The Journal of Socio-Economics 37 (2008) 1729–1745

Contents lists available at ScienceDirect

The Journal of Socio-Economics journal homepage: www.elsevier.com/locate/soceco

Preference variability along the policy chain in Vietnam C. Leigh Anderson a,∗ , Alison Cullen a,1 , Kostas Stamoulis b a b

University of Washington, Seattle, WA, USA Food and Agriculture Organization of the United Nations, Viale delle Terme di Caracalla, 00100 Rome, Italy

a r t i c l e

i n f o

Article history: Received 4 May 2007 Received in revised form 22 February 2008 Accepted 6 April 2008 JEL classification: O1 O2 Keywords: Policy Decision-making Risk Vietnam

a b s t r a c t This paper explores whether there are systematic differences in decision-making between those who regularly allocate public resources and those who are the intended recipients. To test for differences we sample across farmers and policy makers in Vietnam. Our findings suggest that preference parameters such as fairness, risk orientation, discounting and control systematically differ between these two groups, and are predictors of the likelihood that an individual is in a position of allocating public resources or receiving them. Regardless of whether these differences are innate or socialized, they may help to explain the often unexpected outcomes of development policy interventions. © 2008 Elsevier Inc. All rights reserved.

1. Introduction Individuals make decisions both about allocating their own resources, and to varying degrees, about allocating collective resources. Our interest is in whether there are systematic differences in decision-making between those who are regularly involved in allocating public resources for development, and the poorer individuals who are more frequently the intended recipients. To test for differences we sample across two groups of individuals who vary by the control and responsibility they have over public resource allocation decisions in Vietnam. The motivation for our study is that a better understanding of patterns in decision-making can help to explain the often unexpected outcomes of development policy interventions aimed at poverty reduction. The mixed performance of 60 years of development assistance has led to changes in the prevailing paradigms including the search for the proper balance between “state-led” and “market-led” solutions. We are suggesting that this distinction may be artificial or, in any case, not the most important one. Rather, development policy frameworks may have missed salient points of peoples’ behavior which determine the success (or lack thereof) of the policies themselves. First, institutional and program incentives may be premised on a strict model of rationality that ignores experimental findings on risk and fairness perceptions, and second, policy and resource decisions may be premised on the preferences of the policy maker, which can differ from the preferences of the intended recipients. Traditional economic models of decision-making assume that individuals make fully reasoned and consistent choices. Yet arguments of bounded rationality dating back 50 years, regular observation, and repeated experiments fail to support such models, particularly for complex decisions made under uncertainty (Simon, 1955). Individuals often employ decision

∗ Corresponding author at: Evans School, Box 353055, University of Washington, Seattle, WA, USA. Tel.: +1 206 543 0365; fax: +1 206 543 1096. E-mail addresses: [email protected] (C.L. Anderson), [email protected] (A. Cullen), [email protected] (K. Stamoulis). 1 Evans School, Box 353055, University of Washington, Seattle, WA, USA. 1053-5357/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.socec.2008.04.011

1730

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

heuristics; simple rules that allow them to make decisions in the absence of full information, or when they are unable or unwilling to incorporate all the information that is available in the relevant timeframe (Gigerenzer and Selten, 2001). Fairness and some qualitative dimensions of risky outcomes, such as familiarity and a sense of control, can also affect choices (Fischhoff et al., 1978). While there is an impressive inventory of experimental findings, most come from experiments in the U.S. or European laboratory settings with relatively homogeneous and well-off populations (for example, see Thaler, 1991; Camerer et al., 2003). Subsequently we know very little about the nature and prevalence of these behaviors in the field, and how patterns vary across population characteristics. Our paper contributes to the literature in two ways. First, we test whether these behaviours are robust across a sample in Vietnam and how preferences vary with basic demographic and socio-economic characteristics. Second, we explore specifically whether preferences vary across groups – policy makers and program recipients – that vary in their responsibility over allocating development resources. The next section outlines hypotheses related to why we might expect decision-making between these two groups to differ. In Section 4 we describe our sample and present the results from a series of questions posed to farmers and policy makers in Vietnam. Our findings indicate several significant differences in preferred choices and measures of risk, control and fairness. We explore possible reasons for these differences in Section 4 by looking at preference parameters as predictors of respondent job category. Section 5 concludes. 2. Preference differences We have long understood that individuals have different preferences. But given those different starting points, theories of decision-making nonetheless assume some common and predictable responses to changes in opportunities and constraints. The exercise of searching for ways to better model behavior that seemingly deviates from these expectations has often been empirically driven, based on 30 years of experiments, largely conducted in classrooms in the U.S. and Europe. Initially, most experiments focussed solely on recording “behavioural anomalies”. More recently, interest has grown in field research and other strategies to look at differences across individual characteristics, including whether certain characteristics are associated with a greater likelihood of behaviour that contradicts standard axioms. That is, are there certain subpopulations for whom standard models perform better or worse? Thus far research has primarily focused on gender differences, though there is a small literature on stakeholder preferences in health care and “consumer versus citizen” preferences for environmental goods. We briefly review each of these literatures, and the support they provide for positing preference variability between policy makers and program recipients. A priori, why would we expect preferences, and hence choices, to systematically vary between policy makers and program recipients? Certainly in international development policy we typically believe that choices between these two groups will diverge, but that this difference stems primarily from differences in constraints and opportunities. Policy makers are expected to have more formal education, higher income and wealth, and access to a wider range of, though potentially less local and relevant, information. Participatory methods, stakeholder representation, and similar concepts are intended to address these different perspectives. Differences among stakeholders, where income is not the distinguishing feature among groups, have been examined in the health care literature. For example, Shumway et al. (2003) find differences in preferences over schizophrenia treatment outcomes between public policy makers and mental health care providers, and patients and their families. Similarly, Saigal et al. (1999) find differences in preferences for neonatal care between health care professionals and adolescents. In these cases, different decision frameworks and goals are hypothesized to drive the patterns of preferred outcomes. Hence, private attitudes may be similar, but in their role allocating public resources, policy makers may favor broader public goods, and more visible and measurable outcomes. A variation on this theme appears in the environmental literature on consumer versus citizen preferences, whereby the same individual is assumed to have multiple preferences that are distinguished by the degree of altruism arising from their decision-making role (Sagoff, 1988). Levitt (1996) and Shumway et al. (2003), however, suggest that even in the role of public decision maker, individuals may vote according to their private preferences. Levitt finds empirical support that senators give most weight to their own preferences, rather than those of the public they are representing, though he does not explore if, or why, there may be systematic differences. Allocating public resources according to personal preferences has implications when there is a preference divergence between those allocating and designing policy, and those who are the intended recipients. Leaving aside the difficult question of whether the allocation is welfare enhancing, if policies reflect policy maker preferences and these differ from program recipients, we can expect that programs will produce unanticipated results, including low take up rates and an inability to target intended recipients. Hypotheses of biological or evolutionary preference differences appear almost exclusively in studies of gender. Labor market economists have looked at preference differences to explain the “gender gap” and wage discrimination. This fairly large literature on gender differences, presumed both evolutionary and socialized, is summarized by Croson and Gneezy (2004). They suggest three areas where gender preferences may vary systematically: risk taking, social preferences, and reaction to competition. In one developing country example that supports this, Chattopadhyay and Duflo (2004) find that investment in the local public good infrastructure that is of primary concern to women dominates in Indian villages where local village council positions are reserved for female policy makers.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1731

The reproductive cost argument common to hypotheses of differences in innate gender preferences is difficult to carry through to policy makers and program recipients (Croson and Gneezy, 2004) There is, however, the possibility that for biological or other reasons, policy makers will, on average, have different risk attitudes and social preferences. In an evolutionary sense, certain risk attitudes may have both predisposed individuals to seek public decision-making roles, and by exercising their private preferences in those roles, they created policy and institutions that perpetuate the well being of like-minded people. Alternatively, if socialized, policy makers may understand that they are expected to make decisions on behalf of a broader public good. Or the particular informational, political, and economic factors that make up the local environments may lead policy makers and program recipients to view fairness and risk differently. Flynn et al. (1994) and Finucane et al. (2000) have found a “white male effect.” The lower risk perceptions of this group are hypothesized to arise from their greater involvement in creating, controlling and benefiting from technology and other activities seen as hazardous. Since decision makers in many societies are predominantly men of the dominant ethnicity, it may be that it is the decision-making position, rather than gender or race per se, that gives rise to the lower risk perceptions. To what extent these base attitudes are innate and largely immutable, or more endogenous to different environments, is an open debate. What is of interest to us, is whether systematic differences in these latent variables between policy makers and program recipients leads to different choices. Similar to the gender literature, we focus on two factors that may be relevant to differences in one’s position as a policy official allocating public resources, and a private farmer: risk attitudes and social preferences. We do not posit causality – it may be that low income program recipients are more risk averse and have higher discount rates because of the environment they live in – that is, the preference variability is due to income and educational differences (Lawrance, 1991). Or, it may be that risk aversion and impatience played a role in determining who is a policy maker and who is a program recipient—that is, the preference variability is due to latent variables that self-select individuals into different decision-making positions through migration and educational choices (Fisher, 1930). For social preferences, we concentrate on attitudes toward fairness and responsibility. We suggest that development policy and program failures may arise from using behavioural models that fail to account for relevant preference parameters, and that these preference parameters may systematically differ between individuals allocating public resources and the intended recipients due to either innate or socialized differences in risk and social preferences. In the following sections, we describe a series of questions to assess preference differences between these two groups. 3. The Vietnam experiment A stated preference survey instrument was administered in Vietnam during the spring of 2005 by enumerators from the Institute of Sociology in Hanoi. We surveyed 40 relatively poor rural workers (mean monthly income of VND 842,062, or approximately US$ 53) and 47 middle and high level professionals involved in policy making (mean monthly income of VND 4,210,213, or approximately US$ 265). We refer to the former group as farmers, the occupation identified by 90% of this sub-sample. We refer to the latter group as policy makers, though their level of direct involvement in policy varies. We asked all individuals to respond personally, rather than in their public role, to abstract from the political economy question of what individuals in the public sector are trying to maximize, if not their own self interest. The farmers were randomly sampled in Cao Ky commune in Bac Can province. The selection of policy makers was carefully considered to be fairly representative of decision makers who are involved in allocating public resources to rural communes. The policy maker survey focused on senior officers in central agencies including the Ministry of Finance, the Bank of Agriculture and Rural Development and Bank of Social Policies (banks that provide credit for rural areas and the poor), the Ministry of Planning and Investment, the Ministry of Agriculture and Rural Development, the Banking Institute, and the Ministry of Health. Additionally, some provincial, district and communal officers and officers of major hospitals, were surveyed. Ninety percent of policy makers were born in rural communes. Table 1 presents the main descriptive statistics of the sample with further details provided in Appendices A and B. The final survey instrument contained standard questions to assess risk and time preference, views of fairness, and sources of information. Respondents were also asked a series of questions designed to identify the cognitive effort they apply, and their assessment of the qualitative and quantitative dimensions of risky outcomes including their sense of control over, responsibility for, and experience with, various decision outcomes. This choice of questions was driven by three considerations: earlier experimental results, contextual relevance, and priors that preference differences, if they existed, would most likely arise around risk attitudes and social preferences. Each question was repeated for decision-making in health, food, and hiring decisions, to assess different risk perceptions across more or less familiar decision domains, and to assess construct validity. Experimental studies using hypothetical questions have the advantage of allowing the researcher to abstract from the complexities of real decisions, but they introduce new concerns, particularly around validity and the complexities of hypothetical scenarios. Researchers have looked at how stated preferences correspond to revealed preferences, beginning with Binswanger’s (1980) work in rural India where he found that experimentally measured degrees of risk aversion were related to actual economic decisions.2 Davies and Lea (1995) examined a similar question for time preference, and found that student attitudes toward debt were related to actual debt. In experiments with hypothetical rewards, there is also a concern that

2

This issue is particularly important when estimating willingness to pay, as discussed in Carson et al. (2000).

1732

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

Table 1 Descriptive statistics of main characteristics Characteristics

Farmers (n = 40) (%)

Policy makers (n = 47) (%)

Survey respondent category Farmers Officers

46.0 –

– 54.0

Gender Male Female

75.0 25.0

80.9 19.1

Age groups Under 30 30–50 Over 50

12 51 23

14.0 57.3 26.4

Highest level of education Illiterate Primary school Secondary school High school Junior college University Master’s degree Ph.D.

2.5 30.0 47.5 17.5 2.5 0.0 0.0 0.0

0.0 0.0 0.0 2.1 2.1 48.9 17.0 29.8

Household size Two people Three people Four people Five people Six people Seven people Eight people No answer

7.5 30.0 35.0 10.0 7.5 7.5 2.5 –

2.1 23.4 61.7 6.4 4.3 – – 2.1

Household income 150,000–500,000 500,000–1,000,000 1,000,001–2,000,000 2,000,001–4,000,000 4,000,001 and over

30.0 35.0 35.0 – –

– 15.0 5.0 50.0 45.0

Coin toss Cautious Risky

57.5 42.5

40.4 59.6

respondents will have little incentive to think hard about their responses, though in a review of 74 experiments Camerer and Hogarth (1999, p. 8) conclude that for “the kinds of task economists are most interested in, like trading in markets, bargaining in games and choosing among risky gambles, the overwhelming finding is that increased incentives do not change average behavior substantively (although the variance of responses often decreases).” Nonetheless, these cautions still apply in interpreting our results. We expect our decisions across domains to give us some measure of construct validity. That is, we expect, a priori, that farmers will be more familiar with seed choices, policy makers with hiring job candidates, and that both will be equally familiar with having to make decisions to treat sick children, though the options available to each may vary. We also have priors on the expected responses of several questions that were chosen because results are available from previous laboratory experiments in the U.S. Regardless, what is important for our purposes is that any biases are common across farmers and policy makers, since our interest is in differences across these groups. Another common concern of stated preference surveys is the cognitive complexity and market realism of the questions (Viney et al., 2005). This concern can be exacerbated working in countries requiring language and other cultural translations, and with less educated populations. To the extent that we are interested in how education and income affect choice, the variation between our two groups is desirable. Of concern, however, is if the complexity leads one or both groups to disregard the question and supply answers at random. To reduce this possibility and increase confidence in our data, our instrument was pre-tested in the field by a team of experienced local Vietnamese researchers led by a University of Washington trained Ph.D. sociologist, Nguyen Minh. We also asked one of the field managers, Dr. Nguyen Xuan Mai, to record any difficulties that respondents had answering the question, or other anomalies in the process. Nguyen (2005, p. 9) reports that: The subject of economic behavioral survey and the design of questionnaire with suppositions haven’t been used popularly in Vietnam. Therefore, they were strange to interviewed people and even researchers from different institutes

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1733

or industries. Investigators had to practise for a long time, discuss, do trial investigation, localize and explain them in different ways suitable to certain groups of subjects. Despite the question difficulty, we have no reason to believe that respondents in either group answered randomly. On the contrary, Nguyen (2005, p. 9) reports that on average the farmers took over two hours to answer the interview and the policy makers took one and one half hours. Further, Nguyen (2005, p. 9) reports “The thoughts of poor peasants, especially those with low level of education, were very concrete.” And “Most officers considered the mathematical problems deliberately, some even answered them in writing and produced accurate answers.” (Nguyen, 2005, p. 12). This thoughtful consideration of hypothetical tradeoffs was exactly our goal, allowing us to assess if that deliberation produces systematically different responses between groups. 3.1. Question series one: ambiguity and attributes by domain We offered a choice among options that involved multiple attributes, but with uncertainty over the value of any particular attribute. Individuals were asked to choose among options where the higher the sum or mean level of attributes, the greater their variance. The extremes were a high mean, high variance choice with a zero attribute, and a low mean, zero variance choice where each attribute had exactly the same level. We refer to these options as the zero option and the uniform option; options that represent attributes as perfect substitutes or perfect complements for the decision maker. A threshold option represents a middle position. Respondents were not directed to sum attribute values, or perform any other operation, since our interest was precisely in their choice without being given specific directions. Since the ranking of options by their mean and sum is equivalent, we will use these terms interchangeably. We asked respondents to choose among three alternatives with multiple attributes in three domains: a medicine for treating diarrhoea, a modern variety seed, and a loan officer job candidate for a local savings cooperative. Our assumption was that job category would not affect responses to the medicine question, but that farmers would be more familiar with the seed variety choice, and policy makers would be more familiar with the job candidate choice. This variation was introduced to test whether candidates would be more likely to make different mean–variance tradeoffs in cases of greater familiarity or relevance to them. The risk literature suggests that perceptions of risk intensify with outcomes that are dreaded, and with events that are less familiar or perceived as catastrophic or proximate (McDaniels et al., 1997; Slovic et al., 1991). These qualitative dimensions of risky outcomes can lead individuals to have a different willingness to trade off a high mean for a low variance across domains. The question was initially asked with three choices (the base case), and then after supplying respondents with an additional piece of information it was repeated for a different set of choices (the modified case). For shorthand, we refer to the alternative options as: P-0: The zero option (maximum sum and mean but greatest variance and with one zero attribute); P-1: The threshold option (middle sum, mean and variance and all attributes are greater than zero); P-2: The uniform option (low sum, mean and zero variance as all attributes are equal). The base case question on medicine read as: You have been asked to choose among three medicines to help with your child’s diarrhoea. The doctor tells you that each of the three has different strengths and weaknesses in reducing discomfort, killing the bacteria, preventing dehydration, and speeding recovery. You have the scores of each medicine from medical tests, but you don’t know which score applies to which category. Each score is out of 7. Which medicine would you choose? Medicine 1—7, 7, 7, 0 Medicine 2—4, 4, 4, 4 Medicine 3—3, 3, 6, 6 The base case question on seed varieties read as: You need some new seed for your own fields or for the fields of someone you know. The vendor has three varieties that differ in the taste of the product, its yield, and its resistance to pests and disease. An extension officer provides the scores out of 10, given below, for each variety in the three categories, but doesn’t know which score goes with which attribute. Circle the variety you would choose. Variety 1—4, 5, 9 Variety 2—6, 6, 6 Variety 3—10, 10, 0 The base case question on job candidates read as: You have been asked to help select someone to manage a new local credit and savings cooperative. Their scores out of 10 on tests of honesty, hardwork, experience, and knowledge, are below, but the order of the four categories has been mixed up. Circle the candidate you would select.

1734

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

Table 2 Medicine choice: base case Farmer percentage

Policy maker percentage

Total percentage

Zero option Threshold option Uniform option

25.6** 30.8 43.6

8.5** 31.9 59.6

16.3 31.4 52.3

Total count

39

47

86

T-test across respondent category: statistical significance level (**) 0.05. Pearson 2 = 4.892. Sig. two-sided = 0.087, d.f. = 2. Table 3 Seed choice: base case Farmer percentage

Policy maker percentage

Total percentage

Zero option Threshold option Uniform option

33.3*** 20.5 46.2***

4.3*** 10.6 85.1***

17.4 15.1 67.4

Total count

39

47

86

T-test across respondent category: statistical significance level (***) 0.01. Pearson 2 = 16.502. Sig. two-sided = 0.000, d.f. = 2.

Candidate 1—10, 10, 10, 0 Candidate 2—4, 4, 9, 9 Candidate 3—6, 6, 6, 6 In all three domains, the zero option (Medicine 1, Variety 3, Candidate 1) has the largest single attribute, the highest sum, and the greatest variance. The uniform option where all attributes are equal (Medicine 2, Variety 2, Candidate 3) has the smallest sum in the medicine and job candidate case. The threshold option in which all attributes are above zero but uneven (Medicine 3, Variety 1, Candidate 2) has the second highest sum in the medicine and job candidate cases and is tied with the uniform option for seed variety. The attributes were chosen to represent tradeoffs (varying levels of strength or weakness) in desirable properties of the option. That is, a “0” for “killing the bacteria” does not imply that the medicine is necessarily completely ineffective in this regard, but that it receives the lowest possible rating. If attribute levels are interpreted as absolutes, however, it still does not imply that the medicine is completely ineffective, as the body can fight off and kill the bacteria naturally while benefiting from the other medicine properties. Nonetheless, it may be that some respondents do not consider these attributes to be reasonably independent. Again, this matters to our results only if there is some latent reason that the responses within one group are driven by this question construct, and the responses within the other group are not. Our base case results for the choices of program recipients and policy makers are below. In all three domains – medicine, seed, and job candidate – the greatest number of respondents in both job categories opted for a positive value of all attributes, rather than choosing the option with the highest sum and zero attribute. Most respondents chose the uniform option although the frequencies vary across domains and by respondent category. Table 2 shows the percent of respondents choosing each alternative for the medicine case. The ranking of choices for both program recipients and policy makers is: P-2 > P-1 > P-0. T-tests at the 5% level of significance indicate that in two-way comparisons farmers are significantly more likely to choose the zero option than are the policy makers. As expected, differences between policy makers and farmers are more pronounced in the domains that differ by professions – seed and job candidates – than in the medicine domain which is a more common to all households.3 The results in Tables 3 and 4 indicate that the dominant choice is the uniform option P-2. In these two domains, however, the rank ordering between respondent categories differs. The ordering is P-2 > P-0 > P-1 for farmers and P-2 > P-1 > P-0 for policy makers. For the decision about seeds, a significantly greater proportion of farmers than policy makers chose the zero option, and a significantly smaller proportion chose the uniform option. Pair-wise comparisons support significant differences between the groups for both the zero and uniform option; in the case of seeds, at a 1% significance level. For job candidates, the policy makers’ decision procedure ordering remains the same as in the first two domains, but the farmers’ ordering is P-2 > P-0 = P-1. In all three cases the computed 2 , at a 10% or better level of significance, allows us to reject the null hypothesis that the survey respondent categories and choices are independent events. That is, the choice is dependent on whether the survey respondent is a farmer or policy maker. Farmers and policy makers make different mean–variance tradeoffs across domains, and from each other. These results across domain are consistent with our prior that policy makers are unfamiliar with

3

There were six childless respondents.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1735

Table 4 Candidate choice: base case Farmer percentage

Policy maker percentage

Total percentage

Zero option Threshold option Uniform option

25.6* 23.1 51.3**

10.6* 17.0 72.3**

17.4 19.8 62.8

Total count

39

47

86

Farmer percentage

Policy maker percentage

Total percentage

18.2 22.7 59.1

10.5 26.3 63.2

14.6 24.4 61.0

22

19

41

35.3*** 41.2 23.5***

7.1*** 35.7 57.1***

17.8 37.8 44.4

17

28

45

T-test across respondent category: statistical significance levels (**) 0.05 and (*) 0.10. Pearson 2 = 4.651. Sig. two-sided = 0.098, d.f. = 2. Table 5 Medicine choice base case controlled for risk

Cautiousa Zero option Threshold option Uniform option Total count b

Risky Zero option Threshold option Uniform option Total count

T-test across respondent category: statistical significance level (***) 0.01. a 2 = 0.783, d.f. = 2. b 2 = 0.024, d.f. = 2.

choosing seeds and farmers with choosing job candidates, but both share some familiarity with choosing medicines for sick children or siblings. This supports earlier work on how the qualitative dimensions of risky outcomes affect decision-making (Fischhoff et al., 1978). The results across decision maker also have potentially important policy implications. For example, if a policy maker in Hanoi is making decisions over seed technologies to make available in the countryside, they may sacrifice higher yields for less variance than the farmers would choose for themselves, resulting in lower than expected adoption rates. 3.2. Is risk aversion a driver? The zero option has the highest sum in the base decision cases, the largest individual attributes, but the greatest variance. It contains a zero for one (unknown) attribute, suggesting that risk or ambiguity aversion may influence the choice of decision procedure.4 Though several of the attribute values in the other options are low, we presume, following the evidence of experiments in non-linear probability weighting, that the difference between zero and any positive score is more meaningful than between equivalent increments of two positive numbers (Tversky and Wakker, 1995). The survey contained three standard coin toss questions to measure risk aversion. Responses among questions were significantly correlated, so we chose the question that offered a better than fair gamble over a gain. This avoids the confounding effects of loss aversion and defines our risk takers more conservatively. As a group, farmers were more risk averse. On the coin toss question 44% of farmers and 59% of policy makers were risk takers and chose the gamble. When the sample as a whole is divided according to risk attitudes, the results indicate that the respondent’s decision procedure in all three domains is independent of whether they are risk averse or risk taking. This suggests that risk attitudes do not significantly affect the option chosen. When we control for respondent category, however, some interesting results emerge. In all cases, policy makers were more likely to choose uniformity than were farmers. Despite being more risk averse as defined by the coin toss question, farmers were more likely to choose the option with the zero attribute. In all three cases, the null hypothesis of independence was rejected, implying that the choice was dependent on whether the respondent was a farmer or policy maker. Between the risk averse group of farmers and policy makers, this same pattern emerged but was only significant at the 5% level for seed choices (Tables 5–7). To summarize our findings from this first battery of questions: 1. for both respondent categories and in all three domains, the uniform option was the most common choice; 2. policy makers’ decision ordering was constant for all three domains, though proportions varied—uniform, threshold, zero;

4

Each attribute has an equal and calculable probability of being represented by each score.

1736

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

Table 6 Seed choice base case controlled for risk

Cautiousa Zero option Threshold option Uniform option Total count

Farmer percentage

Policy maker percentage

Total percentage

31.8*** 18.2* 50.0**

5.3*** 5.3* 89.5**

19.5 12.2 68.3

22

19

41

35.3*** 23.5 41.2***

3.6*** 14.3 82.1***

15.6 17.8 66.7

17

28

45

b

Risky Zero option Threshold option Uniform option Total count

T-test across respondent category: (*) 0.10, (**) 0.05, (***) 0.01. a 2 = 0.025, d.f. = 2. b 2 = 0.007, d.f. = 2.

3. for seed choice, a greater proportion of farmers chose the zero option than the threshold option for their second choice, and for job candidates an equal proportion chose the zero and threshold option for their second choice; 4. program respondent category and option choice are dependent events for seed variety and job candidate choice, but not for medicine choice; 5. risk attitude appears to affect farmers’ choices more than policy makers’ choices—risk taking farmers were more likely to choose the zero option than risk averse farmers and than risk taking policy makers. 3.3. Modified case Respondents were then asked to complete a second round of “modified” questions choosing among three more medicines, seed varieties and job candidates. Choices changed cardinally and ordinally, with the threshold option now having the highest sum and mean. Respondents were told that an individual whose opinion we posited may matter – an elderly (long lived) local in the case of the medicine, a neighbor in the case of seed, and a friend in the case of the job candidate – had chosen the option that corresponded to what we are calling “zero.” For example, for seed the question was: Another vendor has three other varieties that have been ranked out of 14. You are told that your neighbor has chosen variety 5. Circle the variety you would choose. Variety 4—8, 8, 8 Variety 5—12, 12, 0 Variety 6—6, 8, 11 The options for decision sequences are: P-3: P-4: P-5: P-6:

No switch (preferred option base case = preferred option modified case); Follow recommended (preferred option base case = threshold or uniform; preferred option modified case = zero); Follow the highest sum/mean (preferred option base case = zero; preferred option modified case = threshold); Change to the highest sum/mean (preferred option base case = uniform; preferred option modified case = threshold);

Table 7 Candidate choice base case controlled for risk

Cautiousa Zero option Threshold option Uniform option Total count

Farmer percentage

Policy maker percentage

Total percentage

22.7 22.7 54.5

15.8 10.5 73.7

19.5 17.1 63.4

22

19

41

29.4** 23.5 47.1*

7.1** 21.4 74.1*

15.6 22.2 62.2

17

28

45

b

Risky Zero option Threshold option Uniform option Total count

T-test across respondent category: statistical significance levels (**) 0.05 and (*) 0.10. a 2 = 0.421, d.f. = 2. b 2 = 0.111, d.f. = 2.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1737

Fig. 1. People who switch: decision tree—medicine. All percentages are of the total sample. Due to slight variance in sample size (from 87 to 86), percentages may be slightly off.

Fig. 2. People who switch: decision tree—seed. All percentages are of the total sample. Due to slight variance in sample size (from 87 to 86), percentages may be slightly off.

P-7: Reversal (preferred option base case = zero or threshold; preferred option modified case = uniform). The decision trees in Figs. 1–3 indicate the frequency of decision sequences for the base and modified cases for the sample as a whole. The overall probability of an individual switching does not depend on respondent category. Farmers are more likely than policy makers to switch when their base option was uniform or threshold, and less likely to switch when their base choice is the zero option. The numbers within any one decision procedure are too small for statistical significance, but within this sample suggest that farmers are more likely to switch to the recommended option in response to the new information. Following Simon’s work in the mid-nineteen fifties, a large literature has emerged on bounded rationality from psychologists and economists in the U.S. and Europe (see Tversky and Kahneman, 1981; Gigerenzer and Selten, 2001). This includes aspiration level theories such as Simon’s (1955, 1979) satisficing, and also decisions based on fast and frugal heuristics. These theories can be grouped according to whether or not the implied choice function satisfies the consistency condition, sometimes referred to as the independence of irrelevant alternatives.5 Our interest is in whether there are systematic differences between our two groups of decision makers in the frequency with which they violate this condition.

5

One of Sen’s two properties of Houthakker’s axiom of revealed preference.

1738

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

Fig. 3. People who switch: decision tree—candidate. All percentages are of the total sample. Due to slight variance in sample size (from 87 to 86), percentages may be slightly off.

We consider the base case set of options to be an operationally equivalent subset of the modified set of options. The relative attractiveness of these options may vary if one chooses to consider the new piece of information, but with a few exceptions discussed below, what we consider the important features of the options remain the same: the sum and mean, the variance, risking a zero attribute, and the uniformity of the attributes. None of the first four options (P-3 through P-6) violate the consistency condition. Regardless of the initial mean–variance tradeoff, in P-3, they maintain their ordering in the base and modified case. In P-4, we assume that the decision maker is responding to new information; in this case, they are following the choice or recommendation of a local villager, neighbour, or friend. Although the highest sum has moved from the zero to the threshold option, if the original choice of the zero option was with respect to a single element, rather than the sum (or mean), we cannot rule out consistency by staying with the zero option. Likewise, in P-5 the respondent would be following the choice with the highest sum. Even P-6, at least in the domain of seeds, can be consistent if the preference ordering is more layered than our simple specification. For example, consider a decision rule: first all elements in the set must be greater than zero, then choose the set with the highest sum; as a tie breaker, choose uniform. The result for seeds would be uniform in the base case (when uniform and threshold have the same sum) and threshold in the modified case where (sum threshold > sum uniform). For the medicine and job candidate domains, however, the preference ordering would have to be something such as: first all elements in the set must be greater than zero, then choose uniform unless a set with all elements greater than zero is also the set with the highest sum. It is difficult to induce a choice function that satisfies the consistency condition under decision-making procedure P-7. Zero dominates uniform and threshold in terms of sum in the base case. If zero is initially chosen, then zero (no switch) or threshold (shifting to the highest sum) is expected in the modified case. If threshold is initially chosen because all elements must be greater than zero, then we would expect threshold to be chosen in the modified case since its dominance in terms of the sum grows. If one allows for switching based on relative gains, then in the case of medicine and the job candidate, it can be argued that the sum of the uniform option increases by the same as the threshold option (eight in the medicine case), but more than the zero option (five in the medicine case). So even though the sum is still higher with the zero option, the change has been less dramatic. Hence, one could allow for an ordering where only switching from threshold to uniform was inconsistent. But in the seed case, even this does not hold as the sum of the threshold option increases by the most, and the zero and uniform options by the same amount; it becomes extraordinarily difficult to think of a sensible ordering. Our results indicate that farmers are far more likely to follow decision-making sequence P-7. In the domain of seeds, for instance, up to one fifth of the sample moved from a base case choice of the zero or threshold option, to choosing the uniform option in the modified case. In this same domain of seeds, no policy makers followed P-7. Again, the actual counts become too small to draw any conclusions with confidence, but they are at least suggestive of differences between the policy makers and farmers in the consistency of their sequential decision-making and how they respond to one type of new information (Table 8). Table 8 Decision-making procedure P-7 by respondent category Medicine Farmers (%) (count) Policy makers (%) (count)

12.5 (5) 2.1 (1)

Seeds 20 (8) 0 (0)

Job candidate 15 (6) 8.5 (4)

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1739

Table 9 Percent answering computation questions correctly Decision domain

Socks and shoes

Medicines and vaccination

Chicken and rice

Interest and loan fees

Correct: farmers (%) Correct: policy makers (%)

5 70

0 60

25 90

0 45

3.4. Question series two: computation To assess what has been labelled cognitive effort, respondents were given a series of questions reported in Kahneman (2003), and recently used as part of a cognitive reflection test (Frederick, 2005). Our questions take the form: Medicine and a vaccination cost VND 110,000 in total. The medicine costs 100,000 more than the vaccination. How much does the vaccination cost? The effortless answer is VND 10,000. The correct answer requires recognizing that the cost of the medicine equals 100,000 plus the cost of the vaccination, which requires that the vaccination only cost 5000. In Frederick’s experiments using a bat and a ball and amounts of US$ 1.10 and US$ 1.00, 50% or fewer university students at Princeton and Michigan gave the correct response. In our sample, we asked this question across four domains: for socks and shoes; for medicine and a vaccination; for chicken and rice; and for interest and fees on a loan. The questions differed in their wording and the decision domain, though the algorithm for determining the answer is similar in all cases. The percent of correct responses is given in Table 9. Table 9 shows that for each decision domain a much higher percent of policy makers answered correctly: for example, 70% of policy makers compared to 5% of farmers in the question that referred to socks and shoes. In all cases, policy makers also took a minute or more longer (significant at 10%) to answer the question. Median response times, however, are more similar indicating that these results are driven by a few policy makers who took an extraordinary amount of time to respond. From a large web-based experiment Rubenstein (2006) suggests that instinctive, emotional choices require less response time than cognitively reasoned choices. Oddly enough, neither the mean nor median amount of time decreased for either farmers or policy makers with subsequent questions. A learning curve is not suggested by either the percent of correct responses or the time spent responding. Frederick (2005) found that respondents with lower elicited discount rates, who were more risk seeking over gains, and who were men, scored higher on his cognitive reflection test than did those women who were identified as less patient and more risk averse (over gains, not over losses). His results confirm some earlier studies on the generally positive relationship between patience and intelligence, posited to derive from the greater reasoning and reflection that accompanies intelligence (see Frederick for a review of this empirical work). No similar postulates are advanced for the relationship between cognitive ability and risk preferences, or cognitive ability and gender. We control for discount rates, risk preferences over gains, gender, and unique to our data – the time spent answering and the respondent’s age – in a simple estimation of the likelihood that an individual responds correctly to a computation question. Because the range of incorrect answers is not ordinally meaningful, we recoded responses as a binary outcome and estimated the probability that an individual answers correctly. The sock and shoe question is used because for estimation purposes it has a reasonable balance of correct and incorrect responses. The responses of farmers and policy makers are pooled. We estimated a standard logit model: P = [1/(1 + exp−(␤X + ␧))], where ␤ is a vector of six parameters, and X is an n × k matrix of 86 observations for the k − 1 explanatory variables. Table 10 presents the maximum likelihood estimates, standard errors, p-values, and exponential values of the B coefficients to calculate odds ratios. The small sample size limits confidence that the p-levels can be interpreted as measures of significance for two-tailed tests. Nonetheless, the results do conform to Frederick’s on risk, but not on gender or discounting. Consistent with Frederick’s findings, respondents who are risk seeking over gains are significantly more likely to get the computation question correct; the results are insignificant for gambles over losses. For a respondent who is willing to accept a gamble, the odds of correctly answering the computation question are 2.5 times the odds for a respondent who does not take the gamble. Risk may be picking up some of the effect of the respondent’s job category, which was not included because of concerns of endogenity bias. Age is statistically significant, perhaps because it is strongly correlated with respondent category, though in terms of effect size we note that the odds of getting the right answer rise only marginally for each additional year. With only Table 10 Logit estimate of the probability of computing correctly—socks N = 86

B

S.E.

Sig.

exp(B)

Sex female = 1 Time to answer Coin toss (1, accepts gamble over gains) Age Discount rate, 4M, 3 months Constant

0.189 −0.049 0.910 0.058 −0.006 −3.150

0.582 0.068 0.484 0.026 0.013 1.212

0.745 0.471 0.060 0.024 0.618 0.009

1.208 0.952 2.485 1.060 0.994 0.043

Nagelkerke R2 = 0.16. Chi-square = 10.48, d.f. = 5, sig. = 0.063.

1740

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

nineteen women in our sample, it would be difficult to distil gender effects, if there were any. Response time was not significant. This small sub-sample may also be muting the effect of discount rates, if, as Frederick found, discount rates were more closely associated with the cognitive scores of women whereas risk attitudes were more closely associated with the cognitive scores of men. Based on the field report the questions to infer discount rates were the most problematic, despite using a common format for requesting equivalent monetary amounts over time periods (Thaler, 1981; Benzion et al., 1989; Anderson et al., 2004). For some respondents, questions were too difficult, not relevant, or effective at eliciting interest rates but not discount rates. Nguyen (2005, p. 10) writes: When answering the questions about the amount of money they wanted to get . . . many poor peasants gave answers by their feeling and often based their calculation on the real local situation. For example, “in our place, the interest rate was 10% per day on urgent hot loan, and lower on normal loan” (if they borrow money from those who lend money on high interest), so their calculations of interest were based on urgent loan or normal loan . . . others, including officers, could not calculate the interest and were puzzled when answering because they did not borrow from many sources of credit or had not borrowed. 4. An analysis of decision effort “The study of policy-making is the study of behaviour and its consequences: the behaviour of individuals, groups, and organizations that produce or mediate the social conditions to which policymakers react. It is also the study of the behaviour of policymakers themselves: legislators, elected and appointed executives, bureaucrats, and judges, whose decisions and other actions shape the role of the state in social and economic affairs.” (Lynn, 1986, p. S379) Policy makers convey broad ideas, directions and priorities, and thereby a flow of resources to intermediary groups – government agencies, quasi or non-governmental organizations (NGOs), or members of the private sector – for program design and implementation. These intermediaries convert policy statements into a set of rules that represent the incentives, both constraints and opportunities, faced by recipients. Hence, decisions at each level allocate resources and frame the decisions for the group that follow. Even in our small sample, there are significant differences in decision-making between policy makers and farmers. Their choices result in different tendencies to trade off mean and variance, and different sensitivities to risk. Performance varied significantly on the computation question; the only question in which there was a “correct” answer. Additional survey questions indicated that if given control over funding decisions, farmers and policy makers would allocate funds in quite different proportions, with farmers allocating much more to healthcare and policy makers to the financial sector. Both groups favor vouchers as a form of aid, but farmers were more in favour of unconditional money transfers than policy makers were. Policy makers would allocate more money to subsidized seed, while farmers would allocate relatively more to subsidized medicine and subsidized credit.6 What is behind the differences between farmers and policy makers implied by these results? Decision-making is influenced by a suite of psychological parameters—preferences, ability, and effort. We cannot individually distinguish the impact of each, but we can look for systematic differences between our groups attributable to the cluster. There may be on-the-job factors such as training and exposure to information that affect the decision-making of individuals in these different environments. Circumstances such as income and education may affect risk and time preferences, or causality may run the other way. In our sample, education and respondent category are almost perfectly correlated. All policy makers have education beyond the secondary school level, and most farmers have education at the secondary school level or below. Yet more than 90% of our respondents were born in a rural commune, where education is only available through secondary school. To receive an education above the secondary level in these communes therefore requires migrating, if only temporarily. The decision to migrate may have been determined by parents or other factors; otherwise, it could be driven by either the desire to acquire more education, or by the desire to obtain a high level public sector position which requires more education (recall that for the individuals in our sample, private sector employment was unlikely to have been a relevant consideration in Vietnam at the time). Hence, we begin by assuming that there is a self-selection of individuals who both choose and succeed in policy making and end up in such jobs. Our question is whether or not there are common preference parameters that can predict whether or not one ends up as a policy maker or a farmer, given that most of our sample began in fairly similar circumstances. If there is choice, certain preference parameters may lead one to choose education in order to obtain a public position and/or the associated income. We estimated a logit model of the likelihood that an individual would become a farmer (coded with a zero) or a policy maker (coded with a one) based on a variety of preference parameters that reflect risk attitudes, including the qualitative dimensions of risk that shape risk perceptions, and social preferences. These include risk and time preference, views of fairness, optimism, and questions about the experience, responsibility, and control the individual feels over making three different types of decisions: what help will be given to others less fortunate in the commune;

6

These results are available from the authors.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1741

Table 11 Logit estimate of respondent category

Age Sex (1, female) Discount rate, 4 M, 3 months Discount rate  3 months to 1 year Coin toss (1, accepts gamble over gains) Attitude toward the future Fairness (1, equal fruit) Fairness (1, equal vitamins) Value of good roads and sewage Responsibility for help to others Responsibility for public service priorities Responsibility for productive activities Control over help to others Control over public service priorities Control over productive activities Experience with help to others Experience with public service priorities Experience with productive activities Constant

B

S.E.

Sig.

exp(B)

0.399 −0.714 0.034 0.084 0.496 −2.970 5.380 6.071 0.746 −0.681 −1.473 −3.998 −1.112 3.291 −1.750 −1.873 1.205 5.342 −11.608

0.147 1.599 0.038 0.045 1.534 1.743 2.178 2.380 0.899 1.346 0.895 1.648 1.059 1.372 1.283 1.407 0.688 2.112 6.691

0.007 0.655 0.366 0.059 0.747 0.089 0.013 0.011 0.407 0.613 0.100 0.015 0.294 0.016 0.172 0.183 0.080 0.011 0.083

1.490 0.490 1.035 1.088 1.641 0.051 217.100 433.298 2.108 0.506 0.229 0.018 0.329 26.872 0.174 0.154 3.337 208.872 0.000

Dependent variable: 1, policy maker; 0, farmer. Nagelkerke R2 = 0.851. Chi-square = 85.11, sig. = 0.000, d.f. = 18.

priorities for public services and infrastructure in the commune; and what productive or income earning activities they will do. These questions were designed to range over decisions allocating public resources, and their own private resources, that would be relevant for both groups and not driven by their current position. Results are reported in Table 11. Since senior policy makers tend to be older, it is not surprising that age is statistically significant. There is a modest size effect, i.e., for a one unit change in age (equal to 1 year) the odds are one and one half times greater that the respondent is a policy maker. With a larger sample one could bracket the farmer sample to compare preferences across comparable age ranges for the two groups, but as it stands, we expect these results may be reflecting the smaller, and older, age range for the sample of policy makers. Controlling for gender (not significant) and age, several preference parameters emerge as significant predictors of respondent category. Differences in elicited discount rates for delaying the receipt of four million VND for 3 months are not significant, but changes between short run (3 months) and long run (1 year) discount rates are—bearing in mind the problems experienced with this battery of questions. Having a discount rate that rises with the proximity of an event (that is, a short run discount rate that exceeds your long run discount rate) has a slightly positive effect on the odds of being a policy maker. Risk preference, as measured by willingness to accept a better than fair gamble over gains, is not significant. Though the farmers in our sample were generally more risk averse than the policy makers, they were significantly more optimistic about the future. Recognizing that negative log odds are fractional effects, we can interpret the odds ratio as indicating that for each rank increase on our optimism scale, the odds of being a farmer relative to a policy maker increases by a factor of approximately 20. These results confirm observations from our field manager who was struck by the level of optimism, noting: When they were asked about attitude to the future, most of the poor peasants, to some extent, were optimistic or very optimistic. It seemed that they believed in the possibility of overcoming the poverty situation of their family. Very few people were pessimistic. (Nguyen, 2005, p. 13) A question about the fair allocation of fruit to isolated villagers was intended to reflect social preferences. Answers to this question are both statistically significant and have a very substantial effect in terms of size among the independent variables on the odds of being a policy maker versus a farmer. This question, adapted from Yaari and Bar-Hillel (1984, p. 8) asks7 : An aid shipment containing 12 bananas and 12 papayas is to be distributed between 2 people in a remote village: Nguyen and Viet. The following information is known to all: a. Nguyen derives 100 units of vitamins from each banana eaten, and 0 from papaya. b. Viet derives 50 units of vitamins from each banana and 50 from each papaya eaten. c. Both Nguyen and Viet care only about the vitamins in the fruit, nothing else.

7

We took the question from Rabin (1998, p. 18).

1742

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

d. Nguyen and Viet live too far apart to trade. How should the fruit be divided between Nguyen and Viet if the division is to be fair? Choose one: i. Nguyen: 6 bananas, 6 papayas; Viet: 6 bananas, 6 papayas. ii. Nguyen: 12 bananas, 0 papayas; Viet: 0 bananas, 12 papayas. iii. Nguyen: 8 bananas, 0 papayas; Viet: 4 bananas, 12 papayas. The base case in the regression is the second option, arguably the “efficient” solution that maximizes total welfare by maximizing the total joint consumption of vitamins. The first option equalizes the distribution of fruit, and the third option equalizes the distribution of vitamins. Choosing to equalize fruit compared to the second option of maximizing vitamin use increased the ratio of the odds of being a policy maker by a factor of over 200, all else equal. Choosing to equalize vitamins compared to maximizing vitamin use increased the odds of being a policy maker by double that. Thus, as predictors, measures of fairness have the most significant ability to distinguish farmers from policy makers. If we calculate these odds ratios from simple crosstabs (or from a logit regression where they are the only predictors), the size of the coefficients are reduced by about half relative to their size in the full regression. Hence, controlling for the other variables increases the magnitude of the effect of the respondent’s view on fair allocations on the likelihood of them being a farmer or a policy maker. We can only posit what may be driving these results. But in contrast to the view that policy makers choose for the collective to maximize welfare, farmers were much more likely to choose the “efficient” option and give Nguyen all the bananas and Viet all the papayas. It is possible that this is the crude approximation to option iii for utilizing different vitamin intake. It is also possible that policy makers – just as they preferred uniformity as a decision procedure – prefer equality as an outcome. Though we asked individuals not to respond in their professional capacity, we cannot know if this represents true preferences or a political approach to allocating resources. The third option is also a maximin solution – it has the least poor outcome for Viet – and it provides almost as many total vitamins as option ii (Rawls, 1971).8 Hence, option iii could represent a social justice perspective that satisfies a fairness criterion and exploits some efficiencies. What the data suggest, however, is that for this scenario, policy makers, on average, are substantially more willing to trade off vitamin efficiency for equity than are farmers. Answers to questions about the experience, responsibility and control the respondent has over decisions were scored on a scale of 1, none; 2, a little; 3, medium; 4, a lot; 5, total. Individuals who felt a significantly greater sense of experience and control over decisions on priorities for public services and experience making decisions about their own productive activities had a higher probability of being a policy maker. Experience over one’s own productive activities had a large effect: with every one unit increase on the scale the odds of being a policy maker over a farmer increased by a factor of more than 200. Although this result represents a large impact in the quantitative analysis we assign it less importance in a practical sense, as policy makers are in fact categorized as a result of their job content and experience. Policy makers are more likely to have made a deliberate choice about their livelihood, whereas farming is more likely the default for individuals in this predominantly rural born sample. Surprisingly, however, respondents predicted to be policy makers felt significantly less responsibility over these same two decisions than did farmers, and less control, responsibility and experience over decisions about helping the less fortunate, though not significantly so. Looking at these same results by type of decision is also interesting. Responsibility over one’s own productive activities had a large effect in the estimation, and a relatively larger positive effect on the odds of being a farmer over a policy maker than did responsibility for helping others or deciding on public service prioirites. Experience with decision-making over one’s own productive activities also had the greatest relative impact compared to helping others and public service prioirites. Furthermore, the likelihood of being a policy maker was positively associated with decisions on one’s own productive activities and public service priorities, but negatively associated with helping others (though not significantly). These preference parameters, controlling for age and gender, are reasonably robust with respect to the significance of fairness, time preference, and optimism. The model as a whole is statistically significant, with a p-value less than 0.000, and it explains a large proportion of the variance, even with the Nagelkerke R-squared as an approximation to an OLS R-squared. These results are consistent with other findings that categorize preference differences according to risk attitudes and social preferences (Croson and Gneezy, 2004). And arguably, these preference parameters are of critical importance in constructing public policy and allocating resources. 5. Conclusions We began by asking two questions: first, are patterns of behavior different from those predicted by strict rationality demonstrated in populations outside the laboratory experiments of the U.S. and Europe, and second, are there systematic differences in decision-making between policy makers and program recipients of poverty alleviation programs? We can offer only initial answers to both, but we believe these answers warrant further research.

8

We thank an anonymous reviewer for pointing this out.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1743

This work in Vietnam suggests that the decisions of individuals in a country with a significantly different historical and institutional environment than the U.S. also diverge from the predictions of strict rationality. And there is growing evidence that the behaviours in Vietnam range beyond those examined here (Tanaka et al., 2006), though the nature of the divergences across countries remains an open question. For example, Nisbett (2003) argues that perceptual and cognitive processes are not generalizable across the globe, but rather are quite distinct between Asians and Westerners. More importantly, perhaps, than confirming the regularity of behaviours outside of the U.S. and Europe, is determining whether our traditional assumptions are even less applicable to certain populations such as the poor, the food insecure, and the rurally isolated (Kanbur, 2003; Mullainathan, 2005; Duflo, 2006; Bertrand et al., 2006; Anderson and Stamoulis, 2007). There is evidence, for example, that income, agricultural livelihoods, and levels of market development are related to discount rates and initial offers in ultimatum games (Henrich et al., 2005; Anderson and Gugerty, 2008). If subsistence and isolation do affect decision procedures in ways not predicted by our models, then particularly for developing countries, there are fewer well functioning competitive markets and there is less learning to “average out” anomalous behaviors (Camerer, 1987; List, 2003, 2004). As (Rubinstein, 1998, p. 22) points out, to label certain behaviors as “mistakes” does not make them uninteresting. “If there are many traders in a market who calculate 1 + 1 = 3, then their “mistake” may be economically relevant.” Informal exchange may heighten resource allocation effects, especially when combined with greater price and output variability common to agricultural economies.9 Our second question concerned systematic differences in preference parameters along the policy chain. In many cases the differences emerging between farmers and policy makers are striking: different tendencies to make mean–variance tradeoffs, different sensitivities to risk, and different computational outcomes. There are also differences in what is viewed as fair, and the degree of control, responsibility, and experience that respondents feel over decision-making. Among these, and given our current information, perspectives on fairness seem to carry the most meaningful impacts in a practical sense. It is premature to know whether and how these results derive from differences in cognitive processes, and if so, what is behind this—exposure, training, innate preferences or capacities. But it is not, we believe, premature to recognize that whether these differences are innate and reflect a self-selection into groups, or whether they are socialized, if systematic differences are revealed in these behavioural experiments they may contribute to the seeming disconnect between the predictions and program designs of policy makers and the response of program recipients. Acknowledgment The authors would like to thank the Food and Agriculture Organization of the United Nations for helping to support this research. Appendix A. Other descriptive statistics: whole sample N

Minimum

Maximum

Mean

Standard deviation

Age in 2005 Marital status: 1, married (87.4%); 2, divorced (3.4%); 3, single (6.9%); 4, widowed (2.3%) Approximately household income (monthly) US$ 1 = 15888VND Approximately household expenses (monthly)

86 87

23 1

67 4

43.02

9.90

87 87

170000VND (US$ 10.70) 0VND US$ 0

Approximately household debt

85

0VND US$ 0

Approximately household savings

86

0VND US$ 0

15000000VND (US$ 9441.09) 7000000VND (US$ 440.58) 60000000VND (US$ 3776.44) 71000000000VND (US$ 4468781.47)

2661637.54VND (US$ 167.52) 1814607.00VND (US$ 114.21) 3715294.22VND (US$ 233.84) 891267442.23VND (US$ 56096.90)

2822697.71 (US$ 177.66) 1635661.11 (US$ 102.95) 10430067.60 (US$ 656.47) 7651563194.62 (US$ 481593.85)

Level of decision-making: 0 (41.4%); 1, senior (14.9%); 2, mid-level (29.9%); 3, junior (9.2%) Internet access: 0, yes in home (41.4%); 1, yes outside home (19.5%); 2, no (39.1%)

83

0

87

0

9

2

Fehr and Tyran (2005) have begun to examine under what conditions individual irrationality is decisive for aggregate outcomes.

1744

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

Appendix B. Policy maker job areas Office

Position

Deposit Security Fund Institution of Bank Vietnam Bank for Agriculture and Rural Development, South Hanoi Branch Vietnam Bank for Social Policies State Bank, Ha Tay Province State Budget Department, Ministry of Finance Budget section, State Budget Department, Ministry of Finance Forecast section, State Budget Department, Ministry of Finance Profession section, State Budget Department, Ministry of Finance Informatics and Financial Statistics Department, Ministry of Finance Administrative section, State Budget Department, Ministry of Finance Administrative and synthetic section, Informatics and Financial Statistics Department, Ministry of Finance Statistics section, State Budget Department, Ministry of Finance Locals section, State Budget Department, Ministry of Finance Department of Cooperative and Rural Development, Ministry of Agriculture and Rural Development (MOARD) Central Managing Committee for Rural Projects (MOARD) Department of Science and Technology (MOARD) Department of Planning (MOARD) Department of Legislation (MOARD) Institute of Planning and Agricultural Design, Ministry of Agriculture and Rural Development (MOARD) Counseling Company for Rural Development Rural Development Office, Department of Cooperative and Rural Development (MOARD) Institution of Agricultural Economy AIDS and Community Magazine Finance and Accounting Office, Bach Mai Hospital Department of Planning and Finance, Ministry of Health (MOH) Exchange Department, Vietnam Bank for Agriculture and Rural Development Department of Traditional Medicine, Ministry of Health (MOH) Department of Reproductive Health (MOH) Sustained Development Program (MOPI) Department of Science, Education, Natural Resources, and Environment (MOPI) Synthetics Department (MOPI) Tay Nguyen Managing Committee of Health Projects (MOH) Labour, Invalids, and Social Affairs Cadre of Cao Ky Community, Cho Moi District, Bac Can Province Vietnam Bank for Agriculture and Rural Development, Cho Moi Branch, Bac Can Province Office of Agriculture and Rural Development, Cho Moi District, Bac Can Province Health Center, Cho Moi District, Bac Can Province Vietnam Bank for Social Policies, Cho Moi Branch, Bac Can Province General Administrative Department (MOH)

Chief Executive Officer Deputy Director, Deputy Director, Former Director Director, General Director General Director Deputy Director Deputy Director Deputy Head Deputy Head Deputy Head Deputy Head, Deputy Director Deputy Head Deputy Head Deputy Head Deputy Head Deputy Director, Specialist, Deputy Director Head Director Deputy Director Director Director Director Senior Clerk Deputy Director Assistant Editor Senior Clerk Expert Senior Clerk Deputy Director Specialist Chief of the Secretariat Deputy Director, Specialist Director of Department President Deputy Head, Cadre Director Senior Clerk Director Director Deputy Director

References Anderson, C.L., Gugerty, M.K., 2008. Intertemporal choice and development policy: cross-country evidence on time varying discount rates. University of Washington Evans School Working Paper #2006-11. Anderson, C.L., Stamoulis, K., 2007. Applying behavioral economics to international development policy. In: Mavrotas, G., Shorrocks, A. (Eds.), Advancing Development: Core Themes in Global Economics. Palgrave Macmillan, Hampshire, UK, pp. 664–685. Anderson, C.L., Dietz, M., Gordon, A., Klawitter, M., 2004. Discount rates in Vietnam. Economic Development and Cultural Change 52 (4), 873–887. Benzion, U., Rapoport, A., Yagil, J., 1989. Discount rates inferred from decisions: an experimental study. Management Science 35 (March), 270–284. Bertrand, M., Mullainathan, S., Shafir, E., 2006. Behavioral economics and marketing in aid of decision making among the poor. Journal of Public Policy and Marketing 25 (Spring (1)), 8–23. Binswanger, H., 1980. Attitudes toward risk: experimental measurement in rural India. American Agricultural Economics Association, 395–406. Camerer, C.F., 1987. Do biases in probability judgement matter in markets? Experimental evidence. American Economic Review 77 (5), 981–997. Camerer, C., Hogarth, R., 1999. The effects of financial incentives in economics experiments: a review and capital-labor-production framework. Journal of Risk and Uncertainty 19, 7–42. Camerer, C.F., Loewenstein, G., Rabin, M., 2003. Advances in Behavioral Economics. Princeton University Press, New Jersey. Carson, R.T., Flores, N.E., Meade, N.F., 2000. Contingent valuation: controversies and evidence. Working Paper. Chattopadhyay, R., Duflo, E., 2004. Women as policy makers: evidence from a randomized policy experiment in India. Econometrica 72 (September (5)), 1409–1443. Croson, R., Gneezy, U., 2004. Gender differences in preferences. Working Paper. Davies, E., Lea, Stephen E.G., 1995. Student attitudes to student debt. Journal of Economic Psychology 16, 663–679. Duflo, E., 2006. Poor but rational. In: Banerjee, A., Benabou, R., Mookherjee, D. (Eds.), Understanding Poverty. Oxford University Press, Oxford.

C.L. Anderson et al. / The Journal of Socio-Economics 37 (2008) 1729–1745

1745

Fehr, E., Tyran, J.-R., 2005. Individual irrationality and aggregate outcomes. Journal of Economic Perspectives 19 (4), 43–66. Finucane, M.L., Slovic, P., Mertz, C.K., Flynn, J., Satterfield, T.A., 2000. Gender, race, and perceived risk: the “white male” effect. Health Risk and Society 2, 159–172. Fischhoff, B.S.P., Lichtenstein, S., Read, S., Combs, B., 1978. How safe is safe enough? A psychometric study of attitudes towards technological risks and benefits. Policy Sciences 9, 127–152. Fisher, I., 1930. The Theory of Interest. Macmillan, New York. Flynn, J., Slovic, P., Mertz, C.K., 1994. Gender, race, and perception of environmental health risks. Risk Analysis 14 (6), 1101–1108. Frederick, S., 2005. Cognitive reflection and decision making. Journal of Economic Perspectives 19 (4), 25–42. Gigerenzer, G., Selten, R. (Eds.), 2001. Bounded Rationality—The Adaptive Toolbox. The MIT Press, Cambridge, MA. Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., McElreath, R., Alvard, M., Barr, A., Ensminger, J., Henrich, N.S., Hill, K., Gil-White, F., Gurven, M., Marlowe, F.W., Patton, J.Q., Tracer, D., 2005. Economic man in cross-cultural perspective: behavioural experiments in 15 small-scale societies. Behavioural and Brain Sciences. Kahneman, D., 2003. Maps of bounded rationality: psychology for behavioral economics. American Economic Review 93 (5), 1449–1475. Kanbur, R., 2003. Behavioral development economics. Notes from the Conference on Economics for an Imperfect World. Lawrance, E.C., 1991. Poverty and the rate of time preference: evidence from panel data. Journal of Political Economy 99 (1), 54–77. Levitt, S.D., 1996. How do senators vote? Disentangling the role of voter preferences, party affiliation, and senator ideology. The American Economic Review 86 (3), 425–441. List, J., 2003. Does market experience eliminate market anomalies? Quarterly Journal of Economics 118 (1), 41–71. List, J., 2004. Testing neoclassical competitive theory in multi-lateral decentralized markets. Journal of Political Economy 112, 1131–1156. Lynn Jr., L.E., 1986. The behavioral foundations of public policy-making. The Journal of Business 59 (4), 379–384. McDaniels, T.L., Axelrod, L.J., Cavanagh, N.S., Slovic, P., 1997. Perception of ecological risk to water environments. Risk Analysis 17 (3), 341–352. Mullainathan, S., 2005. Development economics through the lens of psychology. In: Bourguigon, F., Pleskovic, B. (Eds.), Annual Bank Conference in Development Economics. World Bank, Washington, DC. Nguyen, X.M., 2005. Report on the Implementation of Vietnam Economic Behavioural Survey, unpublished paper. Nisbett, R., 2003. The Geography of Thought. The Free Press, Simon and Schuster, Inc., New York, NY. Rabin, M., 1998. Psychology and economics. Journal of Economic Literature 36 (1), 11–46. Rawls, J., 1971. A Theory of Justice. Oxford University Press, Oxford. Rubinstein, A., 2006. Instinctive and cognitive reasoning: a study of response times. Fondazione Eni Enrico Mattei, Nota di lavoro, 36.2006. Rubinstein, A., 1998. Modeling Bounded Rationality. MIT Press, Cambridge. Sagoff, M., 1988. The Economy of the Earth: Philosophy, Law, and the Environment. Cambridge University Press, New York. Saigal, S., Stoskopf, B.L., Feeny, D., Furlong, W., Burrows, E., Rosenbaum, P.L., Hoult, L., 1999. Differences in preferences for neonatal outcomes among health care professionals, parents, and adolescents. Journal of the American Medical Association 281 (21), 1991–1997. Shumway, M., Saunders, T., Shern, D., Pines, E., Downs, A., Burbine, T., Beller, J., 2003. Preferences for schizophrenia treatment outcomes among public policy makers, consumers, families, and providers. Psychiatric Services 54 (8), 1124–1128. Simon, H.A., 1955. A behavioral model of rational choice. Quarterly Journal of Economics 69, 99–118. Simon, H.A., 1979. Models of Thought. Yale University Press, New Haven and London. Slovic, P., Flynn, J., Layman, M., 1991. Perceived risk, trust, and the politics of nuclear waste. Science 254, 1603–1607. Tanaka, T., Camerer, C.F., Nguyen, Q., 2006. Poverty, politics and preferences: field experiments and survey data from Vietnam. Working Paper. Thaler, R., 1981. Some empirical evidence on dynamic inconsistency. Economic Letters 8, 201–207. Thaler, R.H., 1991. Quasi Rational Economics. Russell Sage Foundation, New York. Tversky, A., Kahneman, D., 1981. The framing of decision and the psychology of choice. Science 211, 453–458. Tversky, A., Wakker, P.P., 1995. Risk attitudes and decision weights. Econometrica 63, 1255–1280. Viney, R., Savage, E., Louviere, J., 2005. Empirical investigation of experimental design properties of discrete choice experiments in health care. Health Economics 14, 349–362. Yaari, M.E., Bar-Hillel, M., 1984. On dividing justly. Social Choice Welfare 1 (1), 1–24.