Household Demographic Factors as Life Cycle Determinants of Land Use in the Amazon. Stephen G. Perz 1 or

Transcription

1 1 Household Demographic Factors as Life Cycle Determinants of Land Use in the Amazon Stephen G. Perz 1 (sperz@soc.ufl.edu or sperz@latam.ufl.edu) Department of Sociology and Center for Latin American Studies, University of Florida Prepared for Delivery at the 2000 Meetings of the Latin American Studies Association, Hyatt Regency Miami, March 16-18

2 2 Household Demographic Factors as Life Cycle Determinants of Land Use in the Amazon Abstract This paper seeks to broaden the application of demography to environmental studies by complementing existing macro-level approaches, which feature aggregate populations, with a micro-level approach that highlights household life cycles. I take up the case of small farm households in the Brazilian Amazon to present a theoretical framework that identifies demographic characteristics which dispose families to engage in different forms of land use as household age structures change. Empirical models show that net of the effects of farmer background, neighborhood and institutional contexts, and off-farm incomes, demographic variables indicative of the household life cycle exert significant effects on the prominence of land uses with distinct environmental ramifications. The findings not only reveal micro-level demographic factors which affect Amazon land cover, they yield implications for future changes in rainforest landscapes in northern Brazil, and suggest household life cycle models as an avenue for further demographic research on environmental change in Latin America and other contexts. Key words: Amazon, Demography, Households, Land use, Life cycle

3 3 Introduction Household Demographic Factors as Life Cycle Determinants of Land Use in the Amazon Discussions of population-environment interactions have become more salient in the demographic community over the past decade (e.g., Arzipe, et al. 1994; Mazur 1994; Ness, et al. 1993; Pebley 1998; UN 1994). Most of the attention to demographic factors that affect environmental outcomes is pitched at a macro level, wherein aggregate population size is posited to influence regional or global change (e.g., Bongaarts 1992, 1996; Cohen 1995; Davis & Bernstam 1991; Keyfitz 1991; MacKeller, et al. 1998; Panayotou 1996; Preston 1996; Smil 1994). This is true in part because existing theories of environmental change that feature demographic factors tend to focus on population size, growth or density (e.g., Jolly 1994). Two prominent examples are neo-malthusian notions that population pressure leads to resource degradation (e.g., Ehrlich & Ehrlich 1990; Green 1992; Kates 1996; Smail 1997), and Boserupian arguments that population-induced technological innovations may avoid or alleviate degradation (Boserup 1965, 1981). Less common are micro level frameworks of household demography as an influence on environmental outcomes (Lutzenheiser & Hackett 1993 is one example). Micro level models provide the valuable service of differentiating among groups within populations in order to better grasp the questions of who is responsible for environmental degradation, and why. While there is a large literature on households and environmental change in economics and anthropology, this literature pays only incidental attention to demographic factors. More recognition of micro level frameworks in environmental studies by demographers would thus complement existing macro level models while broadening the contribution of demography to research on environmental change. This paper focuses on land use among small farms in the Amazon in order to link demographic factors to environmental outcomes at the household level. Demographic factors occupy a central role among the "human dimensions" of land use, as articulated in the Land Use and Cover Change Science Plan, produced by the International Geosphere/Biosphere Program for the Study of Global Change and the Human Dimensions Program of the International Social Science Council (Turner, et al. 1995). The Amazon is a crucial case for an examination of household demographic effects on land use, given the extensive deforestation in that region (e.g., Skole and Tucker 1993; INPE 1998) and its many local, regional and global environmental consequences (e.g., Jordan 1986; Fearnside 1990; Gash, et al. 1996). The core argument here is that the "location" of a household in its life cycle is an important determinant of land use and cover change. By "location" of a household's life cycle I refer to a set of demographic characteristics of a domestic group that includes the age of individual members, their collective age composition, and the length of time spent at their present residence. Households that differ in terms of these demographic characteristics occupy different locations along a similar domestic life cycle trajectory. The paper begins with a review of theories that tie a household's life cycle location, and therefore it's demographic characteristics, to predominant land uses. Given a theoretical framework, I draw on a recent survey of small farms on the Transamazon highway, which includes information about household demography and land use practices. The framework and survey provide the basis for an empirical analysis of the effects of household demography and other factors on five land use outcomes. Multivariate models will show that net of the effects of farmer background, neighborhood and institutional contexts, and off-farm income, a household's life cycle location exerts significant effects on four of the five outcomes. These findings bear important implications for our understanding of the determinants of land cover change in the Amazon while broadening the relevance of demography for land use and cover change research and environmental studies in general. Background: Household Demography and Land Use Chayanov's Theory of Demographic Differentiation and Land Use The theoretical foundation of the importance of household life cycles for land use was laid by A.V. Chayanov, who studied farming practices among peasants following the October Revolution in Russia in 1917 (Thorner, et al. 1986). Chayanov observed that peasant households contained families with different age structures, and that those households also farmed different quantities of land. He sought to explain differences in cultivated area in terms of the laborer to consumer ratio in households. Chayanov reasoned that this ratio is high in households with larger numbers of adults and/or smaller numbers of children, both of which allow for greater allocation of labor to agriculture, which in turn enables cultivation of larger land areas. Chayanov drew on this basic insight about

4 4 household demography and labor availability to argue that the laborer/consumer ratio changes through the course of a household's life cycle. He distinguished among life cycle stages, where early on the age of the household head is low (generally under 30 years) and the household age structure is young (due to the presence of infants and young children), so the laborer/consumer ratio is low and relatively little land is farmed due to limited labor available for agriculture. As time passes, the household head ages and the average age among the children increases, so as children grow up and become more economically active, the laborer/consumer ratio rises, allowing expansion of the land area cultivated. By distinguishing households in terms of the age of the household head and the laborer/consumer ratio, Chayanov provided a demographic explanation for differences in land area cultivated among Russian farms. Chayanov's model is relevant to contemporary agriculture in regions such as the Amazon since he studied peasants under conditions of land abundance. In the case of Russia, land was abundant because the October Revolution freed property previously held in private estates (Thorner, et al. 1986). Land abundance also characterizes agricultural frontier areas such as the Amazon, which have low rural population densities (Pichón 1996a). But while Chayanov's model provides a theoretical foundation for examining the demographic dimensions of land use among households, and despite the similar conditions of land abundance in post-revolutionary Russia and the contemporary Amazon, Chayanov made several assumptions about peasants that only partly apply to agriculture in tropical regions of Latin America (Ellis 1993; Netting 1993). First, Chayanov did not treat the question of migration and settlement. In the Amazon, much contemporary deforestation and land use occurs in frontier areas populated largely by migrants from other regions of Brazil, with different biophysical characteristics and farming traditions. It is therefore important to account for a farmer's region of birth, as well as their duration of residence. The latter variable is crucial because it gauges the time since arrival, in which farmers can learn appropriate agricultural techniques in the unique environment of the Amazon (Moran 1989). Second, Chayanov assumed that agricultural input, credit and product markets were absent, precluding the possibility for generation of monetary incomes or investment in capital inputs. Agricultural capital and credit are available (though scarce) in the Amazon, and the region's growing urban population constitutes a significant regional market for local agricultural production. The availability of capital and credit allows for investments in machinery or hired labor that can expand deforested areas despite limited household labor. Moreover, the presence of product markets means that farm households can produce a surplus that can be sold for cash income, which implies cultivation of larger land areas than would be necessary for subsistence alone. Third, he assumed that labor markets do not exist, so that farm families cannot hire or sell labor. In Amazon frontier areas, labor markets do exist, and households frequently hire day laborers or send young adults out for wage work. The presence of labor markets is an important additional factor, as hired labor can offset a shortage of household labor and allow larger tracts of land to be cleared and cultivated. Further, the capacity to sell surplus household labor opens the possibility of cash income that can be invested in new agricultural enterprises which may require more or less land use per unit of output. Finally, Chayanov assumed that land use involves a more or less homogeneous set of agricultural practices shared by all households, which negates recognition of options for diversification or specialization in farming systems. In the Amazon, agricultural households may engage not only in farming (i.e., crop cultivation), but also ranching (raising of livestock) and/or forestry (reforestation). Farm households in the Amazon may thus diversify across or specialize in one of these production activities (e.g., Pichón 1996b, Walker et al. 2000). Theoretical Adaptations of Chayanov to Amazonia Walker and Homma (1996) recognize the shortcomings in the Chayanovian model and provide an adaptation of the household life cycle argument to the case of small farm families in the Brazilian Amazon. They situate farm families in a context where labor can be hired or sold, credit and capital are available, production is often destined for markets, and households may diversify or specialize in their land use. The distinctions among land use activities are of particular importance, as each involves different land, labor and capital requirements, and each also bears a particular set of environmental implications. The economic distinctions among land uses provide a means of linking household demographic characteristics to land cover outcomes. This linkage prompts these authors to argue that household demographic composition and change dispose farms to engage in different land uses through the course of their household life cycle. Further, the environmental impacts of each outcome bear important implications for land cover change that follow demographic change at the micro level. I want to set aside

5 the environmental implications until the closing discussion, in order to focus on how Walker and Homma view the co-evolution of household demographics and land use among small farms in the Amazon. Walker and Homma (1996:68-73) articulate a stylized case wherein colonists migrate to the Amazon frontier as young families who establish farms by clearing plots of rainforest. Having spent much of their savings on the move, and often with responsibility for young children, the parents begin by cultivating annual crops (such as rice, beans, corn and manioc). Annuals require considerable labor inputs for clearing, planting, weeding and harvesting, but land and capital requirements are limited. Because annuals produce soon after planting, they constitute a low-risk agricultural strategy. Given the low capital requirements and low level of risk, young households or recent arrivals with limited labor and high child dependency generally plant annuals to secure a basic subsistence. However, because Amazon soil fertility declines with repeated cultivation on a given plot, households must periodically clear more forest to sustain production of annuals, implying a rise in deforestation on farm lots over time. As the seasons pass, farmers gain experience in Amazon agriculture, the labor of growing children makes larger contributions to the household labor pool, and farms accumulate a stock of deforested land that is unfit for further production of annuals. These changes -- learning about locally appropriate agricultural techniques, expansion of available household labor alongside declining child dependency, and more cleared land -- reduce the risk aversion of colonists. They then use the income from early harvests or proof of land claims to obtain credit, purchase capital or hire labor and diversify into more market-oriented activities, particularly perennial crops and/or pasture for cattle. Older households with larger labor pools often plant perennials, or tree crops (such as cacao, coffee, coconuts, and black peppers). Perennials not only involve substantial labor inputs during planting, harvesting and processing, they also require significant capital inputs in the purchase of seed or saplings (Pichón 1996b). Because they need 4-7 years of growth before the onset of production, and because they are subject to insect and fungal attacks, perennials pose greater economic risk to households than annuals (e.g., Pichón 1996b). However, perennials command higher prices than annuals, so the former are eminently cash crops, with production destined for local or regional markets. Perennials also offer environmental advantages because they can be planted on land formerly under annuals, as tree crops can tap nutrients deeper in the soil, and they contribute to soil remediation by providing cover and reducing erosion (e.g., Serrão and Homma 1993). Older households with less available labor often shift land into pasture for cattle. Pasture itself is valuable because it indicates investment in agriculture, which raises land values (Perz 2000a). In addition, ownership of cattle constitutes a capital reserve that acts as an insurance substitute which can be liquidated to cover unforeseen costs, such as from an illness (Pichón 1996c). Smallholders cannot afford to buy many cattle given the high capital cost and extensive land areas required (generally 1 ha per head). Households often convert several adjoining plots previously used for cropland into pasture and purchase a few cattle for breeding and expansion of the herd. Ranching has often been vilified as a cause of deforestation due to the large land tracts required and the unsustainability of production in most pasture grasses, which leads to weed invasions and severe land degradation (e.g., Nepstad, et al. 1991; Serrão & Toledo 1991). However, the low labor requirements and the insurance function of cattle make ranching an attractive land use option among farm households in the Amazon (Tourrand, et al. 1996). A final land use that merits attention is reforestation. Eventually, households age to the point where the children reach adulthood and may inherit their parents' farmstead. In cases where older colonists seek to pass land on to children, farmers may plant valuable trees on their properties for the sake of long-run production of timber. Reforestation involves substantial capital and labor inputs during the purchase and planting of saplings, but plantations require relatively little attention thereafter. The economic benefit of reforestation is the accumulation of value in a growing timber reserve, which serves an insurance function and creates the prospect of additional future income. The environmental benefit of reforestation lies in the regrowth of tree cover in deforested areas, which contributes to soil remediation. Walker and Homma thus show how the economic constraints of land uses in the Amazon dispose households to engage in different farm activities through their domestic life course. All else equal -- such as capital, credit, hired labor, and other factors -- households with younger age structures and shorter durations of residence tend to focus on annuals, while somewhat older households with longer residencies may have more perennials and/or pasture, depending on their household labor availability, and older households yet may engage in reforestation. 5

6 6 McCracken et al. (1999: ) provide a schematic diagram that further specifies the dynamics articulated in Walker and Homma (1996). The McCracken, et al. diagram (1999:1313) links household demographic composition not only to duration of residence in a frontier area, but also the predominant land use in a family farm system in the Amazon. They distinguish between 5 stages of a household's life cycle, where each stage corresponds to a specific demographic composition, duration of residence, and portfolio of land uses. Stage I involves young parents with young children who are newcomers (residence under 5 years) and primarily engage in cultivation of annuals; Stage II sees parents with growing children, a duration of residence around 5 years, who still emphasize annuals but are planting perennials and investing heavily in pasture and cattle; Stage III witnesses older parents with teenage children, a duration of residence around 10 years, a declining emphasis on annuals and heavy investment in cattle activities; in Stage IV, parents are older still, children begin to reach young adulthood, duration of residence is around 15 years, and cattle ranching becomes less important while production from perennials rises; and in Stage V, after 20 or more years, the parents run the property with their grown children, continue some ranching, and reap ever larger harvests from perennials, while perhaps also planting trees for future timber sales. This diagram, while a schematic presentation, provides a timeline that links the life cycle stage of households -- namely, duration of residence and demographic composition -- to predominant land uses, which change over time from annuals to perennials and/or pasture to reforestation. Household Life Cycle Variables and Existing Work on Land Use in the Amazon Walker and Homma's (1996) discussion and McCracken, et al.'s (1999) diagram point to a set of four variables that together "locate" households along a domestic life cycle. These variables are: the age of the household head, the duration of residence, the number of working-age adults, and the number of dependent children. The household head's age indicates their birth cohort, and thus the location of the parental generation in their individual life course. Duration of residence indicates the arrival cohort of a household on the frontier. While age and duration of residence both serve as proxies for farming experience, it is important to distinguish analytically between birth and arrival cohorts, because colonists may arrive in the Amazon frontier at different ages in the same year, and vice versa. The two variables may therefore indicate different things: a person's age marks the demographic location of the parental generation in their life course and period effects associated with their place of birth, while duration of residence denotes their degree of experience farming in the Amazon and period effects of the regional context at the time of arrival. The number of working age adults indicates labor availability within the household, while the number of dependent children measures a constraint on labor from activities other than production for consumption. These household structure variables are independent of the age and duration of residence variables to the extent that children are born over a period of time before or after the arrival of the household on the frontier. This implies that the growth of the household labor pool and decline in the number of dependent children varies somewhat relative to the head's age and the family's duration of residence. Together, these four variables help one "locate" a household in along a continuum of young to old households while recognizing demographic variation among households that are the same in terms of one of the variables. While the preceding discussion provides a theoretical framework for including age of household head, time since arrival, and number of laborers and children in household models of land use in the Amazon, existing models do not do so. Table 1, adapted and updated from Walker, et al. (2000), summarizes the inclusion of demographic variables in household models of land use and deforestation in the Amazon and neotropical regions of Latin America. While most models include one or more of the four household demographic variables discussed above, most sources include two or fewer, and many do not distinguish between adult laborers and dependent children. Only one model includes all four variables (Godoy, et al. 1998a), and only three sources included three (Godoy, et al. 1997a, 1998b, 1998c), but none of these involve colonists in the Amazon. 2 Moreover, only one other source, a descriptive analysis by Marquette (1998), ties both duration of residence and household demographic composition to land use. Given the extensive land cover conversion in the Amazon, there is a need to more fully consider demographic determinants of household land use in order to better understand the "human dimensions" of deforestation there. (Table 1 about here)

7 7 Data and Methods The Uruará Survey The central question of this paper concerns the significance of household demographic factors (such as duration of residence and adult labor availability) for land use outcomes in the Amazon (such as the cultivation of annual crops or cattle ranching). Data for an empirical assessment of this linkage are available from a recent survey of small farm households in Uruará, a colonist community situated on the Transamazon highway in the Brazilian state of Pará, located in the eastern Amazon (see Map 1). 3 Uruará began in the early 1970s as a colonization project to resettle landless peasants from the Brazilian Northeast (IDESP 1990). Colonists were given lots of 100 hectares (247 acres) and began cultivating annuals, later diversifying into perennials, and most recently moving into cattle pasture, with some reforestation. In the mid-1980s, high prices for perennial crops stimulated a second wave of in-migration, raising the municipality's population to about 25,000 by 1991, with over 11,000 being migrants since 1980 (IBGE 1996). Uruará's population has since risen further, exceeding 37,000 by 1996 (IBGE 1998b). Meanwhile, 1154 km 2 (10% of the municipality land area) was deforested by Because this community consists almost entirely of small farms, because it has experienced deforestation as part of frontier expansion into the region, and because it is situated near research sites in the Amazon where Walker and Homma (1996) and McCracken, et al. (1999) gathered data in developing their theoretical frameworks, Uruará is an appropriate site for an assessment of how demographic factors in small farms affect land use in the Amazon. (Map 1 about here) In June and July 1996, a 9-member research team consisting of North American and Brazilian social and agricultural scientists administered a survey questionnaire to farm households in Uruará. The questionnaire was divided into two parts, where the first addressed household characteristics and the second concerned land use practices. Household items included region of birth and migration history, ownership of durable goods and agricultural capital, and demographic characteristics of the household head and the age composition of the families present. Items pertaining to land use practices included the land area under annuals, the number of productive perennials, the land area under pasture, the number of heads of cattle, and whether reforestation was practiced. The sample includes 261 farm households, or 12% of all rural establishments in Uruará at the time (IBGE 1998b). These households together owned 347 lots, and the same questions were asked for each lot owned by a household. Systematic sampling proved intractable because houses on many lots were not visible from roadsides, and sampling the "nth" house encountered was problematic because residents were frequently absent. Instead, the team sampled on the basis of "first opportunity" and employed a cadastral map from the regional agricultural agency, EMBRAPA/ CPATU, to ensure that samples were not clustered spatially or selective of households by socioeconomic status. The resulting data set thus constitutes as random and representative a sample of Uruará's rural households as possible. Operationalization and Measurement of Outcomes and Independent Variables Table 2 presents operational definitions and descriptive statistics for the land use outcomes, household demographic variables, and a set of other independent variables that may also influence small farm land use. Following Walker and Homma and McCracken, et al., I consider five land use outcomes: 1.) the number of hectares (ha) under rice, beans, corn and manioc (logged to obtain a normal distribution) to indicate land use under annuals, 2.) the (logged) number of productive cacao, coffee and coconut trees and black pepper vines to indicate land use for perennials, 3.) the (logged) ha under pasture to indicate land allocated to ranching,4.) the (logged) head of cattle as a second ranching indicator, somewhat independent from pasture as many lots in Uruará have pasture but no cattle, and 5.) a binomial variable that indicates whether a lot had reforestation (coded 1) or not (coded 0). The descriptive findings (antilogs of mean logs) show that surveyed lots had, on average, 1.77 ha under annuals, 119 productive perennial plants, 10 ha under pasture, 6.2 cattle, and that 6% practiced reforestation. While this might indicate substantial diversification, the standard deviations for each outcome also indicate considerable variation and the potential for specialization in land use in individual lots. (Table 2 about here)

8 8 Before focusing on the household demographic variables, it is worthwhile to consider four groups of competing factors that may also affect land use: farmer background, neighborhood context, institutional context, and off-farm income. 5 I operationalize farmer background in terms of region of birth and initial wealth brought to the frontier. Region of birth indicates whether the head of household was born in relatively developed Southern Brazil or elsewhere, and allows assessment of cultural differences in farming backgrounds. Initial wealth is a factor-weighted index that accounts for ownership of durable goods and initial housing quality, which assesses the initial capital possessed by a household, which can offset limited labor in expanding or diversifying a production system. 6 Neighborhood context involves three indicators: the ordinal lot number owned by a household, the distance to Uruará town, and whether a lot was located in an organized neighborhood. About 25% of households owned more than one lot, and the lot number variable indicates whether a lot was the first, or second or higher order lot owned. This is important as production systems on first lots are generally more advanced from annuals toward other activities than second or other lots, usually obtained later in time. Distance to town measures market distance, a proxy for transportation costs, which are high on the muddy roads of the area and reduce profits from agriculture, in turn eroding incentives for commercial production. Mean lot distance to Uruará town was over 30 km, but this varied substantially within the sample. Neighborhood organization, present among one-third of lots surveyed, indicates whether neighboring households on a given feeder road to the Transamazon highway were mobilized against land invasions and for cooperative labor arrangements, transport of production to town, and similar activities beneficial to commercial production. Institutional context refers to titling status and use of credit. Titling status indicates whether a lot had title (i.e., was claimed formally) or not (i.e., claimed informally). Privatization theory argues that formal titling provides secure tenure and the incentive for longer planning horizons that allow for more risky but productive investments (Alston, et al. 1993, 1996; Wood and Walker 2000). It is not clear in the weak institutional context of the Amazon whether neighborhood organizations or formal titles contribute more to expansion of production systems from annuals to other activities. In general, however, titles are necessary to obtain credit that allows for investment in land use expansion, particularly for cattle. Nearly 60% of surveyed lots were titled, and about 45% had activities funded at least in part by credit. Use of credit is an important control as it supplements or offsets limited household labor by allowing for investments in capital or hired labor. Finally, off-farm income accounts for remittances and income from urban-based businesses. As noted earlier, households may send family members to town or other lots to sell labor for wages, often remitted back to the farm. In the survey, this occurred among households who owned 11% of the lots in the sample. In addition, 9% of lots were owned by households with urban businesses, a potential source of investment funds. An important question is whether those households used remittances and business incomes for investments in agriculture, which would enlarge production systems, or for other activities, which would imply land ownership for speculative rather than productive purposes. Aside from farmer background, neighborhood and institutional contexts and off-farm incomes, household demographic factors should exert important effects on land use. I locate Uruará households along their domestic life cycles using the four demographic variables noted earlier, plus a variable that indicates the hiring of labor. 7 The descriptive statistics show that the mean age of household heads was nearly 50; that duration of residence was nearly 12 years; the number of adults (age 15 and up) was over 4; the number of children (under age 15) was nearly 3; and that the days of hired labor paid in the previous year was just under While these averages suggest older household heads and long durations of residence, as well as significant labor pools, child dependency and modest labor hiring, the standard deviations reveal considerable variation among households, implying a wide range of life cycle "locations" present in the sample. Findings Bivariate Correlations between Outcomes and Independent Variables Table 3 presents bivariate correlation coefficients between each land use outcome with the other outcome variables and every independent variable. Correlations among the outcomes indicate modest associations among different land uses on the lots surveyed. One exception concerns pasture and cattle, with a correlation that is high (r=0.60), but far from perfect. These findings indicate that most of the outcomes are largely independent of one another, and deserve separate consideration with respect to influences by household demographic variables and other factors. The rest of the coefficients in Table 3 bear out this supposition, in that the independent variables with statistically

9 9 significant coefficients vary from one land use to the next. For annuals, farmer background is relatively important while household demographic factors are relatively unimportant; the converse appears true for the other land use outcomes. (Table 3 about here) In general, the non-demographic independent variables exhibit the expected correlations with the land use variables. More land was under annuals among farmers born outside Southern Brazil, those with less initial wealth, second and higher order lots, lots farther from the Uruará market, lots with titles and credit, and lots owned by households without a business income. The correlations for birth and initial wealth are weaker for perennials, but market distance shows a strong negative association with commercial tree crops, while titling and credit conversely show strong positive correlations. While the significant variables for pasture and cattle are similar, lot number is more important for pasture, while market distance is more important for cattle; similarly, titling and especially credit is more crucial for cattle than pasture. Correlations with reforestation are in general weaker than for the other land use outcomes, with titling status, duration of residence and hired labor exhibiting the strongest associations. Before discussing the observed correlations between the outcomes and the demographic variables, it is important to outline expectations derived from the theoretical model of life cycle influences on land use. The columns in Table 3 are organized more or less along the sequence of land uses in which smallholders engage through their life cycle, from annuals to reforestation. Each land use implies specific labor demands, capital and land requirements, and risks and returns for the household. As a low-risk subsistence activity, annuals may be more important among households "early" in their life cycle with young heads, short durations of residence, few adults, and many children. In other words, the land area under annuals may show a negative association with age, length of residence, and number of adults, but a positive association with number of children. Perennials, a laborand capital-intensive activity with higher risks and returns, should be more prevalent "later" in the life cycle, among households with older heads, longer residence durations, more adults, and fewer children, and perhaps more hired labor. Pasture and cattle should have similar associations with demographic variables since these activities take precedence later in the life cycle like perennials, but because ranching is less labor intensive, the importance of adults and the negative effects of children may be weaker, particularly for pasture. Reforestation, which theoretically comes "latest" in the life cycle, should exhibit strong positive associations with age and residence duration, and weak relationships with adults and children, given the limited labor requirements. The observed correlations between demographic and land use variables largely follow expectations. The main exception concerns annuals, which show no statistically significant correlations with any demographic indicator, suggesting that regardless of life cycle location, households around Uruará grow annuals. By contrast, as expected, perennials are more important to households "later" in the life cycle, with longer residence durations, more adults, and more hired labor, a reflection of the lagged onset of cash crop production and the need for family or paid workers. Similarly, both pasture and cattle are more important in households with longer residence durations, more adults, fewer children, and to some extent hired labor. Note that the correlations are stronger for cattle than pasture. Finally, duration of residence is positively associated with reforestation, which follows expectations that tree planting comes "late" in the life cycle. As expected, household labor is not important for reforestation, but it is interesting to note that hired labor is significant, perhaps because of the absence of adult children who have left older households or engaged in other activities. Multivariate Model Results Given the observation of significant correlations between household demographic variables and the land use outcomes, it is important to know the effects of the household life cycle net of other factors. Table 4 presents the results of OLS models of annuals, perennials, pasture and cattle, and a logit model of reforestation, regressed on variables for farmer background, neighborhood context, institutional context, off-farm income and household life cycle. 9 (Table 4 about here)

10 10 The primary determinants of the extent of land under annual crops involve a farmer's background and the neighborhood context. Lots owned by farmers born outside Southern Brazil and those with less initial wealth had more land under annuals, as did second and higher order lots and those farther from Uruará town. The benefit of neighborhood organization is apparent for annuals, while that of formal title and credit is not. None of the household demographic variables show significant net effects on land under annuals, which again suggests that regardless of life cycle location, households cultivated crops such as rice and beans. In contrast, the determinants of having productive perennials emphasize demographic factors and other variables that were not significant for annuals. The number of productive perennials declines substantially with market distance and the availability of business income, which suggests that households with urban businesses invest in other activities. Duration of residence exerts a very strong and positive but nonlinear effect on productive perennials: their number increases over time, but at a decreasing rate, perhaps as a reflection of eventual declines in their production as households reach later stages of the life cycle. The availability of household labor is crucial to having productive perennials, as is the use of hired labor, again a reflection of the high labor demands in harvesting and processing cacao, coffee and other cash crops. The land area planted under pasture varies depending on a wide array of factors. Southerners had more pasture, as did first lots, those closer to town, and those in organized neighborhoods. As expected, pasture area is greater on lots with credit. Both remittance and business income reduced pasture area, again suggesting that wage and commercial income was invested in non-agricultural activities. Demographic variables were very important, and they again show independent effects following the theoretical model. Households with longer durations of residence had more pasture, and the increase attenuates over time. Household composition also influences pasture area, as older families (with more adults and fewer children) had more pasture. The model for cattle is stronger and has similar significant variables, but differences are evident. The background, neighborhood context, and off-farm income variables are less important for cattle than for pasture, while a formal title and especially use of credit were crucial to breeding larger herds. Demographic variables again show strong effects that follow the life cycle model: households with longer durations of residence, more adults, fewer children, and more hired labor had more cattle. Finally, the reforestation model reveals a limited array of significant factors. Lots with title, and those owned by households with business incomes were 3 to 4 times more likely than untitled lots and households without urban businesses to engage in reforestation. This finding might suggest that some households with businesses invested their profits in tree plantations rather than annuals, perennials or cattle. In addition, duration of residence shows a significant positive effect: for every year of residence, the probability of reforesting rose by nearly 50%. Given the low prevalence of reforestation, this implies that only households with very long residence durations reforested. Household composition was unimportant to reforestation, but hired labor was highly significant. This again implies the need for paid workers in the absence of household labor for reforestation, likely to be the case in the latest stages of the household life cycle where grown children have left or engaged in other activities. Discussion This paper outlined a theoretical basis to include household demographic variables as life cycle determinants in models of land use. Findings from empirical models that account for household life cycles show that in the case of small farms in Uruará, in the Brazilian Amazon, demographic factors exert significant effects on land use and land cover change. While this was not true for annual crop cultivation, the models showed the expected effects of residence duration, number of adults, and number of children for productive perennials, pasture, cattle and reforestation. These latter four land uses tend to occur in older households with longer residence durations, and perennials cultivation is more characteristic of households with larger labor pools, while this was less so for pasture and cattle, and not at all the case for reforestation. Thus, most land use outcomes reflect the location of households along their life cycle, following expectations derived from Walker and Homma (1996) and McCracken, et al. (1999). This household level approach provides a means to account for the "demographic dimensions" of land cover change at a micro-level that complements existing macro-level appraisals of population-environment interactions. It also provides a more complete evaluation of how different aspects of household life cycles can influence land use and cover change. A key implication of the theoretical framework concerns the environmental ramifications of aging households in frontier areas of the Amazon. To review briefly, annuals cultivation does not use much land at a

11 given time, but requires substantial deforestation over time since yields on a given plot decline; perennials offer environmental advantages because they can be planted on land previously under annuals, and because they provide shade that contributes to soil remediation; pasture for cattle requires extensive land areas which are subject to soil erosion and compacting, leading to weed invasions that prevent forest succession; and reforestation offers environmental services similar to perennials that aid in soil remediation. The life cycle framework for the case of land use in the Amazon specified the predominance of annuals among young households, followed by cattle pasture (in labor-poor households) and/or perennials (in labor-rich households) among older farms, followed by reforestation among those yet older. There is thus a sequence of predominant land uses through the course of a household's life cycle, from annuals, which degrade land through exposure to wind and rain, followed by potential remediation under perennials or further degradation under pasture, perhaps followed by remediation under reforestation. The findings generally confirm the changing predominance of land uses across household life cycles insofar as older households tend to have more perennials, cattle pasture and reforestation. This indicates that older households may experience greater deforestation and land degradation, but this is dependent on the land use trajectory, specifically the degree to which households continue to farm annuals, favor perennials or pasture, and engage in reforestation. The findings presented here and recent historical turns in Uruará suggest that there is more resource degradation than remediation underway, and that the degradation in part reflects the effects of households advancing along their life cycles. First, the findings indicated no effects of household demographic factors on the cultivation of annuals, which means that households tend to continue clearing forest for plots of rice and beans. While the usual caveats about making longitudinal interpretations from cross-sectional data apply here, similar findings are reported by Walker and Homma (1996:72-73) using data from studies at two points in time in settlements along the Transamazon near Uruará. Second, the descriptive findings presented here show that few households (6% of the sampled lots) engage in reforestation, which suggests that this is not a widespread practice around the study region. Third, the importance of perennials in Uruará has decreased since the mid-1980s due to fungal diseases, which have reduced production, as well as price declines, which have reduced returns to tree crops (IDESP 1990). While these changes constitute period effects on land use, they had their greatest impact on older households with productive perennials. One response to a decline in the importance of perennials among older households is to shift land into pasture for cattle. The Uruará sample data show that the average area under annuals and perennials (calculated as the antilog of the mean logs) was 2.94 ha, while that under pasture was slightly over 10 ha. The emphasis on cattle reflects shifts from crops to pasture in the Amazon as a whole: between 1985 and 1996, the land area under annual and perennial crops declined from 5.9 to 5.7 million ha, while the area under pasture rose from 43.4 to 52.1 million ha (IBGE 1990, 1998b). The limited importance of reforestation and the decline of perennials point to reductions in land use with the potential for soil remediation, while continued cultivation of annuals across life cycle locations and over time and the shift to pasture suggest the perpetuation or rise in land use that risks degradation. The shift from perennials to cattle raises many questions about pasture sustainability, as pasture degradation is common in the Amazon, even to the point where the forest cannot regrow (Nepstad, et al. 1991; Serrão and Toledo 1991). Further, there are growing concerns about timber extraction in remaining forests as a means of financing pasture remediation (Almeida, et al. 1996). Selective logging opens forest canopy gaps and increases fire risks (Nepstad, et al. 1999), as witnessed by the fires in the Amazon during The concern over land degradation under colonist land use systems has prompted many to call attention to indigenous land use strategies such as agroforestry, which combines crop cultivation with long fallow periods, where fallows include economically important tree crops planted among other species to hinder pest attacks (e.g., Dufour 1990). Recent demographic changes underway in the Amazon and the life cycle framework presented here suggest that current land use patterns may not persist. The pace of rural population growth in the Amazon has slowed, from an annual rate of 1.5% per year during to only 0.3% per year during (IBGE 1983, 1996, 1998a). This was due in part to a slowdown in regional population gains due to net migration (Perz 2000b). The slowdown in population growth due to migration implies that contemporary land use in the Amazon is the result of existing populations in frontier areas. As households in these areas progress through their life cycles, the current predominance of pasture may also change. The life cycle framework suggests that as households continue to move to later stages of their life courses, they may engage in more reforestation. However, this remains to be seen. To an extent, this may depend on how land turnover proceeds among farm households. This calls for greater attention to the age structure of colonists in frontier areas, because the young parents who arrived in the early 1970s are now retiring and turning over their properties. A "second generation" of landowners is occupying 11

12 family farms along the Transamazon, and they inherit a landscape very different from their predecessors. At issue is the effect of the change in the demographic composition of land users in the Amazon. On the one hand, if the new households go through their life cycle and use land as their forebears, we might expect reductions in the extent of land use as pasture is allowed to go fallow (or is planted in timber) and the second generation focuses on smaller plots of annuals. On the other hand, the new households may choose to build on the accomplishments of the first generation and continue running cattle on pasture, perhaps even expanding cleared areas. It is very likely that the second generation's life cycle, as it influences land use, will be very different from that of the first. Because the second generation begins with more cleared land and greater knowledge of local agriculture, demographic changes may move small farms along a different future land use trajectory than the historical record shows. The question of the changing influences of household life cycles on present and future land use in the Amazon requires continued attention, and serves as one case among others for the application of variations of the life cycle model to expand and improve environmental research in demography and the social sciences in general. The adaptation of household life cycle models to different contexts, as exemplified by the Walker/Homma adaptation of Chayanov to the Amazon, is important here. While it is true that the household demographic variables employed are in some ways specific to the Amazonian case taken up here, the importance of household life cycles, understood broadly as a series of stages through which families pass, nonetheless applies to different contexts. In any case, the age of individuals and the age composition of households may emerge as key influences on environmental outcomes, whether that means the production of pollution or the consumption or degradation of resources. For example, in the US, resource consumption has been shown to vary over the life course by the age of householders (US Bureau of Labor Statistics 1998, cited in Pebley 1998: ). Demography is beginning to broaden its scope of attention to environmental questions (Pebley 1998). For example, some demographers have made arguments and gathered empirical evidence that show that migration is a key demographic factor in exposure to pollution hazards in Brazil (Hogan 1992), the US (Hunter 1998), and elsewhere (O'Lear 1997). Other students of population have begun to examine demographic factors tied to risks of financial losses in natural disasters (Aguirre, et al. 1993; Smith & McCarty 1996). However, there remains a need for more attention to household models that incorporate demographic factors as life cycle determinants of resource use and environmental pollution. 12