Risk, Ambiguity and the Adoption of New Technologies: Experimental ...

Risk, Ambiguity and the Adoption of New Technologies: Experimental Evidence from a developing economy* Nicholas Ross, Paulo Santos and Tim Capon

October 2010

Despite the expected benefits, the slow adoption of innovations in less developed countries has long been a puzzle. Aside from market constraints, risk-aversion dominates the discussion on the behavioural determinants of technology adoption. This paper investigates the role of ambiguity-aversion as another possible explanation, given farmers may have less information about the outcomes of new technologies compared with traditional technologies. Using primary data from behavioural experiments used to measure risk and ambiguity preferences in the field we find that farmers’ aversion to ambiguity and not risk constrains the adoption of new technologies.

Keywords: risk preferences, ambiguity preferences, technology adoption.

*Acknowledgements: I thank the assistance of Vilas Gobin, Kenekeo Sayarath at SNV Khammouane and my field surveyors Somsy Xayalath and Bouathong Khounxieng. I gratefully acknowledge the University of Sydney and my father, David Ross, for financial support.

1

1. Introduction Despite the potential rewards, the adoption of new technology is often slow and incomplete. Such delays help explain differences in productivity across countries and why poverty can persist. Given the prevalence of poverty in rural areas, agricultural transformation is widely seen as a precondition for overall growth. This explains the great focus of attention on farmers’ decisions to adopt new technologies in developing countries by agricultural and development economists. Feder, Just and Zilberman (1985) is the classical review of much of the earlier research that followed the Green Revolution. More recent developments are reviewed in Zilberman and Sunding (2004). Amongst the explanations, these studies have identified the role played by uncertainty, in particular, the importance of aversion to risk (Feder 1980; Feder et al. 1985; Ghadim et al. 2005, Binswanger, 1980; Knight et al. 2003; Liu 2007; Gong et al. 2010), implicitly assuming that the Subjective Expected Utility (SEU) approach of Savage (1954) is valid and that decision makers do not distinguish between known and unknown probabilities. This last assumption is the focus of Ellsberg (1961) criticism of SEU, where it is suggested that individuals prefer to bet on risky prospects for which they can clearly assess the probability of outcomes over ambiguous prospects for which the probabilities of its outcomes are unknown and the individuals feel less competent. The importance of ambiguityaversion, has been the focus of work by Heath and Tversky (1991), Fox and Tversky (1995) and Chow and Sarin (2002) and has been empirically found to exist in the market (e.g. Sarin and Weber 1993; Halevy 2007; Capon 2009) and among subsistence farmers in developing countries (Akay et al. 2009). The adoption of new technology provides a natural setting to test the importance of ambiguity-aversion for decision-making, given that the probability distribution of outcomes 2

associated with a new technology are rarely known (Liu 2007). Yet the literature on adoption of innovations has systematically failed to differentiate between risk-aversion and ambiguityaversion in explaining these decisions, with few exceptions (e.g. Engel-Warnick et al., 2008). In this paper, we address the importance of ambiguity in explaining the adoption intensity of a new crop, non-glutinous rice, in the context of a developing country, Lao PDR, which is simultaneously very poor and experiencing fast growth. After a brief review of the literature on the role of risk and ambiguity in the adoption process in section 2, we introduce this new technology in section 3. To understand the importance of ambiguity, we collected a unique data set from rice farmers in one village in the province of Khammouane, in the central part of Lao PDR. We use behavioural field experiments to elicit farmers’ risk and ambiguity preferences in conjunction with a household survey. We present our data in section 4 after which we ask how these preferences affect the intensity of adoption of this new technology. We show that ambiguity, but not risk, matter for the adoption decision. We conclude in Section 5 with some policy implications and suggestions for future research. 2. Risk and Ambiguity A farmer’s choice to implement a new technology over an existing one depends on numerous complex factors, some of which are not directly observable in a standard household survey. Examination of the adoption decision began with Grilliches’ (1957) pioneering conclusion that economic variables were significant determinants of the adoption and diffusion of hybrid corn in the US Midwest. The introduction of High Yielding Varieties (HYV) of rice in developing countries during the Green Revolution of the 1960s spurred increased efforts to understand what determines the adoption of new technologies, both within economics and other social sciences (e.g. Rogers 1995). Over time, the focus of

3

research effort has shifted away from market imperfections and market constraints towards the roles played by information, education and social learning. Much of the earlier work focused on the functioning of input markets and their differentiated impact on smallholders (Weil 1970; Binswanger 1978; Gafsi and Roe 1979; Feder et al. 1985, Wills 1972; Khan 1975; Bhalla 1979).The focus soon shifted to the importance of information (Feder 1979) and decision makers capacity to process information, with the analysis of the role of education (Lin 1991; Duraisamy 2002), access to extension (Nkonya et al. 1997) and the role of social learning (Besley and Case 1993; Foster and Rosenzweig 1995; Munshi 2004, Udry and Conley 2010). This change in emphasis has highlighted the importance of the behavioural and social factors that affect farmers’ technology adoption choices under conditions of uncertainty. Theories of decision making under risk in economics are primarily built upon the foundations of Expected Utility Theory (EUT) (von Neumann and Morgenstern1947): A decision maker who is faced with prospects with a probability distribution known a priori (𝑝1, … , 𝑝𝑛 ) over a number of outcomes (𝑥1 , … , 𝑥𝑛 ), is assumed to maximise the probability

weighted sum of the utility of the outcomes, given by ∑𝑖 𝑝𝑖 . 𝑢(𝑥𝑖 ) . A decision maker’s

attitude to risk determines the shape of the utility function, with concavity representing riskaversion. The Subjective Expected Utility (SEU) Hypothesis (Savage 1954) relaxed the assumption that 𝑝𝑖 is objectively known a priori, instead assuming that decisions will be governed by a set of subjectively formed probabilities.

The SEU approach forms the basis of much of the work on the effects of risk on technology adoption. For example, Feder (1980) uses differences in risk-aversion to explain differences in the allocation of land to new crops, under the assumption that the new crop exhibits a higher variability in yield or returns than the existing crop. This result was later 4

expanded by Just and Zilberman (1988) who show that the intensity of adoption depends on whether the new technology is risk increasing or decreasing and whether risk-aversion is increasing or decreasing with wealth. The development of the conceptual frameworks linking risk with the adoption of new technologies was not accompanied by similar advances in empirical work (Feder et.al. 1985). Empirical research was limited by the difficulties encountered in measuring risk preferences, and the corresponding lack of consensus on the best way to recover parameters such as the coefficient of risk-aversion. In practice, two approaches have dominated: the elicitation of subjective probabilities of outcomes and the direct elicitation of risk preferences. 1 The more common method of directly measuring risk in adoption literature is through the elicitation of subjective probabilities. For example, subjective estimates of crop riskiness influence the adoption of grain among Mexican farmers (O’Mara 1980), whilst the risk perceptions of new seed varieties among maize growers in Malawi influenced both their probability and intensity of adoption (Smale et al. 1994). Ghadim (2000) elicits subjective yield and price probabilities from traditional crops and chickpeas from farmers in Western Australia and finds that relative riskiness has a negative effect on adoption. Studies of technology adoption that directly elicit risk preferences are scarce. Binswanger (1980), who conducted experiments characterising risk attitudes in rural India, was the first to directly measure risk-aversion in a low-income environment. He provided respondents with choices between lotteries involving 50:50 chances of higher and lower payoffs (and, intermittently, certainties). Knight, Weir and Woldehanner (2003) use hypothetical questions to divide farmers into risk and non risk-averse groups in Ethiopia, 1

More abundant throughout earlier adoption literature, econometric methods have frequently estimated the probability distribution of output given input, measuring farmers risk attitudes’ through deviations in input choice from profit-maximising input choices (Moscardi and de Janvry 1977; Antle 1989; Pope and Just 1991; Chavas and Holt 1996). As misallocation of inputs can be related to credit and information factors, using this method to measure risk preferences can be highly dubious without a structural model with strong assumptions (Liu 2007; Alpizar et al. 2009).

5

finding risk-aversion is related to lower levels of technology adoption. Liu (2007) conducts lotteries with pair-wise choices to elicit preferences, finding Chinese farmers with higher levels of risk-aversion are slower to adopt new cotton varieties. Such studies commonly ignore the role of ambiguity in decision-making. In the context of the adoption of a new technology, where decisions are made with limited prior knowledge, this is an assumption that can be questioned. We do not make this assumption in our own study of decisions regarding the adoption of non-glutinous rice. Instead, we examine the potential for ambiguity-aversion, in addition to risk-aversion, as a possible explanation for some farmers’ reluctance to adopt new agricultural technologies in developing countries. To understand the potential effects of ambiguity on decision-making, we first review some of the earlier work on ambiguity. Knight (1921) was one of the first to address the role of ambiguity in decision-making, distinguishing between measurable known probabilities and unmeasurable unknown probabilities. However, the concept of ambiguity was scarcely explored until Ellsberg (1961) demonstrated the effect of ambiguity on decision-making with a thought experiment. Essentially, Ellsberg (1961) demonstrated that many people would prefer to bet on a lottery with an objectively known probability rather than a lottery with an unknown yet subjectively equivalent probability and thereby exhibit ambiguity-aversion. Various descriptive theories have attempted to explain attitudes to ambiguity (for a review, see Camerer and Weber 1992). Einhorn and Hogarth (1985) proposed an anchoring and adjustment model, where individuals use an initial estimate of a probability as an anchor and then adjust it according to the ambiguity perceived in the situation and their ambiguity preferences. Kahn and Sarin (1988) generalised the axioms underlying SEU to construct a model accounting for decisions under both risk and ambiguity, with the anchor as the normative expected probability. The individual’s adjustment process is determined by the 6

anticipated amount of disappointment they feel when the ambiguous probability is below or above their normative estimate. Confirming Ellsberg’s conjecture, evidence of ambiguity-aversion has been found in empirical studies (e.g. Becker and Brownson 1964; Slovic and Tversky 1974; Sarin and Weber 1993; Capon 2009). The vast majority of experiments, however, are conducted in laboratory environments with university students, and evidence for ambiguity-aversion amongst populations in developing regions is more limited. For example, Henrich and McElreath (2002) found no evidence for ambiguity-aversion among Chilean farmers, arguing that ambiguity-aversion may be driven by cultural factors and that it does not generalize to a developing agriculture context. Conversely, Akay et al. (2009) found evidence of strong ambiguity-aversion among Ethiopian farmers. Two papers investigate the role of ambiguity in technology choice among farmers in developing regions and are similar, in spirit, to our study. Engle-Warnick et al. (2008) are the first to empirically distinguish the effects of risk and ambiguity-aversion in technology choice among subsistence farmers in rural Peru. Their findings suggest risk and ambiguity-aversion are highly correlated, but were not significantly related to the adoption of new technology. 2 Their approach differs from ours in that it doesn’t focus on a specific technology, instead it defines a farmer as an innovator if she plants any modern crop in a twelve month period, regardless of previous knowledge, either direct or indirect.. Alpizar et al. (2009) conducts experiments with Costa-Rican farmers in the aftermath of extreme climatic conditions, measuring the effects of risk and ambiguity in farmer adaptation 3 to climate change, through the adoption of soil conservation practices. They show

2

Although they are able to show that ambiguity-averse farmers exhibit a less diversified crop portfolio. 3

Adaptation can be seen as the adoption of new technology with the anticipation of new shocks.

7

that both risk and ambiguity preferences significantly influence adoption decisions, concluding that ambiguity-aversion is an important factor in technology choice. Similarly to Engle-Warnick et al. (2008), and contrary to our own work, they define the adoption variable as simply the presence of the new technology, ignoring the possibility that a farmer might only partially adopt a new innovation. 3. Non-Glutinous Rice in Central Laos Lao PDR is a very poor country in SE Asia, heavily dependent on the agricultural sector which in 200X accounts for nearly half of its GDP and employs 77% of its labour force (Bountavy 2006). Poverty remains ubiquitous among farming households, with 87% of the country’s poor living in farmer-headed households (NSC 1999). Rice production, mostly produced on small family farms, dominates the agricultural sector, accounting for 50% of agricultural output. Despite the introduction in 1986 of the New Economic Mechanism (NEC) with the intention of liberalising the economy and broadening the nation’s exposure to international markets, the largest proportion of the country’s agricultural output remains focused on glutinous-rice farming. It consisted of 85% of rice production in 2004 (Schiller 2006). Laos has the highest per capita consumption and production of the variety in the world and this crop represents roughly 20% of its GDP. For the purposes of the present study, glutinous rice is the “existing” technology, to be compared with the introduction of “nonglutinous rice” as a new technology. 4. Unlike glutinous-rice, which is typically grown for direct consumption, non-glutinous varieties are cultivated as a base ingredient in the production of noodles and beer. Widely produced in neighbouring Thailand and Vietnam, the varieties offer farmers greater yields, shorter growth maturity and higher, more stable prices (SNV 2009).

4

We focus on the two varieties most commonly adopted in the region, specifically the varieties VND 95-20 and CR203.

8

This new crop has been promoted in Central Laos by the Dutch Development Agency, in collaboration with local mills, since 200X . Typically, farmers adopting the crop enter a contract with a miller, who agrees to purchase a specified amount of paddy rice at an agreed upon price and time, whilst providing production inputs and credit, whilst the farmer agrees to supply the contracted quantities at specified quality standards (moisture level). The price paid is considerably higher than the price that farmers may receive for glutinous rice (in 2010). The additional revenue seems to translate into additional profits, as the two crops require similar amounts of inputs. Since the adoption of non-glutinous rice requires little initial capital and, although it creates linkages to domestic and foreign markets, it also seems to be less risky with regard to market conditions than glutinous rice (given the guaranteed prices), although there is certainly an element of trust needed, as millers may be personally unknown to farmers so the possibility that contracts may not be honoured exists. That said the adoption of the new crop variety has been slow. In order to explain the adoption of non-glutinous rice, we conducted a household survey and behavioural field experiments in Natai, a village in the central province of Khammouane, Lao PDR during July 2010. The village was selected with the assistance of the Netherlands Development Organisation (SNV), who has conducted research on the adoption for non-glutinous rice in the province (SNV 2009). Natai was chosen because whilst nonglutinous rice was introduced to this region during the 2009 dry season it has so far only been partially adopted. This provides an ideal case study for examining the factors that affect the adoption decision. The size of Natai also means that it was possible to interview 66 of the total number of 69 household heads for this study. Household heads were interviewed prior to the field experiments, leading to a rich data set that, in addition to the standard information collected in these type of surveys (household demographic and socio-economic characteristics, land use and crops produced in 9

the last two seasons) also collected detailed information on the cultivation of glutinous and non-glutinous rice varieties, including yield, price received, capital equipment, labour and exposure to shocks. Table 1 presents summary statistics for the key variables of interest from the household survey. The average interviewee in the sample was 48 years old and had completed 5.2 years of education. Nearly all farmers had received support from an extension service, with an average of around 2 visits from extension services. Glutinous-rice was the predominant crop cultivated in Natai village, with an average of 1.27 hectares per household. Non-glutinous varieties were grown on an average of 0.92 hectares per household, with 54 (82%) of the households growing some area of the crop, although nearly all of them have relatively limited experience with non-glutinous relative to glutinous rice since its introduction in 2009. To understand the degree of agreement between the farmers’ opinion of non-glutinous rice and the opinion of its advocates, we asked farmers for their responses to a set of statements which compared the characteristics of non-glutinous to glutinous varieties. Farmers were asked to rate their agreement on a Likert scale of 1 to 5, where 1 indicated strong disagreement, 3 indicated neutrality, and 5 indicated strong agreement. The responses displayed in Figure 1, illustrate farmers’ perceptions of the new variety’s characteristics as being largely positive. This implies that other behavioural factors lie behind their limited adoption of non-glutinous rice rather than a simple preference for glutinous rice. All survey respondents participated in decision-making experiments designedto measure risk and ambiguity preferences. Given relatively low levels of formal education in Natai, the experiments were implemented with the help of visual aids to assist the respondents in developing a clear understanding of the probabilities of the alternative payoffs. Participants in the experiment were paid for their participation and, depending on outcomes, 10

could receive up to 40000 Lao Kip (LAK) 5.This amount is higher than the average daily household income in the district and close to 2 days of rural wage. Although the number of studies that conducts such experiments has increased, this type of data is still infrequently used and for that reason we describe our procedure in detail regarding the elicitation of risk preferences and ambiguity in more detail. 4. Methods for eliciting risk and ambiguity preferences The vast majority of risk elicitation procedures are based on type of instruments developed by Binswanger (1980) or Holt and Laury (2002). The Holt and Laury (2002) procedures uses choices from a list of binary lotteries that differ in expected payoffs and variance to infer parameters for risk-aversion from the choices made. Our instruments, presented in Table 2, differs from the one employed by these authors, and instead is similar to the approach of Akay et al. (2009) and Capon (2009) in asking respondents to directly compare certain amounts and lotteries, in order to more directly elicit certainty equivalents (CE) for the lotteries.. Obtaining a decision maker’s CE allows a comparison of risk preferences across respondents 6. A CE represents a certain amount that is equally preferred to a risky alternative. Once the CE is elicited, the Risk Premium (RP), defined as the difference between the expected value and the certainty equivalent, allows for inference regarding a decision maker’s attitude towards risk. A positive risk premium implies risk-aversion, a negative risk premium implies risk seeking and no risk premium implies risk neutrality. For each participant, CEs were elicited using two different risky prospects, a coin toss and an urn. The coin toss offered the participant equal probabilities of winning 20,000 Kip and nothing. The urn, containing exactly 5 red and 5 yellow balls, offered the participant the 5 6

At the time of the experiment, the exchange rate was LAK 7,200=AUD $1.00 For prospect 𝑎𝑖 , 𝑈(𝐶𝐸𝑖 ) = 𝐸[𝑈(𝑎𝑖 )]

11

possibility of winning 20,000 Kip if a red ball was drawn and nothing for a yellow ball,. Although both have the same probability distributions and payoffs, the use of different mechanisms allows us to account for potential bias towards a particular way of eliciting preferences, leading to a more balanced assessment of risk preference Participants were offered 11 choices between a certain payoff (option one) and the risky prospect (option two), with the certain payoff increasing in 2000 Kip increments from 0 Kip to 20000 Kip. For small certain payments, most participants would prefer to play the risky prospect, while for very large certain payments, most participants would prefer a sure thing. Given this, at some point most participants will reveal their risk preferences by switching from option 2 to option 1. We calculate the CE in similar fashion to Eggert and Lokina (2007), as the midpoint between the lowest certain payment for which the participant chooses option 1 and the highest certain payment for which they choose option 2. 7 The participants were informed they would receive real payment from the experimenter, depending on the choices they made for each of the 11 options: participants would draw a ticket, numbered from 1 to 11, and would play the prospect corresponding to their selection for the respective choice. One was selected at random for payement, with the participant either receiving a certain payment or playing the prospect depending on the specific choice made when facing the selected option. The distribution of the outcomes for both procedures is presented in Table 3. The results suggest that participants were more risk-preferring for the urn prospect than coin toss. The urn prospect displayed risk-preferring average CE of 11 000 compared to a marginally risk-averse CE of 9939.4 for the coin toss, with 63.3% and 52.7% of participants riskpreferring for the urn and coin toss respectively. 7

All subjects made consistent choices when switching over from option two to option one; that is they switched from gamble 2 to gamble 1 only once. 12

In addition to these games, we followed Anderson et al.’s (1977) suggestion and also asked farmers how they feel about risks. Participants were visually presented with a numbered scale ranged from 1 to 10, where 1 represented the statement “I never like take risks” and 10 represented “I always like take risks”, and asked to rank themselves in that scale. 8 Figure 3 presents the results from participants’ self-assessed risk. The majority (66.6%) of the participants provided Likert scores between 6 and 10, with an average score of 6.48 (and standard deviation of 2.13). The results follow a distribution that is similar to the one already described, suggesting that most participants are willing to take risks and think of themselves as often willing to take risks. We delay the discussion of a formal test of the concurrence between the three procedures for after the discussion of the results regarding the elicitation of ambiguity preferences. Numerous empirical studies of ambiguity have tested Ellsberg’s (1961) thought experiment, mostly in laboratory settings (Becker and Brownson 1964; MacCrimmon and Larsson 1979; Kahn and Sarin 1988; Bowen et al. 1994). Previous field experiments measuring ambiguity preferences in developing countries are scarce, limited to those conducted by Engle-Warnick et al. (2008), Alpizar et al. (2009), and Akay et al. (2009). Our measure of ambiguity preference depicted in Table 4, is a variation of Ellsberg’s 2-colour urn experiment, and is similar to that used by Lauriola and Levin (2001) and Capon (2009) in a laboratory setting. It involves a choice set between an unambiguous urn (option one) and an ambiguous urn (option two), manipulating the objective probability of success in the unambiguous urn whilst leaving the ambiguous urn unchanged.

8

Eliciting individual attitudes to risk using a singular self-assessment question has been frequently used as a proxy for risk aversion by agricultural economists (Kastens and Featherstone 1996; Patrick and Ullerich 1996; Bard and Barry 2000). 13

Prior to the experiment, participants were informed that analogous to the risk procedures, yellow and red balls rewarded them with 20,000 Kip and nothing, respectively. Participants were presented with a set of 11 choices, where each choice asked them to select between playing the unambiguous urn or the ambiguous one.

In each choice, the

unambiguous urn held a known proportion of ten coloured balls, with the proportion of yellow balls (and hence the p of winning) decreasing in increments of 0.1 for each successive choice. This was reinforced visually, where for each choice the participant was shown one yellow ball removed from the urn and replaced by a red ball. Participants were advised the ambiguous urn contained 10 balls, although the number of each type of balls was not revealed. Payment was determined in the same fashion of the risk experiments. We build on Lauriola and Levin (2001) and Capon (2009), who value ambiguity preferences as the objective probability of winning the unambiguous urn prior to crossing over to the ambiguous urn. Our analysis follows similar lines to Lauriola and Levin (2001), and Kahn and Sarin (1985) and set a normative anchor at p=0.5 for ambiguity-neutrality, we correspondingly set our EMV equivalent at 10000 Kip. This implies that ambiguitypreferring participants display an EMV greater than10000 whilst ambiguity-averse participants will display an EMV smaller than 10000. To illustrate, and using the values presented in table 4, a participant who crosses from the unambiguous urn over to the ambiguous one on the third choice prefers the unknown ambiguous prospect rather than the objective p=0.8 of the unambiguous prospect. Their midpoint-calculated EMV of 17000 reveals they are ambiguity-preferring. Conversely, crossing over at the eight choice would they are averse to ambiguity, with an EMV of 7000. There were several factors which made this procedure appealing in the field. The unknown and known probability distributions were visually represented to the participants, permitting a clearer understanding of the question. This was reinforced by the fact that the 14

binary-choice list, payoffs and payment determination closely resembled the procedures used to elicit risk preferences, allowing the participants to understand the procedure through prior experience. Table 5 presents the results of the probability equivalence procedure measuring ambiguity preferences. The distribution of results is evenly spread and suggests the ambiguity attitudes for farmers in Natai were fairly heterogeneous, with slight skewness towards ambiguity-preference (53% of participants elicited EMVs of more than 10000 Kip). Comparing these results to earlier papers using similar samples (Akay et al. 2009) and procedures (Lauriola and Levin 2001), Lao farmers appear to be significantly more ambiguity-preferring. In addition to explore the role of preferences in the process of adoption of new technologies, our data allows us to address two questions: the first, of methodological nature, is whether the way risk preferences are measured matters for our conclusion regarding our classification of respondents’ behaviour. The second, of more substantive nature, is whether ambiguity and risk preferences are so similar that defy any meaningful distinction. To address these questions, we estimate the correlation between the relative rankings of participants’ results for each measurement using Spearman correlation coefficients. The results are presented in Table 6, where values in bold are statistically significant at the 5% level. In addressing the first question, it is important to notice that the correlations between all risk measurements are statistically significant and relatively high: the positive correlation between the coin toss and urn prospects (.5107, p < 0.01) largely suggests that the different procedures lead to similar conclusions. Similarly, both prospects for the CE procedure are also statistically correlated with the self-assessed risk measure.

15

Finally, and contrary to previous work, we find that ambiguity measures exhibit no statistical correlation with the risk measures at the usual 5% level of significance and only in one of the cases, the coin-toss CE procedure, does it exhibit a correlation that is significant at the 10% level: in our data, at least, ambiguity preferences are distinct from risk preferences. 5. Explaining adoption decisions

Our primary focus is to understand whether risk and/or ambiguity preferences have any significant effect on the adoption of innovations. Because most of the households in the village where we conducted our study have already adopted non-glutinous rice, we study their decision in terms of intensity of adoption by specifying a model of the form: 𝑌𝑖∗ = 𝑋𝑖 𝛽 +∈𝑖

(1)

Where 𝑌𝑖∗ is the unobserved latent dependent variable that represents the proportion of

non-glutinous rice planted by farmer i; 𝑋𝑖 is the k x n matrix of observed explanatory

variables expected to influence adoption; 𝛽 is a vector of parameters to be estimated and ∈𝑖 is

a random error term, The observed proportion on non-glutinous rice grown by farmers, 𝑦𝑖 , is left censored at 0 (no adoption) if the unobserved latent variable 𝑌𝑖∗ does not exceed the threshold level 0, after which it becomes a continuous function of the explanatory variables. 𝑌𝑖∗ , 𝑔𝑖𝑣𝑒𝑛 𝑌𝑖∗ > 0 𝑦𝑖 = � 0, 𝑌𝑖∗ ≤ 0

(2)

Under the additional assumption that∈𝑖 ~𝑁 (0, 𝜎 2 ), we can estimate this relation as a

Tobit model (Tobin, 1958), an approach used previously in studies of agricultural technology adoption, including studies of conservation adoption (Norris and Batie 1987; Gould et al. 1989) and the adoption of alternative crop varieties (Adesina and Zinnah 1993). The explanatory variables include, in addition to risk and ambiguity preferences, several other correlates of adoption identified in the literature and for which we have 16

information, collected through the household survey that we conducted: farm size (Binswanger 1978; Akinola 1987), visits from extension services (Polson and Spencer 1991; Nkonya et al. 1997), age (Bultena and Hoiberg 1983; Gould et al. 1989) and years of education (Duraisamy 1989; Liu 1991). Some summary statistics are presented in Table 7. There are two concerns with our data. The first is that we only have cross-sectional data collected after the adoption decision. Previous studies (for example, Besley and Case (1993)) raised the concern that any ex-post measurement of explanatory variables could be affected by the adoption decision, therefore being endogenous. Our selected covariates are unlikely to suffer from this problem as they are unchanging over time, and as such unlikely to be affected by initial adoption decisions that date as late as the 2009 dry season. The second is that, given the correlation between the different measures of risk preferences, multicollinearity may be a problem. To circumvent it, we will estimate a separate Tobit model with each of the risk preference variables included separately. Our estimates are presented in table 8. In columns (1) to (3), we estimate the Tobit model with only one risk measurement, whereas the estimation in column (4) includes all risk measurements. There appears to be little presence of multicollinearity caused by the risk preference variables. Only negligible changes in standard errors are observed across all 4 models and thus with exception of the intercept, the significance of all estimates are consistent across all 4 estimations. For that reason, we concentrate our discussion on the estimates in column 4. Although we are mostly interested in the relative importance of risk and ambiguity, it is important to notice that the sign of all other covariates are as expected, given the results in the ex ante literature. However, the estimates are not precisely estimated at the usual levels of significance of 5%. Focusing on our main concern, the importance of individual preferences, an immediate first conclusion is that ambiguity preferences matter, but risk preferences, irrespective of the specific measurement procedure, do not. 17

As it is known, we cannot directly interpret the Tobit estimates given that, although they allow us to observe both the significance and direction of the relationship between the dependent and explanatory variables, the coefficients only represent the marginal effects of changes on the unobserved latent variable or 𝛽 = 𝜕𝐸( 𝑌𝑖∗ )/𝜕𝑋𝑖 . In order to understand

whether ambiguity also matters in an economic sense, we follow the Tobit decomposition framework suggested in McDonald and Moffit (1980) to obtain the marginal effects of the explanatory variables on the adoption probability and use intensity. If we let the expected value of the dependant variable PROPORTION across all observations be represented by 𝐸(𝑦𝑖 ) , the expected value of the dependant variable

conditional of a farmer growing non-glutinous rice be given as 𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 ) and the

probability of the farmer being uncensored (i.e. the probability of adoption) be represented by 𝐹(𝑧), the cumulative normal distribution of z where 𝑧 =

variables can be shown as:

𝑋𝑖 𝛽 𝜎

. The relationship between these

𝐸(𝑦𝑖 ) = 𝐹(𝑧)𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 )

(3)

Differentiating equation (3), the marginal effects of a change in variable 𝑋𝑖 on 𝐸(𝑦𝑖 )

is expressed as:

𝜕𝐸(𝑦𝑖 ) = 𝐹(𝑧)𝜕𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 )/𝜕𝑋𝑖 + 𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 )𝜕𝐹(𝑧)/𝜕𝑋𝑖 𝜕𝑋𝑖

(4)

Equation (4) reveals that the marginal change in the observed dependant variable 𝑦𝑖

can be decomposed into our two parts of interest, represented in equations (5) and (6). The marginal effect of variable 𝑋𝑖 on the conditional expected value 𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 ), which we can interpret as the change in adoption intensity, is: 18

𝜕𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 ) 𝑧𝑓(𝑧) 𝑓(𝑧)2 = 𝛽𝑖 �1 − − � 𝜕𝑋𝑖 𝐹(𝑧) 𝐹(𝑧)2

(5)

where 𝑓(𝑧) represents the standard normal density and 𝛽𝑖 represents the vector of Tobit

estimates for variables 𝑋𝑖 . The change in the probability of adoption as variable 𝑋𝑖 changes is: 𝜕𝐹(𝑧) = 𝑓(𝑧)𝛽𝑖 /𝜎 𝜕𝑋𝑖

(6)

The results of this decomposition are shown in table 9, distinguishing between the marginal effects of changes on the probability and intensity of adoption. The estimates suggest participants who were more averse to ambiguity had a greater likelihood of either adopting less non-glutinous rice on their land or not adopting it at all. A farmer who has an EMV of 5000 Kip less than another farmer (i.e. they were more ambiguity-averse) subsequently has a 7.3% lower expected probability of adopting the new variety and is expected to grow 10% less of the new variety on their plots if they have decided to adopt. Our finding extends beyond the results of previous attempts to measure this association, by demonstrating that the probability and intensity of technology adoption decreases with ambiguity-aversion (Engle-Warnick et al. 2008; Alpizar et al. 2009). Conversely, risk preferences appear to have no significant relationship with adoption decisions. This contrasts with the long-held notion that risk-aversion prevents the adoption of new technology (Feder 1980; Just and Zilberman 1988; Knight et al. 2003; Liu 2007). Given there is low or no correlation with our ambiguity measure, this suggests the decision making process under ambiguity is different from the decision-making process under risk. Visits from extension services show the strongest significant relationship to the adoption of non-glutinous rice; an additional visit from extension services increases the

19

probability of adoption by 7.2% and the expected intensity by 9.8%.

Although this is

consistent with the findings of Nkonya et al. (1997), care must be taken in determining the impact of this variable due to possibility of reverse causality; where extension visits increased if non-glutinous adoption increased.

Farm size significantly and positively influenced

adoption. A farmer with 1 Hectare more in farm size is 2.4% more probable to adopt and will grow 3.3% more of the new variety if adopted. Larger farmers are likely to have more opportunities to learn about the new technology, have more incentive to adopt it, and are able to bear risks associated with early technology adoption (Feder et al. 1985). The variables of farmer age , household size, education and gender did not significantly influence nonglutinous rice production. 6. Conclusions Given the importance of innovation, the phenomenon of incomplete adoption of new technologies has appropriately received much attention in agricultural and development economics. In addition to a number of market constraints, risk-aversion dominates the discussion on the behavioural determinants to technology adoption. Somewhat paradoxically, given that the outcomes of innovations are, almost by definition, unknown to adopters (at least the earlier ones), not much attention was paid to preferences towards scenarios with unknown probabilities, that Ellsberg (1961) called ambiguity. In this paper we have investigated whether a farmer’s aversion to ambiguity is important in explaining adoption decisions. To answer this, a unique dataset was collected, combining field experiments intended to measure the behavioural parameters of risk and ambiguity preferences with a household survey, collecting information on the technology choices and socioeconomic characteristics of farmers in one developing country, Lao PDR. Given the way we measure innovation, we are able to extend beyond the simple binary 20

adoption variables employed by Engle-Warnick et al. (2008) and Alpizar et al. (2009).Also, we are able to avoid the problems with the definition of innovation that limit earlier studies (for example, Engle-Warnick et al. (2008)). We present two main conclusions. First, farmers have separate risk and ambiguity preferences. Second, and perhaps more importantly, we find that ambiguity-aversion, but not risk-aversion, significantly reduces both the probability and intensity of adoption. These findings are important for two reasons. Firstly, these findings have potential policy implications. The vast majority of the literature that proposes risk-aversion as a possible explanation for hindered adoption in developing countries then prescribes crop insurance (Liu 2007) and money-back guarantees (Sunding and Zilberman 2001) as a means to potentially hedge against production risk and reduce the fear of loss associated with new technology. Our finding that ambiguity, not risk, is important in explaining adoption decisions, implies that policy should be directed at ensuring farmers have access to greater information about the performance of new innovations, allowing them to make more accurate subjective probability evaluations on new innovations. Our additional finding, that adoption responds positively to extension, complements this. The best way of providing extension is a topic of intense debate in the literature, and we do not add to it. We simply argue that, in this case, they seem to be useful. Finally, this study connects the findings of field experiments to tangible decisions in the real world. The external validity of game experiments has been the subject of long standing debate (Samuelson 2005). Unlike experiments conducted in laboratory environments which hypothesise how risk and ambiguity dictate decision-making, our subjects are the decision-makers. The results of this study suggest that game experiments can predict real decisions, hence strengthening their validity. 21

We conclude our paper with suggestions for further contributions to the understanding of the role of ambiguity in technology adoption. Previous studies have demonstrated that farmers update their subjective beliefs of new technology over time through learning from their own experience (e.g. Ghadim 2000) and their social network (e.g. Foster and Rosenzweig 1995), increasing their likelihood of adoption.

Future investigations could

potentially link this with our findings, determining how learning impacts the adoption decisions of farmers constrained by ambiguity-aversion. Our experimental procedures elicited participants’ risk and ambiguity preferences across the domain of gains. Further comprehension of the importance of risk and ambiguity on a farmers adoption decisions could be achieved by measuring preferences over gains and losses. Prospect theory (Kahneman and Tversky 1979) describes a “reflection effect” where a decision-maker exhibits risk-aversion in the domain of gains and is relatively risk-seeking in the domain of losses, perhaps more accurately predicting the behaviour of inexperienced individuals (List 2003). This reflection effect has also been observed under ambiguity, with differing attitudes over gains and losses by Chakravaty and Roy (2009). There lies potential for future research to identify whether this exists among farmers in the developing world and what bearing it has on their preferences to risk, ambiguity and adoption.

22

References

Adesina, A.A. and Zinnah M.M. (1993). Technology characteristics, farmer perceptions and adoption decisions: a Tobit model application in Sierra Leone, Agricultural Economics 9, 297-311. Akay, A., Martinsson, P., Medhin, H. and Trautmann, S. (2009) Attitudes Toward Uncertainty Among the Poor: Evidence from Rural Ethiopia. IZA Discussion Paper No. 4225. Akinola, A. (1987). An application of Peobit analysis to the adoption of tractor hiring service scheme in Nigeria, Oxford Development Studies 16, 70-82. Alpizar, F., Carlsson, F. and Naranjo, M. (2009). The effect of risk, ambiguity, and coordination on farmers’ adaptation to climate change: A framed field experiment, rapport nr.: Working Papers in Economics 382. Anderson, J., Dillon, J. and Hardaker, J. (1977). Agricultural decision analysis, Iowa. USA. Antle, J. (1989). Nonstructural risk attitude estimation, American Journal of Agricultural Economics 71, 774-784. Bard, S. and Barry, P. (2000). Developing a scale for assessing risk attitudes of agricultural decision makers, Management Review 3, 9-25. Becker, S.W. and Brownson, F.O. (1964). What price ambiguity? On the role of ambiguity on decision-making, Journal of Political Economy 72, 65-73. Besley, T. and Case, A. (1993). Modeling technology adoption in developing countries, The American Economic Review 83, 396-402.

23

Bhalla, G. and Alagh, Y. (1979). Performance of Indian agriculture: a districtwise study. Sterling, New Delhi. Binswanger, H. (1978). The economics of tractors in South Asia: An analytical review, New York, USA: Agricultural Development Council. Binswanger, H.P. (1980). Attitudes toward risk: experimental measurement in rural India, American Journal of Agricultural Economics 62, 395-407. Bountavy, S. (2006). International trade and human development : Lao PDR 2006 : the third Lao PDR national human development report / [co-authors, Bhounthavy Sisouphanthong, Charles N. Myers]. Committee for Planning and Investment, National Statistics Centre : United Nations Development Programme, Vientiane. Bowen, J. and Zi-lei, Q. (1994). Robust Tolerance for Ambiguity, Organizational behavior and human decision processes 57, 155-165. Bultena, G. and Hoiberg, E. (1983). Factors affecting farmers' adoption of conservation tillage, Journal of Soil and Water Conservation 38, 281. Capon, T.R. (2009). An experimental study of risk, ambiguity and market behaviour, PhD diss. Griffith University. Camerer, C. and Weber, M. (1992). Recent developments in modeling preferences: Uncertainty and ambiguity, Journal of Risk and Uncertainty 5, 325-370. Chakravarty, S. and Roy, J. (2009). Recursive expected utility and the separation of attitudes towards risk and ambiguity: an experimental study, Theory and Decision 66, 199-228. Chavas, J. and Holt, M. (1996). Economic behavior under uncertainty: A joint analysis of risk preferences and technology, The Review of Economics and Statistics 78, 329-335. 24

Chow, C. and Sarin, R. (2002). Known, unknown, and unknowable uncertainties, Theory and Decision 52: 127-138. DeVellis, R. (1991). Scale Development. Theory and Applications. Applied Social Research Methods Series, Vol 26, Neybury Park: Sage. Duraisamy, P. (1989). Human capital and adoption of innovations in agricultural production: Indian evidence, Anve ak, 1. Duraisamy, P. (2002). Changes in returns to education in India, 1983-94: by gender, agecohort and location, Economics of Education Review 21, 609-622. Eggert, H. and Lokina, R. (2007). Small-scale fishermen and risk preferences, Marine Resource Economics 22, 49. Einhorn, H. and Hogarth, R. (1985). Ambiguity and uncertainty in probabilistic inference, Psychological review 92, 433-461. Ellsberg, D. (1961). Risk, Ambiguity and the Savage axioms, Quarterly Journal of Economics 75, 643-669. Engle-Warnick, J., J. Escobal, et al. (2007). Ambiguity aversion as a predictor of technology choice: Experimental evidence from Peru. CIRANO Working Papers. Feder, G. (1979). Pesticides, information, and pest management under uncertainty, American Journal of Agricultural Economics 61, 97-103. Feder, G. (1980). Farm Size, Risk Aversion and the Adoption of New Technology under Uncertainty, Oxford Economic Papers 32, 263-283.

25

Feder, G., Just, R.E. and Zilberman, D. (1985). Adoption of agricultural innovations in developing countries: a survey, Economic Development and Cultural Change 33, 255297. Foster, A. and Rosenzweig, M. (1995). Learning by Doing and Learning from Others: Human Capital and Technical Change in Agriculture, Journal of Political Economy 103, 1176-1209. Fox, C. and Tversky, A. (1995). Ambiguity aversion and comparative ignorance, The Quarterly Journal of Economics 110, 585-603. Gafsi, S. and Roe, T. (1979). Adoption of Unlike High-Yielding Wheat Varieties in Tunisia, Economic Development and Cultural Change 28, 119-133. Ghadim, A. and Pannell, D.J. (1999). A conceptual framework of adoption of an agricultural innovation, Agricultural Economics 21, 145-154. Ghadim, A., Pannell, D. and Burton, M. (2005). Risk, uncertainty, and learning in adoption of a crop innovation, Agricultural Economics 33, 1-9. Gong, Y., Baylis, K., Xu, J., Kozak, R. and Bull, G. (2010). Risk Aversion and Farm Input Choice: Evidence from Field Experiments in China, Kathy Baylis, 27. Gould, B., Saupe, W. and Klemme, R. (1989). Conservation tillage: the role of farm and operator characteristics and the perception of soil erosion, Land Economics 65, 167182. Grilliches, Z. (1957). Hybrid corn: an exploration in the economics of technological change, Econometrica 25, 501-523. Halevy, Y. (2007). Ellsberg revisited: An experimental study, Econometrica 75, 503-536. 26

Hardaker, J. (2004). Coping with risk in agriculture. CABI. Heath, C. and Tversky, A. (1991). Preference and belief: Ambiguity and competence in choice under uncertainty, Journal of Risk and Uncertainty 4, 5-28. Henrich, J. and McElreath, R. (2008). Are peasants risk-averse decision makers?, Current Anthropology 43, 172-181. Holt, C. and Laury, S. (2002). Risk aversion and incentive effects, American economic review 92, 1644-1655. Just, R.E. and Zilberman, D. (1983). Stochastic Structure, Farm Size, and Technology Adoption in Developing Agriculture, Oxford Economic Papers 35, 307-328. Just, R. and Zilberman, D. (1988). The effects of agricultural development policies on income distribution and technological change in agriculture, Journal of Development Economics 28, 193-216. Kahn, B. and Sarin, R. (1988). Modeling ambiguity in decisions under uncertainty, Journal of Consumer Research 15, 265-272. Kahneman, D. and Tversky, A. (1979). Prospect Theory: An Analysis of Choice Under Risk, Econometrica 47, 263-291. Kastens, T. and Featherstone, A. (1996). Feedforward backpropagation neural networks in prediction of farmer risk preferences, American Journal of Agricultural Economics 78, 400-415. Khan, M. H. (1975). The Economics of the Green Revolution in Pakistan. Frederick A. Praeger, New York.

27

KRLP (Khammouane Rural Livelihoods Project). (2008). Social Impact Assessment in Investment, Department of Planning. (ed.), Thakhek. Knight, F. (1921). Risk, Uncertainty and Profit. Houghton Mifflin, Boston. Knight, J., Weir, S. and Woldehanna, T. (2003). The role of education in facilitating risktaking and innovation in agriculture, Journal of Development Studies 39, 1-22. Lauriola, M. and Levin, I. (2001). Relating individual differences in attitude toward ambiguity to risky choices, Journal of Behavioral Decision Making 14, 107-122. Lin, J. (1991). Education and innovation adoption in agriculture: evidence from hybrid rice in China, American Journal of Agricultural Economics 73,713-723. List, J. (2003). Does Market Experience Eliminate Market Anomalies?, Quarterly Journal of Economics 118, 41-71. Liu, E. (2008). Time to change what to sow: risk preferences and technology adoption decisions of cotton farmers in China, PhD diss. Princeton University. Lynne, G.D., Shonkwiler, J.S. and Rola, L.R. (1988). Attitudes and farmer conservation behaviour, American Journal of Agricultural Economics 70, 12-19. MacCrimmon, K. and Larsson, S. (1979). Utility Theory: Axioms Versus “Paradoxes”, in [2], Expected utility hypotheses and the Allais Paradox: contemporary discussions of decisions under uncertainty with Allais' rejoinder, 333. McDonald, J.F. and Moffit, R.A. (1980). The use of Tobit analysis, Review of Economics and Statistics 62, 318-321. Moscardi, E. and de Janvry, A. (1977). Attitudes toward risk among peasants: an econometric approach, American Journal of Agricultural Economics 59, 710-716. 28

Munshi, K. (2004). Social learning in a heterogeneous population: technology diffusion in the Indian Green Revolution, Journal of Development Economics 73, 185-213. Nkonya, E., Schroeder, T. and Norman, D. (1997). Factors affecting adoption of improved maize seed and fertiliser in northern Tanzania, Journal of Agricultural Economics 48, 1-12. Norris, P. and Batie, S. (1987). Virginia farmers’ soil conservation decisions: An application of Tobit analysis, Southern Journal of Agricultural Economics 19, 79-90. NSC (National Statistics Centre). (1999). The Household of Lao PDR: Social and Economic Indicators, Lao Expenditure and Consumption Survey 1997/1998, LECS 2. State Planning Committee, Vientiane. O'Mara, G. (1980).The Microeconomics of Technique Adoption by Small-holding Mexican Farmers", Report, Development Research Centre, World Bank, Washington D.C. Patrick, G. and Ullerich, S. (1996). Information sources and risk attitudes of large-scale farmers, farm managers, and agricultural bankers, Agribusiness 12, 461-471. Polson, R. and Spencer, D. (1991). The technology adoption process in subsistence agriculture: The case of cassava in Southwestern Nigeria, Agricultural Systems 36, 65-78. Pope, R. and Just, R. (1991). On testing the structure of risk preferences in agricultural supply analysis, American Journal of Agricultural Economics 73, 743-748. Rogers, E. (1995). Diffusion of innovations. Free Pr. Samuelson, L. (2005). Economic theory and experimental economics, Journal of Economic Literature 43, 65-107. 29

Sarin, R. and Weber, M. (1993). Effects of ambiguity in market experiments, Management Science 39, 602-615. Savage, L.J. (1954). The Foundations of Statistics. Wiley, New York. Schiller, J. (2006). Rice in Laos. Int. Rice Res. Inst.

Slovic, P. and Tversky, A. (1974). Who accepts Savage's axiom? Behavioral science 19, 368373. SNV (Netherlands Development Organisation). (2009). Baseline Survey to Identify cash crops for VCD in Nongbok District. Netherlands Development Organisation, Thakhek. Smale, M., Just, R. and Leathers, H. (1994), "Land allocation in HYV adoption models: an investigation of alternative explanations", American Journal of Agricultural Economics 76, 535-546. Spector, P. (1992). Summated rating scale construction: An introduction. Sage Publications, Inc. Sunding, D. and Zilberman, D. (2001). Chapter 4 The agricultural innovation process: Research and technology adoption in a changing agricultural sector. in Bruce, L.G. and Gordon, C.R. (eds.), Handbook of Agricultural Economics. Elsevier, 207-261. Tobin, J. (1958). Estimation of Relationship for Limited Dependant Variables, Econometrica 26, 26-36. Von Neumann, J. and Morgenstern, O. (1944). The Theory of Games and Economic Behaviour. Princeton University Press, Princeton: NJ.

30

Weil, P.M. (1970). The Introduction of the Ox Plow in Central Gambia in McLaughlin, P.F. (eds.) African Food Production Systems: Cases and Theory. John Hopkins University Press, Baltimore Wills, I.R. (1972). Projections of Effects of Modern Inputs on Agricultural Income and Employment in a Community Development Block, Uttar Pradesh, India, American Journal of Agricultural Economics 54, 452-460. World Bank. (2008). Appraisal on the Khammouane Development Project, Rural Development, Natural Resources and Environment Sector Unit, East Asia and Pacific Region.

31

Figures and Tables

Figure 1: Likert Comparison between rice varieties with means denoted I think non-glutinous rice is.. Harder to Eat

Harder to Sell

2.60 1.13

Costs more to maintain

2.34

Needs more Seeds

3.08

More subsceptible to Excess Water

2.60

More subsceptible to Lack of Water

2.75

More subsceptible to Disease

3.00

More subsceptible to Pests

3.03

More subsceptible to Rats

3.18

Needs more Pesticide

3.10

Needs More Fertilizer

2.96

Needs More Labour at Harvest

2.09

Needs More Labour for Planting

2.66

Needs More Labour for Preparing

2.27

32

Figure 2: Results Self-Assessed Risk Attitude

13

14

12 12 10 9

Frequency

10

8 8 5

6

5 4

4 2 0

0

0 1

2

3

4

5

6

7

Self-Assessed Risk Scale

33

8

9

10

Table 1: Descriptive statistics of variables used

Variable

Mean (S.D.)

Age

48.02 (10.85)

Years of Education

5.20 (3.02)

Female

0.18 (0.36)

Household Size

4.14(1.31)

Visits from Extension

2.20 (1.01)

Non-Glutinous Rice Planted (Ha)

0.92 (0.84)

Total Land Size (Ha)

2.43 (1.50) 1.27 (1.13)

Glutinous Rice Planted (Ha) Observations

66

Table 2: Certainty Equivalent Procedure

Turn 1 2 3 4 5 6 7 8 9 10 11

Option One: Certain Payments 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000

Option Two: Urn (P(Payoffs))

Switchpoint from 1 to 2

CE at Switchpoint

0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000) 0.5(0),0.5(20000)

1 to 2 2 to 3 3 to 4 4 to 5 5 to 6 6 to 7 7 to 8 8 to 9 9 to 10 10 to 11

0 1000 3000 5000 7000 9000 11000 13000 15000 17000 19000

34

Implied Risk Preference

Risk Averse Risk Neutral Risk Preferring

Table 3: Ambiguity Preferences: Ellsberg Urns

Turn

𝑎

1 2 3 4 5 6 7 8 9 10 11

Option One: Urn (P(Payoffs)) 1(20000) 0.1(0),0.9(20000) 0.2(0),0.8(20000) 0.3(0),0.7(20000) 0.4(0),0.6(20000) 0.5(0),0.5(20000) 0.6(0),0.4(20000) 0.7(0),0.3(20000) 0.8(0),0.2(20000) 0.9(0),0.1(20000) 1(0)

Expected Monetary Value

EMV 𝒂

Option Two

20000 18000 16000 14000 12000 10000 8000 6000 4000 2000 0

?,? ?,? ?,? ?,? ?,? ?,? ?,? ?,? ?,? ?,? ?,?

35

Crossover from One to Two 1 to 2 2 to 3 3 to 4 4 to 5 5 to 6 6 to 7 7 to 8 8 to 9 9 to 10 10 to 11

EMV at Crossover

Implied Ambiguity Preference

20000 19000 17000 15000 13000 11000 9000 7000 5000 3000 1000

Ambiguity Preferring Ambiguity Neutral Ambiguity Averse

Table 4: Outcomes of CE Procedures

Implied Risk Preference Risk Averse

Risk Neutral

Risk Preferring

Elicited CE

Urn

Coin Toss

Frequency (%)

Frequency (%)

3000

3 (4.5%)

6 (9.1%)

7000

8 (12.1%)

11000

13 (19.6%)

11 (16.7%)

15000

19(28.9%)

10 (15%)

19000

1 (1.5%)

1000 5000 9000

13000 17000 Mean (S.D.) Median

2 (3%)

1 (1.5%)

3 (4.5%)

8 (12.1%)

8 (12.1%)

8 (12.1%)

7 (10.6%) 2 (3%)

11000 (4165.8) 11000

8 (12.1%)

10 (15%) 4 (6%) 0

9939.4 (4299.8) 11000

*Where a CE of 10000 indicated risk-neutrality, less than 10000 indicated risk-preferring and greater than 10000 indicated risk aversion.

36

Table 5: Outcomes of Ambiguity Procedure Implied Ambiguity Preference

Elicited EMV

Frequency (%)

1000

0

Ambiguity Averse

3000

3 (1.5%)

7000

10 (15.15%)

11000

9 (13.6%)

5000

5 (7.6%)

9000

Ambiguity Neutral

13 (19.7%)

13000

10 (15.2%)

17000

4(6.1%)

15000

Ambiguity Preferring

19000

Mean (S.D.) Median

8(12.1%) 4(6.1%)

10845.29 (4202.785) 11000

Table 6: Correlation of Procedure Results Spearman rank correlations 𝒂

Risk: Coin Toss

Risk: Container

Risk: Coin Toss Risk: Urn Risk: Self-Assessed

1 0.511 (0.00) 0.247 (0.045)

1 .2820 (.022)

Ambiguity 𝑎

0.216 (0.082)

P-values are in parentheses

.097 (.440)

37

Risk: SelfAssessed

1 -0.032 (0.798)

Ambiguity

1

Table 7: Definitions of Variables in Empirical Model Dependent Variable PROPORTION

The proportion of total plot area of which nonglutinous rice varieties are planted.

AGE

Age of the household head, measured in years.

HOUSEHOLDSIZE

Number of people in the household

Independent Variables

GENDER

FARMSIZE

VISITFROMEXT. EDUCATION AMBIGUITY COINTOSS URNGAIN PSYCHOMETRIC

Gender of the household head, where male = 1. Total plot area, measured in hectares.

Visits from extension agents; measured as number of visits Education of farmer, measured in years of education.

Ambiguity preferences, measured as the elicited Expected Monetary Value. Elicited risk preferences using urn prospect, measured as the elicited Certainty Equivalent.

Elicited risk preferences using urn prospect, measured as the elicited Certainty Equivalent.

Self-assessed risk, measured on scale of 1 to 10.

38

Table 8: Tobit Left-Censored Regression Estimates Independent Variablea

INTERCEPT AGE

HOUSEHOLDSIZE GENDER

FARMSIZE

VISITFROMEXT. EDUCATION AMBIGUITY

RISK CE: COINTOSS RISK CE: URN

PSYCHOMETRIC

Estimated 𝑌𝑖∗ at means

Variance of Error Term Pseudo 𝑅2

Observations = 66

(1)

(2)

(3)

(4)

-.45764* (.24493) -.00023 (.00298) -.00834 (.02712) .00226 (.09363) .04466* (.02421) .13290*** (.03896) .00799 (.01079) .00003*** (7.50e-06) 8.84e-06 (7.46e-06)

-.37021 (.25304) -.00048 (.00300) 0.0008 (.02689) .01648 (.09439) .04143* (.02426) . 12472*** (.03902) . 00590 (.01110) .00003*** (7.45e-06)

-.39585 (.24469) -.00052 (.00302) -.00061 (.02660) .01393 (.09419) .04103* (.02444) .12471*** (.03925) .00629 (.01084) .00003*** (7.45e-06)

-

-1.57e-06 (7.54e-06)

.34830

-

-

.34900

.00225 (.01433)

-.39895 (.25106) -.00022 (.00299) -.00741 (.02714) .00519 (.09302) .04564* (.02439) .13109*** (.03972) .00579 (.01113) .00003*** (7.47e-06) .00001 (8.73e-06) -8.34e-06 (8.90e-06) .00032 (7.48e-06)

.564

.23633 .533

.02353

.23249

.23430

-

-

.34880 .532

.37490 .5856

Note: standard errors within parentheses. ***: significant at 1%, *: significant at 10%

39

Table 9: McDonald and Moffit Tobit Decomposition Marginal Effect of:

AGE

HOUSEHOLDSIZE GENDER

Adoption Probability

Standard Error

Expected Use Intensity

Standard Error

-.0041168

.01509

-.0056603

.02073

-.0001239 .0029232

FARMSIZE

.0253438*

EDUCATION

.0032156

VISITFROMEXT. AMBIGUITY COINTOSS URNGAIN

PSYCHOMETRIC

.0727838*** .0000145*** 7.23e-06

-4.63e-06 .0001789

.00166 .05308

-.0001704

.00229

.0039528

.07061

.01509

.0348459*

.00624

.0044212

.02876

.1000726***

.01864 .03068 .0085

.00001

.00002***

.00001

.00001

-6.37e-06

.00001

.00001 .00846

***, significant at 1% *, significant at 10% Expected value of 𝐸( 𝑦𝑖 ∣ 𝑦𝑖 > 0 ) at means of X =0.375 Expected probability of adoption at means of X= 0.9260

40

9.95e-06

.0002459

.00001 .01163