Income inequality and adult nutritional status: Anthropometric evidence from a pre-industrial society in the Bolivian Amazon

ARTICLE IN PRESS

Social Science & Medicine 61 (2005) 907–919 www.elsevier.com/locate/socscimed

Income inequality and adult nutritional status: Anthropometric evidence from a pre-industrial society in the Bolivian Amazon Ricardo Godoya,, Elizabeth Byronb, Victoria Reyes-Garcı´ aa, Vincent Vadeza, William R. Leonardc, Lilian Apazad, Toma´s Huancaa, Eddy Pe´reze, David Wilkief a

Heller School for Social Policy and Management, Brandeis University, Heller Building MS 078, Waltham, MA 02454-9110, USA b International Food Policy Research Institute, 2033 K Street, NW, Washington, DC 20006-1002, USA c Department of Anthropology, Northwestern University, Evanston, IL 60208, USA d Proteccio´n del Medio Ambiente Tarija – PROMETA, Calle Alejandro del Carpio Na E-0659, Casilla No 59, Bolivia % e Fundacio´n para el Desarrollo de la Ecologı´a - Estacio´n Biolo´gica Tunquini, Bolivia f Wildlife Conservation Society, 18 Clark Lane, Waltham, MA 02451-1823, USA Available online 4 March 2005

Abstract Evidence has been accumulated about the adverse effects of income inequality on individual health in industrial nations, but we know less about its effect in small-scale, pre-industrial rural societies. Income inequality should have modest effects on individual health. First, norms of sharing and reciprocity should reduce the adverse effects of income inequality on individual health. Second, with sharing and reciprocity, personal income will spill over to the rest of the community, attenuating the protective role of individual income on individual health found in industrial nations. We test these ideas with data from Tsimane’ Amerindians, a foraging and farming society in the Bolivian Amazon. Subjects included 479 household heads (13+ years of age) from 58 villages. Dependent variables included anthropometric indices of short-run nutritional status (body-mass index (BMI), and age- and sex-standardized z-scores of mid-arm muscle area and skinfolds). Proxies for income included area deforested per person the previous year and earnings per person in the last 2 weeks. Village income inequality was measured with the Gini coefficient. Income inequality did not correlate with anthropometric indices, most likely because of negative indirect effects from the omission of socialcapital variables, which would lower the estimated impact of income inequality on health. The link between BMI and income and between skinfolds and income resembled a U and an inverted U; income did not correlate with mid-arm muscle area. The use of an experimental research design might allow for better estimates of how income inequality affects social capital and individual health. r 2005 Elsevier Ltd. All rights reserved. Keywords: Health inequalities; Anthropometrics; Nutritional status; Bolivia; Social capital; Tsimane’ Amerindians

Introduction Corresponding author. Tel.: +1 781 736 2784; fax: +1 781 736 2774. E-mail address: [email protected] (R. Godoy).

In recent years, quantitative evidence has accumulated about the adverse effects of income inequality on individual health in industrial nations (Kawachi, 2000;

0277-9536/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2005.01.007

ARTICLE IN PRESS 908

R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

Wilkinson, 1992, 1996). Income inequality affects individual health through at least three overlapping paths (Coburn, 2000; Mellor & Milyo, 2001b). First, income inequality creates psychosocial stress, which contributes to ailments such as depression and heart disease (Wilkinson, 1996, 1997b). Second, income inequality erodes social capital—trust, safety nets, social networks, and community organizations that enable people to act collectively (Egolf, Lasker, Wolf, & Potvin, 1992; Kawachi & Kennedy, 1997a; Kawachi, Kennedy, Lochner, & Prothrow-Stith, 1997; Kawachi, Kennedy, & Wilkinson, 1999; Wilkinson, 1997a). The erosion of social capital creates shame, distrust, envy, and antisocial behavior that harms health, particularly among the poor (Kawachi & Kennedy, 1999, 2002; Macinko & Starfield, 2001; Wilkinson, 1996, 1997b). Last, income inequality proxies for stratification and for socioeconomic heterogeneity, which makes it harder for people to agree on the provision of public goods that protect or enhance health (Deaton, 2001a; Putnam, Leonardi, & Nanetti, 1993). Support for the idea that income inequality worsens individual health come mainly from industrial nations (Wagstaff & van Doorslaer, 2000). We known less about how income inequality and income might affect individual health in small-scale, contemporary pre-industrial rural societies. Anthropologists who typically study such societies have not done quantitative work on the topic, and social epidemiologists who have developed the field generally do not study such societies. As a result, we do not know whether the positive link between income inequality and individual poor health often found in industrial nations also applies in very different socio-economic and cultural settings, or whether it is unique to a certain type of economy and society. We advance three reasons why income inequality might not harm individual health and why income might not protect individual health in small-scale, pre-industrial societies. First, small-scale, pre-industrial societies typically display strong norms of sharing and reciprocity (Godoy, 2001; Gurven, Hill, Kaplan, Hurtado, & Lyles, 2000; Kaplan & Hill, 1985). These forms of social capital should protect individual health from the harmful effect of income inequality. Estimating the relation between income inequality, social capital, and individual health in industrial nations has proven difficult owing, in part, to the role of intervening variables, such as racial and ethnic heterogeneity, immigration, and government transfers. Small-scale, pre-industrial societies provide a more antiseptic and simpler setting to explore the relations since they lack these confounding variables. Second, if income inequality and strong forms of social capital co-exist, then income inequality might improve individual health. With strong norms of sharing and reciprocity, and with strong kinship links among

people in a community, people who are materially better off might be more likely to help their less fortunate brethren with new medical knowledge and practices, thereby improving the individual health of those at the bottom of the income distribution. When reciprocity, kinship, and sharing permeate a society, those at the top of the income distribution might improve the health of sick villagers, thereby improving the average health of all in the community. Again, small-scale, pre-industrial societies provide an ideal setting to explore the possible positive effect of income inequality on individual health. The final reason why the analysis of income inequality and individual health in small-scale, contemporary preindustrial societies is useful has to do with what it can teach us about the protective role of individual income in highly autarkic societies. Standard economic thinking suggests that a person’s own income relates to a person’s own health in a non-linear fashion—rising income protects health, but at a diminishing rate. At low levels of income, income should have a stronger protective effect on health than at higher levels of income. This is often referred to as the Absolute Income Hypothesis and captures the standard view of diminishing marginal utility of a resource (Wagstaff & van Doorslaer, 2000). Many empirical studies have corroborated the idea (Duncan, 1996; Mellor & Milyo, 2001a, b). If true, then we should expect to find that individual income exerts a large protective effect on individual health in smallscale, pre-industrial societies because the societies are highly autarkic and poor in material resources. But if strong norms of redistribution and sharing induce people to share their personal income with others, then the protective effect of individual income on individual health should wane since individual income will get diluted in the community. In sum, we find several reasons to think that in smallscale, pre-industrial societies with strong norms of sharing and reciprocity, neither income inequality nor individual income should affect individual health. This will occur because norms of sharing and reciprocity should protect all people in the group from the presumably adverse effects of income inequality, and the same norms and behaviors should produce leakages of individual income into the community. Here we contribute to the literature on the effects of income inequality on individual health by exploring whether results from developed societies hold in a very different socio-economic setting. Using cross-sectional data from Tsimane’ Amerindians, a foraging and farming society in the Bolivian Amazon, we estimate the effect of individual income and village income inequality on individual anthropometric indices of short-run nutritional status among household heads (13+ years of age). The Tsimane’ are an apt society to study the effect of income inequality and individual income on individual health. As we shall see, adult

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

Tsimane’ display much variance in anthropometric indices of short-run nutritional status and in individual income, and Tsimane’ villages display considerable variance in income inequality. Despite income inequality, Tsimane’ share frequently and display many other forms of pro-social behavior.

Methods Survey and sample Data come from a survey done during June–November, 2000, among Tsimane’ in the department of Beni. The design and the administration of the survey build on 18 months of prior fieldwork by five researchers in two villages, one close to the market town of San Borja and one relatively more remote. The earlier study included 60 households and 325 subjects followed every quarter for five consecutive quarters. The 18-month panel study allowed us to pilot test methods for collecting data and to measure income with accuracy, and culminated in two dissertations, one on the relation between markets and ethnobotanical knowledge, and one on the relation between markets and health (Byron, 2003; ReyesGarcı´ a, 2001). During May–June, 2000, we tested further the survey used in this study with Tsimane’ living in villages close to the town of San Borja. For the survey, we selected villages in the three main regions inhabited by Tsimane’: Pilo´n-Lajas Reserve, Territorio Multie´tnico, and Territorio Uno. In each region, we selected villages close and far from towns, and villages in between. In each village, we first did a population census and we then took a random sample of 12 households (with replacement if they were absent) in villages with more than 12 households. In villages with fewer than 12 households, we interviewed all households. In each household, we selected at random one of the two household heads for the interview, yielding 479 households or household heads in 58 villages. In each village, we surveyed an average of 8.25 household heads (S.D. ¼ 2.75); in half of the villages, we interviewed more than eight adult subjects (Table 1). The Bolivian census of 2001 puts the Tsimane’ population at about 8000 people. Assuming a mean household size of six people, the Tsimane’ population would contain 1329 households. Since we surveyed 479 households, the survey covered 36.04% of the Tsimane’ population. Dependent variables We use three anthropometric indices of short-run nutritional status: (a) body-mass index (BMI; kg/m2), (b) sex- and age-standardized z-scores of sum of triceps and sub-scapular skinfolds (ZSUMSK2), and (c)

909

Table 1 Sample size of households surveyed in 58 villages # Of households surveys per village

Frequency in sample

% Of total households

Cumulative percent

2 3 4 5 6 7 8 9 10 11 12

1 2 2 7 6 4 7 2 15 4 8

1.72 3.45 3.45 12.07 10.34 6.90 12.07 3.45 25.86 6.90 13.79

1.72 5.17 8.62 20.69 31.03 37.93 50.00 53.45 79.31 86.21 100.00

sex- and age-standardized z-scores of mid-arm arm muscle area (ZMAM). We followed the protocol of Lohman, Roche, and Martorell (1988) and measured all subjects in light clothing without shoes. We recorded stature (standing height) to the nearest millimeter using a portable stadiometer or a plastic tape measure and body weight to the nearest 0.2 kg using a standing scale. Mid-arm circumference was measured to the nearest millimeter using plastic tape measures. Skinfold thickness was measured to the nearest 0.5-mm using Lange callipers. We used Frisancho’s (1990) norm based on information from NHANES I–II in the US to standardize measures of skinfolds and mid-arm muscle area. Explanatory variables We measured income in two different ways owing to the difficulties of measuring income in highly autarkic settings (Deaton, 1997). First, we asked subjects to estimate the area deforested for farming by the household the previous year, and divided the area by the number of people in the household. Area deforested provides a reasonable approximation of gross annual income for the previous year that came from: (a) the value of goods consumed from people’s own farm and (b) the sale of farm goods. However, area deforested underestimates true gross annual income because it excludes annual earnings from wage labor and the sale of forest goods, and it also excludes the value of goods extracted from the forest for own consumption. Subject’s reports of area deforested correlated reliably with actual area deforested as verified by measures taken by an independent team of researchers (Vadez et al., 2003). To overcome the shortcomings just noted, we also measured monetary income in a direct way. We asked

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

910

subjects about the value of goods sold, about the value of goods received in barter, and about earnings from wage labor. We collected the information for the entire household for the 2 weeks before the day of the interview and divided the value by total household size to arrive at an individual measure of monetary income. The earnings variable has low random measurement error from poor recall because we limited the recall to the 2 weeks before the day of the interview, but it has less variance than true earnings because of the short recall period. Also, because the measure of earnings refers to the 2 weeks before the day of the interview, it overlaps in time with the measure of anthropometric indices. As a result, the measure of earnings is more endogenous than area deforested, which refers to the previous year. The partial correlation coefficient between area deforested/person and earnings/person was positive but low (0.29). Control variables included the sex, age, and maximum school attainment of the subject, and distance from the village to the nearest town measured with a geographic positioning systems (GPSs) receiver. We used ordinary least squares with robust standard errors and clustering by village because households are nested within villages. Table 2 contains definition and summary statistics of the variables used in the regressions.

Nutritional status, income, social capital, and income inequality: descriptive analysis In recent work, we provide background historical and ethnographic information on the Tsimane’, so here we limit ourselves to a short description of adult nutritional status, income, income inequality, and social capital among the Tsimane’ to set the stage for the econometric analysis (Byron, 2003; Foster et al., 2003; Huanca, 1999; Reyes-Garcı´ a et al., 2003a, b; Reyes-Garcı´ a, 2001; Vadez et al., 2004) . Nutritional status Adult nutritional status among Tsimane’ is low relative to Western standards. For example, the information in Table 2 suggests that age- and sex-standardized zscores of the sum of triceps and sub-scapular skinfold (ZSUMSK2) and mid-arm muscle area (ZMAM) were 0.61 and 0.55 S.D.’s below norms from the US. Age- and sex-standardized measures of height for age and weight for age (not shown) were 1.93 (height for age) and 0.99 (weight for age) S.D.’s below norms from the US. Income The Tsimane’ have low income and are highly autarkic. In the 18-month panel study of only two

Table 2 Definition and summary statistics of variables used in regressions Variable Dependent ZSUMSK2

ZMAM BMI Explanatory Personal-level Deforestation Earnings

Female Age Schooling

Definition

N

Mean

Age- and sex-standardized z-scores of sum of triceps and subscapular skinfolds (ZSUMSK2) and mid-arm muscle area (ZMAM) using Frisancho’s (1990) norms

476

0.613

0.627

472 476

0.551 23.597

0.913 2.462

479

2.998

2.840

479

31.983

51.047

479 476 479

0.448 37.080 0.849

0.497 14.586 1.476

58 58

0.374 0.541

0.107 0.127

58

35.655

23.913

Body-mass index (kg/m2)

Area/person deforested in 1999 in tareas; 10 tareas ¼ 1 ha. Includes old-growth and fallow forests Cash earnings+value of goods received in barter by households in last 2 weeks divided by household size; in bolivianos (1 US$ ¼ 6.03 bolivianos) Sex of subject; 1 ¼ female, 0 ¼ male Age of subject in years Maximum schooling achieved

Village-level measures of inequality (Gini) Deforest-Gini Gini of area/person deforested Deforest-cash Gini of household earnings Village-level controls Distance Distance in miles from village to closest town using GPS receiver

S.D.

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

911

villages that we did before this study, we measured monetary earnings through surveys and the value of all goods brought into the households through weigh days. For weight days, we selected at random 1 day/quarter for each household, and on that day we identified, weighed, measured, and valued all goods brought into the household from morning until dusk. We found that mean annual personal income from cash earnings and from the imputed value of farm and forest consumption was US$332, a third of the average for Bolivia (US$980/ person) or for all low- and medium-income nations (US$1140/person) (Godoy et al., 2002). Goods bought in the market accounted for only 2.7% of the total value of household consumption. Unfortunately, we did not collect comparable information in the cross-sectional study used here, so we cannot assess how the two measures of income we use—area deforested/person last year and earnings/person over the last 2 weeks— correlate with the more precise measure of income from quarterly surveys and weigh days.

giving, communal labor, and labor help. The share of households that made gifts during the week before the day of the interview were as follows: 71% of household gave home-brewed beer (chicha), 58% cooked food, 45% plantains, 42% meat, 37% rice, 32% fish, 31% manioc, 28% maize, and 12% gave medicines and seeds. During the week before the day of the interview, 22–26% of households helped others or engaged in communal hunting, fishing, miscellaneous work, and farming, 13% of households did errands for others, and 8% offered medical help. Only 7.5% of households did not make any gifts, 39.0% of households did not do any communal work or offered any labor help during the week before the day of the interview, and only 4.45% of households did not make either any gifts or offer any help. The figures suggest that Tsimane’ share a wide range of goods and display a wide range of pro-social behaviors (Gurven, 2004).

Social capital

But offsetting manifestations of sharing and generosity one also finds evidence of accumulation and economic inequalities. The presence or lure of public schools, the encroachment of loggers, ranchers, and small farmers moving into the Tsimane’ territory (Godoy et al., 1998), and the debt peonage into which some Tsimane’ have fallen with outside traders—all create incentives to move less and to accumulate more material possessions. With a more sedentary lifestyle the possibilities for the accumulation of wealth rise. Even without the presence of markets, one finds a strong ethos of economic independence among households, reflecting the fact that most of the diet comes from farm and forest goods produced by each household and not from goods produced communally. Young men who have entered the wage labor market often buy prestige commercial items (e.g., radios). Tsimane’ in some of the more accessible villages build walls to enclose their homes and even put fences around their courtyards. To guard their possessions, some Tsimane’ have started to put locks on their doors when they leave the village (Byron, 2003). Even in meals one finds evidence of lack of sharing. Though people eat communally in smaller villages, they do not go out of their way to invite others to share in their meals. Tsimane’ often turn their backs to others when they eat (Ellis, 1996) and people in the more modern villages often complain that neighbors do not share meat. In the earlier 18-month panel study, we probed how households coped with random mishaps over which they had no control (e.g., crop loss, theft), and found that only 5% of the sample received help from kin or from neighbors after a misfortune. The figures in Table 2 for Gini coefficients of income provide further evidence of inequality. The Gini coefficient ranges from 0 to 1, with higher numbers

At first inspection the Tsimane’ appear as an egalitarian society. Like other indigenous Amazonian groups, the Tsimane’ have a preferential system of crosscousin marriage (men marry mother’s brother’s daughter), which creates a thick and wide web of relatives linked by descent and marriage. Households visit each other often within and across villages to see relatives or to exchange goods and information (Ellis, 1996). This survey shows that only 10% of adults lived in their village of birth. Constant visiting and migration between villages homogenizes many outcomes, such as ethnobotanical knowledge (Reyes-Garcı´ a et al., 2003a, b). Like other indigenous Amazonian populations, the Tsimane’ routinely share home-brewed beer (chicha). Any Tsimane’ can walk into a Tsimane’ household serving chicha and expect to be served. Cooking is often done in open courtyards and eating is communal in the smaller villages. Successful hunters share game with close kin and neighbors. In the earlier 18-month panel study, we found that 11% of all goods entering households from morning until dusk came as gifts or as transfers from friends or from relatives; those goods accounted for 6.70% of the total value of household consumption. Fishing with plant poisons is often done communally (Pe´rez, 2001). In this survey, we found that about a quarter of all fishing events with nets or with fish poison were done communally. Communal work prevails in the construction and in the maintenance of schools, in the clearing of soccer fields and public places, and in village festivities. Descriptive statistics from two subsequent panel survey waves done in 2001 and 2002 with about 350 households in 37 villages highlight the prevalence of gift

Income inequality

ARTICLE IN PRESS 912

R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

indicating more inequality. The average village Gini coefficient for area deforested/person is 0.37, but the S.D. is 0.10, with minimum and maximum values (not shown) of 0.10 and 0.58. The average village Gini coefficient for earnings is 0.54 (S.D. ¼ 0.12), with minimum and maximum values (not shown) of 0.32 and 0.84. The figures suggest considerable variation around mean measures of village income inequality.

Results Table 3 contains the regression results in five sections. Section A contains the main results, and Sections B–E contain the results of additional regressions to ensure robustness. Two findings merit discussion about the main findings in Section A. First, we find mixed support for the Absolute Income Hypothesis—the idea that the link between personal nutritional status and personal income resembles an inverted U. Using area deforested to proxy for income and using skinfolds (ZSUMSK2) as a dependent variable (Section A.1), we see that income protects health at a diminishing rate. The coefficient of income is positive and significant and the coefficient for income squared is negative and significant. Further, the two terms are jointly statistically significant (F ¼ 6:70; po0:002). However, when using monetary earnings as a proxy for income (Section A.2), income and skinfolds bear no statistically significant correlation. Instead we find that income measured with earnings correlates with BMI. The link between BMI and income resembles a U. At low levels of income, income correlates with lower BMI, but at higher levels of personal income the relation turns positive. Both income and income squared are each statistically significant, and so are the two terms jointly (F ¼ 2:59; po0:084). In sum, the relation between anthropometric indices of short-run nutritional status and income resembles either a U or an inverted U; results depend on how one defines income and apply only to BMI or to skinfolds, not to mid-arm muscle area. The second noteworthy (negative) finding from Section A is the absence of a statistically significant correlation between Gini measures of income inequality and anthropometric indices of short-run nutritional status at the conventional 90% confidence level or higher. Irrespective of the dependent variable or of the measure of income chosen, income inequality and anthropometric indices of short-run nutritional status did not display a significant statistical relation. In Section B, we vary the definition of household size. Recall that in Section A we divided income measures by the total number of people in the household. Since people of different ages and sexes have different nutrition and consumption requirements, measures of income/person may not capture well true individual

income once we adjust for the demographic composition of the people in the household. In Section B, we reestimate the parameters of the regression using male adult equivalents, with weights from the earlier 18month panel study. The main results hold up. Personal income measured with area deforested/male adult equivalent continues to bear an inverted U-shaped relation with skinfolds and earnings/male adult equivalent continues to bear a U-shaped relation with BMI. Since results could be sensitive to the measure of inequality used (Kawachi & Kennedy, 1997b), we reestimated the regressions of Section A using the standard deviation of the logarithm of income (Section C) and the coefficient of variation of income (Section D) in the village instead of using the village Gini coefficient of income. Again, we found that measures of inequality bore no statistically significant correlation with any of the anthropometric outcomes. As in previous analysis, income measured as deforestation/person bore an inverted U-shaped relation with age- and sex-standardized measures of skinfolds, and earnings/person bore a U-shaped relation with BMI. Estimates of Gini coefficients with small samples might be imprecise, so we re-estimated the regressions of Table 3 including only villages with a sample size of more than 10 subjects (Section E). The coefficients and results of Section E resemble the coefficients and results of the baseline regression (Section A). Since several studies suggest that income inequality might affect the health of sub-groups rather than the health of everyone in a community in the same way (LeClere & Soobader, 2000; McDonough, Duncan, Williams, & House, 1997; Mellor & Milyo, 2001a), we tested for structural heterogeneity across income and sex groups. To assess whether inequality affected people of one sex more than the other, we re-estimated the regressions of Section A (not shown) by adding an interaction term for Ginisex and found that the interaction term was generally statistically insignificant. To assess whether income inequality affected the poor more than the rich, we created a dummy variable by splitting the sample above and below the median of the income distribution, and interacted the Gini coefficient of income with the dummy variable. We re-estimated the regressions of Section A with the additional interaction term (not shown) and found that the interaction term was statistically insignificant. The analysis suggests that the pooled results tell most of the story.

Paths and extensions One shortcoming of the analysis presented is the absence of data on social capital. Recall that income inequality should affect health through social capital, so failure to control for social capital might bias the

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

913

Table 3 Effects of income inequality measured with Gini coefficients on short-run adult anthropometric indices of nutritional status, Tsimane’ Amerindians, Bolivia, 2000 Explanatory variables

Dependent variables BMI

ZSUMSK2

ZMAM

(A) Baseline (1) Income ¼ deforestation/person Gini Income Income2 R2 N Test: income and income2

0.821 (1.360) 0.032 (0.101) 0.005 (0.006) 0.016 476 0.90 (0.413)

0.247 (0.364) 0.062 (0.022)*** 0.004 (0.001)*** 0.061 476 6.70 (0.002)

0.243 (0.464) 0.035 (0.036) 0.001 (0.002) 0.438 472 0.79 (0.458)

(2) Income ¼ earnings/person Gini Income Income2 R2 N Test: income and income2

1.187 (1.240) 0.010 (0.004)** 0.00003 (0.00001)* 0.030 476 2.59 (0.084)

0.416 (0.309) 0.0005 (0.001) 1.29e09 (4.34e06) 0.055 476 0.47 (0.628)

0.060 (0.400) 0.002 (0.001) 6.36e06 (4.13e06) 0.437 472 1.25 (0.293)

(B) Household size measured with male adult equivalents (1) Income ¼ deforestation/male adult equivalents Gini 0.829 (1.353) Income 0.008 (0.071) Income2 0.001 (0.003) 0.015 R2 N 476 Test: income and income2 0.43 (0.649)

0.231 (0.357) 0.033 (0.015)** 0.001 (0.0005)*** 0.058 476 4.76 (0.012)

0.213 (0.467) 0.026 (0.024) 0.001 (0.001) 0.438 472 0.63 (0.536)

(2) Income ¼ Earnings/male adult equivalents Gini 1.278 (1.233) Income 0.007 (0.003)** 0.00001 (7.58e06) Income2 R2 0.028 N 476 Test: income and income2 2.93 (0.061)

0.513 (0.301)* 0.0005 (0.0009) 8.04e07 (2.34e06) 0.057 476 0.35 (0.704)

0.110 (0.403) 0.001 (0.001) 4.46e06 (2.81e06) 0.438 472 1.26 (0.291)

(C) Measure of inequality ¼ standard deviation of the logarithm of income (1) Income ¼ deforestation/person S.D. log of income 0.083 (0.525) Income 0.011 (0.100) 0.004 (0.006) Income2 R2 0.015 N 476 Test: income and income2 0.70 (0.498)

0.126 (0.144) 0.060 (0.021)*** 0.004 (0.001)*** 0.062 476 7.30 (0.001)

0.075 (0.219) 0.038 (0.037) 0.001 (0.002) 0.438 472 0.85 (0.434)

(2) Income ¼ earnings/person S.D. log of income Income Income2 R2 N Test: income and income2

0.072 (0.094) 0.00002 (0.001) 1.42e06 (3.89e06) 0.051 476 0.34 (0.714)

0.052 (0.097) 0.002 (0.001) 6.34e06 (4.02e06) 0.438 472 1.26 (0.291)

0.122 (0.146) 0.062 (0.022)*** 0.004 (0.001)***

0.048 (0.204) 0.038 (0.036) 0.001 (0.002)

0.249 (0.290) 0.012 (0.004)*** 0.00004 (0.00001)** 0.028 476 4.23 (0.019)

(D) Measure of inequality ¼ coefficient of variation (CV) (1) Income ¼ deforestation/person CV 0.166 (0.612) Income 0.022 (0.100) 0.005 (0.006) Income2

ARTICLE IN PRESS 914

R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

Table 3 (continued ) Explanatory variables

Dependent variables BMI

ZSUMSK2

ZMAM

0.015 476 0.86 (0.428)

0.062 476 6.66 (0.002)

0.438 472 0.84 (0.436)

0.206 (0.370) 0.011 (0.004)** 0.00003 (0.00001)** 0.028 476 3.10 (0.052)

0.074 (0.102) 0.0002 (0.001) 7.6e07 (4.17e06) 0.051 476 0.39 (0.675)

0.009 (0.116) 0.002 (0.001) 5.98e06 (4.04e06) 0.437 472 1.16 (0.320)

(E) Limited to villages with at least 10 households (1) Income ¼ deforestation/person Gini 0.661 (2.037) Income 0.029 (0.133) Income2 0.003 (0.008) R2 0.011 N 289 Test: income and income2 1.00 (0.382)

0.155 (0.574) 0.068 (0.033)** 0.004 (0.002)** 0.056 289 2.50 (0.101)

0.062 (0.741) 0.082 (0.042)* 0.003 (0.002) 0.485 287 4.28 (0.024)

(2) Income ¼ earnings/person Gini Income Income2 R2 N Test: income and income2

0.481 (0.543) 0.002 (0.001) 2.47e06 (3.64e06) 0.060 289 6.28 (0.005)

0.527 (0.727) 0.001 (0.001) 5.01e06 (5.07e06) 0.473 287 0.62 (0.544)

R2 N Test: income and income2 (2) Income ¼ earnings/person CV Income Income2 R2 N Test: income and income2

0.0681 (1.873) 0.005 (0.005) 0.00002 (0.00001)* 0.017 289 6.71 (0.004)

Note: Regressions are OLS with robust standard errors (in parenthesis), clustering by village, and constant (not shown). Controls not shown include: age, schooling, sex, and distance. *, **, and *** significant at the 10%, 5%, and 1% level. Under ‘‘test: income and income2’’, we report results of Wald test and, in parenthesis, the probability of exceeding the F-value. ZSUMSK2 ¼ age- and sexstandardized z-scores of sum of triceps and sub-scapular skinfolds and ZMAM ¼ age- and sex-standardized z-scores of mid-arm muscle area, both using Frisancho’s norms (Frisancho, 1990). BMI ¼ body-mass index (kg/m2).

parameter estimate of income inequality. If social capital correlates negatively with income inequality, but positively with good health, then the omission of social capital from the regression will lower the estimated impact of income inequality on health owing to the influence of the indirect negative effect from the omitted variable. We cannot tackle the shortcoming in a direct way with the survey used in this article because we did not measure social capital at the time of the survey, but we can draw on subsequent information to assess whether social capital, in fact, bears a positive correlation with health and a negative correlation with income inequality. If so, then we could read the parameter estimates of Gini variables reported here as lower bounds of true magnitudes. The use of social-capital variable would also allow us to estimate their direct impact on health while controlling for the level of individual income and income inequality.

To explore we topic we turn to a five-quarter panel collected during 2002–2003 from 679 adult (16+) Tsimane’ of 279 households and 13 villages. We did not use the more recent information for this article because it contained few villages. We equate social capital with gift giving and with communal work or with work done for others by a subject. In defining social capital, we focus on behavior rather than on attitudes or on norms because behavior is a more objective and reliable measure of social capital than norms (Glaeser, Laibson, Scheinkman, & Soutter, 2000). We highlight two forms of social capital—gift giving and communal work or labor help—because they represent the main expressions of generosity in small-scale, pre-industrial societies (Godoy, 2001; Gurven, Hill, Kaplan, Hurtado, & Lyles, 2000; Kaplan & Hill, 1985). To collect information on gift giving we asked subjects how often they had given gifts of fish, meat, rice, manioc, plantains, maize, home-brewed beer or

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

chicha (typically made from manioc), cooked food, and medicines to kin or to other Tsimane’ inside or outside the village during the week before the day of the interview. Most of the goods just mentioned are staples of Tsimane’ diet. To elicit information on labor help or on communal labor, we asked subjects how often they had done unpaid work for or communal work with others during the week before the day of the interview. Questions on labor help centered on hunting, fishing, farming, curing, errands done for others, and miscellaneous forms of labor help. Most households gave gifts or offered labor help; the partial correlation coefficient between the two forms of generosity was low (r ¼ 0:145), so we keep the two expressions of social capital as separate explanatory variables. Before presenting the results, we note where the two surveys differ. In the later panel study, we obtained individual measures of monetary income from all adults, not just from household heads. The distance variable was measured by the numbers of hours it took to walk from the town or from the closest road to the village, not with a GPS receiver. Last, the panel study contains repeated observations from the same subject over five consecutive quarters, so we cluster by subjects rather than by villages. We repeated the analysis clustering by village and results did not differ from those we are about to present in Table 4. For computational ease we used the standard deviation of the logarithm of income for each quarter instead of the Gini coefficient, so the results we are about to present should be compared with the results of Section C.2 in Table 3. In bivariate analysis (not shown), we find that social capital expressed through gifts and labor help correlated positively with income inequality measured with the standard deviation of the logarithm of individual income (gifts ¼ 0.037; labor ¼ 0.692). In bivariate analysis, we also find that the three anthropometric indices

915

of short-run nutritional status bore weak and statistically insignificant correlations with social capital. In the only two cases where social capital correlated in a statistically significant way with anthropometric indices, the correlation had the wrong sign. The partial correlation coefficient between gifts and skinfolds was –0.056 (po0:049) and the partial correlation coefficient between labor help and mid-arm muscle areas was –0.211 (po0:001). The results of bivariate analysis therefore suggest that the bias from omitted variable is negative, but not for the reasons one might have hypothesized. Social capital correlated positively (rather than negatively) with income inequality and weakly and negatively (rather than positively) with individual measures of short-run nutritional status. Regression results from Table 4 suggest a positive correlation between income inequality and anthropometric indices of short-run nutritional status after controlling for standard covariates, including social capital. The coefficient for the inequality variables was positive and statistically significant when using BMI (coefficient ¼ +1.053; po0:038) and sum of skinfolds (coefficient ¼ +0.216; po0:072) as dependent variables, but not when using mid-arm muscle area. The results of Table 4 also suggest that the variables measuring social capital had ambiguous and generally statistically insignificant effects. Jointly the two variables for social capital did not correlate significantly with anthropometric indices of short-run nutritional status except when using mid-arm muscle area as a dependent variable (F ¼ 3:61; po0:032). Last, the variables for income and income squared were rarely significant on their own, but jointly they bore an inverted U-shaped relation with BMI (F ¼ 115:26; po0:0001) and a U-shaped relation with the sum of skinfolds (F ¼ 23:18; po0:0001). Income did not correlate in a linear or parabolic fashion with mid-arm muscle area.

Table 4 Effect of income inequality and social capital on short-run adult anthropometric indices of nutritional status, Tsimane’ Amerindians, Bolivia, 2002–2003 (five-quarter panel data) Explanatory variables

S.D. log of income Income Income2 Gifts Help R2 N Test: income and income2 Test: social capital

Dependent variables BMI

ZSUMSK2

ZMAM

1.053 (0.496)** 0.973 (0.441)** 0.015 (0.052) 0.056 (0.076) 0.048 (0.087) 0.029 2442 115.26 (0.0001) 0.40 (0.671)

0.216 (0.118)* 0.050 (0.094) 0.016 (0.011) 0.010 (0.013) 0.029 (0.016)* 0.158 2439 24.18 (0.0001) 1.66 (0.198)

0.178 (0.180) 0.175 (0.122) 0.021 (0.014) 0.016 (0.014) 0.056 (0.026)** 0.397 2432 1.10 (0.339) 3.61 (0.032)

Note: Income inequality ¼ standard deviation of the logarithm of income. For data description see text. Same notes as in Table 3, except as noted. Under ‘‘test: social capital’’ we include the Wald F-test for the joint significance of gifts and help.

ARTICLE IN PRESS 916

R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

In sum, relative to findings of Section C.2 of Table 3, the findings of Table 4 suggest the following tentative conclusions: (a) the omission of social capital will create a downward bias in the estimate of income inequality, (b) income continues to bear an ambiguous relation with anthropometric indices of short-run nutritional status, and (c) the effects of income or income inequality are most visible when using BMI or age- and sex-standardized measures of skinfolds rather than when using ageand sex-standardized measures of mid-arm muscle area as dependent variables.

Discussion and conclusions Why might income inequality bear a weak correlation with anthropometric measures of short-run nutritional status in Table 3? We offer two possible explanations. One reason might relate to the bias from the omission of social capital. Depending on the magnitude and sign of the indirect effect from the omission of social capital, failure to include social capital could introduce a negative bias and weaken the impact of income inequality on health. The bivariate analysis with panel data from 2002 to 2003 suggests that village income inequality during a quarter correlated with more displays of pro-social behavior, and more frequent expressions of pro-social behavior correlated with worse anthropometric indices of short-run nutritional status (skinfolds and mid-arm muscle area). These figures hint at the possibility that the parameter estimates of income inequality in Table 3 might contain negative biases from the omission of variables related to social capital. Although the sign of the indirect effect between income inequality and anthropometric indicators via social capital might be negative, the hypothesize links between social capital, income inequality, and individual health bore the wrong signs. Social capital and income inequality would correlate positively if people help their less fortunate brethren. Then, one should observe more pro-social behavior in villages with greater income inequality. Pro-social behavior expressed through gifts and labor help might correlate with lower indices of nutritional status if pro-social behavior produces a net drain of energy. These are theoretical possibilities that deserve scrutiny in future empirical work. They buttress the point made at the outset about the value of studying pre-industrial rural societies to spot relations that may be harder to detect in more complex industrial economies. A second reason for the weak correlation between income inequality and anthropometric indices of shortrun nutritional status relates to the small sample size of villages used. We used the village as the geographical unit to compute measures of income inequality, and we only had 58 villages in the sample, some with few

observations. We may not have had enough statistical power to detect significant relations. Perhaps income inequality in the community shapes health, but only when inequality covers a larger geographical area. Measures of inequality may have been too imprecise owing to the small sample of observations in some villages. How do our results compare with the results of other studies, and what are the broader implications of our findings? The absence of a systematic correlation between income inequality and health that we found is not unique to small-scale, pre-industrial societies. In recent years, researchers have found that the link between income inequality and individual health in industrial nations is weak or ambiguous, particularly after controlling for intermediate pathways, such as community or regional-level attributes (e.g., social capital, racial, and ethnic heterogeneity) (Deaton, 2001b; Judge & Paterson, 2002; Osler et al., 2002; Daly, Duncan, Kaplan, & Lynch, 1998; Deaton, 2001a, c; Deaton & Paxson, 1999; Deaton & Lubotsky, 2001; Fiscella & Franks, 2000; Mellor & Milyo, 2002a, b). Contrary to findings from industrial nations, we found ambiguous support for the idea that individual income protects own nutritional status. And contrary to our expectations, we found that the link between individual income and own health was often significantly different from zero, undermining our original hypothesis of strong leakages to the rest of the community. We found that when using deforestation as a measure of income, income protected skinfolds at diminishing rates, as predicted by the Absolute Income Hypothesis. But when using earnings as a measure of income, we found that income bore a U-shaped relation to BMI. We conclude with suggestions for future research. In the short run, one of the major challenges for testing the effect of income inequality on individual health in smallscale, pre-industrial societies centers on the definition and measurement of variables, particularly health and income. For instance, a stronger test of the idea that income inequality hurts health would require obtaining reliable and valid measures of perceived health. Income inequality presumably undermines social capital, thereby inducing stress and behaviors reflecting stress, so it would be more appropriate to focus on how people perceive their health—particularly how they perceive stress—than to focus on objective indicators of health (Wagstaff & van Doorslaer, 2000). The problem with using perceived health is that measures of subjective health will likely contain large random measurement errors. For instance, Byron recently did an analysis of how markets affected perceived health among Tsimane’ children and found very few significant results, in part because of measurement errors in assessing subjective health through proxy respondents (Byron, 2003).

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

Another methodological challenge in the short-run centers on finding a valid and reliable measure of individual income. Perhaps neither of the two measures of income we used mirrors income well. Perhaps a more relevant measure of income would center on direct measures of consumption. Weigh days provides the gold standard for measuring consumption, but weigh days are costly since researchers must spend the entire day in the subject’s house. Beyond the hurdles just noted, one of the most serious challenges with studies of income inequality and individual health has to do with finding reliable estimates in the face of biases from unobserved heterogeneity. The literature on income inequality and health suggests enormous difficulties controlling for both the determinants of income inequality and the paths by which income inequality affects health. If the variables responsible for variation in income inequality also affect individual health, then failure to control for them will bias parameter estimates of income inequality. The same applies to path variables, as we have just seen. We conclude with a suggestion for future empirical research: the use of an experimental research design. At least in small-scale rural societies, researchers collecting primary information on health and income could run a lottery with a staple (or money) as a prize for households in the bottom half of the income distribution. Assuming one sets the level of the prize high enough, one could artificially lower income inequality in communities and, in so doing, introduce random variation in community inequality and levels of income. Allowing for sufficient time to elapse between the transfer and the measurement of health outcomes, one could estimate the effect of a one-shot improvement in income and income inequality on social capital and individual health. Much of the literature reviewed earlier would suggest that the proposed experiment, by reducing income inequality, would strengthen social capital and, in so doing, improve individual health. Whether the experiment produces the expected results remains to be seen.

Acknowledgements Research was funded by two grants from the National Science Foundation (SBR-9731240 and SBR-9904318), a grant from the John D. and Catherine T. MacArthur Foundation, and a grant from the Conservation, Food & Health Foundation. We would like to thank the people who participated in data collection: Susan Tanner (University of Michigan), Zoe¨ Foster, Brian Sandstrom, and Anna Yakhedts (Northwestern University), Yorema Gutierrez (Universidad Nacional Mayor de San Andre´s), and Mario Alvarado. We would also like to thank the following institutions and people

917

for their support with fieldwork and logistics: Gran Consejo Tsimane’, Javier Pache, Alonzo Nate, Damia´n Ista, Paulino Pache, Evaristo Tayo, Lorgio Pache, Claudio Guallata, Nestor Canchi, and Manuel Roca. Preliminary results were presented in a seminar sponsored by the Department of Anthropological Sciences and the Center for Latin American Studies, Stanford University. We would like to thank Russ Bernard, Mike Gurven, Ichiro Kawachi, John Komlos, Jeffrey Milyo, and three anonymous reviewers for comments, ideas, and bibliographic leads.

References Byron, E. (2003). Markets and health: the impact of markets on the nutritional status, morbidity, and diet of Tsimane’ Amerindians of lowland Bolivia. Ph.D. dissertation, Department of Anthropology, University of Florida. Coburn, D. (2000). Income inequalities, social cohesion, and the health status of populations: the role of neoliberalism. Social Science & Medicine, 51(1), 139–150. Daly, M. C., Duncan, G. J., Kaplan, G. A., & Lynch, J. W. (1998). Macro-to-micro links in the relation between income inequality and mortality. The Milbank Quarterly, 76(3), 315–339. Deaton, A. (1997). The analysis of household surveys: a microeconometric approach to development policy. Baltimore, MD: Johns Hopkins University Press. Deaton, A. (2001a). Health, inequality, and economic development. Department of economics, Princeton University, Unpublished manuscript. Deaton, A. (2001b). Inequalities in income and inequalities in health. In F. Welch (Ed.), The causes and consequences of increasing inequality (pp. 285–313). Chicago: The University of Chicago Press. Deaton, A. (2001c). Relative deprivation, inequality, and mortality. Department of economics, Princeton University, Unpublished manuscript. Deaton, A., & Lubotsky, D. (2001). Mortality, inequality and race in American cities and states. Department of economics, Princeton University, Unpublished manuscript. Deaton, A., & Paxson, C. H. (1999). Mortality, education, income and inequality among American cohorts. Department of economics, Princeton University, Unpublished manuscript. Duncan, G. J. (1996). Income dynamics and health. International Journal of Health Services, 26(3), 419–444. Egolf, B., Lasker, J., Wolf, S., & Potvin, L. (1992). The Roseto effect: a 50-year comparison of mortality rates. American Journal of Public Health, 82, 1089–1092. Ellis, R. (1996). A taste for movement: an exploration of the social ethics of the Tsimanes of lowland Bolivia. Ph.D. dissertation, St. Andrews University, Scotland. Fiscella, K., & Franks, P. (2000). Individual income, income inequality, health, and mortality: what are the relationships? Health Services Research, 35(1), 307–318. Foster, Z., Byron, E., Reyes-Garcı´ a, V., Huanca, T., Vadez, V., Leonard, W., & Godoy, R. (2003). Physical growth and nutritional status of Tsimane’ Amerindian children of

ARTICLE IN PRESS 918

R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919

lowland Bolivia. American Journal of Physical Anthropology, in press. Frisancho, R. A. (1990). Anthropometric standards for the assessment of growth and nutritional status. Ann Arbor, MI: University of Michigan Press. Glaeser, E., Laibson, D., Scheinkman, J., & Soutter, C. (2000). Measuring trust. Quarterly Journal of Economics, 65(3), 811–846. Godoy, R. A. (2001). Indians, markets, and rain forests: theory, methods, analysis. New York: Columbia University Press. Godoy, R. A., Jacobson, M., De Castro, J., Aliaga, V., Romero, J., & Davis, A. (1998). The role of tenure security and private time preference in neotropical deforestation. Land Economics, 74(2), 162–170. Godoy, R. A., Overman, H., Demmer, J., Apaza, L., Byron, E., Huanca, T., Leonard, W., Pe´rez, E., Reyes-Garcı´ a, V., Vadez, V., Wilkie, D., McSweeney, K., Cubas, A., & Brokaw, N. (2002). Local financial benefits of rain forests: comparative evidence from Amerindian society in Bolivia and Honduras. Ecological Economics, 40, 397–409. Gurven, M. (2004). Economic games among the Amazonian Tsimane: exploring the roles of market access, costs of giving, and cooperation on pro-social game behavior. Experimental Economics, 7, 5–24. Gurven, M., Hill, K., Kaplan, H., Hurtado, A., & Lyles, R. (2000). Food transfers among Hiwi foragers of Venezuela: tests of reciprocity. Human Ecology, 28(2), 171–218. Huanca, T. (1999). Tsimane’ indigenous knowledge, swidden fallow management, and conservation. Ph.D. dissertation, University of Florida, Gainesville. Judge, K., & Paterson, I. (2002). Poverty, income inequality, and health. Glasgow: Department of Public Health, University of Glasgow. Kaplan, H., & Hill, K. (1985). Food sharing among Ache´ foragers: tests of explanatory hypotheses. Current Anthropology, 26(2), 223–246. Kawachi, I. (2000). Income inequality and health. In L. Berkman, & I. Kawachi (Eds.), Social Epidemiology (pp. 76–94). New York: Oxford University press. Kawachi, I., & Kennedy, B. P. (1997a). Socioeconomic determinants of health: health and social cohesion. Why care about income inequality. British Medical Journal, 314, 1037–1040. Kawachi, I., & Kennedy, B. P. (1997b). The relationship of income inequality to mortality: does the choice of indicator matter? Social Science & Medicine, 45, 1121–1127. Kawachi, I., & Kennedy, B. P. (1999). Income inequality and health: pathways and mechanisms. Health Services Research, 34(1), 215–217. Kawachi, I., & Kennedy, B. P. (2002). The health of nations. Why inequality is harmful to your health. New York: The Free Press. Kawachi, I., Kennedy, B. P., Lochner, S., & Prothrow-Stith, D. (1997). Social capital, income inequality and mortality. American Journal of Public Health, 87, 1491–1498. Kawachi, I., Kennedy, B. P., & Wilkinson, R. G. (1999). Introduction. In I. Kawachi, B. P. Kennedy, & R. G. Wilkinson (Eds.), The society and population health reader: income inequality and health (pp. 1–19). New York: The New Press.

LeClere, F. B., & Soobader, M. (2000). The effect of income inequality on the health of selected US demographic groups. American Journal of Public Health, 90(12), 1892–1897. Lohman, T. G., Roche, A. F., & Martorell, R. (1988). Anthropometric standardization reference manual (abridged ed.). Windsor, Ont: Human Kinetics Publishers. Macinko, J., & Starfield, B. (2001). The utility of social capital in research on health determinants. The Milbank Quarterly, 79(3), 387–427. McDonough, P., Duncan, G. J., Williams, D., & House, J. (1997). Income dynamics and adult mortality in the United States, 1972 through 1989. American Journal of Public Health, 87(9), 1476–1483. Mellor, J. M., & Milyo, J. (2001a). Re-examining the evidence of an ecological association between income inequality and health. Journal of Health Politics, Policy, and Law, 26(3), 485–518. Mellor, J. M., & Milyo, J. D. (2001b). Income inequality and health. Journal of Policy Analysis and Management, 20(1), 151–159. Mellor, J. M., & Milyo, J. (2002a). Income inequality and health status in the United States: evidence from the Current Population Survey. Journal of Human Resources, 37(3), 510–539. Mellor, J. M., & Milyo, J. (2002b). Is exposure to income inequality a public health concern? Lagged effects of income inequality on individual and population health. Health Services Research, forthcoming. Osler, M., Prescott, E., Gronbaek, M., Christensen, U., Due, P., & Engholm, G. (2002). Income inequality, individual income, and mortality in Danish adults: analysis of pooled data from two cohort studies. British Medical Journal, 324, 13. Pe´rez, E. (2001). Uso de la ictiofauna entre los Tsimane’. B.A. thesis, Department of Biology, Universidad Mayor de San Andre´s, La Paz, Bolivia. Putnam, R., Leonardi, R., & Nanetti, R. (1993). Making democracy work. Princeton: Princeton University Press. Reyes-Garcı´ a, V. (2001). Indigenous people, ethnobotanical knowledge, and market economy: a case study of the Tsimane’ Amerindians in lowland Bolivia. Ph.D. dissertation, University of Florida, Gainesville. Reyes-Garcı´ a, V., Byron, E., Godoy, R. A., Vadez, V., Pe´rez, E., Leonard, W. R., & Wilkie, D. (2003a). Measuring culture as shared knowledge: do data collection formats matter? Data from Tsimane’ Amerindians, Bolivia. Field Methods, 15(2), 1–22. Reyes-Garcı´ a, V., Godoy, R. A., Vadez, V., Apaza, L., Byron, E., Huanca, T., Leonard, W., & Pe´rez, E. (2003b). Indigenous people share botanical knowledge. Estimates from an Amerindian society in Bolivia. Science, 299, 1707. Vadez, V., Reyes-Garcı´ a, V., Godoy, R.A., Apaza, L., Byron, E., Huanca, T., Leonard, W.R., Wilkie, D., Pe´rez, E. (2004). Does integration to the market threaten agricultural diversity? Panel and cross-sectional evidence from a horticultural-foraging society in the Bolivian Amazon. Human Ecology, in press. Vadez, V., Reyes-Garcı´ a, V., Godoy, R. A., Leonard, W. L., Apaza, L., Byron, E., Huanca, T., Leonard, W., Pe´rez, E., &

ARTICLE IN PRESS R. Godoy et al. / Social Science & Medicine 61 (2005) 907–919 Wilkie, D. S. (2003). Validity of self-reports to measure deforestation: evidence from the Bolivian lowlands. Field Methods, 15(3), 289–304. Wagstaff, A., & van Doorslaer, E. (2000). Income inequality and health: what does the literature tell us? Annual Review of Public Health, 21, 543–567. Wilkinson, R. G. (1992). Income distribution and life expectancy. British Medical Journal, 304, 165–168.

919

Wilkinson, R. G. (1996). Unhealthy societies: the affliction of inequality. London: Routledge. Wilkinson, R. G. (1997a). Comment: income, inequality, and social cohesion. American Journal of Public Health, 87(9), 1504–1506. Wilkinson, R. G. (1997b). Commentary: income inequality summarizes the health burden of individual relative deprivation. British Medical Journal, 314, 1727–1728.