Subjective Wealth and Rural Development: the Cotton Reform in Burkina Faso as a case in point* Jonathan Kaminski†, Dept of Agricultural Economics, the Hebrew University of Jerusalem P.O. Box 12, Rehovot 76100, Israel

Abstract. The cotton economy of Burkina Faso has been characterized by a changing rural environment for farmers since late nineties, which has come with the cotton reform. There have been slight improvements in living standards and rural households’ income while the subjective wealth indicator has significantly increased. In this paper, we explore the channels through which the elements of the changing rural environment can bridge the wedge between subjective and objective measures of wealth, in addition to the determinants found in the happiness economics literature, namely absolute and relative income measures, health and social status (and expectations of future incomes), and land holdings. We investigate the empirical validity of institutional and technological change as well as the perceptions about the reform. An empirical framework is proposed to deal with seemingly covariant political opinions and evolution of subjective wealth. This is justified by the fact that cotton farmers’ groups have been involved into the political change, which is treated as an advertising effect on subjective wealth. We find that the significantly positive evolution of subjective wealth has been driven by absolute and relative measures of wealth, technological improvements, and by enthusiastic perceptions mostly about the reform’s effects on poverty alleviation (information effect). This evolution has been altered by the beliefs about a larger input access and better agricultural abilities resulting from the reform (comparison effect). JEL Codes: I32, 013, Q16, Q18 Keywords: subjective wealth, Burkina Faso, cotton, rural development, political opinions.

*

I am grateful to the participants of the CSAE Annual Conference on “Economic Development in Africa” held in Oxford in March 2008 and the discussant of this paper, Precious Zikhali. I am indebted to Pierre Dubois and Stéphane Straub for their useful advices and comments. I am grateful to Ayal Kimhi for thoughtful discussions and to the participants of the Hebrew University department Seminar held in Rehovot in November 2008. This paper has benefited from the comments of the participants of the ISSCRI Conference on “Integrating Social Science Research into Cotton Reform Implementation” held in Montpellier in May 2008. I warmly thank ARQADE and Jean-Paul Azam for financial support and advices. I am also grateful to Kimséyinga Savadogo for having welcomed me in Burkina Faso in spring 2006 and having helped me to conduct my survey in cotton areas. † Author’s contact: [email protected]

Subjective Wealth and Rural Development: the Cotton Reform in Burkina Faso as a Case in Point The study of subjective well-being has recently received increasing attention by many researchers. A flourishing literature has been revisiting the issue of income-utility relationship1 (Easterlin, 1974; or Van de Stadt et al., 1985 for instances) while the use of subjective welfare data is more and more accepted as reliable (Krueger and Schkade, 2007) and covering many aspects of welfare2 which are not measurable by surveys. This could be fruitfully used to explore the effects of economic shocks and policy experiences on several aspects of wealth, and not only on absolute income or consumption (Van Landeghem et al., 2008). In the context of development, few micro-economic empirical evidence exist though many country experiences could be used as natural experiments of relevance to contribute to this literature. The so-called “Easterlin Paradox” –stating a cross-sectional positive incomehappiness relationship but a stationary effect over time- should be reexamined in the process of development. Indeed, transient effects can occur during developmental steps when sudden economic changes affect the everyday life of households. The latter may attach subjective values to new economic and social opportunities, as well as to their new institutional environment and technical skills. Furthermore, the issue of development is also a political concern where players differ in their involvement and related satisfaction. Collective action and group mechanisms might provide individuals with divergent opinions about a political change while the latter may impact their own wealth differently. However, political opinions and the pattern of subjective wealth should be linked, because political involvement and empowerment may be a source of additional perceived welfare, and absolute and relative welfare effects matter for political opinions (Bonnet et al., 2006). Common beliefs do also play a role both in the self-assessment of political change and on the perception of own wealth. Finally, the measures of self-assessed wealth and political opinions matter for policymakers when looking for popular adhesion and support. In this paper, we aim to bring new micro-economic evidence about the role of political involvement of rural households on their own subjective wealth, when the former can be captured by opinions on a policy change. In addition to the determinants found in the happiness economics literature, do opinions play a role on the pattern of subjective wealth?

1

We also want to explore if there are additional channels coming from the changing rural environment of households that do play on the pattern of subjective wealth. What happens when the environment of farmers changes rapidly? Are their perceptions modified only by living improvements or also influenced by their participation in the process of change and related outcomes? Last, we want to investigate the causality between the formation of political opinions and the evolution of perceived wealth. To address these points, we need to build a conceptual framework where opinions about a policy and the pattern of subjective wealth are jointly determined as covariant processes, with possible endogeneity of each other. The case of rural development is particularly relevant here, because rural communities are involved, which enables us to further explore the role of group mechanisms and collective action in the process of development and political change. The cotton reform experience of Burkina Faso fittingly lends itself as a case in point. First, cotton sector is the main driving force of agricultural growth and has been one of the major poverty-reduction strategies in the region (Goreux, 2003). Second, the reform is commonly acknowledged to have been one of the few successful across Sub-Saharan Africa, with a unique participation of cotton farmers to agricultural policymaking and empowerment through the establishment of more professional organizations and their influential national union. In Burkina Faso, the reform has led to a pattern of impressive mid-term cotton growth, based on the growth of cotton areas. Kaminski and Thomas (2008) have identified the main channels: influx of labor and capital in land cultivation, new technologies, better designed local institutions and credit access, and better institutional arrangements between farmers and cotton firms. However, national living standards surveys do not report a significant increase in living standards and only slight changes in income on average. Indeed, the price paid to cotton growers has not increased because of the world cotton market environment and more expensive inputs have hampered farmers’ profitability margins. The rise of agricultural income has only concerned farmers having experienced a large increase of cultivated land or those who entered cotton production during the reform and experienced a rapid extension of both cotton and non-cotton cultivated land. The political crisis in Côte d’Ivoire since 2002 has also adversely affected Burkinabé rural households who formerly received remittances from their relatives. Finally, the cotton reform has yielded a more equal distribution of income in rural cotton zones with no significant living standards improvement3. In contrast, the data that we collected in 2006 in representative cotton areas unambiguously show that the perception of wealth has much improved over the reform 2

period. Information about the self-assessment of the effects of cotton reform displays positive values for several opinion indicators. These data contain valuable information about land holdings, land use patterns, technical choices, technical and institutional environment of farmers, as well as evolution of basic living standards, food and non-food consumption, and farm and non-farm sources of income. We use subjective wealth as our variable of subjective well-being, that is, the perceived rank of each rural household on a wealth ladder, which is more directly related to a utilitarian perspective than happiness indicators (as in Ravallion and Lokshin, 2001). Because income and wealth of a reference group matter, these data are valuable since they reflect absolute and relative land holdings, as well as for cotton cultivation, which can bring additional subjective wealth because of their positional role in a poor rural society4. This does not enable us only to determine the appropriate indicator of the reference group effect on wealth (as in Van Landeghem et al., 2008), but also on political opinions and on the value of groups’ political empowerment. We are also able to check if there is significant consumption in positional goods. Last, estimating subjective wealth helps assess the aggregate and distributional welfare effects of the cotton reform, that is, the ones from land cultivation, land use, input access, farm incomes, and so on. Our results highlight that political opinions have been significant for farmers in the pattern of increasing subjective wealth as exogenous advertising effects. In addition to the determinants found in the literature, we show that mechanization (adoption of animal farming) has increased perceived welfare in addition to welfare and social effects. Comparison effects have applied to the pattern of social events and through the opinions about perceived effects from the reform. While a positive advertising effect is associated to poverty alleviation, this has been partly offset by comparison for the access to inputs and farm skills. The two latter are more significant for non-poor households from resident ethnic groups, which can be related to the egalitarian distributional effects of the reform and the new economic opportunities (cotton and agricultural inputs) for village discriminated minorities. The remainder of this paper is as follows. Section 2 briefly introduces the features of the evolving cotton economy of Burkina Faso and describes the evolution of main economic variables of interest during the cotton reform. We also describe the data and subjective indicators of individual wealth and of the appreciation of the cotton reform. Drawn on the literature, section 3 presents a structural model for the evolution of subjective wealth in the context of rural development when players are politically involved. This model is built on a conceptual framework accounting for group mechanisms and collective action in agricultural

3

policymaking. Section 4 discusses econometric estimations and results, and section 5 concludes.

The evolving cotton economy of Burkina Faso Cotton has been one of the leading factors of poverty alleviation throughout the African continent over the twentieth century, and based on a peasant cotton revolution in West Africa (Bassett, 2001). The cultivation of Gossypium has been associated to more food security and more cash income into rural zones. The latter has allowed households to access better health and education commodities while the positive effect on food security has been a consequence of agronomic complementary effects from cotton to other food crops5. In addition, one has not to forget that cotton cropping has brought many agricultural inputs to farmers6, responsible for higher yields, notably in cereal production. As a consequence, the cotton cropping has prevented from rural exodus, in some extent. These features have to be emphasized for Sahelian countries (Mali, Burkina Faso, or Chad for instance) where no alternative cash crop, such as the cotton one, looks relevant in participating to rural poverty reduction and development. The development of cotton economies in Sub-Saharan Africa resulted in more democratization, and education (see Bingen, 1998; for the Malian case), as well as better living standards than in subsistence economies with an active participation (even leading) to the national growth dynamics (see Azam and Djimtoingar, 2004; for Chad). The original reform of Burkina Faso and its implications for growth and employment The cotton reform movement was prompted by several internal and external reasons. The former centralized systems exhibited worsening outcomes, including low producers' incentives (implicit taxation), low managerial performances (high default rate on input credit and inefficient parastatal management), which ended up with macro-economic instability, namely high rates of public debt and inflation. Aid conditionality was also tied to sectoral reforms and to the elimination of former parastatals and official boards in the region, as part as structural adjustment plans within the Washington Consensus. While some elements of these reforms were the same across West and Central Frenchspeaking Africa -especially dismantling parastatal companies, allowing competition to raise price incentives for producers, and improving overall management and the financial situation of the sector- the approach of Burkina Faso has been very original in several regards. The distinctive features of reform were sequencing and gradualism, and a focus on institutional reform with a specific emphasis on increasing the participation of farmers through various

4

new institutional arrangements. This has enabled producers to become professional partners, taking a growing number of responsibilities in managing the industry and influencing government policies. Thereafter, the focus was on strengthening the institutional framework to make it compatible with the ongoing market reforms and improving market coordination in the delivery of crucial public goods. Specifically, the chronology of the reform was as follows: •

1996-1999: introduction of free-adhesion based mechanisms for local groups of cotton farmers, replacing former village groups by market-oriented organizations with the implementation of new local governance rules.

•

1996-2001: progressive establishment of the national cotton union, the Union nationale des producteurs de coton du burkina faso (UNPCB), with the support of the french aid, the government, and SOFITEX (the national cotton parastatal company), based on the membership of local groups and their integration into regional unions.

•

1999: a partial withdrawal of the state, with the partial privatization of SOFITEX. The government gave back half of its shares to the UNPCB

•

2000-2006: progressive delegation of economic activities from SOFITEX and government to UNPCB: provision of cereal input credit, management assistance of cotton groups and participation in quality grading, financial management and price bargaining. The state downsized its involvement in public good investment (research and extension services) and SOFITEX launched new professional technical agents.

•

2002-2006 : progressive introduction of new players : private input providers, new regional private cotton monopsonies (SOCOMA, and FASOCOTON)

•

2004: establishment of an inter-professional agreement with cooperation among wellrepresented stakeholders.

•

2006: change in the price-setting mechanism with more stickiness to the world price levels and the creation of a new smoothing fund

More details and discussions about the reform can be found in Kaminski and Thomas (2008). It is noteworthy to consider the following stylized facts. Burkina Faso has become the current African leader (in 2006 and 2007) in cotton production and exports of lint cotton with the entry of many new producers (with some migrants from Côte d'Ivoire), and production has been multiplied threefold over this period. The cotton growth has been based on an extensive process led by new incentives for production arising from better contractual relationships between/within cotton groups and cotton firms (Kaminski and Thomas, 2008). In this paper,

5

we show that the direct effects from the reform have involved earlier payments of raw cotton to farmers, easier access to inputs and guarantee of selling. This has been accentuated by the positive effect of land extension (and lower need for own-produced food) driven by mechanization (animal farming), better technical assistance, and a larger rural labor force. Before the reform, cotton production accounted for 3.3% of national total agricultural production in constant value while it has been over 8% in 2006 (FAO, 2007). For the other agricultural products, annual average growth rates in constant value have been around 2%. So, cotton production has played an increasing role for agricultural growth, and accounts now for more than 10% of total GDP growth. As for employment, the cotton boom has roughly absorbed 150,000 new farmers (the number of cotton farmers has doubled in these ten years) among which some were already cropping land and others were migrants. This absorption of new labor was remarkable not only because previous influxes of return migrants have been associated with unemployment and economic turmoil (e.g. Ghana in the early 1980s), but also because the cotton reform allowed returning migrants from Côte d'Ivoire to quickly access inputs and form their own farmer groups which have been integrated into the UNPCB and regional unions. Dynamics of poverty, income, and living standards in cotton areas The data set that we use in this paper is the result of a survey of households belonging to GPCs (Groupements de Producteurs de Coton) across 20 villages. This survey was done in March 2006 in representative zones of cotton production with 300 interviewed households, accounting for 0.2 % of national production. An original questionnaire was designed with recall variables and variables about the evolution of agricultural systems and economic decisions within each household. These variables were added to basic variables informing living standards and economic activities -housing, education, health, consumption, credit, savings, crops, cattle-, perceptions of poverty and opinions about the reform. Detailed information on available data is presented in the Table 1 of the appendix. Table 2 and table 3 display summary statistics about the evolution of living standards and for main consumption items. More information on the survey design can be found in Kaminski and Thomas (2008). According to national censuses and permanent surveys (INSD, 2003 and 2006), the evolution of rural incomes have been positive, but admittedly counter-balanced by the negative effect of the political crisis in Côte d'Ivoire, decreasing cotton world prices, and increasing prices of inputs. With a poverty line set at 100,000 CFAF in 2006, 47 % of our sample is below, which corresponds to the national average. Data from INSD (2006) show

6

that poverty indexes have remained constants both at national and at regional levels. The poverty dynamics is subject to several controversies, according to the analytical approach, whether being utilitarian or based on capabilities (Lachaud, 2005). Contrary to the positive trend of poverty reduction found by World Bank (2005), the author claims that both cardinal and ordinal measures (based on Sen, 1976) of poverty have either worsened, or remained stationary from 1994 to 2005. There is neither first-order nor second-order stochastic dominance of poverty distribution during this period. Moreover, monetary and non-monetary measures exhibit similar results. Based on our ordinal measures of the evolution of per-capita consumption and savings (table 3), we conclude to slight increases on average though a significant proportion of the rural population have suffered from decreasing consumption patterns. Largest increases apply to health, energy or clothing while global increase is important for energy, clothing, social events, cereals, animal proteins and condiments. Smallest increases apply to dairy products, alcohol and tobacco, tubers, fruits and education. Diversification of food consumption has not been achieved for many households while savings and investment have been following a positive pattern. This might be a long-term risk strategy for households7. Differences between increases and decreases give us an idea about substitution effects: there are large for clothing, energy, social events, health and cereals and low (likely negative) for dairy products, tubers, fruits and education. Table 2 also shows a slight improvement in living standards, notably for literacy rates, health indicators, and access to water. However, schooling and health constraints8 have remained high for interviewed households. The availability of cash income is likely to have enabled some households to access medicines, to pay for the schooling of their children and to improve their habitat. But the withdrawal of the government from the cotton sector is likely to have had a bad impact on rural infrastructures. One most significant feature of table 2 lies in the moderate shift of health consumption from traditional to conventional fashions9 with a decrease in infantile mortality and in the number of diseases and injuries. Persistence of poverty levels, even in the cotton areas, highlights that positive changes have mostly concerned households above the poverty line. Distributive graphs of income and land (figures 1 and 2) suggest that a small fraction of the rural population could have concentrated most of the benefits generated by the new environment of farmers. Yet, profit-sharing of cotton-related activities has evolved in favor of cotton producers over this period, seeing them reap larger margins. Importantly, cotton reforms were probably highly effective when compared to a counterfactual situation in which the Ivorian crisis and 7

the fall in cotton prices accompanied stagnant production levels. Since this counterfactual scenario would surely have resulted in a large drop in income, the fact that rural incomes have slightly increased and living standards have improved despite these adverse shocks implies that the reforms were highly successful. This is witnessed by our survey in cotton rural areas. Another feature of the reforms is that they are deep institutional reforms that leave the Burkinabé cotton sector well placed to experience future economic shocks (ceteris paribus). Evolution of subjective wealth and income, and perceptions of the reforms’ effects A famous point in the economics of subjective well-being and happiness stands in the “Easterlin paradox” (Easterlin, 1974; Easterlin, 1995). The Easterlin statement involves that average life satisfaction indicators have remained constant in developed countries (Diener et al. 1999) while they experienced high rates of GDP over large periods. For developing countries, in contrast, there is a clear positive trend of the income/happiness relationship. At the individual level empirical papers using cross-sectional data report a significant positive and concave correlation between income and happiness from one country. This holds both for developed (Blanchflower and Oswald, 2004) and developing (Graham and Pettinato, 2002) countries. Our data show that, in spite of slight changes in income and living standards, and under stationary poverty, indicators of subjective wealth have unambiguously increased (figure 3) over the reform period. The pattern is a first-order stochastic dominance of the current distribution of subjective wealth with respect to the one before the reform. Is this a reverse of the “Easterlin Paradox”? The relationship between current income and subjective wealth is, here also, positive and concave across households, albeit of low significance (figure 4 in the appendix). [Figure 3 here] To propose another explanation for the increase in perceived wealth, we look at the correlation between changes in familial land holdings10 and changes in subjective wealth. Here again, the link is not significant (see figure 5 in the appendix). These first observations call for a conceptual framework where we do not only take into account the insights from the literature to work out the “Easterlin Paradox”, but also propose additional channels according to the specific developmental shape of the Burkinabé cotton economy (in the next section). In this regard, we look at the self-expressed opinions about several effects of the cotton reform for rural households. Our data show that these opinions are rather enthusiastic –but heterogeneous among households-, notably for perceived effects on familial income, welfare

8

and input access. One of the empirical challenges in the next section would be to propose an estimation strategy so as to understand to what extent these enthusiastic opinions are linked to the significant increase of subjective wealth and the underlying causality. [Table 4 here]

Empirical framework This section presents an empirical strategy to unveil the mechanisms of subjective wealth increase in the Burkinabé cotton economy, accounting for the determinants of the literature and for the role of opinions and cotton groups’ political involvement. First, we present a general latent utility function, and then discuss the role of opinions and beliefs. Second, we propose an estimation strategy and propose solutions to overcome empirical difficulties such as the role of latent heterogeneity and other potential sources of biases. Conceptual framework and empirical challenges The Easterlin paradox is worked out theoretically when accounting for the role of relative income (Van de Stadt et al., 1985) with respect to a reference group and/or a reference income in time (habituation and preference for increasing wages over time) in the formulation of indirect utility functions. A first effect is the comparison externality when wealthier neighbors negatively affect the social rank of the household. In contrast, Senik (2004) has shown that the welfare of other members of a reference group can bring information on what would be own welfare in the future through expected wage or income profiles. This is the information effect that can dominate the former under high income or social mobility. Other papers point out that information and comparison matter differently for the rich or for the poor (Ravallion and Lokshin, 2005), and according to specific relations between an individual and its reference group (Kingdon and Knight, 2007)11. In rich countries, social comparisons are based on consumption of positional goods, while in poor rural areas, it is likely that land holdings matter (Van Landeghem et al., 2008). However, within cotton groups, people share information about input use, land use, and cotton earnings of members, so that income might well be a source of comparison as well. Information about land, consumption, and income will be useful for us to identify what channels came into play for the reference group effect. This approach of utility functions has also direct implication for poverty analysis, as in our case. Indeed, as Sen (1983) firstly argued, relative concerns such as relative consumption should be taken into account when setting a poverty line or measuring poverty. This would put together income levels and income profiles into the implementation of poverty measures. In the context of poverty, a lower rank on the wealth ladder means that the households will 9

suffer from a lack of access to basic commodities, in case of a crisis. However, a richer neighbor might provide the latter with employment or aid. The first step in our empirical strategy is to define a general latent indirect utility function, which is not directly measurable by available variables. According to the literature, assume indirect utility is affected by own income yi and the average one of a reference group y*. In addition, part of individual income ci (in monetary terms, labor, land, or cattle) is used to contribute to the activities of the cotton group, which participates to the collective action of farmers through their national union. The overall contribution c of local cotton groups increases available individual income through the political bargaining and economic cooperation with other stakeholders. The return of collective action on individual income is assumed to have a common, f(c), and a specific component θi. f(c) is a positive increasing and concave function of the overall contribution of local cotton groups and equals 0 when there is no contribution. Furthermore, collective action provides farmers with an additional source of welfare, which is related to their feeling of political and economic empowerment. We model this advertising effect as a signal σG that is a function of their contribution ci to the group’s political activities and of the group G’s characteristics. The latter are likely to be responsible for differential information channels to the individual farmer about the impact of collective action, according to the level of cooperation, the governance the influence, the composition and the size of the group. This advertising effect can be seen either as a positive or a negative externality, as in the spirit of Becker and Murphy (1983). If collective action increases individual incomes, farmers may also be dissatisfied with the attitude of their leaders, in case of elite capture for instance. So, σG(.) may be either an increasing concave function of individual contribution, either a decreasing convex one. Individual welfare becomes: Wi=Ui ((yi-ci )(1+ θi f(c)) ,(y*-c*)(1+ θ* f(c)))+ σG(ci)

(1)

where * subscripts refer to the average characteristics of the reference group. Let us rename the two arguments of Ui as Yi and Y*. Assuming an interior solution (0 < ci < yi), the individual optimal contribution cio will respect the following first-order conditions: ∂U i ∂U [( yi − ci O )θ i f '(c) − 1 − θi f (c)] + *i [( y* − c* )θ * f '(c)] + σ G '(ci O ) = 0 ∂Yi ∂Y

(2)

The optimal contribution equalizes marginal utility benefits and costs to the marginal advertising effect. Hence, the advertising effect of own contributions to collective action of farmers will rely on the same determinants than those of indirect utility, that is, individual income and average one of a reference group, in addition to group characteristics defining the signal function. Because of the group mechanisms in the case of cotton in Burkina Faso, 10

individual income does not depend on group characteristics in the long-term, because it is now possible to change group and to establish new ones, as well as to control the formation and composition of its group (see the discussion in Kaminski and Thomas, 2008). Controlling for cotton experience –which may also control for habituation to cash income in subsistence rural economies-, group characteristics would only matter for the signaling effect of involvement in collective action, but not for the income effect. So, we can define the general latent utility function : W(yi,y*,Xi,ci(yi,y*,G,Qi))

(3),

where G is a vector of the farmer’s group characteristics, and Xi and Qi are vectors of individual characteristics (with possible common components, but not necessarily the same). While individual and reference income can be measurable, it is pretty difficult to measure what would be the advertising value of own contribution of farmers in their process of change and in the shaping of policies. We argue that one valuable proxy should be their own opinion about the effects of a political change where they have been involved, such as the experience of the cotton reform. The empirical strategy then should test for the significance of political opinion in the assessment of subjective wealth when controlling for the other determinants to valid our hypothesis of advertising effect. Concerning the determinants of the satisfaction with respect to a reform agenda, an empirical study (Bonnet et al., 2006) of the global dissatisfaction about the privatization of infrastructures in Latin America shows that individual beliefs and expectations as well as absolute and relative welfare effects are responsible for the perception of privatization (dissatisfaction in this case). Combining and disentangling welfare effects and shifts in beliefs allow the authors to explain the perception of the reform and the divergence from welfare effects. Based on Martimort and Straub (2009), the belief in the change of corruption patterns –and its increase in privatized utilities-, notably among the middle class who were the big losers of South American reforms, is a significant factor of public discontent. The use of a set of fundamental beliefs and opinions is helpful to control for the propensity to be critical on an economic reform. The role of beliefs is not only relevant for the estimation of political opinions, but also for the self-assessment of well-being. A range of arguments in the literature encompasses the role of beliefs in self-assessment of welfare and self-judgment of policies. Benabou and Tirole (2006) take into account the role of beliefs in the redistributive policies of both US and European policies, emphasizing the importance of “believing in a just world”. Hopkins and Kornienko (2004) bring evidence about the role of beliefs into the relationship between 11

happiness and income distribution.

They introduce conspicuous consumption through

interdependent preferences, showing that inequality matters differently for the rich or for the poor according to expectations on social mobility and social norms, and political opinions. Piketty (1995) highlights the role of individual experience and perception of social mobility in political attitude and redistribution policies while Zaller (1991) focus on how information could influence opinions, in the presence of belief systems. Because political opinions and subjective wealth have common components (beliefs, welfare and advertising effects), with possible endogeneity, we need to consider them as covariant processes in a simultaneous-estimation framework where endogeneity can be tested, as well as causality. Before presenting the structural model, drawn on the conceptual model such as described previously, we tackle the main empirical challenges for the estimation of subjective dependent variables: the treatment of latent heterogeneity and the anchoring effect, and other potential biases. The correlation between the verbal expression of satisfaction and the latent utility is far from trivial with a matter of interpersonal comparability when people exhibit psychological differences. Indeed, as Ravallion and Lokshin (2001) have shown, the identification of welfare effects has to take into account that people have in mind their own ladders of satisfaction and their own way to answer surveys. Moreover, some cognitive biases and misreporting -cognitive dissonance for instance- are often cited as sources of potential biases because subjective data can be subject to manipulation (see Bertrand and Mullainathan, 2001). Finally, the answer to questions on subjective welfare can vary according to mood effects or remind. The use of panel data allows the authors to control for individual-fixed effects to capture personality traits, assuming orthogonal mood effects in standard residual terms. While fixed-effects are useful in order to control for intercept heterogeneity, the socalled “anchoring effect”, latent heterogeneity may also induce different relationships between explanatory variables and the subjective variable, that is, slope heterogeneity. Another technique presented by Clark et al. (2005) consists of latent class estimations to introduce intercept and slope heterogeneity between income and satisfaction with financial situation. Intercept heterogeneities treated by class of individuals are a way to address the treatment of the anchoring effect while slope heterogeneity allows for a non-linear pattern of income influence on happiness. This approach is particularly relevant for their cross-country analysis and data at the national level. Nonetheless, other problems are present: aggregation (forgetting

within-household

inequalities),

income

measurement

(attenuation 12

bias),

misspecification of the relevant income variable, income endogeneity , definition of the 12

relevant reference group and non-linearity of the explanatory variables of the subjective welfare variable. The latter is addressed by the introduction of relative income with respect to a reference group. In addition, these papers put forward the main role played by health, job, and marital status in the income/happiness relationship. Estimation strategy In the case of cotton farmers in Burkina Faso, we firstly follow the approach of Ravallion and Lokshin (2001) to estimate subjective wealth, and then we use a time in-difference approach to get rid of the fixed effects. Our variable of interest, the evolution of subjective wealth, could then be estimated. Several opinion indicators about the perceived effects of the reform are introduced as explanatory variables, according to (3). Because these opinions reflect satisfaction with respect to a policy change that is a sectoral reform, these variables capture the change in the advertising effect over the reform period. Therefore, they should be used as explanatory variables for the in-difference estimation. As we can consider that farmers were not involved before the reform, we can assume that the advertising effect was null before, and use them to estimate the current level of subjective wealth. We also introduce other indicators of wealth, such as absolute and relative familial land holdings, livestock, and consumption in positional goods (social events, oil, and medicines). The role of land rights is not very significant in rural Burkina Faso, as shown by Brasselle et al. (2002). Let us consider this specification of subjective wealth for an individual i at time t: V (Wit ) = α1 ln( yit ) + α 2 ln(

yit A ) + α 3 ln( Ait ) + α 4 ln( it ) + β X it + γ Z it + ηi + ε it yt * At *

(4),

where V (Wit ) is the verbal expression of subjective wealth (traducing latent utility), Ait is a vector of non-income wealth components such as land holdings, livestock, health status, or consumption of positional goods, X it is a vector of household characteristics which are timevarying (including beliefs and aspirations13). Note that this vector may include other elements

than in (3), because of the difference between Wit and V (Wit ) . Z it is a vector of opinion indicators about the effects of the cotton reform. η i is a vector of household fixed effects

designed to capture time-invariant personality traits and personal ladders. The residual term, ε it , is independent across households and time, identically distributed following the normal law, centered in zero, with a homoskedastic variance σ². It corresponds to orthogonal shocks related to mood variability and measurement errors. The reference group corresponds to the set of all cotton groups from the same village. This is quite relevant as much information is shared among members, and cotton production is observable. 13

The available data is a discrete variable, according to a defined scale. Moreover, our variables of interest are ordinal14. Scores are related to a specific ladder where lags are not proportional (8 over 10 does not mean as twice as 4), only the ranks matter. So we consider the following estimation strategy. People transform their latent utility function in a reported well being at time t according to a scale of J+1 discrete numbers. Call their answer vit which belongs to the set of {0, 1… j… J}. The latent continuous utility function Wit , as defined above in (3), can take values on J+1 intervals, separated by J+2 ordered threshold parameters {so = − ∞ , s1, …, sj, …, sJ, sJ+1= + ∞ } such that:

vit = j Ù s j ≤ Wit < s j +1

(5)

Then, the distribution of the observed vit conditional on y it , yt *, Ait, At*, X it and η i is the standard ordered Probit estimate of (4). The parameters estimates are the solution of the Full Information Maximum Likelihood (FIML) under the assumption of exogeneity of independent variables. We then specify the evolution of subjective wealth over the reform period such as: ∆V (Wi ) = α1∆ ln( yi ) + α 2 ∆ ln(

yi A ) + α 3∆ ln( Ai ) + α 4 ∆ ln( i ) + β∆X it + γ Z i + µi yi * Ai *

(6),

where ∆(.) is the time in-difference operator between 1996 and 2006, over the reform period. This directly follows from (4) since we assume time-invariant regressors and Z it =0 in 1996. µi is an independent and identically distributed shock, following a normal law centered in zero and having an homoskedastic variance ν². Then, we follow the same reasoning as above with an observable in-differences discrete measure of satisfaction, ∆vit, which maps the number of won or loosen rungs on the satisfaction ladder to the differential latent utility, as described above in (6). The only difference is now that the threshold parameters will differ from the ones used in (5) and now, they will be 2J+1 intervals corresponding to 2J+2 cut off values. As before, the conditional distribution of ∆vit with respect to its independent variables would be the standard ordered Probit model under the exogeneity assumption. We now turn to the estimation of opinion indicators, which are also discrete variables: Z i = δ1 ln( yi ) + δ 2 ln(

yi A ) + δ 3 ln( Ai ) + δ 4 ln( i ) + φ Qi + ϕ∆V (Wi ) + Gi + ζ i yi * Ai *

(7),

where all parameters are such as defined above, Zi is an opinion indicator, Gi is a vector of

14

group characteristics, and Qi is a vector of household characteristics at the time of the interview, containing both fixed effects and time-varying components, controlling for latent heterogeneity. Note that Qi and Xi have common components, but Qi may have additional elements than in (3), because of the difference between the latent advertising effect and the information captured by the opinion variable. ζ i is a residual term, following a centered normal law with variance ω. As before, the opinion indicator zi is a discrete variable, taking I+2 values on a set of I+1 intervals, which reflects the latent advertising effect of farmers’ involvement in policymaking, σG(ci). Then the distribution of zi conditional on regression parameters can be consistently estimated by ordered Probit FIML estimators. However, as discussed previously, equations (6) and (7) might be correlated, because of covariant evolution of subjective wealth and political opinions. Then, we propose a simultaneous-equation framework where residual terms of each equation can be correlated, and where each dependent variable may be an endogenous explanatory variable of the other one. If residual covariance is not significant, but we have endogenous variables, then a reduced form of the system may exist or we can account for endogeneity biases. If residual covariance is significant, then we can control for exogenous simultaneity of both processes. yi Ai ⎧ ⎪∆V (Wi ) = α1∆ ln( yi ) + α 2 ∆ ln( y * ) + α 3∆ ln( Ai ) + α 4 ∆ ln( A * ) + β∆X it + γ Z i + µi ⎪ i i ⎨ y A ⎪ Z = δ ln( y ) + δ ln( i ) + δ ln( A ) + δ ln( i ) + φ Q + ϕ∆V (W ) + G + ζ 1 2 3 4 i i i i i i ⎪⎩ i yi * Ai *

(8)

( µi ,ηi ) follows a bivariate normal law centered in zero with respective variances ν² and ω² and with correlation coefficient ρ = Cov( µi ηi ,) / (νω). The joint conditional distribution of both dependent variables can be estimated by a consistent FIML Bivariate ordered Probit estimator, such as provided by Sajaia (2007). All parameters can be consistently estimated up to a constant term, including 2J+I+4 cut off values. This approach has been followed in Kaminski and Thomas (2008)15. The possible correlated error terms of the two equations of (8) correct the potential endogeneity of Z i on ∆V (Wi ) and vice versa, whenever γ (resp. ϕ ) is significantly different from zero (Wald-test) and the simultaneity of these processes. We want to test the two hypotheses γ =0 and ϕ =0. As welfare and belief effects are controlled in the second equation, we expect that only the first hypothesis would be rejected, then having possibly endogenous and covariant political opinions in the assessment of change in subjective wealth. But we also aim to explore the reverse causality. These hypotheses can also be tested in the univariate

15

ordered Probit of (6) and (7) by the Rivers-Vuong (1988) approach. We will then estimate system (8) for several opinion indicators, one by one, and use them as exogenous or endogenous, according to the Rivers-Vuong exogeneity tests done before. However, our conclusions should take into account that bivariate estimations can only be partial stories because it is not possible to estimate all potentially endogenous variables with multivariate estimates. The identification of (8) requires an exclusion restriction to be satisfied. One explanatory variable of each equation should not be such in the other one. This variable is interpreted as a valid instrument of the latent opinion on a reform’s effect, which corrects the endogeneity bias. For endogenous political opinions, Gi is a specific vector of instruments, which does not explain changes in subjective wealth. Hence, group characteristics are valid instruments, based on the assumptions of the conceptual framework. If we need to have instruments for changes in subjective wealth, we will need to justify what additional elements of Xi should not be comprised in Qi. There are elements that explain heterogeneity in evolution of subjective wealth, once objective absolute and relative measures of wealth are accounted for (i.e. income and land), but should not be used as welfare effects for opinion indicators. For example, the evolution of positional expenses, once evolution of absolute and relative land holdings are controlled for, can be used as instruments. To tackle the issue of instrumental beliefs in the opinion variables, the joint estimation approach is relevant. If changes in subjective wealth are endogenous in the second equation of (8), then its unobservable related beliefs are instruments. This overcomes the absence of lagged opinion indicators or year-fixed effects that could be used as instruments, as in Bonnet et al. (2006), because we take into account residual covariance and endogenous change in subjective wealth. Finally, we address the issue of slope heterogeneity by splitting the sample according to significant household characteristics, namely ethnic background and income status. We then estimate (8) for the different sub-samples, and use the results as robustness checks. This will enable us to check heterogeneity in the reference group effect, and in the advertising effect. This departs from Clark et al. (2005) where the authors let the data select sub-groups with the use of the EM algorithm. To sum up, our estimation strategy is as follows. We firstly estimate (4) and introduce five political opinion variables. Then, we estimate (6) and (7) separately, with the Rivers-Vuong statistics. We estimate (8) with Bivariate ordered estimates for five different systems of equations, corresponding to five opinion indicators. According to previous exogeneity tests, 16

we treat both changes in subjective wealth and opinion variables as either exogenous or endogenous, and check for significant covariance among residuals. Finally, we run robustness checks and compute marginal effects.

Econometric results and discussion To define wealth in a broad sense, we use a set of different variables, including per-capita income, familial land holdings and livestock value, levels of investment and debt, and consumption in positional goods such as social events, transports, meat, and dairy products. For positional consumption, we only take indicators of relative consumption with respect to the reference group (cotton groups of the village) as absolute consumption is collinear with income, savings (livestock), and investment. To capture the evolution of wealth over the reform period in addition to relative levels of changes, we introduce absolute levels of changes in consumption since we have no reliable income variable in 1996. We also introduce evolution of absolute and relative shares of land allocated to cotton, and changes in non-farm income. The set of control variables is composed by the family size (in consumption units), the ethnic status, the age and education of the household’s chief, and the health state of the family. While family size controls for decreasing marginal costs of living and increasing labor force, ethnicity may control for a differential access to land and productive assets. Education and age are found to be relevant controls in the literature, with a U-shape pattern of the effect of age on subjective well-being (Blanchflower and Oswald, 2007). We also control for aspirations and expectations, as defined by current and past expectations about cereal and cotton prices and own farm production. Finally, we use cotton experience dummies as a control for the changes in subjective wealth and for opinion indicators. We also keep fixed effects in the time in-difference estimation because of possible heterogeneity in answers to retrospective questions. The changing environment of producers over the reform period is characterized by a rise in land holdings and in land-use patterns in favor of cotton, as well as the increasing rate of animal-farming adoption (called mechanization). We will use these additional variables in the estimation of changes in subjective wealth and we will add group characteristics components for the estimation of opinion variables, according to (8). For the latter ones, we will only use relative measures of income evolution (land and cotton share of land) to control for relative welfare effects, in addition to absolute ones. Finally, changes in subjective wealth and opinion

17

indicators will be used either as explanatory variables if exogeneity is rejected by RiversVuong tests of significance. Table 6 (see in the appendix) displays ordered Probit estimates of current subjective wealth with a progressive introduction of control variables and opinions. We learn that absolute income is not very significant, unlike its relative measure with a significant comparison effect. In addition, land holdings significantly affect subjective wealth, with a positive absolute effect that dominates the negative relative one on average16. Other things being equal, an increase in land holdings of neighbors will increase subjective wealth (information effect). While investment and debt have the expected signs, consumption in positional goods exhibit significant impacts for subjective wealth. This is notably the case for the relative consumption in transports and in social events. Transports are important for comparison purposes. Social events might matter more for information or because of positive externalities from community welfare on individuals. The negative effect of relative expenses in social events may also be interpreted by constraints from social norms on actual welfare. The role of control variables appears as crucial, notably for family size (positive) and ethnic status, while the number of diseases and injuries is negatively impacting perceptions of wealth. The latter effect is not robust with respect to the specifications, and education is not significant. This can be attributed to a lack of heterogeneity in education and health levels for our sample, and to colinearity with other objective wealth variables. The ethnicity effect can be interpreted as greater aspirations and future expected consumption for non-resident ethnic groups, due to new economic opportunities and less related institutional constraints, notably for cotton production. Note that positive expectations of future cereal prices are significant while the opinion about the effect of the reform on poverty alleviation is the only significant advertising effect. We now turn to the dynamic analysis of subjective wealth. In tables 7 and 8 (in the appendix), we estimate equations (6) and (7), testing for the exogeneity of our two dependent variables, and considering five potential opinion channels. While the roles of land holdings (dominating information effect) and social events (negative impact of increasing expenses not offset by a positive relative effect) are confirmed in the time in-difference approach, this is not the case for transports, debt or investment. As we may expect, fixed-effects are not significant, even after introducing cotton experience dummies, except for the role of ethnicity (still negative for resident ethnic groups). However, adoption of mechanization brings an additional positive impact on the evolution of wealth perception. Changes in expectations about commodity prices are not significant. Finally, the positive effects driven by opinions about the reform impacts on income and poverty reduction are 18

partly offset by the ones on farm skills and input access. Here again, we suspect that comparison effects were at work. Rivers-Vuong exogeneity tests cannot be rejected, so we should only consider exogenous advertising effect in the following. For the estimation of opinion indicators, absolute and relative welfare effects are significant but differently according to which opinion is considered. Current relative income levels matter for opinions about income and input access effects, while changes in land holdings are significant for the two other opinion indicators. For all these variables, except the opinion about income effect, the change in subjective wealth is negatively and significantly correlated, and its exogeneity is rejected by the Rivers-Vuong statistic. It means that opinions had an overall positive advertising effect (exogenous) but that once we control for objective welfare effects, evolution of wealth perceptions has negatively affected the appreciation of the reform. It can be interpreted by a trade-off between involvement in the reform process of cotton groups, and investment in own individual production. It is also likely the more farmers benefited from the reform or experienced in increase in living standards, the less they were willing to participate to collective action. Finally, unobservable belief effects explaining change in subjective wealth might be negatively correlated with those explaining positive appreciation of the reform on poverty alleviation, farm skills, or input access. However, several effects appear as not robust to the different specifications, so we should estimate consistently (8) by bivariate procedures accounting not only for endogeneity but also for residual covariance among the two processes. [Table 9 here] Tables 9 and 10 display consistent estimates of (8), treating the change in subjective wealth as an endogenous variable in the determination of political opinions about the cotton reform, except for the income effect. This is confirmed by the negative significant impact of change in subjective wealth on opinions in table 10, while residual covariance is always significant and positive. This is justifying the joint-estimation approach. Significant simultaneity between subjective wealth evolution and opinions indicators is changing several estimates. In table 9, land holdings are still positively correlated to subjective wealth evolution with an overall positive effect while this is the reverse for social events. Consumption in goods of first necessity (cereals, energy, or housing) has now negative impacts while mechanization is still positive and robust. The ethnic origin is more significant and also robust to all specifications while changes in the expectations about cotton prices are now significant and positive. The opinion about reform’s impact on poverty alleviation is the positive and robust while the one on income is not. Opinions about input access and farm 19

skills are still negative. That means that the appreciation of the reform on accessing productive factors could have concerned the poorest households and negatively impacted the richest ones. Indeed, it is likely that the more universal access to inputs and technologies allowed by the reform might have negatively impacted the relative wealth of the richest households. Here, this means that the subjective comparison effect would have dominated the information one, in addition to objective channels. The reverse interpretation holds for the opinion about the positive perceived effect of the reform on poverty reduction. [Table 10 here] About opinions, welfare effects are different. While relative income is positively correlated with opinions on income and input access effects, this is not the case for perceived effects on farm skills and poverty reduction where land holdings matter. This is also the case for ethnicity that only matters for the two former opinions. Note that mechanization is not significant in the different indicators of the satisfaction with the reform, even in the perceived effect on farm skills. Welfare effects are already controlled by various measures of wealth and wealth changes, such as land holdings, income and consumption. Note also that group characteristics are not significant, except for the perceived effect on input access. Hence, these variables could not have served as instruments in our estimation strategy, which is in line with exogenous advertising effects of the political involvement of farmers. In contrast, instruments for endogenous change in subjective wealth should be uncorrelated elements of social comparison, such as changes in social events expenses. Another valuable instrument might be the mechanization, which is a robust additional source of subjective wealth apart from other welfare effects. The positive residual covariance means that once we controlled for observable determinants of opinions and subjective wealth evolution, unobservable effects are positively correlated. This might notably be the case for common unobservable beliefs. To get an idea about the size of effects, we have a look at significant marginal effects (figure 6). The largest positive effect is the one of land holdings, followed by the one of mechanization. The overall effect of evolution of expenses in social events is negative and of the same extent than the ethnicity one. For opinion channels, the poverty reduction effect is the largest and dominates the ones of farm skills and input access. [Table 11 here] Finally, we aim to check for different effects according to income and ethnic groups so as to deal with the slope heterogeneity issue. In table 11, we use the second specification of the Bivariate ordered Probit estimation of (8) for four sub-samples. Results show that effects are very specific to each sub-group of the population. The role of social events is only significant 20

for migrant ethnic groups and matters more for the non-poor than for the poor while the one of land holdings matters only for the poor (as relative income). So, the comparison channel is very different among groups. In addition, mechanization has only played a role for resident ethnic groups and non-poor. Interestingly, advertising effects of opinions are also different for each group. The positive effect of opinion on income effect only matters for the poor while the negative farm skills and input access effects are significant for the resident ethnic groups and the non-poor. All this gives support that the negative comparison effect has only affected the richest and the resident ethnic groups with a more equal distribution of wealth or economic opportunities within villages and between groups. Accordingly, note that the negative effect for resident ethnic groups only plays for the richest households. The formerly identified effects on opinions (table 10) are significant only for resident ethnic groups and the non-poor in the case of the poverty-reduction effect. So, only residual covariance positively explains this opinion for the poor and ethnic migrant groups.

Conclusion In this paper, we have identified the impacts of policy change in the context of rural development on wealth not only through the standard determinants existing in the empirical literature, but also through the political involvement and participation of farmers (subjective channel of appropriation). The issue of comparison and information effects has been worked out by the choice of the reference group (village income of cotton group members) where the shaping of policy opinions also takes place. We have found that policy opinions cannot be disentangled only from the impact of the reform on objective wealth but also on common beliefs. Though each opinion channel has a specific relationship with the change in subjective wealth, we have showed that all have an exogenous advertising effect. In addition, opinions are endogenously and negatively determined by changes in subjective wealth. That means that there is a trade-off between the positional change on the social ladder and the political involvement of cotton farmers. More importantly, these effects are very different among groups. Poor and non-poor compare differently, and the changing environment as captured by mechanization and political opinions has impacted ethnic groups in different ways. The particular kind of relationship between farmers, policymakers and foreign cooperation is likely to have influenced these mechanisms because farmers’ groups can be involved into political processes according to the performances of their social organizations, the political regime and the willingness of foreign actors and governments to work with farmers’ representatives. In Burkina Faso, this has led to a strong leadership for cotton farmers, capable 21

to work with policymakers and to be responsible for a growing number of responsibilities. This has been supported by the willingness of government and foreign actors to delegate decisional and management power to farmers to establish a consistent industrial partnership leading to better production incentives. In Mali, the democratic regime is more favorable for farmers to put pressure on government and jeopardizing cotton production to oppose to any reform policy. Democracy has been associated to less coercive power of government and more lobbying power of farmers, yielding to the prevalence of the political status quo. Future research could be done to further explore the mechanisms we have identified in a comparative framework of political change in rural development. The construction of a panel database would be helpful in this regard.

References Azam, J.P., and Djimtoingar, N. (2007) Cotton, War and Growth in Chad (1960-2000) in B.J. Ndulu, S.A. Connell, J.P. Azam, R.H. Bates, A.Fosu, J.Gunning and D. Njinkeu (eds.): The Political Economy of Economic Growth in Africa, 1960-2000, Vol.2: Country Case Studies, pp. 86-115. Harvard University Press, Cambridge, Massachusetts. Bassett, T.J., (2001) The Peasant Cotton Revolution in West Africa, Côte d'Ivoire, 1880-1995, ed. Cambridge University Press, Cambridge, UK, pp. 264. Becker, G.S., and Murphy, K.M., (1993) A Simple Theory of Advertising as a Good or Bad, The Quarterly Journal of Economics 108, 4, pp. 941-964. Benabou, R., and Tirole, J. (2006) Belief in a Just World and Redistributive Politics, The Quarterly Journal of Economics 121, 2, pp. 699-746. Bertrand, M., and Mullainathan, S. (2001) Do People Mean What they Say? Implications for Subjective Survey Data, American Economic Review 91, 2, pp. 67-72. Bingen, R.J. (1998) Cotton, Democracy and Development in Mali, The Journal of Modern African Studies 36, 2, pp. 265-285. Blanchflower, D.G., and Oswald, A.J. (2007) Is Well-Being U-shaped over the Life Cycle? NBER Working Paper 12935. Blanchflower, D.G., and Oswald, A.J. (2004) Well-Being over Time in Britain and the USA, Journal of Public Economics 88, pp. 332-337. Bonnet, C., Dubois, P., Martimort, D., and Straub, S. (2006) Empirical Evidence on Satisfaction with Privatization in Latin America: Welfare Effects and Beliefs. IDEI Working Paper, Toulouse, pp. 60.

22

Brasselle, A., Gaspart, F., and Platteau, J-P. (2002) Land Tenure Security and Investment Incentives: Puzzling Evidence from Burkina Faso, Journal of Development Economics 67, pp. 373-418. Clark, A., Etilé, F., Postel-Vinay, F., Senik, C., and Van der Straeten, K. (2005) Heterogeneity in Reported Well Being: Evidence from Twelve European Countries, Economic Journal 115, pp. 118-132. Clark, A., Frijters, P., and Shields, M. (2007) Relative Income, Happiness and Utility: An Explanation for the Easterlin Paradox and Other Puzzles, IZA Discussion Papers 2840, Institute for the Study of Labor. Bonn, Germany, pp. 57. Di Tella, R., MacCulloch, R., and Oswald, A.J. (2003) The Macroeconomics of Happiness, Review of Economics and Statistics 85, pp. 809-827. Diener, E., and Lucas, R.E (1999) Personality and Subjective Well-being in Kahneman, D., Diener, E., and Schwarz, N. (Eds.), Foundations of Hedonic Psychology: Scientific Perspectives on Enjoyment and Suffering, Chapter 11. Russell Sage Foundation, New York, US. Easterlin, R.A. (1974) Does Economic Growth Improve the Human Lot? Some Empirical Evidence in David, P.A., and Reder M.W. (Eds), Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz. Academic Press, New York, US. Easterlin, R.A. (1995) Will Raising the Income of All Increase the Happiness of All? Journal of Economic Behaviour and Organization 27, pp. 35-47. Easterlin, R.A. (2001) Income and Happiness: Toward a Unified Theory, The Economic Journal 111(473), pp. 465-484. FAO, 2007. FAOSTAT Database for agricultural commodities. Ferrer-i-Carbonell, A., and Frijters, P. (2004) How Important is Methodology for the Estimates of the Determinants of Happiness? Economic Journal 114, pp. 641-59. Goreux, L. (2003) Reforming the Cotton Sector in Sub-Saharan Africa. World Bank, Washington, D.C., World Bank Africa Region Working Paper Series, pp. 87. Graham, C., and Pettinato, S. (2002) Happiness and Hardship: Opportunity and Insecurity in New Market Economies, ed. The Brookings Institution Press, Washington, D.C. Hopkins, E., and Kornienko, T. (2004) Running to Keep in the Same Place: Consumer Choice as a Game of Status, American Economic Review 94, 4, pp. 1085-1107. INSD (Institut national des Statistiques et de la Démographie) prioritaire.

23

(2003). Enquête

INSD (Institut national des Statistiques et de la Démographie) (2006) Rapports annuels des enquêtes socio-démographiques du Burkina Faso, Ouagadougou, Burkina Faso. Kaminski, J. (2007) Cooperative Reform and Outgrower Schemes in the Cotton Sector of Burkina Faso. TSE ARQADE Working paper, Toulouse, France, pp. 50. Kaminski, J., and Thomas, A. (2008) Commodity Reform and Extensive Production Growth: Evidence from Burkinabè Cotton Farmers in Burkina Faso. CAER Working Paper, Hebrew University of Jerusalem, Rehovot, Israel, pp.31. Kingdon, G.G., and Knight, J. (2007) Community, Comparisons, and Subjective WellBeing in a Divided Society, Journal of Economic Behavior and Organization 64, pp. 69-90. Krueger, A., and Schkade, D.A. (2007) The Reliability of Subjective Well-Being Measures. IZA Discussion Papers No. 2724, University of Bonn, Germany. Lachaud, J-P. (2005) A la Recherche de l’Insaisissable Dynamique de Pauvreté au Burkina Faso. Une nouvelle évidence empirique. Documents de travail 117, Centre d'Economie du Développement de l'Université Montesquieu Bordeaux IV. Martimort, D., and Straub, S. (2009). The Political Economy of Private Participation: The Roots of Public Discontent , Forthcoming in Journal of Development Economics. Piketty, T. (1995) Social Mobility and Redistributive Politics, The Quarterly Journal of Economics 110, pp. 551-584. Ravallion, M., and Lokshin, M. (2001) Identifying Welfare Effects from Subjective Questions, Economica 68, pp. 335-357. Ravallion, M., and Lokshin, M. (2005) Who Cares about Relative Deprivation? World Bank Policy Research Working Paper 3782, World Bank, Washington D.C. Rivers, D., and Vuong, Q. (1988) Limited Information Estimators and Exogeneity Tests for Simultaneous Probit Models, Journal of Econometrics 39, pp. 347-366. Sajaia, Z. (2007) Maximum Likelihood Estimation of a Bivariate Ordered Probit Model: Implementation and Monte Carlo Simulations. World Bank, Washington, D.C., mimeo, pp. 24. Sen, A.K. (1976) Poverty: An Ordinal Approach to Measurement, Econometrica, 44(2), pp. 219-31. Sen, A.K. (1983) Poor, Relatively Speaking, Oxford Economic Papers 35, pp. 153169. Senik, C. (2004) When Information Dominates Comparison. Learning from Russian Subjective Panel Data, Journal of Public Economics 88, pp. 2099-2123.

24

Van de Stadt, H., Kapteyn, A., and Van de Geer, S. (1985) The Relativity of Utility: Evidence from Panel Data, Review of Economics and Statistics 67, pp. 179-187. Van Landeghem, B., Swinnen, J., and Vranken, L. (2008) Land and Happiness: Land Distribution and Subjective Well-Being in Moldova. Paper prepared for presentation at the 12th EAAE Congress, Gent, Belgium, pp.39. World Bank (2005) World Development Report, ed. World Bank, Washington, DC. Zaller, J. (1991) Information, Values and Opinion, The American Political Science Review 85, 4, pp. 1215-1237.

0

0

2

Quantiles of pci 500

Quantiles of pcl 4

6

1000

Figures

0

.25

.5 Fraction of the data

.75

1

0

.25

.75

1

Figure 2. Per capita land distribution

0

0

Quantiles of Subjective wealth in 1996 2 4 6 8

Quantiles of Subjective wealth in 2006 2 4 6 8

10

10

Figure 1. Per capita income distribution

.5 Fraction of the data

0

.25

.5 Fraction of the data

.75

1

0

.25

.5 Fraction of the data

.75

Figure 3. Distribution of subjective wealth before the reform and today

25

1

10 Current subjective wealth 2 4 6 8 0

0

500 Per-capita income

1000

95% CI Subjective wealth in 2006

Fitted values

Figure 4. The empirical relationship between per-capita income and perceived wealth constant

increase <1 ha

increase [1,2] ha

increase [2,5] ha

increase >5 ha

0 -5 10 5 -5

0

Evolution of subjective wealth

5

10

decrease

Graphs by Evolution of familial cultivated land

Figure 5. Evolution of subjective wealth by regimes of familial land extension Marginal effects 1 0,8 0,6 0,4 0,2 0 -0,2 -0,4 -0,6 -0,8 -1

Marginal effects 0,08

-1

0

1

2

3

4

5

Relative ∆land

0,06

Absolute + relative ∆social events

0,02

Adoption of animal farming

-0,02

Ethnic status

-0,06

Opinion on income effect

0,04

Opinion on poverty reduction effect

0 -1

0

1

2

3

4

5

-0,04 -0,08

Opinion on farm skills effect Opinion on input access effect

-0,1

Changes in subjective wealth (in number of ranks)

Changes in subjective wealth (in number of ranks)

Figure 6. Marginal effects of objective variables and opinion channels on changes in subjective wealth

26

Tables Table 1. Summary statistics of main variables Variable Family size Age Education level Ethnic group Risk aversion Expected cotton price Expected cereal price Expected crop production Subjective wealth Past subjective wealth ∆ Subjective wealth Income Relative income Land Livestock Non-farm income Debt Investment Social events Transport Health Dairy products Animal proteins Mechanization ∆ Land ∆ Cotton share Cotton experience GPC management GPC quality

Description Mean SE Household main characteristics Number of consumption unities (1 for a man, 0.8 for a woman, 0.6 6.393 3.499 for a child between 6 and 18 years, 0.3 for a child under 6) Age of the household head, in years 33.980 8.082 School degree of the household’s chief: No school (53 %), alphabetization (10 %), coranic school (6 %), 5 years (20 %), 9 years (7 %) Ethnic group of the household: Bobo (21 %), Mossi (24 %), Gourounsi (15 %), Dagara (13 %), Local ethnic groups (24 %), Senoufo (3 %). Mossis and Gourounsis are non-resident ethnic groups Willingness to receive compensation to reduce risk when being 71.242 21.063 paid the harvested production (between 0 and 100% of the harvest value). Anticipation of the trend of cotton price in the future Increase: 29.7 % Anticipation of the trend of cereal prices in the future Increase: 12 % Anticipation of the trend of crop production of the household in Increase: 21.67 % the future Subjective Wealth Perception of wealth on a ladder of [0,10] for the household today 5.313 1.601 Perception of wealth on a ladder of [0,10] 10 years ago 2.960 1.950 Changes in the perception of wealth, in number of ranks 2.354 1.991 Objective wealth indicators Generated household income from crop production, sales of cattle, 137.296 112.815 non-farm income and received transfers in thousands FCFA divided by consumption unities (per capita) Rate of difference between individual per capita income and the 0 .755 average village per capita income Total cultivated land by the household in ha 6.963 4.790 Total value of the livestock of the household in thousands FCFA 657.629 943.749 Value of non-farm generated income in thousands FCFA 13.446 29.003 Value of household non-repaid credits in thousands FCFA 16.075 232.276 Value of investment last year in thousands FCFA 87.931 368.540 Expenses in social events spent last year in thousands FCFA 33.398 45.356 Expenses in transport spent last year in thousands FCFA 31.192 44.562 Expenses in health spent last year in thousands FCFA 30.724 40.693 Value of dairy products consumption last month in thousands .631 1.260 FCFA Value of animal proteins consumption last month in thousands 3.371 3.188 FCFA Agricultural systems and social/technical environment Level of mechanization of the household: traditional farming (20 %), animal farming adopted during the reform (60 %), already mechanized before the reform (20 %) Evolution of total cultivated land by the household: decrease (4 %), same (28 %), increase less than 1 ha (40 %), increase less than 2.5 ha (13 %), increase less than 5 ha (7 %), >5 ha (7 %) Evolution of the land share dedicated to cotton during the reform: decrease (6 %), same (18 %), more (33 %), much more (42 %) Experience with cotton growing: New grower (3 %), Less than 3 years (9 %), Between 3 and 5 years (14 %), Between 5 and 10 years (24 %), More than 10 years (49 %) Perceived quality of management of the cotton group of producers: very good (20 %), correct (66 %), low (13 %), very bad (1 %) Perceived quality of internal relationships within the cotton group: very good (35 %), correct (55 %), low (9 %), very bad (1 %)

27

Table 2. Evolution of living standards during the reform Living standards #Rooms for the household Quality of walls Quality of roof Quality of ground Building cost of habitat (thousands FCFA) Housing changes Property right Water source Water consumption Light Heat source Distance to the main market Distance to the first road Telephone access Distance to the first phone center (en km) Main mean of locomotion At least one person can read At least one person can write At least one person can compute Schooling constraints # diseases/ injuries Consultations Time to the consultancy center Vaccination rates: yellow fever Meningitis Hepatitis Tuberculosis DT Polio Heath state constraints

Infantile mortality

Today Ten years ago 5.27 (3.5) 3.25 (2.19) banco 91 % briks 6 % banco 93 % briks 3 % iron 27 % clay 24 % iron 13 % clay 27 % banco 24 % straw 21 % banco 27 % straw 30 % clay 78 % cement 11 % clay 81 % cement 8 % banco 9 % banco 11 % 566.61 (1076.98) 275.29 (539.06) quality improvement 23 % quality improvement 17 % size increase 20 % size increase 10 % owner 76 % loan 15 % owner 74 % loan 15 % drill 68 % well 28 % drill 46 % well 45 % 288.05 (248.23) 157.01 134.64 lamp/ candles 97 % lamp/ candles 95 % wood 99 % wood 99 % no change : 7.8 km no change: 6.0 km 33 % 32 % 14.89 (16.45) 27.21 (23.08) bike 64 % moto 32 % bike 83 % moto 12 % 58 % 40 % 52 % 33 % 53 % 38 % cost (5.77) distance (1.94) cost (5.66) distance (2.41) need for labor force (1.7) need for labor force (2.32) 2.73 (2.15) 3.41 (5.03) nurse: 74 % doctor: 20 % nurse: 63 % doctor: 18 % healer: 4 % healer:16 % 44.6 min (56.01) 56.0 min (71.36) 73 % 56 % 93 % 76 % 44 % 14 % 52 % 40 % 86 % 77 % cure prices (6.64) cure prices (6.53) distance to care center (3.22) distance to care center (3.43) consultations prices (2.18) consultations prices (2.42) 9.2 % (12.54) 12.8 % (15.98)

Note: standard deviations in parenthesis if present except for schooling and health state constraints (mean of a graduation on [0,10]).

28

Table 3. Evolution of consumption, savings, and investment during the reform Changes in consumption by households Savings Investment Social events Energy Transports Clothing Housing Education Health Alcohol/ Tobacco Beverages Condiments Fat nutrients Dairy products Animal proteins Fruits Vegetables Tubers Cereals

Large increase 12 % 20 % 20 % 21 % 17 % 21 % 10 % 9% 21 % 7% 20 % 16 % 5% 5% 17 % 6% 10 % 5% 19 %

Slight increase 45 % 29 % 47 % 46 % 40 % 47 % 40 % 28 % 39 % 16 % 38 % 47 % 48 % 21 % 47 % 34 % 44 % 33 % 53 %

Constance 43 % 29 % 13 % 18 % 19 % 16 % 26 % 40 % 14 % 55 % 23 % 24 % 31 % 43 % 14 % 36 % 28 % 37 % 17 %

Slight Big decrease decrease (No accumulation) 16 % 6 % 12 % 8% 11 % 4% 15 % 9% 11 % 5% 18 % 6% 16 % 7% 19 % 7% 12 % 10 % 14 % 5% 10 % 4% 10 % 6% 16 % 15 % 14 % 7% 18 % 6% 15 % 3% 18 % 7% 7% 4%

Table 4. Perceptions of reform’s effects (on a scale of [0,10]) Perceived effects of the reform On income On welfare On input access On agricultural knowledge and abilities On relative/world price of cotton On poverty reduction

Mean 5.74 5.13 5.83 2.97 0.82 3.07

SE 2.88 2.82 2.94 3.05 2.23 3.03

Min 0 0 0 0 -5 0

Max 10 10 10 10 10 9

Median 6.5 6 6 3 0 3

Table 5. Matrix of subjective wealth mobility (number of observations) ---------------------------------------------------------------------Subjective| Wealth | Subjective Wealth 10 years ago | 0 1 2 3 4 5 6 7 8 9 ----------+----------------------------------------------------------0 | 1 1 2 | 1 2 1 2 3 | 7 5 17 1 6 1 4 | 10 1 17 10 1 3 1 2 5 | 18 17 19 14 3 2 1 1 6 | 10 3 9 24 14 2 7 | 10 1 2 13 20 1 1 8 | 7 9 2 9 | 1 1 10 | 3

29

Interquart. 3 3 3 6 0 6

Table 6. Ordered estimates of the current subjective wealth Current Subjective Wealth Explanatory variables Log p.c.income Relative p.c.income Log land Relative land Log livestock Relative livestock Relative social events Relative transports Relative animal proteins Relative dairy Log debt Log investment Health # diseases/ injuries Family size Age Age² Resident ethnic group Education dummies Expected cotton prices Expected cereal prices Expected production Risk aversion Opinion on income effect Opinion on welfare effect Opinion on poverty reduction effect Opinion on farm skills effect Opinion on input access effect Constants: 9 cut off values

Ordered Probit 1

Ordered Probit 2

Ordered Probit 3

Ordered Probit 4

.286 (.155)* .160 (.095)* .562 (.190)*** -.303 (.209) .112 (.056)** -.031 (.068) -.100 (.054)* .093 (.049)* -.010 (.076) -.006 (.044) -.173 (.035)*** .088 (.033)***

.302 (.170)* .330 (.112)*** .505 (.235)** -.470 (.223)** .052 (.061) -.020 (.072) -.105 (.055)** .109 (.051)** -.035 (.077) -.006 (.043) -.216 (.054)*** .090 (.034)*** .009 (.008) -.067 (.032)** .073 (.024)*** .024 (.040) -.000 (.001) -.332 (.132)*** yes

.262 (.166) .337 (.109)*** .578 (.248)** -.502 (.238)** .050 (.066) .011 (.072) -.106 (.052)** .087 (.053)* -.037 (.082) -.003 (.041) -.198 (.051)*** .084 (.034)*** .008 (.008) -.054 (.033)* .075 (.024)*** .047 (.042) -.001 (.001) -.388 (.141)*** yes .167 (.142) .792 (.199)*** -.056 (.158) .116 (.302)

.194 (.161) .305 (.112)*** .800 (.250)*** -.637 (.243)** .027 (.068) .033 (.073) -.105 (.054)** .090 (.049)* -.046 (.085) -.006 (.042) -.212 (.047)*** .080 (.035)** .012 (.008) -.053 (.034) .059 (.023)*** .033 (.042) -.000 (.001) -.460 (.146)*** yes .059 (.147) .757 (.204)*** -.153 (.168) -.166 (.326) .013 (.030) .036 (.031) .090 (.025)*** -.018 (.023) -.025 (.022)

6 last cut off values 6 last cut off values 7 last cut off values 6 last cut off values are significant are significant are significant are significant Pseudo R² .066 .089 .105 .121 Observations 300 300 300 300 Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %.

30

Table 7. Ordered estimates of the change in subjective wealth (number of ranks) over the reform period with Rivers-Vuong exogeneity tests Change in Subjective Wealth Explanatory variables Relative income ∆land Relative ∆land ∆cotton land-use Relative ∆cotton landuse ∆cereals ∆animal proteins ∆energy ∆transports Relative ∆transports ∆Housing ∆social events Relative ∆social events ∆Savings ∆Investment ∆Non-farm income ∆Health expenses Mechanization Cotton experience Family size Age Resident ethnic group ∆Expected cereal prices ∆Expected cotton prices Opinion on income Opinion on welfare Opinion on poverty reduction Opinion on farm skills Opinion on input access Rivers-Vuong Statistic

Ordered Probit 1

Ordered Probit 2

Ordered Probit 3

Ordered Probit 4

Ordered Probit 5

.133 (.107) -.218 (.113)** 1.086 (.410)*** .056 (.389) -.148 (1.648)

.116 -.165 1.023 -.047 .262

(.103) (.118) (.421)** (.389) (1.698)

.154 (.295) -.177 (.120) 1.052 (.419)*** -.018 (.426) .127 (1.855)

.093 (.107) .139 (.253) -.097 (.905) -.098 (.402) .358 (1.688)

.076 (.124) -.146 (.121) .878 (.463)** -.053 (.400) .277 (1.689)

-.137 (.088) -.103 (.069) -.080 (.080) -.104 (.144) .749 (.457)* -.073 (.069) -.476 (.163)*** 1.413 (.489)*** .206 (.119)* .016 (.066) .061 ( 069) -.007 (.063) Adoption of animal farming*** Not significant .022 (.023) .003 (.009) -.284 (.166)* .071 (.159) .254 (.156)*

-.106 (.088) -.111 (.068) -.097 (.086) -.028 (.147) .480 (.462) -.125 (.075)* -.515 (.162)*** 1.520 (.515)*** .136 (.120) -.041 (.067) .027 ( 072) -.013 (.065) Adoption of animal farming** Not significant .013 (.024) .004 (.009) -.323 (.175)* .036 (.169) .230 (.155) .044 (.027)* .013 (.030) .082 (.027)***

-.105 (.094) -.104 (.090) -.031 (.173) -.029 (.145) .473 (.485) -.123 (.080) -.516 (.164)*** 1.538 (.507)*** .170 (.293) -.036 (.082) .023 ( 072) -.009 (.075) Adoption of animal farming* Not significant .020 (.055) .005 (.010) -.373 (.440) .048 (.224) .231 (.156) .010 (.342)

-.142 (.092) -.134 (.071)* -.194 (.111)* -.077 (.148) .474 (.457) -.182 (.087)** -.673 (.202)*** 2.102 (.687)*** .019 (.154) -.087 (.074) -.030 ( 081) .026 (.070) Adoption of animal farming** Not significant -.008 (.029) .012 (.011) -.438 (.196)** -.023 (.177) .185 (.157) .051 (.023)**

-.143 (.099) -.121 (.071)* -.102 (.086) -.082 (.163) .693 (.550) -.096 (.087) -.545 (.163)*** 1.591 (.517)*** .143 (.121) -.028 (.067) .040 ( 076) -.008 (.064) Adoption of animal farming** Not significant .005 (.027) .004 (.009) -.255 (.185) .040 (.168) .250 (.159) .051 (.023)**

.082 (.027)***

.350 (.201)*

.080 (.028)***

-.041 (.022)* -.062 (.024)**

-.039 (.022)* -.060 (.024)*** Opinion on income

-.035 (.022) -.034 (.022) -.065 (.025)*** .017 (.105) Opinion on Opinion on input poverty access reduction Constants: 11 cut off 8 cut off values 7 cut off values 8 cut off values 8 cut off values 7 cut off values values are significant are significant are significant are significant are significant Wald Chi² 171.61*** 248.15*** 236.61*** 253.72*** 239.33*** Pseudo R² .140 .159 .159 .159 .159 Observations 293 293 293 293 293 Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %. Three ordered Probits are processed with three different Rivers-Vuong tests of endogeneity, related to the perceived effects from the reform: the first one is the effect on income, the second is the effect on poverty reduction, and the third is the effect on input access.

31

Table 8. Ordered estimates of the opinion indicators about the reform effects with RiversVuong exogeneity tests Political opinions

Ordered Probit 1: Income effect

Ordered Probit 2: Poverty reduction effect

Ordered Probit 3: Farm skills effect

Ordered Probit 4: Input access effect

.415 (.107)*** .046 (.147) -.139 (.539) .145 (.304) -.707 (1.273) .290 (.109)*** .079 (.062) -.002 ( 073) .036 (.061) Traditional farming*** Not significant .067 (.023)*** .002 (.009) -.505 (.162)*** Not significant Not significant .290 (.154)* -.040 (.139) .044 (.104)

.153 (.112) -.681 (.173) 2.694 (.679)*** .109 (.366) .190 (1.590) .322 (.113)*** .055 (.071) .166 ( 081)** -.047 (.064) Not significant

.109 (.127) -.597 (.163)*** 2.920 (.568)*** -.430 (.355) 1.754 (1.489) .218 (.108)** .073 (.068) -.100 ( 081)** -.105 (.057)* Not significant

.266 (.086)*** -.039 (.157) .671 (.567) .175 (.307) -.475 (1.253) .087 (.105) -.081 (.066) .021 ( 071) -.047 (.061) Not significant

Significant*** .056 (.021)*** -.013 (.010) -.081 (.173) Not significant Not significant .168 (.183) .153 (.158) -.470 (.135)***

Not significant .015 (.024) -.013 (.010) -.073 (.183) Not significant Not significant .071 (.175) .113 (.163) -.545 (.123)***

Not significant .073 (.024)*** .005 (.009) -.630 (.165)*** Not significant Significant*** .071 (.175) .113 (.163) -.339 (.112)***

Explanatory variables Relative income ∆land Relative ∆land ∆cotton land-use Relative ∆cotton land-use ∆Savings ∆Investment ∆Non-farm income ∆Health Mechanization Cotton experience Family size Age Resident ethnic group Group relationships Group management ∆Expected cereal prices ∆Expected cotton prices Change in subjective wealth Rivers-Vuong Statistic Constants: 9 cut off values

Not significant Significant*** Significant*** Significant* 3 last cut off values 3 last cut off values last cut off value 4 last cut off values are significant are significant are significant are significant Wald Chi² 97.90*** 82.58*** 92.92*** 116.78*** Pseudo R² .073 .117 .084 .064 Observations 293 293 293 293 Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %. The Rivers-Vuong tests of endogeneity are related to the changes in subjective wealth.

32

Table 9. Bivariate ordered estimates of changes in subjective wealth Change in Subjective Wealth Explanatory variables Relative income ∆land Relative ∆land ∆cotton land-use Relative ∆cotton landuse ∆cereals ∆animal proteins ∆energy ∆transports Relative ∆transports ∆Housing ∆social events Relative ∆social events ∆Savings ∆Investment ∆Non-farm income ∆Health Mechanization

Bioprobit 1: Exogenous opinion on income .162 (.095)* -.176 (.127) 1.038 (.451)** -.034 (.336) .183 (1.401)

Bioprobit 2 Exogenous opinion on poverty reduction .157 (.094)* -.289 (.123)** 1.465 (.438)*** -.088 (.324) .487 (1.350)

Bioprobit 3: Exogenous opinion on farm skills .154 (.095)* -.285 (.123)** 1.406 (.438)*** .080 (.322) -.143 (1.348)

Bioprobit 4: Exogenous opinion on input access .084 (.096) -.187 (.126) 1.059 (.447)** -.011 (.333) .122 (1.389)

-.105 (.082) -.109 (.071) -.097 (.079) -.023 (.159) .470 (.501) -.119 (079)* -.498 (.164)*** 1.472 (.511)*** .173 (.113) -.036 (.064) .022 ( 074) -.009 (.064) Adoption of animal farming** Not significant .021 (.023) .005 (.009) -.382 (.160)** .054 (.142) .231 (.130)*

-.037 (.062) -.064 (.054)* -.128 (.059)** -.021 (.118) .245 (.381) -.134 (051)*** -.388 (.126)*** 1.163 (.398)*** .201 (.111)* -.003 (.061) .020 ( 072) -.012 (.065) Adoption of animal farming** Not significant .021 (.023) .001 (.009) -.317 (.157)** .050 (.140) .239 (.129)* .009 (.018)

-.105 (.060)* -.111 (.053)** -.074 (.058) -.162 (.114) .656 (.367)* -.075 (.051) -.242 (.128)** .827 (.401)** .161 (.111) -.040 (.061) .057 (.072) -.008 (.061) Adoption of animal farming** Not significant .026 (.023) .002 (.009) -.302 (.157)** .077 (.140) .252 (.129)** .050 (.018)*** .038 (.019)***

-.157 (.077)** -.083 (.068) -.101 (.075) -.122 (.151) .880 (.474)* -.039 (070) -.448 (.164)*** 1.327 (.506)*** .174 (.113) -.019 (.063) .046 ( 073) -.037 (.064) Adoption of animal farming*** Not significant .013 (.023) .004 (.009) -.275 (.160)* .067 (.142) .263 (.130)** .036 (.023) .050 (.028)*

Cotton experience Family size Age Resident ethnic group ∆Expected cereal prices ∆Expected cotton prices Opinion on income Opinion on poverty .079 (.027)*** reduction Opinion on farm skills -.038 (.024) -.065 (.017)*** -.058 (.023)*** Opinion on input access -.058 (.023)*** -.042 (.018)** -.066 (.018)*** ρ .131 (.064)** .775 (.065)*** .782 (.070)*** .390 (.161)** Constants: 11 cut off 8 cut off values are 7 cut off values are 7 cut off values are 7 cut off values are values significant significant significant significant Wald Chi² 162.56*** 164.10*** 161.56*** 158.33*** Pseudo R² .109 .142 .129 .106 Observations 293 293 293 293 Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %.

33

Table 10. Bivariate ordered estimates of political opinions Political opinions

BioProbit 1: Income effect

BioProbit 2: Poverty reduction effect

BioProbit 3: Farm skills effect

BioProbit 4: Input access effect

.422 (.092)*** .035 (.133) -.080 (.468) .117 (.312) -.574 (1.311) .284 (.108)*** .070 (.059) -.004 ( 072) .031 (.059) Not significant Not significant .068 (.023)*** .002 (.009) -.532 (.157)*** Not significant Not significant .291 (.143)** -.023 (.130)

.134 (.093) -.489 (.135)*** 1.979 (.469)*** .041 (.320) .319 (1.335) .204 (.113)* .018 (.064) .124 ( 075)* -.043 (.061) Not significant Significant* .040 (.023)* -.008 (.009) -.135 (.165) Not significant Not significant .168 (.183) .153 (.158) -.407 (.052)***

.104 (.094) -.455 (.133)*** 2.249 (.474)*** -.343 (.317) 1.435 (1.326) .136 (.111) .019 (.064) -.078 ( 074) -.092 (.060) Not significant Not significant .012 (.023) -.008 (.009) -.120 (.165) Not significant Not significant .049 (.145) .122 (.138) -.504 (.046)***

.256 (.091)*** -.041 (.132) .658 (.467) .137 (.318) -.315 (1.340) .079 (.107) -.082 (.060) .020 ( 072) -.049 (.059) Not significant Not significant .069 (.023)*** .005 (.009) -.612 (.161)*** Not significant Significant*** .071 (.175) .113 (.163) -.346 (.081)***

Explanatory variables Relative income ∆land Relative ∆land ∆cotton land-use Relative ∆cotton land-use ∆Savings ∆Investment ∆Non-farm income ∆Health Mechanization Cotton experience Family size Age Resident ethnic group Group relationships Group management ∆Expected cereal prices ∆Expected cotton prices Change in subjective wealth ρ Constants: 9 cut off values

.131 (.064)** .775 (.065)*** .782 (.070)*** .390 (.161)** 2 last cut off values 2 last cut off values last cut off value 4 last cut off values are significant are significant are significant are significant Wald Chi² 97.90*** 164.10*** 161.56*** 158.33*** Pseudo R² .109 .142 .129 .106 Observations 293 293 293 293 Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %.

34

Table 11. Robustness checks: heterogeneity of effects by ethnic status and income groups Change in Subjective Wealth Explanatory variables Relative income ∆land Relative ∆land ∆cereals ∆animal proteins ∆Housing ∆social events Relative ∆social events ∆Savings Mechanization Cotton experience Family size Age Resident ethnic group Opinion on income Opinion on poverty reduction Opinion on farm skills Opinion on input access Constants: 11 cut off values ρ Pseudo R² Observations Opinion on poverty reduction effect Relative income ∆land Relative ∆land ∆Savings Mechanization

BioProbit 2 for migrant ethnic groups

Significant* .067 (.040)* .020 (.013)

BioProbit 2 for resident ethnic groups .115 (.119) .007 (.162) .483 (.597) .030 (.123) -.183 (.097)* -.075 (.109) -.127 (.320) -.296 (1.198) .034 (.167) Adoption of animal farming* Not significant .007 (.031) -.004 (.013) .035 (.034)

Not significant -.009 (.036) -.009 (.017) -.235 (.305) .097 (.035)***

BioProbit 2 for non-poor households -.033 (.132) .108 (.208) -.353 (.770) -.123 (.132) -.000 (.105) -.049 (.121) -.939 (.255)*** 2.780 (.826)*** -.139 (.188) Adoption of animal farming** Not significant .007 (.033) .024 (.012)** -.665 (.243)*** .049 (.035)

.019 (.041) -.011 (.040) -.009 (.046) 4 cut off values are significant .476 (.095)*** .193 114

-.052 (.026)** -.092 (.034)*** 4 cut off values are significant .847 (.039)*** .108 179

-.072 (.038)* -.030 (.039) 5 cut off values are significant .387 (.057)*** .123 137

-.011 (.033) -.064 (.032)** 5 cut off values are significant .896 (.043)*** .176 156

.156 (.210) -.562 (.428) 2.042 (1.576) .005 (.241) Not significant

.247 (.196) -.873 (.277)*** 4.010 (1.256)*** .622 (.215)*** Adoption of animal farming* Significant*** -016 (014)

.335 (.201)* -.325 (.303) 1.392 (1.126) .010 (.137) -.082 (.125) -.104 (.119) -.851 (.251)*** 1.990 (.724)*** .221 (.179) Not significant

BioProbit 2 for poor households .980 (.397)*** -.106 (.253) 1.349 (.745)* -.292 (.129)** -.153 (.105) -.074 (.110) -.522 (.295)* 1.891 (.809)** .216 (.180) Not significant

.856 (.496)* .061 (.180) -.648 (.367)* -.866 (.271)*** 1.292 (1.420) 4.195 (1.169)*** .277 (.207) .584 (.2262)*** Traditional Not significant farming** Cotton experience Not significant Not significant Significant*** Age -.012 (.016) -006 (018) -038 (012)*** Resident ethnic group -.491 (.416) .245 (.297) Group relationships Not significant Not significant Not significant Not significant Group management Not significant Not significant Not significant Not significant Change in subjective wealth -.195 (.485) -1.465 (.702)** -.761 (.474)* -.608 (.533) Constants: 9 cut off values 2 last cut off values One cut off value is One cut off value is One cut off value is are significant significant significant significant Note: robust standard errors in parentheses, * is significant at 10 %, ** is significant at 5 %, *** is significant at 1 %. Poor households are the ones who are located below the income poverty line of 100,000 FCFA per consumption unit.

35

End Notes 1

The use of subjective data has regularly challenged economic theory. Several empirical evidences point out the shape of interdependent preferences and the complex impact of income from a reference group on individual utility (see for instances Graham and Pettinato, 2002; or Senik, 2004). For a full survey of issues, see Clark et al. (2007). 2 In spite of severe criticisms claiming that answers to subjective questions are pure noise, a cautious treatment of subjective data –subject to manipulation- (Bertrand and Mullainathan, 2001) has been proved to yield convincing results. 3 The improvement in living standards subsequently to a rise of income can occur with some delay as it requires some mid-term investment in infrastructures (housing, building schools and hospitals, deep wells, roads). Hence, the dynamic processes of living standards and income evolutions after the cotton reform are different and it is not very surprising that agricultural income could have risen without a significant improvement of living standards. 4 Cotton earnings are the main source of farm cash income for rural households of Southern Burkina Faso, and are also a factor of social prestige. 5 Food crops benefit from the remainder of mineral and/or organic fertilizers in soils formerly planted in cotton as well as from less sanitary problems. Cotton is known as a very good starting crop in a rotating crop system in many dried tropical agro-ecological systems. 6 These inputs are often delivered by cotton companies, through in-kind credit schemes repaid by cotton purchases from customers. Being a cotton grower is often the only way for rural producers to access agricultural inputs so that the availability of agricultural inputs through cotton growing reveals economic complementarities between cotton and other crops. 7 Notice that savings have more increased than investment. 8 These variables are self-assessed (subjective) evaluations of difficulties in sending children to schools and reaching at a satisfactory health state. 9 From the tradi-praticien (traditional healer) to the doctor or the nurse. 10 This relationship may arise because the main cotton reform effect on agricultural production relied on land extension. Furthermore, land holdings better explain the pattern of subjective wealth (as claimed by Van Landeghem et al., 2008) because they capture non-economic benefits, such as social prestige. 11 This approach of utility functions has also direct implication for poverty analysis, as in our case. Indeed, as Sen (1983) firstly argued, relative concerns such as relative consumption should be taken into account when setting a poverty line or measuring poverty. This would put together income levels and income profiles into the implementation of poverty measures. In the context of poverty, a lower rank on the wealth ladder means that the households will suffer from a lack of access to basic commodities, in case of a crisis. However, a richer neighbor might provide the latter with employment or aid. 12 Income is also driven by social effects and correlated with unobserved latent personality traits in subjective welfare or other missing variables. 13 As shown by Easterlin (2001), material aspirations matter for subjective well-being and the impact of income. If aspirations are the same along time, then income growth will make subjective well-being increase. But aspirations also increase with income and time, which makes experienced well-being different from expected one. 14 The use of ordinal variables is related to the assumption of ordinal utility made by economists. Working with cardinal well-being variables, as assumed in psychology, has been shown to have a limited incidence on the results (Ferrer-i-Carbonell and Frijters, 2004). 15 See the paper for more details about the model and the computation of joint probability pairs. 16 On average, log(land)=1.75 and relative(land)=1 so the positive absolute effect dominates the negative relative one, according to the estimates of table 6.

36