Objectives. I estimated the association between parents’ education, mothers’ vocabulary, and early child cognitive development in a sample of poor children in rural Ecuador.

Methods. I used regression analysis to estimate the association between parents’ education, mothers’ vocabulary, and the vocabulary, memory, and visual integration skills of children at early ages, controlling for possible confounders. The study is based on a longitudinal cohort of children in rural Ecuador (n = 2118).

Results. The schooling and vocabulary levels of mothers were strong predictors of the cognitive development of young children. Household wealth and child's height, weight, and hemoglobin levels explained only a modest fraction of the observed associations. The vocabulary levels of mothers and children were more strongly correlated among older children in the sample, suggesting that the effects of a richer maternal vocabulary are cumulative.

Conclusions. Differences in children's cognitive outcomes start very early, which has important implications for the intergenerational transmission of poverty and inequality. Programs that seek to increase early stimulation for disadvantaged children, perhaps through parenting programs or high-quality center-based care, hold promise.

A recent review estimates that in developing countries, more than 200 million children younger than 5 years fail to reach their potential in cognitive development because of poverty, inadequate nutrition, and insufficient stimulation.1 Children with higher levels of cognitive development at young ages are more likely to be successful throughout their lives in a number of dimensions. Long-term panels that have followed children into adulthood show that cognitive tests taken at ages 5 and 10 years explain 27% to 35% of the variation in the logarithm of wages at age 30 years in the 1970 British Cohort Study, and tests taken at ages 7 and 11 years predict 28% to 37% of the variation in the logarithm of wages at ages 33 and 42 years in the United States National Child Development Study.2

Development in early childhood is malleable. In the United States, experimental estimates of the impact of high-quality, intensive preschool interventions such as the Perry Preschool Program and the Abecedarian Program suggest large impacts on school attainment, higher wages, and lower rates of criminal behavior in adulthood, most robustly among girls.3,4 In developing countries, long-term longitudinal studies have shown sustained benefits from better nutrition in early childhood in Guatemala5–8 and from stimulation and parenting interventions in Jamaica.9–11 A variety of other interventions have also shown promise in some settings.12

I used longitudinal data from rural Ecuador to analyze the relationship between the schooling level of parents, the vocabulary of mothers, and cognitive development in young children. As is well-known, there are substantial disparities in cognitive development associated with family background in many settings.1,13,14 I investigated when these disparities set in and how they evolve as children age.

I based the analysis on a longitudinal data set with information on child development at 2 points in time. I built on, and substantially extended, earlier work on Ecuador that used a single cross-section and a single cognitive outcome.15 The sample was drawn from rural areas in 6 of the 22 provinces in Ecuador. Of the provinces, 3 are in the “Sierra” or highlands region of the country, and 3 are on the coast. No data were collected on children from the sparsely populated “Oriente” or jungle region.

The data I used were originally collected for an evaluation of the impact of a cash transfer program, the Bono de Desarrollo Humano (Human Development Bond).16 They include only households in the poorest half of the nationwide distribution of a composite measure of wealth because these are the households that are eligible or almost eligible for transfers from the Bono program. Furthermore, only households with at least 1 child younger than 6 years in 2003, but with no children older than 6 years, were eligible for the variant of the Bono program that was the subject of this evaluation. As a result, my sample consisted mainly of young mothers with relatively small families. The sample was chosen in 2 phases. In the first phase, 79 rural parishes were randomly selected within the 6 study provinces. In the second phase, up to 50 families that were eligible for the study were randomly selected in each parish (some parishes had < 50 eligible families).

The first of the surveys analyzed, which I refer to as baseline, was conducted between September 2005 and January 2006, whereas the second, referred to as follow-up, was carried out between May and July 2008. On average, 32 months elapsed between baseline and follow-up, with the exact number depending on the date of the 2 surveys for a given family. Habitus Investigación, a specialized firm in Quito, Ecuador, collected both the baseline and follow-up surveys. The enumerators had previous experience in the collection of household survey data and received intensive training specific to this survey, in particular with regard to the cognitive tests that were applied.

I focused on children aged 36 to 71 months at baseline. I did not use cognitive tests for children younger than 36 months in the sample. At baseline, there were 2128 children aged between 36 and 71 months; of these, 660 were 3, 811 were 4, and 657 were 5 years of age. The average age was 54 months at baseline and 85 months at follow-up. To avoid any biases that could potentially result from children not understanding the tests used to measure cognitive development, in particular the vocabulary test, I did not use data on 10 children (< 0.5% of the original sample) who were reported to speak a language other than Spanish. However, given the small number of children involved, the results were very similar if these children were included in the regressions.


The study enumerators administered 3 cognitive tests to children aged 36 months and older. The first, the Test de Vocabulario en Imágenes Peabody (TVIP), is the Spanish version of the Peabody Picture Vocabulary Test, a widely used test.17 Children are shown slides, each of which has 4 objects, and are asked to identify an object named by the enumerator. The test is a measure of receptive vocabulary because children do not have to name the objects themselves and need not be able to read or write. The test was carefully piloted to ensure that the words were culturally appropriate for the context. Importantly, the TVIP was also administered to children's mothers.

The other 2 tests were drawn from the Woodcock-Johnson-Muñoz battery of cognitive abilities.18 One is a test of long-term retrieval, or associative memory. Children are gradually introduced to a series of space creatures with nonsensical names, and are then shown groups of space creatures that they are asked to identify. The last test measures visual integration, or visual-spatial thinking. Children are shown a series of pictures of common objects that have been distorted in various ways, and are asked to identify the object. Long-term retrieval and visual-spatial thinking are 2 of the primary broad factors in the Cattell-Horn-Carroll theory of cognitive abilities.19–21

The 3 cognitive tests were administered at baseline and follow-up. To keep the number of results in the article manageable, I present only the baseline tests in the main set of results. However, the results were very similar when I used the follow-up data or focused on changes in development between baseline and follow-up.

Tests were administered in children's homes, and every attempt was made to preserve reasonable testing conditions. As a robustness check, I corrected for differences across households on a number of dimensions of the testing environment, as reported by enumerators. Cronbach α values for the tests were reasonable (0.74 for the TVIP, 0.87 for the memory test, and 0.89 for the visual integration test), suggesting that the tests had good internal consistency. The pairwise correlation coefficients across tests were positive and statistically significant, ranging from 0.40 to 0.50, well below unity. This finding suggests that different tests picked up on different underlying dimensions of child development, although measurement error in the tests is also likely to play a role.

To assess the extent to which cognitive development at baseline predicted a child's performance in school 2.5 years later, the study included 3 additional tests at follow-up that measured knowledge of letters and words, basic mathematics, and numeric series. These tests were drawn from the Woodcock-Johnson-Muñoz battery of tests of achievement (as opposed to cognitive abilities). In the test of letters and words, children are asked to read out words of increasing difficulty. The mathematics test is a test of basic arithmetic (addition, subtraction, and multiplication); children are asked to answer as many questions as possible within 3 minutes. Finally, the test of numeric series asks children to complete series that are missing a given number—for example, one question listed the numbers 2, 4, and 8, with a blank space between 4 and 8; the correct answer was 6.

An important question for my analysis was whether, for any given test, one should use the score standardized with the guidelines provided by test developers or some other transformation of the test score. The TVIP was standardized by the test developers on samples of Mexican and Puerto Rican children and young adults to have an average score of 100 at all ages and an SD of 15.17 The mean standardized TVIP score of children in my sample was 84, more than 1 SD behind the norm. The performance of mothers was even lower: the average score was 73, almost 2 SD below the norm. Performance on the tests of memory and visual integration was also low. In this case, the guidelines provided by the test developers made it possible to convert the raw scores into percentiles of the population that was used to standardize the test. On average, children in the sample scored at the 17th percentile of the memory test and the 9th percentile of the visual integration test.

Working with the externally standardized scores for the tests is attractive, as in principle it makes it possible to assess cognitive deficits relative to how these children “should” be performing. However, it is not clear whether the reference populations that were used to standardize the tests are appropriate for rural Ecuador. To be conservative, I therefore constructed internally standardized, age-specific z scores for every test. For children, I did this by subtracting the month-of-age-specific mean of the raw score and dividing by the month-of-age-specific standard deviation. (On average, there were 59 children for each month of age, and at no age were there fewer than 34 children.) These z scores therefore have a mean of 0 and an SD of 1 at every age. For mothers, I followed the same procedure, but did so for the sample as a whole rather than separately for mothers of different ages. (In the sample, younger and older mothers did not have significantly different average scores on the test.) In practice, however, the results were broadly similar if, instead of using the internally standardized scores, I worked with the raw scores or the scores that used the external reference populations.

Explanatory Variables

I focused on the association between children's cognitive development, parents’ education (years of completed schooling, separately, for each parent), and mothers’ vocabulary (as measured by their TVIP scores). In some regressions, I also included a household wealth index and measures of child height, weight, and hemoglobin status. To calculate the wealth index, I aggregated a number of household assets and dwelling characteristics by principal components. The wealth index used here is the value of the first principal component. Similar wealth indices have been used extensively in the medical, demographic, and economics literature,22–25 including research on the population I studied.15 The results I present are not sensitive to the exact choice of variables included or to alternative ways of aggregating them. Enumerators measured height in the field using stadiometers (Seca, Hanover, MD) and measured weight with electronic scales (Tanita, Illinois, IL). I converted height into height-for-age z scores, and weight into weight-for-height z scores, using standards provided by the World Health Organization (WHO). Enumerators measured hemoglobin levels in the field using portable photoreflectometers (Hemocue, Angelholm, Sweden). Because higher levels of hemoglobin are required at greater elevation above sea level, before analysis I adjusted hemoglobin levels for differences in elevation across parishes using procedures published by the US Centers for Disease Control and Prevention26; elevation was measured at the household level using handheld global positioning system (GPS) units (Garmin, Olathe, KS).

Missing Data

I discuss 3 potential sources of missing data: attrition (loss to follow-up), children who were missing 1 or more tests, and children who were missing 1 or more covariates.

As with any longitudinal study, attrition (loss to follow-up) was a potential source of concern. Figure A (available as a supplement to the online version of this article at http://www.ajph.org) shows the number of children in the 2 survey waves and the proportion who were missing 1 or more cognitive tests. Attrition between baseline and follow-up surveys in the study was low. There were 2118 children aged 3, 4, and 5 years at baseline, 122 (5.8%) of whom could not be found at follow-up. The characteristics of parents and children who were observed in both survey waves were very similar (Table A, available as a supplement to the online version of this article at http://www.ajph.org).

A somewhat higher percentage of children (17.4%) were missing 1 or more of the baseline cognitive development tests. Younger children were more likely to be missing tests, as were children of lower household wealth, parental education levels, maternal TVIP scores, and nutritional status (Table A).

I analyzed the test performance of children who missed a given test at baseline but took the same test at follow-up. Compared with children who took the test in both survey waves, children who missed a test at baseline did significantly worse on the same test if they took it at follow-up. More children of mothers with low TVIP scores, and of parents with lower education levels, were missing tests than were other children. In principle, this could introduce biases into the reported results. As a robustness test, I therefore present results in which I imputed the baseline test scores for children who were missing tests by using the percentiles of the distribution of the tests these children took at follow-up. For example, a child who did not take the TVIP at baseline but took it at follow-up, and scored in the 25th percentile of the distribution, would be given a baseline TVIP score equivalent to the score at the 25th percentile of the baseline distribution.

A modest percentage of children (3.9% and 5.5%, respectively) did not complete the tests of knowledge of letters and words and of numeric series at follow-up. A much higher percentage of children (32.4%) did not take the basic math test. In the majority of cases (95.4%), this was because children could not add or subtract—an obvious prerequisite for this test. The small number of missing values for the tests of letters and words and of numeric series make it unlikely that missing data would bias the results. For the basic math test, I report whether or not a child took the test and how the child scored on the test, assigning children who did not take the test a score of zero. (Results were qualitatively similar if, instead, I discarded data for children who did not take the math tests from these estimates.)

A little more than one quarter of children (31.6%) in the sample were missing data on 1 or more covariates—in most cases, data on either child hemoglobin levels (missing for 20.9% of children) or the TVIP scores of mothers (missing for 6.6% of children). In the main set of results, I limited the analysis to children with no missing data. However, as a robustness test, I also present results in which I replaced the missing values with the sample averages.

Statistical Methods

I report 2 sets of results. In the first set, I used the baseline data to describe (1) the association between parents’ education, mothers’ vocabulary scores, and children's cognitive development; (2) how this association changed as children aged; and (3) whether the observed relationship appeared to be mediated by household wealth and child nutritional status. In keeping with the literature on socioeconomic status and child health,27–30 I refer to the differences in development between children of parents with more or less schooling (or higher or lower mother TVIP scores) as schooling or maternal vocabulary gradients.

To account for differences across parishes (e.g., differences in the disease environment or in the availability of health or education infrastructure), all regressions include a set of parish fixed effects. The estimated coefficients can therefore be interpreted as differences in child cognitive development between households living in the same parish. For each outcome, specification 1 reports the coefficients on the measures of maternal schooling, paternal schooling, and mother's vocabulary scores without additional controls. In specification 2, I also included the wealth index and the measures of child nutritional status.

Because my main interest was the association between cognitive development, parents’ education, and mothers’ vocabulary levels, I report only the coefficient and standard errors on these variables. However, I also report the P value on an F test of joint significance of the parish fixed effects, an F test of joint significance of the child nutrition variables, and the P value on the t test of significance on the wealth index in specification 2. A comparison of the coefficients on parents’ education and mothers’ vocabulary in specifications 1 and 2 is informative about the extent to which child nutrition or household wealth appears to mediate the observed associations.

I next report results on age patterns in the relative development of children of parents with more or less formal schooling, or of mothers with higher or lower vocabulary. The first results were based on a series of descriptive graphs that focused on differences between children in the upper and lower thirds of the distribution of mother TVIP scores, by month of age. There were relatively few children at any given month of age, so I took 7-month moving averages.

The graphs show strong age patterns in the association between the vocabulary scores of mothers and children. Because no such age patterns were apparent for the measures of memory and visual integration, the next set of results was limited to child vocabulary only. I report the results of regressions of child vocabulary scores separately for single year of age of children at baseline.

The final set of results shows the association between parents’ education, mothers’ vocabulary, and school-aged children's performance on achievement tests. For this purpose, I regressed the standardized scores on the tests of letters and words, basic math, and numeric series at follow-up on the measures of parents’ education and mothers’ vocabulary, with and without controls for the baseline scores of children on the tests of language, memory, and visual integration. A comparison of the coefficients across the 2 specifications gives a sense of the extent to which any association between parents’ education and mothers’ vocabulary, on the one hand, and children's achievement, on the other, was mediated by cognitive development before children enter school.

To account for possible correlation of outcomes within villages, all regressions clustered standard errors at the parish level. For the statistical analysis, I used Stata version 10.1 (StataCorp LP, College Station, TX).

The baseline characteristics of households and children in the sample are summarized in Table B (available as a supplement to the online version of this article at http://www.ajph.org). Mothers in the sample were young (mean age = 26.9 years). Households were very poor: total per capita expenditures were US $2.10 per day for the average household, less than $2 per day for 52% of households, and less than $1 per day for 11% of households. (Ecuador adopted the US dollar as its national currency in January 2000.) On average, parents had completed 7.5 years of schooling.

A little more than half the children in the sample were boys, and the average age was 54 months. Children had serious nutrition problems; 42% were anemic (i.e., hemoglobin levels below 11.5 g/dL, after adjustment for differences in elevation above sea level) and 21% were stunted (i.e., had height for age that was more than 2 SD below that of a reference population as defined by the WHO; Table A). On the other hand, as in most samples from Latin America, the fraction of children who were wasted (i.e., had weight for height that was more than 2 SD below that of a reference population as defined by the WHO) was low—less than 2%.

Table 1 shows the results of regressions of children's cognitive development on the measures of parents’ education and mothers’ vocabulary scores, with and without additional covariates. These results suggest a strong association between children's cognitive development and both mothers’ years of schooling and mothers’ vocabulary. For example, when the additional covariates were not included in the regression, every year of maternal schooling increased the vocabulary, memory, and visual integration scores by 0.053, 0.023, and 0.037 standard deviations, respectively. An increase in maternal vocabulary of 1 standard deviation increased children's scores on the vocabulary, memory, and visual integration scores by 0.24, 0.24, and 0.20 standard deviations, respectively, with years of completed schooling held constant. Inclusion of the measures of household wealth and child nutrition reduced the coefficients on mothers’ years of schooling somewhat. The coefficient on maternal vocabulary never decreased by more than a quarter when these additional variables were included in the regression. The association between fathers’ years of completed schooling and children's cognitive development was weaker, and was significant only for the test of visual integration. The wealth index was statistically significant in all 3 specifications. In those regressions that included parents’ education and mothers’ vocabulary, child nutrition was significantly associated with vocabulary scores and had a borderline-significant association with visual integration scores (P = .054), but it did not significantly predict memory scores (P = .115).


TABLE 1— Regressions of Cognitive Development in Early Childhood on Parents’ Education, Mothers’ Vocabulary, Household Wealth, and Children's Nutritional Status: Ecuador, 2005–2006

TABLE 1— Regressions of Cognitive Development in Early Childhood on Parents’ Education, Mothers’ Vocabulary, Household Wealth, and Children's Nutritional Status: Ecuador, 2005–2006

Vocabulary (n = 1349), B (SE)
Memory (n = 1398), B (SE)
Visual Integration (n = 1291), B (SE)
Specification 1Specification 2Specification 1Specification 2Specification 1Specification 2
Mother's schooling, y0.053*** (0.011)0.043*** (0.011)0.023* (0.013)0.013 (0.014)0.037*** (0.009)0.022** (0.009)
Father's schooling, y0.013 (0.010)0.006 (0.010)0.001 (0.009)−0.006 (0.009)0.024*** (0.008)0.015* (0.008)
Mother's TVIP score, z0.241*** (0.036)0.205*** (0.036)0.241*** (0.036)0.207*** (0.036)0.200*** (0.029)0.160*** (0.029)
Model Statistics, P
    F test: parental educationa< .001< .001< .001< .001< .001< .001
    F test: parish fixed effectsb< .001< .001< .001< .001< .001< .001
    t test: asset indexc.007.003< .001
    F test: child nutritiond< .001.115.054

Note. TVIP = Test de Vocabulario en Imágenes Peabody (the Spanish version of the Peabody Picture Vocabulary Test).17 All regressions included a set of parish fixed effects. Specification 1 included only the measure of parents’ education and mothers’ scores on the TVIP; specification 2 added the wealth index and the measures of child nutritional status (height-for-age z score, weight-for-height z score, hemoglobin levels). SEs adjusted for clustering at the parish level.

aGives the P value on a test of joint significance of the parental education and mothers’ TVIP variables.

bGives the P value on a test of joint significance of the parish fixed effects.

cGives the P value on a test of significance on the wealth index.

dGives the P value on a test of joint significance of the child nutrition variables.

*P < .1; **P < .05; ***P < .01.

The results from 2 important robustness checks on the main results are shown in Table C (available as a supplement to the online version of this article at http://www.ajph.org). The first specification, specification 1, corresponded to the basic results in Table 1: the sample was limited to children without missing covariates or missing test data. In specification 2, I replaced missing covariate data with the sample mean for the relevant variable (so that, for example, all children who were missing hemoglobin data were assigned a value of 11.27). In this specification, I also imputed missing test data, as described in the Methods section. In specification 3, I controlled for a number of characteristics of the testing environment. Specifically, I included controls for whether the child was distracted by other children, adults, or animals some of the time or most of the time (2 variables); whether testing materials were placed on a flat surface, such as a table; whether both the enumerator and the child were seated during the tests; and whether tests were conducted inside the home rather than outside.

The coefficients on the basic regressions were qualitatively similar if I replaced the missing covariate values with the sample averages and imputed the missing test values, and also if I controlled for a number of characteristics of the testing environment, as described previously (Table C).

Figure 1 graphs the average scores on the tests of vocabulary, memory, and visual integration for children in the top, middle, and bottom thirds of the distribution of mothers’ TVIP scores. Consistent with Table 1, Figure 1 shows that children whose mothers had smaller vocabularies also had lower levels of cognitive development. In the case of the vocabulary test, but not the memory or visual integration tests, these gradients appeared to be steeper among older children.

Figure 2 extends the analysis of age patterns in vocabulary. The figure compares the distributions of vocabulary scores among the youngest children in the sample, aged 3 years at baseline, and the oldest children, aged 5 years at baseline. Figure 2, a and b graphs the distribution of outcomes for the sample as a whole and overlays a normal distribution, whereas figure 2, c and d graphs the distributions separately for children in the top and bottom thirds of the distribution of maternal vocabulary scores.

Figure 2 shows clear differences in the distribution of vocabulary scores between children aged 3 and 5 years. For 3-year-old children, the distribution was unimodal and right skewed, perhaps suggesting that there were not enough easy questions on the test for the youngest children. Differences between children of mothers with larger and smaller vocabularies were modest. For 5-year-old children, the distribution was bimodal. The 2 modes of the distribution corresponded well with the underlying distributions of vocabulary scores for the children of mothers with relatively high and low performances on the vocabulary test.

Table D (available as a supplement to the online version of this article at http://www.ajph.org) complements the results in the figures, focusing on performance on the vocabulary test. The association between the vocabulary scores of mothers and children appears to be larger as children become older. An F test for the equality of the coefficients on mothers’ TVIP scores for children aged 3 and 5 years at baseline had a P value of .001 in the specification without additional controls and a P value of .044 in the specification that also controlled for wealth and child nutritional status. On the other hand, the null hypothesis of equal coefficients on completed years of schooling of mothers or fathers for children of different ages could not be rejected.

Table 2 shows the results of regressions of children's school achievement on parents’ education and mothers’ vocabulary. In those regressions that did not control for cognitive development at baseline, every year of maternal schooling increased the score on the tests of letters and words, basic math, and numeric series by 0.038, 0.024, and 0.038 SD, respectively; every year of paternal schooling increased the score on the tests of letters and words, basic math, and numeric series by 0.028, 0.016, and 0.023 SD, respectively. An increase in mothers’ vocabulary score of 1 standard deviation was associated with increases in children's scores on the tests for letters and words, basic math, and numeric series of 0.172, 0.170, and 0.130 SD, respectively. A comparison of the coefficients across specifications suggested that a substantial portion of the association between parents’ characteristics and children's performance on achievement tests in school was mediated by the cognitive development of children before they entered school. For example, the coefficients on maternal vocabulary decreased by roughly one half when the measures of baseline child cognitive development were included in the regressions.


TABLE 2— Regressions of Achievement Tests for School-Aged Children on Parents’ Education, Mothers’ Vocabulary, and Performance on Achievement Tests for School-Aged Children: Ecuador, 2008

TABLE 2— Regressions of Achievement Tests for School-Aged Children on Parents’ Education, Mothers’ Vocabulary, and Performance on Achievement Tests for School-Aged Children: Ecuador, 2008

Score: Letters and Words (n = 1182), B (SE)
Took Basic Math Test (n = 1233), B (SE)
Score: Basic Math Test (n = 1215), B (SE)
Score: Numeric Series Test (n = 1173), B (SE)
Specification 1Specification 2Specification 1Specification 2Specification 1Specification 2Specification 1Specification 2
Mother's schooling, y0.038*** (0.010)0.024** (0.009)−0.004 (0.005)−0.002 (0.005)0.024*** (0.012)0.011 (0.012)0.038*** (0.011)0.028** (0.011)
Father's schooling, y0.028*** (0.009)0.024*** (0.008)−0.012*** (0.004)−0.010** (0.004)0.016 (0.010)0.012 (0.010)0.023** (0.010)0.020** (0.009)
Mother's TVIP score, z0.172*** (0.029)0.081*** (0.030)−0.041*** (0.017)−0.028 (0.018)0.170*** (0.035)0.092*** (0.034)0.130*** (0.038)0.067* (0.038)

Note. TVIP = Test de Vocabulario en Imágenes Peabody (the Spanish version of the Peabody Picture Vocabulary Test).17 All regressions include a set of parish fixed effects. Specification 1 included only the measure of parental education and mothers’ scores on the TVIP; specification 2 added the measures of cognitive development at baseline (vocabulary, memory, visual integration). SEs adjusted for clustering at the parish level.

*P < .1; **P < .05; ***P < .01.

It is widely accepted by neuroscientists, developmental psychologists, and economists that there are early sensitive or critical periods in child development when environmental experiences are particularly important.31–33 Understanding the age at which deficits in cognitive development set in, and how these correlate with parental education, maternal vocabulary, and various measures of household socioeconomic status, is critical for the design of effective social policies. I used longitudinal data from rural Ecuador to show that the relative advantage of children whose mothers have more education and richer vocabularies begins early in a child's life. This finding has important implications for the intergenerational transmission of poverty and inequality—a critical consideration given that the distribution of income in Ecuador and other Latin American countries is among the most unequal in the world.34 Broadly similar patterns, which show that the bulk of the difference in test scores by socioeconomic status is established before children enter school, are also apparent in research in the United States.35,36 The study also shows that disparities in cognitive development in early childhood were associated with school performance, as has been found in other settings.1,22,37,38

The most striking results reported here concern vocabulary acquisition. In the sample, changes in the distribution of overall vocabulary scores between ages 3 and 5 years corresponded remarkably well with the emergence of distinct distributions for children with mothers who had richer and poorer vocabularies. Vocabulary acquisition is a cumulative process, so a child's ability to learn more words depends in good part on his or her existing vocabulary at a given age. In rural Ecuador, it appears that small deficits in vocabulary at early ages beget larger deficits later on.39

An important question is whether the observed parental education and maternal vocabulary gradients in child cognitive development are mediated by child nutritional status or household wealth. Household wealth, parents’ education levels, mothers’ vocabulary, and children's nutritional status are not randomly assigned, so causal inferences can only be drawn with some caution. I showed that the coefficients on maternal schooling and (especially) vocabulary generally change only modestly when variables for child nutritional status and household wealth are included in the regressions. Thus, it does not appear that the main reason for the better performance of children with mothers who have more schooling and richer vocabulary levels is the fact that these children are less poor, or have better nutritional status.

The study has some limitations. The sample was drawn from the lower half of the distribution of wealth in Ecuador, was limited to relatively young families, and only covered rural areas; it is not clear whether the results would hold for a nationally representative sample of children in Ecuador. Because the period covered by the study was also relatively short, it is not clear whether the disparities between children of parents with more or less education, and between those of mothers with more or less vocabulary, remain as children age, and how they affect a larger set of outcomes.

The study's results have important policy implications. First, even in settings where poverty is widespread, as is the case in rural Ecuador, children of mothers with lower education and lower vocabulary levels have worse cognitive outcomes than do other children before they enter school. Arguably, priority for interventions in early childhood should therefore be given to these children, if targeting of interventions on these characteristics is administratively feasible.

Second, the lower levels of development among children with parents who have less education and a more limited vocabulary suggest that these children may receive less early stimulation (e.g., they may be read to less), have home environments that are less nurturing, or have less access to preschool or other programs that may improve child development and school readiness. Programs that seek to increase early stimulation for disadvantaged children may therefore hold considerable promise, as has been found with home-based parenting programs in Jamaica9–11 and preschool in Argentina,40 among other settings.

My analysis leaves many questions for future research. Two are particularly important. The first of these is the extent to which the associations reported represent causal effects of parents’ education and mothers’ vocabulary on children's development, or an unobserved, perhaps genetically determined correlation between the cognitive endowments of parents and children. Answering this question is particularly difficult because it would require an exogenous increase in the education of one group of people but not another, presumably while they are still school aged, and long-term follow-up to assess differences in the cognitive development of their children. The second critical question is the extent to which specific policies and programs can help address the observed deficits in cognitive development, both for children who have not yet entered school and for those who are school aged. Careful evaluation of interventions that improve child nutrition, provide early stimulation to children, or assign children to better teachers are all promising.


Data collection was funded by the World Bank, Princeton University, and the Government of Ecuador.

I have benefited from very useful comments from 3 referees and from Harold Alderman, Maria Caridad Araujo, Jere Behrman, Julian Cristiá, Santiago Cueto, Patrice Engle, Deon Filmer, Sally Grantham-McGregor, Florencia López-Boo, Karen Macours, Hugo Ñopo, Christina Paxson, and Aimee Verdisco. Rodrigo Azuero and Yanira Oviedo provided outstanding research assistance.

Human Participant Protection

This project was approved by the institutional review board at Princeton University, and by a comparable body in the Government of Ecuador.


1. Grantham-McGregor S, Cheung YB, Cueto S, et al. Developmental potential in the first 5 years for children in developing countries. Lancet. 2007;369(9555):6070. Crossref, MedlineGoogle Scholar
2. Case A, Paxson C. Stature and status: height, ability, and labor market outcomes. J Polit Econ. 2008;116(3):499532. Crossref, MedlineGoogle Scholar
3. Anderson MA. Multiple inference and gender differences in the effects of early intervention: a reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects. J Am Stat Assoc. 2008;103(484):14811495. CrossrefGoogle Scholar
4. Currie J. Early childhood education programs. J Econ Perspect. 2001;15(2):213238. CrossrefGoogle Scholar
5. Pollitt E, Gorman KS, Engle PL, et al. Early supplementary feeding and cognition: effects over two decades. Monogr Soc Res Child Dev. 1993;58(7):1118. Crossref, MedlineGoogle Scholar
6. Hoddinott J, Maluccio JA, Behrman JR, Flores R, Martorell R. Effect of a nutrition intervention during early childhood on economic productivity in Guatemalan adults. Lancet. 2008;371(9610):411416. Crossref, MedlineGoogle Scholar
7. Maluccio JA, Hoddinott J, Behrman JR, Martorell R, Quisumbing AR, Stein AD. The impact of improving nutrition during early childhood on education among Guatemalan adults. Econ J. 2009;119:734763. CrossrefGoogle Scholar
8. Behrman JR, Calderón C, Preston S, Hoddinott J, Martorell R, Stein AD. Nutritional supplementation of girls influences the growth of their children: prospective study in Guatemala. Am J Clin Nutr. 2009;90(5):13721379. Crossref, MedlineGoogle Scholar
9. Grantham-McGregor SM, Powell CA, Walker SP, Himes JH. Nutritional supplementation, psychosocial stimulation, and mental development of stunted children: the Jamaican Study. Lancet. 1991;338(8758):15. Crossref, MedlineGoogle Scholar
10. Grantham-McGregor SM, Walker SP, Chang SM, Powell CA. Effects of early childhood supplementation with and without stimulation on later development in stunted Jamaican children. Am J Clin Nutr. 1997;66(2):247253. Crossref, MedlineGoogle Scholar
11. Walker SP, Chang SM, Powell CA, Grantham-McGregor SM. Effects of early psychosocial stimulation and nutritional supplementation on cognition and education in growth-stunted Jamaican children: prospective cohort study. Lancet. 2005;366(9499):18041807. Crossref, MedlineGoogle Scholar
12. Engle P, Black M, Behrman J, et al. Strategies to avoid the loss of developmental potential in more than 200 million children in the developing world. Lancet. 2007;369(9557):229242. Crossref, MedlineGoogle Scholar
13. Brooks-Gunn J, Duncan GJ. The effects of poverty on children. Future Child. 1997;7(2):5571. Crossref, MedlineGoogle Scholar
14. Ardila A, Rosselli M, Matute E, Guajardo S. The influence of parents’ educational level on the development of executive functions. Dev Neuropsychol. 2005;28(1):539560. Crossref, MedlineGoogle Scholar
15. Paxson C, Schady N. Cognitive development among young children in Ecuador: the role of health, wealth and parenting. J Hum Resour. 2007;42(1):4984. CrossrefGoogle Scholar
16. Paxson C, Schady N. Does money matter? The effects of cash transfers on child health and development in rural Ecuador. Econ Dev Cult Change. 2010;59(1):187229. Crossref, MedlineGoogle Scholar
17. Dunn LM, Padilla ER, Lugo DE, Dunn LM. Test de Vocabulario en Imágenes Peabody. Circle Pines, MN: American Guidance Service; 1986. Google Scholar
18. Woodcock RW, Muñoz-Sandoval AF. Batería Woodcock-Muñoz Pruebas de Aprovechamiento-Revisada. Chicago, IL: Riverside Publishing; 1996. Google Scholar
19. Schrank FA, McGrew KS, Ruef ML, Alvarado CG, Muñoz-Sandoval AF, Woodcock RW. Batería III Woodcock-Muñoz: Assessment Service Bulletin Number 1, Overview and Technical Supplement. Chicago, IL: Riverside Publishing; 2005. Google Scholar
20. Schrank FA. Woodcock-Johnson III Normative Update: Assessment Service Bulletin Number 7. Chicago, IL: Riverside Publishing; 2006. Google Scholar
21. McGrew KS. The Cattell-Horn-Carroll Theory of Cognitive Abilities: past, present and future. In: Flanagan DP, Harrison PL, eds. Contemporary Intellectual Assessment: Theories, Tests and Issues. 2nd ed. New York, NY: Guilford; 2005:136182. Google Scholar
22. Stein AD, Behrman JR, DiGirolamo A, et al. Schooling, educational achievement, and cognitive functioning among young Guatemalan adults. Food Nutr Bull. 2005;26(2 suppl. 1):S46S54. Crossref, MedlineGoogle Scholar
23. Mohan P. Inequities in coverage of preventive child health interventions: the Rural Drinking Water Supply Program and the Universal Immunization Program in Rajasthan, India. Am J Public Health. 2005;95(2):241244. LinkGoogle Scholar
24. Filmer D, Pritchett L. Estimating wealth effects without expenditure data—or tears: an application to educational enrollments in states of India. Demography. 2001;38(1):115132. MedlineGoogle Scholar
25. Wagstaff A, Watanabe N. What difference does the choice of SES make in health inequality measurement?Health Econ. 2003;12(10):885890. Crossref, MedlineGoogle Scholar
26. Centers for Disease Control. CDC criteria for anemia for children and childbearing-aged women. MMWR Morb Mortal Wkly Rep. 1989;38(22):400404. MedlineGoogle Scholar
27. Adler NE, Boyce T, Chesney MA, et al. Socioeconomic status and health: the challenge of the gradient. Am Psychol. 1994;49(1):1524. Crossref, MedlineGoogle Scholar
28. Case A, Lubotsky D, Paxson C. Economic status and health in childhood: the origins of the gradient. Am Econ Rev. 2002;92(5):13081334. Crossref, MedlineGoogle Scholar
29. Currie A, Shields MA, Price SW. The child health/family income gradient: evidence from England. J Health Econ. 2007;26(2):213232. Crossref, MedlineGoogle Scholar
30. Currie J. Healthy, wealthy, and wise: socioeconomic status, poor health in childhood, and human capital development. J Econ Lit. 2009;47(1):87122. CrossrefGoogle Scholar
31. Knudsen EI, Heckman JJ, Cameron JL, Shonkoff JP. Economic, neurobiological and behavioral perspectives on building America's workforce. Proc Natl Acad Sci USA. 2006;103(27):1015510162. Crossref, MedlineGoogle Scholar
32. Fox NA, Rutter M. Introduction to the special issue on the effects of early experience on development. Child Dev. 2010;81(1):2327. Crossref, MedlineGoogle Scholar
33. Fox SE, Levitt P, Nelson CA. How the timing and quality of early experiences influence the development of brain architecture. Child Dev. 2010;81(1):2840. Crossref, MedlineGoogle Scholar
34. World Bank. World Development Report: Equity and Development. New York, NY: Oxford University Press; 2006. Google Scholar
35. Heckman JJ. Skill formation and the economics of investing in disadvantaged children. Science. 2006;312(5782):19001902. Crossref, MedlineGoogle Scholar
36. Smith JR, Brooks-Gunn J, Klebanov PK. Consequences of living in poverty for young children's cognitive and verbal ability and early school achievement. In: Duncan J, Brooks-Gunn J, eds. Consequences of Growing up Poor. New York, NY: Russell Sage Foundation; 1997:132189. Google Scholar
37. Sigman M, McDonald MA, Neumann C, Bwibo N. Prediction of cognitive competence in Kenyan children from toddler nutrition, family characteristics and abilities. J Child Psychol Psychiatry. 1991;32(2):307320. Crossref, MedlineGoogle Scholar
38. Victora CG, Barros FC, Lima RC, et al. The Pelotas Birth Cohort Study, Rio Grande do Sul, Brazil, 1982–2001. Cad Saude Publica. 2003;19(5):12411256. Crossref, MedlineGoogle Scholar
39. Carneiro P, Heckman J. Human Capital Policy. Boston, MA: National Bureau of Economic Research; 2003. NBER Working Paper 9495. Google Scholar
40. Berlinski S, Galiani S, Gertler P. 2009. The effect of pre-primary education on primary school performance. J Public Econ. 2009;93(1–2):219234. CrossrefGoogle Scholar


No related items




Norbert Schady, PhDNorbert Schady is with the Inter-American Development Bank, Washington, DC. “Parents’ Education, Mothers’ Vocabulary, and Cognitive Development in Early Childhood: Longitudinal Evidence From Ecuador”, American Journal of Public Health 101, no. 12 (December 1, 2011): pp. 2299-2307.


PMID: 22021308