Objective: The aim of the study was to report the association of socio-economic factors with child's height.

Design: Cross-sectional study based on a representative national sample of government schools.

Setting: Trinidad and Tobago in 1999.

Subjects: A total of 2608 boys and girls mean age 5.8 y, range 4.38–6.99 y and 3080 mean age 8.6 y, range 7.00–10.44 y olds.

Outcome: Measurement of height and a questionnaire completed by parents. In the analysis height was expressed as standard deviation scores (s.d.s.) based on the British height curves (1990) or height below −1.5 s.d.s.

Results: Ethnicity, parental heights, birthweight, maternal age at child's birth and number of children in the family were the main factors associated with children's height. Lack of piped water supply in the home was the only socio-economic factor consistently associated with height (mean difference in s.d.s. adjusted only for age group, gender and ethnicity −0.192, 95% CI −0.257 to −0.127 and in addition adjusted for the variables listed above −0.080, 95% CI −0.141 to −0.019). Parental education, household overcrowding and employment status were weakly associated with height in the partially adjusted model only. Analysis of severe growth failure gave similar results.

Conclusion: The impact of socio-economic factors on height is marginal in Trinidad and Tobago. As socio-economic factors may have an impact on a broad range of health indicators, height and rates of undernutrition should not be used as sole criteria for assessing progress in decreasing health differentials caused by social inequalities.

Sponsorship: None.

‘Growth as a mirror of society’ was proposed by Tanner (1986) to convey the idea that height, in particular, was a good marker of social inequality in a community. A lack of variation in height between social classes in Swedish children was interpreted as indicating that the society had reached a high degree of social development and its individuals had reasonable access to the countries' wealth (Lindgren, 1976). Since that paper was published several cross-sectional studies have shown only small differences or no differences in height according to social stratification in children and young adults in many European countries (Gulliford et al 1991; Rona, 2000). Exceptions to this general trend come from less developed countries in the region such as Poland, Portugal and Turkey (Belicki & Szklarska, 1999; Nebigil et al, 1997; Padez & Johnston, 1999). Analyses using an ecological approach (Reading et al, 1993) or a case–control design (Voss et al, 1998) have shown an important contribution of social factors to the variation of height. The message from these studies may indicate that children's height is associated with socio-economic variables but socio-economic differences have greatly diminished over time. If adjustment is made for confounding variables, especially for the child's birthweight, parental heights and number of children in the family, socio-economic differences in height are negligible in Britain and other Northern European countries (Gulliford et al, 1991; Herngreen et al, 1994). However, it is possible that serious failure to thrive linked to social inequalities can still be found in a small minority of children in developed countries.

The relationship between children's height and socio-economic variables in middle-income countries based on representative national samples has rarely been documented. A survey using a representative sample was carried out long ago in Cuba (Jordan et al, 1975), but the findings were not analysed in relation to social variables. In China five large scale cross-sectional surveys of growth were carried out contrasting rural and urban areas (Shen et al, 1996) and in Mexico a national study of preschool children was carried out in 1988 (Hernandez-Diaz et al, 1999). Other studies have explored social issues in relation to height using a convenience sample or linked to a group in the community (Amigo et al, 2001). Many middle-income countries have experienced profound changes in economic and health conditions over relatively short time periods. In the Caribbean under-nutrition in children was an important public health problem two to three decades ago (Sinha, 1988). Now rising standards of living have resulted in obesity being more important than under-nutrition. In Trinidad and Tobago the per capita GNP was estimated as US$4520 in 1998 (United Nations Development Programme, 2000). Based on data from the 1981 Household Budgetary Survey in Trinidad and Tobago, the GINI coefficient, an indicator of inequality in income distribution, was estimated to be 40.3, indicative of moderate to high income inequality (World Bank, 2001).

The purpose of this report is to analyse children's height in relation to indicators of parental education and employment, household composition and amenities in a representative sample of government primary schools in Trinidad and Tobago. We aimed to determine whether there were socio-economic variations in children's height after adjusting for confounding variables.

Subjects

We carried out a cross-sectional survey of school children in Trinidad and Tobago. There are 468 government primary schools in Trinidad and Tobago, 433 in Trinidad and 35 in Tobago. A sample of 66 schools was drawn stratified by health administrative areas in the country and randomly selecting schools with probability proportional to size (Gulliford et al, 2001). The sample of schools was drawn by the Central Statistical Office for an earlier survey carried out in 1989. The same sample was used for this survey because the geographical distribution of children in that survey corresponded closely to the distribution observed in the 1990 census. Fieldwork was carried out in the first 6 months of 1999, using the methods of the National Study of Heath and Growth (Rona & Altman, 1977). Within each school we measured all children in the first year classes (mean age 5.8 y, range 4.38–6.99 y) and in the classes for children aged 8–9 y (mean age 8.6 y, range 7.00–10.44 y). Further details were reported elsewhere (Gulliford et al, 2001).

Measurements

Height was measured, to the last 0.1 cm on a Holtain stadiometer using the method described by Cameron (1986) and 0.05 cm was added to correct the bias. Fieldwork was carried out by the nutritionists and food demonstrators from the Nutrition Division of the Ministry of Health. They were trained in measurement techniques before the start of the study, but measurement error was not specifically quantified.

Questionnaires

The parents of each child were asked to complete a questionnaire. Where necessary the questionnaire was interview-administered by a class teacher or a fieldworker. For the present analyses the following items were used: child's ethnicity, parents' self-reported height, child's birthweight, maternal age at child's birth, the number of children in the family, mother's education and employment, paternal education and employment, grandparents living at home, water supply at home and household overcrowding. We also included initially birth order in the analysis, but as it did not add further information when number of children was included in the model it was omitted in subsequent analyses. Continuous variables were divided into quartiles and adding a ‘not known’ category as a fifth group to minimize losses in the analyses. The variables were categorized as shown in Tables 1 and 2. Ethnicity was classified using a shortened form of the groupings used in the Trinidad and Tobago national census (Central Statistical Office, Trinidad and Tobago, 1994). The mother's age at the child's birth was estimated from the mother's year of birth and the child's age.

Analysis

In order to analyse data for children from more than one age group it was necessary to standardize measurements for age and gender. As measurements increase in their variation with age, as well as mean value, a standard deviation score (s.d.s.) was calculated for each measurement. The s.d.s. for height is given by the difference between the child's measurement and the mean for a child of the same age and gender from a reference population divided by the standard deviation for that age and sex in the reference population. The British (1990) height curves were used for reference (Freeman et al, 1995). By definition each s.d.s. had a normal distribution mean 0.0 and standard deviation 1.0 for the English white 1990 population.

Extreme values for s.d.s. (greater than 10 s.d.s. or less than −10 s.d.s.) were examined and s.d.s. and associated measurements set to missing where these appeared impossible. Eight items of data were excluded in this way. A conservative approach was taken to allow the possibility that some extreme departures from the UK reference data would occur. Children were divided into two age groups, less than 7 y and 7 y or more, in order to examine the hypotheses in relation to age. Random effects models (with school as the random effect) were used to estimate associations between height and explanatory variables. Models were fitted using the ‘xtreg’ command in Stata with the maximum likelihood option (Stata Corporation, 1999). Multiple logistic regression analyses with height dichotomized using the cut-off ≤−1.5 s.d.s., equivalent to centile 6.7 of the British height curves (1990), (Freeman et al, 1995), were also carried out. Robust standard errors were estimated to allow for clustering by school. The purpose of this second analysis was to assess possible causes of moderate and severe short stature that could attract medical and social care provision.

The sample characteristics by ethnicity have been published recently (Gulliford et al, 2001). Children from the Indo-Trinidadian group were lighter at birth, mothers in that group were younger at child's birth, had lower body mass index (BMI), lower education level, were less likely to be in paid employment and the household had a higher overcrowding level, but it was more likely that a child was part of a two-parent family than in the Afro-Trinidadian group. Table 1 shows the association between biological variables and height adjusted for all other variables in the table. Parents' heights, child's birthweight and number of children in the family were markedly associated with height (P<0.001). Maternal age at child's birth also had a linear association with height (P<0.001), but the association was less marked than for the other variables in the model. Afro-Trinidadian children were taller than Indo-Trinidadian children and the mixed group was intermediate (P<0.001).

In contrast to the biological variables, factors related to socio-economic background were only slightly associated with height in the partially adjusted model (Table 2). Thus within these models the level of association between overcrowding was equivalent to the association between mother's age at child's birth and child's height after full adjustment shown in the previous table. The association of the other independent variables and height in the analyses based on partial adjustment showed a difference of less than 0.2 s.d.s., equivalent to approximately 1 cm. With the exception of water supply at home (P=0.03), none of the socio-economic variables was associated with child's height in the fully adjusted model (Table 2).

The associations between short stature (≤−1.5 s.d.s.) and biological and demographic factors were similar to those given in Table 1 (not shown). Short stature was more frequent in children who were Indo-Trinidadian (P<0.001), had parents of short stature (P<0.001), had low birthweights (P<0.001) or were from families with a large number of children (P<0.001), but was insignificantly associated with mother's age at child's birth. There was evidence that lack of piped water supply at home was associated with short stature (P<0.001; Table 3). There was little evidence of an association with overcrowding (P=0.41) and parental unemployment (P=0.28). There was an unexpected result showing that children whose parents had a university education had greater risk for having short stature than other children. However, only 5% of the parents in the total sample had university education.

The large differences in mean height by socio-economic factors shown at age 7 in several developing countries in the 1980s (Martorell & Habicht, 1986) were not confirmed in our studies. They showed differences that varied from 4 to 12 cm between low and high socio-economic status in eight countries, including Jamaica. The standard of living in Trinidad and Tobago is higher than the countries in that report, including Jamaica. In our analysis the variables associated with height in pre-adolescent Trinidadian children were very similar to those shown in Britain (Gulliford et al, 1991). Parental height, child's birthweight, ethnic background and number of children in the family were the variables most markedly related to height in the analysis using height as a continuous variable. Only piped water supply, among the socio-economic variables in the analysis, was significantly associated with height after adjustment for other factors. Other social factors were associated with height before adjustment, but the size of differences in mean height was small, between 1.0 and 1.5 cm.

We expected that an analysis assessing factors associated with extreme short stature would be more fruitful in showing associations with social factors, as others have reported an impact of socio-economic factors at the lower end of the distribution (Guimaraes et al, 1999; Voss et al, 1998). Such a finding would be important because it would have indicated that severe malnutrition associated with poverty is still in existence in Trinidad and Tobago. However, the results were remarkably similar in the dichotomized analysis and the analysis using height as a continuous variable. We used as cut-off point for such analysis −1.5 s.d.s. instead of the conventional −2.0 s.d.s. because of the low frequency of children with a height below −2.0 s.d.s. in Trinidad and Tobago. A cut-off above −2.0 s.d.s. has been also used in another study in Chile for the same reason (Amigo et al, 2001).

The strengths of our study are that it is based on a national representative sample of government primary schools, the participation rate was above 90% and the measurement techniques were standardized. The only minor problem in the study was that quality control of measurements was not assessed during the study. This may have produced bias. However, as that deficiency is common to other studies it does not explain the lack of impact of socio-economic factors on height in comparison to other studies (Martorell & Habicht, 1986; Guimaraes et al, 1999; Amigo et al, 2001).

We need to discuss whether our results can be explained by over adjustment of social factors in the analysis as some factors such as parental height, child's birthweight and number of children may be also reflecting socio-economic circumstances. It is also important to assess whether number of children in the family is a proxy measure of socio-economic circumstances.

Adjustment for parental height, ethnicity, birthweight and number of children in the family is an attempt to account for the effects of genes and maternal environment in uterus. Giving results before and after adjustment provides a manner for assessing the possible over-adjustment or, conversely, the possible under-adjustment of excluding the biological and demographic factors in the analysis. In our analysis even before adjustment, the association between height and social factors was slight. It is, however, telling that a factor such as level of education, a key indicator of social deprivation and social exclusion, was not associated with height in a country in which 30% of the population do not reach secondary education. It is clear that lack of piped water supply is still associated with height, but residual confounding as an explanation of the association cannot be excluded. If the association is causal, repeated infections may be a possible explanation in rural areas (Stephensen, 1999). There is evidence that diarrhoea may affect growth (Torres et al, 2000) and the level of helminth infection may be high in Caribbean countries (Wong et al, 1994). It has been reported that antihelminth treatment does not affect growth (Northrop-Clewes et al, 2001), but it may benefit performance in children of poor nutritional status (Simeon et al, 1995). The other social variables, such as household overcrowding and unemployment, are unlikely to have an effect on height, but if they do it would be at most marginal.

Number of children in the family may be a proxy measure of socio-economic level and also provides some information on environment in utero. Birth order instead of number of children would be preferable for assessing environment in utero (Dowding, 1981). In our study the inclusion of birth order was not associated with height when number of children was included in the model. As the two variables are highly associated, and birthweight was also included in the analysis, we decided to omit birth order. In our study, low parental education, lack of piped water supply and father's unemployment were more common in families with a large number of children. It is not axiomatic that number of children is inevitably related, as a socio-economic factor, to growth, as in the NSHG in Britain height was associated with size of sibship in white, but not in Afro-Caribbean children, or in those originating in the Sub-Indian continent (Gulliford et al, 1991). Thus number of children in the family is a multifaceted variable in which its association, or absence of an association, with height may depend on the cultural characteristics related of subsections of society.

Although ethnicity is a powerful genetic trait influencing height, in many countries social stratification is also highly associated with race and ethnicity. In Trinidad and Tobago race used to be an important factor for social stratification during the colonial era, but it has become less so since independence (Gulliford & Mahabir, 2001). Those of Indian descent, representing the most recent group, tended to have lower socio-economic status, but this difference has been disappearing over time. Children of West African descent are the tallest in ethnic groups in the UK and the USA, despite being relatively poorer than other ethnic groups in these countries (Chinn et al, 1998; Garn et al, 1973). Thus we would expect that the differences in height between ethnic groups prevailing in Trinidad and Tobago would not be accounted for by environmental factors.

Mueller (1976) showed that parent–offspring height correlations were higher in developed countries than in less developed countries. He hypothesized that the parent–offspring correlations were higher in those countries with a higher level of stability of the environment over time. In the NSHG the midparent–child correlation was 0.56 while in the Trinidad and Tobago study it was 0.36. Following Mueller's interpretation of international data, this may indicate that the socio-economic conditions have changed over the period of one generation more markedly in Trinidad and Tobago than in Britain in the last 30 y.

In conclusion the main factors associated with child's height in Trinidad and Tobago are not dissimilar from those in Britain. Socio-economic variables are not important in explaining height variations. The impact of the most relevant social factors on growth is only marginal. With the exception of lack of piped water supply, severe failure to thrive was not associated with socio-economic factors. In our view the paradigm suggested by Tanner (1986) that ‘growth was a mirror of society’ may not be satisfactory for assessing the effects of social inequalities in countries of intermediate socio-economic development. Instead, we suggest that the assessment of social inequalities in health should be based on a broad range of measurements and indicators.

The authors thank the Chief Medical Officer of Trinidad and Tobago for permission to report this work. They also thank the Principal Medical Officer (Community Services), the Ministry of Education, and the staff, pupils and parents of the participating schools for their support of the survey. Special thanks are due to the staff of the Nutrition Division for their skill and dedication in working on the survey.

Author notes

1. • D Mahabir
   •  & M C Gulliford

   Guarantor: MC Gulliford and D Mahabir.

2. • R J Rona
   • , D Mahabir
   • , B Rocke
   • , S Chinn
   •  & M C Gulliford

   Contributors: All authors participated in the design of the study. DM, BR and MCG organized and monitored the data collection. SC calculated height standard deviation scores (s.d.s.). RJR and MCG discussed the analysis and MCG carried out the analysis. RJR wrote the paper and all the contributors commented and approved the final version of the paper.

Affiliations

1. Department of Public Health Sciences, King's College London, London, UK

   • R J Rona
   • , S Chinn
   •  & M C Gulliford

2. Nutrition and Metabolism Division, Ministry of Health, Trinidad and Tobago

   • D Mahabir
   •  & B Rocke

Corresponding author

Correspondence to R J Rona.

To obtain permission to re-use content from this article visit RightsLink.

Publication history

DOI

https://doi.org/10.1038/sj.ejcn.1601508