Commodity price volatility and the psychological well-being of farmers

2.1 Background

Coffee production in Vietnam has accelerated over the last 3 decades; from accounting for just 1.2% of world output in 1989, Vietnam is currently the second largest coffee producer in the world (after Brazil). There are three key features of coffee production in Vietnam. First, the majority of the coffee grown in the country is the “robusta” variety. Compared to “arabica” (the other main variety grown globally), robusta is more resilient to variations in growing conditions and weather shocks, and typically has higher yields, but also fetches a lower price in the international market.

Second, there is a stark difference between the Central Highlands and other regions of Vietnam with respect to coffee production. In the Vietnam Access to Resources Household Survey (VARHS) data used in this paper (details regarding the data are provided in the following section), although 75% of the households residing in the Central Highlands report producing coffee, less than 0.5% of the households outside the Central Highlands did the same.3 Furthermore, the coffee sector plays an outsized role not just in agriculture but in the labor markets of the Central Highlands as well. As per the 2022 Vietnam Labor Force Surveys, the coffee sector provided employment to 34.3% of individuals between the ages of 15–64 in the Central Highlands, and only 35% of these were farming. The coffee sector provided employment for less than 0.2% of individuals in non-Central Highlands provinces.

Third, more than 85% of coffee in Vietnam is grown by smallholder farmers, with the average size of the holding being between 1 and 2 hectares (Luong & Tauer, 2006). So although in aggregate the country is a large producer, individual farmers are “price takers” in the international markets.

International coffee prices are quite volatile and driven by a mix of weather conditions in the largest coffee-producing countries, expectations about future prices, changes in demand and interest rates, as well as speculation (Deaton, 1999). Together this implies that future coffee prices are difficult to predict, especially for the smallholder farmer. The fluctuations in the world price for robusta coffee over the period 2004–2020 are illustrated in Figure 1 (details regarding the data are provided in the following section).

This volatility in coffee prices has important implications for farmers: newly planted coffee trees take about 3–4 years to bear fruit and tend to have a life span of more than 50 years. The trees are also costly to cut down (Giovannucci et al., 2004), altogether making it very difficult for farmers to abandon/invest in coffee trees in response to short-term fluctuations in coffee prices. The inability of coffee farmers to adjust the area dedicated to coffee trees quickly to international coffee prices has been well documented for countries such as Vietnam (Beck et al., 2019) and Uganda (Hill, 2010). Results presented later under the identification subsection further substantiate this assumption. In sum, small Vietnamese coffee farmers are uniquely exposed to the vagaries of the international coffee markets, providing an appropriate setting to examine the effects of price volatility on the mental health of farmers.

2.2 Conceptual framework

Our central hypothesis is that volatility in coffee prices increases uncertainty of future economic prospects as farmers do not know what revenue to expect at the time of harvest. In this subsection, we discuss six—potentially intersecting—reasons why one would expect income uncertainties to increase depressive symptoms.

The first theory is that constant worries about income can exhaust mental resources or “mental bandwidth” (Dean et al., 2019). Increased cognitive load can impair memory, attention spans, and overall cognitive functioning (Lichand & Mani, 2020; Mani et al., 2013; Ridley et al., 2020), which in turn have been found to predict diagnosis of depression (Phillips et al., 2010).

Second, a consistent body of evidence finds that an increased cognitive load can inhibit self-control (Dean et al., 2019; Spears, 2011). For example, Cotti et al. (2015) find fluctuations in the Dow Jones Industrial Average index increased engagement in risky health behavior such as binge drinking and cigarette smoking, which have strong associations with mental disorders.4

Third, income uncertainty could also lead to pessimistic expectations of future economic prospects and aspirations. For example, constant financial worries may change the perceived probability of future events such that higher probabilities are placed on bad states of the world materializing in the future. Such expectations can then lead to pessimistic views, which play an important role in the etiology and continuation of depression (de Quidt & Haushofer, 2019; Lybbert & Wydick, 2018).

Fourth, impending financial scarcity and the associated stress could lead individuals to withdraw from social events—either as a way to curtail expenses or due to an increase in antisocial behavior or reduction in social trust in the community (Haushofer et al., 2023; Prediger et al., 2014). Social isolation and low interpersonal trust have been shown to be significant predictors of depression (Cacioppo et al., 2006; Kim et al., 2012).

Fifth, looming financial distress may lead to a lower locus of control (i.e., whether one believes one's life outcomes are determined by their own actions or luck/fate), which is an important determinant of mental health (e.g., Churchill et al., 2020). Last, increased financial worries and associated cognitive load may reduce the utility derived from present consumption (Schofield & Venkataramani, 2021). Reduced gratification from consumption—not just of goods and services but also that of personal relationships and experiences—is a well-established symptom of depression (de Quidt & Haushofer, 2019).

In light of the above observations, we hypothesize that coffee price volatility will negatively affect the mental well-being of small coffee farmers in rural Vietnam. The underlying pathways discussed here are empirically tested later in the mechanisms section.

3.1 Data

We use the 2016, 2018, and 2020 waves of the Vietnam Access to Resources Household Survey (VARHS hereafter). The VARHS is a long-running panel survey of almost 2600 rural households conducted between May and July every second year since 2006 in 12 provinces across Vietnam.5^,6 The data were representative at the level of the province in 2012 (see Ayala-Cantu et al., 2017 for details). The VARHS data includes three of the four provinces in the Central Highlands—Dak Lak, Dak Nong, and Lam Dong. The sample size declined slightly during the analysis period: from 2669 in 2016 to 2605 in 2018 and 2583 in 2020. For the primary analysis, we use the balanced panel of 2581 households and discuss concerns related to attrition later in the robustness subsection.

We use the last three waves of the VARHS as mental health was only measured in these waves. Mental health is measured in the VARHS using the 10-item Center for the Epidemiological Studies of Depression scale (CES-D). The original CES-D scale consisting of 20 items was developed by Radloff (1977) as a screening tool to measure depressive symptoms, and a number of epidemiological studies show that it strongly predicts clinical diagnoses of depression and anxiety disorders (Weissman et al., 1977). We use the modified 10-item short-form version developed by Andresen et al. (1994).7 This version of the CES-D has been shown to have good psychometric properties in a variety of contexts (Andresen et al., 1994; Björgvinsson et al., 2013; Boey, 1999).

The CES-D is only administered to the primary respondent of the household—the household head or his/her spouse in most cases. Respondents were asked to indicate how often they had certain feelings in the last week on a 0–3 scale—“never (0 days in a week),” “sometimes (1–2 days of the week),” “often (3–4 days of the week),” and “all the time (5–7 days of the week).” A higher CES-D score reflects poorer mental health. We use the CES-D scale to construct two dependent variables. First, we use the composite score of the 10 questions (ranging from 0 to 30). Second, the cutoff of 10 (out of a maximum of 30) has been recommended as an indicator for the presence of significant depressive symptoms (Andresen et al., 1994; Björgvinsson et al., 2013; Boey, 1999). We use this threshold to construct a dummy variable, “severe depressive symptoms,” that takes the value 1 for CES-D scores greater than or equal to 10, and 0 otherwise. This latter measure of mental health has been used widely in other studies (e.g., Singhal, 2019; Sharma et al., 2021).

To investigate the effect of coffee price volatility we merge the household survey data with the monthly international robusta coffee prices published by the International Coffee Organization (ICO). The ICO publishes monthly indicator (or spot) prices for robusta coffee, which are a function of the average demand and supply of robusta in key global markets.8 These are shown in Figure 1 for the period 2004–2020. We convert the international coffee price from US cents to 1000 Vietnamese Dong (VND) in real December 2018 prices and capture volatility faced by the household leading up to the survey interview by defining our key explanatory variable (“coffee price volatility”) as the standard deviation in the international coffee price in the 2 years preceding the survey month (the survey takes place every 2 years).9 Fluctuations in the standard deviation of prices over the study period of 2016–2020 are shown in Figure 2. The mean coffee price volatility during the VARHS survey months in 2016 is 4.75 SD, 4.16 SD in 2018, and 3.37 SD in 2020. We also control for the level effect of the coffee price using the average coffee price in the preceding 2 years.

The remainder of the control variables come from the household survey data. We report the summary statistics for the main variables from the balanced panel dataset in Table 1. Panel A shows that the average score on the CES-D is 7.28 for the sample, and 28% of the sample exhibits severe depressive symptoms. Figure A3 in the online appendix presents the distribution of the CES-D scores for each year of the VARHS. The Cronbach's $$$ \alpha $$$ is 0.77 for 2016, 0.80 for 2018, and 0.81 for 2021, which indicates a high level of internal consistency.

The average age of the respondent is 54, and 41% are female. Some 71% have at least completed primary school, and nearly 80% are married. The survey also asks whether the household experienced any natural disasters (floods, droughts, etc.), pest attacks, and health shocks (death or illness of a household member) in the preceding 2 years. Exposure to these shocks ranges from 6% to 9%. Table 1 also shows the summary statistics separately for households residing in and outside the Central Highlands. Column 4 shows that respondents in the Central Highlands are more likely to exhibit severe depressive symptoms. They also differ significantly on several predetermined characteristics and exposure to shocks; we account for these in the analysis that follows.

3.2 Estimation strategy

The vector $$$ {X}_{it} $$$ includes respondent characteristics (gender, age, age squared, marital status, primary school completion dummy) and time-varying indicator variables to control for household exposure to the following shocks in the preceding 2 years: natural disasters (floods, droughts, etc.), pest attacks, and health shocks (death of a household member or illness). In addition to the dispersion in the coffee prices, the level of coffee price could also affect mental health. To account for this possibility, we also include $$$ \mathrm{Coffee}\kern0.5em {\mathrm{Price}}_t $$$, defined as the mean international coffee price in the 2 years preceding the survey, interacted with the Central Highlands indicator. All household-specific time-invariant characteristics are captured by the household fixed effect, $$$ {\eta}_i $$$. We also control for any other unobserved country-wide macroeconomic conditions at the time of the surveys using month-by-year fixed effects ( $$$ {\phi}_t $$$). They also control for seasonality. Last, $$$ {\nu}_{ict} $$$ is an idiosyncratic error term. Following Abadie et al. (2023), we cluster the standard errors at the level of the commune to account for the cluster sampling process of the VARHS (random sampling of communes within provinces, followed by random sampling of households within the communes).11

3.3 Identification assumptions

Before proceeding to the results, we discuss the validity of our study design. First, for international robusta coffee prices to matter, they must sufficiently transmit down to the farmgate prices received by farmers. We assess this assumption by comparing the international robusta coffee prices to the average monthly farmgate coffee prices observed in Dak Lak province in the Central Highlands during 2016–2022.12 We find the monthly international coffee price to be highly correlated with the monthly farmgate price in Dak Lak (pairwise correlation coefficient = 0.93). Next, we test the transmission mechanism by regressing log of the farmgate price on log international prices and month and year fixed effects. Results presented in Table A2 of the online appendix show that a 10% increase in international prices translates into a significant increase of 7.1% in farmgate prices in Dak Lak province.

Next, we test the assumption that coffee farmers are uniquely exposed to volatility in the international coffee markets due to their inability to significantly alter production decisions (at least in the short run). We limit the VARHS sample to all households that report harvesting coffee at least once in the 3-year panel and estimate Equation (2) for the following measures of production: (i) amount of coffee produced (tons); (ii) area under coffee cultivation (hectares); (iii) share of coffee in total area cultivated; and (iv) value of inputs used in the production of all crops.13 As results presented in Columns 1–4 of Table A3 in the online appendix show, the scale of coffee production is unresponsive to coffee price volatility, underpinning our assumption that coffee cultivation is a long-term investment.

We also examine the possibility that farmers could adjust the timing of sales in an effort to smoothen the revenue received from coffee. In Column 5, the outcome variable is an indicator that takes the value of 1 if the household sold coffee that year and 0 otherwise. In Column 6, the outcome variable is the change in the amount of stored coffee from the preceding year. Once again, we do not find these variables to be affected by coffee price volatility. Taken together, these results suggest that coffee farmers have limited ability to adjust coffee production at both the intensive (input use) and extensive (production scale) margins in response to short-run price movements. They also have limited capacity to hold onto output in periods of price uncertainty.

The study design could be undermined if farmers can anticipate future prices based on past price fluctuations. Although there often are swings in coffee prices (see Figure 1), we believe it is unlikely that smallholder farmers in rural Vietnam can anticipate changes in future international prices that are determined by international supply and demand factors (Bazzi & Blattman, 2014; Deaton, 1999), and forecasting of commodity prices is difficult (Ghoshray, 2011). Furthermore, results from the Phillips-Perron test indicate that the coffee price time series is nonstationary, making predictions based on this difficult.

Last, self-reported mental health measures can be a cause for concern as respondents may understate mental health issues due to social stigma against mental health problems or if they do not wish to be viewed in a negative light by the enumerator (social desirability bias). If this measurement error is random, then it will only reduce the precision of the estimates, without biasing the results. However, systematic bias in responses can occur if, for example, richer households care more about their social image and do not want to be perceived as depressed/unhappy (e.g., Reisinger, 2022). A recent study on the impacts of cash transfers tests this by randomly telling some respondents that a lower/higher self-report of depression was expected from them. This did not affect responses, indicating social desirability bias may not be a substantial issue in developing country settings (Haushofer, Mudida, & Shapiro, 2020). We also include a wide range of individual and household-level controls along with household fixed effects to capture sources of reporting bias.

4.1 Main results

Table 2 shows our key results using Equation (1). The dependent variable in Column 1 is the CES-D index. We find that an increase in coffee price volatility reduces mental health in CH relative to households residing outside the CH. The point estimates show that an increase of one standard deviation of 1000 VND prices increases the CES-D score by 0.9 units; given the control mean of 7.3, this translates into a 12.3% increase in depressive symptoms of individuals in CH relative to individuals residing outside the CH. To put the effect size into perspective, we note that 1 SD of coffee price is approximately 5800 VND. In the VARHS sample of coffee producers the average production of coffee is 3132 kg (Table A3, Col. 1), so assuming no additional costs, an increase in 1 SD of coffee price translates into 18.1 million VND or 11% of the household income of coffee cultivating households, which reflects a substantial risk to the household. The outcome variable in Column 2 is an indicator variable for the presence of significant depressive symptoms. Although statistically insignificant, the sign of the coefficient is consistent with our hypothesis and with the estimates of the impact on the CES-D index noted in Column 1.

Results in Columns 3 and 4 show that these results are robust to the addition of control variables. Column 3 shows that a one standard deviation increase in international coffee prices is associated with a 1.127 unit increase in the CES-D index (15% of the control mean). Results in the second row show that the mean level of international coffee prices does not affect mental health. We also find that women have lower mental health relative to men. This is supported by global evidence on the female mental health penalty (WHO, 2017) and other studies in Vietnam (Giang et al., 2010; Sharma et al., 2021).

4.2 Robustness

The results in Table 2 are also robust to a number of sensitivity checks. First, we check the robustness of the results to alternative definitions of coffee growing regions in Table A4 of the online appendix: (i) we construct a district-level dummy variable that takes value 1 if the district reported coffee cultivation in 2020, and 0 otherwise14; (ii) we use the Global Agro-Ecological Zones version 4 (GAEZv4) database that estimates the suitability of land for various crops to compute the mean coffee suitability index for every district and then proxy the potential exposure of a district to coffee price variability by constructing an indicator variable that takes the value 1 if the mean district suitability index for coffee is either “high” or “very high.”15 We find that the results are robust to these alternative specifications.

Second, we show that the results are robust to using several alternative measures of volatility in Table A5 of the online appendix as follows. (i) In Columns 1–2 of Panel A we compute volatility as the standard deviation of the logarithm of prices (1000 VND) in the 24 months preceding the survey; (ii) in Columns 3 and 4 of Panel A we use the coefficient of variation; and (iii) in Columns 5 and 6 we use the interquartile range. We then (iv) allay any concerns that realized coffee prices in the recent harvest months may be biasing the results by recomputing coffee price volatility as the standard deviation of international coffee prices in the 24 months preceding the survey excluding the harvest months of October–December (Columns 1 and 2 of Panel B).16 For all these alternative measures we find that an increase in volatility in international coffee prices is significantly related to worse mental health.

Next, (v) we conducted a placebo analysis using volatility in palm and crude oil prices (two commodities produced only at small scale in Vietnam) to test whether we are picking up the effects of international coffee price volatility or a more general effect of volatility in commodity markets. For both palm oil and crude oil, we do not find volatility in these commodity markets to be associated with mental well-being (Columns 3–6 of Panel B).17

In Panel C of Table A5, (vi) we report the results from using smaller time windows to calculate coffee price volatility. Instead of calculating volatility over the 24 months preceding the survey (the time between VARHS rounds), we iteratively reduce the time span to 22 months, 20 months, and 18 months. We find that although the results hold for price volatility over 22 months preceding the survey, the coefficient loses statistical significance upon reducing the window further.18

Third, in Table A7 of the online appendix, we undertake a series of robustness checks to confirm that recently realized coffee price shocks are not driving the effects of coffee price volatility. Although the preferred specification reported in Columns 3–4 of Table 2 controls for the average price of coffee in the 2 years preceding the survey, one may be concerned that coffee prices observed in the 12 months preceding the survey (including the most recent harvest season) matter more for the mental health of the farmer. We find that the results are robust to controlling for the average coffee price during the last 12 months leading up to the survey (Columns 1–2 of Panel A).

Alternatively, it may be the case that rather than the average coffee price, farmers are affected by large price movements (shocks). We test this by constructing the following indicators of price shocks: (i) We use data on coffee prices over 2004–2020 to calculate the deviation of average monthly coffee prices in the year preceding the survey from the historical mean (normalized using the historical standard deviation of coffee prices); and (ii) we generate a dummy variable that takes value 1 for every month where the coffee price faced by the farmer falls below 1 SD of the historical mean and then calculate the proportion this occurs during the 12 months preceding the survey. Controlling for these measures of price shocks does not change the results (Columns 3–6 of Panel A, Table A7).

Further, in the context of coffee price shocks, it may be the case that farmers are more concerned about price shocks during the harvest months of October–December rather than other times in the year. We test this by constructing the following three separate indicators of price shocks during the harvest months: (i) we compute the average coffee prices faced by farmers during the harvest months in the year preceding the survey; (ii) we compute the (normalized) deviation of the coffee prices during the harvest months in the year preceding the survey from the historic mean of coffee prices during harvest months; and (iii) we generate a dummy variable that takes value 1 for the months where the coffee prices experienced during harvest months in the preceding year fall below 1 SD of the historic mean of coffee prices during harvest months. We then calculate the proportion this occurs during the 12 months preceding the survey. Results reported in Columns 1–6 of Panel B, Table A7 show that the effects of volatility continue to hold, demonstrating that price shocks are not driving the results.

Fourth, we also conducted several analyses to examine if the results are sensitive to alternative formulations of the CES-D index. In particular, cardinal treatment of an ordinal dependent variable (such as the CES-D index) in a linear regression can lead to biased conclusions if the empirical findings are not robust to monotonic increasing transformations of the dependent variable (Schröder & Yitzhaki, 2017). Therefore, we examine the robustness of the results to (i) a wide range of smooth convex and concave monotonic transformations of the CES-D index recommended in Bloem (2022); and (ii) using the inverse hyperbolic sine transformation of the CES-D index. The results continue to hold for all these transformations, suggesting that cardinal treatment of the CES-D index may not be biasing the results (Figure A4 and Table A8, in the online appendix). Table A8 shows that the result is also robust to a principal component analysis of the CES-D questions, where we retain the first component and use the standardized value of this component as the dependent variable.

Fifth, we check whether the results are affected by attrition. Across 2016–2020, the VARHS sample experienced a 3% attrition. Although this is small, it may bias the results if attrition is driven by volatility in the international coffee markets. For example, if households with greater depressive symptoms were more likely to drop out of the VARHS when faced with increasing coffee price volatility, then we may be underestimating the true relationship between coffee price volatility and mental health. We investigated the extent of this bias in the following way: (i) We check if coffee price volatility drives attrition by regressing a dummy variable that takes value 1 if the household is missing in the next survey round (i.e., at $$$ t+1 $$$) on all the right hand side variables in Equation (1). We find that coffee price volatility does not significantly affect the probability of attrition from the VARHS sample (Column 1 of Table A9 of the online appendix); and (ii) we re-estimate the results using the unbalanced panel. Results reported in Columns 2 and 3 of Table A9 of the online appendix are similar to those in Table 2. Taken together, these checks suggest that our results are not biased due to attrition.

Sixth, we check whether some unobserved factors are driving the results. In addition to the vector of controls mentioned in Equation (1), we control for a host of other time-varying variables that can matter for mental health: household size, land owned, real value of livestock owned, real yearly household income, asset index, and respondent's risk aversion.19 Although these are potentially “bad controls,” comparing results with and without these variables can allay concerns that other factors are driving the results. Results reported in Table A10 of the online appendix show that the effects of coffee price volatility are qualitatively similar to those reported in Columns 3–4 of Table 2.

Relatedly, differences in current mental well-being could be due to unobserved unequal access to health care. We check if this is the case by using distance of the household to the nearest hospital (in km) as an outcome variable. Results indicate that unequal access to health care does not explain our findings (Column 1 of Table A11 of the online appendix). One may also be concerned that using the 2020 VARHS survey contaminates the results due to the COVID-19 pandemic. However, it is important to highlight that Vietnam had a tight control over the spread of COVID-19 during 2020. By the end of 2020, Vietnam had only experienced 1465 cases and zero deaths. Nonetheless, the month-year fixed effects should account for any time varying effects of COVID-19 (e.g., announcement of successful vaccine trials) on mental health. We also drop all communes that detected COVID-19 or were placed under a lockdown before the VARHS 2020 and find that the results are robust to this change (Column 2 and 3 of A11 of the online appendix).

4.3 Additional analysis

This subsection explores the implications of our main results further. First, we investigate whether the results vary by the characteristics of the respondent and the household. Second, we examine additional effects on related health and other well-being indicators. Third, we assess various mitigation strategies undertaken by households.

4.4 Mechanisms

Our central hypothesis is that volatility in coffee prices increases uncertainty of future economic prospects. The underlying assumption—validated earlier—is that farmers have limited ability to adjust the scale of coffee production or smoothen revenue by holding onto output for better prices. Farmers could also ex-ante try to stabilize revenue by (i) insuring against price movements and (ii) insuring against crop loss. Signing preharvest contracts with traders that specify the selling price has been found to be an effective method of protecting against adverse price movements. For example, in a recent study with 1200 farmers in Madagascar, Bellemare et al. (2021) find participation in contract farming to be associated with lower income variability, due to the transfer of price risk from growers to processors. However, we find participation in contract farming is very uncommon among coffee farmers in Vietnam: In our VARHS sample none of the coffee farmers report signing a preharvest contract with traders in 2016, and only nine households did so in 2020. Furthermore, of the nine households that had a preharvest contract in 2020, only two had contracts that prespecified the selling price.26 Although contract farming is uncommon, we find that farmers are more concerned about crop loss when coffee prices are uncertain. In the VARHS, farmers are asked if they would be willing to purchase insurance against crop damage/loss. In Column 1 of Panel A, Table 6, we find that a one standard deviation increase in coffee price volatility leads to a 12 percentage points significant increase in the willingness to purchase crop insurance.

Altogether, the discussion above implies that volatility in coffee prices creates substantial economic uncertainty for coffee farmers, as they do not know what revenue they can expect at the time of harvest (assuming they have a reasonable expectation of output), generating stress and adversely affecting their mental health. We now empirically test the underlying pathways discussed in the conceptual framework.

First, given the previous result that coffee price volatility increases problems with remembering things (Table 4, col 4), we note that coffee price volatility possibly increases cognitive load. Second, we investigate the role of reduced self-control by examining the effects of coffee price volatility on alcohol and tobacco consumption. In the VARHS, the respondents are asked about daily consumption of wine and beer, and whether they smoke. The results are reported in Panel A of Table 6. Column 2 shows that an increase in coffee price volatility significantly increases the probability of consuming beer daily by 14.3 percentage points (or 35% of the sample mean). We do not find any significant effects on the daily consumption of wine or the likelihood of smoking (Columns 3 and 4, respectively).27

Next, we check whether coffee price volatility leads to pessimistic expectations of future economic outcomes. In the VARHS, respondents were asked to report what quantile their household falls into based on the income distribution in their commune at the time of the survey. Following this, they are asked what position they would like their household to be at after 2 years. The difference between these two responses can be viewed as the expectation of the household's income mobility over the next 2 years, which we call “aspired income mobility.” We report the effects on this variable in Column 5 of Panel A, Table 6. Results show that an increase in coffee price volatility significantly reduces aspired income mobility.

Fourth, we examine the role of reduced social participation and social trust in depth in Panel B of Table 6. The first three columns examine the effects of coffee price volatility on the household's participation in festivals, weddings, and funerals in the community. We find that the likelihood of participating in at least one festival and a funeral in the commune in the last 12 months preceding the survey falls significantly with an increase in coffee price volatility (Columns 1 and 3).28 The next two columns examine the role of social capital. The VARHS measures interpersonal trust by asking respondents about their agreement with the statement that “most people are basically honest and can be trusted” on a 4-point Likert scale (strongly disagree, disagree, agree, strongly agree), where higher values indicate higher trust in others. Results reported in Column 4 show that an increase in coffee price volatility significantly reduces interpersonal trust. The VARHS respondents are also asked to state how many people they could turn to for help if they needed money in an emergency.29 Using this as the outcome variable in the last column, we find that an increase in coffee price volatility is associated with a significant reduction in the number of people who could help in case of a financial emergency.

Fifth, fluctuations in international coffee prices could result in a lower internal locus of control. We measure locus of control based on responses to four statements (“I feel hopeless in dealing with problems in life?”; “what happens to me in the future mostly depends on me?”; “I have little control over things that happen to me?”; “there is little I can do to change many of the important things in my life”). Responses were coded on a 1–4 Likert scale, where a higher score indicates more internal locus of control; that is, the respondents believe that events in their life are more under their control. We construct an index by summing up the responses to these four statements. Results in Column 1 of Table A14 of the online appendix show that an increase in coffee price volatility is associated with a decline in locus of control, although this is marginally insignificant ( $$$ \mathrm{p}\hbox{-} \mathrm{value}=0.105 $$$). The results for each statement are shown separately in Columns 2–5; all coefficients are negative but not statistically significant.

Last as discussed in the conceptual framework, it could be the case that coffee price volatility and the associated financial worries reduce utility from consumption thereby contributing to mental distress. We do not, however, have data to examine the importance of this pathway.