Theorizing Land Cover and Land Use Change: The Peasant Economy of Amazonian Deforestation

Abstract

This article addresses deforestation processes in the Amazon basin, using regression analysis to assess the impact of household structure and economic circumstances on land use decisions made by colonist farmers in the forest frontiers of Brazil. Unlike many previous regression-based studies, the methodology implemented analyzes behavior at the level of the individual property, using both survey data and information derived from the classification of remotely sensed imagery. The regressions correct for endogenous relationships between key variables and spatial autocorrelation, as necessary. Variables used in the analysis are specified, in part, by a theoretical development integrating the Chayanovian concept of the peasant household with spatial considerations stemming from von Thünen. Results from the empirical model indicate that demographic characteristics of households, as well as market factors, affect deforestation in the Amazon basin associated with colonists. Therefore, statistical results from studies that do not include household-scale information may be subject to error. From a policy perspective, the results suggest that environmental policies in the Amazon based on market incentives to small farmers may not be as effective as hoped, given the importance of household factors in catalyzing the demand for land. The article concludes by noting that household decisions regarding land use and deforestation are not independent of broader social circumstances, and that a full understanding of Amazonian deforestation will require insight into why poor families find it necessary to settle the frontier in the first place.

Key Words:

• Amazon
• deforestation
• peasant economy

Acknowledgments

This research was supported by the National Aeronautics and Space Administration under the project “A Basin-Scale Econometric Model for Projecting Amazonian Landscapes—NCC5-694,” by the National Science Foundation under the projects “Patterns and Processes of Landscape Change in the Brazilian Amazon: A Longitudinal, Comparative Analysis of Smallholder Land Use Decision-Making—BCS137020” and “Tenure Security and Resource Use in the Amazon,” and by the National Geographic Society under a Field Support Grant. We are particularly indebted to all who have participated in our various field activities, and to the anonymous reviewers of the journal who greatly improved our original manuscript. We remain responsible for any remaining errors.

Notes

Soil Type 1: High potential, ranching. Mostly latisols, with good structure, but nutrient poor (EMBRAPA 2001).

Soil Type 2: High potential, crop agriculture. Eutrophic soils; saturation of bases > 50 percent (EMBRAPA 2001).

Soil Type 3: Medium potential, ranching. Mostly latisols, with good structure, but nutrient poor and hilly (EMBRAPA 2001).

Soil Type 4: Low potential. Gravelly soils with minimal A horizon (EMBRAPA 2001).

Note: Standard deviations are in parentheses.

^aThe amount of day labor paid the preceding year (person-days).

^bDependents: children + elderly + women.

^cThe survey asked if the property had ever received agricultural credit.

^dThe census data include large properties.

Notes: Wealth was defined on specific durable goods possession as queried in the survey:

Category 1: household possesses none of the surveyed goods.

Category 2: household possesses stove or chainsaw.

Category 3: household meets category 2, plus refrigerator, or generator, or television, or satellite dish, or motorcycle.

Category 4: household meets category 3, plus car or tractor.

The dummy variables used in regression (see Table 5) are given in brackets, as defined by the categories; the omitted dummy for the regression analysis comprises the poorest group of farmers.

Notes: Robust standard errors are in parentheses below the coefficients. Two-tailed probabilities. AIC=Akaike Information Criterion.

IV (1) and (2) are two-stage estimators; title to land is used as an instrumental variable for credit. The same credit equation was used in both IV (1) and (2). The title coefficient is 0.88 with a significance level of 0.001.

Spatial (1) and (2): spatial lag model; credit variable used in IV (1) and IV (2) estimations. Breuch-Pagan tests do not allow rejection of homoskedastic error hypotheses in either regression.

^aPaid non-family labor used the year before the survey, in men per day.

^bOne-tailed probability value is 0.068.

^cThe wealth dummy variables (1)–(3) are defined in Table 4. Category 2=wealth 1; Category 3=wealth 2; Category 4=wealth 3, and the omitted dummy is Category 1.

^dSoil class 4, the low potential category, is omitted because variables are defined as percentages, which sum to 1 for individual properties. Including all four variables would introduce a linear dependency, thereby rendering estimation impossible.

^eWater is a binary variable with value 1 if a stream flows within 500 m of the front of the lot (facing the settlement road), and 0 otherwise. Multilot properties were measured for the lot of residence, typically the first one occupied.

^fDummy variables that measure the household access to agricultural credit. For how access to credit was determined, refer to the main text and to Note 13.

^***significant at 1 percent;

**significant at 5 percent;

*significant at 10 percent.

Note: These are the Robust Lagrangian statistics. The Moran's I statistic is always greater than the expected value, indicating positive spatial autocorrelation.

1. Indeed, loggers and colonists engage in mutually beneficial interactions with immediate implications for forest cover (Walker 2003). Aggregate and disaggregated approaches may avoid this because their dependent variable is indifferent to deforestation associated with different types of agents such as loggers, miners, and so on. But they cannot attribute the amount of deforestation to the individual types.

2. We by no means wish to suggest that research on household livelihoods, land use, and agriculture in forest frontiers of the Amazon basin and elsewhere is limited to the three countries indicated. Land cover change in Bolivia has been studied (Godoy, Jacobson, and Wilkie 1998; Mertens et al. 2004; Hecht 2005), where colonization processes have a long history (Stearman 1984). In addition, land use and livelihoods of both “peasant” and indigenous peoples have received considerable attention in Hondurus (Godoy, Groff, and O'Neill 1988; Stonich 1993; McSweeney 2004a, 2004b).

3. The McCracken effort was largely descriptive and the socioeconomic variables implemented in their regression models were not theoretically derived (McCracken et al. 1999, 1316). On the other hand, sample size severely constrained the Walker study (n=32).

4. Markets for inputs, labor, and land certainly existed in nineteenth- and early twentieth-century Russia. Nevertheless, Chayanov abstracted from the empirical setting to focus on the case of autarkic households, at least with respect to labor markets. Although nonseparability can arise from weaker conditions than a completely nonexistent labor market (Benjamin 1992), our treatment abstracts from such complexities to focus on the household demand for land, rather than labor. In fact, it is common in forest frontiers for households to be detached from labor markets, but engaged in product markets (Maertens, Zeller, and Birner 2006).

5. In other words, the decline in transportation costs over time allows a switch from nonseparable to separable production (Lopez 1984, 1986).

6. The assumption, then, is that although product markets exist, those for labor are imperfect. This is consistent with such frontier settings (Maertens, Zeller, and Birner 2006).

7. Transportation costs affecting farm-gate prices can be interpreted as transactions costs (Omamo 1998; Key, Sadoulet, and de Janvry 2000).

8. This is equivalent to an open access situation (terra devoluta), where land is free for the taking. In colonization areas, boundaries are demarcated, but the land grants typically exceed the amount of land that can be cleared in the short to midrun.

9. These results are derived in a technical appendix available from the authors upon request.

10. Good soils could encourage agricultural intensification and reduced deforestation, although they might also increase deforestation given enhanced profit potential, as Pichón (1997) observed. Similarly, human capital might include acquisition of environmental knowledge and values, at the same time that enhanced farming ability could lead to more clearing.

11. Distance to the Transamazon Highway is used because transportation price structures are mostly defined on this distance, at least within the vicinity of Uruará. Travel on BR-230 is easy; travel on the settlement roads is difficult. Uruará is the commercial market for the goods, if not the population of final consumption. Dependency is defined as number of nonworkers. Chayanov's variable (the number of dependents divided by the number of workers) confounds the number of dependents with the active workforce. Our definition yields a pure consumption effect, to the extent nonworkers do not contribute to farm labor.

12. We defined our wealth variable based on possession of durable goods. Because we did not have their prices, we did not create a single monetary value (Junming 1997; Morris, Carlleto, and Hoddinott 2000). Rather than impose a weighting scheme, we chose categories that could be used to define dummy variables (Weil 1989; Menon, Ruel, and Morris 2000). Wealth is correlated with durable goods possession (Morris, Carlleto, and Hoddinott 2000). The four wealth categories are given in Table 4. The poorest households (category 1) possess none of the durable goods listed. Category 2 households possess either a stove or a chainsaw, and nothing else, and category 3 households possess a stove or a chainsaw, plus either (i) some good depending on electricity generation or (ii) a motorcycle. The wealthiest households (category 4) have what category 3 households own, but also possess either a car or a tractor. In essence, our categorization here elaborates the two categories used in Walker, Perz, et al. (2002).

13. Credit is a binary variable indicating whether the property currently has or has ever had bank credit for agriculture. Colonists were queried in the survey for a yes/no response, and how the credit was used. At some time, 52.2 percent of the households had had agricultural credit. Of these households, 16 percent reported using it strictly for pasture, 4.1 percent strictly for annuals crops, and 2.3 percent strictly for perennials. The remainder used their loans for combinations of these three basic crop groups.

14. Distance for a single-lot property is simply the distance from the lot to the Transamazon, BR-230 (cf. Note 11). For multilot holdings, distance is from the lot of residence, typically the first lot occupied by the colonist.

15. The 1996 Brazilian population count (IBGE 1996a) and 1995/96 Brazilian agricultural census (IBGE 1996b) allow for comparisons to assess sampling bias. The Uruará sample had a mean household size of 7.5; the 1996 population count figure for the municipality of Uruará was only 5.6, but it is not clear from census documentation whether families beyond the first were counted. If we exclude people outside the first family, household size in the Uruará sample is also 5.6. The 1995/96 agricultural census indicated the following land use allocation in Uruará: 65 percent in primary forest, 5.6 percent under cropland, 23 percent under pasture, and 5.9 percent under secondary growth. This is very similar to the sample (see Perz 2004).

16. The term “pasture” as used by colonists does not reflect our common understanding. In particular, colonists broadcast grass seeds onto deforested land, which they abandon until they can stock it with cattle. Any regrowth seeded with grass is subsequently referred to as “pasto sujo” (dirty pasture), and comprises part of the farming system, even if it shows up on satellite imagery as “regrowth.” For this reason we calculate our dependent variable by summing cleared land and other lands not forested, which we call regrowth, for lack of a better term. In the region, truly fallowed land has two names, capoeira, which may possibly be used again in the long-term, and terra abandonada, which is land that has been abandoned for some reason (too wet, too rocky, etc.). Self-reported magnitudes in the sample for these two categories are 1.73 and 0.20 ha, respectively, amounts that are small compared to average deforestation on the properties, which is 49 ha (Perz and Walker 2002).

17. Household wealth might also be regarded as endogenous to deforestation, since increased clearance should yield more production and greater potential for market success. To overcome a possible endogeneity issue, we use wealth endowment, or the productive assets in the possession of the colonists when they arrived on the property, and not in the year of the survey, 1996.

18. Other assets could be posted, but land may be the only possession of sufficient value. Evidently, the Fundo Constitucional do Norte (FNO) program in Brazil, which provides credit to smallholders, did not require much in the way of collateral.

19. For models using discrete dependent variables defined on pixel states, researchers have been stymied in this regard because theoretically valid approaches to resolving the problem—namely, through implementing spatial regression—have been unavailable until recently (Lesage 2000). Thus, modelers have typically resorted to ad hoc solutions, such as defining new independent variables based on spatial “lags” of the dependent variable (Nelson and Hellerstein 1997; Wear and Bolstad 1998; Nelson, Harris, and Stone 2001).

20. Although women clearly work on the farm, there is a distinct gendered division of labor, and men undertake most of the land clearing and crop tending. The men and women cohorts are for individuals aged 15–64. Children are younger, and the elderly are older (Table 5).

21. We experimented with a third demographic specification, aggregating total men, total women, and the elderly, and leaving children as a separate variable. This model yielded similar results. The aggregated labor force variable was positive (3.4) and significant (α=0.005), and the children “dependents” variable was negative (−1.2) and insignificant (α=0.178).

22. Some of the households (36 percent) obtain off-farm income via external activities (e.g., business in town, wages), retirement funds (Brazil's Funrural program), or remittances. We experimented with a binary dummy variable defined on whether a household had access to off-farm income. OLS regression did not change the significant variables of Table 5 or their magnitudes in a meaningful way.

23. Arima, Barreto, and Brito (2005) show for example that cattle ranching in the Amazon is more profitable than in the choice locations in the southern part of the country.