21
Beyond Population and Environment: Household Demographic Life Cycles and Land Use Allocation Among Small Farms in the Amazon Stephen G. Perz & Robert T. Walker & Marcellus M. Caldas Published online: 6 July 2006 # Springer Science+Business Media, Inc. 2006 Abstract Most research featuring demographic factors in environmental change has focused on processes operating at the level of national or global populations. This paper focuses on household-level demographic life cycles among colonists in the Amazon, and evaluates the impacts on land use allocation. The analysis goes beyond prior research by including a broader suite of demographic variables, and by simultaneously assessing their impacts on multiple land uses with different economic and ecological implications. We estimate a system of structural equations that accounts for endogeneity among land uses, and the findings indicate stronger demographic effects than previous work. These findings bear implications for modeling land use, and the place of demography in environmental research. Key words Population . environment . land use . land cover change . Amazon. Introduction Concern about demographic factors and environmental damage has generated a large literature featuring a population and environmentdiscourse (e.g., Arizpe et al., 1994; Lutz et al., 2002; Mazur, 1994; Ness et al., 1993; Pebley, 1998). Much of this literature draws on Malthusian and Boserupian perspectives and focuses on demographic phenomena in large-scale aggregates, particularly nation-states or the planet. One often encounters the use of demographic techniques such as population projections (e.g., MacKellar et al., 1998) or decompositions (e.g., Bongaarts, 1992), often involving the IPATidentity (e.g., York et al., 2003). The population and environmentdiscourse in other disciplines also tends to Hum Ecol (2006) 34:829849 DOI 10.1007/s10745-006-9039-8 S. G. Perz (*) Department of Sociology, University of Florida, 3219 Turlington Hall, PO Box 117330, Gainesville, FL 32611-7330, USA e-mail: [email protected] R. T. Walker : M. M. Caldas Department of Geography, Michigan State University, 116 Geography Building, East Lansing, MI 48824-1117, USA

Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Embed Size (px)

Citation preview

Page 1: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Beyond Population and Environment: HouseholdDemographic Life Cycles and Land Use AllocationAmong Small Farms in the Amazon

Stephen G. Perz & Robert T. Walker &

Marcellus M. Caldas

Published online: 6 July 2006# Springer Science+Business Media, Inc. 2006

Abstract Most research featuring demographic factors in environmental change hasfocused on processes operating at the level of national or global populations. This paperfocuses on household-level demographic life cycles among colonists in the Amazon, andevaluates the impacts on land use allocation. The analysis goes beyond prior research byincluding a broader suite of demographic variables, and by simultaneously assessing theirimpacts on multiple land uses with different economic and ecological implications. Weestimate a system of structural equations that accounts for endogeneity among land uses,and the findings indicate stronger demographic effects than previous work. These findingsbear implications for modeling land use, and the place of demography in environmentalresearch.

Key words Population . environment . land use . land cover change . Amazon.

Introduction

Concern about demographic factors and environmental damage has generated a largeliterature featuring a “population and environment” discourse (e.g., Arizpe et al., 1994;Lutz et al., 2002; Mazur, 1994; Ness et al., 1993; Pebley, 1998). Much of this literaturedraws on Malthusian and Boserupian perspectives and focuses on demographic phenomenain large-scale aggregates, particularly nation-states or the planet. One often encounters theuse of demographic techniques such as population projections (e.g., MacKellar et al., 1998)or decompositions (e.g., Bongaarts, 1992), often involving the “IPAT” identity (e.g., Yorket al., 2003). The “population and environment” discourse in other disciplines also tends to

Hum Ecol (2006) 34:829–849DOI 10.1007/s10745-006-9039-8

S. G. Perz (*)Department of Sociology, University of Florida, 3219 Turlington Hall,PO Box 117330, Gainesville, FL 32611-7330, USAe-mail: [email protected]

R. T. Walker :M. M. CaldasDepartment of Geography, Michigan State University, 116 Geography Building,East Lansing, MI 48824-1117, USA

Page 2: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

focus on aggregate level processes, such as research by geographers on population-inducedagricultural intensification (e.g., Turner et al., 1993).

At the same time, it is widely acknowledged that human impacts on environmentsinvolve scale-dependent processes. This is in part due to recognition by biophysicalscientists that distinct processes are at work as one moves from the cellular to the landscapelevel (e.g., O’Neill et al., 1986). Attention to scale is also becoming more evident inenvironmental social science frameworks, which seek to incorporate factors ranging fromindividual agency to international politics (e.g., Gibson et al., 2000; Wood, 2002). Giventhe growing literature on scale dependency in human-environment interactions, demo-graphic environmental research needs to move beyond the “population and environment”discourse to consider demographic processes operating on other scales and their impacts onenvironmental change.

This paper presents an analysis of household-level demographic processes and theirenvironmental consequences. We take up the case of frontier colonist households in theBrazilian Amazon and assess how the “demographic location” of a household affects itsland use allocation. By “demographic location” we refer to a constellation of factors—duration of residence, age structure, and generational transitions—which together delineatethe position of a household along its life cycle. We draw on economic anthropology andhousehold economics and present a household-level framework based on Chayanoviantheory fused with household production theory.

The analysis models the effects of household demographic life cycles on land useallocation, and advances beyond previous efforts in three respects. First, we consider abroader suite of demographic variables than has generally been the case in household landuse models. Second, the analysis evaluates land allocation among several land uses, ratherthan deforestation and other “single-outcome” approaches. In particular, we distinguishdifferent types of crops, which is theoretically important but rarely done empirically. Third,the modeling approach explicitly recognizes that allocation of land to one use constitutes anopportunity cost for allocation to other uses, making land use decisions mutuallyendogenous. We therefore use three-stage least squares (3SLS) estimation and specify asystem of equations which accounts for endogeneity among land uses and yields unbiasedestimates of the effects of household demographic factors. The findings show strong effectsof the household life cycle variables, which bears implications for understandingdemographic impacts on environmental processes at different scales.

Theoretical Background

Household Demographic Life Cycles and Land Use

The link between household demographic life cycles and land use was first featured byChayanov (1986[1966]), who observed that peasant households in post-revolutionaryRussia contained families with different age structures, and that those households alsofarmed different quantities of land. He reasoned that age structures are older in householdswith larger numbers of economically active adults and/or smaller numbers of dependentchildren. Both allow for greater allocation of labor to agriculture, which in turn enablescultivation of larger land areas. As a result, Chayanov argued that “demographicdifferentiation” among farm households explained differences in their land use. Chayanovthus characterized the “peasant economy” via its emphasis on family labor availability (e.g.,Harrison, 1975; Hunt, 1979).

830 Hum Ecol (2006) 34:829–849

Page 3: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Chayanov also distinguished among stages in household life cycles on the basis of thetiming of demographic events, particularly fertility and the onset of labor contributions byolder children. Initially, land use by young parents is limited due to low labor availability.With the onset of childbearing, households exhibit increased dependency without addedlabor. The rise in consumption therefore impels expanded land use via the increased“drudgery” of the young parents. As the children grow up, they add to the household laborpool, facilitating further expansion of family land use. In late stages of the life cycle, childrenreach adulthood and either leave the household or remain to inherit the farm, formingmultigenerational households. At this “generational transition,” consumption and laboravailability may decline, resulting in a decrease in land use. However, this trend may bereversed when the next generation begins to move through its life cycle.

Chayanov’s insights have been generalized from post-revolutionary Russia to othercontexts by economic anthropologists and other scholars (e.g., Chibnik, 1984). In developingregions, households also rely on agricultural production via family farming, land is abundant,and markets and access to capital are limited.

Markets and Household Production in Developing Regions

Nonetheless, contemporary developing countries exhibit changes left untreated by Chayanov,such as the formation of markets for land, labor, and capital. These alter the peasant economiccalculus and potentially undermine the importance of household demography as anexplanation for land use. As a result, household production theory emerged to account formarkets via linked production and consumption decisions (e.g., Ellis, 1993; Singh et al.,1986; Walker, 2003).

Household production theory features the role of markets in stimulating agriculturalcommercialization, proletarianization, and the shift of livelihoods toward non-farm activities.First, developing regions have markets for credit, agricultural inputs and agriculturalproducts. Such markets open possibilities for households to acquire capital and substitute itfor labor in land use, such as using a chainsaw to cut trees. Markets also make it possible toproduce cash crops for commercial sales rather than household consumption, which canexpand household land use beyond subsistence demand. Second, developing regions alsohave labor markets. The presence of labor markets implies that farm households cannot onlyhire labor, which changes the effective labor pool available for agriculture, but also sell labor,which provides an income stream for investment.

Study Region: The Brazilian Amazon

We take up the case of the Brazilian Amazon as a developing region in which to assess theimpact of household demography on land use. The Amazon is an important study casebecause it is the world’s largest contiguous rainforest biome and is experiencing rapiddemographic expansion and land cover change. The population of the Legal Amazon rosefrom 5.3 million in 1960 to 19.6 million in 2000 (IBGE, 1962, 2000). Similarly, forestclearing in the Amazon has risen from 152,200 km2 in 1978 to 587,700 km2 in 2000(INPE, 2002). The Amazon is also a good choice for the study of household farming. In1996, the Legal Amazon had some 900,000 rural establishments (IBGE, 1998a). Of these,over 800,000 establishments were under 200 ha in size, and in most cases these are family-run operations.

There are aspects of family demography and land use specific to the Amazon. First,households on the Amazon frontier contain colonists who migrated from other regions of

Hum Ecol (2006) 34:829–849 831

Page 4: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Brazil. This requires attention to a household’s duration of residence. The length of timesince arrival has been viewed as important for allowing adaptation of land use strategies tothose suitable in the Amazonian environment (Moran, 1989). Second, colonist agriculturalsystems in the Amazon vary substantially. Some households emphasize subsistence crops,others feature cash crops, yet others focus on cattle ranching, and some engage indiversified systems (e.g., Serrão and Homma, 1993; Walker et al., 2002). These farmsystem components require different quantities of labor inputs and have distinct economicand ecological ramifications. Consequently, it is not adequate to assess land use in terms ofthe extent of land in use as in Chayanovian theory, or in terms of agricultural and non-agricultural activities as in household production theory. And third, Amazonian agricultureinvolves agricultural fallows and degraded land. Chayanovian and household productiontheories both focus on productive activities, neglecting fallowing practices wherein land istemporarily allowed to rest, or land degradation where poor soils are taken out ofproduction (Perz and Walker, 2002; Walker, 2003). It is therefore crucial to account forsecondary vegetation, which appears in fallows and on degraded land.

Household Demographic Life Cycles and Land Use Allocation in the Amazon

This section presents a theoretical framework that fuses Chayanovian thought withhousehold production theory, and adapts the two to the case of the Amazon. We draw onprevious articulations of demographic processes involved in household life cycles in theAmazon (Brondizio et al., 2002; Marquette, 1998; McCracken et al., 2002; Perz, 2002;Perz and Walker, 2002; Walker, 2003; Walker and Homma, 1996; Walker et al., 2002). Ourcentral argument is that even given markets and factors specific to frontier areas of theAmazon, land use allocation should still vary among households at different points in theirlife cycles because they have distinct demographic characteristics.

We present our theoretical framework via the stylized case of a colonist household. Thelife cycle begins when migrants relocate to the frontier. They come as young families,whether as childless couples or parents with young children, and establish land claims byclearing primary forest. Having spent much of their savings on the move, and often withresponsibility 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 annualsproduce soon after planting, they constitute a low-risk agricultural strategy. However,because Amazon soil fertility declines with repeated cultivation on a given plot, householdsmust periodically fallow land and clear more forest to sustain production of annuals. Thus,early in the life cycle of a colonist household in the Amazon, primary forest area declines,and land allocated to annual crops and regrowth expands.

As the seasons pass, farmers gain experience in Amazonian agriculture, the labor ofgrowing children makes larger contributions to the household, and farms accumulate astock of deforested land. These changes reduce the risk aversion of colonists, who seek toobtain credit to purchase capital or hire labor and engage in market-oriented farmingactivities, particularly perennial crops and pasture for cattle. Thus, later in the householddemographic life cycle, primary forest declines further as colonists allocate more land toperennials and pasture.

Older households with larger labor pools often plant perennial crops such as cocoa,coffee, coconuts, and black peppers. Perennials not only involve substantial labor inputsduring harvesting and processing, they also require significant capital inputs for purchaseand maintenance. Because perennials require several years of growth before the onset of

832 Hum Ecol (2006) 34:829–849

Page 5: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

production, and because they are subject to insect and fungal attacks, they pose greatereconomic risks to households than annuals. However, produce from perennials oftencommands high prices. Perennials also offer environmental advantages because they can beplanted on land formerly under annuals, and they protect soils by providing more permanentland cover.

Households with less labor often allocate land to pasture for cattle. Pasture is valuablebecause it indicates investment in agriculture, which raises land values, and ownership ofcattle constitutes a capital reserve that acts as an insurance substitute to cover unforeseenexpenses, such as for illness (e.g., Tourrand and Veiga, 2003). But smallholders cannot affordto buy many cattle given the high initial investment involved, and ranching has often beenvilified environmentally due to the large land tracts required, and because many pastureshave not been managed sustainably, leading to land degradation (e.g., Serrão and Toledo,1990). However, cattle are an attractive land use option due emerging urban markets forbeef in the Amazon (Faminow, 1998).

Late in the household life cycle, different trajectories may occur. One involves out-migration of young adults as they leave to establish their own farms or find urbanemployment (McCracken et al., 2002). Labor availability and subsistence demand declines,leading to a reduction in the land area under crops and pasture, and further expansion ofsecondary growth. However, another trajectory is possible if grown children stay in theparental household (Perz and Walker, 2002). This reflects a “generational transition” as onegeneration passes control of the property to the next. This is particularly likely if the youngadults are parents with young children, for the farm provides the security of an establishedenterprise. In this scenario, forest clearing may expand to make way for agriculture asyoung children expand demand for subsistence.

Household Life Cycles and Land Use in Previous Research

Following the foregoing discussion, we find it necessary to account for a suite of householddemographic factors: duration of residence, age structure, and generational transitions.Length of residence captures the effects of agricultural experience on land allocation. Inaddition, the age structure of a household, measured in terms of the number of children,working age adults, and elderly, is necessary to evaluate dependency and labor availability.Because colonists do not arrive on the frontier at exactly the same ages, and because thetemporal distribution of fertility events varies among households, age structure effects aredistinguishable from residence duration effects. We view generational transitions as occurringin multigenerational households with grandparents and grandchildren, for this implies apassing of responsibility of the farm from grandparents to new parents.1

Few previous studies on household land use have considered duration of residence, agestructure, and generational transitions. Most prior models of household land use in LatinAmerica consider only one of these factors, if any (Walker et al., 2002). This oversight is

1 We considered other approaches to measuring household demographics, but found no satisfactoryalternatives. One reviewer argued to aggregate cohorts instead of using duration of residence in single years,but this presents problems because there are many factors to consider for defining cohorts, which could resultin many possible cohorts, and greatly affect the findings. Many analysts employ the age of the householdhead as a life cycle indicator, but this says little about past fertility events or overall household age structure.Others have employed “Chayanovian” dependency ratios, calculated as the units of labor divided by units ofconsumption, but these fail to distinguish between youth and elderly dependency. We also avoid “true”dependency ratios because they are unstable at the household level.

Hum Ecol (2006) 34:829–849 833

Page 6: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

intriguing because studies that have incorporated most or all of these demographic factorshave found strong effects on land use (e.g., Coomes et al., 2000; Perz, 2002; Pichón, 1997).

Our theoretical framework also features five land uses: primary forest, annual crops,perennial crops, cattle pasture, and secondary vegetation. Previous work on household landuse in the Amazon rarely involves analyses with more than one or two of these uses. Suchwork fails to distinguish among land uses with different economic and ecologicalconsequences, and which become important at different moments in a household’s life cycle.

Data and Methods

Study Site

Our study case in the Amazon is Uruará, a colonist community situated on the Transamazonhighway with a township located at Lat. 03° 42’ 54” S, Long. 53° 44’ 24’’W in the Brazilianstate of Pará (IDESP, 1990). Uruará was founded in the 1970s as a colonization project toresettle rural families from the Brazilian Northeast. The state land-titling agency, INCRA,surveyed and distributed lots of roughly 100 hectares (ha) to a first wave of colonists. In themid-1980s, perennials such as cocoa and black pepper commanded high prices, whichprompted households to expand their clearings for cash crops. This stimulated a second waveof in-migration, raising the municipality’s population to 25,000 by 1991 (IBGE, 1996).

This dynamism gave way to difficulties in the 1990s, as pest attacks reduced cash cropproduction and price declines reduced agricultural incomes. This crisis led to a shift in landuse toward pasture for cattle (Toni, 2003), and catalyzed the emergence of social movementsseeking to improve colonist living standards (Nascimento and Drummond, 2003). Localorganizations served as conduits for new credit programs aimed at small producers, andused the financing for pasture formation and livestock purchases (Toni, 2003). By 2000,23% of the forest cover in the municipality had been cleared (Nepstad et al., 2000, cited inNascimento and Drummond, 2003, p. 126).

Uruará is an appropriate site for an assessment of how household demographic life cyclefactors affect land use allocation. First, this community consists almost entirely of smallfarms that rely primarily on family labor. Second, the Transamazon highway corridoraround Uruará exhibits substantial deforestation for various land uses, but also substantialsecondary growth.

Data Collection

In June and July 1996, a nine-member research team consisting of North American andBrazilian social and agricultural scientists administered a survey questionnaire to farmhouseholds in Uruará. The questionnaire was divided into two components, where the firstaddressed household characteristics and the second concerned the lot(s) held by households.The household component included items such as family age composition, sources ofincome, and material wealth. The lot component included items such as land use, access tocredit, use of agricultural technologies, and distance to market.

Systematic sampling of farm lots proved intractable because not all lots had houses.Moreover, systematic sampling of houses encountered was problematic because residentswere sometimes absent. We therefore sampled by “first opportunity” of residents encounteredon their lot. We employed a cadastral map of Uruará from the Pará state office of Brazil’s

834 Hum Ecol (2006) 34:829–849

Page 7: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

agricultural research agency, EMBRAPA/CPATU, as our sampling frame, to ensure thatsampling was not clustered spatially or selective of households by socioeconomic status.2

The sample includes 261 households, or 12% of all rural establishments in Uruará in1996 (IBGE, 1998a). The sample also includes 347 lots, as 25% of households held morethan one lot, and the same questions were asked about each lot. The sample consists ofhouseholds with one family (71%) and two or more families (29%), indicating somemultifamily households working the same land.3 In addition, 12% of households had one ormore elderly members, indicating multigenerational families.

Outcome Variables: Land Use Allocation

The outcome variables are measures of land allocation among primary forest, annual crops,perennial crops, cattle pasture, and secondary vegetation.4 Separation of annual andperennial crops is an advance beyond previous work and difficult to do. Annuals are ofteninterplanted, and data for perennials refers to trees (and vines) rather than area planted. Butbecause annuals differ from perennials in many respects, it is crucial to separate the two.We therefore used other data from the 1996 survey, information for Uruará from the 1995/1996 agricultural census (IBGE, 1998a) and our field notes (from yearly visits beginning in1996) to separately estimate land areas for annuals and perennials.5

Table I presents descriptive statistics and correlations for the land use variables. Thefarms in the survey had most of their land in primary forest (65 ha), with substantialpasture (22 ha), some annuals and perennials (3.9 ha each), and secondary growth(6.8 ha). Standard deviations indicate substantial variation in land allocation among thelots in the sample. Skew values for all five raw measures were large, so we transformedthem into natural logarithms (ln) with smaller skew values, indicating more normaldistributions that are more appropriate for regression analysis.

2 The 1996 Brazilian population count (IBGE, 1998b) and 1995/1996 Brazilian agricultural census (IBGE,1998a) allow for comparisons to assess sampling bias. The Uruará sample had a mean household size of 7.5,while the 1996 population count figure for the municipality of Uruará was only 5.6, but it is not clear fromcensus documentation whether families beyond the first were counted. If we exclude people outside the firstfamily, household size in the Uruará sample is also 5.6. The 1995/1996 agricultural census indicated thefollowing land use allocation in Uruará: 65% in primary forest, 5.6% under cropland, 23% under pasture, and5.9% under secondary growth. Table I indicates a very similar distribution. We conclude that sampling bias islimited.3 We recognize that different families in a given household may be at different life cycle locations. However,Chayanov left open the possibility of multifamily households. For purposes here, it is crucial to recognize thelabor contributions and dependency of families rather than exclude them from the analysis, for their presenceaffects land use. Nonetheless, we ran models keeping only the lots held by one family, and the results aresimilar to those presented.4 These measures refer to land use reported by households, which may or may not correspond to physicalland cover. Land use is still analytically important because use categories reflect distinctions and decisionsmade by households.5 We assumed that beans and corn are interplanted, so if both were planted, we divided their combined areain half, and added the result to other annual crops to estimate the total land area under annuals. Forperennials, we assumed that the tree crops (cocoa, coffee, oranges, cupuaçu, etc.) were planted with 3m by3m spaces, yielding 1,111 trees per hectare, while vine crops (i.e., black peppers) were planted with 2m by2m spaces, yielding 2,500 vines per hectare. This allowed conversion from plantings to areas, which we thensummed for all perennials reported. We validated the accuracy of our estimates by adding the areas underannuals and perennials to the reported areas under primary forest, cattle pasture and regrowth, and comparingthe sum to the reported total land area. The summed total was 101.1 ha, and the reported total was100.7 ha, a difference of 0.4 ha; the correlation between the two figures was very high (r > 0.99). Weconclude that the estimates are valid.

Hum Ecol (2006) 34:829–849 835

Page 8: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Table I indicates that the land use outcomes are interrelated. Primary forest exhibitsnegative correlations with the other land uses, which is expected given that forest is clearedfor agriculture. Primary forest area shows the strongest inverse relationship with pasture,which is not surprising given the large extent of pasture relative to crops and regrowth.Annuals, perennials and pasture are positively correlated, a reflection of their expansionduring the process of farm establishment. Secondary growth also shows positiveassociations with the agricultural land uses, a reflection of the need to fallow or abandonland. That said, the correlations are moderate, which may reflect the different points in thehousehold life cycle at which specific land uses become important.

Explanatory Variables: Household Life Cycle Location and Other Factors

Table II presents seven groups of explanatory variables: socioeconomic background, initialland cover, context of lot, institutional context, remittances and hired labor, landmanagement practices, and household life cycle location. We feature the role of householddemographic life cycle variables from Chayanovian theory, and include the other variablesas controls following household production theory and specificities of the Amazonfrontier.6

Table I Descriptive Statistics and Correlations for Land Use Outcomes, Farm Lots, Uruará, Pará, Brazil,1996 (n = 347)

Land use Mean Std.dev.

Skewness Correlation with:primary forest

Annualcrops

Perennialcrops

Pasture Secondarygrowth

Primary forestHectares (ha) 65.05 33.67 3.15 1.00Natural log (ln) ha 3.98 0.92 −4.79 1.00

Annual cropsHa 3.92 4.52 2.43 −0.08+ 1.00Ln ha 0.41 1.78 −0.62 −0.07 1.00

Perennial cropsHa 3.92 7.56 4.19 −0.03 0.23*** 1.00Ln ha 0.13 1.74 −0.07 −0.10+ 0.41*** 1.00

PastureHa 22.17 17.89 1.30 −0.25*** 0.04 −0.04 1.00Ln ha 2.32 1.91 −1.67 −0.20*** 0.27*** 0.18** 1.00

Secondary growthHa 6.79 8.69 2.28 −0.03 0.04 0.11* −0.02 1.00Ln ha 0.70 2.02 −0.53 −0.08+ 0.14** 0.14* −0.02 1.00

+p < 0.15, *p < 0.05, **p < 0.01, ***p < 0.001.

6 Table II indicates which variables are from the household questionnaire, and which come from the lotquestionnaire. The statistics in Table II are calculated for lots, including for the household variables, so thefigures are weighted toward households with more than one lot. However, the values do not change much ifcalculated for households, since 75% held one lot.

836 Hum Ecol (2006) 34:829–849

Page 9: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Table II Descriptive Statistics for Explanatory Variables and Correlations with Land Use Outcomes, FarmLots, Uruará, Pará, Brazil, 1996 (n = 347)

Explanatory variable Unit1 Mean Std.dev.

Correlation withprimary forest

Annualcrops

Perennialcrops

Pasture Secondarygrowth

Socioeconomic backgroundPrevious job of householdhead (0 = non-agricultural,1 = agriculture)

H 0.67 0.47 −0.04 0.03 0.21*** 0.07 0.03

Initial wealth(factor index)

H 0.00 1.28 0.11* −0.19*** −0.17** −0.11* −0.08

Initial agriculturalcapital (factor index)

H 0.00 2.47 −0.02 −0.11* −0.04 0.01 −0.11*

Initial land coverLn ha cleared uponacquisition

L 0.26 2.38 −0.07 −0.04 0.00 0.11* 0.08+

Context of lotOrdinal lot number(1 = 1st, 2 = 2nd...6th)

L 1.25 0.43 0.03 −0.41*** −0.33*** −0.34*** −0.23***

Kilometers to Uruará town L 31.16 15.49 0.27*** −0.18** −0.28*** −0.35*** −0.13*Neighborhood organization(0 = No, 1 = Yes)

L 0.34 0.47 0.05 0.01 −0.04 −0.01 −0.12*

Damage by fire setby neighbor (0 = No,1 = Yes)

L 0.21 0.41 −0.17** 0.15** 0.11* 0.22*** 0.19***

Institutional contextUse of credit (0 = No,1 = Yes)

L 0.46 0.50 −0.12* 0.25*** 0.30*** 0.42*** 0.08+

Extension agency assistance(0 = No, 1 = Yes)

L 0.16 0.37 −0.19** 0.08 0.20*** 0.15** 0.07

Commercial business(0 = No, 1 = Yes)

H 0.09 0.29 −0.03 −0.14** −0.13* −0.13* −0.12*

Remittances and hired laborRemittance income(0 = No, 1 = Yes)

H 0.11 0.31 0.08+ −0.06 −0.10+ −0.13* 0.07

Ln days of labor hired H 2.25 2.24 0.03 −0.11* 0.11* −0.03 −0.04Land management practicesAgricultural inputs(factor index)

L 0.00 2.12 −0.14* 0.11* 0.24*** 0.22*** 0.06

Pasture rotation(0 = No, 1 = Yes)

L 0.69 0.46 −0.13* 0.28*** 0.27*** 0.61*** −0.04

Life cycle locationYears on lot L 10.12 6.70 −0.13* 0.17** 0.30*** 0.23*** 0.27***Number of adults(ages 15–65)

H 4.33 2.65 −0.07 0.02 0.22*** 0.22*** −0.11

Number of adults squared HNumber of children(under age 15)

H 2.93 2.83 −0.02 0.06 0.11* −0.01 0.04

Number of elderly(ages 66+)

H 0.15 0.46 0.02 0.10+ 0.07 0.06 0.05

Generational transition(elderly × children)

H

+p < 0.15, *p < 0.05, **p < 0.01, ***p < 0.001.1H Household-level characteristic, L lot-level characteristic. All statistics are calculated for lots.

Hum Ecol (2006) 34:829–849 837

Page 10: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Control Variables

“Socioeconomic background” refers to three variables that characterize assets held byhouseholds upon their arrival in Uruará. Previous job indicates whether the household headhad worked in agriculture.7 Previous agricultural experience should be particularlyimportant for cultivating crops, given their labor intensity. Most household heads did haveprevious agricultural experience, which shows a positive correlation with perennials. Initialwealth refers to financial assets. Households who brought wealth were better able toliquidate assets, facilitating farm implementation and altering land allocation. We measureinitial wealth using indicators of durable goods possession and housing quality. These wereconverted to z-scores, weighted by factor loadings from principle components analysis, andsummed to form an index with a mean of zero.8 Initial agricultural capital is also a factor-weighted index, constructed using three measures of whether a household owned specificagricultural implements at the time of their arrival.9 Agricultural technologies such aschainsaws may afford more rapid implementation of farming systems. The standarddeviations for both wealth indexes indicate asset inequality in the sample, and both exhibitsignificant correlations with land use.

“Initial land cover” is operationalized in terms of the ln ha deforested when the householdacquired a lot. The antilog of the ln mean was only 1.3 ha, though the standard deviationindicates considerable variation. More initial deforestation facilitates farm implementation.This reduces the labor inputs necessary for agricultural land use, but also makes moreextensive regrowth possible. The correlations, though weak, confirm these expectations.

“Context of lot” comprises four indicators that situate a lot in a household’s farmingsystem and among neighboring lots. First, we consider the order in which a householdacquired a lot. The first lot acquired is generally the most heavily used, so second and laterlots (25% of all lots in the survey) should have more forest and less cropland, pasture andregrowth, expectations confirmed by the correlations. Second, we account for distance tomarket, especially important for commercial land use decisions because transport costs arehigh on unpaved roads in the study area, reducing the profitability of more distant lots. Lotsaveraged about 30 km from Uruará town, though this varied substantially. Larger distancesshould correspond to more primary forest and less land under the other use types,expectations confirmed by the correlations. Third, the presence of neighborhoodorganizations indicates whether neighboring households were mobilized for cooperativelabor arrangements, against land invasions, and/or to secure agricultural credit, all of whichshould allow for greater agricultural land use. About 34% of lots were in organizedneighborhoods, but weak correlations suggest ambiguous effects on land allocation. Andfourth, we consider damage to vegetation from fires set by neighbors. Fire damage mayreduce primary forest, facilitating the expansion of agricultural land, but the damage mayexceed a household’s ability to use the burned land productively, leading to substantial

7 We also considered the household head’s region of birth and years of schooling. However, neither of thesevariables exhibited significant effects.8 Variables and factor weights from principle components analysis for the initial wealth index are: house intown 0.80, brick walls 0.50, electricity, 0.64, generator 0.57, gas stove 0.67, sewing machine 0.54,refrigerator 0.79, radio 0.53, television 0.81, satellite dish 0.70, bicycle 0.66, and car 0.50. The eigenvaluefor this factor was 5.08, and the common variance was 42.4%.9 Variables and factor weights from principle components analysis for the initial agricultural capital indexare: chainsaw 0.81, cocoa dryer 0.63, and tractor 0.48. The eigenvalue for this factor was 1.28, and thecommon variance was 42.8%.

838 Hum Ecol (2006) 34:829–849

Page 11: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

secondary growth. About 20% of lots had incurred fire damage, and it shows significantland use correlations in the expected directions.

“Institutional context” comprises three variables that tie a lot to public and private agenciesand the urban economy. First, the use of credit indicates the importance of lendinginstitutions. Because credit can offset capital scarcity, it facilitates commercialization.Consequently, use of credit should lead households to allocate less land to forest and more toperennials and cattle.10 Nearly half of the lots surveyed were owned by households withcredit, and credit exhibits the anticipated associations with land use. Second, extensionassistance indicates whether government agricultural agents had ever visited a given lot.Extension agents in Uruará focus on commercial activities, so assistance should correspondto less forest and more of the other uses, especially perennials and cattle. Only 16% of lotshad been visited by extension agents, but it shows the expected correlations. And third,some farm households ran local businesses in Uruará town. Investment in commercialenterprises initially diverts resources from agriculture, though earnings may help financeexpanded land use. About 9% of lots were owned by households with businesses, and thecorrelations suggest that those lots had less agricultural land, perhaps due to divertedinvestment.

“Remittances and hired labor” are included to assess the effects of labor markets. Theremittances variable refers to whether a household had absent family members sendingmoney, and this occurred among households who owned 11% of the lots surveyed. Likecredit, remittances can offset capital scarcity and facilitate greater land use. However, thecorrelations are weak and run in the other direction, which implies that remittances are putto uses other than agriculture. Hired labor, measured as the ln days of labor paid by ahousehold in the previous year, can offset family labor scarcity and encourage forestclearing, especially for commercial agriculture. On average, households paid for 9.5 daysof hired labor. The positive association with perennials is consistent with the use of hiredlabor to expand cash crops.

“Land management practices” refers to two strategies households may employ to sustainproduction on their lots, namely the use of agricultural inputs and pasture rotation. Theagricultural inputs measure is a factor-weighted index calculated using indicators of use ofpesticides and fertilizers to sustain crop productivity.11 While some households may employinputs to reduce the land area in use, others may do so to sustain production in larger areas.The correlations suggest that the latter interpretation is correct, via the negative associationwith forest and positive associations with crops and pasture. Pasture rotation requires moregrazing land for a given number of cattle. Rotation, used on 69% of the lots surveyed,shows a positive correlation with pasture and a negative association with forest as expected,but also positive associations with crops.

Demographic Life Cycle Variables

Demographic variables that define a household’s life cycle location should also influenceland allocation. Table II measures life cycle location using six variables: time on lot,

11 Variables and factor weights from principle components analysis for the agricultural inputs index are:insecticides 0.74, fungicides 0.54, herbicides 0.53, chemical fertilizers 0.81, and organic fertilizers 0.58. Theeigenvalue for this factor was 2.12, and the common variance was 42.3%.

10 We considered using measures of tenure status, but land titles are usually necessary to obtain credit, andtitles have a high correlation with credit (r > 0.60). Because credit is more proximate to land use, and becausecredit exerted stronger effects, we exclude tenure status.

Hum Ecol (2006) 34:829–849 839

Page 12: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

number of adults, adults squared, number of children, number of elderly, and a child–elderly interaction term.

Time on lot captures a household’s duration of residence with reference to their land.This indicates a household’s experience with a property via exploration of its resources andexperimentation with agricultural techniques. Long-term ownership should yield less landallocated to forest and more to the other uses. The survey data indicate a mean duration of10 years with substantial variation, and significant correlations with the land use outcomesin the expected directions.

The next four variables assess age structure effects on land use allocation. Theoretically,these four variables change in tandem with time on lot. But as shown in Table III, they areindependent to the extent that children are born over time and households acquire lots atdifferent moments in their life cycles. The number of working-age adults (persons age 15–65)measures household labor availability.12 More adults should lead to larger productionsystems with less primary forest and more agriculture. Because crops require heavy laborinputs, the effect of adults should be especially important for annuals and perennials. Wealso consider the square of adults because households with especially large labor pools mayincreasingly allocate labor to off-farm activities such as wage work in town. The effect ofthe adults squared term should be the opposite of the adults effect. Hence, the overallimpact of adults should be non-linear, with declining marginal effects, such that forestdecline and agricultural expansion attenuate in especially large households due toincreasing off-farm labor allocation. Table II shows an average above four adults for thesample, with substantial variation, and correlations with land use largely as expected.

The number of children (persons under age 15) measures the impact of young householdmembers on land use.13 Children constitute pressure to plant annual crops to meetsubsistence demand, but older children expand the household labor pool, allowing forlarger areas of commercial crops. Table II shows a mean of nearly three children for thesample, with a large standard deviation. Correlations with land use outcomes are somewhatweak, which may reflect the countervailing effects of children, though there is a significantpositive effect on perennials, consistent with an interpretation emphasizing child laborcontributions.

The number of elderly (persons age 66+) measures the extent of aging among colonisthouseholds. Elderly household members imply that children are grown and some have leftto start their own farms or other enterprises. This suggests a decline in household size,reducing agricultural land areas and increasing secondary growth. Table II indicates fewelderly on average but substantial variation, and weak correlations with land use.

We operationalize “generational transition” using a child–elderly interaction term.Conceptually, this term defines multigenerational households as those where the farm isbeing handed from one generation to the next, which often happens when the grandchildrenarrive. The interaction term allows for evaluation of the generational transition effect onland use net of the distinct influences of elderly members and children by themselves.Households with elderly members as well as children are taken to exhibit transitions fromone generation’s life cycle to another, implying a rise in subsistence demand, which shouldprompt greater agricultural land use and a decline in secondary growth.

12 One might object that men and women should have separate variables to assess their distinct effects onland use. However, correlation analysis indicated a strong association between the number of men andwomen (r > 0.60), and models with a single variable for adults were stronger.13 One might object that aggregating children ages 0–15 mixes true dependents and those contributing labor.We recognize other possible age cutoffs but use the 0–15 due to limitations in the survey data. This stillprovides an indication of the net effect of young household members on land use.

840 Hum Ecol (2006) 34:829–849

Page 13: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Modeling Land Use Allocation

Land allocation must be viewed in terms of joint decisions among competing land uses.This makes land allocation decisions mutually endogenous (i.e., simultaneous), for thedecision to allocate more land to one use on a lot of a given finite size constitutes anopportunity cost and limitation on the quantity of land left to allocate to other uses. Plantingof annuals comes at the expense of forest; later, perennials and pasture replace forest andannuals; and eventually, secondary growth replaces cropland and pasture.

But household models of land use rarely account for endogeneity in land allocationdecisions (e.g., Jones et al., 1995). Most common are models that assume independenceamong the various outcomes (e.g., Perz, 2001; Pichón, 1997). Such efforts overlookendogeneity and the consequent problems of estimation bias and inconsistency, with theresult that conclusions about factors affecting land use may be incorrect. As a result,analysts have used other approaches, such as seemingly unrelated regression (SURE),which accounts for correlated error terms (Pan et al., 2001; Perz, 2002). The limitation ofSURE is that it only indirectly accounts for the effect of one outcome variable on another,and does not allow direct observation of whether, for example, more pasture or somethingelse is planted at the expense of perennials.

We therefore employ three-stage least squares (3SLS) estimation. This involves creationof a system of structural equations where the error terms are correlated and one or moredependent variables are endogenous explanatory variables in other equations. Like 2SLS,3SLS uses instrumental variables to produce consistent estimates of the endogenousvariables. And like SURE, 3SLS uses 2SLS estimation for each equation to adjust forcorrelated errors and obtain a consistent error covariance matrix. But 3SLS then uses GLSestimation, which adjusts for correlated errors and incorporates the instrumented variablesto simultaneously estimate the entire system of equations. 3SLS thus goes beyond SURE bycreating instrumented variables; 3SLS also goes beyond 2SLS by generating results for theentire system. Both advances are necessary to adequately account for the endogeneity ofland allocation decisions and simultaneously evaluate the effects of household demographicvariables for multiple land uses.

Table III Correlations Among Life Cycle Demography Variables, Farm Households and Lots, Uruará,Pará, Brazil, 1996

Life cycle demography variable Time in Uruará/on lot Numberof adults

Numberof children

Numberof elderly

Households (n = 261)Duration of residence in Uruará 1.00Number of adults (ages 15–65) 0.22** 1.00Number of children (under age 15) 0.02 0.46*** 1.00Number of elderly (ages 66+) 0.19** 0.16** 0.18** 1.00

Lots (n = 347)Time on lot 1.00Number of adults (ages 15–65) 0.12* 1.00Number of children (under age 15) −0.02 0.45*** 1.00Number of elderly (ages 66+) 0.12* 0.15** 0.15** 1.00

+p < 0.15, *p < 0.05, **p < 0.01, ***p < 0.001.

Hum Ecol (2006) 34:829–849 841

Page 14: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Model specification worked from results from SURE models (Perz, 2002), rerun for allfive outcomes, in order to identify instrumental variables. We then evaluated the per-formance of the SURE models, and constructed a 3SLS system by using significantvariables from the SURE equation for a given land use outcome in that outcome’s equationin the 3SLS model. We then iteratively tested the 3SLS system, dropping variables thatwere insignificant in a given equation if used in more than one, and excluded variables thatnever reached significance and whose removal did not significantly change or weaken thesystem. We constrained the process of specification by including all of the householddemographic life cycle variables in each equation, and using the control variables toidentify the system. This reflects our focus on life cycles and facilitates comparisons of theireffects among the land use outcomes. It also reflects our expectation that the life cyclevariables do not have the same effects on each land use outcome.14

Findings

Table IV presents results from our modeling effort: a system of five equations, each withcoefficients for instrumented land use variables, selected control variables, and thehousehold demographic life cycle variables. All five equations have significant chi-squarevalues.

Primary Forest

The weakest equation is the first, for primary forest. None of instrumented land usevariables exerted independent effects on forest area. However, lots had more primary forestif they 1) were farther from Uruará town, 2) they had not been damaged by fire, and 3) theyhad not been visited by extension agents. Household demographic variables exhibit limitedimpacts on primary forest. The number of children reduces forest area, likely a reflection ofsubsistence demand early in the household life cycle.

Annual Crops

The story for annual crops is considerably different, in large part due to significant effectsof household demographic variables. Among the instrumented land use variables, pasturearea has a positive effect on annuals. This suggests that households manage risk not only byplanting annuals but also by running cattle as a form of rural insurance. It is likely also aperiod effect, for in the years just before the 1996 survey, problems with perennials ledmany households to focus on annuals for food security and cattle for their marketability.

With respect to the control variables, lots had larger areas planted under annuals if they1) belonged to households who arrived with less initial wealth, and 2) were the first lotacquired. While the initial wealth effect is weak, both of these findings are consistent withthe interpretation that annuals provide a subsistence. Households generally live on the first

14 One potential problem with 3SLS is that misspecification of one equation yields inconsistent and biasedestimates of coefficients in the other equations. We worked from a SURE system with equations with r2

values ranging from about 0.20 to 0.50 and significant F-ratios (p < 0.001). This suggests that there wereeffective instruments for the land use outcomes. By systematically changing model specification andevaluating the results, we were able to evaluate specifications by iterating toward equations such that furtheralterations produced similar but weaker models. Through this process, we distinguished the most effectiveinstruments, which allowed us to identify the system and satisfy the order condition.

842 Hum Ecol (2006) 34:829–849

Page 15: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

lot acquired, which is also the lot where food is grown, and poor households are especiallyconcerned to minimize risks by planting annuals.

That said, the most important explanation for land allocation to annual crops involvesthe demographic life cycle variables, especially age structure. The number of adults has astrong and positive but non-linear effect on annuals, a reflection of the importance of

Table IV Three-stage Least Squares Model of Land Use Allocation with Life Cycle Location and OtherVariables, Farm Lots, Uruará, Pará, Brazil, 1996.

Explanatory variable Primaryforest

Annualcrops

Perennialcrops

Pasture Secondarygrowth

Equation parameters 14 1 12 15 16 16Model chi-square 60.62*** 106.24*** 103.92*** 186.68*** 78.18***Intercept 3.81*** −0.78 −1.11 4.77* 0.01Endogenous land use variables

Ln ha under primary forest −0.01 −0.37 −0.92+ 0.2Ln ha under annual crops 0.15 −0.94* 1.01** 0.41Ln ha under perennial crops −0.1 −0.06 −0.14 0.35Ln ha under pasture −0.09 0.28* 0.38* −0.66***Ln ha under secondary growth −0.04 −0.02 0.02 0.13

Socioeconomic backgroundPrevious job (0 = non-agricultural,1 = agricultural)

0.39*

Initial wealth (factor index) −0.03+ −0.08**Initial agricultural capital (factor index) −0.16* −0.24*

Initial land coverLn ha cleared upon acquisition −0.01 0.31***

Context of lotOrdinal lot number (1 = 1st, 2 = 2nd...6th) −1.43*** −2.27** 0.94+Kilometers to Uruará town 0.01*** 0.0002Neighborhood organization (0 = No, 1 = Yes) 0.33+Damage by fire set by neighbor (0 = No,1 = Yes)

−0.18+ 0.76*

Institutional contextUse of credit (0 = No, 1 = Yes) 0.41*Extension agency assistance (0 = No, 1 = Yes) −0.2+ −0.62+Commercial business (0 = No, 1 = Yes) −1.01*

Remittances and hired laborRemittance income (0 = No, 1 = Yes) 0.56+Ln days of labor hired 0.13**

Land management practicesAgricultural inputs (factor index) 0.11*Pasture rotation (0 = No, 1 = Yes) 1.45***

Life cycle locationYears on lot 0.003 −0.01 −0.002 −0.01 0.14***Number of adults (ages 15–65) 0.06 0.41** 0.84*** −0.21 −0.27Number of adults squared −0.001 −0.04*** −0.07** 0.03 0.02Number of children (under age 15) −0.05** 0.09* 0.13* −0.18** 0.09Number of elderly (ages 66+) 0.03 0.56* 0.25 −0.79* 0.42Generational transition (elderly × children) −0.0004 −0.06+ −0.03 0.1* −0.09+

+p < 0.15, *p < 0.05, **p < 0.01, ***p < 0.001.1 Valid cases after listwise deletion of cases with missing values: n = 310.

Hum Ecol (2006) 34:829–849 843

Page 16: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

household labor as well as livelihood diversification into other activities if the labor pool isespecially large. Children have a positive effect on annuals, which we interpret as primarilyreflecting subsistence demand for food. Contrary to expectations, elderly members alsohave a positive effect on annuals, which suggests that old age dependency generatesmeasurable subsistence demands. Also unexpected is a negative effect of generationaltransitions on annuals. However, this effect is weak and does little to offset the largerindependent effects of children and elderly members.

Perennial Crops

The third column presents the equation for perennial crops. Less land under annuals leadsto more under perennials, while more pasture expands perennials. These findings suggestthat perennials tend to replace annuals, and that perennials and pasture tend to expandsimultaneously. Both results are consistent with the life cycle framework outlined earlier.

Focusing on the control variables, a lot had more land under perennials if 1) thehousehold head had prior agricultural experience, 2) the household arrived with less initialwealth, 3) it was the first lot acquired by a household, 4) the household hired more labor,and 5) it received more agricultural inputs. The finding for agricultural experience isconsistent with arguments that knowledge about agriculture facilitates commercialization ofproduction. The negative coefficient for initial wealth combines aspects of settlement cohortand period effects. Earlier settlement cohorts, who in general brought less wealth with them,also had more time to cultivate perennials, putting them in a better position to manage pestoutbreaks in the early 1990s. A similar interpretation applies to the concentration ofperennials on the first lot acquired. Households initially expand their farming activities ontheir first lot, and it takes time to plant and maintain perennials before they becomeproductive, but the price declines and pest attacks in the years prior to the 1996 surveylikely slowed new perennial plantations, especially on newly acquired lots. The positiveeffects of hired labor and agricultural inputs on perennials follow expectations, and reflectthe importance of emerging labor, input, and product markets in the study site, as perhousehold production theory.

Net of these factors, household demographic variables are very important for landallocation to perennials, particularly age structure. As with annuals, the number of adultsexerts a large, positive, and non-linear effect on land planted under perennials. The effect ofchildren on perennials is also positive and significant. This likely reflects the laborcontributions of older children, which follows our expectations that households with olderchildren expand plantings of perennial crops.

Pasture

The strongest equation is that for pasture. Among the instrumented endogenous variables,primary forest shows a weak negative effect, suggesting that pasture expands to someextent at the expense of forest. Annual crops show a large positive impact on pasture,indicating a period effect resulting from problems with perennials that led many householdsto shift their farming systems toward annual crops and cattle.

Larger pasture areas occur on lots if they 1) were held by a household that arrived withless initial agricultural capital, 2) were the second or later lot acquired, 3) were located in anorganized neighborhood, 4) received investments made with bank loans or other credit, and5) practiced pasture rotation. The negative coefficient for initial agricultural capital mayreflect settlement cohort and period effects, wherein households who arrived earlier (and

844 Hum Ecol (2006) 34:829–849

Page 17: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

tended to be poorer) had more time to plant more pasture, especially in the yearsimmediately prior to the 1996 survey during problems with perennials. The order ofacquisition effect, while weak, is the opposite of that seen for crops, and suggests thathouseholds with more than one lot tended to allocate their pasture to newer lots. This islikely a period effect, reflecting the expansion of pasture when more recent lots wereacquired. Further, there is a weak positive effect of neighborhood organizations on pasture,and a stronger positive effect of credit. The influences of both of these variables also reflectperiod effects. In the early 1990s, new credit lines appeared, notably the “FNO-e” programaimed at smallholders (Toni, 2003). Loans were made to households via localorganizations, who used credit to expand pastures and cattle herds. However, the singlemost important variable is pasture rotation. By dividing pastures into two or four fields withfences to rotate cattle and lengthen the productive life of grazing land, households aremaking a serious commitment to ranching, which multiplies the grazing area required.

The demographic variables show weaker and different effects on pasture than perennials.Adult labor is not important for pasture, a reflection of the low labor requirements forranching. The negative child effect suggests that it is older households that expand theirpasture area, but the negative effect of elderly members indicates a decline in pasture areaamong the oldest households. Both findings are consistent with the theoretical framework.The positive effect of the instrumented perennials variable suggests that older householdsshift their labor toward perennials even as they also expand pasture for cattle. This is afinding overlooked in the literature on pasture expansion, and not easily discerned without asystem of equations for multiple land uses. The negative elderly effect may reflect the factthat 43% of the households in the sample with elderly members received state retirementincome, which would constitute an alternative form of insurance to cattle. However, there isa positive generational transition effect, which indicates that there is more pasture on lotsheld by households with both elderly and young members. Thus, the decline in pasture latein the household demographic life cycle is dampened by a generational transition, which isconsistent with the theoretical framework.

Secondary Growth

In the last equation, among the instrumented endogenous variables, only pasture is sig-nificant, but it exerts a very strong negative impact, suggesting a tradeoff in land allocationwith secondary growth. This follows expectations, in that older pasture is often taken out ofuse and becomes secondary growth.

Several control variables exhibit significant effects. Regrowth was more extensive onlots that 1) were held by households with less initial agricultural capital, 2) had more forestcleared upon acquisition, 3) had fire damage to vegetation, 4) had not been visited byextension agents, 5) were held by households without a commercial business in town, and6) were held by households receiving remittance income. The effects of initial capital, forestcleared upon arrival, fire damage, and lack of extension visits all reflect in various ways thelimited ability of colonist households to keep land in production. Capital scarcity, largecleared areas upon acquisition, widespread fire damage, and lack of extension assistancesooner or later necessitate fallowing or risk of land degradation, both of which result inexpanding secondary growth. The commercial business effect suggests that householdswith enterprises in Uruará town also had capital to invest in larger farming systems,reducing the extent of regrowth on their lots. The remittances effect, while weak, suggests alivelihood diversification strategy wherein some households opt to focus more on wagelabor and less on farming, expanding regrowth.

Hum Ecol (2006) 34:829–849 845

Page 18: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Regarding the demographic life cycle variables, time on lot has a strong positive impact onthe extent of secondary growth, while age structure is of no consequence, and a weakgenerational transition effect appears. The finding for time on lot confirms the importance ofresidence duration for regrowth, consistent with the theoretical framework, which suggestedmore secondary growth among long-established households. The weak generationaltransition effect follows expectations that households with elderly and young members havesomewhat less secondary vegetation, which complements findings of larger productionsystems in multigenerational households.

Discussion

Household Demographic Processes and Land Use

Household demographic life cycle variables have diverse effects on different land uses, andthese impacts are consonant with expectations based on Chayanovian and householdproduction theories. Duration of residence is vital for understanding the extent of secondarygrowth; age structure is crucial for labor-intensive land uses; and generational transitionsaffected several aspects of land allocation. These findings indicate that householddemographic processes have environmental consequences even in the presence of markets.

Estimation of a system of equations offers insights about land allocation that go beyondprevious modeling efforts and reveal some contrasts with prior work. First, the significance ofendogenous land use variables shows which land uses constitute opportunity costs for otherland uses. The significant effects occurred among the agricultural land uses and secondarygrowth. Although agriculture replaces primary forest, the extent of forest did not vary muchamong lots in the sample, whereas the other land uses varied substantially. Second, the use ofa system of equations indicates indirect effects of many explanatory variables on the land useoutcomes. Variables important for annuals, such as adult labor and child dependency, areindirectly important for cattle pasture because the instrumented annuals variable is significantin the pasture equation. Similarly, credit and pasture rotation are important for understandingthe extent of secondary growth via their effects on pasture, which exhibits a significant effecton regrowth. Thus, complex causal pathways become evident when we model multiple landuses simultaneously. And third, household demographic life cycle variables often exhibitedstronger effects in the foregoing analysis than our previous modeling approaches for the studycase. Perz (2001) used OLS estimation and found no significant effects for householddemographic variables on annual crops, and Perz (2002) employed SURE and foundsignificant life cycle impacts on annuals but not on pasture.

Statics and Dynamics in Household Life Cycle Demography and Land Allocation

A caveat to the findings is that they come from cross-sectional data, whereas the theoreticalmodel we evaluated was framed in dynamic terms. However, other evidence suggests thattemporal interpretations of differences in land allocation in the cross-sectional data herereflect dynamic processes. Populations are aging in frontier areas of the Amazon. Braziliancensus data indicate that Uruará’s under-15 population declined from 44.7% in 1991 to37.0% in 2000 (IBGE, 1996, 2000). The 1996 Uruará survey showed that 40% of thepeople in the households sampled were under age 15, consistent with the census data. Landuse change in the Amazon follows our theoretical framework. Walker and Homma (1996)compare land use profiles of households along the Transamazon highway for the 1970s and

846 Hum Ecol (2006) 34:829–849

Page 19: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

early 1990s. In 1975, households had 2.5 ha under rice and 6.4 ha under pasture, and by1993, households had 4.1 ha under rice and 37 ha of pasture, confirming pastureexpansion. Other authors, using retrospective questions for land use (e.g., Coomes et al.,2000) and multitemporal analysis of satellite images (e.g., Brondizio et al., 2002) havefound similar changes. We conducted another survey on land use in Uruará in 2002, andpreliminary analysis of those data indicates dynamics consistent with our models. House-holds aged because the number of children per household declined, while the number ofelderly members rose. In parallel, land allocation changed such that the area under annualsdeclined, while perennials and pasture both expanded. We plan to pursue dynamic modelingto confirm whether these demographic and land use changes correspond at the householdlevel.

Implications for Demographic Environmental Research

Demographic environmental research has largely advanced via methodological innovations(e.g., Pebley, 1998; Lutz et al., 2002). Demographers and other social scientists havedeveloped creative means of integrating data for land cover research using GIS (e.g., Walshand Crews-Meyer, 2002), which has fostered “agent-based” dynamic simulation modeling(e.g., Parker et al., 2003). However, such efforts have rarely incorporated demographicprocesses (cf. An et al., 2005).

That said, theoretical innovations will also be necessary to advance demographiccontributions to environmental science. By moving across scales in the social sciences, fromfamilies to the global system, one encounters different theoretical traditions that help accountfor environmental outcomes (Wood, 2002). Demographic environmental studies wouldbenefit from following this lead by diversifying beyond the “population and environment”discourse. Such movement is beginning, such as via efforts that feature environmentalconsequences of demographic processes in social networks (e.g., Curran, 2002), but muchmore stands to be done.

Acknowledgments This research was supported by a grant from the US National Science Foundation(SBR-9511965). We thank Charles Wood for support in the US, Adilson Serrão and Alfredo Homma forsupport in Brazil, and research team members André Caetano, Roberto Porro, Fabiano Toni, Célio Palheta,Rui Carvalho, and Luiz Guilherme Teixeira, as well as the people of Uruará, for insights about the study site.Errors are the responsibility of the authors.

References

An, L., Linderman, M., Qi, J., Shortridge, A., and Liu, J. (2005). Exploring Complexity in a Human–Environment System: An Agent-Based Spatial Model for Multidisciplinary and Multiscale Integration.Annals of the American Association of Geographers 95(1): 54–79.

Arizpe, L., Stone, M.P., and Major, D.C. (eds.) (1994). Population and Environment: Rethinking the Debate,Westview, Boulder.

Bongaarts, J. (1992). Population Growth and Global Warming. Population and Development Review 18:299–319.

Brondizio, E.S., McCracken, S.D., Moran, E.F., Siqueira, A.D., Nelson, D.R., and Rodriguez-Pedraza, C.(2002). The colonist footprint: toward a conceptual framework of land use and deforestation trajectoriesamong small farmers in the Amazonian frontier. In Wood, CH, and Porro, R (eds.), Deforestation andLand Use in the Amazon, University of Florida, Gaineville, pp 133–161.

Chayanov, A.V. (1986[1966]). The Theory of Peasant Economy, University of Wisconsin, Madison.Chibnik, M. (1984). A Cross-National Examination of Chayanov’s Theory. Current Anthropology 25(3):

335–340.

Hum Ecol (2006) 34:829–849 847

Page 20: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Coomes, O.T., Grimard, F., Burt, G.J. (2000). Tropical Forests and Shifting Cultivation: Secondary Forest FallowDynamics among Traditional Farmers of the Peruvian Amazon. Ecological Economics 32: 109–124.

Curran, S. (2002). Migration, social capital and the environment: considering migrant selectivity and networksin relation to coastal ecosystems. In Lutz, W, Prskawetz, A, and Sanderson, WC (eds.), Population andEnvironment: Methods of Analysis, Population Council, New York, pp 89–119.

Ellis, F. (1993). Peasant Economics: Farm Households and Agrarian Development, 2nd edn, CambridgeUniversity Press, Cambridge.

Faminow, M.D. (1998). Cattle, Deforestation and Development in the Amazon: An Economic, Agronomicand Environmental Perspective, CAB International, Oxford.

Gibson, C., Ostrom, E., and Toh-Kyeong, A. (2000). The Concept of Scale and the Human Dimensions ofGlobal Change: A Survey. Ecological Economics 32: 217–239.

Harrison, M. (1975). Chayanov and the Economics of the Russian Peasantry. Journal of Peasant Studies 2(2): 379–417.

Hunt, D. (1979). Chayanov’s Model of Peasant Household Resource Allocation. Journal of Peasant Studies6(3): 247–285.

IBGE—Instituto Nacional de Geografia e Estatística (1962). Censo Demográfico de 1960, IBGE, Rio deJaneiro.

IBGE (1996). Censo Demográfico de 1991, IBGE, Rio de Janeiro.IBGE (1998a). Censo Agropecuário de 1995/1996, IBGE, Rio de Janeiro.IBGE (1998b). Contagem Populacional de 1996, IBGE, Rio de Janeiro.IBGE (2000). Censo Demográfico de 2000, IBGE, Rio de Janeiro.IDESP—Instituto de Desenvolvimento do Estado do Pará (1990). Uruará, IDESP, Belém.INPE—Instituto Nacional de Estudos Espaciais (2002). Monitoramento da Floresta Amazônica Brasileira por

Satélite: Projeto PRODES Website available at www.obt.inpe.br/prodes/index.htmlJones, D.W., Dale, V.W., Beauchamp, J.J., Pedlowski, M.A., O_Neill, R.V. (1995). Farming in Rondonia.

Resource and Energy Economics 17: 155–188.Lutz, W., Prskawtez, A., and Sanderson, W.C. (eds.) (2002). Population and Environment: Methods of

Analysis Supplement to Population and Development Review vol 28.MacKellar, F.L., Lutz, W., McMichael, A.J., Suhrke, A., Mishra, V., O_Neill, B., Prakeesh, S., and Wexler, L.

(1998). Population and climate change. In Rayner, S., and Malone, E.L. (eds.), Human Choice andClimate Change, vol 1: The Societal Framework, Battelle, Columbus, pp 89–193.

Marquette, C.M. (1998). Land Use Patterns among Small Farmer Settlers in the Northeastern EcuadorianAmazon. Human Ecology 26(4): 573–598.

Mazur, L.A. (ed.) (1994). Beyond the Numbers: A Reader on Population, Consumption, and the EnvironmentWashington, District of Columbia: Island.

McCracken, S.D., Siqueira, A.D., Moran, E.F., and Brondizio, E.S. (2002). Land use patterns on an agriculturalfrontier in brazil: insights and examples from a demographic perspective. In Wood, C.H., and Porro, R.(eds.), Deforestation and Land Use in the Amazon, University of Florida, Gainesville, pp 162–192.

Moran, E.F. (1989). Adaptation and maladaptation in newly settled areas. In Schumann, D., and Partridge, W.(eds.), The Human Ecology of Tropical Land Settlement in Latin America, Westview, Boulder, pp 20–41.

Nascimento, E.P., and Drummond, J.A. (eds.) (2003). Amazônia: Dinamismo Econômico e ConservaçãoAmbiental, Garamond, Rio de Janeiro.

Ness, G.D., Drake, W.D., and Brechin, S.R. (eds.) (1993). Population–Environment Dynamics: Ideas andObservations, Ann Arbor, University of Michigan.

O’Neill, R.V., DeAngelis, D.L., Waide, J.B., and Allen, T.F.H. (1986). A Hierarchical Concept of Ecosystems,Princeton University Press, Princeton.

Pan, W., Murphy, L., Sullivan, B., and Bilsborrow, R. (2001). Population and Land Use in Ecuador’sNorthern Amazon in 1999: Intensification and Growth on the Frontier Paper presented at the PopulationAssociation of America meetings, Washington, District of Columbia.

Parker, D.C., Manson, S.M., Janssen, M.A., and Hoffmann, M.J., and Deadman, P. (2003). Multi-AgentSystems for the Simulation of Land-Use and Land-Cover Change: A Review. Annals of the Associationof American Geographers 93(2): 314–337.

Pebley, A.R. (1998). Demography and the Environment. Demography 35(4): 377–389.Perz, S.G. (2001). Household Demographic Factors as Life Cycle Determinants of Land Use in the Amazon.

Population Research and Policy Review 20(3): 159–186.Perz, S.G. (2002). Household Demography and Land Use Allocation among Small Farms in the Brazilian

Amazon. Human Ecology Review 9(2): 1–16.Perz, S.G., Walker, R.T. (2002). Household Life Cycles and Secondary Forest Cover among Small Farm

Colonists in the Amazon. World Development 30(6): 1009–1027.

848 Hum Ecol (2006) 34:829–849

Page 21: Beyond Population and Environment: Household Demographic ...rwalker/pubs/perz et al. human ecology.pdf · Beyond Population and Environment: Household Demographic Life Cycles and

Pichón, F.J. (1997). Colonist Land-Allocation Decisions, Land Use, and Deforestation in the EcuadorianAmazon Frontier. Economic Development and Cultural Change 45(4): 707–744.

Serrão, E.A.S., and Homma, A.K.O. (1993). Brazil. In Sustainable Agriculture and the Environment in theHumid Tropics, National Academy, Washington, pp 265–351.

Serrão, E.A.S., and Toledo, J.M. (1990). The search for sustainability in Amazonian pastures. In Anderson,A.B. (ed.), Alternatives to Deforestation: Steps Toward Sustainable Use of the Amazon Rain ForestColumbia University Press, New York, pp 195–213.

Singh, I., Squire, L., and Strauss, J. (eds.) (1986). Agricultural Household Models: Extensions, Applicationsand Policy, Johns Hopkins University Press, Baltimore.

Toni, F. (2003). Uruará: Pecuarização na Fronteira Agrícola. In Toni, F., and Kaimowitz, D. (eds.),Municípiose Gestão Florestal na Amazônia, AS Editores, Natal, pp 175–218.

Tourrand, J.-F., Veiga, J.B. (eds.) (2003). Viabilidade de Sistemas Agropecuários na Agricultura Familiar naAmazônia Belém: Embrapa Amazônia Oriental.

Turner, B.L. II, Hyden, G., and Kates, R.W. (eds.) (1993). Population Growth and Agricultural Change inAfrica, University of Florida, Gainesville.

Walker, R.T. (2003). Mapping Process to Pattern in the Landscape Change of the Amazonian Frontier.Annals of the Association of American Geographers 93(2): 376–398.

Walker, R.T., and Homma, A.K.O. (1996). Land Use and Land Cover Dynamics in the Brazilian Amazon:An Overview. Ecological Economics 18:67–80.

Walker, R.T., Perz, S.G., Caldas, M.M., and Teixeira Silva, L.G. (2002). Land Use and Land Cover Changein Forest Frontiers: The Role of Household Life Cycles. International Regional Science Review 25(2):169–199.

Walsh, S.J., Crews-Meyer, K.A. (eds.) (2002). Linking People, Place and Policy: A GIScience Approach,Kluwer, Boston.

Wood, C.H. (2002). Introduction: land use and deforestation in the Amazon. In Wood, C.H., and Porro, R.(eds.), Deforestation and Land Use in the Amazon, University of Florida, Gainesville, pp 1–38.

York, R., Rosa, E.A., and Deitz, T. (2003). Footprints on the Earth: The Environmental Consequences ofModernity. American Sociological Review 68(2): 279–300.

Hum Ecol (2006) 34:829–849 849