WU Hai-xia ,SONG Yan ,YU Le-shan ,GE Yan

1 Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing 100081, P.R.China

2 Sino-Danish College, University of Chinese Academy of Sciences, Beijing 101408, P.R.China

3 School of Public Policy and Management, University of Chinese Academy of Sciences, Beijing 100040, P.R.China

4 International Business School, Shaanxi Normal University, Xi’an 710100, P.R.China

5 School of Public Finance and Taxation, Central University of Finance and Economics, Beijing 100081, P.R.China

Abstract Based on the microdata of 705 wheat farmers in the Loess Plateau,this study empirically analyzes the impact of uncertainty on farmers’ adoption of innovative seeds using a field experiment.The results indicate that farmers are generally ambiguity-averse and risk-averse.In addition,farmers with higher ambiguity aversion and risk aversion are less likely to adopt innovative wheat seeds,where their risk aversion plays a dominant role.Enhancing information access will alleviate the negative influence of ambiguity aversion on farmers’ adoption of innovative seeds,and interlinked insurance and credit contracts will be beneficial to ease the adverse effect of risk aversion on the adoption of innovative wheat seeds.Meanwhile,heterogeneity analysis reveals that the inhibitory effects of ambiguity aversion and risk aversion on innovative seed adoption are more significant among farmers with lower education and household income.The government can establish both ex-ante and ex-post relevant guarantee mechanisms to help farmers preferably cope with various uncertainties in the production process,remitting farmers’ ambiguity aversion and risk aversion to enhance new agricultural technology adoption rates.

Keywords: ambiguity aversion,risk aversion,technology adoption,field experiment

1.lntroduction

Complex and diverse geographic and climatic conditions augment the frequency of natural disasters and the systemic risks of agricultural production in China (Du and Wang 2020;Xu and Sun 2022).Compounded by the long agricultural production cycle and insufficient risk management tools (Caoet al.2019;Tanget al.2019),the huge production losses resulting from natural disasters are difficult to disperse,which even causes farmers to fall into poverty traps (Nakano and Magezi 2020),particularly for the small householders.A strand of studies has shown that the adoption of new agricultural technologies is one of the important ways for rural households to deal with natural risks,enhance farming income,and stabilize agricultural production (Ragasaet al.2021;Yang Det al.2021;Zhenget al.2022),which is essential for improving agricultural production efficiency,ensuring food security,and alleviating rural poverty (Mikulaet al.2020;Adebayoet al.2022).

Among agricultural technologies,innovative seeds are attracting increasingly widespread attention (Xu and He 2022;Wu and Li 2023;Yuet al.2023).The Chinese government has been committed to developing seed technology.Therefore,the importance of seed innovation,which contributes to more than 45% of China’s increase in grain production,cannot be overstated (http://www.gov.cn/xinwen/2020-12/19/content_5571094.htm).On the other hand,how to stimulate farmers’ willingness to adopt innovative seeds has become a key issue in the face of high costs as well as the intensification of seed wars in various countries.This study will focus on farmers’innovative seed adoption behaviors and provide valuable suggestions for innovative technological diffusion.

Substantial studies have shown that the effects of policy interventions (Yaoet al.2018;Wang and Yang 2021),agroecological environment (Mariano 2012;Qiet al.2021;Wanget al.2021),information access(Uwanduet al.2018;Campenhout 2021;Timothyet al.2022),socioeconomic status (Sekabiraet al.2022;Ullahet al.2022) and individual’s characteristics (Montes de Oca Munguiaet al.2021;Okelloet al.2022) on farmers’technology adoption behavior have been a hot issue of concern in agricultural technology adoption research and has generated relatively consistent findings.For instance,government penalties and regulatory measures(Wang and Yan 2022) and economic subsidies (Yang X Jet al.2021) can enhance farmers’ acceptance of new technologies.Socioeconomic status (Chenet al.2022),the scale of operation (Jianget al.2018),condition of information (Shiferawet al.2015;Magruder 2018),and information access (Perosaet al.2021;Veettilet al.2021)have positive effects on agricultural technology adoption.Individual and household characteristics such as farmers’age,education level,agricultural income,and household size can facilitate new technology adoption (Foguesatto and Machado 2021;Okyere and Ahene-Codjoe 2022).

Nonetheless,most extant theoretical and empirical literature concerning farmers’ technology adoption has been conducted under certain conditions.As a result,the research conclusions always lack practical guidance to actual production (Isik and Khanna 2003;Jinet al.2019).Given the increased weather extremes (Marmaiet al.2022;van Tilburg and Hudson 2022),ambiguous technological inputs and outputs (Tevenart and Brunette 2021),and global trade fluctuations (Zhouet al.2019;Denget al.2022),farmers’ production decisions are often made under uncertainty.Hence,it is obvious that individual attitudes toward uncertainty play a crucial role in the perception,diffusion,as well as adoption of agricultural technologies (Freudenreich and Musshoff 2022;Wuet al.2022).

Individuals’ uncertainty aversion can be classified as ambiguity aversion and risk aversion,which are related to the probability distributions of the outcomes of uncertainty events (Barhamet al.2014;Aliet al.2021;Priyo and Nuzhat 2022).As individuals make decisions in the context of uncertainty,external factors like lack of information and internal factors like cognitive level,beliefs,and emotions influence their attitudes towards uncertain events (Heath and Tversky 1991;Pulford and Colman 2007;Fairleyet al.2022).Barhamet al.(2014)and Qiuet al.(2020) point out that farmers are ambiguityaverse and risk-averse,which brings in differential effects on farmers’ technology learning and adoption.Consequently,it is necessary to distinguish between ambiguity aversion and risk aversion among farmers and explore their differences in the mechanisms and impacts on the adoption of agricultural technologies.

Compared to existing studies,the academic contributions of this paper are as follows: First,new technology adoption is a typically risky decisionmaking behavior,as the benefits of implementing new technologies are uncertain.This paper analyzes how uncertainty affects farmers’ agricultural technology adoption decisions by distinguishing between ambiguity and risk.Second,to overcome the subjectivity of questionnaire research,we conducted a field experiment to measure farmers’ ambiguity aversion and risk aversion as well as their technology adoption behavior.By taking real farmers as the subjects and using experimental information from real situations,we tackle the lack of external validity of questionnaire research,and the experimental results more precisely capture the uncertainty aversion and adoption willingness of farmers.Third,part of the research confirms the hindering effect of uncertainty aversion on farmers’ adoption of technology,but the mechanism of influence has not yet received extensive attention and validation.Therefore,we incorporate information access,interlinked insurance,and credit contracts into the analytical framework to explore the mechanisms for alleviating farmers’ uncertainty aversion,then promote innovative technology adoption.

The rest of the paper is organized as follows.Section 2 presents the conceptual framework.Section 3 describes the data and methods,including sample selection,field experiment design,data description,and econometric models.Section 4 analyzes and discusses the empirical results.Finally,Section 5 discusses the conclusions and policy implications.

2.Conceptual framework

2.1.The effect of ambiguity aversion on agricultural technology adoption

Assuming that the production decisionxis made in the statee,with a benefitπ(x,e),and the distribution ofe,which explicitly depends on the parameterv.Based on the expected utility theory,farmers are inclined to maximize the utility that the decision is carried out,which is defined as {Ee/vU(π(x,e)):x?X}.First,we consider the case in which the probability distribution of the payoffsvis unknown,the probability distribution of which is assumed to beG(v).The farmer chooses the production decisionxto maximize the following eq.(1):

where in eq.(1),h[·] is a strictly increasing function(Barhamet al.2014).Assuming that farmer is ambiguityaverse,h[·] is a concave function according to Klibanoffet al.(2005) and Neilson (2010).

Referring to Barhamet al.(2014),for a givenx?X,we use the notationEvto replace the notationEv(v) and define the certain amount of cashR(x) as the ambiguity premium,satisfying the following eq.(2):

where in eq.(2),Ra(x) represents the cash that the farmer is willing to pay to eliminate the ambiguity by the new technology.We allow the parametervto be replaced by its mean valueEv,which indicates thatRa(x) measures the cost of the ambiguous part of the uncertainty cost.This portion of the cost is determined by both the ambiguity exposure (determined by the distribution of functionG(v)) and the ambiguity aversion (determined by the curvature ofh[·]).

Rational farmers,while making production decisions in an uncertain environment,need to compare new and traditional technologies,from which they choose the technology to maximize eq.(2).Under a certain degree of ambiguity exposure,the more ambiguity-averse farmers will have a lower perceived value of the new technology and,therefore,will be less willing to adopt it.

Accordingly,we propose the first hypothesis of this paper:

Hypothesis 1: Farmers with high ambiguity aversion are reluctant to adopt innovative wheat seeds.

Information access fosters the adoption of new agricultural technologies by farmers.That is,information accumulation will facilitate farmers’ proactive adoption of agricultural technologies (Jianget al.2021;Maet al.2022).For instance,Takahashi (2013) notes that when farmers become familiar with new technologies through learning,ambiguity aversion will no longer have an impact on their decision-making behavior;Crentsilet al.(2020) reveal that access to information about new agricultural technologies mitigates the impact of ambiguity aversion on farmers.As farmers obtain more and more information concerning certain technologies,their knowledge of the new technologies in terms of yield and profitability becomes adequate,which effectively enhances their confidence in employing new agricultural technologies.

Thus,we propose the second hypothesis of this paper:

Hypothesis 2: Information access can alleviate the inhibitory effect of farmers’ ambiguity aversion on innovation seed adoption.

2.2.The effect of risk aversion on agricultural technology adoption

Once ambiguity diminishes,which means the probability distribution of outcomes is known,uncertainty will merely be the remaining risk (Takahashi 2013),and the probability distribution ofvis known.

As studies have shown,under the combined effects of imperfect information,unpredictable climatic conditions,and potentially dramatic fluctuations in market prices,farmers’ incomes are in a precarious state,and farmers will,therefore,generally behave as risk-averse (Hazell 1982;Guiso and Paiella 2008;Haushofer and Fehr 2014).

It is important to note that farmers’ risk aversion demonstrates different effects on new technology adoption (Barhamet al.2014;Bryan 2019).These effects vary depending on the impact of the new technologies on the variance of returns.For instance,lodging-resistant seeds are conducive to reducing the variance of yield variation,which are preferred by risk-averse farmers.Whereas innovative seeds increase agricultural risk,and farmers’ risk-averse attitude will inhibit their adoption of innovative seeds.Notably,in addition to their concern for the expected cultivation income,farmers are also interested in the variability of cultivation income (Chavas and Nauges 2020),which is well captured by the variance of returns.

It is assumed that farmers are risk averse and thereforeU(·) is a concave function withU′′<0.Following the analysis of Lusk and Coble (2005),we assume that the expected benefit of adopting the innovative seeds isa,with a varianceσ2,namely,

Furthermore,assuming that a farmer with a certain incomewownshha of land,and its rental price ispoper ha of land.Meanwhile,the general input per ha of land ispc,and the cost of adopting the innovative seeds isc,satisfying 0

Referring to Qiuet al.(2020),by taking the Taylor series expansion and taking the expectation of eq.(5),we can obtain:

Hence,we propose the third hypothesis of this paper.

Hypothesis 3: Risk aversion negatively impacts farmers’ adoption of innovative wheat seeds.

Agricultural insurance is an essential and effective response to agricultural risks,and it can alleviate risks,reduce farmers’ economic losses,and assist them in overcoming low-risk,low-return production conditions(Aliet al.2020).It can guarantee a smooth ex-post consumption of farmers’ transition and encourage farmers to adopt ex-ante production strategies with higher variance and expected payoffs (Freudenreich and Musshoff 2018).

Interlinked insurance and credit contracts are more effective than single agricultural insurance contracts at transferring risk and expanding credit to subsistence farmers while reducing self-selected risk rationalized by farmers (Tadesse 2014;Abateet al.2016).The variance of new technology payoffs will diminish after farmers purchase interlinked insurance and credit contracts.In other words,σ2decreases,which can lead to an increase in the negativeSo interlinked insurance and credit contracts can moderate farmers’ risk aversion to technology adoption by stabilizing return fluctuations.

Then,we propose the fourth hypothesis of this paper.

Hypothesis 4: Interlinked insurance and credit contracts can mitigate the effect of risk aversion on innovative seed adoption.

2.3.Dominant effect of risk aversion on agricultural technology adoption

Risks persist whether the probability distribution of cultivation payoffs is known or not.For uncertain events,whether an ambiguity exists or not,the risk always exists.Hence,risk receives more widespread and substantial attention than ambiguity.This may indicate that risk aversion plays a sufficiently vital role in agricultural technology adoption relative to ambiguity aversion.For instance,Olijslagers and van Wijnbergen (2019) reveal that ambiguity aversion takes on less importance than risk aversion when studying the effect of risk aversion,ambiguity aversion,and intertemporal substitution elasticity on the willingness to pay for the avoidance of climate change risk.This finding is in line with Aliet al.(2021),who note that the dominant effects of risk aversion in impacting insurance uptake behavior: the former focuses on how individual preferences influence the present-day valuation of future outcomes,while the latter concentrates on the effect of uncertainty on farmers’agricultural insurance participation decisions.They essentially examine the impact of ambiguity aversion and risk aversion on individual decision-making behavior.This paper explores the effects of both on farmers’ technology adoption behavior,which is another branch of individual decision-making behavior.Thus,the fifth hypothesis of this paper is as follows.

Hypothesis 5: Risk aversion plays a dominant role in agricultural technology adoption than ambiguity aversion.

3.Data and methods

3.1.Sample selection

This study obtained data through household surveys in four locations,including Yongshou County in Xianyang City and Heyang County in Weinan City in Shaanxi Province and Pinglu County in Yuncheng City and Yaodu District in Linfen City in Shanxi Province during July and August 2021.The sample areas were selected for the following reasons: First,Shaanxi and Shanxi provinces are major dryland wheat-producing areas,and wheat is the dominant crop grown by farmers in these areas.Therefore,it is expected that farmers in these regions will have a higher level of knowledge and understanding of innovative wheat seeds.Second,these areas are characterized by low precipitation and frequent disasters.Local farmers are expected to be highly ambiguityaverse,risk-averse,and with a greater comprehension of interlinked insurance and credit contracts.Third,the transmission of agricultural practice information in the selected regions is quite slow.

For the selection of sample farmers,we randomly selected five towns in each county based on the representativeness of the grain crop planting system and the advantages of grain crop production.Then five villages were randomly selected from each town.In each village,we first obtained the list of villagers from the village committee and sorted them alphabetically by the surname of the household head.For villages with 50 and less,51–100,and 101 and more households,the sample selection distances are 2,4,and 10 in that order,respectively,by which households were randomly selected.

First,we obtained farmers’ levels of ambiguity aversion and risk aversion and their willingness to adopt innovative seeds through the field experiment.Then,following field visits and interviews,questionnaires were used to collect basic personal and household information,wheat and corn input and output,agricultural insurance participation,and farmer social capital in 2020.A total of 745 questionnaires were collected.Excluding samples with outliers,705 valid questionnaires were obtained finally,with an efficiency rate of 94.63%.Among them,382 households are from Shanxi Province,and 323 are from Shaanxi Province.

3.2.Field experiment for measuring farmers’ uncertainty aversion

Some studies attempt to understand farmers’ risk behaviors through questionnaires and directly measure farmers’ uncertainty aversion (Blais and Weber 2006).However,some scholars have pointed out that the multifaceted heterogeneity of the subjects makes the survey questions too simplistic (Dinget al.2010).Meanwhile,since this measure lacks direct material incentives and is very subjective,it is difficult to ensure that it accurately reflects the underlying risk attitude(Charnesset al.2013).Besides this,the experimental method based on the expected utility model is also commonly used to measure individuals’ uncertainty attitudes.Heet al.(2016) note that when subjects with different risk aversion may have different risk willingness from the questionnaire and experimental method,and the experimental method can effectively distinguish between ambiguity aversion and risk aversion.Meanwhile,the subjects in this experiment are from rural areas where the education conditions are relatively low,so the experiment setting should be easy to understand.Based on the aforementioned studies,we conduct the following experimental design to measure farmers’ uncertainty aversion,referring to Holt and Laury (2002) and Qiuet al.(2020).The experimental procedure consists of the following three stages.

Preview experimentBefore beginning the experiment,the testers explained the experiment details to the farmers and informed them of the reward.The reward was set as points and converted into cash at a ratio of 10:1 based on the final results of the experiment.If farmers earn more points during the experiment,they will receive a higher amount of money at the end.So,in the experiment,they have to choose the best strategy to maximize their experimental gain,which avoids arbitrary choice behavior.Additionally,all farmers can receive a fixed appearance fee of 10 CNY,regardless of the experiment outcome.

Farmers were informed that there were three black cards and three red cards in the non-transparent bag.Rewards for drawing the black and red cards are shown in Table 1.We tested their understanding of this experiment based on their choices.Regardless of whether a red card or a black card is drawn,the payoff is greater when choosing reward plan B.Therefore,if reward plan A is chosen,it means that the farmer does not understand the rules yet and needs further explanation;if reward plan B is chosen,it means that the farmer understands the rule well and can proceed to the next formal experiment.

Table 1 Pre-experimental reward plan and amount

Ambiguity aversion measurement experimentFarmers were only aware that there were a total of six cards in the non-transparent bag,but the exact number of black cards or red cards was unknown.The rewards for drawing black cards and red cards are shown in Table 2 for reward plans C and D.We conducted 10 rounds of experiments with them,and in each round,they can choose between reward plan C and reward plan D.Farmers choosingreward plan C will receive 20 CNY for drawing either a red or a black card.The variance of the reward amount is 0,so it is a risk-free reward plan;for those choosing reward plan D,the reward for drawing a red card is higher than that for drawing a black card.As the experiment proceeds,the reward amount obtained by drawing a red card increases gradually from 22 to 60 CNY,while the reward obtained by drawing a black card decreases gradually from 18 to 0 CNY.The variance of the reward amount is larger,so it is a high-risk reward plan.Ambiguity aversion was measured based on the choices made by farmers,and the higher frequency they chose reward plan C,the more ambiguity averse they were.

Table 2 Ambiguity aversion measurement experimental reward plan and amount

Risk aversion measurement experimentIn this experiment,farmers were explicitly informed that the nontransparent bag contained three black cards and three red cards.The risk aversion measurement experiment was repeated 10 rounds with the same rewards shown in Table 2,and farmers still chose either reward plan C or reward plan D.Depending on the number of times farmers choose reward plan C to measure their risk aversion level,the higher frequency they choose reward plan C,the higher their level of risk aversion.

3.3.Field experiment for technology adoption

Technology settingTo address the lack of external validity,we have designed a field experiment similar to the one conducted by Tanget al.(2019) to measure farmers’ willingness to adopt innovative seeds.Subjects were randomly selected from the entire farmer population and were divided into control and treatment groups to ensure that farmers’ decisions are not influenced by any conditions outside the experiment.Farmers were faced with two types of production decisions: traditional seeds with low costs,low planting returns,and low exposure to weather,or innovative seeds with high costs,high planting returns,and high exposure to weather.Since innovative seeds have higher costs,farmers adopting innovative seeds need to take a loan from a bank to purchase innovative seeds,whereas farmers adopting traditional seeds are not limited by these financial constraints.Farmers in both the control and treatment groups were required to choose between the two seeds,and the key difference between the two groups was that farmers in the treatment group were offered interlinked insurance and credit contracts.

Production conditions settingThe experiment was performed using several red and black cards and one non-transparent bag.Weather conditions in the experiment were determined by the results of the farmers’card drawings,which ensured that farmers were not aware of the weather conditions of the year when making decisions.In a non-transparent bag containing four red cards and two black cards,each farmer drew a card at random.Drawing a red card indicates good weather conditions,while a black card indicates poor weather conditions,resulting in no income.The probability of disaster weather occurring in the experiment was determined to be 1/3 based on local meteorological data.The probability of the weather event is shown in Table 3 below.

Table 3 Weather occurrence probability

Experiment scenario settingTo better simulate real agricultural production scenarios,this field experiment involves a combination of weather conditions,loans,and agricultural insurance.At the beginning of the year,farmers have 4 200 CNY as the start-up capital and 0.667 ha of cultivated land,and they have to choose between the two different types of seeds before plowing.In the treatment group,we offer loans to farmers who choose to plant innovative seeds and provide them with an insurance contract.The input and output of these two seeds,depending on the weather conditions,are shown in Table 4 below.

Table 4 Input and output of seeds under different weather conditions (CNY)

In the control group,if a farmer chooses to use traditional seeds,the start-up capital of the farmer is 4 200 CNY,which is 200 CNY more than the 4 000 CNY needed for production inputs,so the farmer does not need to take a loan from the bank.In the case of good weather for the year,the farmer can earn 7 000 CNY fromplanting,so the year-end balance is 7 200 CNY.In the case of bad weather,the farmer has no harvest and will not earn income from planting,at which point the yearend balance is 200 CNY.If the farmer chooses innovative seeds,the start-up capital of 4 200 CNY is not enough to cover the total input cost of 6 000 CNY.To produce,the farmer needs to borrow 1 800 CNY from the bank for normal inputs.In addition,to obtain this loan,the farmer is required to provide the bank with collateral items worth 1 800 CNY.At the end of the year,if the weather is suitable for wheat growth,the farmer can obtain 12 000 CNY and get back the collateral items after paying the 1 800 CNY loan,leaving a balance of 10 200 CNY for the year.If the weather is bad,the planted wheat will have no harvest,and the cultivation income will be 0 CNY,resulting in the farmer being unable to pay the loan and having the collateral items confiscated,losing 1 800 CNY and finishing the experiment.

In the treatment group,farmers choosing to adopt traditional seeds will face the same input–output situation under different weather conditions as those in the control group.The difference between the control group and the treatment group is that farmers who choose to plant innovative seeds have purchased interlinked insurance and credit contracts,and if the weather is good that year,they will get 12 000 CNY in cultivation income at the end of the year and can get back the collateral after liquidating the loan of 2 000 CNY.If bad weather occurs,they will receive no cultivation income but can still get back the collateral items after the 2 000 CNY payout,and the yearend balance will be 0 CNY at this time.Farmers with a good year-end using this technology adoption experiment will be paid,while those with a bad year-end will receive nothing.

3.4.Variables description

Agricultural technology adoptionWe measured farmers’ choice of agricultural technology adoption based on the technology adoption experiment.The distribution of farmers’ choices of innovative and traditional seeds in the field experiment is shown in Table 5.The mean difference for innovation seed adoption in the control group and treatment group is given in Table 6.In the treatment group with interlinked insurance and creditcontracts,a total of 236 farmers choose to adopt innovative seeds,while 108 farmers choose traditional seeds,with the adoption rate of innovative seeds being 68.6%.In the control group without interlinked insurance and credit contracts,innovative seeds are adopted at a rate of 57.9%,10.7% lower than the treatment group.

Table 5 Distribution of farmers’ choice between innovative seeds and traditional seeds

Table 6 Absolute mean difference for innovation seeds adoption in the control group and treatment group

Uncertainty aversion(1) Ambiguity aversion.Based on the experiment results of the ambiguity aversion measurement,we measured farmers’ ambiguity aversion by the ambiguity index calculated in eq.(8).

The ambiguity aversion index calculated from eq.(8)has a range of [0,1].If the ambiguity aversion index is 0,it means that the farmer’s ambiguity aversion type is extreme ambiguity-loving;if the ambiguity aversion index is 1,it means that their ambiguity aversion type is extreme ambiguity aversion.

(2) Risk aversion.Like the calculation of the ambiguity aversion index,based on the experiment results of the risk aversion measurement,the farmers’ risk aversion index can be calculated by using the following eq.(9):

The risk aversion index derived from eq.(9) also has a range of [0,1].When the risk aversion index is 0,the farmer is extremely risk-loving,and when the risk aversion index is 1,the farmer is extremely risk-averse.

The frequency distribution of farmers’ ambiguity aversion index and risk aversion index is shown in Fig.1.Both extreme ambiguity-averse and risk-averse farmers have the highest proportions.This indicates that during the experiment,the majority of farmers consistently prefer the risk-free option that provides them with stable rewards.In both plots,the percentage of farmers with scores larger than 0.5 is greater than 50%,showing that most farmers are ambiguity-and risk-averse,which is similar to the findings of Barhamet al.(2014) and Qiuet al.(2020).

Fig.1 Frequency distribution of ambiguity aversion index and risk aversion index.

lnformation accessThis paper characterizes the farmers’ information access based on their agreement with “Can you get useful information (e.g.,marriage and schooling) from people around you?” with values ranging from [1,5].Farmers’ information access is divided into high and low groups based on the mean value of information access,where samples with the range of [1,4]are classified into the low information-access group,and those with the range of [4,5] are classified into the high information-access group.

lnterlinked insurance and credit contractAccording to the technology adoption experiment,the treatment group samples are considered to be the samples with interlinked insurance and credit contracts,which have the value of 1.The control group sample is considered the sample without interlinked insurance and credit contracts,which has a value of 0 since interlinked insurance and credit contracts are provided to the farmers planting innovative seeds in the treatment group.

Control variablesThe control variables selected in this paper are gender (Gender),age (Age),education degree (Education),whether the household is a leader in the village (Leader),total household income (Income),household size (Familysize),number of wheat plots(Numland),the proportion of wheat area (Wheatland),and distance from the county (Discity).Given that the wheat cultivation plot number and wheat cultivation area proportion exhibit quite distinct geographical characteristics,we introduce a province dummy variable(Province),which takes the value of 1 if the farmer is from Shanxi and 0 if the farmer is from Shaanxi.

3.5.Statistical description

From the variables descriptive statistics in Table 7,it is evident that among 705 sample farmers,63% adopt innovative seeds,indicating that the adoption rate is not high enough.The average ambiguity aversion index of farmers is 0.73,and the average risk aversion index is 0.65,illustrating that most farmers are ambiguity-averse and risk-averse.Furthermore,the ambiguity aversion level is higher than the risk aversion level,and they are generally reluctant to participate in activities with high ambiguity and high risk.

Table 7 Variables meaning and descriptive statistics

At the individual level,the majority of the farmers are male (68%),with an average age of approximately 58 years old and an elementary school diploma (7 years of education on average).Additionally,13% of them are rural cadres.At the household level,farm households have a family size of about five persons and a total household income of 46 700 CNY.The average number of wheat cultivation plots is 2,accounting for an average of about 72% of the total cultivated land area.

3.6.Econometric models construction

As the explanatory variable is a binary choice variable,to explore how ambiguity aversion and risk aversion affect agricultural technology adoption,the following probit regression models are constructed for econometric analysis as shown in eqs.(10)–(12):

where the subscriptiin eqs.(10)–(12) represents theith farmer,yiis technology adoption which is a binary explanatory variable,whereAmbiguityiin eq.(10) is the ambiguity aversion index to characterize ambiguity aversion level,andControlsirepresents a series of control variables like gender and age.In eq.(11),Riskiis the risk aversion index,which is used to characterize thelevel of risk aversion.In eq.(12),Ambiguityi×Riskiis the interaction term between the ambiguity aversion index and the risk aversion index.αis the constant term,βandγdenote regression coefficients,andεiis assumed to be a random error term obeying normal distribution.

4.Results and discussion

4.1.Baseline regression

Baseline regression estimation resultsProbit models are applied to estimate the effects of ambiguity aversion and risk aversion on innovative seed adoption,and the estimation results are presented in Table 8,respectively.Model (1) reports the effect of ambiguity aversion on innovative seed adoption.Farmers’ ambiguity aversion affects innovative seed adoption in the same direction as expected: the higher the level of ambiguity aversion,the lower the probability of adopting innovative seeds.This also substantiates Hypothesis 1: farmers with high ambiguity aversion levels are reluctant to adopt innovative seeds.This finding is consistent with the results reported by Rosset al.(2012),who find ambiguity aversion negatively affects farmers’ adoption of innovative seeds in Laos.

Table 8 Baseline regression estimation results

Model (2) presents the results of the effect of risk aversion on innovative seed adoption.The estimation results show that risk aversion is highly significant at the 5% significance level with a marginal effect coefficient of–0.132,which indicates that,for every 0.1 increase in the level of risk aversion,the probability of adoption of innovative seeds by farmers decreases by 1.32%,ceteris paribus.This finding is in line with Hypothesis 3.Risk aversion hinders the adoption of new technologies that increase the variance of payoffs,so more risk-averse farmers will lack the willingness to adopt innovative seeds,which coincides with the findings of Tan and Lu (2021),Wang and Zhang (2022),and Xu and He (2022).

Econometric estimations shown in Table 8 report that in the regression results of ambiguity aversion and risk aversion,the coefficient of information access is statistically significant at the 1% level,which implies that farmers are more likely to adopt innovative seeds if they are well-informed and knowledgeable about the new technology.Providing interlinked insurance and credit contracts has a significant positive effect on innovative seed adoption at the 1% level.From the results of marginal effects,the marginal effect coefficients are 0.125 and 0.123 in the regression results of ambiguity aversion and risk aversion,respectively.Hence,providing interlinked insurance and credit contracts to farmers is the key to motivating them to adopt new technologies.

As for factors of farm household characteristics,total household income passed the significance test at the 5%level with a positive coefficient,suggesting that farmers with higher total household income are more inclined to adopt innovative seeds.This is probably due to the fact that farmers with higher total household incomes have more leisure time and will spend more time focusing on innovative seeds,and with high disposable incomes,they have sufficient funds to try innovative seeds to maximize their returns.Regarding factors related to farmlandcharacteristics,the number of planted plots negatively affects agricultural technology adoption,with marginal utility coefficients of–0.013 in the regression results of ambiguity aversion,indicating that the probability of adopting innovative seeds will decrease by 1.3% for each increase in the number of planted plots,ceteris paribus.It is not surprising,given that the number of plots characterizes land fragmentation,and it increases the difficulty of agricultural mechanization operations and leads to an increase in the cost of machinery use and labor costs (Lvet al.2014;Jianget al.2018).Besides,the proportion of the wheat planted area is highly significant at the 1% significance level,suggesting that farmers with a higher proportion of wheat are more likely to accept innovative wheat seeds.This is probably because farmers with a low proportion of wheat planted tend to care more about other crops’ yield,while the importance of wheat and the utility it generates for farmers are quite limited,so they tend to be indifferent to related technologies.

Robust checkThree methods are applied to test the robustness of the baseline regression results: (1) Altering regression model.Replace the probit model that replaces the probit model with the logit model to test the robustness of the baseline regression results previously obtained using the probit model.(2) Replacing core independent variables.Assign the level of ambiguity aversion index in the interval [0,0.5] to 0 and the level of the ambiguity aversion index in the interval [0.5,1] to 1 to generate a binary variable characterizing ambiguity aversion and the binary variable of risk aversion obtained by treating the risk aversion index in the same way.(3) Winsorizing.To reduce the impact of extreme data values on the estimation results,we treated the core variables ambiguity aversion and risk aversion with a 15% tail reduction for further robustness testing.

The results of the three robustness tests suggest that the risk aversion is significantly negative,while ambiguity aversion negatively affects innovative seed adoption(Table 9).The coefficients of information access and interlinked insurance and credit contracts on innovative seed adoption are significantly positive,further confirming the robustness of the above baseline regression results and the conclusions of this paper.

Table 9 Robustness check of baseline regression estimation results

4.2.Dominance analysis

Ambiguity aversion and risk aversion are introduced simultaneously in the model and decentered,and then their interaction terms are added.The estimation results of the strength of ambiguity aversion and risk aversion affecting the adoption behavior of farmers’ innovative seeds are shown in Table 10.Model (1) estimates ambiguity aversion,risk aversion,their interaction terms,information access,and interlinked insurance and credit contracts.The estimation results show that after introducing their interaction terms,ambiguity aversion is no longer significant,while risk aversion is significant at the 10% level of significance.In model (2),with the addition of control variables,the same result is obtained that ambiguity aversion is no longer significant while risk aversion remains significant at the 10% level of significance.This confirms the dominant role of risk aversion in the influence of uncertainty on innovative seed adoption.

Table 10 Dominance test of ambiguity aversion and risk aversion

Ambiguity aversion and risk aversion are vital factors influencing farmers’ adoption of innovative seeds.Moreover,information access and interlinked insurance and credit contract supply also affect farmers’willingness to adopt innovative seeds.As suggested by the theoretical analysis in Section 2,the influence of information access and interlinked insurance and credit contract supply on innovative seed adoption can also be achieved by moderating ambiguity aversion and risk aversion.

4.3.Moderating effect analysis

Moderating effect of information access on ambiguity aversionFarmers are generally positive about the dissemination of agricultural technologies and methods,and most respondents agree that “they can get useful information (e.g.,marriage and schooling) from people around them” indicating that most farmers have high access to information.The results of the moderating effect of information access and interlinked insurance and credit contract are shown in Table 11.Models (1) and (2)report the results of the effect of ambiguity aversion on innovation seed adoption in the high and low information access groups.It can be seen that the coefficients of ambiguity aversion are–0.401 and–0.160,respectively,and only the coefficient of ambiguity aversion in the high information access group is significant.Farmers with higher information access have lower levels of ambiguity aversion,and in Model (1),the ambiguityaversion coefficient is–0.401,which is smaller than the baseline regression group,implying that this group of farmers is more likely to adopt innovative seeds.This can be attributed to information largely eliminating farmers’ apprehension about innovative seeds,and high information access can mitigate the negative effect of farmers’ ambiguity aversion to innovative seed adoption.

Table 11 Regression results of moderating effects

Moderating effect of interlinked insurance and credit contracts on risk aversionModel (3) shows the results of the effect of risk aversion on innovative seed adoption in the case of providing interlinked insurance and credit contracts,and it is quite intuitive that risk aversion severely inhibits farmers’ willingness to adopt innovative seeds with a coefficient of–0.555.Model (4) reports the effect of risk aversion on innovative seed adoption in the absence of interlinked insurance and credit contracts,and the results indicate that risk aversion has no significant effect on the adoption of innovative seeds in this case.This indicates that farmers who enjoy interlinked insurance and credit contracts have lower risk aversion and are more likely to adopt technology,implying that the inhibitory effect of risk aversion on innovation seed adoption is diminished.

From the above analysis,we can see that interlinked insurance and credit contracts can moderate the effect of risk aversion on innovative seed adoption and that it is negatively moderated by interlinked insurance and credit contract,confirming that interlinked insurance and credit contract negatively moderates the effect of risk aversion on innovative seed adoption and that farmers who purchase interlinked insurance and credit contracts have a greater willingness to adopt innovative seeds.

4.4.Heterogeneity analysis

Since education degree reflects farmers’ knowledge of technology and learning ability to some extent (Li and Ma 2021),farmers with different levels of education are likely to differ significantly in understanding and mastering the seed technology,while households with different income levels behave differently in their ability to coordinate and utilize productive resources.Hence,we study how education degree and income level influence farmers’ambiguity aversion and risk aversion when adopting new seeds.

The farmers’ education degrees are divided into lowand high-education groups.Farmers with an education level below the sample median are classified in the loweducation group.The results of sub-sample regressions are shown in Table 12.Model (1) shows the regression results for the low-education group and Model (2) reports the regression results for the high-education group.The results show that ambiguity aversion and risk aversion are significant in the low-education group,while they are not significant in the high-education group,which perhaps because of the higher learning ability as well as the ability to acquire and use information in the high-education group (Sang and Luo 2021),where the role of ambiguity reduced dramatically,and thus ambiguity aversion will appear to be less sensitive in the high-education group.As for risk,people with higher levels of education tend tobe more risk-taking,which means that their risk aversion is also less sensitive than that of the low-education group.

Table 12 Regression results for education and income level subgroups

Using 0.5 quantiles as the boundary,we divide samples into the low-income group and the high-income group according to total household income.Model (3) is the regression result for the low-income group,and Model(4) is the regression result for the high-income group.Low-income farmers are more averse to ambiguity and risk when compared to high-income farmers.This is likely because low-income farmers are limited by their means of production and will face more ambiguity and risk in agricultural production activities,so ambiguity aversion and risk aversion will show stronger sensitivity.

Table 12 indicates that ambiguity aversion and risk aversion among less educated and lower-income farmers greatly hinder the adoption of innovative seeds.Providing less educated farmers with interlinked insurance and credit contracts,as well as enhancing their access to information,would aid in the adoption of innovative seeds.However,this supporting effect is not significant among the low-income farmers in the sample.

5.Conclusion and policy recommendation

5.1.Conclusion

Ambiguity aversion and risk aversion are crucial factors influencing farmers’ adoption of innovative seeds.Using survey data from 705 wheat farmers in Shaanxi and Shanxi provinces and employing the experimental economics method to measure innovative seed adoption,ambiguity aversion,and risk aversion,we examined the relationship between farmers’ ambiguity aversion and innovative seed adoption under the moderating effect of information access as well as the relationship between risk aversion and innovative seed adoption under the moderating effect of interlinked insurance and credit contract.Our findings indicate that:

(1) Both ambiguity aversion and risk aversion negatively influence innovative seed adoption.A higher level of ambiguity aversion makes farmers less likely to adopt innovative seeds,while risk aversion also significantly lowers the likelihood of adopting innovative seeds.

(2) Information access contributes to alleviating inhibition caused by ambiguity aversion on adopting innovative seeds.Specifically,farmers with higher information access tend to have lower ambiguity aversion levels and a higher willingness to adopt innovative seeds.

(3) Interlinked insurance and credit contracts can mitigate the negative impact of risk aversion on innovative seed adoption.Farmers purchasing interlinked insurance and credit contracts have lower risk aversion levels,which further encourages them to adopt innovative seeds.

5.2.Policy recommendation

Ambiguity and risk influence innovation seeds adoption differently.As a result,policymakers can design appropriate safeguard mechanisms from both ex-ante and ex-post perspectives to reduce various uncertainties in the agricultural production process.They should focus on addressing the current challenges of farmers’ ambiguity aversion and risk aversion to facilitate the adoption of new technologies by farmers.

First,the government can manage ambiguity through ex-ante mechanisms.For ambiguity-averse farmers,providing technical training and other relevant information can help alleviate their ambiguity aversion.Relevant government departments need to take full advantage of grassroots cadres and improve farmers’ understanding and awareness of new agricultural technologies so that they can adopt new technologies earlier and benefit from them.

Second,policymakers can provide effective support to farmers through ex-post mechanisms.By providing them with interlinked insurance and credit contracts,the variations in planted yields will be greatly reduced,various risks in the process of agricultural production will be effectively mitigated,and farmers’ risk tolerance to natural disasters will be improved.

Last but not the least,according to the research findings,different levels of ambiguity aversion and risk aversion lead to various levels of willingness to adopt innovative seeds.Hence,the government can tailor personalized recommendations for farmers with different levels of ambiguity aversion and risk aversion and accordingly provide more targeted and effective technologies and corresponding point-to-point training to different farmers to realize the matching of supply and demand of agricultural technologies.

There are some limitations to this paper.First,farmers’ adoption of seed technology can also be influenced by social service organizations,while it is not adequately considered in this paper.Second,ambiguity and risk exist throughout every stage of agricultural production,while this paper only studies technology adoption.Hence,future research on uncertainty in agricultural practices can be extended to cover the entire process of production.Third,even though we strictly adhere to the research content and research needs by conducting household research and field experiments,data from only four counties in Shaanxi and Shanxi cannot provide a full examination and analysis of agricultural technology adoption in China.Future studies can further expand the sample to obtain more general and generalizable findings.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (71973087 and 72003215),the 72nd General Program of China Postdoctoral Science Foundation (2022M720170),the Soft Science Project of the Department of Science and Technology of Shaanxi Province,China (2022KRM131),and the Special Fund Project of Basic Scientific Research Operation Funds of Central Universities,China (20SZYB21).

Declaration of competing interest

The authors declare that they have no conflict of interest.