Ecological Economics 126 (2016) 112–124
Contents lists available at ScienceDirect
Ecological Economics journal homepage: www.elsevier.com/locate/ecolecon
Analysis
Forest clearing, livelihood strategies and welfare: Evidence from the Tsimane' in Bolivia Emilie Perge a,⁎, Andy McKay b a b
Earth Institute, Columbia University, 535 West 116th Street New York, NY, USA University of Sussex, Jubilee Building, University of Sussex, Falmer, Brighton BN1 9SL, UK
a r t i c l e
i n f o
Article history: Received 8 February 2015 Received in revised form 10 January 2016 Accepted 13 March 2016 Available online 17 April 2016 Keywords: Forest clearing Welfare Livelihoods Tsimane', Bolivia
a b s t r a c t This study analyzes the relationship between forest households' livelihood strategies, and forest clearing, and the relationship of both to welfare. The analysis relies on a rich panel dataset collected on the Tsimane' communities in Bolivia to highlight how forest households combine sales and wage activities. Forest clearing is associated with all livelihood strategies, but the association with welfare differs depending on the strategy pursued. Households that clear more and that generate their earnings by combining agricultural sales and wage activities are better-off (judged in terms of assets) than those undertaking other strategies. By contrast, households in a subsistence strategy are not able to accumulate assets in the long run. Overall, households clear only small areas of forest, which has a positive effect on welfare and enables accumulation of assets. © 2016 Elsevier B.V. All rights reserved.
1. Introduction A significant number of people across the world live in forest areas. Estimates of this vary from 300 million (WWF, 2014) to around 800 million (Chomitz, 2007), so between 4.3% and 11% of the world's population. Those living in forests are widely believed to be very poor and largely excluded from public services, in part because forest areas are typically remote and badly connected to the rest of the country (Sunderlin et al., 2005). But such households are typically not adequately captured in conventional approaches to assessing living standards such as household surveys. Even if these surveys do cover such populations, they typically only account for a small share of the sample, not allowing any significant conclusions to be derived about them (Sunderlin et al., 2007). Our current knowledge of living conditions in such environments mostly stems from anthropological and ethnographic studies (Godoy et al., 2007) or from one off specialist cross sectional surveys, such as those conducted as part of the Poverty Environment Network (PEN) coordinated by the Center for International Forestry Research (CIFOR). These studies provide important insights on questions of livelihoods and forest clearing among other issues.
⁎ Corresponding author at: Agriculture and Food Security Center, Earth Institute, Columbia University, 61 Route 9w, Palisades, NY 10964, USA. E-mail addresses:
[email protected] (E. Perge),
[email protected] (A. McKay).
http://dx.doi.org/10.1016/j.ecolecon.2016.03.017 0921-8009/© 2016 Elsevier B.V. All rights reserved.
In this paper we seek to add to this literature, by looking at longitudinal data to explore the links between livelihood strategies, forest clearing and welfare, as well as the heterogeneity within this. To do so we use a five wave panel dataset collected yearly between 2002 and 2006 on the indigenous Tsimane' communities in the Bolivian Amazon (Leonard and Godoy, 2008; Leonard et al., 2015). This survey data was complemented by a qualitative investigation conducted in the region by one of the authors. These data help identify four livelihood strategies based on households' reported sources of cash earnings (sale, wage, diversified or subsistence strategy). The data also provide a measure of clearing of old-growth and fallow forest, as well as different measures of household welfare or wellbeing (body-mass index, perceived happiness, and assets). We construct an asset index as a measure of welfare to avoid the limitations of consumption or income measures, especially in these environments (Sahn and Stifel, 2000; Günther and Klasen, 2007; Moser and Felton, 2007). In almost all cases households undertake agricultural activity as part of their livelihood strategy, with much of this being for sale even if some adopt a subsistence strategy. While agriculture is enabled by forest clearing, the extent of clearing is small in aggregate terms, and the majority of this is clearing of fallow forest. Some households combine this with working in wage activities. Over time, adults' body-mass index has not changed and if anything perceived welfare decreases slightly in 2006, probably due to the flood that affected the communities that year. Judging welfare in terms of assets, those with a diversified (sales and wage) strategy tend to be among the best off and those with a
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
subsistence strategy the worst off. Over the time period studied, most households have accumulated assets and so become better off. The analysis highlights the fact that households with the least assets accumulate most; and that asset-poor households with the highest growth rate are engaged in a wage strategy. Although these households accumulate assets faster than other households, they do not seem to reach levels of assets similar to households engaged in a sale or a diversified strategy. Wage activities in the research area combined with other sources of income are associated with higher levels of welfare. The remainder of this paper is structured as follows. Section 2 summarizes existing evidence on forest households' livelihoods, forest clearing and welfare, as well as providing a description of the research tools (data, outcome variables and empirical strategy). Section 3 discusses in descriptive terms households' livelihood strategies, their links to forest clearing and welfare, followed by a multivariate analysis of the factors affecting welfare as measured by assets in Section 4. A discussion of our findings and conclusions is presented in Section 5. 2. Conceptual and Empirical Framework 2.1. Empirical Evidence on Forest Households' Livelihoods and Welfare Studies of welfare can provide a great deal of information on the conditions in which households live. Looking at welfare over time allows one to assess what can help some households to become better off while others seem to stay in persistent poverty. In most microeconomic studies, welfare is assessed through consumption, income or assets. The first two measures though can be hard to measure and subject to volatility and seasonality in remote areas (Günther and Klasen, 2007), including forest environments. Persistent poverty seems then to characterize many communities and households in remote rural areas (Carter and Barrett, 2006; Barrett and Carter, 2013). According to these studies and others (Lybbert et al., 2004; Barrett and McPeak, 2006; Barrett et al., 2006; McKay and Perge, 2013), persistent poverty can take the characteristics of a poverty trap. In this situation welfare may be slowly improving (single equilibrium poverty trap) or there may exist non-linearities in the welfare improvement process that prevent households from reaching higher levels of welfare (multiple equilibria poverty trap) (Barrett and Carter, 2013). Applying this model of poverty trap to the studied population in the Bolivian Amazon, the Tsimane', previous studies have shown that households are likely to be ensnared in low levels of development accumulating slowly assets (Perge, 2010; McKay and Perge, 2013). But households commonly have diversified strategies; in addition to agriculture and environmental activities, households can diversify in livestock raising, wage or self-employed activities (Angelsen et al., 2014). Poor households rely more on forest resources than rich households even though the latter extract more products (Wunder, 2001; Fisher, 2004; Debela et al., 2012; Angelsen et al., 2014). In the studies conducted as part of the Poverty Environment Network (PEN) project, researchers found that in Latin America, forest income contributed to 28.6% of forest households' total income, being then the largest source of income followed by wage income. Through its design, the PEN collection of research studies took place over a single year providing explanations on intra-year variability but not on inter-year changes. The Tsimane' data used here help us to assess changes over time in terms of welfare, forest clearing and livelihood strategies. As of today, many studies listed in Leonard et al. (2015) have explained how the Tsimane' live. With respect to our question on welfare and its relationships to forest clearing and livelihood strategies, ethnobotanical and anthropological studies from this panel research study find that over time the Tsimane' have improved their welfare in terms of body-mass index and consumption (Godoy et al., 2009a, 2009b), that their agricultural
113
incomes have a positive effect on levels of forest clearing (Vadez et al., 2008; Godoy et al., 2009c), and that they are engaged in different activities (Godoy et al., 2007). Inspired by all this work, the present study aims to bring together models of welfare changes with specificities on forest households. Our study emphasizes the importance of knowing more about the relationships between welfare dynamics, livelihood strategies and forest clearing. Although we cannot infer causality from one factor to another, we want to provide evidence on relationships in a dynamic framework, using the Tsimane' Amazonian Panel Study data from 2002 to 2006 to do so. 2.2. Tsimane' Amazonian Panel Study (TAPS) 2002–2006 The Tsimane' population represent one of the largest indigenous population of the Lowlands of Bolivia with an indigenous territory of 330,000 ha (Chumacero, 2011 as cited in Paneque-Gálvez et al., 2013). In the early 2000s about 8000 adults were living in 100 remote communities in the Amazon rainforests of Beni Department mostly connected to each other via the Apere and Maniqui rivers (Appendix Fig. 1) (Reyes-García, 2001; Apaza et al., 2002; Reyes-García et al., 2014). The data come from the Tsimane' Amazonian Panel Study (TAPS) 2002– 2006, a panel dataset collected every year from May to October in thirteen villages, with an average sample size of 250 households, and nearly 1800 individuals.1 We have to acknowledge that the data analyzed are not necessarily representative of the whole Tsimane' population; the villages in the sample are relatively close to San Borja, the main market town, and so have easier interactions with the market economy than those in more remote locations. The data cover a series of indicators, including income sources and activities, assets, psychological status, health, and human capital holdings, enabling consideration of different aspects of the welfare of this population (Leonard and Godoy, 2008). Additional references on the numerous studies published with this panel data and its extended version (2002–2010) are available in Leonard et al. (2015). The activities covered include agriculture, forest-based activities and wage work. The Tsimane' are engaged in agricultural markets through the sales of rice and corn, in Non-Timber Forest Product (NTFP) markets through the sales of jatata (geonoma deversa a local thatch palm; Couvreur, 2011), and in wage activities when working for cattle ranchers, timber loggers, or as teachers. The TAPS team also collected forest clearing information asking households how much fallow and old-growth forest they had cleared in the last twelve months, the number of plots they cultivated, and the land used for rice or corn. In relation to welfare, the research team gathered a large amount of data on consumption, earnings, assets, anthropometric and psychological data, including heights and weights (sufficient to compute body-mass-index and height-forage). The analysis presented in this paper is based on the balanced panel of 176 households2 who were included in each wave between 2002 and 2006. In addition to these quantitative data, one of the authors undertook in summer 2008 a qualitative survey in nine villages, asking questions about households' uses of forest resources, their activities and insurance mechanisms, as well as questions about community organization and households' interactions with external actors. Historically autarkic, the Tsimane' have interacted more with the market economy through selling agricultural products and NTFPs,
1 The survey interviewed all people living in the villages in 2002 and the sampled adult population represented around 10% of total adult Tsimane' populations (Leonard et al., 2015). 2 The research team has collected data on around 250 households since 2002, but when keeping only households in all five years and after computing a price deflator using 2006 prices the sample reduced to 176 households. The following analysis relies on this number of observations.
114
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
and undertaking more wage work with cattle ranchers and timber loggers than they used to (Godoy et al., 1998; Apaza et al., 2002; Reyes-García et al., 2014). Lately these outsiders have been reported to enter illegally the indigenous territory intensifying tensions with the communities, destroying their crops and depleting timber species and game (Godoy et al., 1998). In addition to these human forces, environmental shocks have increased in the territory and in 2006, the Tsimane' villages were hit by the worst flood of early 2000s (Godoy et al., 2008).
one adult member has been happy in the past seven days (Appendix Table 2). In the following section we present a descriptive analysis of the above key variables in the panel data set which precedes a section modeling changes in welfare, assessed based on the asset index.
3. Descriptive Analysis: Socioeconomic Characteristics, Livelihood Strategies, Forest Clearing and Welfare 3.1. Tsimane' Socioeconomic Characteristics
2.3. Livelihood, Forest Clearing and Welfare Variables We constructed livelihood strategies depending on households' cash earnings from different sources as asked in the questionnaire. Three strategies are defined based on the relative balance of cash earnings from sales and from wages: those with more than 75% of their cash earnings from sales (a sale strategy); those with more than 75% from wages (a wage strategy); and those in between these two (a diversified strategy). A fourth strategy is then identified for those whose earnings are by barter (subsistence strategy). The cash earning data were related to the previous two weeks (to facilitate accurate response). As such the earning data may not be representative of the whole year; but as they were collected at the time period every year, we feel confident to compare the data over time. In relation to forest clearing, households were asked a direct question about the amount of forest they had cleared in the past twelve months. While self-reported, other analysis suggests the responses tend to be quite accurate (Vadez et al., 2003). In terms of measures of welfare, while the TAPS survey collects data on earned cash income of households and on consumption, we privileged other measures of welfare to avoid limitations of income and consumption measures. First the recall period for cash earnings is short; and anyway the survey does not enable consumption from own production to be reliably distinguished from overall consumption to enable computation of total income. Second, in the forest environment such monetary measures are likely to be highly volatile, creating potentially greater biases than other non-monetary alternatives as measures of living conditions (Günther and Klasen, 2007; Moser and Felton, 2007). Welfare measures based on assets or household nutrition are likely to be preferable, and are chosen here. We also use a measure of happiness collected by the TAPS research team to assess households' well-being. Nutrition is summarized through the body-mass index measures which indicate whether individuals have appropriate weight for their height for specific age categories and gender. The TAPS research team constructed the variables recording the height and weight of surveyed individuals (Godoy et al., 2009b). In relation to assets, as households can have many different types, we combine these into a single measure by constructing an asset index through factor analysis (Lawley and Maxwell, 1973; Friel, 2007; Perge, 2010; McKay and Perge, 2013). Our preferred index is inspired from Ellis's five capitals (physical, financial, natural, human, and social) (Ellis, 2001) and aggregates material and non-material assets employed in agricultural production, used for hunting and fishing, human capital through the ability to speak Spanish and mathematical skills, assets facilitating communication with the other communities and the external world, and assets facilitating interactions among Tsimane' households according to their culture (Table 1 in Appendix). However, we also examine an asset index based only on physical assets collected in all five waves to identify changes resulting only from an accumulation of material assets. We also separately assess changes in terms of human and social capital. These measures of physical, human and social capital are presented in Appendix Table 2. Perceived happiness was asked at the individual level in each year of the survey; we construct a dummy variable for whether at least
Shifting agriculture is the main source of livelihood of the Tsimane' population. In the sampled 13 communities, 157 households (out of 176) in 2004 and 161 in 2006 cultivate rice, with the corresponding numbers for corn being 87 and 82 respectively. Agriculture is based on cleared forest by slash-and-burn cultivation; between June and September the Tsimane' cut either fallow (secondary) or old-growth (primary) forest with simple tools (machetes, axes, more recently chainsaws) and let the vegetation dry before burning it (Vadez et al., 2003, 2004). Conversion is only temporary, for one or two years, after which another plot is cleared and the previous plot is left in transition for forest regeneration for at least five years before being cleared again (Godoy et al., 1998; Reyes-García, 2001; Vadez et al., 2003).3 The transition of the Tsimane' from a nomadic way of life to a sedentary one has pushed households to clear more fallow than old-growth forest since much of the latter has already been cleared near their homesteads; the Tsimane' usually cultivate land near their homes (Godoy et al., 2009c). When clearing a plot, the Tsimane' become an informal owner of the plot for the duration of cultivation; however once it is left back to fallow, any Tsimane' households can clear the plot or use products from the plot (Godoy et al., 2009c). Consequently, the Tsimane' do not accumulate land per se since every year they have to clear anew. Between 2002 and 2006 the Tsimane' households have cleared on average 1 ha of forest each year (0.15 ha per capita), opening around 0.6 ha of fallow forest and 0.4 ha of old-growth forest. Overall the extent of forest clearing is tiny relative to the total area of the forest (Killeen et al., 2007; Paneque-Gálvez et al., 2013) and the environmental impact of this activity is small compared to the impact of cattle ranching or large scale logging activities in the area.4 Further the process of clearing fallow forest is conducted in such a way that forest can still naturally regenerate (Vadez et al., 2003). The Tsimane' generate their earnings from sales or barter5 of agricultural products and NTFPs, from wage work, and marginally remittances. A majority of households sell (62.5%) and a small majority also participate in a wage activity (55%), the latter slowly increasing over time. In all five years wage earnings generate on average more cash income than sales. As stressed above we believe the earning data are sufficiently comparable to assess changes over time. Agricultural sales are mainly made to traders coming to the villages or through transporting to the nearest town, San Borja (Vadez et al., 2008); on average households sell around half of their rice harvest. The Tsimane' also sell jatata, but the majority sells this thatch palm to meet an upcoming expense (Reyes-García et al., 2012; Zycherman, 2013). Similarly, households mostly engage in wage work to cope with unexpected shocks and so stabilize their
3 Although the Tsimane' forest clearing may not be a huge driver of deforestation in the research area, Paneque-Gálvez et al. (2013) show that forest clearing by Tsimane' households increases levels of forest degradation. 4 However more recently, the Tsimane' have undertaken more logging activities in their territory to regain control over their lands and to generate cash earnings (Zycherman, 2013). 5 The survey uses village selling prices to find the values of bartered products.
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
income and consumption flows. The qualitative survey revealed that most households in the sample with a member working in a wage job report that they usually do these activities on a day-by-day basis, for a specific task, and sporadically when they need extra cash (see also Zycherman, 2013). Those few working more permanently in a wage activity tend to be young, single men without the responsibility of a household, so that they may live in the logging camp or at the ranch. Nonetheless, wage activities can contribute to a significant share of income for some households while others have more balanced sources of income (Godoy et al., 2007).
3.2. Households' Livelihood Strategies and Forest Clearing Using the livelihood classification explained above, during the first four years of the panel (2002–2005), more Tsimane' were engaged in a sale strategy than any other strategy; between 34 and 42% of the households obtained 75% or more of their cash earnings from sales (Table 1). Only in 2006 is the proportion of households undertaking wage work (32%) to generate most of their earnings greater than the proportion of households in any other strategies. This is likely to be a consequence of the flood in this year. Over the five years, between 16% and 22% of households have diversified earnings with a relatively equal contribution from wages and sales, and between 12% and 20% of the Tsimane' are engaged in a subsistence strategy. Through this classification we can distinguish cases where wage activities are implemented as a safety net mechanism (sale strategy) from cases where wage work is the main source of income (wage strategy). It is also clear that some households remain nonintegrated with the market consuming all they produce, and relying on barter to acquire non-produced goods.
Table 1 Livelihood strategies: participation and household characteristics (mean and standard deviation). Strategies 2002 Number of households Participationa(%) Household members age Household head school grade Household size 2004 Number of households Participation (%) Household members age Household head school grade Household size 2006 Number of households Participation (%) Household members age Household head school grade Household size
Sale
Wage
Diversified
Subsistence
67 38.0 24.5* (14.5) 1.7*** (1.9)
51 28.9 20.5 (13.2) 3.7*** (3.6)
29 16.5 18.5 (6.3) 1.8 (1.2)
24 16.5 21.9 (14.6) 2.2 (2.0)
6.4 (3.0)
6.7 (3.3)
7.2 (2.2)
6.4 (2.3)
64 36.6 22.6 (13.2) 1.7*** (1.7)
62 35.2 16.1*** (5.5) 4.1*** (3.4)
28 15.9 22.0 (12.0) 1.5** (1.6)
22 12.5 31.8*** (24.6) 1.5* (1.7)
6.2 (2.8)
6.6 (2.4)
7.6** (2.7)
51 28.9 25.8* (17.7) 1.9* (1.9)
56 31.8 18.4** (7.7) 3.4*** (3.3)
6.3 (2.9)
7.2 (2.8)
115
With respect to their characteristics, households engaged in a wage strategy have younger and more educated household members and heads than other households. Wage work requires young people able to perform strenuous tasks for a ranch or a timber logging company, or more educated people working as a teacher for example. Households engaged in a diversified strategy are larger on average; having more household labor may allow them to participate in a more diversified portfolio of activities. Looking at mobility from one year to the next, many households engaged in sale and wage in 2002 remain in this activity at the end of the period in 2006. For instance between 2002 and 2006, 46% of households in a sale strategy in 2002 are still in this strategy in 2006 and 53% of households in a wage strategy in 2002 are in the same strategy in 2006. 41% of the households in a subsistence strategy in 2002 are still in this strategy in 2006. Households in a diversified strategy are much more likely to have switched between 2002 and 2006, either to a wage strategy (41% between 2003 and 2004) or to a sale strategy (53% between 2004 and 2005). This is not surprising given how this strategy was defined. But the relative stability in other livelihood categories between 2002 and 2006 does not mean that households stayed in the same activity for all five years; only 7 and 6% of the Tsimane' have stayed in a wage and sale strategy respectively for all five years. Households engaged in a subsistence strategy are mobile between 2003 and 2004 and between 2004 and 2005, with on average only 13% of them staying in this strategy during these periods. 27% of households engaged in a diversified strategy are still in this strategy from one year to the next except between 2005 and 2006 when this proportion rises to 40%. Households in a diversified strategy switch either to a wage strategy (41% between 2003 and 2004) or to a sale strategy (53% between 2004 and 2005). Over the 5 years, 45% of the households engaged in a diversified strategy in 2002 have switched to a wage activity in 2006. Agriculture remains the dominant or substantial activity in all cases except where households engage in a wage strategy; but even in this last category all households also engage in agriculture. In each livelihood strategy, the extent of forest clearing varies. Table 2 shows the relationship between forest clearing and livelihood strategies. Households engaged in a wage or subsistence
Table 2 Fallow, old-growth and number of plots (mean and sd; ha). Strategies
Sale
Wage
Diversified
Subsistence
5.9 (2.9)
2002 Fallow forest Old-growth forest Number of cultivated plots
0.60* (0.67) 0.51 (1.28) 1.7 (0.89)
0.37** (0.36) 0.37 (0.47) 1.5 (0.87)
0.67** (0.61) 0.27 (0.49) 1.5 (0.82)
0.36 (0.5) 0.50 (0.53) 1.6 (0.99)
34 19.3 21.1 (7.8) 1.5** (2.0)
35 19.9 24.9 (16.7) 2.9 (3.2)
2004 Fallow forest Old-growth forest Number of cultivated plots
0.64 (0.66) 0.29 (0.45) 1.5 (0.87)
0.51 (.50) 0.30 (0.42) 1.5 (0.72)
0.61 (0.57) 0.57*** (0.70) 1.8* (0.86)
0.45 (0.42) 0.32 (0.50) 1.2** (.81)
7.7** (2.5)
5.9** (2.5)
2006 Fallow forest Old-growth forest Number of cultivated plots
0.73 (0.82) 0.49 (0.63) 1.5 (0.81)
0.62 (0.55) 0.48 (0.67) 1.6 (0.87)
0.75 (0.79) 0.47 (0.68) 1.8* (0.94)
0.48 (0.53) 0.38 (0.52) 1.4 (0.81)
% of households in 2006 after being in a […] strategy in 2002 Sale 46.2 14.9 23.9 Wage 11.7 52.9 15.7 Diversified 27.6 44.8 17.2 Subsistence 20.7 20.7 17.2
14.9 19.6 10.3 41.4
Test of significance against the whole sample *p b 0.1, **p b 0.05, ***p b 0.01. Note: We are testing against the whole sample because we are interested in testing how different households engaged in one strategy were from households engaged in all other strategies. Testing against the sale category for each remaining strategy brings similar results. a Participation represents the percentage of households in each strategy.
% of households with increases in […] between 02 and 06 Forest clearing 63% 71% 64%
60%
Test of significance against the whole sample *p b 0.1, **p b 0.05, ***p b 0.01. Note: We tested against the whole sample so as to measure how different households engaged in one strategy were from households engaged in all other strategies. We tested significance of these variables for households in one strategy against the ones in a sale strategy and we found similar results.
116
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
strategy clear less forest in all years than those engaged in a sale or diversified strategy, confirming that these latter groups are likely to be those with the highest agricultural production. The Tsimane' engaged in a subsistence strategy have lower forest clearing than other households; and households in a subsistence strategy in both 2002 and 2006 have decreased their forest clearing between these two years. Forest clearing appears to be important for households across all strategies but it has a distinct role; it supports all livelihood generation in a sale strategy but has a smaller effect on agricultural production and sales in a wage strategy. 3.3. Assessment of Tsimane' Welfare Welfare outcomes for the households in our sample show that Tsimane' adults (male and female) have an average BMI of 23 kg/cm2 which is not low according to World Health Organization standards (Appendix Table 2). Overall the Tsimane' households do not seem to suffer from malnutrition and there is some evidence that the children are getting better-off (Godoy et al., 2009b). With respect to happiness, in the first years of the panel a large majority of households have at least one member happy in the last seven days. In 2006, 88% of households have at least one member feeling happy in the last seven days. In the sample between 2002 and 2006, households seem to feel less happy than at the beginning, though the flood in 2006 may have contributed to this (Appendix Table 2). Over this period the asset index value has increased6 (Table 2 in Appendix). If one was to define a cut-off (or poverty threshold) at the 40th percentile of the distribution of the asset index in 2002, there is a significant reduction over time in the percentage of the population lying below this level, from 40.1% in 2002 to 22.7% in 2006. A poverty trap test developed by Perge (2010) and McKay and Perge (2013) shows that the Tsimane' are engaged in a smooth (if slow) asset accumulation dynamic showing an improvement in welfare. Distinguishing households with respect to their livelihood strategy, the average BMI values are very similar across strategies and change very little over time. As previously noted the average levels for adults are quite high in all cases (Table 3). In terms of happiness, households across strategies seem to have similar levels of happiness. More variation though is seen between strategies and over time in the average level of the asset index as well the percentage lying below the 40th percentile of the asset index. The average value of the asset index is highest, and the percentage below the 40th percentile is lowest, for households engaged in a sale strategy or a diversified strategy compared to the others. The average value of the asset index is highest, and the percentage below the 40th percentile is lowest, for households engaged in a sale strategy or a diversified strategy compared to the others. The average asset index is lowest, and the percentage below the 40th percentile is highest, in the subsistence strategy, where the overall level of production is likely to be lower. There are clear improvements over time for those engaged in the diversified strategy, and these indicators are much better in 2006 than in 2002 for those engaged in a sale strategy. This latter point is true to a lesser extent for the wage strategy, but not for those in the subsistence strategy. In addition, switching from any strategy to a diversified strategy in 2006 allows households to have more assets and to be less likely to be
6 In appendix, we look at an asset index composed only of physical assets; we can see that households are accumulating physical assets over time. We report also measures of human and social capital and we can see that over time households are less social capital but the same levels of human capital (Appendix Tables 1 and 2).
Table 3 Adult women and children BMI, household happiness, asset index and asset poverty (mean and sd). Strategies 2002 Adult women BMIa Adult men BMI Household happinessb (%) Asset indexc Asset povertyd (%) 2004 Adult women BMI Adult men BMI Household happiness (%) Asset index Asset poverty (%) 2006 Adult women BMI Adult men BMI Household happiness (%) Asset index Asset poverty (%)
Sale
Wage
Diversified
Subsistence
22.8* (2.7) 22.6 (2.2) 94.0 (23.8)
23.2 (2.8) 23.3 (2.4) 94.1 (23.7)
23.6 (2.8) 22.3 (2.2) 89.6 (31.0)
23.8 (2.30) 24.0 (2.5) 96.5 (18.5)
1.95 (1.02) 34.3 (47.8)
1.92 (1.22) 41.1 (49.7)
1.94 (0.96) 37.9 (49.4)
1.65 (0.94) 48.2 (50.8)
23.6 (3.2) 22.6 (2.2) 95.3 (21.3)
23.5 (2.6) 23.6 (2.7) 98.4 (12.7)
22.3* (2.4) 22.7 (2.0) 92.8 (26.2)
21.6** (3.6) 22.6 (2.3) 95.4 (21.3)
1.88 (0.90) 35.9 (48.3)
1.77 (0.72) 38.7 (49.1)
2.04*** (1.42) 21.4* (41.8)
1.68 (1.18) 50.0 (51.1)
23.0* (3.2) 22.9 (2.2) 84.3 (36.7)
24.0 (3.2) 23.1 (3.2) 92.8 (26.0)
24.3 (2.2) 22.9 (1.7) 97.0* (17.1)
23.3 (3.1) 23.0 (2.3) 77.1** (52.6)
2.43 (1.05) 19.6 (40.1)
2.20 (0.84) 28.7 (45.6)
2.95*** (1.18) 5.9* (23.9)
1.80*** (0.94) 45.7** (50.5)
% of households with increases in […] between 02 and 06 Asset index 66% 71% 75%
63%
Test of significance against the whole sample *p b 0.1, **p b 0.05, ***p b 0.01. Note: We tested against the whole sample so as to measure how different households engaged in one strategy were from households engaged in all other strategies. We tested significance of these variables for households in one strategy against the ones in a sale strategy and we found similar results. a Individual body-mass index in kg/cm2. b % of households with at least one member feeling happy in the last seven days. c Asset index including social and human capital. d % of households below the fourth percentile of the distribution of asset index values in 2002.
asset-poor compared to other switches. Those generating most of their income from wage activities do not have as many assets as households in other strategies; these activities are mostly low-skilled, strenuous activities. It seems then that adopting one livelihood strategy rather than another is associated with specific levels of both clearing and welfare. As a consequence, forest clearing would have specific linkages with welfare depending on the strategy household is engaged in. The following section now analyzes more carefully the links between clearing and livelihood strategies on welfare changes over time taking into account the panel structure of the data.
4. Factors Affecting Welfare Changes 4.1. Empirical Strategy The results above have shown that nutrition appeared to be stable over time, as generally was the case for happiness, apart from its reduction in 2006. Happiness may anyway be linked to individuals' characteristics hard to summarize in an econometric model. Our analysis therefore focuses on changes in the asset index. Using the panel structure of the data, we estimate the changes over time in the asset index, either from one year to the next or from beginning to end, as a function of the previous level of the asset index, household characteristics, livelihood strategies, clearing of old-growth and fallow forest, and when available shocks and coping mechanisms. In different specifications we use lagged, current or baseline values of the asset index and forest clearing. We estimate this first
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
as a log-linear specification (for simplicity, subscripts i and t are omitted). K
ΔY ¼ α þ β y þ γ FC þ ∑ φk Xk þ δ T þ μ k¼1
in which ΔY is the change in asset index; y is value of asset index at baseline (t-1) or 2002 in the case of beginning-to-end changes; FC stands for forest clearing and is a composite of old-growth (OF) and fallow (FF) forest; X are household characteristics and T are time dummies (only for yearly changes). But to allow for non-linearities in asset accumulation such as the ones assumed when testing for a poverty trap (Lybbert et al., 2004; Barrett and McPeak, 2006; Carter and Barrett, 2006; McKay and Perge, 2013) and in testing for an environmental Kuznets curve (Celentano et al., 2012), the models estimated are polynomials taking the following form: 3
3
K
a¼1
b¼1
k¼1
ΔY ¼ α þ ∑ βa ya þ ∑ γb FCb þ ∑ φk Xk þ δ T þ μ:
Table 4 Summary statistics for variables used in models (1), (2), (3) and (4) (mean and sd). Variables
Definition
Mean (s.d)
Yearly changes in asset index 0.11 (0.88) ΔAIt-1 (model 1/3) ΔAI02-06 (model 2/4) Changes in asset index 2002–2006 0.43 (1.06) Explanatory variables Asset index AIt-1 Lagged values of asset index 1.97 (1.08) AI02 Values of asset index in 2002 1.89 (1.06) sqAIt-1 Squared lagged values of asset index 5.07 (6.24) sqAI02 Squared values of asset index in 2002 4.70 (5.90) cub AIt-1 Cubic lagged values of asset index 16.38 (36.10) cubAI02 Cubic values of asset index in 2002 14.81 (31.96) Forest characteristics Ag plots Number of agricultural plots 1.68 (0.97) FF Size of fallow forest clearing (ha) 0.59 (0.59) OG Size of old-growth forest clearing 0.42 (0.64) sqFF Squared size of fallow forest clearing (ha) 0.70 (1.32) sqOG Squared size of old-growth forest 0.59 (3.53) clearing (ha) cubFF Cubic size of fallow forest clearing (ha) 1.13 (3.46) cubOG Cubic size of old-growth forest clearing 1.85 (33.90) Livelihood strategies Diversified Dummy for whether household in 0.17 (0.38) diversified strategy Wage Dummy for whether household in 0.30 (0.46) wage strategy Subsistence Dummy for whether household in 0.16 (0.37) subsistence strategy Ag diversification Dummy for whether household plant 0.54 (0.50) plantain and cassava Household characteristics Gender head Dummy for whether head is male 0.96(0.182) Age head Age of household head 43.94 (16.95) Sq age head Squared age of household head 2218.30 (1689.99) HH size Size of household 6.70 (2.85) Children Number of children 3.18 (2.52) Credit Dummy for whether household 0.65 (0.47) received credit past 2 weeks Travel SB Dummy for whether household 0.41 (0.49) traveled to San Borja past 7 days Shocks & coping mechanisms (2002–2006 model only) Shock flood Dummy for whether household faced 0.31 (0.46) shock from flood in 2004, 2005, or 2006 Shock health Dummy for whether household faced a 0.59 (0.49) shock from health in 2004, 2005 or 2006 0.44 (0.50) Pub assistance Dummy for whether household received public assistance after 2006 flood
117
Summary statistics for the variables used in the regression are presented in Table 4. Based on earlier research on poverty traps, we expect lagged, baseline, values of asset index to have a negative sign, meaning that asset-poor households accumulate more assets than asset-rich households and providing some evidence of convergence. With respect to forest clearing, we assume that size of cleared forest is positively linked to changes in asset index but at a decreasing rate (Celentano et al., 2012). In relation to livelihood strategies, a positive effect might be associated with wage and diversified strategies and a negative one with the subsistence strategy. We assume that the variable “ag diversified” has a positive link to asset change; households diversifying their agricultural production are more likely to have increased their holdings of assets. Data on shocks in the previous year are available only for the latest years (2004–05–06) and for these years we take into account covariant (flood) and idiosyncratic (health) shocks. We also construct a dummy for public assistance using households' responses on assistance they received in 2006. In both types of models we try current and lagged values of forest clearing and livelihood strategy. Reflecting the panel structure of the data, we use a household fixed-effect to control for unobserved household characteristics when estimating yearly changes in asset index with current and lagged values, including also time fixedeffects to control for unexpected events that could affect the outcome. Changes between 2002 and 2006 are estimated using ordinary least-squares (OLS) with village dummies. To check the robustness of our results we estimate the same models with the asset index composed only of physical assets (Appendix Tables 3 and 4). As already stated and as confirmed by Fig. 1, households have accumulated assets between 2002 and 2006. In this figure, most of the points are above the 45 degree line highlighting that most households have more assets in 2006 than in 2002. Fig. 1 also shows that asset accumulation varies according to the 2002 livelihood strategy of the household. As explained above, we model both yearly changes (models 1 and 3) and changes between 2002 and 2006 (models 2 and 4). In models 1 and 2, we use lagged/baseline values of asset index as well as current values of forest clearing, livelihood strategies, and credit. In models 3 and 4, we use lagged/baseline values of asset index, forest clearing, livelihood strategies, and credit. We start first with a simple log-linearized model of asset changes (Table 5) before extending these models to include polynomials for asset and forest clearing (Table 6). 4.2. Models With Log-Linearized Distribution of Asset Index Table 5 reports that lagged and baseline asset values have a significant negative effect on asset changes over time; households with the least assets are able to accumulate more than households with more assets. This result is robust over the different models. Forest clearing in the current period (models 1 and 2 in Table 5) does not have any significant effect on households' yearly changes except for agricultural plots; households with larger agricultural plots in current period have been able to increase more the value of their asset index. For changes between 2002 and 2005, clearing of fallow forest in 2006 is positively linked to asset changes between 2002 and 2006. With respect to livelihood strategies in the current period (models 1 and 2 in Table 5), our most important result is that households engaged in a subsistence strategy have been less likely to change their level of asset index both from one year to the next or between 2002 and 2006. With respect to households' characteristics, the age of household head does not seem to have any effect on households' changes of asset levels. Between 2002 and 2006, households with a male head are more likely to have changed their asset holdings between 2002
118
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
Fig. 1. A scatterplot of asset holdings 2002 and 2006 in 2002 livelihood strategies.
and 2006; since most households are headed by a male, this confirms that households have improved their levels of asset from the beginning of the survey and its end. Household size across all models has a positive relationship to changes in the asset index; larger households have changed most their levels of assets over time. Between
2002 and 2006, households with more children have been less able to change their asset levels (Table 5). Finally, households with access to credit in the current year are more likely to have changed their levels of asset index from one year to the next compared to households without any credit.
Table 5 Log-linearized regressions of asset changes: yearly changes and 2002–2006 changes. Variables
lnAI Ag plots FF OF Diversified Wage Subsistence Ag divers Gender head Age head Sq age head Household size Children Credit Travel SB Shock flood Shock health Pub assistance Dummy 2003 Dummy 2004 Dummy 2005 Dummy 2006 Constant Observations R-squared Number of hhid Rho Sigma_u Sigma_e
Yearly changes (FE)
2002–2006 Changes
Yearly changes (FE)
2002–2006 Changes
(Model 1)a
(Model 2)b
(Model 3)c
(Model 4)d
−1.219*** (0.053) 0.035* (0.020) 0.001 (0.004) 0.001 (0.004) 0.041 (0.039) 0.024 (0.048) −0.153** (0.068) 0.012 (0.035)
−0.722*** (0.063) 0.040 (0.051) 0.011* (0.006) 0.010 (0.006) 0.075 (0.089) −0.003 (0.079) −0.180* (0.092) 0.005 (0.065) 0.324** (0.146) 0.009 (0.011) −0.000 (0.000) 0.173*** (0.033) −0.183*** (0.039) −0.033 (0.063) 0.081 (0.063) 0.104 (0.067) −0.004 (0.061) 0.006 (0.063)
−1.198*** (0.058) 0.032 (0.021) −0.002 (0.004) −0.005 (0.003) 0.057(0.053) 0.075 (0.062) 0.056 (0.050) 0.013 (0.038)
−0.692*** (0.064) 0.064 (0.045) 0.001 (0.007) 0.010*** (0.003) 0.147* (0.087) −0.050 (0.079) −0.069 (0.090) −0.045 (0.068) 0.398*** (0.140) 0.005 (0.011) −0.000 (0.000) 0.230*** (0.028) −0.242*** (0.035) −0.166** (0.069) 0.008 (0.062) 0.121* (0.064) 0.008 (0.060) −0.058 (0.063)
0.005 (0.011) −0.000 (0.000) 0.128*** (0.016) −0.011 (0.011) 0.086** (0.034) −0.032 (0.052)
0.000 (0.000) 0.023 (0.050) 0.143*** (0.038) 0.258*** (0.047) −0.440 (0.292) 585 0.659 176 0.741 0.509 0.301
−0.659** (0.284) 136 0.636
0.010 (0.011) −0.000 (0.000) 0.134*** (0.019) −0.017 (0.011) −0.013 (0.044) −0.037 (0.055)
0.000 (0.000) −0.018 (0.052) 0.149*** (0.042) 0.243*** (0.045) −0.490 (0.309) 585 0.636 176 0.731 0.520 0.315
Models 2 and 4 include village dummies. Robust standard errors in parentheses *p b 0:1, **p b 0:05, ***p b 0:01. a Model 1 includes lagged values for asset index and current ones for forest clearing, for livelihood strategies and credit. b Model 2 includes 2002 values for asset index and current ones for forest clearing, livelihood strategies and credit. c Model 3 includes lagged values asset index, forest clearing, livelihood strategies and credit. d Model 4 includes 2002 values for asset index, forest clearing, livelihood strategies and credit.
−0.609** (0.281) 136 0.661
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
119
Table 6 Polynomial regressions of asset changes: yearly changes and 2002–2006 changes. Variables
AI SqAI CubAI Ag plots FF SqFF CubFF OF SqOF CubOF Diversified Wage Subsistence Ag divers Gender head Age head Sq age head Household size Children Credit Travel SB Shock flood Shock health Pub assistance Dummy 2003 Dummy 2004 Dummy 2005 Dummy 2006 Constant Observations R-squared Number of hhid Rho Sigma_u Sigma_e
Yearly changes (FE)
2002–2006 Changes
Yearly changes (FE)
2002–2006 Changes
(Model 1)a
(Model 2)b
(Model 3)c
(Model 4)d
−1.415*** (0.217) 0.095 (0.072) −0.009 (0.007) 0.094** (0.046) 0.032 (0.302) 0.130 (0.317) −0.048 (0.084) 0.495* (0.279) −0.487* (0.278) 0.109* (0.062) 0.141 (0.087) 0.070 (0.091) −0.182* (0.099) 0.056 (0.061)
−0.641 (0.391) −0.021 (0.157) −0.003 (0.0176) 0.029 (0.106) 0.582 (0.414) −0.182 (0.340) 0.022 (0.074) −0.065 (0.507) 0.368 (0.530) −0.081 (0.116) 0.331* (0.173) −0.015 (0.155) −0.153 (0.187) 0.170 (0.134) 0.228(0.245) −0.008 (0.022) 0.000 (0.000) 0.434*** (0.068) −0.448*** (0.077) −0.0399 (0.126) 0.154 (0.131) 0.275* (0.161) −0.009 (0.103) −0.097 (0.110)
−1.402*** (0.231) 0.100 (0.075) −0.010 (0.008) 0.050 (0.046) −0.010 (0.295) 0.081 (0.293) −0.032 (0.075) −0.014 (0.163) −0.030 (0.084) 0.001 (0.007) 0.077 (0.105) 0.239** (0.106) 0.109 (0.101) 0.013 (0.076)
−0.075 (0.496) 0.009 (0.212) −0.004 (0.024) −0.070 (0.147) 0.511 (0.719) −0.074 (0.679) −0.043 (0.178) 0.436 (0.576) 0.091 (0.375) −0.014 (0.032) 0.098 (0.300) −0.091 (0.213) 0.033 (0.216) −0.244 (0.232) −0.060 (0.370) −0.022 (0.0230) 0.000 (0.000) 0.314*** (0.082) −0.363*** (0.098) −0.245 (0.222) −0.174 (0.195) 0.146 (0.201) −0.202 (0.153) −0.080 (0.197)
0.034 (0.022) −0.000 (0.000) 0.274*** (0.035) −0.036 (0.025) 0.167** (0.065) 0.004 (0.109)
0.000 (0.000) −0.002 (0.100) 0.228*** (0.079) 0.426*** (0.085) −0.491 (0.539) 586 0.642 176 0.735 0.957 0.575
0.048 (0.519) 137 0.758
0.043** (0.021) −0.000** (0.000) 0.290*** (0.038) −0.051** (0.025) −0.039 (0.080) −0.026 (0.116)
0.000 (0.000) −0.151 (0.099) 0.191** (0.081) 0.358*** (0.084) −0.436 (0.543) 587 0.629 176 0.735 0.995 0.597
0.506 (0.789) 135 0.382
Models 2 and 4 include village dummies. Robust standard errors in parentheses *p b 0:1, **p b 0:05, ***p b 0:01. a Model 1 includes lagged values for asset index and current ones for forest clearing, for livelihood strategies and credit. b Model 2 includes 2002 values for asset index and current ones for forest clearing, livelihood strategies and credit. c Model 3 includes lagged values asset index, forest clearing, livelihood strategies and credit. d Model 4 includes 2002 values for asset index, forest clearing, livelihood strategies and credit.
Households frequently traveling to San Borja are not significantly more likely to have changed their asset levels over time. Time dummies show that in 2005 and 2006, changes in levels of asset index were significantly larger than in 2003 and 2004. Using lagged/baseline values for forest clearing, livelihood strategies and credit gives in most instances similar results than when using current values for these variables except that clearing of fallow forest is not significant anymore. On the other hand, in model 4 clearing of old-growth forest is significant and households having cleared more old-growth forest in 2002 have larger changes in the levels of their asset index. Households who in 2002 cleared more old-growth forest may still enjoy benefits from newly cleared forest. With respect to livelihood strategies, being in a subsistence strategy the year before (model 3) or in 2002 (model 4) does not have any longer any significant effect on asset changes; only households engaged in a diversified strategy in 2002 have significantly changed their levels of asset index between 2002 and 2006. Household characteristics identified as significant in models 1 and 2 remain significant in models 3 and 4. Quite unexpectedly households having received credit in 2002 (model 4) have smaller changes between 2002 and 2006. The impact of facing shocks does not have a significant impact on asset accumulation at the 5% level. In addition, controlling for whether or not households received public assistance does not affect
these results.7 In the Appendix Table 3, estimations using the asset index only composed of physical assets show that the results hold; households accumulate assets over time. In addition, only for changes between 2002 and 2006 human capital seen through literacy has a positive effect on asset accumulation; social capital has on the contrary no effect. 4.3. Cubic Regressions With Asset Index and Forest Clearing As an extension to these models, we develop non-linear cubic functions to control for non-linearities in the asset accumulation process as suggested in tests for poverty traps (Barrett et al., 2006; Carter and Barrett, 2006) and tests for the forest clearing boomand-bust hypothesis as in Celentano et al. (2012). Although forest cover remains in the Tsimane' territory, to some extent old-growth forest has been overly degraded around villages (Paneque-Gálvez et al., 2013) which suggests that there could be such a boom-and7 There is a positive effect of having faced a flood shock on asset accumulation between 2002 and 2006 in model 4. While perhaps unexpected, there is also no obvious reason why a flood should have an adverse effect on accumulation of most of the assets considered here (as opposed to income for example). Those areas more affected by the flood may have been better connected and more easily able to accumulate assets. The interviews revealed that floods mainly affected riverine land that are sometimes cultivated but did not touch villages and homesteads where assets are stored.
120
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
bust episode happening. Table 6 reports results for yearly changes and 2002–2006 changes with current values of forest clearing, livelihood strategies and credit (models 1 and 2) as well as same changes but with lagged/baseline values of these variables. R-squared values in Table 5 and Table 6 are nearly equal and explain a fair share of the variance. It is interesting to report that as in Perge (2010) and McKay and Perge (2013) there does not seem to be non-linearities in the accumulation of assets; only first-order lagged or baseline values of asset index have the expected negative effects on changes in asset index. Second-order values of asset index have a positive, but not significant, relationship which would show that asset-poor households are changing the most their levels of asset index at a positive rate. Also we can point out that for 2002–2006 changes (models 2 and 4), once we control for non-linearities, the baseline values of asset index do not have any effect on changes; over a large period of time, effects seem to follow a convergent, linear, process. With respect to forest clearing, only for yearly changes current levels of clearing of old-growth forest have the assumed signs (model 1 in Table 6): clearing more old-growth forest has a positive link with changes in asset index but at a decreasing rate. However a certain level of clearing seems also to have a positive link with changes in asset index (Celentano et al., 2012). This result can to some extent be explained that Tsimane' households have cleared large amount of old-growth forest which has been replaced now by newly regenerating forest (Paneque-Gálvez et al., 2013). However we can only identify this effect for yearly changes with current values of forest clearing (model 1); this effect does not appear in the other three models. Current participation in diversified or subsistence strategies (models 1 and 2) shows that the former has a positive relationship on 2002–2006 changes while the second has a negative one on yearly changes. Also it seems that participating in a wage strategy the year before has a positive effect on changing levels of asset index (model 3 Table 6). Household size and the age of household head have a positive relationship with asset changes: households with more members are able to change most their asset levels and so do households with older head. On the other hand, again households with more children are less able to change their levels of assets. Again we find the positive relationship between credit and yearly changes showing that households with credit in the current period have been able to change most their levels of assets. We again find that households affected by flooding accumulate assets between 2002 and 2006 and that receiving public assistance again does not have any effect. Developing the same models for the asset index composed only of physical assets give the same results and report that human and social assets do not have any effects on asset accumulation (Table 4 in Appendix). In all models, the baseline asset value (except for non-linear models of 2002–2006 changes) has the expected significant negative effect: asset-poor households are the most able to change their levels of asset index. This finding corroborates the slow growth process found when testing for a poverty trap (Perge, 2010; McKay and Perge, 2013) and economic, health and psychological improvements found in Godoy et al. (2009b). The results on forest clearing are also interesting and show how higher levels of fallow and old-growth forest are linked to higher levels of welfare for households. Livelihood strategies have specific relationships to welfare: households engaged in a diversified strategy and to some extent in a wage have been more able to change their levels of assets between 2002 and 2006 or from one year to the next. On the contrary being engaged in a subsistence strategy is negatively linked to changes in levels of asset index: subsistence strategy cannot be linked to welfare improvements.
In what follows we further discuss all the results we obtained from these models and the descriptive analysis before providing a short conclusion and policy recommendations.
5. Discussion and Conclusion The TAPS panel data is a rich dataset providing a lot of information on households' livelihood strategies, forest clearing and welfare. Previous studies referred to above have considered the impact of income on forest clearing; here the reverse relationship is considered. Welfare dynamics confirm that the Tsimane' have improved their welfare over time as it was found in Godoy et al. (2009b); Perge (2010) and McKay and Perge (2013). Over time households have more assets and there is a slow growth process with poorer households accumulating more assets than rich households. Only because households' welfare levels are low, it is reasonable to assume that households won't escape poverty over their lifetime. With respect to their livelihoods, agriculture is the main source of food for forest households even for those engaged in wage activities. Forest clearing is the means of enabling agricultural production, which plays a key role in almost all household's livelihood strategies; and more forest clearing seems to enable higher levels of welfare to be achieved. Those reliant mainly on wage activities also clear quite large areas of forest (they also farm). Those in a subsistence strategy clear less, but they are also worse off than other households. Forest clearing in this research area has a positive effect on welfare and could potentially help households to escape poverty. What is crucial to keep in mind is that in the research area, households clear still small areas of forest. From this analysis, livelihood strategies matter and are linked to different levels of welfare and welfare changes over time. Forest households are engaged to different degrees in sales of agricultural and non-timber forest products, and wage activities. Being engaged in a diversified strategy has a positive effect on asset changes in all specifications. Combining both types of earnings is the most effective way of raising their assets, though those engaged in a sale strategy are also able to increase their assets significantly. Those households not engaged in the market and reliant on their production and on barter to acquire goods they cannot produce are the poorest of all and accumulated least assets. Overall, agriculture based on small amounts of forest clearing remains an important source of livelihood for the Tsimane' households. The limited good wage opportunities reduce their capacity to improve welfare and combining both agriculture and wage activities seem to be the best option they have to improve welfare. However even if for the time being, forest regeneration is still happening and forest clearing is not a major issue, the nature of available wage activities is more worrying. Cattle rancher or timber logging companies are more environmentally destructive than small agricultural production and it is important to find ways of developing more sustainable and environmentally-friendly wage activities or markets so as to encourage forest households to move away from degrading wage activities and not to increase forest clearing beyond forest capacity.
Acknowledgments We are very grateful for helpful comments on an earlier draft from Ricardo Godoy and Stefano Pagiola as well as from four anonymous referees. We would like to thank the TAPS team for the data and their assistance throughout the fieldwork and the DFID (UK) funded Chronic Poverty Research Center in Manchester for financial support to undertake the fieldwork. Usual disclaimers apply; all errors remain ours only.
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
121
Appendix A. Appendix
Appendix Fig. A. Tsimane' settlements in the Beni Department, Bolivia. Source: Reyes-García et al. (2014).
Appendix Table 1 Description of assets included in asset index (pooled data). Variable
Description
Mean (s.d)
Ax Bike Bow Canoe Cow Hook Knife Machete Mosquito net Net Radio Rifle Shotgun Gifts Spanish Maths
Number of axes used for agriculture and timber logging Number of bikes used to go to market to sell NTFPs and agricultural products Number of bows used for hunting Number of canoes used for fishing and to go to market to sell NTFPs and agricultural products Number of cows owned by households Number of hooks used for fishing Number of knives used for hunting, fishing and agriculture Number of machetes used for hunting, agriculture and NTFPs Number of mosquito nets used as first protection against insects and snakes Number of nets used for fishing Number of radios used to communicate between communities, with traders and with markets Number of rifles used for hunting Number of shotguns used for hunting Dummy for whether households give to other households Number of household members speaking Spanish Dummy for whether households have a member having maths skills
1.40 (0.98) 0.35 (0.73) 1.61 (1.38) 0.47 (0.69) 0.48 (2.09) 5.47 (3.57) 3.50 (2.28) 3.51 (1.96) 4.19 (2.28) 0.79 (1.02) 0.89 (0.87) 0.51 (0.59) 0.42 (0.56) 0.59 (0.49) 1.22 (1.14) 0.92 (0.26)
122
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
Appendix Table 2 Non-monetary measures of poverty: BMI, perceived happiness and asset index (mean and standard deviations). Year
2002
2003
2004
2005
2006
Adult women BMI Adult men BMI Household happiness (number of members happy at least once in last 7 days) Asset index AI with physical assets only Dummy for households receiving gifts Number of member speaking Spanish Dummy for whether households w/members with math skills Welfare headcount (% of households below the 40th percentile of the distribution of asset index values in 2002)
23.2 (2.6) 23.1 (2.3) 93.8 (24.0)
23.7 (2.8) 23.3 (2.2) 94.3 (23.2)
23.2 (2.9) 23.1 (2.4) 96.0* (19.6)
23.3 (2.7) 23.3 (2.5) 90.9 (28.8)
23.6 (3.0) 23.1 (2.4) 88.0 (32.5)
1.89 (1.06) 1.90 (1.05) 0.64 (0.48) 1.36 (1.25) 0.93 (0.25) 40.1 (49.1)
1.90 (1.03) 1.92 (1.01) 0.59 (0.49) 1.27 (1.13) 0.91 (0.28) 44.2** (49.8)
1.95*** (1.03) 1.98*** (1.03) 0.59 (0.49) 1.18 (1.12) 0.92 (0.27) 36.5** (48.3)
2.17** (1.18) 2.18** (1.15) 0.60 (0.49) 1.10 (1.05) 0.91 (0.28) 29.7** (45.8)
2.34 (1.06) 2.34 (1.04) 0.59 (0.49) 1.22 (1.15) 0.93 (0.25) 22.7 (42.0)
Changes 2002–2006
−0.43*** (1.06) −0.44*** (1.00) 0.05 (0.68) 0.14** (0.9) −0.15***
Note: significance tests marked at the lagged value; for instance if 2003 marked significant it is compared to 2004. For changes between 2002 and 2006, tests of significance indicate significant differences between 2002 and 2006. Test of significance *p b 0.1, **p b 0.05, ***p b 0.01.
Appendix Table 3 Log-linearized regressions of physical asset changes: yearly changes and 2002–2006 changes. Variables
lnAI Ag plots FF OF Diversified Wage Subsistence Ag divers Gender head Age head Sq age head Household size Children Credit Travel SB Literate Fluent Spanish Gifts Shock flood Shock health Pub assistance Dummy 2003 Dummy 2004 Dummy 2005 Dummy 2006 Constant Observations R-squared Number of hhid Rho Sigma_u Sigma_e
Yearly changes (FE)
2002–2006 Changes
Yearly changes (FE)
2002–2006 Changes
(Model 1)a
(Model 2)b
(Model 3)c
(Model 4)d
−1.184*** (0.051) 0.036* (0.020) 0.002 (0.004) 0.002 (0.004) 0.056 (0.039) 0.041 (0.047) −0.114* (0.066) 0.018 (0.034)
−0.706*** (0.068) 0.048 (0.053) 0.009 (0.006) 0.013** (0.006) 0.035 (0.091) −0.056 (0.084) −0.235** (0.097) −0.011 (0.067)
−1.158*** (0.056) 0.032 (0.021) −0.003 (0.004) −0.005 (0.003) 0.042 (0.050) 0.059 (0.063) 0.037 (0.050) −0.000 (0.037)
0.005 (0.011) −0.000 (0.000) 0.117*** (0.017) −0.012 (0.010) 0.090*** (0.033) −0.042 (0.052) 0.063 (0.045) −0.006 (0.041) 0.029 (0.030)
0.001 (0.012) 0.000 (0.000) 0.178*** (0.034) −0.182*** (0.040) −0.024 (0.064) 0.059 (0.066) 0.148** (0.073) 0.039 (0.068) −0.056 (0.064) 0.111 (0.069) −0.043 (0.063) 0.063 (0.064)
0.009 (0.011) −0.000 (0.000) 0.133*** (0.019) −0.017 (0.011) −0.013 (0.042) −0.050 (0.055) 0.042 (0.048) −0.010 (0.041) 0.035 (0.031)
−0.699*** (0.068) 0.085* (0.046) −0.001 (0.007) 0.009*** (0.003) 0.146 (0.091) −0.053 (0.080) −0.094 (0.091) −0.049 (0.069) 0.494*** (0.142) 0.002 (0.012) 0.000 (0.000) 0.225*** (0.029) −0.238*** (0.035) −0.166** (0.070) 0.007 (0.063) 0.135* (0.069) 0.013 (0.066) −0.058 (0.062) 0.131** (0.066) −0.013 (0.062) −0.045 (0.064)
0.000 (0.000) 0.024 (0.046) 0.152*** (0.038) 0.254*** (0.046) −0.460 (0.283) 585 0.660 176 0.741 0.489 0.289
−0.248 (0.249) 136 0.641
0.000 (0.000) −0.019 (0.051) 0.144*** (0.041) 0.234*** (0.044) −0.486* (0.292) 585 0.641 176 0.736 0.503 0.302
Models 2 and 4 include village dummies. Robust standard errors in parentheses *p b 0:1, **p b 0:05, ***p b 0:01. a Model 1 includes lagged values for asset index and current ones for forest clearing, for livelihood strategies and credit. b Model 2 includes 2002 values for asset index and current ones for forest clearing, livelihood strategies and credit. c Model 3 includes lagged values asset index, forest clearing, livelihood strategies and credit. d Model 4 includes 2002 values for asset index, forest clearing, livelihood strategies and credit.
−0.705** (0.289) 136 0.680
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
123
Appendix Table 4 Regressions of physical asset changes: yearly changes and 2002–2006 changes. Variables
AI SqAI CubAI Ag plots FF SqFF CubFF OF SqOF CubOF Diversified Wage Subsistence Ag divers Gender head Age head Sq age head Household size Children Credit Travel SB Literate Fluent Spanish Gifts Shock flood Shock health Pub assistance Dummy 2003 Dummy 2004 Dummy 2005 Dummy 2006 Constant Observations R-squared Number of hhid Rho Sigma_u Sigma_e
Yearly changes (FE)
2002–2006 Changes
Yearly changes (FE)
2002–2006 Changes
(Model 1)a
(Model 2)b
(Model 3)c
(Model 4)d
−3.130*** (0.651) 0.741*** (0.202) −0.060*** (0.019) 0.050* (0.029) −0.040 (0.198) 0.079 (0.165) −0.025 (0.040) 0.135 (0.217) −0.150 (0.202) 0.021 (0.045) 0.148* (0.082) 0.059 (0.067) −0.175** (0.079) 0.001 (0.044)
−0.657* (0.370) 0.0201 (0.147) −0.00965 (0.016) 0.0159 (0.112) 0.504 (0.399) −0.0537 (0.332) −0.00457 (0.0711) 0.275 (0.477) 0.154 (0.513) −0.0462 (0.112) 0.350* (0.178) −0.0557 (0.143) −0.224 (0.177) 0.103 (0.128) 0.268 (0.267) −0.009 (0.022) 0.000 (0.000) 0.387*** (0.0654) −0.419*** (0.0762) 0.0849 (0.115) 0.207 (0.135) 0.0961 (0.125) 0.00604 (0.111) −0.136 (0.106) 0.285* (0.168) 0.0487 (0.0967) −0.160 (0.106)
−3.086*** (0.655) 0.740*** (0.206) −0.061*** (0.019) −0.004 (0.029) 0.134 (0.243) −0.010 (0.189) −0.011 (0.048) 0.119 (0.118) −0.038 (0.062) 0.001 (0.005) −0.006 (0.056) 0.101 (0.071) 0.112 (0.104) 0.030 (0.056)
−0.266 (0.380) −0.159 (0.151) 0.012 (0.017) 0.161 (0.109) −0.623 (0.608) 0.516 (0.568) −0.091 (0.141) 0.123 (0.420) 0.049 (0.278) −0.003 (0.024) 0.278* (0.141) −0.060 (0.151) 0.030 (0.168) −0.022 (0.128) 0.440 (0.268) −0.013 (0.022) 0.000 (0.000) 0.505*** (0.053) −0.530*** (0.060) −0.275** (.131) 0.125 (0.136) −0.013 (0.125) 0.067 (0.127) −0.166 (0.113) 0.344** (0.152) 0.087 (0.103) −0.271** (0.128)
0.022 (0.014) −0.000 (0.000) 0.162*** (0.027) −0.002 (0.015) 0.083 (0.060) −0.046 (0.064) 0.075 (0.071) −0.012 (0.067) 0.053 (0.045)
0.000 (0.000) 0.078 (0.074) 0.217*** (0.059) 0.297*** (0.057) 1.616** (0.657) 586 0.593 176 0.690 0.647 0.433
0.229 (0.530) 137 0.768
0.024* (0.013) −0.000* (0.000) 0.153*** (0.025) −0.008 (0.015) −0.069 (0.056) −0.046 (0.068) 0.078 (0.077) −0.053 (0.061) 0.047 (0.044)
0.000 (0.000) 0.021 (0.069) 0.199*** (0.059) 0.242*** (0.053) 1.735*** (0.638) 587 0.581 176 0.677 0.638 0.440
0.102 (0.533) 135 0.778
Models 2 and 4 include village dummies. Robust standard errors in parentheses *p b 0:1, **p b 0:05, ***p b 0:01. a Model 1 includes lagged values for asset index and current ones for forest clearing, for livelihood strategies and credit. b Model 2 includes 2002 values for asset index and current ones for forest clearing, livelihood strategies and credit. c Model 3 includes lagged values asset index, forest clearing, livelihood strategies and credit. d Model 4 includes 2002 values for asset index, forest clearing, livelihood strategies and credit.
References Angelsen, A., Jagger, P., Babigumira, R., Belcher, B., Hogarth, N.J., Bauch, S., Börner, J., Smith-Hall, C., Wunder, S., 2014. Environmental income and rural livelihoods: a global-comparative analysis. World Dev. 64, S12–S28. Apaza, L., Wilkie, D., Byron, E., Huanca, T., Leonard, W., Prez, E., Reyes-García, V., Vadez, V., Godoy, R., 2002. Meat prices influence the consumption of wildlife by the Tsimane' Amerindians of Bolivia. Oryx 36 (4), 2–13. Barrett, C.B., Carter, M.R., 2013. The economics of poverty traps and persistent poverty: policy and empirical implications. J. Dev. Stud. 49 (7), 976–990. Barrett, C.B., McPeak, J.G., 2006. Poverty traps and safety nets. In: de Janvry, A., Kanbur, R. (Eds.), Poverty, Inequality and Development: Essays in Honor of Erik Thorbecke. Alain de Janvry and Ravi Kanbur (eds.). Springer. Barrett, C.B., Marenya, P.P., McPeak, J., Minten, B., Murithi, F., Oluoch-Kosura, W., Place, F., Randrianarisoa, J.-C., Rasambainarivo, J., Wangila, J., 2006. Welfare dynamics in rural Kenya and Madagascar. J. Dev. Stud. 42 (2), 248–277. Carter, M.R., Barrett, C.B., 2006. The economics of poverty traps and persistent poverty: an asset-based approach. J. Dev. Stud. 42 (2), 178–199. Celentano, D., Sills, E., Sales, M., Verssimo, A., 2012. Welfare outcomes and the advance of the deforestation frontier in the Brazilian Amazon. World Dev. 40 (4), 850–864. Chomitz, K.M., 2007. At loggerheads? Agricultural expansion, poverty reduction, and environment in the tropical forests. Policy Research Report. The World Bank, Washington D.C., USA. Chumacero, J., 2011. Territorios Indgena Originario Campesinos en Bolivia. Entre la Loma Santa y la Pachamama. Fundacíon Tierra, La Paz. Couvreur, T.L., 2011. Palms of the lower Madidi River in Northern Bolivia. Palms 55 (1), 37–45.
Debela, B.L., Shively, G., Angelsen, A., Wik, M., 2012. Economic shocks, diversification, and forest use in Uganda. Land Econ. 88 (1), 139–154. Ellis, F., 2001. Rural Livelihood and Diversity in Developing Countries. Oxford Press University, Oxford. Fisher, M., 2004. Household welfare and forest reliance in Southern Malawi. Environ. Dev. Econ. 9 (2), 135–154. Friel, C.M., 2007. Notes on Factor Analysis. Criminal Justice Center. Godoy, R., Jacobson, M., Wilkie, D., 1998. Strategies of rain-forest dwellers against misfortunes: the Tsimane' indians of Bolivia. Ethnology 37 (1), 55–69. Godoy, R., Reyes-García, V., Huanca, T., Tanner, S., Seyfried, C., 2007. On the measure of income and the economic unimportance of social capital: evidence from a native Amazonian society of farmers and foragers. J. Anthropol. Res. 63 (2), 239–260. Godoy, R., Reyes-García, V., Leonard, W., Tanner, S., Huanca, T., Vadez, V., Wilkie, D., Team, T.B.S., 2008. Short and long-run effects of household income on household forest clearance among native Amazonians: panel evidence (2002–2006) from the Tsimane', Bolivia. Working Paper 43. Tsimane' Amazonian Panel Study. Godoy, R., Nyberg, C., Eisenberg, D., Magvanjav, O., Shinnar, E., Leonard, W., Gravlee, C., Reyes-García, V., McDade, T., Huanca, T., Tanner, S., Team, T.B.S., 2009a. Short but growing: catch-up growth among native Amazonian Bolivian children. Working Paper 48. Tsimane' Amazonian Panel Study. Godoy, R., Reyes-García, V., Gravlee, C., Huanca, T., Leonard, W.R., McDade, T., Tanner, S., Team, T.B.S., 2009b. Moving beyond a snapshot to understand changes in the wellbeing of native Amazonians. Curr. Anthropol. 50 (4), 563–573. Godoy, R., Reyes-García, V., Vadez, V., Leonard, W., Tanner, S., Huanca, T., Wilkie, D., Team, T.B.S., 2009c. The relation between forest clearance and household income among native Amazonians: results from the Tsimane Amazonian panel study, Bolivia. Ecol. Econ. 68, 1864–1871.
124
E. Perge, A. McKay / Ecological Economics 126 (2016) 112–124
Günther, I., Klasen, S., 2007. Measuring chronic non-income poverty. Draft Paper. Department of Economics, University of Gottingen. Killeen, T.J., Calderon, V., Soria, L., Quezada, B., Steininger, M.K., Harper, G., Solrzano, L.A., Tucker, C.J., 2007. Thirty years of land-cover change in Bolivia. Ambio 36 (7), 600–606. Lawley, D., Maxwell, A., 1973. Regression and factor analysis. Biometrika 60 (2), 331–339. Leonard, W.R., Godoy, R., 2008. Tsimane Amazonian Panel Study (TAPS): the first 5 years (2002–2006) of socioeconomic, demographic, and anthropometric data available to the public. Econ. Hum. Biol. 6 (2), 299–301. Leonard, W.R., Reyes-García, V., Tanner, S., Rosinger, A., Schultz, A., Vadez, V., Zhang, R., Godoy, R., 2015. The Tsimane Amazonian Panel Study (TAPS): nine years (2002–2010) of annual data available to the public. Econ. Hum. Biol. 19, 51–61. Lybbert, T.J., Barrett, C.B., Desta, S., Coppock, D.L., 2004. Stochastic wealth dynamics and risk management among a poor population. Econ. J. 114 (498), 750–777. McKay, A., Perge, E., 2013. How strong is the evidence for the existence of poverty traps? A multicountry assessment. J. Dev. Stud. 49 (7), 877–897. Moser, C., Felton, A., 2007. The construction of an asset index measuring asset accumulation in Ecuador. Working Paper 87. Chronic Poverty Research Center CPRC, Manchester, UK. Paneque-Gálvez, J., Mas, J.-F., Gueze, M., Luz, A.C., Macía, M.J., Orta-Martínez, M., Pino, J., Reyes-García, V., 2013. Land tenure and forest cover change. The case of southwestern Beni, Bolivian Amazon, 19862009. Appl. Geogr. 43, 113–126. Perge, E., 2010. Testing a poverty trap mechanism with Tsimane panel data. Working Paper 158. Chronic Poverty Research Center CPRC, Manchester, UK. Reyes-García, V., 2001. Indigenous people, ethnobotanical knowledge, and market economy. A Case Study of the Tsimane' Amerindians in Lowland Bolivia. University of Florida, Gainsville, Florida Ph. D. thesis. Reyes-García, V., Ledezma, J.C., Paneque-Galvez, J., Orta, M., Gueze, M., Lobo, A., Guinart, D., Luz, A.C., 2012. Presence and purpose of nonindigenous peoples on indigenous
lands: a descriptive account from the Bolivian Lowlands. Soc. Nat. Resour. 25 (3), 270–284. Reyes-García, V., Paneque-Galvez, J., Bottazzi, P., Luz, A.C., Gueze, M., Macía, M.J., OrtaMartínez, M., Pacheco, P., 2014. Indigenous land reconfiguration and fragmented institutions: a historical political ecology of Tsimane' lands (Bolivian Amazon). J. Rural. Stud. 34, 282–291. Sahn, D.E., Stifel, D., 2000. Poverty comparisons over time and across countries in Africa. World Dev. 28 (12), 2123–2155. Sunderlin, W., Angelsen, A., Belcher, B., Burgers, P., Nasi, R., Santoso, L., Wunder, S., 2005. Livelihoods, forests, and conservation in developing countries: an overview. World Dev. 33 (9), 1383–1402. Sunderlin, W., Dewi, S., Puntodewo, A., 2007. Poverty and forests: multi-country analysis of spatial association and proposed policy solutions. Occasional Paper 47. Center for International Forestry Research CIFOR, Bogor, Indonesia. Vadez, V., Reyes-García, V., Godoy, R., Williams, L., Apaza, L., Byron, E., Huanca, T., Leonard, W., Prez, E., Wilkie, D., 2003. Validity of self-reports to measure deforestation: evidence from the Bolivian Lowlands. Field Methods 15 (3), 284–304. Vadez, V., Reyes-García, V., Godoy, R., Apaza, L., Byron, E., Huanca, T., Leonard, W., Prez, E., Wilkie, D., 2004. Does integration to the market threaten agricultural diversity? Panel and cross-sectional data from a horticultural-foraging society in the Bolivian Amazon. Hum. Ecol. 32 (5), 635–646. Vadez, V., Reyes-Garca, V., Huanca, T., Leonard, W., 2008. Cash cropping, farm technologies, and deforestation: what are the connections? A model with empirical data from the Bolivian Amazon. Hum. Organ. 67 (4), 384–396. Wunder, S., 2001. Poverty alleviation and tropical forests: what scope for synergies? World Dev. 29 (11), 1817–1833. WWF, 2014. Overview: forests. http://www.worldwildlife.org/habitats/forests. Zycherman, A., 2013. The Changing Value of Food: Localizing Modernity among the Tsimané Indians of Lowland Bolivia PhD Dissertation Columbia University, New York.