in the labor market also have more opportunities to diversify income sources through diversifying employment choices. This brings an additional adaptive capacity to such households. For instance, in regions with high rainfall variability, it is less likely that all members of the same household will be self-employed in agriculture (Bandyopadhyay and Skoufias 2015) and more likely that household members will diversify their employment choices (Chuang 2019).

Earlier literature also suggests that men and women differ in their labor market decisions when faced with temperature and precipitation shocks. While men are more likely to shift their time from non-agricultural work to leisure as a result of heat and drought exposure, there is no such effect for women’s employment decisions (Huang et al. 2020). However, due to differences in access to resources and social networks, women are also less likely to adopt drought-tolerant agricultural production technologies (Fisher and Carr 2015), making female-headed households more vulnerable to climate variability (Flatø, Muttarak, and Pelser 2017).

Finally, since women are more risk averse (Croson and Gneezy 2009; Loewenstein et al. 2001), it is more likely that they will make an employment choice in favor of less risky activities and possibly encourage other household members to do so (Azmat and Petrongolo 2014). Thus, we hypothesize that persons in households with a greater degree of women’s empowerment will more likely prefer non-agricultural activities and salaried employment as compared to households with a lower degree of women’s empowerment.

Thus, our hypothesis regarding the role of women’s empowerment is as follows:

H2:

Employment choices in households with a greater degree of women’s empowerment are less affected by climate variability.

Background

Climate and economy of Uzbekistan

Uzbekistan is a “doubly” landlocked lower-middle-income economy and the most populated country in Central Asia. It borders Kazakhstan, Turkmenistan, Afghanistan, Tajikistan, and Kyrgyzstan, which are also landlocked. The area of Uzbekistan is approximately equal to that of Spain and the population is over 33 million, with equal shares of women and men. Almost half of the population of the country (49.6% in 2019) lives in rural areas (World Bank 2019).

The northeastern and southeastern parts of the country are mountainous. In the north-central and western parts, deserts are located. Deserts and steppes comprise 80% of the country’s territory and mountains occupy the remaining 20%. The administrative division of Uzbekistan includes 12 provinces, one autonomous republic, and one independent city, which can be classified into 7 geographic-economic zones.²

The climate of Uzbekistan is dry and continental. The average temperature in January varies from −5°C in the north to + 5°C in the south, while the average temperature in July varies from + 26°C and + 30°C in the north to + 32°C and + 41°C in the south. The average annual precipitation varies from 80 mm in the north, 200–300 mm in the west, and 1000 mm in mountainous areas.

Uzbekistan is especially prone to climate risks. Over the last 50 years, the average annual temperature increased in Uzbekistan by 1.5°C, which is twice the global average for the same period. According to different estimates, average annual temperature in Uzbekistan is projected to increase by 3–8°C by 2040 (Boehlert et al. 2013). The average annual precipitation has fallen by 10 mm over the last 50 years. Projections suggest an increasing rainfall variability in the future with even drier climate in most of the country and increasing precipitation in mountainous areas (Boehlert et al. 2013). Table A1 in the online Appendix shows the shares of districts in our analysis that are in the top and bottom quartiles of the temperature and precipitation distribution. As shown, the country is diverse in terms of climatic conditions and their variability over time.

Water scarcity, water salinization, desertification, and land erosion are the main climate-related challenges of Uzbekistan. Currently, 90% of surface water is used for irrigation purposes. The World Bank predictions suggest that water scarcity will worsen in the future, leading to an increase in the water deficit from 2,000 m³ in 2005 to 13,000 m³ by 2050 and reducing the yields of almost all crops by 20–50% by 2050 (Boehlert et al. 2013). This brings significant risks to the economy of Uzbekistan. According to the World Development Indicators (World Bank 2019), agriculture is one of the major economic sectors. It contributes 25.5% to the country’s gross domestic product (GDP) and employs 23.9% of the labor force. Uzbekistan is one of the largest producers and exporters of cotton in the world. Services and industry contribute approximately equal shares of GDP (32.2% and 33.2%, respectively); 46.6% of the labor force is employed in services and 29.5% in industry, including construction.

Women’s empowerment in Uzbekistan

Traditional gender roles play an important role in Uzbekistan, especially in rural areas. Surveys indicate that 80% of the population in Uzbekistan supports the family model, in which men are breadwinners and women are homemakers (FAO 2019). Families in Uzbekistan are often extended and have a patriarchal power structure with a high authority of elders, and boys controlling girls even when the girls are older (Bhat 2011). Women are responsible for most household chores in the family, including taking care of children and the elderly, cooking (regularly and on special family occasions), and housekeeping (Tokhtakhodjaeva 1997). Moreover, especially in rural areas, the chores also include home-based activities to support household current consumption, including working in the garden plot, looking after livestock and poultry, and delivering water for drinking and domestic needs (Bhat 2011). Women also have fewer opportunities for economic activities than men, and are often criticized for deviating from traditional roles and engaging in self-employment activities such as shuttle trade (Bhat 2011; Kamp 2005; Welter and Smallbone 2008).

Traditional gender roles are also reflected in the labor market. As shown in Table A2 in online Appendix A, according to the World Development Indicators (World Bank 2019), female labor force participation in Uzbekistan is 52% of working age women, which is much lower than in neighboring Kazakhstan (72%) and Russia (69%). Low female labor force participation and low preprimary school enrollment rates for both boys and girls also confirm that women are often housekeepers and primary caregivers in Uzbekistan (FAO 2019). Moreover, as compared to Kazakhstan and Russia, women in Uzbekistan are more likely to be self-employed or employed in agriculture and less likely to be employed in industry and services. Traditional gender norms also imply that priority in receiving education is given to boys in the family (Bhat 2011; Tokhtakhodjaeva 1997), which is confirmed by a lower rate of enrollment in tertiary education of women than men.

Data

Survey details

We use cross-sectional household-level survey data collected in Uzbekistan from November 2015 to January 2016. The initial sample includes 600 households with 3,000 individual observations from 95 districts (mahalla).³ For the analysis we use data on working age population, which is about 1,600 observations. The primary sampling unit is a household, and the respondent is the household head or the most knowledgeable person in the household. 95% of all interviews were conducted in the Uzbek language and 5% in Russian.

The survey is country representative. In each geographic-economic zone, a random sampling of regions and districts is undertaken, and a random selection of households in each district is performed. Sampling quotas for each geographic-economic zone are calculated based on the shares of urban and rural population. Of 600 interviews, 51% (304) were conducted in urban areas and 49% (296) in rural areas. The survey covered regions from each geographic-economic zone (see Figure A1 in online Appendix A).

The survey questionnaire includes information on household and individual socioeconomic characteristics, including: age, gender, education, employment status, and the economic sector of main occupation of each household member; administrative unit and mahalla of residence; rural or urban area of residence; migration patterns in household; and women’s empowerment experiences in the household. Despite a relatively small sample size of the survey, it gives us an opportunity to provide fresh insights into the relationship between climate variability and employment dynamics in the context of Uzbekistan. For this, we enrich the survey data by georeferencing each mahalla and matching the survey to long-term data on temperature and precipitation (for details, see the section entitled “Temperature and rainfall variability”).

Table 1 presents the descriptive statistics for all variables used in our analysis. Table A3 in online Appendix A also provides the sample descriptive statistics for rural and urban areas and by gender.

Table 1. Descriptive statistics.

	N = 1,618
Variables:	Mean	S.D.	Min.	Max.
Occupational choice:
Having own business	0.10	0.29	0.00	1.00
Salaried employment	0.47	0.50	0.00	1.00
Self-employed	0.08	0.28	0.00	1.00
Irregular employment	0.12	0.32	0.00	1.00
Out of labor force	0.14	0.34	0.00	1.00
Unemployed	0.10	0.30	0.00	1.00
Employed in agricultural sector	0.16	0.36	0.00	1.00
Employed in non-agricultural sector	0.61	0.49	0.00	1.00
Women’s empowerment	2.05	1.16	0.00	4.00
Average temperature	14.39	1.76	10.12	27.60
Rainfall variability (RV)	0.22	0.03	0.17	0.29
Male	0.51	0.50	0.00	1.00
Age	35.51	11.53	18	76
Higher education	0.20	0.40	0.00	1.00
Urban	0.44	0.50	0.00	1.00
Household size	3.34	1.42	1.00	7.00

Note: S.D. stands for standard deviations. Min. and Max. are minimum and maximum values of variables, respectively.

Employment choices

We first focus on the sector of main occupation, including the agricultural sector, non-agricultural sector, being unemployed, and being out of the labor force. We then take a closer look at employment choices in rural and urban areas and analyze several more specific employment choices, including: (1) having one’s own business; (2) being employed in the state or private sector; (3) being self-employed in a farm or a croft (tomorqa); (4) having an irregular remunerated activity; (5) being unemployed; or (6) being out of labor force.

Having one’s own business as a main occupation means that a respondent has a registered business in the non-agricultural sector: i.e. has a patent to operate as an independent entrepreneur or this business is registered as a legal entity; has an unregistered business in non-agricultural sector; or is a farm owner. Thus, this category includes both formal and informal businesses.

Being employed means that a respondent has a salaried job in the public or private sector, including being on maternity leave. Self-employed are those respondents who work in a family farm or a croft, including those who work in a croft but are perceived as unemployed. Having an irregular remunerated activity means being a mardikor, i.e. having temporary, one-time, or seasonal work. Being self-employed and having an irregular remunerated activity are both also typically informal.

We consider unemployed under the definition of the International Labor Organization, i.e. those who currently do not work, but actively search for a job and are ready to start working immediately given a job opportunity.⁴ Those who are out of labor force are working-age respondents who currently do not work and for any reason do not search for a job. Our sample excludes individuals younger than 18 years old, students, retired, and individuals with severe disabilities.⁵

It is important to account for informal employment while studying occupational choices in developing and transition economies (Lehmann 2015; Lehmann and Pignatti 2018; Williams and Lansky 2013). Informal employment typically includes work without an employment contract, self-employment without registering this activity, or irregular remunerated activities (Staneva and Arabsheibani 2014; Lukiyanova 2015; Williams and Lansky 2013). Due to the nature of informal employment, it is often difficult to distinguish between unemployment and informal employment. The occupational choices we studied include both formal and informal activities, and we account for self-employment and irregular activities as separate categories.⁶ Also, all choices in our sample are mutually exclusive. Thus, the respondents are likely to have no incentive to state that they are unemployed when they are in fact employed informally.

Finally, in a separate model specification we also account for secondary occupations from the same set of choices as the main occupation (business, employment, self-employment, and irregular activity).

Women’s empowerment

The question on women’s empowerment in our survey is formulated as follows: “Do you agree or do not agree with the following statements: (i) women in our family take active participation in decision making of main issues (i.e. planning of family expenses, large expenditure items, organization of family events, and education of children); (ii) women in our family can make independent decisions; (iii) if women in our family want to work, no one will impede them; and (iv) women in our family can go to a wedding, visit relatives living far away, go shopping without formal authorization from their spouse, father, brothers, etc.” This question is answered by the most knowledgeable person in a household. If a respondent agrees with a statement, the answer is coded as one and zero otherwise.

Based on this survey question, we construct a categorical variable for women’s empowerment that equals zero if a respondent disagrees with all statements (this category is used as a default in our analysis), 1 if a respondent agrees with one statement, 2 if a respondent agrees with two statements, 3 if a respondent agrees with three statements, and 4 if a respondent agrees with all four statements. Using this categorical variable helps to capture a possible non-linearity of women’s empowerment in a household from 0 (the lowest) to 4 (the highest). Figure 1 presents the distribution of answers to the women’s empowerment question.

Figure 1. The distribution of answers to the women’s empowerment question. The figure shows the shares of respondents who (0) disagreed with all statements regarding women’s empowerment, (1) agreed with one statement, (2) agreed with two statements, (3) agreed with three statements, and (4) agreed with all four statements, as described in the article text.

Source: Authors’ construction.

In Figures A2–A4 in online Appendix A, we disentangle the responses to this question by gender and urban/rural area. In rural areas, the responses of women and men are similar, while in urban areas, women feel more empowered than men evaluate them to be. Also, women and men in urban areas are more likely to agree with women’s empowerment statements than women and men in rural areas. The responses to the women’s empowerment question by gender also confirm that men are more conservative than women and in rural areas women are less empowered than in urban areas.

Temperature and rainfall variability

The data on temperature (°C) and precipitation (mm) are taken from Earth Map for the period from 1979 to 2015 at the district (mahalla) level. These data are based on the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric re-analysis of global climate (ERA5).⁷ To match the climatic data to the survey data, we first georeferenced each mahalla in our survey and then based on the latitude and longitude of each mahalla, collected data on temperature and precipitation for the period from 1979 to 2015.

The key variable that we use in our analysis is rainfall variability (RV).⁸ RV is an index of climatic risk and is the average deviation in annual departures from normal rainfall in percent of the long-term average (Conrad 1941, 1941; Schulze 2007a, 2007b). It is constructed as follows:

$\mathit{R}{V}_s=\frac{\mathit{Standard}\mathit{\hspace{0.17em}}\mathrm{deviatio}{\mathit{n}}_{\mathit{s},1979-2015}}{\mathit{Annual}\mathit{\hspace{0.17em}}\mathit{precipitatio}{n}_{\mathit{s},\mathit{\hspace{0.17em}}1979-2015}}$

where $\mathit{Standard}\mathit{deviatio}{n}_{\mathit{s},1979-2015}$ is the average standard deviation from the mean during the 1979–2015 period in district s. $\mathit{Annual}\mathit{precipitatio}{n}_{\mathit{s},1979-2015}$ is the average annual precipitation (in mm) during the 1979–2015 period.

Higher RV reflects more variable year-to-year rainfall in a district and a lower average precipitation in a district, i.e. the areas with frequent droughts would suffer both from low average annual precipitation and high RV (Schulze 2007a). In our sample, the average RV index is 0.22, ranging from 0.17 in the higher rainfall districts to 0.29 in the lower rainfall districts (see Table 1). Such RV can be considered as moderate,⁹ although, as discussed above, it is expected to grow. Higher RV also implies a less predictable amount of year-to-year rainfall, and thus serves as a measure of climatic risks in an economy (Conrad 1941; Schulze 2007a). Therefore, we hypothesize that higher RV creates the need for households to cope with the risks of unpredictable rainfall and induces changes in household occupational choices, and households with more empowered women may have better adaptive capacities.

As explained in the next section, in all model specifications, we also control for average annual temperature (in °C) during the 1979–2015 period to capture the long-run climatic differences between districts.

Econometric model

An individual i has the following employment choices: (1) to be involved in agricultural activities; (2) to be involved in non-agricultural activities; (3) to stay out of the labor force; or (4) to remain unemployed. Following Cameron and Trivedi (2005), we use the additive random utility model for multiple alternatives in which the individual utility associated with the n^th alternative is as follows:

(1) $U_{\mathit{ins}}=V_{\mathit{ins}}\left(x_{\mathit{is}},\mathit{R}{V}_s,\mathit{\hspace{0.17em}}\mathit{tem}{p}_s\right)+{\epsilon }_{\mathit{ins}}\text{break}\mathit{\hspace{0.17em}}{V}_{\mathit{ins}}\left({\mathit{x}}_{\mathit{is}},\mathit{R}{V}_s,\mathit{\hspace{0.17em}}\mathit{tem}{p}_s\right)={\delta }_n\mathit{tem}{p}_s+{\theta }_n\mathit{R}{V}_s+{\mathit{x}}_{ins}^{\prime }{\mathit{\beta }}_n\mathit{\hspace{0.17em}}$(1)

where the subscripts i$,\mathit{n}=\overline{1,4},$ and s stand for an individual, occupation alternatives, and a district, respectively. $U_{\mathit{ins}}\mathit{\hspace{0.17em}}$stands for the utility of an individual i, who decides upon an occupation n in a district s. $V_{\mathit{ins}}\left(\cdot \right)$ is the deterministic component of an individual i’s utility. $x_i$ is a set of individual characteristics such as age, gender, level of education, living in urban or rural areas, and a particular geographic-economic zone. $\mathit{R}{V}_s$ and $\mathit{tem}{p}_i$ stand for rainfall variability and the average annual temperature (°C) for the 1979–2015 period, respectively, that an individual i has faced in a district s. ${\epsilon }_{\mathit{ins}}$ is the random component of an individual i’s utility and stands for individual unobserved characteristics. ${\delta }_n$, ${\theta }_n$, and ${\beta }_n$ are a set of parameters to be estimated.

The individual’s decision regarding occupation is based on choosing the alternative with the greatest utility. That is, she chooses occupation n if the utility for this occupation is greater than the k^th alternative, implying that $U_{\mathit{ins}}$≥ $U_{\mathit{iks}}$ for ∀ n≠k. In this study, remaining unemployed is used as the reference alternative. Thus, the probability that an individual i chooses occupation n is as follows:

(2) $Pr\left[\mathit{Occupatio}{n}_{\mathit{is}}=\mathit{n}\right]=Pr\left[U_{\mathit{ins}}\ge U_{\mathit{iks}}\right]=\mathrm{Pr}\left[V_{\mathit{ins}}\left(x_{\mathit{is}},\mathit{R}{V}_s,\mathit{tem}{p}_s\right)+{\epsilon }_{\mathit{ins}}\ge V_{\mathit{iks}}\left(x_{\mathit{is}},\mathit{R}{V}_s,\mathit{tem}{p}_s\right)+{\epsilon }_{\mathit{iks}}\right]=Pr\left[V_{\mathit{ins}}\left(x_{\mathit{is}},\mathit{R}{V}_s,\mathit{tem}{p}_s\right)-V_{\mathit{ik}}\left(x_{\mathit{is}},\mathit{R}{V}_s,\mathit{tem}{p}_s\right)\ge {\epsilon }_{\mathit{iks}}-{\epsilon }_{\mathit{ins}}\right]\mathit{for}\mathrm{\forall }\mathit{n}\ne \mathit{k}$(2)

where $\mathit{\epsilon }$ are independent identically distributed type 1 extreme values and have the following density function:

(3) $\mathit{f}\left({\epsilon }_{\mathit{ins}}\right)=e^{-{\mathit{\epsilon }}_{\mathit{ins}}}exp\left(-e^{-{\mathit{\epsilon }}_{\mathit{ins}}}\right)\mathit{for}\mathit{n}=\overline{1,4}$(3)

Given (2) and (3), we obtain the multinomial logit model:

(4) $Pr\left[\mathit{Occupatio}{n}_{\mathit{is}}=\mathit{n}\right]=\frac{{\mathit{e}}^{{\mathit{V}}_{\mathit{ins}}}}{e^{{\mathit{V}}_{\mathit{i},\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{non}-\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{out},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{unempl}.,\mathit{s}}}}$(4)

Following a similar methodology, we then model the following occupational choices of an individual i: (1) to have one’s own business; (2) to be employed; (3) to be self-employed in a farm or a croft; (4) to have irregular remunerated activities; (5) to stay out of labor force; or (6) to remain unemployed. In all model specifications, robust standard errors are clustered at the district (mahalla) level. In addition, we estimate all models for urban and rural areas separately. In a separate model specification, we also analyze the occupational choices by gender.

To analyze whether women’s empowerment serves as a channel underlying the impact of climate variability on employment decisions, we include women’s empowerment variable in Equation 4(4) $Pr\left[\mathit{Occupatio}{n}_{\mathit{is}}=\mathit{n}\right]=\frac{{\mathit{e}}^{{\mathit{V}}_{\mathit{ins}}}}{e^{{\mathit{V}}_{\mathit{i},\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{non}-\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{out},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{unempl}.,\mathit{s}}}}$(4) ) with four separate categories, as described in the data section above. If the main effect of climate variability changes in magnitude or loses its statistical significance when women’s empowerment categories are included, this implies that women’s empowerment may serve as a channel for the climate variability and employment decisions relationship.

In addition, we provide several robustness checks to our results, as described in online Appendix B.

Results

Main results – agricultural vs. non-agricultural jobs

Tables 2 and 3 present the main estimation results (marginal effects) for Equation 4(4) $Pr\left[\mathit{Occupatio}{n}_{\mathit{is}}=\mathit{n}\right]=\frac{{\mathit{e}}^{{\mathit{V}}_{\mathit{ins}}}}{e^{{\mathit{V}}_{\mathit{i},\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{non}-\mathit{agriculture},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{out},\mathit{s}}}+e^{{\mathit{V}}_{\mathit{i},\mathit{unempl}.,\mathit{s}}}}$(4) . Unemployment is used as a default category and marginal effects are interpreted in comparison to this category. Given that individuals in rural and urban areas face different employment opportunities, we disentangle the results by urban and rural areas (Table 2). We then also present these results by gender (Table 3). For simplicity, Tables 2 and 3 present only the main results regarding the effects of temperature and rainfall variability (for full regression results, see Tables A4 and A5 in online Appendix A).

Table 2. Agricultural and non-agricultural sectors – marginal effects.

	Agricultural sector	Non-agricultural sector	Out of labor force
Urban area
Average temperature	−0.010	0.065**	0.009
	(0.013)	(0.027)	(0.006)
Rainfall variability	−1.699**	−0.155	1.022
	(0.792)	(1.749)	(1.226)
Observations	708	708	708
Rural area
Average temperature	−0.014	−0.004	0.052**
	(0.026)	(0.025)	(0.020)
Rainfall variability	1.688	3.070**	−3.356***
	(2.077)	(1.466)	(1.228)
Observations	910	910	910

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. Multinomial logit results (marginal effects) are presented. Robust standard errors are clustered at the locality level. All specifications include individual characteristics and regional fixed effects. Individual characteristics include gender, age and its square, a dummy for having higher education, and household size.

Table 3. Agricultural and non-agricultural sectors – marginal effects – by gender.

	Urban			Rural
	Agricultural sector	Non-agricultural sector	Out of labor force	Agricultural sector	Non-agricultural sector	Out of labor force
Women
Average temperature	−0.021	0.024	0.006	−0.050**	0.035	0.099***
	(0.057)	(0.038)	(0.020)	(0.024)	(0.028)	(0.031)
Rainfall variability	−0.786	−3.388	2.657	3.945**	0.828	−6.046***
	(1.098)	(2.952)	(2.301)	(1.698)	(1.329)	(1.871)
Observations	365	365	365	430	430	430
Men
Average temperature	−0.004	0.158	0.045	0.019	−0.025	0.011
	(0.009)	(0.295)	(0.000)	(0.049)	(0.050)	(0.017)
Rainfall variability	−2.087	5.337	−3.084	−0.620	4.955**	−0.595
	(1.269)	(16.253)	(0.000)	(2.599)	(2.520)	(1.313)
Observations	343	343	343	480	480	480

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. All regression results include individual characteristics and regional fixed effects. Individual characteristics include gender, age and its square, a dummy for having higher education, and household size. Robust standard errors are clustered at the locality level.

As shown in Table 2, rainfall variability reduces the probability of being in the agricultural sector in urban areas by 1.7% points (p.p.) and increases the probability of being in the non-agricultural sector in rural areas by 3.1 p.p. Thus, because of increasing rainfall variability, individuals in an urban area would rather be unemployed than involved in agricultural activities, while in a rural area, individuals prefer working in the non-agricultural sector. In addition, in a rural area, with an increase in rainfall variability, individuals are also more likely to become unemployed instead of being out of labor force by 3.6 p.p. This result supports our Hypothesis 1a.

We further disentangle these results by gender. The results are presented in Table 3. In urban areas, employment choices of both women and men are unaffected by temperature and rainfall variability, while in rural areas, we uncover several important differences in the impact of temperature and rainfall variability on employment choices of men and women. A 1 p.p. increase in rainfall variability leads to an increase in women’s involvement in the agricultural sector by 3.9 p.p. and to an increase in men’s involvement in the non-agricultural sector by 4.96 p.p. At first glance, this may seem counterintuitive that women prefer involvement in the riskier agricultural sector when rainfall variability rises. However, non-agricultural jobs are scarce in rural areas and traditional gender norms imply that men should be given priority when employment opportunities are low. Since rainfall variability may tighten the liquidity constraints for rural households, the agricultural sector may become the only possibility for a woman in a rural area to earn money instead of remaining unemployed. This explanation is also confirmed by a substantial decrease in the likelihood of being out of labor force for women in a rural area (by 6 p.p.). These findings are also consistent with the results presented in Table 2.

Occupational choices in urban and rural areas

Tables 4 and 5 present the estimation results (marginal effects) for several specific occupational choices in urban and rural areas, including having own business activities, salaried employment, self-employment in a farm or a croft, having irregular remunerated activity, and staying out of the labor force, holding unemployment as a default choice. Similar to the results above, we first present these results for urban and rural areas (Table 4) and then disentangle these results by gender (Table 5).¹⁰

Table 4. Occupational choices – marginal effects – rural and urban areas.

	Business	Employed	Self-employed	Irregular	Out of labor force
Urban area
Temperature	−0.002	0.068***	−0.012	−0.005	0.015**
	(0.002)	(0.019)	(0.040)	(0.047)	(0.006)
Rainfall variability	2.817***	−1.376	−0.639	−2.767***	1.123
	(0.617)	(1.825)	(0.672)	(1.038)	(1.252)
Observations	708	708	708	708	708
Rural area
Temperature	0.012**	0.014	−0.024	−0.018	0.051**
	(0.006)	(0.029)	(0.017)	(0.013)	(0.020)
Rainfall variability	−0.710*	1.195	1.350	2.764***	−3.297***
	(0.410)	(1.305)	(0.993)	(0.769)	(1.191)
Observations	910	910	910	910	910

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. Multinomial logit results (marginal effects) are presented. Robust standard errors are clustered at the locality level. All specifications include individual characteristics and regional fixed effects. Individual characteristics include gender, age, and its square, a dummy for having higher education, and household size.

Table 5. Main results – marginal effects – by gender.

	Urban					Rural
	Business	Employed	Self-employed	Irregular	Out of labor force	Business	Employed	Self-employed	Irregular	Out of labor force
Women
Temperature	0.000	0.043**	−0.022	−0.037	0.022	0.006	0.020	−0.048*	0.011	0.094***
	(0.006)	(0.020)	(0.056)	(0.031)	(0.014)	(0.018)	(0.026)	(0.026)	(0.011)	(0.035)
Rainfall variability	3.325***	−4.305*	−0.881	−1.387***	1.966	−0.685	2.303*	2.022	0.931*	−5.992***
	(0.301)	(2.612)	(1.109)	(0.408)	(1.904)	(1.063)	(1.233)	(1.272)	(0.560)	(2.076)
Observations	365	365	365	365	365	430	430	430	430	430
Men
Temperature	−0.018	0.044	-	0.141	0.038	0.009	0.004	0.023	−0.036	0.008
	(0.050)	(0.146)		(0.108)	(0.295)	(0.017)	(0.046)	(0.028)	(0.023)	(0.018)
Rainfall variability	3.364	2.152	-	−2.689	−2.769	−0.856	−0.441	1.388	4.027***	−0.561
	(2.256)	(9.841)		(5.272)	(19.938)	(0.981)	(1.976)	(1.225)	(1.103)	(1.351)
Observations	343	343		343	343	480	480	480	480	480

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. Robust standard errors are clustered at the locality level. All regressions include regional fixed effects. The sample contains only one observation for a self-employed male in urban areas, and therefore this category is not estimated.

As shown in Table 4, a 1°C increase in temperature raises the likelihood of being out of labor force in both urban and rural areas (by 1.5 and 5.1 p.p., respectively). This finding is consistent with the literature on the allocation of time during heat (Graff Zivin and Neidell 2014). In addition, an increase in temperature raises the probability of being employed in an urban area by 6.8 p.p. and raises the likelihood of having a business in a rural area by 1.2 p.p.

Regarding rainfall variability, its increase by 1 p.p. raises the probability of having one’s own business in an urban area by 2.82 p.p. and reduces the probability of having irregular remunerated activity by 2.77 p.p. when compared to unemployment. This result might be explained by increasing opportunities for running a business in the service sector on a regular basis instead of being involved in irregular activities. Thus, we find support for Hypothesis 1b in urban areas in the case of irregular remunerated activities, but not in the case of business activities.

In rural areas, we find that rainfall variability decreases the probability of having one’s own business by 2.82 p.p. and increases the probability of having irregular remunerated activity by 2.76 p.p. Rainfall variability in a rural area also reduces the likelihood of being out of labor force by 3.3 p.p., i.e. individuals become unemployed and start looking for a job.

These results suggest that individuals in rural and urban areas indeed react to climate variability differently. In urban areas, where economic sectors are more diversified, in response to rainfall variability individuals prefer to run their own business and reduce irregular activities. In rural areas, where most activities are related to either agriculture or its related services, higher rainfall variability increases the risks for farming, and as a result, reduces income. Thus, to secure their livelihood, individuals choose to be involved in irregular remunerated activities rather than remain unemployed. In addition, we find that in both rural and urban areas, rainfall variability has no statistically significant effect on the likelihood of having salaried employment. This finding is in line with earlier literature. For instance, Otrachshenko and Popova (2022) show that extreme precipitation in Russia does not affect the shares of employment in different economic sectors. It is likely because labor regulations do not allow for labor demand to adjust quickly in response to rainfall variability. Thus, salaried employment becomes unaffected by the risks associated with rainfall variability.

The choice to be self-employed in a farm or a croft instead of being unemployed is also unaffected by either temperature or rainfall variability in both rural and urban areas. When faced with the risks of temperature and rainfall variability, individuals may not differentiate between unemployment and self-employment in a farm or a croft, since the latter occupational choice may not bring much additional income.

We then disentangle the results by gender. The results are presented in Table 5. For women in a rural area, rainfall variability marginally increases the likelihood of being employed (by 2.3 p.p.) and having irregular activities (by 0.9 p.p.) and significantly decreases the likelihood of being out of labor force (by 6 p.p.). This suggests that with an increase in rainfall variability, women start looking for a job, but the job opportunities for women in a rural area might be scarce. For men in a rural area, rainfall variability raises the likelihood of moving from unemployment to irregular activities by 4 p.p., providing opportunities to earn when own business is risky or regular employment possibilities are scarce. In an urban area, rainfall variability affects women’s employment choices only and does not affect the employment choices of men. For women, the likelihood of being involved in business increases by 3.3 p.p. in response to a 1 p.p. increase in rainfall variability, while the likelihood of being involved in irregular activities and in employment decreases by 1.4 and 4.3 p.p., respectively. Temperature has no effect on the employment choices of men in both rural and urban areas. With an increase in the average temperature by 1°C, women in a rural area are more likely to leave the labor force compared to being unemployed, and are less likely to be self-employed, while in urban areas, women are more likely to become employed.

The role of women’s empowerment

Next, we test whether women’s empowerment smooths the impact of climate variability on occupational choice.¹¹ Table 6 presents the results for employment in the agricultural and non-agricultural sectors. In the presence of women’s empowerment, the magnitude of the marginal effect of rainfall variability on being employed in the agricultural sector in urban areas has slightly increased. In rural areas, the magnitude of the marginal effect of rainfall variability on being employed in the non-agricultural sector becomes not statistically significant, while the magnitude of the marginal effect of rainfall variability on being out of the labor force decreases and becomes marginally statistically significant. These results point out that under climate risks, women’s empowerment in urban areas shifts the labor force away from the risky agricultural sector. In rural areas, employment choices are shifted from being out of the labor force to unemployment and from unemployment to non-agricultural jobs, providing the possibility of receiving income that is less vulnerable to climate risks. This is in line with our Hypothesis 2 and with the literature that shows that women are more risk averse. However, these findings hold for rainfall variability but not for mean temperature, since adding women’s empowerment does not alter the marginal effect of temperature on agricultural and non-agricultural employment.

Table 6. Women’s empowerment – agricultural and non-agricultural sectors – marginal effects.

	Agricultural sector	Non-agricultural sector	Out of labor force
Urban area
Average temperature	−0.013	0.069**	0.007
	(0.013)	(0.028)	(0.007)
Rainfall variability	−1.739**	0.224	0.767
	(0.848)	(1.818)	(1.272)
Women’s empowerment (default = disagree with all statements):
Agree with one statement	0.019***	0.078	−0.058**
	(0.005)	(0.047)	(0.026)
Agree with two statements	0.051***	0.038	−0.056
	(0.011)	(0.059)	(0.043)
Agree with three statements	0.010**	0.097**	−0.108***
	(0.005)	(0.042)	(0.041)
Agree with four statements	0.010	0.115***	−0.117***
	(0.011)	(0.034)	(0.032)
Observations	708	708	708
Rural area
Average temperature	−0.004	0.003	0.039**
	(0.029)	(0.028)	(0.020)
Rainfall variability	1.043	2.898	−2.514*
	(2.337)	(1.824)	(1.335)
Women’s empowerment (default = disagree with all statements):
Agree with one statement	0.095**	−0.082*	−0.070**
	(0.043)	(0.049)	(0.028)
Agree with two statements	0.102*	−0.009	−0.104***
	(0.058)	(0.066)	(0.034)
Agree with three statements	0.011	0.057	−0.099***
	(0.059)	(0.068)	(0.030)
Agree with four statements	0.039	−0.025	−0.043
	(0.079)	(0.058)	(0.043)
Observations	910	910	910

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. All regression results include individual characteristics and regional fixed effects. Individual characteristics include gender, age and its square, a dummy for having higher education, and household size. Robust standard errors are clustered at the locality level. See the section entitled “Women’s empowerment” for the definition of the women’s empowerment variable.

Table 7 presents the results with the women’s empowerment for occupational choices. Compared to the results in Table 4, we find that in an urban area the magnitude of marginal effects of rainfall variability on having one’s own business has decreased by 13 p.p. (=[2.453–2.817]/2.817) and on having irregular remunerated activities has increased by 5.7 p.p., respectively. In rural areas, the marginal effect of rainfall variability on the choice of having one’s own business becomes not statistically significant and on the choice of irregular remunerated activity increases by 20 p.p. In rural areas, women’s empowerment also decreases the magnitude of the effect of rainfall variability on the choice of being out of the labor force (26 p.p.) and makes it marginally significant. Thus, women’s empowerment affects the relationship between rainfall variability and having one’s own business in both rural and urban areas. A possible explanation is that running a business is not only associated with high risk and uncertainty but also requires some efforts and investment. In this case women, who are more risk averse than men, may remain unemployed or advise their family members to remain unemployed. On the other hand, we find that the marginal effect of rainfall variability on having irregular remunerated activity increases in the presence of women’s empowerment, suggesting that such empowerment encourages finding a job when climate risks increase. These findings suggest that women’s empowerment may help to secure their households’ livelihoods under climate risks, providing support to our Hypothesis 2.

Table 7. Women’s empowerment – occupational choices – marginal effects.

	Business	Employed	Self-employed	Irregular	Out of labor force
Urban area
Average temperature	−0.001	0.078***	−0.027	−0.002	0.015**
	(0.002)	(0.020)	(0.038)	(0.046)	(0.007)
Rainfall variability	2.453***	−0.439	−0.758	−2.925***	0.897
	(0.830)	(2.240)	(0.706)	(0.976)	(1.305)
Women’s empowerment (default = disagree with all statements):
Agree with one statement	−0.052**	0.081*	0.126***	−0.022	−0.075***
	(0.025)	(0.047)	(0.023)	(0.020)	(0.023)
Agree with two statements	−0.023	0.035	0.145***	−0.030	−0.075**
	(0.030)	(0.063)	(0.021)	(0.028)	(0.038)
Agree with three statements	−0.065	0.106	0.140***	−0.031	−0.132***
	(0.053)	(0.082)	(0.025)	(0.020)	(0.049)
Agree with four statements	−0.012	0.116***	−0.006	0.026	−0.112***
	(0.031)	(0.037)	(0.018)	(0.032)	(0.032)
Observations	708	708	708	708	708
Rural area
Average temperature	0.012	0.044	−0.028	−0.027**	0.038*
	(0.008)	(0.034)	(0.025)	(0.013)	(0.020)
Rainfall variability	−0.421	−0.460	1.293	3.318***	−2.431*
	(0.515)	(1.755)	(1.499)	(0.829)	(1.285)
Women’s empowerment (default = disagree with all statements):
Agree with one statement	−0.038	0.021	0.062	−0.034	−0.062***
	(0.029)	(0.059)	(0.048)	(0.030)	(0.022)
Agree with two statements	−0.041	0.143	0.035	−0.051	−0.096***
	(0.026)	(0.094)	(0.078)	(0.033)	(0.028)
Agree with three statements	−0.004	0.159*	−0.019	−0.073*	−0.092***
	(0.025)	(0.090)	(0.078)	(0.038)	(0.026)
Agree with four statements	−0.012	0.082	−0.050	−0.015	−0.034
	(0.026)	(0.101)	(0.085)	(0.053)	(0.036)
Observations	910	910	910	910	910

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. All regression results include individual characteristics and regional fixed effects. Individual characteristics include gender, age and its square, a dummy for having higher education, and household size. Robust standard errors are clustered at the locality level. See the section entitled “Women’s empowerment” for the definition of the women’s empowerment variable.

Concerning mean temperature, the introduction of women’s empowerment makes the marginal effect on the choice of having one’s own business in rural areas not statistically significant. Moreover, the impact of temperature on the choice of being out of labor force becomes marginally significant and of lower magnitude, while the negative impact of temperature on the choice of irregular activities becomes statistically significant. Women’s empowerment also reinforces the positive impact of temperature on the likelihood of being employed in an urban area. These findings suggest that women’s empowerment serves as a channel through which the relationship between temperature and choice is explained.

Secondary jobs

7.3% of the respondents in our sample also have a secondary occupation in addition to their main one. The secondary occupation is from the same set of the main employment choices – i.e. having one’s own business, being employed, being self-employed, or having irregular activities. Based on this information, we create a dummy variable that equals 1 if an individual has a secondary occupation and zero otherwise. The choice to have a secondary job is conditional on having a main one. Therefore, we exclude individuals who are unemployed or out of labor force as their main occupational choice.

As shown in Table 8, an increase in rainfall variability of 1 p.p. raises the likelihood of having a secondary job in a rural area by 2.67 p.p., while in the whole sample and in the subsample of urban areas there is no such effect. This implies that rainfall variability substantially increases the risk of income loss for individuals in rural areas, so individuals are forced to look for additional income sources and choose to have a secondary occupation in addition to the main one.

Table 8. Secondary occupation – marginal effects.

	Baseline results			Results with women’s empowerment
	All	Urban	Rural	All	Urban	Rural
Average temperature	−0.009	−0.006	−0.017	−0.006	−0.003	−0.020
	(0.011)	(0.006)	(0.014)	(0.011)	(0.006)	(0.020)
Rainfall variability	0.877	−2.448	2.673***	0.744	−2.013	2.501***
	(0.737)	(1.841)	(0.735)	(0.648)	(1.911)	(0.823)
Women’s empowerment (default = disagree with all statements):
Agree with one statement	-	-	-	−0.013	0.029	−0.029
				(0.026)	(0.081)	(0.031)
Agree with two statements	-	-	-	0.037	0.066	0.005
				(0.031)	(0.083)	(0.039)
Agree with three statements	-	-	-	0.008	0.109	−0.112
				(0.051)	(0.076)	(0.085)
Agree with four statements	-	-	-	0.010	0.090	−0.047
				(0.049)	(0.082)	(0.075)
Observations	1,240	536	704	1,240	536	704

Notes: ***, **, and * correspond to 1%, 5%, and 10% significance levels, respectively. All regression results include individual characteristics and regional fixed effects. Individual characteristics include gender, age and its square, a dummy for having higher education, and household size. Robust standard errors are clustered at the locality level.

When we include women’s empowerment in Table 8, the magnitude of the marginal effect of rainfall variability in rural areas becomes slightly less (by 6.4 p.p.). This suggests that women’s empowerment may partially serve as a channel between rainfall variability and the choice to have a secondary job. With women’s empowerment, it is likely that women in a household also work, providing an additional income to the household. This smooths the impact of rainfall variability.

Conclusion

This paper contributes to the literature by analyzing the impact of climate variability on individual labor supply decisions in Uzbekistan, a country at high risk of experiencing the adverse economic consequences of global warming due to its dependency on agriculture and significant climatic risks. Earlier studies on the role of climate variability on the labor market have mostly focused on the effects of extreme temperatures. Our findings underscore that it is important to account for both temperature and rainfall variability. We find that temperature and rainfall variability affect the decisions to be active in the agricultural or non-agricultural sector as well as the decisions to have one’s own business activities, to have irregular remunerated activities, or to be out of labor force. The effects of climate variability differ in rural and urban areas. Interestingly, women’s empowerment helps to smooth the effects of climate variability and shifts employment choices to less risky activities. This implies that women’s empowerment is an important instrument in protecting households from income losses in the presence of global warming.

Our results open several avenues for future research that can be pursued upon the availability of panel data at the regional and individual levels. First, we examine the individual labor supply allocation specifics at a given point of time. It would also be interesting to analyze the role of climate variability on occupational choices over time. By showing that occupational choices do not depend on the period over which climate variability is calculated (see online Appendix), we provide a starting point for such an analysis. Given that the frequency and intensity of extreme weather events increase over the course of global warming, future studies may examine how the labor allocation changes over time, which occupational choices and industries are affected more, and whether women’s empowerment plays an increasing role over time. This is especially important for lower-middle-income and developing economies, many of which suffer from the consequences of global warming given their geographic location and traditionally high involvement in agriculture.

Another important dimension to consider is the role of climate variability and women’s empowerment on labor demand. Finally, it would be important to take a closer look at the regional dimension and investigate whether and how climate variability affects the regional-level allocation of labor between different industries and understand the extent to which rural-urban migration plays a role in this process.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/1060586X.2024.2309822.

Additional information

Funding

Vladimir Otrachshenko acknowledges the funding by the German Academic Exchange Service (DAAD) from funds of the Federal Ministry for Economic Cooperation (BMZ), SDGnexus Network (Grant No. 57526248), program “exceed – Hochschulexzellenz in der Entwicklungszusammenarbeit.” Nargiza Alimukhamedova acknowledges the support from the Czech Science Foundation (GAČR) project No. 19-19158S.

Notes

1. Another way of smoothing consumption and mitigating the consequences of climate change is participation in informal networks. For the role of informal networks in the climate-livestock relationship in Uzbekistan, see Otrachshenko, Ilyas, and Alimukhamedova (2024).

2. These are: Southern (Qashqadaryo and Surkhandaryo regions), Northern (Karakalpak, Khorezm), Fergana (Andijan, Fergana, Namangan), Central (Samarkand, Bukhara, Navoi), Tashkent capital, Tashkent region, and Mirzachul (Djizzakh, Sirdaryo).

3. Mahalla is derived from the Arabic mahali, which means “local.” In Uzbekistan, the term mahalla means neighborhood, local community, or state administrative unit.

4. Unfortunately, we do not have information on unemployment duration.

5. The retirement age in Uzbekistan is 60 years old for men and 55 years old for women, but the retired can also continue working while receiving retirement benefits. If a respondent has reached the retirement age but considers business, salaried employment, self-employment, or irregular activity as his/her main occupation despite being retired, we retain such respondents in our sample. Such persons constitute 1% of our sample.

6. The sample size does not allow us to distinguish between formal and informal businesses as separate categories.

7. For more details, see http://www.openforis.org/tools/earth-map.html (accessed 20 November 2020).

8. While we analyze how rainfall variability affects occupational choices in Uzbekistan, several studies use an alternative approach and examine the impact of rainfall shocks on labor market outcomes in India (Jayachandran 2006; Kaur 2019; Shah and Steinberg 2017). In such a methodology, rainfall shocks are defined as indicators of excess rainfall and/or rainfall scarcity (droughts) in specific years. While this approach helps in identifying the impact of short-run rainfall shocks, it does not account for the variability of rainfall and for climatic risks in the long run, which is our aim. For a general discussion on the differences in studying the effects of weather and climate, also see Dell, Jones, and Olken (2014).

9. Compared, for instance, to calculations done for South Africa (Schulze 2007a, 2007b, 2012).

10. Full results for Table 4 are presented in Table A6 in the online Appendix A.

11. The interpretation of results with women’s empowerment included in the model should consider the potential presence of unobserved factors that may affect both women’s empowerment and employment choices. We thank an anonymous reviewer for this comment.