Contents - Previous - Next

This is the old United Nations University website. Visit the new site at

Javanese family models

In the Javanese models (figs. 8.3-8.7), separate versions of the model were estimated using growth, the Vineland social quotient (SQ), and mental development (IQ) as outcome variables. Although these variables could be treated as alternate measures of a more general index of development, it is useful to examine first the size and significance of the paths in the separate models before making a decision on whether they are similar enough to be combined. Descriptions of the measured variables may be found in the Appendix to this chapter.

The latent variables in the models can be divided into three groups. The first group consists of family dynamics variables which we call "family management, beliefs, and caring behaviours": they are Academic Stimulation, Parental Affection, Feeding Practices, and Health Practices. The second group consists of resource variables and includes Social Resources, Material Resources, Support, and Community Endowment. The third group of latent variables represents outcomes such as Growth and Mental Development.

Each path between a pair of latent variables represents a conjecture about the causal relationships between these variables. For example, Social Resources (measured by literacy and mother's years of education) is postulated to affect directly Academic Stimulation and expression of Affection. Material Resources (measured by household possessions and housing quality) directly affects only Feeding Practices and Health Practices.4 If Material Resources affects Academic Stimulation and Affection, we postulate it is through a covariance pathway to Social Resources.5

The indirect pathways between Material Resources and Academic Stimulation and Affection are explained by the assumption that wealthier people are more literate, more educated, and more inclined to stimulate their children. The links between the expression of Affection and Family Resources (Social and Material), however, are less well established. We have little hard evidence that Social and Material Resources affect Parental Affection, although such influences are documented in the cross-cultural literature.

The measured variables that we have available are strongly identified with the latent variables they represent. For example, household possessions and quality of house are strongly related to the Material Resources of the household. The only weak measure is the measure of protein adequacy for the Feeding Practice latent variable in the SQ and IQ models.

Growth model

In the growth model (fig. 8.3), Social Resources are strongly linked to Academic Stimulation; however, Social Resources surprisingly are not linked to the expression of Affection. The insignificant linkage between Social Resource and Affection may reflect cultural norms in Java, or behaviour that was learned by mothers during their own childhood; therefore, there is little variation in the degree of maternal affection with respect to their social status. The degree of Support, however, reflecting the mothers' emotional satisfaction and fathers' involvement in child care, is strongly associated with Affection, and in turn is strongly linked to Growth. Affectionate mothers (measured by warmth and hours spent with children - HrsWkd) had a positive impact on the growth of their children.

Academic Stimulation is not found to affect Growth directly. This result seems plausible for children between the ages of two and five years, and suggests that the pathway linking Academic Stimulation to Growth might not be part of models for older children. Previous findings using ANOVA (Chomitz 1992) show that some components in Academic Stimulation are strong predictors of the child's diet. Parents who stimulate their children are likely to be those who also provide good diets; therefore, the link between Academic Stimulation and Growth is probably through its association with Feeding Practice. This hypothesis was investigated by including a covariance pathway between Academic Stimulation and Feeding Practice, but it showed insignificant association. This insignificant association may be due to the ages of the children in this study. Since only older, more active children are included in this data set, the more stimulated children might also have a higher energy expenditure, so that no direct effect on growth would be expected. Also, Academic Stimulation could be a source of stress in the absence of physical and emotional nurturance and hence could negatively influence growth if the malnourished child underwent pressure to perform. The insignificant association also may be due to the large number of alternate paths of influence included in the structural model.

Fig. 8.3 Path diagram of growth, Java. N=185 2- to 5-year-old children and their families. GFI (adjusted) = 0.668. Ovals: latent variables; rectangles: measured variables; statistical significance of path coefficients: *0.025 < p < 0.050; D 0.010 < p < 0.025;# p < 0.010

Feeding Practice is strongly determined by Material Resources and Community Endowment, and showed a strong effect on Growth. This comes as no surprise, given the use of direct information on the child's diet.

The link between the Health Practice variable (measured by total immunization and vitamin A tablets) and Growth does not have the expected sign and is statistically insignificant. It may be that a simultaneity problem exists between Health Practice and Growth, in which malnourished children were more likely to be exposed to the health centre, and more likely to receive immunizations than healthier children. If this is so, a return pathway from Growth to Health Practice should show a negative sign, and may alter the path from Health Practice to Growth. When the model was estimated with this modification, no real support for the hypothesis was found. Even though Growth showed a negative effect upon Health Practice, the coefficient was not significant, and the effect of Health Practice on Growth remained the same. Since this result is counter-intuitive, it is likely that there is some other error in the specification that explains the findings. A possible non-linear effect may cause the insignificant negative effect of Health Practice on Growth: the nutritionally worse-off children may be brought to the health centre more often than the better-off children. Beyond some point, however, an improvement in nutritional status would not increase the child's presence in the health centres, and as the level of nutritional status is higher, the participation in the nutrition and health programme increases.

The significant negative link between Community Endowment (measured by the community health indicators - crude birth rate and children-under-five mortality rate) and Health Practices is also puzzling and may relate to the negative effect of Health Practice on Growth. It is possible that in villages where the quality of health facilities are low, mothers (or enumerators) tend to overreport the immunization information, or that immunization is conducted through highly memorable campaigns. If this is so, then there is a systematic bias in the data that cannot be eliminated by reformulating the model.

Mental development model

Two observed variables are used to measure the Mental Development construct: these are the Vineland SQ and the Stanford Binet IQ. Before we combined the two variables into a single Mental Development index, we estimated separate models with each variable. Except for the pathway linking Health Practice and Child Developmental outcome, the coefficients estimates are very similar for these two forms of the model (figs. 8.4 and 8.5). This stability supports the belief that the two measured variables represent a single concept, which we call Mental Development. This construct was used in subsequent work.

The relationship between Social Resources and Academic Stimulation (.68#), and Academic Stimulation and Mental Development (.48#) are both strong in the Mental Development model (fig. 8.6). This result conforms to the widely held belief that moreeducated parents provide more academic encouragement to their children, and that children respond with greater performance.

The latent variable, Support, also is found to have an effect on Academic Stimulation (.41#). It seems that stimulation for their children is more likely to come from mothers who are emotionally satisfied and whose husbands are involved in child-care activities.

None of the other pathways into Mental Development appear to be significant in this model. This could be due to misspecification, measured variables that poorly represent the latent variable, too little variation in the data, or too few observations. Affection and Feeding Practices may affect Mental Development in more complex ways than can be represented as direct pathways in the model. Growth could be an intervening factor in the underlying relationships, since it is expected that Growth will affect Mental Development, as discussed in the following section.

Fig. 8.4 Path diagram of IQ, Java. N=185 2- to 5-year-old children and their families. GFI (adjusted) = 0.725. BB Index=0.639. Ovals: latent variables; rectangles: measured variables; statistical significance of path coefficients: *0.025 < p < 0.050; D 0.010 < p < 0.025;# p < 0.010

Fig. 8.5 Path diagram of SQ, Java. N=185 2- to 5-year-old children and their families. GFI (adjusted) = 0.727. BB Index=0.640. Ovals: latent variables; rectangles: measured variables; statistical significance of path coefficients: *0.025 < p < 0.050; D 0.010 < p < 0.025;# p < 0.010

Fig. 8.6 Path diagram of mental development, Java. N=185 2- to 5-year-old children and their families. GFI (adjusted) = 0.731. BB Index=0.639. Ovals: latent variables; rectangles: measured variables; statistical significance of path coefficients: *0.025 < p < 0.050; D 0.010 < p < 0.025;# p < 0.010

Overall child development model

The most complete model included all outcome (child development) indicators. This model was designed to provide the most complete test of the influences of family-level variables on child development.

In building up to this model we started with only one outcome indicator in each step of model testing. This was an attempt to understand the nature of the relationships between family-level variables and each outcome indicator before more complex relationships were explored.

As discussed previously, we found that only Academic Stimulation directly affects both IQ and SQ, or the latent variable Mental Development. The latent variables Affection and Feeding Practice have direct links to Growth. Since the model postulates the influence of Growth on Mental Development, Affection and Feeding Practice may have some indirect influences on Mental Development through their effects on Growth; therefore, pathways linking Growth to IQ and SQ were included.

It also has been argued that Mental Development may affect Growth as well (see Zeitlin, Ghassemi, and Mansour 1990; Myers 1992). To test this conjecture we examined models with a path from Mental Development to Growth, and a covariance between the disturbance terms of the two latent variables: model 1 included both reciprocal influences and a covariance term; model 2 included a covariance term; model 3 included reciprocal influences; model 4 did not include either.

When model 2 and model 3 were tested against model 1, the pvalues were both slightly greater than 0.05, indicating that omitting either the reciprocal influences or the covariance of the disturbances cause very little deterioration in the fit of the model. When model 4 was tested against the other models, the p-values were all greater than 0.05, which indicates that the combined effect of the reciprocal influence and the covariance of the disturbance terms is statistically not significant; therefore, we could omit either reciprocal influence or covariance terms, and eliminate both at the same time. For this reason, the model shown in figure 8.7 does not include covariance terms nor reciprocal influence in older children's development model (see Nigerian model, chapter 8 for explanation).

Fig. 8.7 Path diagram of overall development, Java. N=185 2- to 5-year-old children and their families. GFI (adjusted) = 0.740. BB Index=0.650. Ovals: latent variables; rectangles: measured variables; statistical significance of path coefficients: *0.025 < p < 0.050; D 0.010 < p < 0.025;# p < 0.010

In another variation of the basic model, we treated IQ and SQ as separate variables. This allowed us to see if their relationship to Growth and other outcome indicators will differ. A covariance term was added to take into account the reciprocal relationship between the two variables. This model variant yielded coefficients and significance levels similar to those in the previous models. This model also revealed some interrelationships between the Growth and Mental Development indicators.

Academic Stimulation had a strong effect on IQ, and a less strong effect on SQ. Its effect on SQ differs from the previous model (fig. 8.5). After taking into account the effect of Growth on SQ and the covariance of IQ with SQ, the effect of Academic Stimulation on SQ evaporated. This suggests that Academic Stimulation is affecting SQ through Growth.

Notice that the relationships between Affection and Growth (.28 # ), Growth and IQ (.17 D ), and Growth and SQ (.21#), are all strong; in fact, the two-link pathways from Affection to IQ and SQ through Growth are about .05, respectively (.28 x .17 and .28 x .21). This is in contrast to the finding in the Mental Development model (fig. 8.6) in which the effect of maternal Affection was negative and insignificant. This model confirms our expectations about the role of Growth as an intermediary element connecting Maternal Affection (also Feeding Practice) to child Mental Development.

This model again shows the strong effect of the latent variable Support on Academic Stimulation and Maternal Affection. It clearly indicates that a stable family environment, as measured by maternal emotional satisfaction and paternal involvement in child-rearing activities, is beneficial in producing positive child development.

We failed to find an association between the Health Practices variable and the child outcome indicators. As mentioned previously, this may be due to bias in the data-gathering process. Establishing precise estimates of this linkage may require more information on Health Practices and better methods of gathering that information.

We conclude that after controlling for socio-economic status (SES), Maternal Affection and Feeding Practice positively affect Growth. Given the interrelationships between Growth and Mental Development indicators, Maternal Affection and Feeding Practice also positively affect mental development. Academic Stimulation showed a direct and positive effect on IQ in older children; however, its effects on SQ and Growth are much less certain. It appears that a stable family environment helps to generate positive maternal attitudes towards the children, which in turn promotes positive child development.

Total effects of indicator variables

Tables 8.1-8.4 present a summary of the findings from the model by showing the total effects of each latent variable that could be considered an "input" on the variables that could be considered an "output." The total effects summarize the individual effects of all possible pathways connecting the input indicators to the output indicators. Since the entries are scaled in standardized units, they show the relative strengths of the influences of the input variables on the output indicators and intervening variables.

Table 8.1 Total effects in Javanese growth model

  Stimulation Affection Feeding Health Growth
Social .191 .006 - - - .053
Material - - .347 .225 .146
Stimulation - - - - - .294
Affection - - - - .445
Feeding practices - - - - .451
Health practices - - - - - .045
Support .609 2.061 - - .737
Community endowment - - .365 - .545 .189

Table 8.2 Total effects in Javanese mental development model

  Stimulation Affection Feeding Health MDa
Social .275 - .015 - - .248
Material - - .057 .206 .009
Stimulation - - - - .893
Affection - - - - - .216
Feeding practices - - - - .275
Health practices - - - - - .035
Support .128 .384 - - .032
Community endowment - - .075 - .579 .041

a. Mental development (IQ and SQ).

Table 8.3 Total effects in Javanese overall child development model

  Stimulation Affection Feeding Growth IQ SQ
Social .230 - .025 - - .008 .198 .095
Material - - .076 .079 .021 .024
Stimulation - - - - .871 .427
Affection - - - .335 .089 .101
Feeding - - - 1.248 .332 .337
Health - - - - .080 - .021 - .024
Support .117 .409 - .137 .139 .092
Community endowment - - .099 .167 .044 .051

Table 8.4 Legend for Javanese models

Variable code Description
MOMED Mother's education
LITERACY Reading and writing Bahasa Indonesian
SOCNET Mother's involvement in social organization
HOUSING Housing quality
TOTITEM Total household possessions
TEACH Learning and teaching environment
STIMUL Academic stimulation and physical environment
AFFECT Maternal affection
HRSWKD Length of time a mother spent with her child per day
PROTCAL Child's protein adequacy
VITA Child's vitamin A adequacy
TOTIMM Total immunization score
VITASUP Vitamin A tablets
EMOSUP Mother's emotional satisfaction with family support
FATHER Father's involvement in child care
IQ Stanford Binet Index
SQ Social Quotient Index
WAZ Weight-for-age Z-score
HAZ Height-for-age Z-score

The coefficients of most variables are relatively stable from one model to the next. One exception was the impact of Affection on Mental Development: Affection was negative in affecting Mental Development (table 8.2); in the overall model, however (table 8.3), the effect of affection was positive. This provides an example of how ignoring the indirect effects of a variable through other variables may give a very different picture of its overall effect. One of the benefits of using structural equation analysis is that we can obtain an estimate of the indirect effect of a variable, which cannot be done by using a reduced-form model for each variable of interest.

Social (literacy and years of education) and Material Resources are usually the strongest predictors of Growth and other child developmental outcomes in other studies that do not control for the effect of family dynamics variables. When some intervening proximate "family management and caring behaviour" factors are added to the model (table 8.3), Stimulation, Affection, and Feeding Practice emerge as the dominant factors. This demonstrates the importance of the dynamics within the family, a point often overlooked by those interested in child welfare.

Contents - Previous - Next