Comparisons of Intergenerational Schooling Mobility ...and Székely basically define intergenerational mobility to be the extent to which school gaps of children who co-reside with

Comparisons of Intergenerational SchoolingMobility Estimates Based on Household Surveys

from Latin America and the Caribbean

by

Jere R. Behrman and Miguel Székely1

November 9, 2000

(Preliminary version. Please do not quote.)

1 Behrman is the William R. Kenan, Jr. Professor of Economics at the University of Pennsylvania([email protected]). Székely is Research Economist in the Research Department, Inter-American Development Bank ([email protected]). Behrman collaborated in this paper as a consultant tothe Carnegie Endowment for International Peace. We thank Alejandro Gaviria and Nancy Birdsall forsuggestions and help on this paper. Only Behrman and Székely, and not Carnegie nor IDB, are responsiblefor the content of this paper.

1

Introduction

There is a growing consensus that income inequality has increased in a large number of

countries in the world in recent years.2 The main concern is that, precisely because of the

high inequalities, absolute poverty rates are much greater and that economic growth has

not been immediately translated into poverty reduction.

It is important to know, however, the extent to which inequality is driven by individual

differences in ability, work ethic, or simply good luck, rather than by differences in

opportunities. Thus, if some individuals prefer to work more hours or to invest more

energy in their work than others, the income inequality that will occur as a result of these

differences would not necessarily be a policy issue. In fact, reducing this type of

inequality through policy interventions could well lead to reductions rather than increases

in welfare.

But this type of “efficient” inequality probably does not explain the extent of inequalities

in most countries in the world. Rather, inequality likely originates substantially from the

absence of opportunities for large segments of the population. The outright (or implicit)

exclusion of some groups on the basis of their gender, ethnic origin, place of residence or

social status can in turn explain inequality of opportunity. Two different societies with

the same “snapshots” of income distribution at a point of time may have different levels

of social welfare if in one of them these characteristics are a more important determinant

of individual’s fates in life.

Since income inequality measurements in cross-sectional data are “snapshots” each at a

point of time which can change due to many factors, they do not necessarily inform on

whether welfare in society is changing, for instance, due to reduced equality of

opportunity. Friedman (1962) argues that a given extent of income inequality in a rigid

system in which each family stays in the same position in each period may be more a

cause for concern than the same degree of income inequality due to great mobility and

dynamic change associated with equality of opportunity. Thus, if the objective is to

2

determine whether welfare levels are improving, it is essential to characterize the degree

of social mobility both across generations and within generations.

For such reasons there have been increasing recent efforts to measure actual or perceived

social mobility and changes in such mobility in the region. Several of these are

summarized in Birdsall and Graham (2000). But such efforts are limited by the limited

availability of longitudinal data with which to follow the same individuals over time.

Therefore some of the recent studies have based their estimates on cross-sectional data,

with focus on intergenerational mobility in schooling – in particular, Behrman, Birdsall

and Székely (2000) and Dahan and Gaviria (2000) both have used multiple household

surveys to assess the extent of intergenerational schooling mobility. Behrman, Birdsall

and Székely basically define intergenerational mobility to be the extent to which school

gaps of children who co-reside with their parents are not associated with parents’

characteristics, primarily schooling. Dahan and Gaviria characterize mobility to be

greater when the between family variance in child schooling success is small relative to

the total variance in schooling success using data on siblings who are co-residing with

their parents (where “success” is defined as not lagging more than a grade below the

median for each age level).

This paper compares and contrasts empirical estimates of intergenerational schooling

mobility based on cross-sections of household surveys from LAC. The paper focuses on

the Behrman, Birdsall and Székely and Dahan and Gaviria approaches, because to our

knowledge these are the most recent attempts to characterize inter-generational mobility

with cross sectional data consistently for a number of countries. To perform the analysis

we use the micro data from 123 household surveys. Of these, 87 belong to 18 Latin

American countries, for the period 1977-1999. These countries cover around 95% of the

total population of the region. We use 19 Current Population Surveys (for 1980-1998) for

the United States, and 17 Family Income and Expenditure Surveys for Taiwan, for 1980-

1996.

2 See for instance Székely and Hilgert (1999) for a recent comparison for the 1990s decade for 35 countries.

3

Our analysis illustrates that even when we restrict the measurement of mobility to only

few options, the way in which inter-generational mobility is measured has strong

implications for three important issues. The first is country rankings. Economic policies

and the performance of countries are commonly evaluated in terms of social indicators,

and countries where inter-generational mobility is relatively high are thought of as being

more successful. Our results show that if countries are ranked according to different

mobility measures, a very different idea is obtained about their performance relative to

others.

The second issue is the evolution of inter-generational mobility over time. We compare

the trends followed by three measures over 20 years for Latin America and the Caribbean

(LAC), the United States and Taiwan, and conclude that depending on the measure

chosen to estimate mobility, a different picture emerges. This has important implications

because depending on the measure chosen, the policy conclusion may shift from giving

priority to compensatory measures for sectors of the population that have been “losers”

during the past years, to giving priority to policies aimed at leveling the “play field”.

The third issue is the correlation between mobility and economic reform. One of the main

attractive features of reforms is that they are supposed to improve market functioning,

and this would be expected to act in the direction of leveling the “play field”. Our

preliminary results on this show that the correlation between reforms and mobility also

depends on the specific methodology selected to measure inter-generational mobility.

The paper has three sections. The first summarizes some conceptual issues regarding

measuring social mobility in the literature. The second presents empirical estimates for

some of these alternatives. The third concludes.

4

Section 1. How is Social Mobility Statistically Modeled and Measured?

How social mobility has been modeled and measured has varied, not surprisingly,

depending on aspects of social mobility that are discussed in Behrman (2000), as well as

the available data. We review some of the more prominent approaches to statistically

modeling social mobility in this section, focusing on integenerational schooling mobility

as an example. This is because schooling is normally thought to be an important factor in

income distribution and this is the type of mobility that perhaps can be most easily

characterized with cross-sectional household data because such data permit the

representation of schooling for two generations that are co-residing.

Section 1.1 Intergenerational Correlations or Markov Statistical Models of Social

Mobility for Continuous Socioeconomic Indicators

A common statistical characterization of mobility dating back at least to Galton’s (1889)

model of regression towards the mean is the first-order Markov model in which the

relevant socioeconomic indicator -- say schooling -- for entity i in period t (Sit) depends

on the value of that indicator in the previous period (Sit-1) and a stochastic term (wit) that

is independent of the previous period indicator and that is independently distributed

across individuals and across periods:3

(1) Sit = bSit-1 + wit.

This approach is basically what Behrman, Birdsall and Székely (2000) use. In that

context, each period can be a generation and i refers to the family dynasty. Thus the

previous period indicator carries all relevant past information about family i, including

the past experience regarding transitory (in the sense of being for only one generation)

shocks (which means that the variance of S increases with t). The parameter b is positive

3 With data on three (or more) periods the first-order Markov assumption can be tested by its implicationthat the correlation in Y between periods t and t + 2 equals the product of the correlations between t and t +1 and between t + 1 and t + 2.

5

and is greater than one if there is real growth in S. If Sit is defined relative to the mean of

its distribution, then the parameter b affects the relative position in the distribution and b

< 1 implies regression towards the mean (that is more rapid the smaller is b). The

parameter b is a measure of immobility. Another frequently utilized measure of mobility

is r, the inter-period correlation in S. As t goes to the limit, b and r coincide, but along

the path b consistently exceeds r (Atkinson, Bourguignon and Morrisson 1992, p. 9).

Estimates of relation (1) or of extensions of this relation (e.g., with population groups

with different mobility) may be used to characterize intergenerational (or

intragenerational) social mobility with continuous socioeconomic indicators such as

schooling (or income or earnings) measured in either absolute or relative terms.

Section 1.2 Inter- Versus Intra-Familial Variations

In this approach, which is basically what Dahan and Gaviria (2000) use (and which dates

back at least to Conlisk 1974), the variance in schooling success is decomposed between

within and between family components, and mobility is characterized as being greater the

greater is the intrafamily share of the variance. That is, if Sjic is the schooling success of

the jth child in the ith family, the total variance in schooling success across children

(perhaps conditioned on some age range) can be composed into the within family

variance Sj2 and the between family variance Si

2:

(2) Sij2 = Sj

2 + Si2, which implies:

(2A)1 = Sj2/ Sij

2 + Si2/ Sij

2.

The larger is the share of the interfamily variance in the total variance, the greater is

immobility.

6

Section 1.3 Markov Transition Matrix-Based Measures of Intergenerational

Mobility

Schooling can be represented by categorical variables. A convenient and standard way to

characterize intergenerational mobility with categorical variables is to use transition

probability matrices for movements among segments of the distribution (e.g., relevant

categories, terciles, deciles) between generations. These generally are used in first-order

Markov processes with the assumption, as in relation (1) above, that the previous period

indicator carries all relevant past information about the family dynasty. In certain

respects transition matrices allow greater flexibility in characterizing mobility than do the

approaches based on continuous variables that are discussed in the previous sections (at

least as they usually are used) because they allow asymmetries and other non-linearities.

For example, transition matrices easily may capture a situation in which the probabilities

of moving in a large jump from the bottom of the schooling distribution to the top may be

larger than the probability of moving from the top to the bottom, with the difference

balanced out by differences in the probability of moving to the middle.

A transition probability matrix (P) is an n x n matrix, where n refers to the number of

categories. The element in the jth row and kth column of a transition probability matrix

(pjk) gives the probability that an entity moves from the jth category to the kth category

between generations (periods). The sum across elements in each row must be one

because every family that initially is in the jth category must end up in one of the

categories ( kpjk = 1 for each j), assuming that all family lines continue to the next

generation.

In general the sum of elements in each column need not be one. If the categories have

equal numbers in them and there is relative or exchange mobility (see Behrman 2000) so

that distribution does not change between generations, the sum of the elements in each

column is one.4 Following are examples of such matrices for population terciles (so there

4 As noted above, the latter term is frequently used by sociologists concerned with social mobility incontrast to “structural” mobility if the distribution is changed. If the sum of the elements in each of therows and of the elements in each of the columns is one, the matrix is said to be “bi-stochastic.”

7

are three segments) for two special cases of interest: no intergenerational mobility (PN)

and “complete” intergenerational mobility (PC):

1.00 0.00 0.00 0.33 0.33 0.33

(3) PN = 0.00 1.00 0.00 ; PC = 0.33 0.33 0.33

0.00 0.00 1.00 0.33 0.33 0.33

For PN, there is no relative intergenerational mobility, so there are no nonzero off-

diagonal elements. Children end up in exactly the same part of the distribution as their

parents – the completely rigid social system noted in the introduction to which Friedman

(1962) refers. For PC, there is “complete” intergenerational mobility in the sense that the

probabilities are equal of ending up in any of the three terciles after the transition,

independent of the initial starting point.

One important strain in the literature is concerned with how to infer the extent of

intergenerational (or other types of social) mobility from transition probability matrices

of the types indicated above. In essence, the problem is how to reduce such a probability

matrix to a scalar that characterizes the extent of mobility (immobility). A number of

possibilities have been proposed in the literature and are summarized by Dardanoni

(1993), to which the interested reader is referred for further discussion and references:

1. Trace [trace (P) - 1)/(n - 1)]: The intuition behind this measure is that the

greater the concentration on the diagonal the less the mobility. The obvious

limitation is that the trace only distinguishes between being on and off the

diagonal, not whether the latter are close or far from the diagonal.

2. Determinant [|P|1/(n-1)]: The determinant incorporates information about off-

diagonal elements as well as the diagonal elements. But the determinant is zero

(implying complete mobility) if any two rows or any two columns are identical no

matter what is the distribution of elements elsewhere in the matrix, which is a

definite limitation as a mobility indicator.

8

3. Mean First Passage Time: If two individuals are drawn from the population at

random and there is a steady-state Markov chain of transitions, the mean first

passage time is the expected number of periods which must pass before the first

individual achieves the state of the second individual (Conlisk 1990).

4. Bartholomew’s (1982) Measure [ j k jk|k-j|]:5 The expected number of

category boundaries crossed from one generation to the next when the chain is in

steady state.

5. Second Largest Eigenvalue: This index has been proposed to represent the

speed of escape from the initial conditions and of regression to the mean.

Unfortunately, as Dardanoni (1993) emphasizes, these mobility indicators do not

consistently rank different transition matrices. He illustrates with the following three

transition matrices:

0.60 0.35 0.05 0.60 0.30 0.10 0.60 0.40 0.00

(5)P1 = 0.35 0.40 0.25 ;P2 = 0.30 0.50 0.20 ; P3 = 0.30 0.40 0.30

0.05 0.25 0.70 0.10 0.20 0.70 0.10 0.20 0.70

He shows that any of these three matrices may be considered to represent the greatest

mobility, depending on which mobility index is used:

Dardanoni therefore derives a partial Social Welfare Function (SWF) ordering with an

additive SWF placing greater weights on individuals who start in lower positions in

society. Conditional on this SWF, he derives a condition for the partial ordering of

alternative transition matrices.6 He also shows that most of the mobility indices

5 is the equilibrium probability vector, which exists and is the unique solution to ’ = ’P if P is regular (i.e.,for some large enough integer m, Pm is strictly positive).

6 He shows that welfare is greater for transition matrix P than for Q (if both are monotone regular transitionmatrices for SWF weights and instantaneous utilities that are nondecreasing in income) if T’ [P( )-Q( )]T 0

9

mentioned above (with the exception of Bartholomew’s measure and a modified

eigenvalue measure) do not necessarily imply greater mobility for transition matrices that

give families starting with lower initial schooling better schooling lotteries.

Such considerations point to the facts that currently there is no one “correct” way to

measure relative mobility with transition matrices, different approaches may yield

different rankings for the same transition matrices, and to make much progress in such

cases may require explicit assumptions about the SWF – but even with such assumptions

complete orderings of transition matrices may not be possible.

Partly for this reason, Fields and Ok (1996) contribute an axiomatic approach to

characterizing mobility from longitudinal data. Based on seven axioms,7 they derive a

measure of total mobility that is additively decomposable into mobility from (i) the

transfer of resources among individuals with total resources held constant (relative or

exchange social mobility) and (ii) a change in the total resources available.

In particular, they define the relative mobility due to resource transfers (MR) in a growing

economy to be twice the amount lost by losers (twice because every unit lost by a loser is

gained by a winner):

(4A) MR = 2 ( (yjt - yit+1)), where j is summed over all losers.

where T is a n x n summation matrix with zeros below the main diagonal and ones elsewhere and 0 < 1 isthe discount factor.

7 The axioms are (i) linear homogeneity (i.e, an equiproportional change in all income levels both in theinitial and the final distributions results in exactly the same percentage change in the mobility measure), (ii)translation invariance (i.e., if the same amount is added to everybody’s income in the initial and the finaldistributions, the new situation has the same mobility as the initial one), (iii) normalization (i.e., a onedollar income gain and a one dollar income loss both produce one unit of mobility for that individual), (iv)strong decomposibility (i.e., the level of total income mobility experienced by a population is a function ofthe levels of mobility experienced by two disjoint and exhaustive subpopulations) (they also consider weakdecomposibility), (v) population consistency (i.e., in the contexts of populations of different sizes, if equalsare added to equals the results are equal), (vi) growth sensitivity (i.e., if unequals are added to equals theresults are unequal), and (vi) individualistic contribution (i.e., the contribution of one individual’s incomechange to total mobility depends only on the amount of his/her income change).

10

Relative mobility due to transfers in a shrinking economy is defined to be twice the

amount gained by winners:

(4B) MR = 2 ( (yjt+1 - yit)), where j is summed over all winners.

Mobility due to resource growth (MY) in a growing economy is defined to be:

(4C) MY = yjt+1 - yit, where j is summed over all members of the population.

Mobility due to resource contraction in a shrinking economy is defined to be:

(4D) MY = yjt - yit+1, where j is summed over all members of the population.

Total mobility (M) is defined to be:

(4F) M = |yjt+1 - yit|, where j is summed over all members of the population.

They show that with these definitions of mobility:

(4G) M = MR + MY.

This approach has several advantages over the previous literature. Total mobility is

decomposed into relative (exchange) mobility and resource change mobility as in relation

(4G) -- which is the first exact decomposition of resource mobility in the literature. This

approach can be applied in per capita terms to allow for comparisons across periods (or

surveys) with different numbers of people, which also permits comparisons with base

resources so that statements can be made such as “average mobility is 10% of initial

resources.” This approach does not depend on Markovian assumptions that have been

rejected by empirical studies of Britain, France and the United States (Atkinson,

Bourguignon and Morrisson 1992, Atoda and Tachibanaki 1991, Shorrocks 1976). It

also does not depend on normative assumptions regarding social welfare functions.

11

While their formulation is stated in terms of income, the resources of relevance could be

total schooling or any other resources.

Section 2. Alternative Estimates of Intergenerational Schooling Mobility in Latin

America and the Caribbean Based on Cross-Sectional Household Surveys

Section 2.1 Data

For our exploration of alternative intergenerational schooling mobility estimates for LAC

we use a compilation of household surveys for Latin America, as well as a series of the

Current Population Survey for the United States, and a series of household surveys for

Taiwan. We believe that the comparison between this particular region and countries is

illustrative about the differences across levels of development. Since we have access to

the micro data in these surveys, we are able to construct comparable indicators on the

critical information for this paper, which are the measures of schooling attainment for

multiple co-resident family members – parents and children for the Markovian methods

in Sections 1.1 and 1.3, and siblings for the variance decomposition in Section 1.2. Some

details of the data can be found in Appendix Table A1.

From each survey we use only a restricted sample of children ages 13 to 21. This age-

group is selected so that individuals are still co-resident with their parents, however, their

schooling attainment is censored in many cases because they still are in schooling at the

time of the survey. For these reasons, studies such as Behrman, Birdsall and Székely

(2000) and Dahan and Gaviria (2000) focus on children who are young enough so that

they are still co-resident with their parents but old enough so that many of these children

will have completed their schooling – typically their late teens. In addition, these studies

use measures for the children’s schooling that indicate to what extent children have

relatively low schooling conditional on their age, even if some may subsequently

continue in school after the surveys. In particular, Behrman, Birdsall and Székely use the

schooling gap, defined as the difference between the number of schooling grades that

would have been completed if a child entered school at the earliest legal age (e.g., six)

12

and progressed satisfactorily one grade each subsequent year minus the number of actual

completed grades. This indicator indicates various possible degrees to which children of

a given age lag behind what would be possible in their schooling attainment due to late

entry, grade repetition or dropping out. Dahan and Gaviria use an dichotomous index for

lagging behind in schooling that distinguishes between children of a given age who have

completed the median number of schooling grades for that age versus those who have at

most one grade less than the median number of completed grades for that age. This

indicator indicates whether or not children are below some cutoff in their schooling

attainment to date. The latter index is analogous to the headcount poverty index with the

poverty line at the median schooling, with the former index is analogous to a poverty

index that indicates how far one is below the poverty line with the poverty line at the

schooling level that would have been attained with entry on time and successful

promotion each year. Both of these indices, or variants of them (including one analogous

to the distribution-weighted poverty index) can be constructed and used with any of the

measures of intergenerational schooling mobility that are discussed in Section 1.

Although the data used to compute these mobility measures is of high quality, relative for

instance to income data from household surveys, which varies considerably in terms of

coverage, definitions, and quality, it is not without its problems. For instance, household

surveys not always inform on whether a child residing at a household is effectively a

child of the head. In the cases where this can be verified, the proportions are surprisingly,

not that high. For instance, Table 1 shows the proportion of children that are children of

the head. In countries such as Venezuela, the proportion is only around 78 percent, and

only 76 percent of all children ages 13-19 reside in a household where the head is the

parent, and where the parent has a spouse who is also a parent of the same child. Due to

the age restriction, sample sizes for computing the mobility measures examined here are

relatively small, so imposing an additional constraint of kinship between head and

children stretches the data in many cases.

13

Section 2.2 Alternative Estimates of Intergenerational Schooling Mobility

Behrman, Birdsall and Székely (BBS henceforth) consider schooling gaps separately for

four age groups: 10-12, 13-15, 16-18, and 19-21 years old because family background

may have differential effects depending on how close a child is to marginal schooling

decisions. Additionally, they consider separately five parental schooling quintiles. These

quintiles are defined by parental schooling because parental schooling represents not only

important components of permanent household income, but also possibly important non

income characteristics such as genetic endowments and preferences regarding schooling

as well as parental price-of-time considerations. For each subgroup of the population, two

indexes are computed. The first (which we label BBS1) consists on regressing schooling

gaps of children on parental schooling, household income and other household

characteristics. From the estimated coefficients of the three family background variables

two basic intergenerational schooling mobility indices are constructed: (1) the

Aproportional intergenerational schooling mobility index@ defined as the proportion of the

variance in the schooling gap that is associated with the weighted average of parental

schooling and household income, where the weights are the regression coefficient

estimates for these three variables (labeled BBS1). (2) the Agap-adjusted intergenerational

schooling mobility index@ defined as the proportion of the variance in the schooling gap

that is associated with the weighted average of parental schooling and household income

multiplied by the average gap relative to the expected schooling for that subsample)

(labeled BBS2).

The measure by Dahan and Gaviria (DG henceforth), is defined in two steps. The

first uses the following correlation:

∑∑

∑ ∑∑

= =

= ==

−

−−=

F

f

S

ssf

F

f

S

kfkf

S

ssf

gf

ff

gg

Sgggg

1 1

2

1 1

2

1

2

)(

/)()(

ρ ,

where F is the number of families in the sample, Sf is the number of teenage

siblings in family f, gsf is the binary indicator of socioeconomic failure of individual s in

14

family f, and g is the average indicator in the entire sample. Since ρg could yield positive

values even if family background is inconsequential -as will be the case, for example,

when children are assigned to families randomly, the authors define an alternative index

as follows:

FS

Sga −

−−−=

1)1(1 ρρ , (2)

where S is the number of children in the sample. The new index (ρa), which corresponds

now to the adjusted R2 obtained by regressing earnings on family dummies, will yield

positive values only if the previous index (ρg) is greater than would be expected purely

by chance. Positive values of ρa can thus be unambiguously interpreted as evidence that

family background does play a role in the determination of schooling outcomes.

The BBS1, BBS2, and DG indexes just described, refer to immobility, rather than

mobility indexes. For ease of interpretation, in what follows we slightly transform these

indexes by defining: Index=1-original index.

Figures 1a, 1b and 1c plot the transformed BBS1, BBS2, and DG mobility indexes,

respectively, and correlates them with the Gini coefficient for the distribution of

household per capita incomes. In all three cases the correlation is quite strong (.44, .55

and .56, respectively), but the figures clearly illustrate that inequality and

intergenerational mobility are not the same. Countries with high mobility in general tend

to have lower inequality, but this is not always the case.

Sensitivity of Country Rankings

Figures 2a, 2b and 2c plot the correlation between the three indexes. As expected, BBS1

and BBS2 in panle 2c, have a stronger correlation (of .87). The correlation between

BBS1 and DG is of .35, while the correlation between BBS2 and DG is of .49. But even

though the correlation is significant in statistical terms, the use of different indexes may

lead to different conclusions about the extent of social mobility in one country compared

to others. Table 3 illustrates this. In the first column, countries are ranked according to

BBS1. The country with the highest mobility appears to be the United States, and the one

15

with the lowest is Paraguay. The second column shows the ranking according to BBS2.

There are several reversals. Peru, Uruguay and Argentina now appear to have higher

mobility, while Guatemala has lower mobility. The number of re-rankings is much larger

when the comparison is made between BBS2 and DG. Mexico changes from being a

country with moderate mobility to being one of the countries with lower mobility. Similar

changes are observed for Argentina and Honduras. Paraguay and Peru, on the other hand,

appear to have much more mobility in terms of DG than in terms of BBS2. The United

States and Taiwan are the two countries that consistently appear as having high mobility,

independently of the measure chosen.

Sensitivity of Time Trends

Figures 3a, 3b and 3c show the evolution of intergenerational mobility for the BBS1,

BBS2 and DG indexes, for the average LAC country, the United States and Taiwan. The

United States seems to be the only case where the trend is very similar, independently of

the measure chosen to measure mobility. However, BBS1 and BBS2 show a continuous

increase in mobility for LAC, while DG shows a decline after around 1990. For Taiwan,

the choice of index also makes an important difference. According to BBS2 it has been

increasing since the mid 1980s. According to BBS1 it declined sharply towards 1985 and

has been erratic thereafter, while there has been a considerable increase since 1980

according to the DG measure.

Figures 4a and 4b plot the three measures for Costa Rica and Peru. Of the sample of 20

countries we consider, these are the two cases where the conclusion about the evolution

intergenerational mobility is most sensitive to the choice of measure. In Costa Rica,

BBS2 shows an increase, while DG declines sharply. In Pery BBS2 declines, and DG

raises in the 1990s.

Sensitivity of Correlation with Economic Reform

Finally, it is also of interest to determine whether the choice of measure leads to different

conclusions about the correlation between mobility and economic reform. To explore

this, we regress mobility indexes for LAC on several indexes of economic reform. The

16

analysis is restricted to the LAC data because reform indexes are only available for this

region.

To characterize the pace and depth of different types of reforms, we use reform indices

developed by Lora (1997) and modified and extended by Morley, et al. (1999). These

indices summarize information on trade reform, financial liberalization, tax reform,

liberalization of external capital transactions, and privatization for the period 1970-1995.

Unlike proxies commonly used in the literature, these reform indices have the advantage

that they are based on direct indicators of governmental policies, so that they reflect

policy “effort”. The trade reform index is the average of the average level of tariffs and

the average dispersion of tariffs. The index of domestic financial reform is the average of

an index that controls for borrowing rates at banks, an index of lending rates at banks,

and an index of the reserves to deposit ratio. The index for international financial

liberalization averages four components: sectoral controls of foreign investment, limits

on profits and interest repatriation, controls on external credits by national borrowers and

capital outflows. The tax reform index averages four components: the maximum marginal

tax rate on corporate incomes, the maximum marginal tax rate on personal incomes, the

value added tax rate, and the efficiency of the value-added tax. The tax reform index is

higher, the lower is the average of the marginal tax rates. The privatization index is

calculated as one minus the ratio of value-added in state owned enterprises to non-

agricultural GDP. Finally, the labor market reform index considers firing costs after 1

and 10 years of work, mandatory costs for over-time work, restrictions on temporary

contracts, and the value of contributions to social security. All the indices are normalized

between 0 and 1, where in each case, 0 refers to the minimum value of the index across

all Latin American countries in the relevant time period (including those that do not

appear in our data on wage differentials), and 1 is the maximum registered in the whole

sample. Thus, the indices are comparable across countries in the region.

Table 4a presents the results from regressing BBS1 on an average of the five reform

indexes. The regression corresponds to a fixed effects estimation performed by linking

the average mobility index across age groups and parental education quintile for each

17

country and year, with the reform index for the same country and year. Even though we

control for all country fixed characteristics, we interpret this only as a correlation since

we do not control for time-varying unobserved variables that are correlated with reforms,

and that might affect BBS1. The correlation is positive and statistically significant,

indicating that intergenerational mobility, as characterized by BBS1 is strongly correlated

with economic reform in Latin America. Table 4b presents similar results for BBS2, and

leads to the same conclusions.

In Table 4c we performed a similar fixed effects regression, but the BBS indexes are

substituted by the DG measure. Interestingly, the correlation between reforms and

mobility is still positive, but it is not statistically significant.

Tables 5a, 5b and 5c, BBS1, BBS2 and DG are regressed on each of the reform indexes

included independently. The conclusion about the correlation between different measures

and each reform varies depending on the mobility index chosen. Tax reform and financial

market reform are positively correlated with BBS1, while capital account liberalization is

negatively and significantly correlated with this same measure. There is no statistically

significant association between trade reform and privatization and BBS1. The main

difference with BBS2 is that for this index we do not find a significant correlation with

capital market liberalization, and that the correlation with privatization is negative, rather

than positive as with BBS1. A totally different picture emerges when DG is used as

dependent variable. Trade and Capital market reforms now appear to have a positive

correlation, while tax reform has a negative significant correlation with this mobility

index.

3. Conclusions

18

References

Armitage, Jane and Richard Sabot, 1990, "Education Policy and IntergenerationalMobility," in J.B. Knight and R. Sabot, eds., Education, Productivity and Inequality: TheEast African Natural Experiment, Oxford: Oxford University Press.

Atkinson, A.B., F. Bourguignonne, and C. Morrisson, 1992, Empirical Studies ofEarnings Mobility, Chur, Switzerland: Harwood Academic Publishers.

Bartholomew, D.J., 1982, “Stochastic Models for Social Processes,” 3rd edition, NewYork: Wiley.

Behrman, Jere R., 2000, “Social Mobility: Concepts and Measurement in Latin Americaand the Caribbean” in Nancy Birdsall and Carol Graham, eds., New Markets, NewOpportunities? Economic and Social Mobility in a Changing World,, Washington, DC:The BrookingsInstitution and the Carnegie Endowment for International Peace, 69-100.

Behrman, Jere R. and James C. Knowles, 1999, “Household Income and Child Schoolingin Vietnam,” World Bank Economic Review 13:2 (May), 211-256.

Behrman, Jere R , Nancy Birdsall and Miguel Székely, 2000, “Intergenerational Mobilityin Latin America: Deeper Markets and Better Schools Make a Difference” in NancyBirdsall and Carol Graham, eds., New Markets, New Opportunities? Economic andSocial Mobility in a Changing World, Washington, DC: The Brookings Institution andthe Carnegie Endowment for International Peace, 135-167.

Behrman, Jere R., Suzanne Duryea and Miguel Székely “Schooling Investments andMacroeconomic Conditions: A Micro-Macro Investigation for Latin America and theCaribbean”, OCE Working Paper Series No., 407Research Department, Inter AmericanDevelopment Bank, October, 1999.

Behrman, Jere R., Robert A. Pollak, and Paul Taubman, 1995, From Parent to Child:Intrahousehold Allocations and Intergenerational Relations in the United States,Chicago, IL: University of Chicago Press.

Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman, 1994, "Endowments and theAllocation of Schooling in the Family and in the Marriage Market: The TwinsExperiment," Journal of Political Economy 102:6 (December), 1131-1174.

Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman, 1996, "College Choice andWages: Estimates Using Data on Female Twins," Review of Economics and Statistics73:4 (November), 672-685.

Behrman, Jere R. and Paul Taubman, 1985, “Intergenerational Earnings Mobility in theUnited States: Some Estimates and a Test of Becker’s Intergenerational Endowments

19

Model” (with Paul Taubman), Review of Economics and Statistics 67:1 (February 1985),141-5; reprinted in From Parent to Child: Intrahousehold Allocations andIntergenerational Relations in the United States (with Robert A. Pollak and PaulTaubman), Chicago: University of Chicago Press, 1995.Behrman, Jere R. and Paul Taubman, 1990, "The Intergenerational Correlation betweenChildren's Adult Earnings and Their Parents' Income: Results from the Michigan PanelSurvey of Income Dynamics," The Review of Income and Wealth 36:2 (June), 115-127.

Bénabou, Roland, 1996, “Heterogeneity, Stratification, and Growth: MacroeconomicImplications of Community Structure and School Finance,” American Economic Review86:3 (June), 584-609.

Berry, Albert, 1997, “The Income Distribution Threat in Latin America,” Latin AmericanResearch Review 32:2, 3-40.

Birdsall, Nancy and Carol Graham, 1998, “New Ways of Looking at Old Inequities:Market Reforms, Social Mobility and Sustainable Growth (Latin America inComparative Context),” Washington, DC: The Brookings Institution and the Inter-American Development Bank, mimeo.

Birdsall, Nancy and Carol Graham, eds., 2000, New Markets, New Opportunities?Economic and Social Mobility in a Changing World,, Washington, DC: The BrookingsInstitution and the Carnegie Endowment for International Peace.

Birdsall, Nancy, Carol Graham, and Richard Sabot, eds., 1998, Beyond Tradeoffs: MarketReforms and Equitable Growth in Latin America, Washington, DC: Brookings Instituteand Inter-American Development Bank.

Conlisk, John, 1974, “Can Equalization of Opportunity Reduce Social Mobility?”American Economic Review 64:1 (March), 80-90.

Conlisk, John., 1989, “Ranking Mobility Matrices,” Economic Letters 29, 231-235.

Conlisk, John, 1990, “Monotone Mobility Matrices,” Journal of Mathematical. Sociology15, 173-191.

Dahan, Momi and Alejandro Gaviria, 2000, “Sibling Correlations and IntergenerationalMobility in Latin America,” Washington, DC: Inter-American Development Bank,mimeo.

Dardanoni, Valentino, 1993, “Measuring Social Mobility,” Journal of Economic Theory61, 372-394.

Deaton, Angus and Christina Paxson, 1994, "Intertemporal Choice and Inequality,"Journal of Political Economy 102:3 (June), 437-467.

20

Featherman, David L., F. Lancaster Jones, and Robert M. Hauser, 1975, “Assumptions ofSocial Mobility Research in the U.S.: The Case of Occupational Status,” Social ScienceResearch 4:4 (December), 329-360.

Fields, Gary S. and Efe A. Ok, 1996, “The Meaning and Measurement of Income Mobility,”Journal of Economic Theory 71, 349-377.

Friedman, Milton, 1962, Capitalism and Freedom, Princeton, NJ: Princeton University.

Friedman, Milton and Simon Kuznets, 1954, Income from Independent Professional Practice,New York: NBER.Friesen, P.H. and D. Miller, 1983, “Annual Inequality and Lifetime Inequality,” QuarterlyJournal of Economics 98, 139-155.

Galton, F., 1889, Natural Inheritance, London: Macmillan.

Geweke, J., R.C. Marshall, and G.A. Zarkin, 1986, “Mobility Indices in Continuous TimeMarkov Chains,” Econometrica 54, 1407-1423.

Grusky, David B. and Robert M. Hauser, 1984, “Comparative Social Mobility Revisited: Modelsof Convergence and Divergence in 16 Countries,” American Sociological Review 49:1(February), 19-38.

Hauser, Robert M., 1986, “Reinventing the Oxcart: Jones’ Obsolete Proposal for MobilityAnalysis,” Social Forces 64:4 (June), 1057-1065.

Hauser, Robert M. and David B. Grusky, 1988, “Cross-National Variation in OccupationalDistributions, Relative Mobility Chances, and Intergenerational Shifts in OccupationalDistributions,” American Sociological Review 53 (October), 723-741.

Hauser, Robert M. and John Allen Logan, 1992, “How Not to Measure IntergenerationalOccupational Persistence,” American Journal of Sociology 97:6 (May), 1689-1711.

Hill, M. Anne, 1990, Intercohort Differences in Women’s Labor Market Transitions,” AEAPapers and Proceedings 80:2 (May), 289-292.

Hout, Michael and Robert M. Hauser, 1992, “Symmetry and Hierarchy in Social Mobility: AMethodological Analysis of the CASMIN Model of Class Mobility,” European SociologicalReview 8:3 (December), 239-266.

Kanbur, S.M.R. and J.E. Stiglitz, 1986, “Intergenerational Mobility and Dynastic Inequality,”Woodrow Wilson Discussion paper No. 111, Princeton: Princeton University, mimeo.

Lam, David, 1996, "Demographic Variables and Income Inequality," in Mark R. Rosenzweig andOded Stark, eds., Handbook of Population and Family Economics, Amsterdam: North-HollandPublishing Company.

Lam, David and Deborah Levison, 1991, "Declining Inequality in Schooling in Brazil and itsEffect on Inequality in Earnings," Journal of Development Economics 37:1/2 (November),199-226.

21

Lam, David and Deborah Levison, 1992, "Age, Experience and Schooling: DecomposingEarnings Inequality in the United States and Brazil," Sociological Inquiry 62:2, 218-245.

Lam, David and Robert F. Schoeni, 1993, "Effects of Family Background on Earnings andReturns to Schooling: Evidence from Brazil," Journal of Political Economy 101:4 (August),710-140.

Lam, David and Robert F. Schoeni, 1994, "Family Ties and Labor Markets in the United Statesand Brazil," Journal of Human Resources 29:4 (Fall), 1235-1258.

Lillard, Lee and Robert Willis, 1994, "Intergenerational Education Mobility: Effects of Familyand State in Malaysia," Journal of Human Resources 29:4 (Fall), 1126-1166.

Londoño, Juan Luis and Miguel Szekely, 1997, “Distributional Surprises After a Decade ofReforms: Latin America in the Nineties,” Washington, DC: Inter-American Development Bank,mimeo.

Lustig, Nora and Miguel Szekely, 1997, “ “Hidden” Trends in Poverty and Inequality in Mexico,”Washington, DC: Inter-American Development Bank, mimeo.

Mare, Robert D., 1996, “Demography and the Evolution of Educational Inequality,” in James N.Baron, David B. Grusky and Donald J. Treiman, eds., Social Differentiation and SocialInequality: Essays in Honor of John Pock, Boulder, CO: Westview Press, 117-51.

Markandya, A., 1984, “The Welfare Measurement of Changes in Economic Mobility,”Economica 51, 457-471.

McMurrer, Daniel P. and Isabel V. Sawhill, 1997, Getting Ahead: Economic and Social Mobilityin America, Washington, D.C.: The Urban Institute, mimeo.

Miller, Paul W. and Paul A. Volker, 1985, “On the Determination of Occupational Attainmentand Mobility,” Journal of Human Resources 20:2 (Spring), 197-213.

Morley, Samuel, 1994, Poverty and Inequality in Latin America, Baltimore: John HopkinsUniversity Press.

Piketty, Thomas, 1995, “Social Mobility and Redistributive Politics,” Quarterly Journal ofEconomics CX:3 (August), 551-584.

Prais, S.J., 1955, “Measuring Social Mobility,” Journal of the Royal Statistical Society Ser. A118, 56-66.

Psacharopoulos, George, Samuel Morley, Ariel Fiszbein, Haeduck Lee and Bill Wood, 1992,Poverty and Income Distribution in Latin America: The Story of the 1980s, Washington, DC:World Bank.

Schemo, Diana Jean, 1998, “Brazil’s Reformist Chief Rides a Bucking Bronco,” New York TimesFebruary 8, 1998, p. 12.

22

Shorrocks, Anthony F., 1976, “Income Mobility and the Markov Assumption,” EconomicJournal 86, 556-578.

Shorrocks, Anthony F., 1978, “The Measurement of Mobility,” Econometrica 46, 1013-1024.

Slesnick, Daniel T., 1986, “Welfare Distributional Change and the Measurement of SocialMobility,” The Review of Economics and Statistics LXVII:4 (November), 586-593.

Solon, Gary R., 1989, "Biases in the Estimation of Intergenerational Earnings Correlations,"Review of Economics and Statistics 21:2, 172-174.

Solon, Gary R., 1992, "Intergenerational Income Mobility in the United States," AmericanEconomic Review, 82:3 (June), 393-408.Sommers, P.M. and J. Conlisk, 1978, “Eigenvalue Immobility Measures for Markov Chains,”Journal of Mathematical Sociology 6, 253-276.

Székely, M. and M. Hilgert, 1999a, “The 1990s in Latin America: Another Decade of PersistentInequality”, OCE Working Paper Series No. 410, Research Department, Inter AmericanDevelopment Bank, December.

Székely, M. and M. Hilgert, 1999b, “What’s Behind the Inequality we Measure?: AnInvestigation Using Latin American Data”, OCE Working Paper Series No. 409, ResearchDepartment, Inter AmericanDevelopment Bank, December.

Warren, John Robert and Robert M. Hauser, 1997, “Social Stratification Across ThreeGenerations: New Evidence from the Wisconsin Longitudinal Study,” American SociologicalReview 62 (August), 561\-572.

Zimmerman, David J., 1992, "Regression Toward Mediocrity in Economic Stature," AmericanEconomic Review 82:3 (June), 409-429.

23

Table 1Table 1. Most Mobile Matrix Conditional on Mobility Index Used

Trace Determinant Mean First

Passage Time

Bartholomew’s

Measure

Eigenvalue

P1, P3 P1 P3 P1, P2, P3 P2

Table 2

Table 3

Share of Teenagers (age 13-19)Children

Children Related Other Own household of Headof Head to Head non-relative head or spouse (Two Parents)

Arg* 88.71 7.59 1.22 2.47 83.72Bol 84.84 9.40 2.12 3.65 83.06Chl 84.11 13.52 0.99 1.39 83.26Bra 83.70 10.29 1.42 4.59 81.60Ven 78.63 17.56 1.58 2.22 76.61

*GBA onlySource: Duryea, Edwards and Ureta 2000. "Adolescents and Youth in Latin America"

Correlation CoefficientsGini Index BBS1 BBS2 DG

Gini Index 1BBS1 -0.4482 1BBS2 -0.551 0.8701 1DG 0.5658 -0.3505 -0.4959 1

24

Table 4country bbsr country bbsag country dg

Paraguay 0.929258 Brazil 0.7329 El Salvador 0.387343Brazil 0.935106 Paraguay 0.745058 Mexico 0.416248Peru 0.94028 Guatemala 0.749411 Nicaragua 0.427939Bolivia 0.940713 Nicaragua 0.790979 Guatemala 0.44914El Salvador 0.94611 El Salvador 0.814268 Brazil 0.449868Uruguay 0.946572 Bolivia 0.848708 Ecuador 0.453114Guatemala 0.947686 Peru 0.850669 Argentina 0.467544Nicaragua 0.949789 Mexico 0.865842 Bolivia 0.474531Mexico 0.953171 Ecuador 0.868582 Costa Rica 0.492356Costa Rica 0.956957 Uruguay 0.868625 Colombia 0.493189Ecuador 0.958991 Costa Rica 0.875188 Honduras 0.500401Panama 0.961621 Dominican Republic 0.897136 Dominican Republic 0.522026Dominican Republic 0.962695 Colombia 0.900487 Venezuela 0.524479Argentina 0.96671 Venezuela 0.905804 Chile 0.550535Venezuela 0.967292 Panama 0.909905 Peru 0.562818Colombia 0.970244 Honduras 0.92553 Uruguay 0.587608Taiwan 0.972216 Argentina 0.932881 Panama 0.588539Chile 0.975943 Chile 0.946481 Paraguay 0.629256Honduras 0.979925 Taiwan 0.970515 Taiwan 0.828759United States 0.98686 United States 0.97967 United States

25

Table 4a. xtreg bbsr totidx, i(id) fe

Fixed-effects (within) regression Number of obs = 48Group variable (i) : id Number of groups = 15

R-sq: within = 0.1859 Obs per group: min = 1 between = 0.0105 avg = 3.2 overall = 0.0474 max = 7

F(1,32) = 7.31corr(u_i, Xb) = -0.0034 Prob > F = 0.0109

------------------------------------------------------------------------------ bbsr | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- totidx | .0275703 .0102001 2.703 0.011 .0067934 .0483472 _cons | .9363158 .0072426 129.279 0.000 .9215631 .9510685------------------------------------------------------------------------------ sigma_u | .01818868 sigma_e | .00700375 rho | .87087372 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(14,32) = 17.62 Prob > F = 0.0000

Table 4b

. xtreg bbsag totidx, i(id) fe



F(1,32) = 13.79corr(u_i, Xb) = 0.0709 Prob > F = 0.0008

------------------------------------------------------------------------------ bbsag | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- totidx | .2000376 .0538637 3.714 0.001 .0903208 .3097544 _cons | .6995948 .0382461 18.292 0.000 .62169 .7774996------------------------------------------------------------------------------ sigma_u | .08742295 sigma_e | .03698483 rho | .8481932 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(14,32) = 19.67 Prob > F = 0.0000

26

Table 4c. xtreg dg totidx, i(id) fe



F(1,31) = 0.57corr(u_i, Xb) = 0.1565 Prob > F = 0.4579

------------------------------------------------------------------------------ dg | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- totidx | .0597081 .079426 0.752 0.458 -.1022823 .2216985 _cons | .4797011 .0568667 8.436 0.000 .3637206 .5956816------------------------------------------------------------------------------ sigma_u | .08126226 sigma_e | .05352345 rho | .69743716 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(14,31) = 7.64 Prob > F = 0.0000

27

Table 5a. xtreg bbsr kapidx tradeidx finidx prividx taxidx, i(id) fe



F(5,28) = 4.50corr(u_i, Xb) = -0.6181 Prob > F = 0.0039

------------------------------------------------------------------------------ bbsr | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- kapidx | -.0314912 .015394 -2.046 0.050 -.0630244 .0000421tradeidx | -.0098675 .0125867 -0.784 0.440 -.0356502 .0159151 finidx | .0148581 .0076401 1.945 0.062 -.0007919 .0305082 prividx | .0101721 .0128029 0.795 0.434 -.0160535 .0363977 taxidx | .0420371 .0168488 2.495 0.019 .007524 .0765502 _cons | .9484621 .0102614 92.431 0.000 .9274427 .9694815------------------------------------------------------------------------------ sigma_u | .02102453 sigma_e | .00617939 rho | .92048387 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(15,28) = 12.78 Prob > F = 0.0000

Table 5b. xtreg bbsag kapidx tradeidx finidx prividx taxidx, i(id) fe



F(5,28) = 6.50corr(u_i, Xb) = -0.2121 Prob > F = 0.0004

------------------------------------------------------------------------------ bbsag | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- kapidx | -.0364358 .0801527 -0.455 0.653 -.2006212 .1277496tradeidx | -.0714882 .0655356 -1.091 0.285 -.2057318 .0627555 finidx | .0667371 .0397801 1.678 0.105 -.0147487 .1482229 prividx | -.0587651 .0666614 -0.882 0.386 -.1953149 .0777847 taxidx | .2537579 .087727 2.893 0.007 .0740573 .4334584 _cons | .7880331 .0534281 14.749 0.000 .6785905 .8974756------------------------------------------------------------------------------ sigma_u | .08623046 sigma_e | .03217445 rho | .87779378 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(15,28) = 17.39 Prob > F = 0.0000

28

Table 5c. xtreg dg kapidx tradeidx finidx prividx taxidx, i(id) fe



F(5,27) = 2.00corr(u_i, Xb) = -0.4730 Prob > F = 0.1107

------------------------------------------------------------------------------ dg | Coef. Std. Err. t P>|t| [95% Conf. Interval]---------+------------------------------------------------------------------- kapidx | .1396252 .1255803 1.112 0.276 -.1180442 .3972947tradeidx | .2514425 .1011321 2.486 0.019 .0439365 .4589485 finidx | -.052167 .0611277 -0.853 0.401 -.1775908 .0732567 prividx | -.0662397 .1051264 -0.630 0.534 -.2819413 .1494619 taxidx | -.2457972 .1369205 -1.795 0.084 -.5267349 .0351406 _cons | .4161292 .0847136 4.912 0.000 .2423113 .589947------------------------------------------------------------------------------ sigma_u | .10075596 sigma_e | .04943549 rho | .80597497 (fraction of variance due to u_i)------------------------------------------------------------------------------F test that all u_i=0: F(15,27) = 7.27 Prob > F = 0.0000

29

Figure 1a

Figure 1b

Figure 1c

(mean) sdgini

bbsr Fitted values

.365103 .661942

.909011

.991575

(mean) sdgini

bbsag Fitted values

.365103 .661942

.586349

.987788

(mean) sdgini

dg Fitted values

.365103 .661942

.357098

.828759

30

Figure 2ª

Figure 2b

Figure 2c

bbsr

dg Fitted values

.909011 .991575

.357098

.828759

bbsag

dg Fitted values

.586349 .987788

.357098

.828759

bbsag

bbsr Fitted values

.586349 .987788

.909011

.991575

31

Figure 3a

Figure 3b

Figure 3c

year

bbsrlac bbsrusa bbsrtai

1980 1999

.928184

.991575

year

bbsaglac bbsagusa bbsagtai

1980 1999

.790935

.987788

year

bbsdglac bbsdgtai

1980 1999

.456006

.828759

32

Figure 4ª: Costa Rica

Figure 4b: Peru

year

bbsr bbsag dg

1987 1998

.398019

.965046

year

bbsr bbsag dg

1985 1997

.492406

.960831

33

Table A2 (Incomplete)

Household Surveys Used in Growth Regressions

Country Year Name of the survey Reference

Month for Incomes Households Individuals Labor Property Rent Capital Rent Transfers Non-Monetary Imputed Rent

1 Bolivia 96 Encuesta Nacional de Empleo June 8,311 35,648 X X X X n.a. n.a.

2 97 Encuesta Nacional de Empleo November 8,461 36,752 X X X X n.a. n.a.

3 Brazil 81 Pesquisa Nacional por Amostra de Domicilios September 103,193 481,480 X X X X n.a. n.a.

4 83 Pesquisa Nacional por Amostra de Domicilios September 113,599 511,147 X X X X n.a. n.a.







11 Chile 87 Encuesta de Caracterización Socioeconómica Nacional November 22,719 97,044 X X X X X X

12 90 Encuesta de Caracterización Socioeconómica Nacional October 25,793 105,189 X X X X X X




16 Colombia 95 Encuesta Nacional de Hogares - Fuerza de Trabajo September 18,255 79,012 X Xb Xb Xb n.a. n.a.

17 97 Encuesta Nacional de Hogares - Fuerza de Trabajo September 32,442 143,398 X Xb Xb Xb n.a. n.a.

18 Costa Rica 81 Encuesta Nacional de Hogares - Empleo y Desempleo July 6,604 22,170 X n.a. n.a. n.a. n.a. n.a.

19 83 Encuesta Nacional de Hogares - Empleo y Desempleo July 7,132 23,449 X n.a. n.a. n.a. n.a. n.a.

20 85 Encuesta Nacional de Hogares - Empleo y Desempleo July 7,351 23,960 X n.a. n.a. n.a. n.a. n.a.

21 87 Encuesta de Hogares de Propósitos Múltiples July 7,510 34,591 X Xb Xb Xb n.a. n.a.






27 Dominican Republic 96 Encuesta Nacional de Fuerza de Trabajo February 5,548 24,041 X X X X n.a. n.a.

28 Ecuador 95 Encuesta de Condiciones de Vida July-Sept. 5,810 26,941 X X X X X X

29 El Salvador 95 Encuesta de Hogares de Propósitos Múltiples 1995 8,482 40,004 X Xb Xb Xb n.a. X

30 Honduras 89 Encuesta Permanente de Hogares de Propósitos Múltiples August 8,727 46,672 X n.a. n.a. n.a. n.a. n.a.

31 92 Encuesta Permanente de Hogares de Propósitos Múltiples August 4,757 24,704 X n.a. n.a. n.a. n.a. n.a.

32 96 Encuesta Permanente de Hogares de Propósitos Múltiples August 6,428 33,172 X n.a. n.a. n.a. n.a. n.a.

33 98 Encuesta Permanente de Hogares de Propósitos Múltiples February 6,493 32,696 X Xb Xb Xb n.a. n.a.

34 Mexico 84 Encuesta Nacional de Ingreso Gasto de los Hogares Third quarter 4,735 23,985 X X X X X X

35 89 Encuesta Nacional de Ingreso Gasto de los Hogares Third quarter 11,531 57,289 X X X X X X




39 Nicaragua 93 Encuesta Nacional de Hogares Sobre Medicion de Niveles de Vida February to June 4,455 24,542 X X X X X X

40 Panama 79 Encuesta Continua de Hogares - Mano de Obra July 8,593 24,284 X n.a. n.a. n.a. n.a. n.a.

41 91 Encuesta Continua de Hogares - Mano de Obra July 8,867 38,000 X Xa Xa X n.a. n.a.

42 95 Encuesta Continua de Hogares July 9,875 40,320 X Xa Xa X n.a. n.a.

43 97 Encuesta de Hogares July 9,897 39,706 X Xa Xa X n.a. n.a.

44 Paraguay 95 Encuesta de Hogares - Mano de Obra August to November 4,667 21,910 X X X X n.a. n.a.

45 Peru 85-86 Encuesta Nacional de Hogares sobre Medición de Niveles de Vida July 1985 to July 1986 5,108 26,323 X X X X X X

46 91 Encuesta Nacional de Hogares sobre Medición de Niveles de Vida August-October 2,308 11,507 X Xa Xa X X X

47 94 Encuesta Nacional de Hogares sobre Medición de Niveles de Vida April-June 3,623 18,662 X Xa Xa X X X

48 96 Encuesta Nacional de Hogares sobre Niveles de Vida y Pobreza April-June 16,744 88,863 X Xa Xa X X X

49 97 Encuesta Nacional de Hogares sobre Niveles de Vida y Pobreza August-October 3,843 19,575 X Xa Xa X X X

50 Venezuela 81 Encuesta de Hogares por Muestra Second semester 45,421 239,649 X n.a. n.a. n.a. n.a. n.a.

51 86 Encuesta de Hogares por Muestra Second semester 129,713 682,636 X n.a. n.a. n.a. n.a. n.a.



54 95 Encuesta de Hogares por Muestra Second semester 18,702 92,450 X Xb Xb Xb n.a. n.a.

55 97 Encuesta de Hogares por Muestra Second semester 15,948 76,965 X Xb Xb Xb n.a. n.a.

a. Can not separate between property and capital rent.

b. Can not separate between property rent, capital rent, and transfers.

Sample size Income

Documents

Comparisons of Intergenerational Schooling Mobility ...and Székely basically define intergenerational mobility to be the extent to which school gaps of children who co-reside with