2003-2007: Amazonian Deforestation and the Structure of Households (Phase II) (funded by NICHD)

Description The initial phase of this project focused on the initial demographic composition of the households of smallholder settlers entering the agricultural frontier, how those demographic characteristics changed over time, and how these changes interacted with other characteristics of households to determine patterns of forest clearing and regrowth. This initial phase involved collecting social survey data from landowning households on 402 properties in one study area in a colonization area to the west of the city of Altamira (including parts of the current municípios of Altamira, Brasil Novo, Medicilândia, and Uruará) in 1997 and 1998. These data were linked to satellite data, classified into areas with various land uses, by using property boundaries. This project was, to our knowledge, the first to make the social survey satellite data linkage through the use of cadastral maps updated with field data, a practice that is now relatively common among population and environment researchers (e.g. Walsh et al. 2005, Fox et al. 2003), and that developed methods that permitted querying and analysis at the level of the individual property as well as the landscape (Moran and Brondizio 1998).

This early work drew on a conceptual model of the developmental cycle of domestic groups that indicated that deforestation patterns would change in a predictable way over the lives of households as their available labor and consumption needs changed. Young households have little available labor and relatively high consumption needs. Over time, labor increases and consumption needs stabilize. We argued that this resulted in a distinctive pattern of land use change over time. Surprisingly, we found that it is upon arriving in the frontier, when most frontier households are youngest, that one finds the highest rates of deforestation (Brondizio et al. 2002). Surprising, too, is the finding that within five years, most households are on a trajectory of declining rates of deforestation that persists over the next 10 year period (McCracken et al. 2002).  There is a short-term increase in deforestation associated with consolidation of the areas farmed (approximately 15 years after settlement), followed by a steady decline in deforestation rates. Each cohort appears to follow this trajectory no matter at what time they arrived, even if they came during a period of hyperinflation and tight credit policy (Brondizio et al. 2002).  While the magnitude of an individual household's deforestation is a product of period effects, increasing when credit rates are lower and inflation is lower, the trajectory of deforestation is the same across arrival cohorts. In fact, our analysis showed that aggregate deforestation rates declined during the 1980s because the much larger population representing earlier arriving cohorts had entered into a period of declining rates of deforestation, compounded with the unfavorable conditions for farm expansion (Moran and Brondizio 2001).

Our conceptual model initially suggested a shift from pasture to perennial crops as a time - and capital-dependent shift in land use strategy associated with the aging of household members. However, our analyses showed that the decision between perennials and pasture is explained better by differences in the physical endowments of the farm (Moran et al. 2002; Moran et al. 2000).  We found a linear relationship between the proportion of high quality soils on the farm and the proportion of the property in pasture.  Whereas properties that are characterized by acid, low nutrient soils have over 80% of their deforested area in pasture, this percentage drops in proportion to the amount of good soils on the property until it reaches a balanced amount of pasture and perennial crops (mostly cocoa and sugar cane) on properties with mostly good soils (i.e. terra roxa estruturada eutrófica or alfisols).  This suggests a learning process where farmers learn over time the differences that soils make in yield, and those with good soils develop balanced portfolios of land use, whereas those with poor soils tend to focus only on pasture and are unable to evolve their strategy in the direction of perennial crops.

Based on these main findings of the initial phase of the project, our second phase proposed a further test of the conceptual model and exploration of the mechanisms underlying it (migration and marriage as components of changing household labor availability, off-farm employment, as well as biophysical constraints on household decisions). We expanded the geographic extent of our project by adding a study area in the agricultural area to the south of the city of Santarém (including parts of the current municípios of Santarém and Belterra).  The Altamira study area settlement scheme began only in 1971 and thus we were able to capture only the actions of first generation settlers.  In the second phase of the project, we examined a much older Amazon region where we were able to examine trajectories of land use change among later generations of settlers, and collected a second wave of data in Altamira.  We collected new social survey data on households and farms in Santarém in 2003, including in this data collection all households (owners and residents) on sampled rural properties. We used a multistage cluster sampling approach: within each of four regions in the study area (each with a different settlement history), we selected a random sample of grid cells (each being nine square kilometers); we then selected a random sample of properties from existing cadastral maps within each grid cell; we then interviewed every household in each selected property. This spatial sample and the interviewing of all households in each property produced a sample that allows results of models including specific subsamples of the data to be generalized to the population of households as well as to the populations of properties or landowning households. The follow up data in Altamira provided a prospective look at changes in households and properties between 1997/98 and 2005. We followed properties, owning households, and children who had been living in the owning households in 1997/98. Thus, this phase of the project allows us to more persuasively examine the reciprocal relationships between population and environment.

The data collected in Santarém allowed us to conduct comparative analyses of the effects of household life cycle stage on land use and land cover in Altamira and Santarém. The first of these comparative analyses examines the independent effects of household life cycle (or household age) and of time since acquisition of the property on land use (VanWey et al. in revision). Our past work on the cycles of deforestation used the time of settlement as a proxy for the start of the household life cycle, based on the observation that most settlers were young families (McCracken et al. 1999). However, there is considerable variation in the time between and sequence of formation of a household (marriage) and acquisition of a property. Our results show that what we had previously seen in an examination of trajectories of deforestation based on satellite data overlaid with property boundaries was actually a cycle of deforestation following a standard pattern over the time since acquisition of a property by a household. These results, showing a property life cycle,. are consistent with patterns shown by Pan and Bilsborrow (2005) in the Ecuadorian Amazon and Perz et al. (in press) in Uruará. There is no independent effect of head of household age once the time since acquisition of the property is controlled. Our recent analyses further show that this property life cycle is only evident in Altamira, arguably because Altamira is a new frontier and settlers are arriving on largely forested properties about which they know relatively little. In contrast, in Santarém, we find no effect of household age or time since acquisition, finding instead evidence of property level production specialization. The initial land use on a property at the time of acquisition has a lasting effect, suggesting that households select land with a given use in mind in this region where generations of settlers have learned the most appropriate uses of properties given biophysical characteristics and proximity to markets.

The second set of comparative analyses examines the mechanisms put forward in our and others development of the household life cycle model (Brondizio et al. 2002; McCracken et al. 1999; Perz et al. in press; Perz and Walker 2002; Walker 2003). We explicitly test the effects of changing household demographic composition on changes in forest area and used area (measured using both survey data and satellite data) using fixed effects regression models (VanWey, et al. under review). We find no support for the argument that production depends on available household labor (in this context, male adolescents and adults). Models show instead that changes in the number of children and women, particularly young women, have the most significant effects on land use and land cover change. We speculate that this is in part due to government subsidies for education and in part reflects endogenously determined migration decisions of young women. These two sets of comparative analyses indicated that households, in fact, are engaged in production for local and regional economies with expansion in mind rather than production primarily for household consumption (as assumed in Chayanov 1966; Perz et al. in press; Perz and Walker 2002; Walker 2003).

In addition to these questions about the relationship between household life cycle and land use / land cover, our second phase of the project addressed directly some of the family processes underlying changes in households and residents of properties over time. In the first phase of this project, we found a rapid decline in fertility across birth cohorts in the frontier, which arguably should be characterized by high fertility given land abundance and labor scarcity. We found that women use a variety of methods of fertility control and reduction with more than 80% having used some form of fertility control in the past (Siqueira and McCracken 2001) and the two youngest cohorts of reproductive age women being aware of their contraceptive options (Siqueira et al. under review). Most notable among these is the widespread use of sterilization, usually after having two children. We found that over 43% of women in the Altamira study area were sterilized by age 25-29 and that this procedure was tied closely (i.e. in 70% of cases) to cesarean birth events. To follow up on this finding, we specifically examined sterilization in our Santarém study area (after verifying the same fertility decline among the population). Our work on sterilization in Santarém confirms the prevalence of sterilization, and offers a contrast to findings from the Northeast of Brazil where it is tied to political clientilism (Caetano and Potter 2004). In our study area the use of sterilization is a strategy pursued by women and financed by their own resources (Siqueira et al. under review). Reversible contraceptive methods are difficult and costly to access outside of the urban area of Santarém, and those women who have used hormonal methods (overwhelmingly the pill in this area) in the past report high rates of adverse health effects due to the high hormonal content of the available pills and lack of medical follow up.

We additionally have examined the determinants of inter-generational co-residence on farm properties, focusing on the differences between daughters and sons in the determinants of co-residence (VanWey and Cebulko, in revision). In our earlier work in Altamira, we found that daughters were substantially more likely to leave their parents' farm, and that they were more likely to be moving to the city (McCracken and Siqueira 2000). We verified this result in Santarém and further found that there are no significant differences in the effects of children.s or parents. characteristics on living out of the parents' household versus living off of the parents' property. Congruent with fieldwork observations, living together on a property is virtually equivalent to sharing a household. We also found that not only were daughters more likely to live off the farm, it was the daughters with the most education and from wealthier backgrounds (farm-owning rather than farm workers) who were the most likely to live away from their parents. Finally, our second phase of the project involved analyses of the importance of the entry of large farmers, associated land consolidation, and urbanization in the Santarém region. Though these issues were not specific aims of our phase II proposal, they were immediately evident in the field. In 2003, we arrived in Santarém months after the opening of a deep water port operated by the multinational Cargill to facilitate the export of soybeans from the Center of Brazil (the Mato Grosso region). This port (and the anticipation of it) spurred the growth of soybean production in the region and the arrival of large capital to transform the landscape from one of small farms to one of large mechanized farms in a favorably flat plateau. When we went to the field with our sample drawn from cadastral maps, we found that fully half of the sampled properties were no longer small farms, with most now part of large mechanized farms. In an analysis of the locations of these large farms and other forms of property consolidation, we found that while there were high rates of consolidation close to the better roads to Santarém, there were also high rates of consolidation in forested regions far from good roads and far from the city (VanWey, et al. manuscript).

We further explore the impacts of land consolidation only among small farmers by examining the impact of property size on forest cover and forest cover change (D.Antona, VanWey and Hayashi, in press). We find that larger properties are able to preserve a larger proportion of their area in forest, and also are able to allow previously used area to go through longer fallow cycles, with vegetation growing for long enough to be indistinguishable from forest in satellite imagery. In this new phase of the project, we will further explore this process of consolidation and mechanization in order to assess its long-term impacts on forest cover among other things.

In considering urbanization, we have also drawn on field experiences. Using our experiences in several sampled properties which at the time of the survey had an unexpected combination of agricultural lands and residential parcels, we developed a typology of property trajectories. These show that some properties in the rural zones around Santarém are in fact increasing in both housing density and infrastructure, a sort of rural urbanization (D.Antona and VanWey, under review). Associated sometimes with processes of inheritance and sometimes with explicit household strategies, properties are being subdivided below the threshold for viable agricultural use. In some cases, households are self-financing some infrastructure (bars, churches) in an effort to leverage public investment in infrastructure (electrification, piped water, schools, health centers).

The two phases of the project to date have also made useful technical and methodological contributions. Of particular note in the first phase of the project is our development of an approach that permits querying at both a landscape and household/property level (McCracken et al. 1999; McCracken et al. 2003; Brondizio et al. 2002; Moran et al. 2002; McCracken et al. 2003). This technical approach represented a breakthrough in the analysis of population and environment. Previous work, for an area this size (3800 sq km), had typically analyzed and described land-cover change from the perspective of the landscape, without addressing the differential and multiple land use dynamics at the level of individual farms and households, or had analyzed land use from aspatial survey data. This dynamic querying at both property and landscape levels has proven productive, as it helps identify different patterns on the landscape and relate the behavior of households to their economic, demographic, and biophysical characteristics in a spatially explicit fashion.

In the second phase of the project, we focused on improving sampling methods to allow us to examine the reciprocal relations between population and environment. As described above, we designed a multi-stage cluster sampling approach in Santarém that allows results from the full sample or various subsamples to be generalized to the population of households or to the population of properties (the landscape). In our first phase of data collection in Altamira, we took advantage of properties usually containing only one household to collect data simultaneously at the household and property level, making that initial sample also generalizable to both the population of households and the landscape. In 2005, we conducted a three-pronged follow up in Altamira: (1) we re-interviewed the previously interviewed owner (and spouse); (2) we conducted interviews with owners and all resident households on previously interviewed properties; and (3) we interviewed children of the initial owner who had been living in the owners' household in the first wave. To our knowledge, we are the first team to conduct a simultaneous follow up of both households and properties, allowing us to look longitudinally at both environmental effects on population change and population effects on environmental change.