HOUSEHOLD LIFE CYCLE, PROPERTY LIFE CYCLE AND DEFORESTATION IN - - PDF document

1

HOUSEHOLD LIFE CYCLE, PROPERTY LIFE CYCLE AND DEFORESTATION IN BRAZILIAN AMAZON: THE REGION OF MACHADINHO D’OESTE

Vanessa Cardoso Ferreira - Department of Demography – CEDEPLAR/UFMG Gilvan Guedes - Department of Demography – CEDEPLAR/UFMG Alisson Barbieri - Department of Demography – CEDEPLAR/UFMG This work was developed with the support of CNPq and CEDEPLAR/UFMG. XXVIII IUSSP International Population Conference in Session “Population and vulnerability to environmental change” - Friday, 3rd November 2017 Abstract: It is crucial to study the causes and consequences of deforestation on the global climate balance given the potential impact it can have. The Brazilian Amazon suffered a great increase on its deforestation rates, part of which is result of human occupation (fields and pasture). The purpose of this paper is to understand how socio-demographic characteristics

• f smallholders are related to deforestation of Machadinho D'Oeste, Rondônia, Brazilian
• Amazon. Theoretical frameworks of the border stages, household life cycle, and the

property life cycle were used as a backdrop for understanding the context of deforestation in this region, as consequence of the actions of established family farms in the region in two different stages of the frontier. The proposed model has the dependent variable expressed as a proportion of total lot area. Following Beta regressions were used: Generalized model, correction per fraction, and regressions Beta Zero inflated. Also, the proportion multivariate regression was used for combined beta-logistic distributions. Results indicate that the household life cycle and property life cycle markers are good candidates to explain deforestation in this region. It is one of the first times that an Amazon region was studied under the perspective of the beginning of the colonization project. Keywords: Brazilian Amazon, Deforestation, Household Life Cycle, Property Life Cycle.

2 INTRODUCTION The theoretical field of demography and environment has grown in recent years, mainly due to the increase of problems related to the global environment. Human action is increasingly seen as responsible for generating various types of environmental degradation such as desertification, drought, deforestation soil pollution, water pollution, and atmospheric pollution. The Amazon has been a recurrent focus of this debate, since this biome has increasingly been threatened by human activities. The Amazon forest is responsible for planetary balance. Changing this balance can lead to a number of environmental consequences on a global scale. Deforestation is a major problem associated with Amazonian environmental issues and has often been addressed on a macro scale (Perz et al. 2005; Barbieri 2007). However, in recent years within the field of demography, there was a very influential tendency to explain the context of Amazonian frontier under a micro perspective, i.e. at the household

• level. Under this interpretation, deforestation may be strongly related to the agro-pastoral

practices adopted by settlers (families) (Walker et al. 2002; Guedes 2010; Schmink and Wood 2012; McCracken et al. 1999). Similar to other recent studies, (Vanwey 2007; Guedes 2010) this analysis utilizes a micro perspective of the household life cycle to interpret domestic demographic effects on land use and land cover. According to this approach, small farmers adjust their survival strategies according to their consumption needs and their ability to work as they age. The property life cycle approach shows that the time of residence in the field influences the behavior of the settlers regarding land use. Using the Settlement Project Machadinho D'Oeste, located in the Brazilian Amazon, as a case study, this paper analyzes how the change in household life cycle and property life cycle influence the dynamics of land cover in the agricultural frontiers of the

• Amazon. The studies already done in this research area used the effects of variables (such

as residence time on the property, age of household head, number of adults and children in the lot) on land cover (Walker et al. 2002; Guedes 2010; Vanwey 2007). These variables

3 were considered in this study, and additional relevant variables were incorporated to improve model performance. Unlike many studies that have made this type of analysis on already established frontiers, this paper tests these relations since the beginning of the settlement project – as well as Guedes et al. (2017b). It uses data on the settlers and the socioeconomic, demographic, and land cover of lots from their time of arrival at the

• frontier. This is one of the firsts studies in The Amazon that considers this complete time

span on its data analysis. RELATION OF DEFORESTATION WITH THE HOUSEHOLD LIFE CYCLE, PROPERTY LIFE CYCLE AND BORDER STAGES IN MACHADINHO D’OESTE As demonstrated in previous studies (McCracken et al. 1999; Vanway et al. 2007; Guedes et al. 2011, Guedes et al. 2017b), this paper uses the Household Life Cycle (HLC), Property Life Cycle (PLC), and evolutions of frontiers to explain the change in soil cover in two different stages of the agricultural frontier in Machadinho D'Oeste. HLC is assumed as the role of family composition (number of inhabitants, age and sex of a household unit) in decision making and definition of survival and (re) production strategies as families age. The family composition is essential, since it determines the consumption needs, the amount of labor available, the accumulation of capital, and the generation of surplus families over time (McCracken et al. 1999; Walker et al. 2002; Brondízio et al. 2002; Walker 2003; Caldas et al. 2003; Vanwey et al. 2007; Barbieri et. al. 2005; Guedes et al. 2011). When considering the property/batch component in the analysis, some authors (Barbieri et al. 2005; Guedes et al. 2011; Guedes et al. 2017a) showed there are two elements that should be observed. One element is regarding to the time of existence of the lot (i.e. how old is the lot) and the second one is related to the residence time of the head of the household in the lot (i.e. how long does the household head leave in the lot). The present analysis considered the concept of Life Cycle in the Property, i.e., the one that concern how long the head of the domicile resides in the lot.

4 PLC was also important in the present study because, according to Barbieri et al. (2005) and Vanwey et al. (2007), the time a family lives in a property determines the level

• f knowledge about the biophysical characteristics of their property. Thus, it is expected

that in the beginning of the occupation process in a region, newly arrived settlers, having little knowledge about the soil of their property, need to perform experiments to verify which crops best fit that soil. The opposite is expected from older residents or inhabitants from older lots, since knowledge allows them to specialize in crops appropriate to the type

• f soil in their property (Barbieri et al. 2005; Vanwey et al. 2007)

The following is presented here as the two approaches to communicate with the stages of the agricultural frontier in relation to the soil cover. Studies show that deforestation is more sensitive to the effects of HLC at the beginning of the border, with HLC more influential than PLC at that time. This is justified by the fact that in moments of initial occupation of a region, the inexistence, or minimal existence of interactions with markets, causes production (and, consequently, land use and cover) to be closely associated with the structure domicile, depending heavily on the families' ability to provide labor. The survival of the families will be associated with everything they can produce on the lot, which generates the need to deforest a larger proportion of land. However, when the frontier evolves, this relationship is inversed. The evolution of borders is associated with the increase of market relations and market exchanges with urban

• centers. As the boundaries evolve, the PLC becomes more influential, especially as a result
• f the relationship with the market, which will determine the production demands based on
• commercialization. It is also worth noting that as time passes, a natural deterioration (wear)
• f the soils is verified. This is another factor that incorporates importance to the PLC in

more advanced stages of the border, since the knowledge of the most appropriate crops for the soil of the region becomes essential in the face of the reduction of fertility of the soils. Recent results for the micro relationship between demographic dynamics and deforestation at Amazonian agricultural frontiers have shown that the home life cycle approach has received little empirical support since these effects have been complex and inconsistent (Vanway et al. 2007; Barbieri 2007; Guedes et al. 2011; Côrtes and D'Antona 2014). However, the above results referred to analyses for periods in which the frontier was

5 in a more advanced or consolidated stage. More specifically, the period of analysis of these studies was the 1990s and early 2000s. The lack of empirical support attributed to HLC in more consolidated frontiers is based on the fact that, in stages of greater economic integration, home production would be more tied to external market demand and less internal characteristics of the home. Guedes et al. (2011), VanWey et al. (2007) and Guedes et al. (2017b), argued that household dynamics would have a greater influence on deforestation, especially at more recent agricultural frontiers. Therefore, we identify that part of the aforementioned lack of relationship is related to the time window in which the analysis was performed. Unlike other studies, this article, as well as Guedes et al. (2017b), analyzes the results of demographic dynamics on deforestation using a time window that begins with the

• ccupation of a region. In this way, it is possible to study the influence of the demographic

composition (CVD) on deforestation, precisely at the moment in which the theory indicates that this relation is stronger. DATA AND METHODS Area of study This study uses areas of Machadinho D’Oeste as a case study. Those areas consisted mainly of a settlement process in rural regions, started by the federal government during the 1980s in Brazil. This region is located between the municipalities of Ariquemes and Jaru, around 400km from the Porto Velho (capital of the State of Rondônia), Brazil (Miranda et

• al. 2015).

The Settlement Project was approved in 1981 and partially financed by the World Bank, in the Polonoroeste project, implemented in 1982 by INCRA (“Instituto Nacional de Colonização e Reforma Agrária”) (Barbieri et al. 2014; Miranda et al. 2015). According to Monte-Mór (2004), Machadinho Project of Colonization presented a better infrastructure when compared with other colonization projects at Rondônia, which held more investments

6 to this region. Another unique feature of this region was the location of roads between the watershed dividers. These roads allowed the flow of water in the lots, and reduced the soil

• erosion. This project represented a great advance regarding the traditional models of
• ccupation that followed the “fishbone” format previously adopted in the entire State

(Monte-Mór, 2004). Most of the migrants to this region came from the southern states of Brazil, which at that time were densely populated (Monte-Mór 2004; Miranda et al. 2015). The occupation of agricultural lands in Machadinho D’Oeste started in 1984. In July

• f 1985 the urban nucleus of this lands was already a small village in expansion. In 1987,

this location had more than 1,500 houses, although a great number of them were not finished or their use was tied to external market demand and less internal characteristics of the home (only as a second residence for rural families) (Monte-Mór 2004; Guedes et al. 2013). Figure 1 shows a map of the region. The region was determined as a municipality in 1988, its limits were expanded, new areas were incorporated (four other colonization projects and 8 urban centers), resulting in a total urban area of 11,800 km2. In 1989, the municipality had around 30,000 inhabitants, with over 2,000 families in the urban nucleus. The urban nucleus was endowed with basic infrastructure, including an elementary school, a hospital, a bank agency, an SUCAM (“Superintendência de Campanhas de Saúde Pública”) post, an INCRA office, and a Technical Center, where the general support administration worked (MIRANDA et al., 2015). According to Monte-Mór (2004), around 1990 this region presented typical characteristics of a consolidated frontier. The Center for Development and Regional Planning (Cedeplar) of Universidade Federal de Minas Gerais (UFMG) conducted research at Machadinho D’Oeste, and collected information in five field surveys, covering 25 years of history in the region. This research provided the necessary information to verify the relationship between HLC, PLC, frontiers evolution and land cover. These data are the most extensive collection on rural areas in the Amazon, and is the only one that has followed a region since the beginning of the settlement project.

7 Figure 1: Location of Machadinho D’Oeste and sectors

Source: Barbieri et al. 2014.

Survey Data Survey data collected from settlers of Machadinho D'Oeste from the beginning of the settlement project was used in this analysis. It contains information from around 1985 through 2010, from the project developed by Cedeplar/UFMG. In the years of 1985, 1986, 1987 and 1995 the surveys corresponded to all families

• f farmers from Original Colonization Project (Sector 1 and Sector 2). In 2010, a

representative sample of these regions was collected. For this reason, we chose to focus

• nly in the years 1987 and 1995. The year 1987 correlates with the starting point of the

frontier, giving a margin of 2 years for more effective establishment of families and return

• f the activities produced in the batch. The year 1995 represents a more consolidated stage

8

• f the frontier. The focus on two distinct periods better captures the role of the household

and property life cycle. The work started with a total of 765 lots, 820 households and 3,965 individuals in

• 1987. This was reduced to 751 lots, 804 households and 3,889 individuals after excluding

lots with missing information. In 1995, it consisted of 946 lots, 1,078 households and 5,035 individuals, reduced to a sample of 941 lots, with 1,067 households and 4,978 individuals after excluding missing data. As for those two years the interviews correspond to all farming families in Machadinho D'Oeste, the number of lots, households, and individuals increase between 1987 and 1995. This result mirrors the growth in the region for the period (Table 1).

Table 1: Description of the Sample Results – Machadinho D’Oeste – 1987 and 1995 Year Universe Analytical Sample Household Lot Individuals Household Lot Individuals 1987 820 765 3.965 804 751 3.889 1995 1.078 946 5.035 1.067 941 4.978

In addition to these data, we also used the information produced by the Land Use, Climate and Infections in Western Amazonia (LAI/LUCIA), and the “Dinâmica Demográfica e Uso Da Terra na Amazônia: Um Estudo Longitudinal para a Região de Machadinho, Rondônia” project, coordinated by Cedeplar/UFMG. For the analysis of deforestation, data from Imazon and satellite images were used (CEDEPLAR, 20014). The Deforestation Model: Deforestation was analyzed using a model that considers the dependent variable as a proportion, i.e. the dependent variable is expressed as a proportion of total lot area. The theoretical model was applied and estimated by lots, instead of aggregated units.

9 In this model the following Beta regressions were used: Generalized model, correction per fraction, and regressions Beta Zero inflated, i.e. the proportion multivariate regression for combined beta-logistic distribution. The deforested area was measured in square meters, and built from Imazon data and satellite images. Data on land cover were divided into the following groups: water, field, Amazon field, exposed soil, arboreal vegetation, and shrubby vegetation. Therefore, the deforested area was constructed as follows:

Deforested area = total area – (arboreal vegetation + shrubby vegetation + water)

The dependent variable “proportion of deforested area” was obtained by the relation between the deforested area and the total area in each lot. The independent variables were grouped into four categories: (a) variables of the household life cycle; (b) variables of the property life cycle; (c) variables of market integration; and (d) control variables. Three steps were performed: (1) Model of household life cycle and property life cycle, represented as “A”; (2) Model of household life cycle and property life cycle with market integration, represented as “B”; and (3) Model of household life cycle and property life cycle with market integration, including the control variables, represented as “C”. Beta regressions: The linear regression model is not appropriate for situations where the response variable is restricted to an interval (say, 0,1). This is because linear regression models can produce adjusted values for the variable of interest that exceeds its upper and lower bounds (0 or 1). In addition, the effect of explanatory variables tends to be non-linear and the variance tends to decrease as the mean approaches the limit values (in this case, 0 or 1). According to Ferrari and Cribari-Neto (2004), the beta regression model is adapted to situations in which the dependent variable (y) is continuously measured in a standard unit interval, that is, 0 <y <1.

10 This model is based on the assumption that the response has a beta distribution, using a parameterization of the beta law, which is indexed by the mean and dispersion

• parameters. The beta distribution is very flexible to model proportions, since its density can

take different forms, depending on the values of the parameters in which the distribution will be indexed (Ferrari and Cribari-Neto 2004; Paolino 2001). In the beta distribution, the density is given by:

𝜌 (y; p, q) = 𝛥 (𝑞 + 𝑟) 𝛥 (𝑞) 𝛥 (𝑟) 𝑧𝑞−1(1 − 𝑧)𝑟−1, 0 < 𝑧 < 1

where p > 0, q > 0 and Γ (.) is the gamma function. The mean and variance of y are given by:

𝐹(𝑧) = 𝑞 (𝑞 + 𝑟) 𝑤𝑏𝑠(𝑧) = 𝑞𝑟 (𝑞 + 𝑟)2(𝑞 + 𝑟 + 1)

The beta regression model naturally accommodates non-constant variances and asymmetry (Ferrari and Cribari-Neto 2004; Paolino 2001). The beta distribution is a flexible distribution that can produce a unimodal, uniform or bimodal distribution (Paolino 2001). The beta regression model is described as:

𝑕(µ𝑗) = 𝑦𝑗

𝑈𝛾 = 𝜃𝑗

Where: i = 1, ..., n; β = (β1, ..., βk)T is a vector kx1 of unknown regression parameters (and where k < n); xi = (xi1, ..., xik)T is a vector of k regressors and ηi is a linear predictor; g (.) is a strictly monotonic and twice differentiable function that can assume the values (0 ; 1) in the sets of real numbers (IR) (Ferrari and Cribari-Neto 2004; Cribari-Neto, Zeileis 2010). In this case, the regression parameters of the beta regression model are interpretable in terms of the mean of the response. When the logit link is used, the parameters are interpreted by means of an odds ratio, contrary to the parameters of a linear regression employing a transformed response variable. The estimation in this type of model is

11 performed by maximum likelihood, with strictly asymptotic interpretation (Ferrari and Cribari-Neto 2004). The beta regression has the problem of dealing with the dependent variable (y) only in the interval 0 <y <1, that is, boundaries are not included. There are two possibilities for including these values: (i) assuming that the values 0 's and 1's represent very low or very high proportions, which can "accidentally" result in a ratio of 0 or 1 (note that here 0's and 1's occur through the same process as the other proportions); (ii) assuming that the values

• f 0's and 1's represent processes different from those experienced by the other proportions,

which implies zero-one-inflation values. When considering the possibility of inclusion (i), only the mean is modeled, but the variance is not. The mean was calculated using the betafit glm command in combination with the link(logit) family(binomial) robust command options of the statistical package Stata / SE 12.00. When considering the possibility of inclusion (ii), the model is estimated in three parts: logistic regression for proportions that are equal to 0, inflated to 0, but not to 1; logistic regression for proportions that are equal to 1, inflated in 1, but not 0; model for the proportions between 0 and 1, inflated in both. This alternative was calculated using the zoib command in combination with the oneinflated and robust command options of the statistical package Stata / SE 12.00. Description of the Model Variables Dependent variable The dependent variable (deforested area) was measured in square meters and restricted to the lot area. As mentioned before, it was built from Imazon data and satellite

• images. The intensity of deforestation was defined as the proportion of deforested area.

Independent variables The following variables were used as indicators to HLC: household head age; number of children (persons between 0 and 9 years old); number of adolescents (persons

12 between 10 and 14 years old); number of adults (persons between 15 and 59 years old); number of elderly (persons above 60 years old). The household head age is important, because it is an indicator of the experience with agricultural activities (even before the arrival at the region of study), and it can be positively or negatively related with the deforestation (Perz et al. 2001; Walker 2002; Guedes et al. 2017b). Number of adults (and eventualy the number of adolescents) represents the family agricultural potential labor force. It is expected to show a positive relation between this variables and the deforestation (McCracken et al. 1999; Vanwey et al.2007). The number

• f children, adolescents, and elderly in the lot is a way to estimate the level of dependency.

As well as occur with the household head age, total dependents can have dubious effect on deforestation. In general, a higher ratio of youth dependency is expected to increase deforestation due to the initial growth in consumption needs. Regarding the elderly dependency, it is expected that in situations where the land use system is predominantly extensive with low labor needs (empty nest scenario), the effect is positive (Walker et al. 2000). However, in situations where there are multiple families, the labor-intensive soil system tends to prevail (generational shift scenario), and this effect is negative (Perz 2001; Walker et al. 2002; Browder et al. 2004) Time of residence in lot was used as a as indicators to PLC, as well as in VanWey et

• al. (2007) e Barbieri et al. (2005). As mentioned at the theoretical part of this paper, the

time of residence in lot represents a specific knowledge of the lot, which reduces the need for experimentation with land. For the analysis of integration with the market the indicator distance to the urban center was used. The distance from Machadinho urban center was used to approximate the external influences on land use and the degree of connection of the border with the markets. It expresses, therefore, the relation between cost and return of the agricultural practice destined to the market (sales). The income variables (agricultural income – percentage; and family income – value) were incorporated with the purpose of measuring the intensity of the relationship with the market or the degree of market dependence of the lot (Caldas et al. 2007; Summers 2008).

13 FINDINGS Descriptive Analysis: Table 2 shows the descriptive analysis of the data for all the variables considered in the rotated models. The last two of the table represent a test of paired correlation between the dependent variable (deforestation rate) and each one of the independents variables. In general, the tendency presented is quite consistent with the conceptual frameworks suggested in the theoretical part of this work. Note that in 1987, practically all lots had already been deforested to some extent in Machadinho D'Oeste (only one lot had not been cleared). The proportion of deforested area per lot varied from 0 to 89.33% (that is, no lot had been completely deforested), with an average proportion of 14.27%. In 1995, the proportion of deforested area varied from 0 to 100%, with an average proportion of 43.62%. Figure 21 above show the evolution of deforestation rates between 1987 and 1995. It suggested that the deforestation rates grew (i.e., the points were less green) from rates between 0-25(%) and 26-50(%), in 1987, to 51-75(%) and 76-100(%) in 1995. The HLC and PLC indicators show that there is an aging of families over time, a result that was expected. The average number of children reduced slightly (from 1.6 to 1.2), while the average number of adults (2.3 to 2.4) and the average number of 0.2) is slightly

• increased. The average increase of only 4.3 years of residence in the lot between 1987 and

1995 suggests a turnover effect of owners in the period, typical of the initial stages of the border. Average schooling of household head almost doubled in the period (from 1.5 to 2.5 years) although it remained very low. The percentage of non-agricultural income of farmers in 1987 (88.1%) fell, representing, in 1995, only 25% of the total income of the household, suggesting a process of ruralization of the settlement.

1 The authors are thankful to Malia Jones, Assistant Scientist from Applied Population Laboratory at

University of Wisconsin-Madison, for the assistance in the construction of this map.

14

Table 2: Descriptive Analysis – Machadinho D’Oeste – 1987 and 1995 Indicators Year Paired Correlation 1987 1995 1987 1995

• Obs. Mis.

Mean

• St. Dev. Obs. Mis.

Mean

• St. Dev.

Deforestation rate Deforestation rate Dependent Variable Deforestation rate 751 14.28 10.58 941 43.62 22.12 1.0000 1.0000 Independet Variables Household Life Cycle Household head age 741 10 39.8 11.1 936 5 42.73 12.03 0.0649* 0.0248 Number of persons between 15 and 59 years 751 2.34 1.32 941 2.46 1.36 0.2286*** 0.1115*** Number of persons between 0 and 9 years 751 1.56 1.54 941 1.23 1.32

• 0.0624
• 0.0007

Number of persons between10 and 12 years 751 0.44 0.69 941 0.45 0.69 0.1093*** 0.0952*** Number of persons between 13 and 14 years 751 0.30 0.55 941 0.31 0.54 0.0641 0.0488 Number of persons above 60 years 751 0.11 0.37 941 0.18 0.47 0.0262 0.0748 Depandency ratio 718 33 1.21 1.06 887 54 1.10 0.98

• 0.0331

0.0334 Dependents 751 2.63 1.98 941 2.45 1.69 0.0048 0.0634* Property Life Cycle Time of Residence (years) 750 1 1.6 1.0 920 21 5.9 3.6 0.0888** 0.0954*** Market integration Distance to the urban center 751 23,677.5 10,663.5 941 24,546.5 10,471.3

• 0.0412
• 0.1162***

Control Variables Household education (years) 737 14 1.53 1.89 937 4 2.45 2.14 0.0006 0.0313 Agriculture income (percentage) 591 160 11.67 16.72 843 98 61.02 43.71 0.0207 0.1421*** Outside income (percentage) 591 160 88.14 17.14 843 98 38.98 43.71

• 0.0467
• 0.1421***

Family Income (value) 630 121 3,664.48 2,734.99 856 85 11,864.42 32,270.73 0.1411*** 0.0585* Note: Standart Deviation: *** p<0.01, ** p<0.05, * p<0.

15

Figure 2: Deforestation Rates in Machadinho D’Oeste – 1987 and 1995

16 Regressions Beta Models: Taking, for example, the number of adults in 1987, we can observe that having an

• lder adult at home is associated with an increase in the mean proportion deforested by

0.013 (Model A) (Ferrari, Cribari-neto, 2004). The results of the beta model are consistent with the literature on the HLC and the PLC on deforestation, both for 1987 and 1995. This model was first corrected by assuming that the 0's and 1's values represent very low or very high proportions, which may "accidentally" result in a ratio of 0 or 1 ("Beta glm"). In this case, 0's and 1's occur through the same process as the other proportions, and here only the mean was modeled (equidispersion assumption of the precision parameter). The results show that results from the previous models (A, B and C) were underestimated, both in 1987 and 1995 (Table 3 and Table 4). The beta regression model was corrected assuming that the values of 0's and 1's represent processes different from those experienced by the other ratios, which implies zero-one-inflation values. Here, in addition to the mean, the variance was also modeled, relaxing the equidispersion assumption of the beta distribution precision parameter. Results presented for this model were very similar to those presented in the first beta regression models. Given that the distribution of the 0's and 1's followed the distribution of the other proportions of the original beta regression models, this result may indicate that there were not many observations containing 0's and 1's in this sample, therefore not significantly influencing the result. It is worth mentioning that in 1987 there were no 1 values in the sample, so the correction "1-inflated" was not performed, this

• ccurred only in 1995.

The variables: number of adults, number of people aged 10 to 14, number of elderly, time of residence in the lot and income variables were positively correlated with the proportion of the deforested area of the lot. The variable Euclidean distance with respect to the nearest urban center presented negative correlation with the proportion of the deforested area of the lot, which was also expected. Lastly, the variable number of children (persons between 0 and 9 yr. old) was negatively correlated with the proportion deforested in almost

17 all models (except for 1995, model A of the beta regression and of the zero-one-inflation regression). Despite the use of the various models, the effect of each variable on the average change in the proportion of the deforested area remained very close between the different modeling strategies, both in 1987 and 1995. This advocates in favor of the consistent effect found under different hypotheses, reinforcing the empirical evidence. Results indicate that the effect of the integration variable with the markets (distance to the urban center) on deforestation was higher in 1995 than in 1987. However, the same is true for the other variables of the life cycle and the lot.

18

Table 3: Beta Regression Models of Proportion of Deforestation – 1987 Variables A B C A B C A B C Household head age

• 0.000570
• 0.000569 -0.000848**
• 0.000559
• 0.000541
• 0.000880*
• 0.000569
• 0.000568 -0.000847*

(0.000372) (0.000371) (0.000421) (0.000427) (0.000427) (0.000465) (0.000396) (0.000395) (0.000435) Number of persons between 0 and 9 years

• 0.00438*
• 0.00453*
• 0.00455*
• 0.00679*** -0.00696***
• 0.00664**
• 0.00438*
• 0.00452*
• 0.00454

(0.00236) (0.00236) (0.00267) (0.00262) (0.00264) (0.00286) (0.00247) (0.00249) (0.00279) Number of persons between10 and 14 years 0.0106*** 0.0105*** 0.0112*** 0.00730* 0.00726* 0.00948** 0.0105*** 0.0105*** 0.0112*** (0.00357) (0.00357) (0.00410) (0.00405) (0.00405) (0.00479) (0.00349) (0.00350) (0.00422) Number of persons between15 and 59 years 0.0126*** 0.0127*** 0.0112*** 0.0162*** 0.0163*** 0.0149*** 0.0126*** 0.0127*** 0.0112*** (0.00247) (0.00248) (0.00273) (0.00342) (0.00345) (0.00360) (0.00362) (0.00364) (0.00387) Number of persons above 60 years 0.0110 0.0113 0.0210* 0.0135 0.0138 0.0216** 0.0110 0.0112 0.0209* (0.0101) (0.0101) (0.0114) (0.0103) (0.0101) (0.0110) (0.0107) (0.0106) (0.0117) Time of residence (years) 0.0121*** 0.0120*** 0.00812** 0.0104*** 0.0103*** 0.00555 0.0121*** 0.0120*** 0.00810* (0.00330) (0.00330) (0.00376) (0.00382) (0.00380) (0.00446) (0.00365) (0.00366) (0.00430) Distance to the urban center

• 2.86e-07
• 3.85e-07
• 5.02e-07
• 5.62e-07
• 2.85e-07
• 3.85e-07

(3.24e-07) (3.60e-07) (3.51e-07) (3.85e-07) (3.25e-07) (3.58e-07) Agriculture income (percentage) 0.000276 0.000475* 0.000276 (0.000239) (0.000253) (0.000287) Family Income (value) 4.06e-06*** 5.34e-06*** 4.05e-06** (1.51e-06) (1.83e-06) (1.74e-06) Constant

• 2.053*** -1.998***
• 1.970***
• 2.080***
• 1.989***
• 2.013***
• 2.053*** -1.998***
• 1.970***

(0.128) (0.142) (0.166) (0.155) (0.168) (0.201) (0.145) (0.161) (0.185) Observations 739 739 579 740 740 580 740 740 580 Constant (ln_phi) 2.382*** 2.383*** 2.402*** 2.382*** 2.383*** 2.402*** (0.0524) (0.0524) (0.0592) (0.0733) (0.0731) (0.0813) Observations 739 739 579 740 740 580 Constant (z.inflate)

• 6.605*** -6.605***
• 6.361***
• 1.001
• 1.001
• 1.002

Observations 740 740 580 Note: Standart Deviation: *** p<0.01, ** p<0.05, * p<0.1 BETA BETA (glm) BETA (zoib)

19

Table 4: Beta Regression Models of Proportion of Deforestation - 1995

Variables A B C A B C A B C Household head age

• 0.00172**
• 0.00792**
• 0.00213***
• 0.00198**
• 0.00221***
• 0.00239***
• 0.00171**
• 0.00195**
• 0.00212***

(0.000779) (0.00314) (0.000796) (0.000793) (0.000791) (0.000818) (0.000789) (0.000784) (0.000813) Number of persons between 0 and 9 years 0.000143

• 0.00452
• 0.00190
• 0.00121
• 0.00228
• 0.00338

0.000143

• 0.00111
• 0.00189

(0.00575) (0.0231) (0.00583) (0.00619) (0.00615) (0.00623) (0.00633) (0.00625) (0.00637) Number of persons between10 and 14 years 0.0142* 0.0595* 0.0156** 0.0146* 0.0150** 0.0165** 0.0142* 0.0146* 0.0156** (0.00766) (0.0308) (0.00776) (0.00748) (0.00752) (0.00742) (0.00744) (0.00753) (0.00737) Number of persons between15 and 59 years 0.0234*** 0.101*** 0.0276*** 0.0227*** 0.0242*** 0.0265*** 0.0233*** 0.0250*** 0.0275*** (0.00579) (0.0233) (0.00599) (0.00593) (0.00592) (0.00590) (0.00598) (0.00598) (0.00595) Number of persons above 60 years 0.0723*** 0.312*** 0.0765*** 0.0751*** 0.0792*** 0.0796*** 0.0720*** 0.0767*** 0.0762*** (0.0188) (0.0755) (0.0189) (0.0198) (0.0194) (0.0199) (0.0194) (0.0188) (0.0192) Time of residence (years) 0.00558*** 0.0237*** 0.00228 0.00614*** 0.00642*** 0.00295 0.00556*** 0.00585*** 0.00227 (0.00204) (0.00821) (0.00215) (0.00207) (0.00205) (0.00217) (0.00215) (0.00213) (0.00229) Distance to the urban center

• 1.28e-05*** -3.51e-06***
• 2.76e-06*** -3.07e-06***
• 3.15e-06*** -3.49e-06***

(2.78e-06) (7.13e-07) (7.56e-07) (7.93e-07) (8.18e-07) (8.76e-07) Agriculture income (percentage) 0.000936*** 0.000959*** 0.000933*** (0.000185) (0.000181) (0.000177) Family Income (value) 5.73e-07* 7.71e-07** 5.71e-07** (3.08e-07) (3.59e-07) (2.70e-07) Constant

• 0.409***
• 0.0758
• 0.194
• 0.409***
• 0.113
• 0.252
• 0.409***
• 0.0758
• 0.194

(0.139) (0.156) (0.168) (0.143) (0.162) (0.176) (0.142) (0.161) (0.174) Observations 912 912 820 915 915 823 915 915 823 Constant (ln_phi) 1.419*** 1.443*** 1.506*** 1.419*** 1.443*** 1.506*** (0.0424) (0.0425) (0.0451) (0.0545) (0.0516) (0.0554) Observations 912 912 820 915 915 823 Constant (z.inflate)

• 6.816***
• 6.816***
• 6.709***
• 1.001
• 1.001
• 1.001

Observations 915 915 823 Constant (o.inflate)

• 6.122***
• 6.122***
• 6.016***

(0.708) (0.708) (0.708) Observations 915 915 823 BETA BETA (glm) BETA (zoib) Note: Standart Deviation: *** p<0.01, ** p<0.05, * p<0.1

20 Analysis of the predicted values of the proportion deforested by age of the household head (Figure 3) reveals a pattern consistent with what has been demonstrated in the literature. In 1987, it was observed that the proportion of deforestation has an inverted U-shape curve, i.e. it starts with low values and the slope changes when the household head is about 45 to 50 years of age. Consistency with the literature comes from the fact that the household heads reach the frontier when they are young, and most often do not yet have children. Therefore, with little manpower to work in lots, the deforested area ratio is low. As the household heads get

• lder and so do their children, the number of workers in the household increases allowing

greater use of the soil, and also increases the deforestation of the lot. As time passes, the household heads age and the children begin to leave home, reducing the pressure on demand for additional forest, since the family begins to devote less intensive activities on hand labor. Results for 1995 also followed a very similar pattern, although the initial clearing rates were slightly higher (which was expected, since families were already in the frontier for a longer time).

Figure 3: Predicted Value of the proportion of deforestation by age of the household head – 1987 e 1995

Similarly, analysis of predicted values of the proportion deforested by time of residence (Figure 4) reinforces the considerations made in the literature. In 1987, the graph has an "inverted U" curve, suggesting that in the first years there is an intensification of

,11 ,12 ,13 ,14 ,15 10 20 30 40 50 60 70 80

Household Head Age

1987 ,42 ,43 ,44 ,45 ,46 10 20 30 40 50 60 70 80

Household Head Age 1995

21 demand for land (deforestation), reflecting a phase of years of learning (land experimentation), in which the settlers, in an attempt to produce on soil, end up generating high deforestation. This no longer appears for the year of 1995, where the tendency is descending from the first years. This result suggests a "learning effect" of younger farmers at more advanced stages of the frontier (Vanwey et al. 2012; Guedes et al. 2017b). Among older farmers the additional demand for land is lower and the trend continues. This was supported even over a long period of stay, represented by the simulated values (“SimA”, “SimB” and “SimC”). There are two possible causal explanations for the result found. The first is that there has been an increase in the specific capital of the land, and the second is that the systems of use have become less intensive on land, reducing deforestation (Vanwey et al. 2012, Guedes et

• al. 2017b).

Figure 4: Predicted Value of the proportion of deforestation by residence time – 1987 e 1995

22 CONCLUSION A consequence of Machadinho D'Oeste region’s occupation was the increased deforestation in the region. Between 1987 and 1995 the average rate of deforestation in rural areas of this region increased from 14.28% to 43.56% of the total area. Results presented in this study for household head age and residence time in the lot show interesting findings. The deforestation patterns increase non-linearly with the household life cycle, increasing its intensity in the period of lot consolidation (nuclear families and young children), and reversing the trend of deforestation when households get

• lder. At the same time, the non-linear effect of residence time in lot on deforestation

suggests that the early years are predominantly years of soil experimentation, as advocated by VanWey et al. (2007), VanWey et al. (2012), and Guedes et al. (2017b). Despite innovations in this article, with a longitudinal base that captures the early years of colonization and a model that comes from a class of probability distribution more suitable for the analysis of proportions, some limitations are noteworthy. In this study we do not control models by lot size in the previous period, which is highly correlated with the lot carrying capacity in the current period, which may affect results. Regardless of the limitations and possibilities, this study presents an analysis of demographic dynamics and forest cover in agricultural frontiers in the setting time of the frontier finding consistent effects to the household life cycle and property life cycle. The possibility to work with data from the initial time of the settlement provided us an analytical advantage for this type of study, rarely found in population and demographic studies in frontier environments.

23 REFERENCES Barbieri, A. F.; Bilsborrow, R. E.; Pan, W. K. 2005. Farm household lifecycles and land use in the Ecuadorian Amazon. Population and Environment, New York, v. 27, n. 1, p. 1- 27, Sept. Barbieri, A. F. 2007. Mobilidade populacional, meio ambiente e uso da terra em áreas de fronteira: uma abordagem multiescalar. Revista Brasileira de Estudos de População, v. 24,

• n. 2: 225-246.

Barbieri, A. F.; Guedes, G. R.; Antigo, M. 2013. Demographic Dynamics, Livelihoods and Land Use: a Twenty Five Years Longitudinal Study for the Brazilian Amazon. In: Annals of the IUSSP Conference, Busan, South Korea. Barbieri, A. F.; Santos, R. O.; Guedes, G. R. 2014. The migration, environment and development nexus in the frontier: a review of the literature based on empirical evidences from the Brazilian Amazon. In: Determinants of International Migration Conference, 2014, Oxford. Bilsborrow, R. E., A. S. Oberai, et al. 1984. Migration Surveys in Low Income Countries: Guidelines for Survey and Questionnaire Design. London and Sydney, Croom Helm, cap. 2 e 3. Bilsborrow, R. and H. Okoth-Ogendo. 1992. Population-driven changes in land use in developing countries. Ambio 21(1): 37-45. Brondízio, E. S., McCracken, S. D., Moran, E. F., Siqueira, A. D., Nelson, D. R., e Rodriguez-Pedraza, C. 2002. The colonist footprint: Toward a conceptual framework of land use and deforestation trajectories among small farmers in the Amazonian frontier. In:

• C. H. Wood & R. Porro (Eds.). Deforestation and land use in the Amazon. Gainsville, FL:

University Press of Florida, pp. 133–161. Browder, J. and Godfrey, B. 1990. Frontier Urbanization in the Brazilian Amazon: A theoretical framework for urban transition. Conference of Latin American Geographers, 16: 56-66. Caldas, M. et al. 2007. Theorizing land cover and land use change: The peasant economy of Amazonian deforestation. Annals of the Association of American Geographers, v. 97, n. 1: 86-110. Cedeplar – Centro de Desenvolvimento e Planejamento Regional. 2014. Project Land Use, Climate and Infections in Western Amazonia (IAI/LUCIA). Research Report 2014. Belo Horizonte. Chatterjee, S. and Hadi, A. S. 2015. Regression analysis by example. John Wiley & Sons.

24 Cribari-Neto and F.; Zeileis, A. Beta Regression. 2016. In: R. Universidade Fedaral do Pernambuco, 2010. Available at: <http://citeseerx.ist.psu.edu/viewdoc/summary?doi= 10.1.1.489.52 91>. Last acess in 8th January of 2016. Ferrari, S.L.P and Cribari-Neto, F. 2004. Beta Regression for Modelling Rates and

• Proportions. Journal of Applied Statistics, 31(7): 799-815.

Guedes, G. R. 2010. Ciclo de vida domiciliar, ciclo do lote e dinâmica do uso da terra na Amazônia rural brasileira: um estudo de caso para Altamira, Pará. PhD. Theses. Cedeplar/UFMG, Belo Horizonte. Guedes, G. R.; Queiroz, B.L.; Barbieri, A.F.; Vanwey, L. 2011. Ciclo de vida domiciliar, ciclo do lote e mudança no uso da terra na Amazônia Brasileira. Revista Brasileira de Estudos de População. Guedes, G. R.; Antigo, M. F.; Barbieri, A. F. 2013. Poverty, income dynamics, and returns to capitals in agricultural frontiers: a case study for the Brazilian Amazon. IUSSP. Guedes, G. R.; Barbieri, A. F.; Santos, R. O.; Antigo, M. 2015. Estratégias de subsistência e do ciclo de vida na Amazônia brasileira: o caso de Machadinho d'Oeste, Rondônia. Territórios e Fronteiras (Online), v. 8:196-217. Guedes, G.R; Queiroz, B.L.; Barbieri, A. F.; Vanwey, L.K. 2017a. Ciclos de vida de la propriedade y del hogar, mercados y cambios en el uso y la cobertura de la tierra en la Amazonia brasileña. Notas de Población, v.44, n. 104. Guedes, G. R.; Barbieri, A. F; Santos, R. O.; Ferreira, V. C. 2017b. Composição demográfica domiciliar e dinâmica do uso do solo em Machadinho d’Oeste, Rondônia: evidências baseados nos estágios iniciais da fronteira. Revista Brasileira de Estudos de População, v.34, n.2: 271-299. Hammel, E. A. 2005. Chayanov revisited: a model for the economics of complex kin units. PNAS 102(19): 7043-7046. Henkel, R. 1982. The move to the oriente: colonization and environmental impact. In: Tempe, L.; Jerry, R. (Ed.). Modern day Bolivia: legacy of the revolution and prospects for the future. Arizona State University, Center for Latin American Studies. McCracken, S. D. et al. 1999. Remote sensing and GIS at farm property level: demography and deforestation in the Brazilian Amazon. Photogrammetric Engineering and Remote Sensing, v. 65, n. 11: 1311-1320. Miranda, E. E. et al. 2015. Sustentabilidade Agrícola na Amazônia - Machadinho d'Oeste. Campinas: Embrapa Monitoramento por Satélite, 2005. Available at: <http://www.machadinho.cnpm.embrapa.br>.

25 Monte-Mór, R. L. 2004. Modernities in the jungle: extended urbanization in the Brazilian

• Amazonia. PhD. Theses. University of California, Los Angeles (UCLA), Los Angeles.

Paolino, P. 2001. Maximum likelihood estimation of models with beta-distributed dependent variables. Political Analysis, 9(4): 325–346. Perz, S. G.; Aramburú, C.; Bremner, J. 2005. Population, land use and deforestation in the Pan Amazon Basin: a comparison of Brazil, Bolivia, Colombia, Ecuador, Perú and

• Venezuela. Environment, development and sustainability, v. 7, n. 1: 23-49.

Schmink, M.; Wood, C. 2012. Conflitos sociais e formação da Amazônia. [Tradução de Noemi Miyasaka Porro e Raimundo Moura]. Ed.Ufpa, Belém. Summers, P. M. 2008. The post-frontier: land use and social change in the Brazilian Amazon (1992 – 2002). PhD. Theses. Environmental Design and Planning, Virginia Polytechnic Institute and State University, Blacksburg, Virginia. Vanwey, L. K.; D’Antona, A. O.; Brondízio, E. S. 2007. Household demographic change and land use/land cover change in the Brazilian Amazon. Population and Environment, New York, v. 28, n. 3: 163-185. Vanwey, L. K.; Guedes, G. R.; D’Antona, A. O. 2012. Out-migration and land-use change in agricultural frontiers: insights from Altamira settlement project. Population and environment, v. 34, n. 1: 44-68. Walker, R. T. et al. 2002. Land use and land cover change in forest frontiers: the role of household life cycles. International Regional Science Review, Philadelphia, v. 25, n. 2: 169- 199. Walker, R. T. 2003. Mapping process to pattern in the landscape change of the Amazonian

• Frontier. Annals of the Association of American Geographers, Washington, v. 93, n. 2:

376-398.