Introduction

Agricultural production is affected by preferences of farmers in several ways. Risk preference is associated with the adoption of biotechnology (Liu, 2013), diversification of operations and crop insurance (Hellerstein et al. 2013) and crop portfolio choices (Kurosaki and Fafchamps, 2002). Similarly, social preferences influence productivity (Carpenter and Seki, 2011), prosocial behavior is linked to the allocation of land rights (Yiwen and Kant, 2022), discounting is associated to farm decisions (Duquette et al. 2012) and fairness in the irrigation system influences crop choices (Buisson and Balasubramanya, 2019).

In assessing how farmers’ preferences affect agricultural decisions and long-term sustainability of agriculture, the traditional economics assumes stability of preferences over time and personal experiences (for a review please see Chuang and Schechter, 2015). Contrarily, psychological studies argue that recent personal experiences exert a larger influence on preferences (Hertwig et al. 2004; Malmendier and Nagel, 2011). The recent literature also suggests that culture and political factors in the upbringing environment of individuals influence beliefs and preferences (Guiso et al. 2004; Alesina and Fuchs-Schündeln, 2007; Malmendier and Nagel, 2011). Several economic studies also suggest that negative events and natural shocks alter preferences (for example: Cameron and Shah, 2015; Becchetti et al. 2017). Overall, the literature on preference stability does not lead to unanimous outcomes. Therefore, it is difficult to foresee with certainty how the external shocks will impact farmers’ preferences.

One of the most important shocks of this kind is the Coronavirus Disease 2019 (Covid-19). The direct effects of the pandemic have surfaced in the shape of acute health challenges, loss of lives and strained medical systems. The spillover effects of the pandemic have stalled the economic progress and raised substantial challenges specifically in the developing and underdeveloped economies (Mahmud and Riley, 2021). It is important to examine the impact of the pandemic on preferences because they fundamentally govern both individual and collective decisions and play a critical role in the aggregate growth.

Since the arrival of Covid-19, researchers have meticulously examined the influence of both direct and indirect impacts of the pandemic on preferences. The recent survey paper by Umer (2023b) nicely summarizes this rapidly expanding literature by focusing on a rich set of preferences (altruism, cooperation, trust, inequity aversion, risk, and time preferences) elicited from diverse subjects (students, Amazon Mechanical Turk (AMT) workers, traders, and general population). However, despite a rapid growth in pandemic literature, the influence of pandemic-driven agricultural shocks on preferences of farmers is quite unknown. It is vital to examine the impact of the pandemic-driven agricultural shocks on preferences because agriculture and farmers specifically in developing countries make up an integral component of the economy, farmers grow food for masses and help in fighting food security challenges that are further intensified by the pandemic, and traditional and pre-pandemic coping mechanisms are less likely to provide buffer against the ongoing challenges of farmers.

Another important aspect of studies that examined preference stability during the pandemic is their subject pool, most of them relied on students (for example: Buso et al. 2020; Lohmann et al. 2020; Shachat et al. 2021 and others) and resultantly, their external validity is quite limited. Even studies based on samples from the general population have overwhelming relied on unincentivized surveys (for example: Bogliacino et al. 2021; Cappelen et al. 2021; Heap et al. 2021; Umer, 2023a and others). The experimental literature frequently reports that behavior across incentivized and hypothetical settings differ (for example: Bühren and Kundt, 2015; Clot et al. 2018 and others). Moreover, the psychological mechanism used to reach a decision also differs under the hypothetical and incentivized decisions (Vlaev, 2012). Therefore, hypothetical decisions even though from a general population might be ineffective in predicting behavior when money is at stake.

In this study, we assess how the Covid-19 shocks to agricultural income affect farmers’ preferences and how the impact differs from natural shocks (such as bad weather, floods) to agricultural income. As both Covid-19 and natural shocks are unfavorable events that affect agricultural income, they could have similar effects on preferences.Footnote 1 On the other hand, natural shocks affect only a segment of the society and partially predictable by farmers to some extent based on their experiences, while the income shock driven by Covid-19 can potentially affect farmers on a massive scale and contain more uncertain elements. Such differences may result in different effects on farmers’ preferences. Therefore, an investigation about the comparative effects of these shocks is essential to see if both shocks alter preferences in a similar or a different manner. To improve the external validity, we attempt this with the help of an incentivized field experiment performed with 100 farmers in Pakistan in mid-2022. The Covid-19 shock is proxied by combining the impact of the pandemic on crop inputs and outputs and the sale of livestock products (milk, meat and eggs) in the last one year. The natural shock measure is constructed by combining the annual impact indicators of bad weather, floods and insect attacks on the crop output. Four preferences are examined that include impatience, risk-taking, generosity (average of donations toward an anonymous citizen, mosque and a charity organization)Footnote 2 and fairness.

Our study intends to contribute to the existing literature through several avenues. First, we will perform a comparative analysis about the impact of the Covid-19 driven and natural calamity driven agricultural shocks on farmer’s preferences. The comparative analysis will provide novel evidence on the differential impact of the Covid-19 and natural agricultural shocks on preferences. Second, we will report results from a field experiment with actual farmers, which can improve the external validity as compared to studies solely based on lab experiments. Third, we rely on incentivized field experiments and therefore, our results are less likely to be influenced by biases observed in decisions when no money is at stake. These outcomes are expected to provide useful insights on the malleability of preferences when exposed to different unfavorable events.

The rest of the study is structured as follows. The second section reports experimental details and the construction of dependent and primary independent variables. The third section examines the possible endogeneity bias in the components of the natural and Covid-19 shocks. The fourth section reports main results while the fifth section reports sub-group analysis based on the wealth and age of farmers and performs robustness checks. The last section concludes the paper.

Experimental details

With the cooperation from Zarai Taraqiati Bank (ZTBL), we conducted the field experiment as a part of a broader studyFootnote 3 in June–July 2022 with 100 farmers in the Gujranwala region of the Punjab province. Within Gujranwala there are 21 branches of ZTBL that are spread across the region. Because the 21 ZTBL branches spread widely in the Gujranwala region, following the zonal office advice we chose one typical branch (Hafizabad Branch) for the main sampling. Hafizabad is a name of TehsilFootnote 4, District, and the city where the district headquarters is located. The Hafizabad District has an area of 2367 square kilometers, a population of 1.2 millionFootnote 5 and 97% of it is Muslim, and agriculture is the major source of earning for majority of the population with rice and wheat being the primary crops. Because of Hafizabad’s proximity to larger city centers and motorway, the region is well connected to the country and is famous for rice exports.

The 100 sample farmers were recruited through three routes. First, farmers visiting the ZTBL Hafizabad branch office were randomly approached and inquired if they would like to participate in a survey, resulting in 72 observations. Second, farmers participating in the two sites where farmer’s training was being administered by ZTBL were randomly approached and invited for the survey. We recruited 6 farmers from this route. Lastly, a subset of randomly selected villages with a higher ratio of borrowers was visited by our enumerator who recruited 22 farmers. All of the sample farmers were male, and their residence was spread over 39 villages in Hafizabad District. Although our random sample of male 100 farmers is relatively small, we have farmers from 39 villages in the Hafizabad district with average land size of about 8.5 acres and current borrowers of ZTBL. Therefore, our findings can be generalized to male farmers in the Hafizabad district with a similar set of characteristics. Moreover, a sample of 100 farmers in our study is reasonable for an incentivized field experiment. Other studies with farmers from Pakistan (for example: Khan et al. 2019)Footnote 6 have also used a similar sample size for their analysis.

Implementation of experiments

We hired a local resident with a university degree to act as enumerator and implement the field experiment. The enumerator had previous experience of performing surveys with farmers, was familiar with the study area and did not know the research questions examined in this paper. We further trained the enumerator via pilot survey before proceeding to the actual work. We synchronized our field work with crop harvest time (wheat harvest) so that the farmers are free from their cropping activities and can participate in survey without any time pressure.

The farmers who agreed to participate in the survey (irrespective of the data collection point) were first briefed about the survey, average time required to complete the survey and the amount of money they can earn through their participation. The farmers were promised a fixed fee of 500 rupees for their participation in a one hour long survey irrespective of the extent of survey completed. The participation fee was about half of the daily wage and therefore enough for just an hour survey. The farmers were also informed that they can earn up to 500 rupees from the economic games at the end of the survey. The farmers provided an informed oral consent before starting the survey. At the end of the survey, farmers were paid in cash for both their participation fee (500 rupees) and the money earned in the economic games, and a signed receipt was obtained from each participant. The survey questionnaire had six major sections that elicited information about demographics and income, agricultural land, crop input and output information, agricultural assets, informal and formal ZTBL loans, miscellaneous details, Covid-19 and natural shock information and economic games. All farmers in our experiment are current borrowers of ZTBL and own a significant amount of land (about 8.5 acres on average). These farmers are therefore not credit-constrained.

Preference elicitation

The four domains of farmers’ preference were elicited through economic games. The games arranged in the same order for all farmers included time preference questions, a binary lottery choice and the bomb risk elicitation task (BRET), dictator games with three different recipients (mosques, fellow citizens, charity), and the ultimatum game both as proposer and as responder. Before the implementation of games, farmers were briefed that one of these games is already selected by the enumerator for the payment purposes and will be revealed at the end of the survey. Therefore, they should consider each game as an independent choice and carefully exercise every decision.

As respondents participated in multiple economic games, there is a possibility that the ordering of games might have influenced their behavior. However, the existing literature suggests that incentivizing one of the multiple decisions in experiments is an incentive compatible mechanism that does not distorts behavior, and minimizes the impacts of eliciting multiple decisions from same subjects (Beattie and Loomes, 1997; Azrieli et al. 2018; Azrieli et al. 2020). A recent meta-analysis by Umer (2023c) also shows that random payment in case of multiple decisions does not alter behavior. As one of the multiple games in the current field experiment was incentivized and the respondents had no prior information about that game, we believe the impact of playing multiple games and the ordering effects of those games are likely to be trivial in our data.

Risk taking

A binary lottery choice and Bomb Risk Elicitation Task (BRET) were used to elicit risk taking. The binary lottery had two options; 500 rupees sure payment or 750 rupees with 0.5 probability and 250 rupees with 0.5 probability. As the binary lottery choice does not show variation (only one of the 96 respondents selected the risky option), we do not use it in the analysis. In BRET, the total number of boxes is 100 including 99 boxes with monetary reward and one box designated as the empty box (we did not use bomb to avoid negative connotations). The payoff for each box opened without picking the empty box was five rupees and risk increased in the number of boxes opened. The farmers were explained the task and monetary outcomes with the help of a matrix containing 100 boxes and two examples; in the first example farmers did not pick the empty box, and in the second example they picked the empty box. In both examples the location of the empty box remained fixed, while the experimenter changed the number of boxes opened. We relied on BRET instead of other risk-elicitation tools because it is relatively easier to explain to the farmers, requires only one decision, and therefore, saves time and keeps the experiment brief.

Prosocial behavior/generosity

We used dictator games with three different recipients (any mosque, fellow citizen (stranger), and a charity organization called Edhi Foundation) to elicit prosocial behavior. The prosocial behavior in the dictator game is sensitive to the recipient type (see meta-analysis by Umer et al. 2022). Therefore, to eliminate any recipient dependent bias in prosocial behavior, we use three recipients.Footnote 7 While fellow citizens and charity recipients are common in the dictator game experiments, not many studies use mosques as recipient. We incorporated any mosque as a recipient because Pakistani citizens have a higher trust in donations channeled through mosques as compared to the civil organizations and charities.Footnote 8 The size of the monetary pie in each of the three dictator games was 500 rupees and farmers could donate any amount in positive integers. We use average of the three donated amounts as a proxy for pro-social behavior.

Fairness

We used an ultimatum game to elicit fairness. The farmers acted both as proposer (division of 500 rupees with an anonymous responder) and as responder (minimum fraction of 500 rupees acceptable from an anonymous proposer).

Impatience/time preferences

We elicited time preferences through a standard question:

500 rupees tomorrow or 525 rupees after a month

The time preference question can reflect present bias or higher discounting, and we cannot distinguish them in our data. Ideally, we should use more questions to isolate these two impacts, but practically, it was difficult to add further questions because our main survey was already too long. In fact, based on pilot surveys, we had to eliminate several questions to avoid mentally fatiguing farmers to a point where they either leave the survey unfinished, or impact the quality of the elicited information.

Covid-19 and natural shocks elicitation

Prior to the elicitation of preferences, we collected information about the impact of Covid-19 and natural shocks in the last one year on different aspects of agriculture.Footnote 9 We used a one-year window to mitigate the possibility of recall bias. The existing studies (for example: Beegle et al. 2012) show that a recall period of approximately one year does not lead to a significant upward or downward bias in the agricultural information obtained from farmers. Moreover, recall bias for the Covid-19 and natural shocks is expected to be minimal because both are salient events and more likely to stay fresh in the farmer’s mind. As it might be relatively difficult for farmers to precisely quantify the impact of shocks on the amount of agricultural earnings, we focused on qualitative information corresponding to different channels through which negative shocks can cause possible fluctuations in the agricultural earnings.Footnote 10 We used four questions to elicit the impact of the Covid-19 and three questions to elicit the impact of natural shocks on the agriculture. All seven questions were binary (1 = Experienced the impact in the last one year; 0 = No impact). The summary statistics of responses for all these seven items are reported in online Appendix 1. We aggregate these questions into two indices, one for the Covid-19 shock and the other for the natural shock.

Covid-19 shock

Negative impact of the Covid-19 on: (1) availability of crop inputs (seeds, fertilizers, etc.) (2) crop sales, (3) crop prices, and (4) sale of milk, meat, and eggs (livestock products).

Natural shock

Impact of (1) bad weather, (2) floods, and (3) insect attack on crop production.

To obtain shock indices we first standardized every item (deducted mean value and then divided by the standard deviation) and obtained an average value based on the four standardized items for the Covid-19 shock index and three standardized items for the natural shock index. As a last step, we standardized the indices once again to obtain comparable effect sizes across the two shock indices.

While individual items used to generate the two shocks are different, it is important to note that at the core each item elicits information about the impact of shock on the agricultural production that ultimately translates into the financial wellbeing of the farmers. Therefore, both shocks have similar end-result in the form of fluctuations in the farmer’s wellbeing. Furthermore, natural shocks such as floods, insect attacks and bad weather have a direct influence on crops while the Covid-19 shock indirectly influences crop by disrupting both input and output markets. Therefore, to precisely capture the impact of the Covid-19 and natural shocks, we had to rely on a different but pertinent set of questions for each of these shocks.Footnote 11

All farmers in our sample reported agriculture as their main source of income, and grow rice and wheat as their main crops. We synchronized our field experiment with the harvesting time of wheat. A 1-year window used in the shock elicitation questions covers the sowing as well as harvesting of rice and wheat crops and encompasses the impact of Covid-19 induced input and output price shocks as well as natural shocks on both rice and wheat. Therefore, if a farmer responded “Yes” to the impact of Covid-19 on crop input, it shows a negative impact on the production decisions of either rice or wheat, or both crops in the last 1 year. A similar setting applies to the crop production related components of the Covid-19 and natural shocks. Due to the nature of questions used for shocks elicitation, we cannot identify the intensive margins for the impact of shocks on preferences.

There are three agricultural markets in Hafizabad (two are fully functional while one has a limited business). One can expect that shocks to these agricultural markets driven by Covid-19 would have a similar impact on all farmers. However, the observance of Covid-19 protocols varied across the three agricultural markets. Resultantly, the supply of crop inputs and their prices, and the purchase of crop output were not consistent across the three markets. This is the main cause of heterogeneous impacts of market driven agricultural shocks due to the Covid-19.

There is a possibility that farmers mistakenly reported health impacts of the pandemic as the Covid-19 agricultural shock. The severity of the Covid-19 in Hafizabad district even during the peak of the pandemic in Pakistan was very low (one infected person per square km and ten deaths per million persons) compared to the urban districts (Saddique et al. 2021). Therefore, health impacts of the pandemic in Hafizabad are unlikely to contaminate the Covid-19 agricultural shock. Furthermore, our instruments of the Covid-19 shock specifically inquire about the agricultural outcomes and do not discuss health outcomes at all. As a result, we expect that the Covid-19 agricultural shock is less likely to be contaminated with the health-related aspects of the pandemic. On the other hand, the effects of macroeconomic policy to combat the economic impacts of the Covid-19 might be reflected in our Covid-19 shock variable.

Empirical specification and data summary

We estimate the impact of the Covid-19 and natural shock on preferences using the following OLS regression.Footnote 12

$${{Preference}}_{i}={\beta }_{0}+{\beta }_{1}{{Covid}}\; {{shock}}_{i}+{\beta }_{2}{{Natural}}\; {{shock}}_{i}+{\beta }_{3k}\,{X}_{i}+{\epsilon }_{i},$$
(1)

where \({\rm{Preference}}_{i}\) is one of the empirical measures representing the four preference domains (generosity, fairness, risk-taking, patience). Covid Shock and Natural Shock are the primary explanatory variables of interest.Footnote 13 A negative and significant β1 and β2 would reflect a negative impact of exposure to these shocks on preferences. The vector Xi contains control variables while ϵi is the error term. We estimate equation 1 without any controls and with two sets of controls. The first set includes controls for the demographic variables, religious differences and agricultural experience while the second set (extended control) adds information about pre-shock wealth (value of inherited land). In regressions with extended control, we restrict our analysis to those farmers who inherited land and did not purchase any land in the last 1 year (only one farmer purchased land in the last 1 year).

Table 1 provides the data summary and statistics for the dependent variables and all controls. Regarding risk preferences, we do not use data from the binary lottery task because it is highly skewed towards risk-averse choice. Regarding pro-sociality, we use the average amount donated to the three recipients as the most representative measure of generosity.Footnote 14 We normalize risk-taking (boxes opened divided by 100), generosity and fairness (amounts divided by 500) to make the coefficient interpretation easier. In online Appendix 2 we report the distribution of choices in the economic games.

Table 1 Data summary.
Full size table

Our sample consists of relatively older farmers with an average age of 54 years and an average of 9 years education. The average family size is also relatively large (about 16 members). We do not include information about annual income, agricultural and livestock assets in regressions because these can be influenced by shocks and lead to endogeneity in regressions. However, we do incorporate the value of inherited land owned by the farmers as an additional control for wealth. We restrict the wealth analysis to farmers (n = 99) who inherited their land and did not purchase it in the last 1 year. As land inherited is a pre-shock asset, it is highly unlikely to be influenced by shocks.

Examination of endogeneity bias

The natural shock consists of exposure to flood, bad weather and insect attacks. There is an obvious concern that people exposed to floods and those who remain safe might be different from each other. For example, it is possible that wealthier farmers can shift to places where floods are unlikely to impact agriculture and subsequently take more risk because of their wealth (Cameron and Shah, 2015). The evidence from Pakistan also shows that farmers with larger land holdings (wealthier farmers) are better able to mitigate the negative effects of floods on consumption smoothing (Kurosaki, 2015) and the impact of flood can be idiosyncratic at the household level even within a village (Kurosaki and Khan, 2011). In the presence of this endogeneity, our estimates for the impact of natural shock on risk-taking could be biased and render us unable to specify the precise impact of the natural shock on risk-taking and other preferences as well.

To examine the possible endogeneity bias, we first examined whether farmers exposed to floods are any different from those unexposed in their observable characteristics. We only find a significant difference between the family size of the farmers exposed and unexposed to floods (Table 2). For all other variables we do not find any significant differences between the two groups.

Table 2 Comparison of farmers exposed to with those unexposed to flood.
Full size table

To further examine whether wealthier farmers can mitigate floods by moving to safer locations or higher altitude areas, we regressed exposure to floods variable on the farmer’s aggregate wealth. We added the current values of livestock, agricultural assets and land owned to obtain the aggregate wealth of farmers. For brevity reasons, we only provide the output for aggregate wealth in Table 3 (Panel A). The coefficient for aggregate wealth is insignificant in all regressions without and with controls and indicates no endogeneity bias.

Table 3 Examination of Selection Bias.
Full size table

As floods can diminish wealth, ideally, we should use pre-flood wealth as the primary explanatory variable to examine whether wealthier farmers are able to escape floods (Cameron and Shah, 2015). While we do not have complete information about pre-flood wealth, we do have information about purchase of the agricultural land in the last one year. All farmers except one in our data inherited their land (99% made no purchases in the last one year) and therefore, the value of land is a useful proxy for wealth accumulated prior to floods. Using value of land as the primary explanatory variable and restricting the sample to those farmers who inherited land, we re-examine if the wealthier farmers can escape floods. Again, we do not find a significant coefficient for the value of land (Table 3: Panel B) and in fact the coefficients are like the ones reported in Panel A. Overall, we do not find any evidence to suggest that wealthier farmers are able to escape floods.Footnote 15 As our sample size is relatively small, we examined the robustness of the coefficients by using bootstrapped standard errors (with 50 replications which vary by sample size, minimum 48 replications). The significance of the relevant coefficients remains same.

While bad weather and insect attacksFootnote 16 are likely to be exogenous covariates and influence almost everyone in a small region where we conducted experiments, there is also a possibility that these elements are idiosyncratic at the household level. For example, rainfall and climate can show variations even within a couple of kilometers (Graef and Haigis, 2001) which in turn can influence breeding of insects. Resultantly, even exposure to bad weather and insect can differ at the household level and might cause endogeneity bias if they are dependent on the household characteristics or wealth. We therefore analyzed whether bad weather and insect attacks have an endogenous aspect by repeating the endogeneity tests that we did for floods (output in the online Appendices 3–6). We only find marginal evidence that farmers exposed to bad weather have lower levels of education and those exposed to insect attack are predominantly Sunni sect. However, both variables have nothing special to do with the weather and insect attack. Wealth or value of inherited land has no significant influence on exposure to bad weather and insect attacks. Overall, we do not find any significant evidence for endogeneity bias for all three components of the natural shock.

Although Covid-19 is an aggregate shock by definition and hence very likely to disrupt the agricultural inputs and sales in the study region, how individual farmers were affected by the disruptions is idiosyncratic at the household level for several reasons. First, a subset of farmers with enough resources might have purchased the crop inputs in advance to minimize the anticipated impact of the Covid-19 on inputs. Second, it is quite possible that a subset of farmers either sold crops in their village or stored the output and hence avoided the disruptions in the agricultural markets. However, all farmers in our sample sold at least one crop in the market in the last one year and therefore, we are certain that the Covid-19 driven fluctuations in crop sales and prices affected all farmers in our sample, but with different degrees of intensity. Third, farmers close to the city center under lockdown restrictions might have been affected differently from the farmers farther from the city center. These lockdown restrictions might have affected both cropping activities as well as sale of crop output and livestock products. Due to these reasons, we expect the Covid-19 shock to be idiosyncratic at the household level and hence the response to the Covid-19 shock variables can have variation.

As household level impact of the Covid-19 and coping strategies can be different, there might be some endogeneity in the four instruments used to construct the Covid-19 shock. Therefore, we analyzed the possible bias by following the procedures used to check the endogeneity in the instruments of the natural shock and report the complete output in online Appendices 7–14. We do find evidence that farmers who reported the impact of the Covid-19 on inputs are relatively younger with lesser years of education and agricultural experience and located a bit further from the city center. Similarly, farmers who reported the impact of the Covid-19 on crop sales also have lower education, located farther from the city center and have lower wealth while those reporting the effect on crop prices are located farther from the city. Lastly, farmers who reported the impact on livestock products have a lower wealth, but the difference is insignificant.

Overall, the evidence suggests that wealthier farmers and those located closer to the city center are less influenced by the Covid-19 driven agricultural shocks. Therefore, to control for these differences, we add all these variables in the regressions. We also perform a sub-group analysis to partially control for endogeneity that can be caused by wealth and age and report it in section 5.

Results

We report results for the main explanatory variables in the text and complete output in appendices. Table 4 summarizes the impact of shocks on preferences while online Appendices 15–17 report complete regression results. Regression 1 is without controls, regression 2 and 3 are with controls and extended control, respectively. As we incorporate an additional control (value of land purchased prior to the last one year) in regression 3, we restrict regression 3 to those respondents who did not purchase land in the last one year (only one observation is dropped). We follow this pattern in the subsequent sections except in the pre-shock wealth-based comparisons, where it is natural to restrict all three regressions to only those respondents who did not purchase land in the last one year. We consider the coefficient to be different from zero if at least two of the three regressions lead to significant outcomes.

Table 4 Impact of Shocks on Preferences.
Full size table

We do not find any significant impact of the Covid-19 shock on risk taking and fairness (Table 4: Panels A, C, and D) in any of the three regressions. However, we do find a negative impact of the Covid-19 shock on generosity (Panel B, one standard-deviation increase in the severity of the Covid-19 shock reduces generosity by 3–5 percentage points, or 12–17% in comparison to the mean of the dependent variable, depending on the specification, ceteris paribus) and a negative effect on impatience (Panel E, one standard-deviation increase in the severity of the Covid-19 shock reduces impatience by 12 percentage points, or 20% in comparison to the mean of the dependent variable, ceteris paribus) of the farmers.Footnote 17 Both these effects remain significant even when we control for wealth (value of land owned) accumulated prior to the last 1 year. The size of these coefficients is large and economically significant as shown in their relative magnitudes.Footnote 18

The exposure to natural shock has a negative and significant impact on risk-taking (Table 4: Panel A, one standard-deviation increase in the severity of the natural shock reduces the ratio of boxes opened by 2 percentage points, or 9–10% in comparison to the mean of the dependent variable, ceteris paribus), on minimum acceptable amount in the Ultimatum game (Panel D, one standard-deviation increase in the severity of the natural shock reduces minimum acceptable offers by 2 percentage points or 3% in comparison to the mean of the dependent variable, ceteris paribus), while it has no significant impact on other preferences.

Discussions

As reported in Table 4 (Panels B and E), we find a negative impact of the Covid-19 shock on generosity and impatience. The negative impact of the Covid-19 shock on generosity can be due to the reduced agricultural earnings of farmers. Moreover, it is also possible that the shock made farmers more cautious about future uncertainties in the agricultural markets, and to smoothen their consumption they control their current spending. This delayed spending also appears to be a plausible explanation for reduced impatience with exposure to the Covid-19 shock. The current findings are consistent with several pandemic studies. For example, Brañas-Garza et al. (2022) report decreased generosity while Meunier and Ohadi (2021) report increased patience post pandemic. Similarly, Angrisani et al. (2020), Lohmann et al. (2020), Drichoutis and Nayga (2021) and Heap et al. (2021) find no change in risk-taking.

In Table 4 (Panel A) we find a negative impact of the natural shock on risk taking. The increased risk aversion with exposure to the natural shock can be driven by at least two possible channels. First, exposure to the natural shock can increase the perceived likelihood that the negative shocks will occur again. Second, the current exposure to the natural shock can increase fear of the negative events. Both these factors can reduce risk-taking (Cassar et al. 2017).

The existing literature also suggests that risk-taking can alter due to the changes in wealth or income accompanied by the natural shocks (Cameron and Shah, 2015). However, even when we control for the pre-shock wealth (Table 4: Panel A, Regressions 3) the negative impact of exposure to the natural shock on risk-taking persists and remains significant. This suggests that the exposure to the natural shock itself alters risk-taking possibly by increasing the perceived likelihood and fear of such negative events. These findings are consistent with the work of Cameron and Shah (2015) who also obtain a negative effect of exposure to floods on risk-taking even when pre-flood wealth is added to the analysis. Overall, the negative impact of shocks on risk-taking is largely consistent with the findings from the existing studies. For example, Cameron and Shah (2015) find that exposure to flood or earthquake, Cassar et al. (2017) report that exposure to tsunami while Sakha (2019) find that exposure to agricultural shocks leads to increased risk-aversion. Similarly, Kurosaki et al. (2022) also find that exposure to violence leads to enhanced risk-aversion in the former FATA region of Pakistan.

The decrease in minimum acceptable offer in the Ultimatum game due to natural shock (Table 4, Panel D) can be due to an increased salience of money (Abatayo and Lynham, 2020). Moreover, negative shocks can make people interpersonally more pleasant (Burton et al. 2010), and hence can decrease their desire to punish unfair behavior. Overall, the negative impact of natural shocks on the minimum acceptable offers resonates with recent work by Abatayo and Lynham (2020), who also find that exposure to natural shock (typhoons) eliminates rejections by responders in the Ultimatum game.

Overall, the current findings show that each of the Covid-19 and natural shock has a differential impact on preferences. While the Covid-19 shock decreased generosity and impatience, natural shock decreased risk-taking and fairness concerns.Footnote 19 The differential impact of these shocks on preferences is most likely due to the differential underlying mechanisms; the effect of the Covid-19 shock is transmitted primarily through market-driven forces while natural shocks have a direct influence on crops. Also, Covid-19 is an aggregate shock on a much larger scale compared to the natural shocks mostly concentrated in a region. The comparative analysis also makes it evident that the natural and pandemic driven agricultural shocks might not necessarily have an identical influence on preferences. Therefore, we need to exercise caution while predicting the impact of a pandemic on behavior based on the studies that examined other natural shocks on preferences.Footnote 20

Further analysis and mechanisms

To infer the potential mechanisms underlying the findings in the previous section, we report results from further analysis. First, we examine whether the relations were heterogeneous depending on the characteristics of farmers. Second, we try a dependent variable different from direct measures of preferences and examine whether natural disaster and Covid-19 shocks affect it. Third, we run several robustness checks. The results are reported below.

Are preferences of wealthier farmers less susceptible to shocks?

The existing literature suggests that poverty via stress can influence the revealed preferences (for a review of this literature please see Haushofer and Fehr, 2014, and Mullainathan and Shafir, 2013). Therefore, farmers with relatively lower wealth might be more susceptible to preference alterations if exposed to the negative shocks because shocks have a high tendency to enhance stress levels. The endogeneity analysis reported in section “Examination of endogeneity bias” also provided some evidence that wealthier farmers are less susceptible to the Covid-19 shock. Therefore, to examine the role of wealth in coping with the natural and Covid-19 shocks, we performed further analysis based on their value of inherited land owned (pre-shock wealth proxy). We construct four variables of Covid 19 shock (c_shock) and natural shock (n_shock) measures interacted with the dummy variable corresponding to whether the land value is larger than its median value. As pre-shock wealth is our main source of sub-group formation, we drop one respondent who purchased land in the last 1 year in all regressions to perform clean comparisons. We report main findings in Table 5 and provide complete results in online Appendix 19–21.

Table 5 Impact of shocks on preferences: pre-shock wealth based analysis.
Full size table

The impact of natural shock on generosity (Table 5, Panel B) and time preferences (Panel E) is insignificant across wealthy and less wealthy farmers. The negative and significant effect of natural shock on risk-taking is observed only for wealthy farmers (Panel A), the coefficients for wealthy and less wealthy farmers are however not significantly different from each other, as per F-stat (natural shock) reported at the end of Panel A. We also find that natural shock has a positive impact on proposed amount in the Ultimatum game for wealthy farmers, and has a negative impact on less wealthy farmers (Panel C). The coefficients across the two groups are also significantly different (F-stat). Similarly, only for wealthy farmers, we find that natural shock has a negative impact on the minimum acceptable amount by responders (Panel D). The findings for the Ultimatum game show that money is relatively less salient for wealthy compared to less wealthy farmers.

The wealth-based analysis for the Covid-19 shock shows no significant impact on risk preferences of wealthy and less wealth farmers (Panel A). The Covid-19 shock still shows a negative and significant impact on generosity irrespective of the wealth group (Panel B) and the F-test also corroborates that the coefficients for the two wealth groups are statistically indistinguishable. Similarly, there are no significant wealth dependent differences in the proposer behavior in the Ultimatum game (Panel C), while the Covid-19 shock has a positive and significant impact on the minimum acceptable amount of less wealthy farmers (Panel D). The F-stat however does not provide enough evidence to suggest that the coefficients for wealthy and less wealthy farmers for the Ultimatum game are different. Lastly, the Covid-19 shock has a significant and negative impact on impatience of only wealthy farmers (Panel E). Again, F-test does not provide evidence that the coefficients for impatience are different for wealthy and less wealthy farmers.

The sub-group analysis overall does not provide strong evidence to suggest that preferences of relatively less wealthy farmers are more malleable to unfavorable shocks. One possible reason for the absence of this effect is that farmers in our sample own a significant amount of agricultural land (average land owned = 8.5 acres), are relatively well-off and therefore incomparable to the typical poor people examined in the previous studies.

Are preferences of older farmers more susceptible to change due to shocks?

The existing literature suggests that life experiences can influence preferences (Alesina and Angeletos, 2005; Alesina and Fuchs-Schündeln, 2007; Chuang and Schechter, 2015). As older farmers are more likely to have faced such experiences, their preferences relative to younger farmers might be more prone to change when struck with negative shocks. We examined this possibility by using median age (53 years) as the cutoff. We construct four variables similarly as before. The main findings are in Table 6 while the complete output is in online Appendix 22–24.

Table 6 Impact of shocks on preferences: age based analysis.
Full size table

In case of natural shock, the coefficients for all preferences (except risk-taking) are insignificant for both older and younger farmers. Only in case of risk taking, we find a significant and negative impact of natural shock on older farmers (Panel A). For the Covid-19 shock, we do not find significant impact on risk and time preferences of both older and younger farmers. The Covid-19 shock however reduces generosity of older farmers and has insignificant impact on the generosity of younger farmers (Panel B). Furthermore, the Covid-19 shock has positive (negative) impact on the amount shared by younger (older) farmers (Panel C), and has a positive (no significant) impact on minimum acceptable amount by older (younger) farmers in the Ultimatum game (Panel D). These findings from the dictator and ultimatum games together suggest that the Covid-19 shock has increased salience of money only for the older farmers. The findings are particularly stronger in the Ultimatum game; F-stat proves that the coefficients for younger and older farmers afflicted with Covid-19 shock are significantly apart. One possibility for the age contrast on generosity could be that pre-Covid-19, older farmers were more altruistic than the younger ones, and therefore, the room for decline in altruism is larger for them. However, as we did not measure altruism pre-Covid-19, we cannot judge the validity of this possibility.

Overall, we find very weak evidence that preferences of older farmers are more likely to change due to the natural shocks. However, the differential impact of the Covid-19 shock on older and younger farmers provides support that preferences of older farmers are more malleable to the Covid-19 shock. It is important to mention here that our sample is skewed toward older farmers (average age is almost 54 years) and the youngest farmer in our sample is 35 years. Therefore, we do not have typical young farmers in our sample and this might be a possible reason for absence of stark differences in the effect of natural shocks among relatively older and younger farmers in our data.

Do shocks influence agricultural attitudes?

Attitudes are negative or positive feelings or beliefs about a certain object and have been used frequently to explain and predict the behavior of farmers (Dimara and Skuras, 1999). It is possible that the negative shocks indirectly influence farmer’s agricultural choices by influencing their agricultural attitudes. To examine this possibility, we employ the data on answers to the following instrument.

(1) Do you want your children to continue agriculture? (1 = Yes; 0 = No) [Mean = 0.69]

This instrument elicits information about intergenerational transfer of agricultural activities and shows reasonable variation. We examined the impact of shocks on this attitude and do not find any significant impact (main output in Table 7, complete output in online Appendix 25). The findings suggest that the attitude towards intergenerational agricultural transfer is quite resilient to both types of shocks and reflects the strong desire of farmers to pass down the agricultural activities to the next generations.

Table 7 Impact of shocks on agricultural attitude.
Full size table

Are results robust to a different specification of shocks?

We performed additional checks to examine whether results are robust if the Covid-19 and natural shocks are constructed in a different manner. We redefined shocks by aggregating all components of the Covid-19 related shock (Covid-19 shock = Impact on crop inputs + crop sales + crop prices + sale of milk, meat, and eggs) and natural shock (Natural shock = bad weather + floods + insect attack) and used these as primary explanatory variables in regressions. Most findings remain consistent with the main findings reported in Table 4; the Covid-19 shock has a negative effect on generosity and impatience while the natural shock has a negative effect on risk taking (output in online Appendices 26–28).

Adding the interaction of Covid-19 and natural shocks

As a last part of the analysis, we added the interaction term of the Covid-19 and natural shocks to the analysis to examine if both shocks together influence preferences in an additional way (output in online Appendices 29–31). The individual impact of the Covid-19 and natural shocks on preferences remains similar to the one reported in Table 4. The interaction term is insignificant in all cases (insignificant coefficient in at least two of the three regressions) except one: the interaction term has a negative and significant impact on impatience (online Appendix 30). The impact of the interaction term is in the direction of strengthening the impact of the Covid-19 shock.Footnote 21

Conclusions

The current study reported a systematic comparison of the effect of agricultural income shocks driven by either the Covid-19 or natural calamities on four preferences (patience, generosity, risk-taking and fairness) of farmers. Despite similar end results of both shocks in the form of wellbeing fluctuations, we find that each of the shocks influenced different preferences. The Covid-19 shock had a negative impact on generosity and a positive impact on patience while the natural shock had a negative impact on risk-taking and minimum acceptable offers in the Ultimatum game. We also analyzed whether these shocks have dissimilar impacts across the relatively wealthy and poor and between relatively older and younger farmers. We do not find conclusive evidence to suggest that preferences of relatively poor farmers are more malleable to shocks and find weak evidence that older farmers showed significant decrease in risk taking only when exposed to the natural shock. In case of the Covid-19 shock, the evidence is mixed. We also examined if these shocks influence agricultural attitude and do not find any evidence to support this preposition. Overall, our findings show that the Covid-19 and natural shocks have a differential impact on preferences, and this perhaps is due to the underlying heterogenous mechanisms of these shocks.

The current findings make a novel contribution to the literature by reporting a comparative analysis of the Covid-19 and natural shocks on preferences. The analysis shows that these shocks impact different preferences, and therefore, we need to measure both to assess their heterogenous impact on preferences. The results also suggest that in future as well, we would need to measure other detrimental shocks like the Covid-19 to predict precise changes in behavior. Moreover, the heterogeneous impact of the Covid-19 and natural shocks on preferences can modify the agricultural decisions of the farmers in the short as well as long run, and therefore, might change their growth trajectories. Overall, we expect these findings will provide relevant and useful information to researchers examining the heterogenous impacts of shocks on preferences, and hopefully, spur more research in this direction.

We close the paper by highlighting some of the possible caveats in our work. The Covid-19 shock used in the study might be contaminated with Covid-19 related health and macroeconomic policy variations. Therefore, the Covid-19 shock can to some extent capture the impact of these variables. Second, the mechanism we suggested from the age and wealth heterogeneity analysis is only suggestive. Age and wealth can proxy other factors. Identifying the mechanism through further experiments is left for future research. Third, we study the impact of exposure to shocks on preferences in the short run. Whether and to what extent these changes prevail in the long run is difficult to predict based on the existing data and would require further research. Fourth, our findings are based on male farmers only, and therefore, outcomes might change if we use data containing either both male and female or female farmers only. Lastly, we examined the impact of only negative shocks on preferences. Whether positive shocks will have an opposite impact on preferences cannot be extrapolated from the current data, and it is a potential future research avenue.