Cover Image

PAPERBACK
$60.00



View/Hide Left Panel

3
Small Families and Large Cohorts: The Impact of the Demographic Transition on Schooling in Brazil

David A. Lam and Letícia Marteleto




The demographic transition that has been observed throughout the developing world is associated with dramatic changes in family size and the size of birth cohorts. During a substantial period of the demographic transition, it is common to observe family size decreasing at the same time that cohort size is increasing. From the standpoint of a child entering school, these changes may imply offsetting effects. Smaller families may mean less competition for resources at the family level, leading to higher school enrollment and better school outcomes. Larger cohorts may mean more competition for resources at the population level, however, leading to more school crowding and worse school outcomes.

This chapter analyzes these issues for the case of Brazil. Brazil provides an interesting case for examining the impact of the changing cohort size and family size on children’s school enrollment. During recent decades the country has experienced a large and rapid fertility decline combined with persistent low levels of schooling and high educational inequality. School enrollment improved little in the 1980s, but began to increase rapidly in the 1990s, a period in which the size of the school-age population began to decline. This chapter explores how changing demographics at the family and population levels may help explain the changing patterns in school enrollment in Brazil in recent decades.

We begin by providing an overview of the demographic transition in Brazil and the resulting changes in cohort size, drawing on census data. We focus in particular on the size and growth rate of the population ages 7 to 14. We then describe the annual Pesquisa Nacional de Amostra de Domicílios (PNAD) household survey data from 1977 to 1999 that will be



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies 3 Small Families and Large Cohorts: The Impact of the Demographic Transition on Schooling in Brazil David A. Lam and Letícia Marteleto The demographic transition that has been observed throughout the developing world is associated with dramatic changes in family size and the size of birth cohorts. During a substantial period of the demographic transition, it is common to observe family size decreasing at the same time that cohort size is increasing. From the standpoint of a child entering school, these changes may imply offsetting effects. Smaller families may mean less competition for resources at the family level, leading to higher school enrollment and better school outcomes. Larger cohorts may mean more competition for resources at the population level, however, leading to more school crowding and worse school outcomes. This chapter analyzes these issues for the case of Brazil. Brazil provides an interesting case for examining the impact of the changing cohort size and family size on children’s school enrollment. During recent decades the country has experienced a large and rapid fertility decline combined with persistent low levels of schooling and high educational inequality. School enrollment improved little in the 1980s, but began to increase rapidly in the 1990s, a period in which the size of the school-age population began to decline. This chapter explores how changing demographics at the family and population levels may help explain the changing patterns in school enrollment in Brazil in recent decades. We begin by providing an overview of the demographic transition in Brazil and the resulting changes in cohort size, drawing on census data. We focus in particular on the size and growth rate of the population ages 7 to 14. We then describe the annual Pesquisa Nacional de Amostra de Domicílios (PNAD) household survey data from 1977 to 1999 that will be

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies used for most of our analysis. Using these data, we document the decreases in family size and increases in parental education that are observed in Brazil from 1977 to 1999. We then describe the evolution of educational measures over the past three decades, pointing out the improved performance of the 1990s. We then analyze the effect of the growth rate of the school-age population, numbers of siblings, and parental schooling on school enrollment using probit regressions. We estimate a negative effect of both cohort growth and family size on school enrollment. These effects are statistically significant, but are relatively small in magnitude. Interactions with age, gender, and father’s schooling indicate that the group most negatively affected by rapid growth of the school-age population is older boys from poorer households. This supports our theoretical predictions that school enrollment pressures will tend to push out students who are closest to the margin of leaving school. We also estimate positive effects of both mother’s and father’s schooling on enrollment, effects that are considerably larger in magnitude than the effects of cohort size and family size. Using the coefficients from our regressions, we simulate the impact of macro- and micro-level demographic change on school enrollment during the late 1970s, 1980s, and 1990s. We find that the growth rates of cohort size of the school-age population tended to reduce school enrollment rates in the 1980s and helped increase enrollment in the 1990s. Decreasing family size and increasing parental schooling both tended to increase enrollment in all periods, with parental schooling having the largest impact. Taking all variables together, the combination of our regression coefficients and the observed changes in independent variables explain more than 60 percent of the observed increase in school enrollment between 1977 and 1999. PREVIOUS RESEARCH ON COHORT SIZE, FAMILY SIZE, AND SCHOOLING This chapter explores the impact on school enrollments of both cohort size and family size. Both of these variables have been the focus of extensive discussion in previous theoretical and empirical research on the impact of rapid population growth in developing countries. Without attempting to thoroughly review this large literature, we briefly focus on some of the studies that provide a background for this chapter. The possible negative effect of rapid growth of the school-age population on schooling outcomes has often been raised as one of the potential negative consequences of rapid population growth (Jones, 1971; Knodel, 1992; World Bank, 1984). There has not been strong empirical evidence of a negative impact of the size of the school-age population on school outcomes, however. In one of the most comprehensive analyses of the issue, Schultz (1987) analyzed the economics of school finance in the presence of changing size of the

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies school-age population relative to the adult population. Using aggregate cross-national data on age structure, school enrollments, and school expenditures, Schultz found no significant effect on school enrollment rates of the proportion of the population in school age. He also found no noticeable effect of relative cohort size on the shares of gross national product (GNP) allocated to education, although he did find a negative association between the proportion of the population of school age and public school expenditures per child. Kelley (2001) notes that several other studies based on cross-country data also suggest that there is no impact of relative cohort size on the share of national budgets allocated to schooling. Kelley (1996) updated Schultz’s analysis using data from the 1980s and continued to find no significant effect of cohort size on the share of educational spending in GNP, although he did not look directly at the impact on enrollment. In the case of Brazil, several studies have mentioned the potential benefits generated by lower population growth rates and decreases in the relative and absolute size of the school-age population. Birdsall and Sabot (1996) point to Brazil’s rapid increase in the number of children of school age in the 1970s and 1980s as potential cause for the poor educational performance of the 1980s. Rigotti (2001) argues that the decline in the population pressure and resulting smaller cohorts of school-age groups may have helped the performance of the educational system. Along the same lines, Castro (1999) has pointed to the high proportions of the population of school age of north and northeast Brazil as one of the potential reasons for lower enrollment rates in these regions. Although past research has recognized the importance of cohort size on children’s schooling in developing countries, this research has typically relied on cross-national regressions using aggregate data. Our analysis will take a different approach, using a combination of time-series and cross-state variation in cohort size, and using household survey data that make it possible to look at the impact of household-level variables as well as aggregate variables. In addition to the literature on cohort size, there is an even larger literature analyzing the impact of family size on schooling outcomes. As pointed out in the reviews by Lloyd (1994) and Kelley (1996), previous research in this area has produced mixed results, ranging from negative effects to statistically insignificant effects to positive effects. Most empirical studies on educational attainment in developing countries have found that children from large families attain less schooling on average than children with fewer siblings (Anh et al., 1998; Knodel and Wongsith, 1991; Marteleto, 2001; Parish and Willis, 1993; Patrinos and Psacharopoulos, 1997). This is often attributed to a dilution of resources, with a smaller share of financial and interpersonal resources allocated to each child in larger families. Some studies, however, have found a positive association between family size and education (Chernichovsky, 1985; Hossain, 1988; King et al., 1986; Mueller, 1984; Zajonc, 1976), a result that Kelley (1996) argues could be theoretically plau-

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies sible if there were large economies of scale in the production of human capital within families. Still other studies have found no statistically significant effect of sibship size on children’s educational outcomes (Mason, 1993; Shavit and Pierce, 1991). As emphasized in the review by King (1987), whatever the relationship between family size and schooling observed in the data, giving a causal interpretation to the association is difficult, because fertility and children’s schooling are choices made jointly by parents. In the case of Brazil, two studies have examined the role of family size on children’s education and have showed an overall negative relationship. Psacharopoulos and Arriagada (1989) found small overall negative impact of number of siblings on school enrollment and attainment, but no effect on school dropout rates. Marteleto (2001) found negative effects of number of siblings on mean years of schooling and school enrollment for cohorts of children born pre- and postdemographic transition. We will include measures of the numbers of siblings in our analysis of schooling outcomes below, and will use the estimated impact of the numbers of siblings to predict how declining family size might be a factor driving the increasing school enrollment rates. The Demographic Transition and Cohort Size Brazil’s demographic transition is fairly typical of those observed throughout the developing world in recent decades. Table 3-1 provides an overview of Brazil’s demographic transition based on census data from 1940 to 2000. As shown in the first row of Table 3-1, the Total Fertility Rate (TFR) for all Brazil was around 6.20 from 1940 to 1960, declining rapidly to 4.35 in 1980, 2.70 in 1991, and 2.31 in 2000. Brazil’s rapid fertility decline occurred during a period of far-reaching social change that included periods of both rapid economic growth and economic crisis (Lam and Duryea, 1999; Martine, 1996; Wood and Carvalho, 1988; Potter, Schmertmann, and Cavenaghi, 2002). The precise reasons for Brazil’s fertility decline are still subject to debate and are outside the scope of this chapter. For the purpose of this chapter, it is important to note the pace of the decline and the substantial regional differences in the timing of the decline. As Table 3-1 shows, the poorer north and northeast regions have consistently had the highest regional fertility rates and began fertility decline somewhat later than the higher income south and southeast regions. In 1970, the southeast’s TFR had fallen to 4.6, while the TFR for the northeast remained at 7.5. In 1991, the regional differences persisted as the southeast showed a TFR of 2.4 and the northeast had a TFR of 4.0. By 2000 the TFR for the southeast had declined to slightly below replacement level at 2.0, with a TFR for the northeast of 2.6. This regional unevenness of demographic indicators mirrors trends and patterns in socioeconomic development. The TFR in the more developed southeast and south regions

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies TABLE 3-1 Total Fertility Rates by Regions, Total Populations, and Annual Growth Rates, Brazil, 1940-2000   Year 1940 1950 1960 1970 1980 1991 2000 TOTAL FERTILITY RATES Brazil 6.16 6.21 6.28 5.76 4.35 2.70 2.31 North 7.17 7.97 8.56 8.15 6.45 4.00 3.05 Northeast 7.15 7.50 7.39 7.53 6.13 4.00 2.60 Southeast 5.69 5.45 6.34 4.56 3.45 2.40 2.00 South 5.65 5.70 5.89 5.42 3.63 2.30 2.25 Center-West 6.36 6.86 6.74 6.42 4.51 2.90 2.34 Total population (millions) 41.23 51.94 70.07 93.14 119.00 146.83 169.54 Annual intercensal growth rate 2.31 2.99 2.85 2.45 1.91 1.60   SOURCES: Instituto Brasileiro de Geografia e Estatística (1996, 2002); Wong (2000). is similar to those of high-income countries, while the higher TFR in the north and northeast regions reflects the lower income, education, and industrialization levels of those regions. Table 3-1 also shows the population size and annual growth rates for the country from 1940 to 2000. Brazil experienced rapid population growth during the second half of the twentieth century, with the annual growth rate peaking at 3 percent in the 1950-1960 period. This growth was driven primarily by falling mortality, especially infant and child mortality. In the 1970-1980 period, the growth rate was still about 2.5 percent per year, but had clearly begun to decline as falling fertility rates began to catch up with falling mortality. The annual growth rate fell to 1.9 percent in the 1980-1991 intercensal period, and to 1.6 percent in the 1991-2000 period. As the second to last row of Table 3-1 shows, Brazil more than quadrupled its total population over this period, from 41 million in 1940 to 169 million in 2000. The dramatic changes in fertility, mortality, and population growth in Brazil are associated with large changes in the size of birth cohorts. These are shown graphically in Figure 3-1, which combines the overlapping single-year age distributions from the censuses of 1970, 1980, 1991, and 2000. The figure shows the size of the birth cohort as reported in two overlapping censuses (when possible), without any adjustment for mortality, using the age distributions from age 0 to age 20 in each census. For example, the two numbers shown for the 1975 cohort are the number of 5-year-olds in the 1980 census and the number of 16-year-olds in the 1991 census.1 Because 1   The census is taken in October of each census year. For simplicity we assume that those reported at age 0 were born in the same year as the census, those age 1 were born in the previous calendar year, and so on.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies FIGURE 3-1 Size of birth cohorts in Brazilian censuses of 1970, 1980, 1991, and 2000. SOURCE: Authors’ tabulations of Brazilian censuses 1970, 1980, 1991, and 2000. our interest is in estimating the size of the school-age population, rather than the actual number of births that originally occurred for each cohort, adjustments for mortality will be relatively unimportant to our calculations. Figure 3-1 shows an increase in cohort size throughout the 1950s, 1960s, and 1970s, peaking around 1982-1983. The rapid increase in cohort size in the 1950s was the result of rapidly falling infant and child mortality. Figure 3-2 makes it clear that the pace of increase varied over time, with much slower growth in the latter half of the 1960s than in the 1970s. These changes in the pace of the increase in cohort size are the result of the complex interaction between falling fertility rates and increasing numbers of women of childbearing age, with the growth of the childbearing population driven by past declines in mortality. The decline in cohort size after the peak in the early 1980s is also uneven, with cohort size actually increasing again during the early 1990s. This is not due to an increase in fertility, which falls rapidly throughout the period, but to an increase in women of childbearing age as an “echo” of the rapid cohort growth of the

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies FIGURE 3-2 Number and growth rate of population ages 7-14, Brazil, 1965-2000 (number relative to 1965 = 100). SOURCE: Authors’ tabulations of Brazilian censuses 1970, 1980, 1991, and 2000. 1970s. The cohort size changes shown in Figure 3-1 imply that the child-dependency ratio was falling before the peak in cohort size occurred. The ratio of 0- to 14-year-olds to 15- to 59-year-olds fell from 0.8 to 0.7 between the 1970 and 1980 censuses, and continued falling to 0.6 in 1991 and 0.5 in 2000. As pointed out by Carvalho and Wong (1995), the demographic “window of opportunity” created by a rising proportion of the population of working age may have a positive impact on many aspects of Brazilian society. The changes in cohort size shown in Figure 3-1 translate into large changes in the absolute size and growth of the school-age population. We use the cohort size numbers in Figure 3-1 to generate the number of children ages 7 to 14 and their growth rate for each calendar year. Figure 3-2 shows the absolute size of the population ages 7-14 in each year, using 1965 as a reference year, along with the annual percentage growth rate for

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies that population. As Figure 3-2 shows, the 7-14 age group grew rapidly in the 1965-1975 period, reaching growth rates exceeding 3 percent per year. The age group grew at a much slower rate in the 1975-1980 period, falling to an annual growth rate of approximately 0.5 percent around 1978. The growth rate increased rapidly again in the 1980s, peaking at a growth rate of approximately 2.5 percent around 1988, followed by a rapid decline in the 1990s. The population ages 7 to 14 actually began to decline in absolute numbers around 1995. These rapid changes in the size of that population during the 1980s and 1990s are the result of a complex combination of the pace of fertility decline and the numbers of women entering childbearing age, reflecting past population dynamics. Figures 3-1 and 3-2 demonstrate that both the relative and the absolute size of the school-age population were declining by the mid-1990s. Brazil’s Household Survey Data Our analysis is based on large household surveys collected by the Instituto Brasileiro de Geografia e Estatística (IBGE), the Brazilian statistical bureau. The PNAD is a nationally representative sample of 80,000 to 100,000 households surveyed collected annually to provide data on employment and earnings. The PNAD contains standard demographic and economic variables such as employment status, occupation, income, and schooling for all members of the household. In this chapter we use the PNAD from 1977 to 1999. There was no survey in 1980 or 1991 because of the censuses conducted in those years, and there was no survey in 1994. The PNAD is appropriate for this study because the repeated cross-sections and comparability of the questionnaires provide us with consistent measures of school enrollment, family composition, and parental characteristics for more than two decades, covering the period in which the school-age population peaked. To analyze the extent to which changes in population size and family size affected children’s school enrollment over the period 1977-1999, the sample will be limited to children ages 7 to 17. This focuses the attention on children at primary and secondary school levels. We will look at the impact of population size using the growth rate of the population ages 7 to 14, estimated using a combination of state population estimates based on census data and single-year age distributions for each state estimated using the PNAD. We take advantage of the rich microdata of the PNAD to generate measures of family size and other characteristics such as age and parents’ schooling measured at the household level. To explore the effect of family characteristics such as mother’s and father’s education, we restrict the sample to children who are classified as children of the household head who have both parents present in the household.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Table 3-2 provides details on our sample selection. The table gives the number of children in the full sample, the number who are children of the household head, and the number who have both parents present at ages 10, 14, and 17. The table demonstrates the large sample sizes we are using. We begin with 187,000 10-year-olds in the pooled PNAD samples. Selecting only those who are children of the household head who have both parents present, this number falls to 149,000, or about 80 percent of the full sample. The size difference between the full and analytical samples increases with age, as more children leave the parental home at older ages. At age 14 and age 17 our analytical sample is 76 percent and 67 percent, respectively, of the full sample. The table also gives two measures that are useful for identifying how our selected sample differs from the sample of all children: mean schooling and the proportion currently enrolled in school. For example, the table shows that 89 percent of all 10-year-olds were enrolled in school, while 90 percent of the 10-year-old children of household head and living with both parents were enrolled. Mean schooling is slightly higher for our selected sample compared to the full sample, as we would expect, but the differences are quite small. The differences between the full and analytical samples increase with age but are never large. TABLE 3-2 Comparison of Selected Sample with Sample of all Children, Brazil Pesquisa Nacional de Amostra de Domicílios, 1977-1999   Full Sample Children of Household Head Children of Head with Both Parents Present Age 10   N 186,943 169,547 149,265 Weighted N 64,310,952 58,464,389 51,717,218 Percentage of full sample 100.0 90.9 80.4 Mean schooling 1.63 1.65 1.66 Enrollment rate (%) 88.9 89.2 89.5 Age 14   N 181,200 160,860 135,570 Weighted N 62,121,671 55,678,826 47,186,042 Percentage of full sample 100.0 89.6 76.0 Mean schooling 4.03 4.09 4.12 Enrollment rate (%) 75.3 76.4 77.0 Age 17   N 170,708 138,206 112,039 Weighted N 57,956,268 47,607,592 38,851,303 Percentage of full sample 100.0 82.1 67.0 Mean schooling 5.42 5.58 5.63 Enrollment rate (%) 50.7 54.5 55.5 SOURCES: PNADs 1977-1999.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Changing Family Characteristics In addition to large changes in cohort size and age structure, the major shifts in the Brazilian demographic and socioeconomic patterns of recent decades have also affected the micro conditions in which children’s schooling takes place by changing family size and levels of parents’ education. Brazilian families have been changing in recent decades in ways that are similar to trends observed in other developing countries. Children are sharing family resources with fewer siblings as fertility declines. In addition, as a result of steady improvements in schooling, children from recent cohorts have better educated parents than their parents had. Table 3-3 shows the mean number of siblings ages 0 to 6 and 7 to 17 in the household for children ages 7 to 17, estimated from the annual PNAD surveys. The average number of siblings ages 0 to 6 declined more than 50 TABLE 3-3 Number of Siblings in Household and Schooling of Parents, Brazilian Children Ages 7-17, 1977-1999 Pesquisa Nacional de Amostra de Domicílios Year Number of Siblings in Household N Ages 0-6 Ages 7-17 Mother’s Schooling Father’s Schooling Urban % 1977 107,105 1.17 2.45 2.51 2.75 60.54 1978 110,447 1.15 2.40 2.53 2.77 61.19 1979 90,035 1.11 2.36 2.65 2.88 61.55 1981 97,705 1.05 2.23 2.78 2.98 64.95 1982 102,134 1.03 2.15 2.82 2.98 65.30 1983 102,097 1.00 2.10 2.94 3.12 65.97 1984 100,944 0.97 2.05 3.10 3.24 66.25 1985 101,010 0.94 1.98 3.28 3.38 67.25 1986 56,007 0.90 1.91 3.48 3.58 67.24 1987 58,316 0.87 1.88 3.59 3.67 67.38 1988 58,465 0.80 1.85 3.77 3.77 68.37 1989 59,171 0.77 1.82 3.96 3.96 68.78 1990 60,333 0.73 1.77 4.08 4.05 68.40 1992 58,964 0.64 1.66 4.31 4.30 73.83 1993 60,450 0.61 1.63 4.45 4.38 73.74 1995 61,420 0.55 1.53 4.73 4.59 74.96 1996 59,395 0.52 1.49 4.93 4.80 75.09 1997 60,398 0.52 1.42 5.02 4.84 74.96 1998 58,291 0.50 1.38 5.23 5.02 74.56 1999 57,693 0.49 1.34 5.12 5.07 74.24 TOTALS 1,520,380 0.80 1.85 3.82 3.85 68.98 1999 minus 1977   −0.68 −1.10 2.61 2.32 13.70 Percentage change   −58.10 −45.18 104.11 84.38 22.63 SOURCE: PNADs 1977-1999. Sample is children of household head with both parents present.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies percent, from 1.2 to 0.5, from 1977 to 1999. The average number of siblings ages 7 to 17 declined 45 percent, from 2.5 to 1.3 over the same period. As shown by Marteleto (2001), the distribution of children across family sizes has changed considerably over the past three decades, with a large reduction in the number of families with more than four children. Table 3-3 also shows that the level of parents’ education has increased substantially throughout the period studied. The mean years of mother’s education nearly doubled over the 22 years covered by our data, from 2.5 years in 1977 to 5.1 years in 1999. Similarly, father’s years of schooling increased more than 80 percent throughout this period. This significant improvement in parents’ education may have contributed to the substantial increase in children’s school enrollment. It may also be that the magnitude of these improvements in educational outcomes is different for specific groups. Table 3-3 also shows the proportion of children in urban areas. In 1977 about 60 percent of children ages 7 to 17 lived in an urban area. By 1999 the urban proportion had increased to 74 percent. In the next section, we show how educational measures for Brazilian children changed during the period of rapid demographic and social changes of the 1970s, 1980s, and 1990s. Children’s Schooling in Brazil Figure 3-3 shows the mean years of schooling for males and females ages 12, 14, and 16 during the past three decades in Brazil. For consistency with FIGURE 3-3 Mean years of schooling by age, males and females, Brazil, 1977-1999. NOTE: Sample is children of household head with both parents present. SOURCE: Authors’ tabulations of PNADs 1977-1999.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Year 1983   0.183 (0.007)a 0.167 (0.007)a Year 1984 0.194 (0.007)a 0.173 (0.007)a Year 1985 0.203 (0.007)a 0.180 (0.007)a Year 1986 0.206 (0.009)a 0.178 (0.009)a Year 1987 0.223 (0.009)a 0.191 (0.009)a Year 1988 0.254 (0.009)a 0.215 (0.009)a Year 1989 0.238 (0.009)a 0.195 (0.009)a Year 1990 0.249 (0.009)a 0.202 (0.009)a Year 1992 0.122 (0.009)a 0.066 (0.009)a Year 1993 0.174 (0.009)a 0.115 (0.009)a Year 1995 0.269 (0.009)a 0.202 (0.009)a Year 1996 0.321 (0.009)a 0.251 (0.009)a Year 1997 0.448 (0.009)a 0.377 (0.010)a Year 1998 0.572 (0.010)a 0.501 (0.010)a Year 1999 0.676 (0.011)a 0.602 (0.011)a Urban 0.406 (0.003)a 0.378 (0.003)a North 0.024 (0.007)a 0.036 (0.007)a Southeast 0.027 (0.004)a −0.010 (0.004)b South −0.095 (0.005)a −0.139 (0.005)a Central-West 0.042 (0.005)a 0.008 (0.005) Constant 0.492 (0.004)a −0.665 (0.007)a −0.419 (0.009)a Sample size 1,413,275 1,413,275 1,413,275 Pseudo R-squared 0.150 0.215 0.220 Log likelihood −617810 −570827 −567072 aSignificance level = 0.01. bSignificance level = 0.05. NOTES: Robust standard errors in parentheses. Omitted categories: Year 1978, Northeast region, age 7. Sample is children of household head ages 7 to 17 with both parents present.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies about 80 percent of 10-year-olds and about 67 percent of 17-year-olds, with most other children classified as “other relative” of the head (including, e.g., stepchildren of the head who are biological children of the mother). We make this sample restriction in order to simplify interpretation of the family variables included in the regression, such as parental schooling and number of siblings. When we use the full sample of age-eligible children, our estimated effects of cohort size growth are quite similar to those reported below. As shown in Table 3-4, we have very large samples in our pooled cross-sections, with more than 1.4 million observations used in each regression.3 Regression 1 looks only at the effect of the school-age population growth rate on enrollment, leaving out family-level variables such as number of siblings and parental schooling. Dummy variables are included for single-year age groups and female. The measure of cohort growth is log[Pt/Pt-1], where Pt is the number of children ages 7 to 14 in year t (in the case of a 2-year interval between surveys, the average growth rate over 2 years is used). We also include interactions of the cohort growth rate with both female and age. To simplify interpretation, the age variable in the interaction is defined as age minus 14, meaning that the main effect for cohort growth is measured for 14-year-olds. Regression 1 shows a statistically significant negative effect of the school-age population growth rate on the school enrollment rate, consistent with our expectations. The interaction with the female dummy is positive, indicating that the enrollment rate of girls is less negatively affected by the growth of the school-age population. As discussed further below, we interpret this as evidence that boys are closer to the margin of dropping out of school, perhaps because of a higher tradeoff between school and work. The negative interaction with age indicates that the impact of cohort growth becomes more negative at higher ages. The fact that the effects of cohort growth on enrollment increase with age suggests that a growing school-age population tends to push students out of school at the top of the school-age distribution rather than deterring enrollment by new school entrants. This is also consistent with a view that the children who are closest to the margin of leaving school are the ones affected by enrollment pressures, with older boys being the group most affected. 3   Because the PNAD data are used to construct the estimates of age-specific population totals for each state, and because we use the change in population as an independent variable, we lose the 1977 data in the regressions. We are thus pooling the PNAD data for the years 1978-1979, 1981-1990, 1992-1993, and 1995-1999. When the PNAD surveys are 2 years apart, the growth rate variable is 0.5 times the 2-year growth rate.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Although Regression 1 shows a statistically significant negative effect of the 7-14 growth rate on enrollment, the magnitude of the effect is relatively modest. For a 14-year-old boy, the estimated effect of changing from a zero growth rate to a 3 percent annual growth rate would be a decline in the predicted probability of enrollment from 76.5 percent to 74.4 percent. It is important to recall, however, that this coefficient may be significantly biased downward in absolute value due to measurement error in the cohort size growth rate. The estimated effect of cohort growth in Regression 1 may be biased due to the correlation of cohort growth with a number of omitted variables. Cohort growth may be correlated with a wide variety of variables, changing across both time and space, that affect school enrollment. One of the most important time-varying covariates is parental schooling, a variable we are also interested in directly. Regression 2 in Table 3-4 adds a number of additional variables to the regression. We include the schooling of the mother and father, the square of these variables, and an interaction between father’s schooling and the cohort growth variable. To control for aggregate effects across time and space, we include dummy variables for every year, an urban dummy, and regional dummies for Brazil’s five major geographic regions.4 The results in Regression 2 indicate that these additional variables reduce the estimated effect of the growth rate of the 7-14 population. The effect continues to be statistically significant and negative, however. We estimate a statistically significant positive interaction with father’s education. We interpret this as supporting our hypothesis that children from better-off households are less affected by the growth rate of the school-age population. The interaction with female continues to be positive, and the interaction with age continues to be negative in Regression 2. All of these effects are consistent with our theoretical framework, with older boys from poorer households being the group closest to the margin of dropping out of school. As in previous research, we estimate large effects of both mother’s and father’s schooling on school enrollment, with a slightly larger effect of mother’s schooling. Effects are largest at low levels of schooling. For a 14-year-old boy in the urban northeast in 1999, an increase in each parent’s schooling from 2 to 3 years increases the probability of enrollment from 88.6 percent to 91.9 percent. Regression 3 adds variables for family size. We use a variable for the number of siblings ages 0 to 6 and the number ages 7 to 17. We also include 4   Results for all variables are very similar if dummy variables are included for every state. The regional dummies are shown here for ease of interpretation.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies interactions between the number of siblings ages 0 to 6 and three variables: the female dummy, age, and father’s schooling. The results indicate that there is a negative effect of the number of siblings ages 0-6 on enrollment. For an urban 14-year-old girl in the northeast in 1999, going from one to two siblings ages 0 to 6 would lower her probability of school enrollment from 95.0 percent to 93.7 percent. There is also a negative, but much smaller, effect of siblings in the 7-17 age group. The interactions with the number of siblings ages 0 to 6 indicate that preschool siblings have a more negative effect on the enrollment of girls than boys. This result suggests that there is a larger child care role for girls than for boys, as suggested in previous research (Marteleto, 2001). The effect is also slightly more negative when the father is better educated. This result is somewhat counterintuitive, but the magnitude of the interaction is very small. There is no significant interaction with age. The coefficient on cohort growth changes little between Regression 2 and Regression 3, indicating that controlling for family size has little impact on the estimated effect of cohort growth. Although we estimate a statistically significant negative effect of cohort growth on school enrollment, the magnitude of the effect is relatively small. For a 16-year-old boy in the urban northeast in 1999 with no parental schooling and one sibling in each age group, the impact of an increase in cohort growth from 0 to 3 percent per year would be a decline in the probability of school enrollment from 59.4 percent to 57.8 percent. The year dummies in Regression 2 and Regression 3 indicate large increases in enrollment over time, even after controlling for parental schooling, cohort growth, and family size. We also estimate significant regional differences in enrollment, some of which may appear surprising. For the full set of independent variables in Regression 3, we estimate that the south and southeast actually have lower enrollment than the northeast, controlling for parental schooling and family size. This is similar to a result observed in Barros and Lam (1996), who found that enrollment rates for 14-year-olds in the northeast and southeast were quite similar, even though children in the poorer northeast region had completed almost two fewer grades of school. Children in the south and southeast are so much farther ahead in school by ages 16 and 17 that their enrollment rates begin to drop below those of other regions, especially when variables such as parental schooling are held constant. All of the regressions in Table 3-4 indicate a statistically significant enrollment advantage for females, a result that is consistent with previous research on schooling in Brazil. For an urban 14-year-old in 1999 with both parents’ schooling at 4 years, the probability of enrollment for a female is about 2 percentage points higher than for a male. This may be another

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies indication that males are more affected than females by a tradeoff between work and school enrollment in Brazil. Counterfactual Simulations of Enrollment Rates To analyze the extent to which the variables in our regressions can explain the trends in school enrollment in Brazil during the 1980s and 1990s, it is useful to simulate enrollment under various counterfactual assumptions. Figure 3-5 takes the coefficients from Regression 3 in Table 3-4 and combines them with the actual values for the independent variables for 1977 to 1999. The coefficients are used to predict the probability of school enrollment for a 16-year-old urban male for every year from 1977 to 1999. Calibrating the enrollment rate to the actual 1977 level for 16-year-old males in urban Brazil, the enrollment rate in each subsequent year is simulated using various counterfactual assumptions. Series 1 in Figure 3-5 simply plots the baseline 1977 enrollment rate of 64 percent as a benchmark for comparison. Series 2 plots the actual school enrollment of 16-year-old boys who are children of the household head for urban Brazil for each year, smoothed as 3-year moving averages.5 As already shown above, the enrollment rate shows little improvement during the 1980s, actually falling to 62 percent around 1986. The actual enrollment rate turns sharply upward around 1989, and rises steadily to nearly 85 percent by 1999. Series 3 in Figure 3-5 simulates the school enrollment rate using the coefficients from Regression 3 and the actual changes in the school-age population growth rate, parental schooling, and family size observed over the period (estimated using the sample of 16-year-old urban males). We omit the regional variables from the simulation, implicitly holding the regional distribution constant. We also omit the year dummies because our goal is to see the extent to which the demographic and family changes can explain the observed changes in enrollment. Series 3, which can be thought of as a baseline for our counterfactual simulations, indicates that the changes in cohort growth, family size, and parental schooling from 1977 to 1999, combined with the coefficients from Regression 3, can explain a significant fraction of the increase in school enrollment over this period, predicting an increase from 64 percent in 1977 to 77 percent in 1999. This is 63 percent of the actual change, suggesting that our variables can explain a high fraction of the observed increase in enrollment. The remaining 37 percent of the increase is presumably due to 5   Although our samples are large, there is still considerable sampling variability when we limit the sample to urban boys age 16. Three-year moving averages are used in Figure 3-5 to reduce the volatility in the series.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies FIGURE 3-5 Actual and simulated school enrollment rates for 16-year-old males, urban Brazil, 1977-1999 (based on Regression 3 in Table 3-4). SOURCE: Authors’ tabulations of PNADs 1977-1999. other variables for which we have not controlled, such as educational expenditures, school policies, and macroeconomic conditions. The most notable difference between our predicted series and the actual series is that our variables predict relatively steady improvement across both decades, in contrast to the actual pattern of relatively constant enrollment rates in the 1980s and sharper increases in the 1990s. The difficult economic situation of the 1980s, which are swept away in our estimation by the single-year dummies, is one possible reason why enrollment grew more slowly in the 1980s than changes in parental schooling, family size, and cohort growth would have predicted. Series 4 shows the effect of holding all other variables constant while allowing the parental schooling variables to vary. This simulation indicates that increasing parental schooling can explain a high fraction of the improvements in children’s school enrollment. Rising parental schooling alone would have predicted a school enrollment rate for urban boys of about 77 percent in 1999, 52 percent of the actual observed increase. As with Series 3, we see that the changes in parental schooling alone would have caused relatively steady increases in school enrollment over the 1980s and 1990s, in contrast to the declining enrollment observed in the 1980s. Series 5 shows the effect of holding all other variables constant while allowing the family size variables (numbers of siblings ages 0 to 6 and 7 to

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies 17) to vary. This series rises steadily throughout the 1980s and 1990s, indicating that falling family size contributed to rising school enrollment. The predicted increases are considerably smaller than the effects of parental schooling, however. Falling numbers of siblings alone would have only increased school enrollment to 66 percent, an increase that is 11 percent of the actual observed increase. Series 6 simulates the effect of changes in the growth rate of the population ages 7 to 14. An important feature of this series is that it is the only simulation that predicts a decline in school enrollment during the 1980s. This is the result of the increase in the growth rate of the population ages 7 to 14 that occurs in the 1980s. Predicted enrollment in Series 6 turns upward in the late 1980s when the cohort growth rate peaks, coinciding closely with the upturn in enrollment observed in the actual data in Series 2. The magnitudes of the changes in enrollment predicted by changes in the school-age population growth rate are quite small, however. Although we are able to predict rising enrollment rates in the 1990s due to the falling rates of cohort growth, the predicted increases in enrollment from this source are only 1 or 2 percentage points. Series 6 shows that we predict a slight fall in enrollment rates after 1997. This is because the growth rate of the school-age population reaches a minimum in 1997, remaining negative but moving toward zero in subsequent years, as shown in Figure 3-2. Although the role of cohort growth appears relatively modest in these counterfactual simulations, we reiterate the point that there may be substantial measurement error in the cohort growth variables, potentially creating substantial downward bias in the estimated effect. The cohort size variables are the only variables estimated at the state level, and our estimates are necessarily based on indirect methods that may have considerable error as a measure of the true cohort size variables that affect the enrollment of particular children. In contrast, our family-level effects are estimated using cross-section variation across many thousands of household records each year. It is also important to remember that by including a full set of year dummies, we have not used any time-series association between cohort growth and school enrollment to estimate the effect of cohort growth. For example, our estimate is not influenced by the fact that the sharp rise in school enrollment rates around 1990 coincides with sharp declines in the growth rate of the school-age population. If we thought the true coefficient on cohort growth was several times larger than our estimated value, the simulations in Figure 3-5 would begin to much more closely track the actual values. It is especially significant that a large effect of cohort growth leads to predicted enrollment that is much flatter in the 1980s, corresponding closely to the actual pattern. Because both family size and parental schooling grow quite steadily over the 1980s and 1990s, cohort growth is the only variable in our regression with the

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies potential to predict the poor performance of the 1980s combined with large improvements in enrollment in the 1990s. We believe future research using more precise measures of cohort growth may produce results that provide stronger evidence that the rapid growth of the school-age population in the 1980s played a role in Brazil’s poor educational performance in that period. On the other hand, the modest impact we estimate for the effect of cohort growth on enrollment simply may be new evidence in support of the finding in previous literature that the size of the school-age population has little effect on school enrollment rates (Kelley, 1996; Schultz, 1987). Our effects of family size, while they show a statistically significant negative association between numbers of siblings and school enrollment, are also relatively modest in size. This is also consistent with the surveys by Lloyd (1994) and Kelley (1996), who conclude that previous literature does not point to large negative effects of family size on schooling. The biggest effects we estimate are those for parental schooling. Our results indicate that increased schooling of both mothers and fathers plays the largest role by far in explaining the rise in school enrollments in Brazil. CONCLUSIONS Brazil’s demographic transition, like that of many other developing countries, produced large changes in cohort size and family size during recent decades. In considering the impact of these changes on schooling outcomes, one of the important features of the demographic transition is that family size and cohort size move in opposite directions during much of the transition. Declining fertility rates compete with population momentum to determine the size of birth cohorts, with the increasing numbers of childbearing-age women outpacing the declining fertility rates for many years of the transition. We show that the size of birth cohorts continued growing in Brazil until around 1982, even though fertility rates and family size had been falling since the 1960s. The school-age population continued growing until the early 1990s, with the growth rate of the school-age population dropping sharply in the 1990s. Cohorts born after 1982 are the first cohorts in Brazil to experience both falling cohort size and falling family size relative to previous cohorts, a fact that may have important implications for school outcomes. The peak of Brazil’s school-age population coincides with the beginning of a period of rapid improvement in school enrollment and school attainment in Brazil, beginning around 1990. Although many factors may have affected these improvements in schooling, this chapter has focused on the role of cohort size and family size. Using Brazil’s large household sur-

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies veys to estimate regressions that control for the growth rate of the 7-14 age population, the number of coresident siblings, and parental schooling, we analyze the impact of these variables on school enrollment from 1977 to 1999. Our results indicate that school enrollment is affected negatively by the growth rate of the population ages 7 to 14, with the most negative effects on older males from poorer households. We interpret this result as evidence that school crowding has the biggest impact on students who are closest to the margin of dropping out of school, with older male students from poor households being most sensitive due to the tradeoff between work and school. Our counterfactual predictions indicate that enrollment rates would have improved faster in the 1980s if cohort growth rates had not increased, and that enrollment rates were positively affected by the declining growth rates of the 1990s. We also estimate significant effects of family size on enrollment, with the number of siblings ages 0 to 6 and 7 to 17 both having a negative effect on enrollment. The effect of siblings ages 0 to 6 is much greater than the effect of siblings ages 7 to 17, and is slightly more negative on girls than boys. Our simulated counterfactuals suggest that declining family size was one of the factors contributing to the rising school enrollment rates of the 1990s. By far the most important explanatory factor in our analysis is parental schooling, with large positive effects of both mother’s and father’s schooling on children’s school enrollment. Our simulated counterfactuals suggest that increases in parental schooling alone can explain a substantial fraction of the increase in school enrollment between 1977 and 1999. Taken in combination, our results imply that changes in the growth rate of the school-age population, number of siblings, and parental education can explain more than 60 percent of the observed increases in school enrollment between 1977 and 1999. REFERENCES Anh, T., Knodel, J., Lam, D., and Friedman, J. (1998). Family size and children’s education in Vietnam. Demography, 35, 57-70. Barros, R., and Lam, D. (1996). Income and education inequality and children’s schooling attainment in Brazil. In N. Birdsall and R. Sabot (Eds.), Opportunity foregone: Education in Brazil (pp. 337-366). Washington, DC: Inter-American Development Bank. Birdsall, N., and Sabot, R. (1996). Opportunity foregone: Education in Brazil. Washington, DC: Inter-American Development Bank. Carvalho, J.A., and Wong, L. (1995). A window of opportunity: Some demographic and socioeconomic implications of the rapid fertility decline in Brazil. (Discussion Paper No. 91.) Pampulha, Brazil: CEDEPLAR, Universidade Federal de Minas Gerais. Castro, M.H.G. (1999). As desigualdades regionais no sistema educacional Brasileiro. Paper presented at the Instituto de Pesquisa Economica Aplicada IPEA seminarry, May, Rio de Janeiro, Brazil.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Chernichovsky, D. (1985). Socioeconomic and demographic aspects of school enrollment and attendance in rural Botswana. Economic Development and Cultural Change, 33, 319-332. Hossain, M. (1988). Credit for alleviation of rural poverty: The Grameen Bank in Bangladesh (IFPRI Research Report No. 65.) Washington, DC: International Food Policy Research Institute. Instituto Brasileiro de Geografia e Estatística. (1980). Anuário estatístico do Brasil. Rio de Janeiro, Brazil: Instituto Brasileiro de Geografia e Estatística. Instituto Brasileiro de Geografia e Estatística. (1996). Anuário estatístico do Brasil. Rio de Janeiro, Brazil: Instituto Brasileiro de Geografia e Estatística. Instituto Brasileiro de Geografia e Estatística. (1998). Anuário estatístico do Brasil. Rio de Janeiro, Brazil: Instituto Brasileiro de Geografia e Estatística. Instituto Brasileiro de Geografia e Estatística. (2002). Anuário estatístico do Brasil. Rio de Janeiro, Brazil: Instituto Brasileiro de Geografia e Estatística. Jones, G. (1971). Effects of population growth on the attainment of educational goals in developing countries. In National Academy of Sciences, Rapid population growth (pp. 315-367). Baltimore, MD: Johns Hopkins University Press. Kelley, A.C. (1996). The consequences of rapid population growth on human resource development: The case of education. In D.A. Ahlburg, A.C. Kelley, and K. Oppenheim Mason (Eds.), The impact of population growth on well-being in developing countries (pp. 67-137). Berlin, Germany: Springer. Kelley, A.C. (2001). The population debate in historical perspective: Revisionism revised. In N. Birdsall, A.C. Kelley, and S.W. Sinding (Eds.), Population matters: Demographic change, economic growth, and poverty in the developing world (pp. 24-54). Oxford, England: Oxford University Press. King, E.M. (1987). The effect of family size on family welfare. In D.G. Johnson and R.D. Lee (Eds.), Population growth and economic development: Issues and evidence (pp. 373-411). Madison: University of Wisconsin Press. King, E.M., Peterson, J.R., Moeriningsih Adioetomo, S., Domingo, L.J., and Syed, S.H. (1986). Changes in the status of women across generations in Asia. Santa Monica, CA: RAND. Knodel, J. (1992). Fertility decline and children’s education in Thailand: Some macro and micro effects. (Policy Research Division Working Paper No. 40.) New York: Population Council. Knodel, J., and Wongsith, M. (1991). Family size and children’s education in Thailand: Evidence from a national sample. Demography, 28, 119-131. Lam, D. (2001). Generating extreme inequality: Schooling, earnings, and intergenerational transmission of human capital in South Africa and Brazil. Paper presented at the annual meeting of the Population Association of America, March, Washington, DC. Lam, D., and Duryea, S. (1999). Effects of schooling on fertility, labor supply, and investments in children, with evidence from Brazil. Journal of Human Resources, 341, 160-192. Lam, D., and Marteleto, L. (2000). Grade repetition, school enrollment, and economic shocks in Brazil. Paper presented at the 2000 meeting of the Population Association of America, March, Los Angeles, CA. Lam, D., and Schoeni, R. (1993). Effects of family background on earnings and returns to schooling: Evidence from Brazil. Journal of Political Economy, 101, 710-740. Lloyd, C.B. (1994). Investing in the next generation: The implications of high fertility at the level of the family. In R. Cassen (Ed.), Population and development: Old debates, new conclusions (pp. 181-202). Washington, DC: Overseas Development Council.

OCR for page 56
The Changing Transitions to Adulthood in Developing Countries: Selected Studies Marteleto, L. (2001). A cohort analysis of children’s schooling in Brazil: Do number and composition of siblings matter? Paper presented at the 2001 annual meeting of the Population Association of America, March, Washington, DC. Martine, G. (1996). Brazil’s fertility decline, 1965-95: A fresh look at key factors. Population and Development Review, 22, 47-75. Mason, A. (1993). Demographic change, household resources, and schooling decisions. In G. Johnson and R.D. Lee (Eds.), Human resources in developing countries along the Asia-Pacific rim (pp. 259-280). Singapore, Thailand: Oxford University Press. Mueller, E. (1984). The value and allocation of time in rural Botswana. Journal of Development Economics, 15, 329-360. Parish, W., and Willis, R. (1993). Daughters, education, and family budgets: Taiwan experience. Journal of Human Resources, 28, 863-898. Patrinos, H., and Psacharopoulos, G. (1997). Family size, schooling, and child labor in Peru: An empirical analysis. Journal of Population Economics, 10, 387-405. Potter, J., Schmertmann, C., and Cavenaghi, S. (2002). Fertility and development: Evidence from Brazil. Demography, 39(4), 739-761. Psacharopoulos, G., and Arriagada, A.M. (1989). The determinants of early age human capital formation: Evidence from Brazil. Economic Development and Cultural Change, 37, 683-708. Rigotti, I. (2001). A transição da escolaridade no Brasil e as desigualdades regionais. Paper presented at the International Union for the Scientific Study of Population conference, August, Salvador, Brazil. Schultz, T.P. (1987). School expenditures and enrollments, 1960-80: The effects of income, prices, and population growth. In D.G. Johnson and R.L. Madison (Eds.), Population growth and economic development: Issues and evidence (pp. 413-476). Madison: University of Wisconsin Press. Shavit, Y., and Pierce, J.L. (1991). Sibship size and educational attainment in nuclear and extended families: Arabs and Jews in Israel. American Sociological Review, 56, 321-330. Wong, L. (2000). A projeção da fecundidade: Um exercício aplicado ao Brasil para o período 1991-2000. In Annals of the XII Meeting of the Brazilian Association of Population Studies. Caxambu, Brazil: Associação Brasileira de Estudos Populacionais. Wood, C., and Carvalho, J.A.M. (1988). The demography of inequality in Brazil. Cambridge, England: Cambridge University Press. World Bank. (1984). World development: Report on population change and economic development. New York: Oxford University Press. Zajonc, R.B. (1976). Family configuration and intelligence. Science, 192, 227-236.