Data source and selection
Secondary data sets of Indonesian Family Life Survey (IFLS) 1993, 1997, 2000 were downloaded from http://www.rand.org. The survey was conducted by RAND Coorporation and Institute of Demography University of Indonesia at the baseline, supported by the National Institute of Child Health and Human Development, USAID, Ford Foundation and WHO.
The IFLS survey was designed as a longitudinal study and covered 13 out of 27 provinces in 1993, or 83% of Indonesian population representative at both the urban-rural and national level.
Data were collected by trained enumerators. To ensure the quality of the data, supervisors were engaged to check the completeness of the data as well as the logical route of the data every time enumerators completed the field work. The completed data was then sent to the research center for coding, entry and cleaning data done by trained personnels, as has been published elsewhere [26].
Population and samples
The study population was Indonesian households in 1993. The sampling frame refers to Susenas, which is a national frame designed by the Indonesian Central Bureau for Statistics (BPS) to ensure representativeness of data at both the urban-rural and national level. The survey covered 13 out of 27 provinces all over Indonesia in 1993, accounting for 83% of the population of Sumatera, Java, Bali, West Nusa Tenggara and Sulawesi (classified as Celebes in the IFLS study).
The sampling was a two stage cluster sampling design; i.e. stratification at provincial level through ‘PPS’ proportional to population size. Households were randomly selected at the baseline (1993). As many as 60,000 households were covered at the baseline, and 7,730 households were selected based on the criteria. Of the eligible households, 7,039 participated, achieving a high contact rate of 91.1%.
The child population in this study was children born between July 1st, 1991 to 1993, and were followed up to age of 9 years old in 2000. Inclusion criteria were that they had to be the biological children of the mother, were a single birth (not a twin or triplets), were living with the parents, and had information on birth weight and age of pregnancy when delivered. The exclusion criteria was born with genetic abnormalities or developmental disorders .
Study variables
Several new variables were constructed to effectively observe growth patterns at the early postnatal period. As data on the birth length of newborns was unavailable in the data set, a new variable was constructed for children aged 0-2 to address this. LBW could not accurately represent the real birth length as she or he may have been born with a normal weight but short length, or normal length but short and/or skinny. As such, it was deemed necessary to transform the variable. Birth weight was defined as either normal (n=284) or low for less than 2500 grams (n=17). This variable was then merged with the height variable to determine stunting, which resulted in children with normal height (n=168) and children with stunting (n=133). Lastly, this merged data was classified into the results of the height measurement and categorized into children at age 0-2 year who were stunted (n=121) and not stunted (n=180) see graph 1.
Growth at the early age was calculated by creating a new variable which combined data on low birth weight (weight <2500 gram) and stunting (length <48 cm) at birth. At the following ages, growth was indicated by anthropomentric measurements of Z-score. Calculation of the Z-scores were based on body length by age and converting a child’s variables of identification number, gender, age (in months or year), and body length using WHO anthropometry software (WHO AnthroII.PC2007). Stunting was defined as a Z-score less minus two standard deviations (<-2SD) from the median [27].
Infants’ demographic characteristics (i.e., gender, stunting, as well as morbidity at age 0-2 years old); feeding behaviors (i.e. beast-feeding initiation, whether the colostrum was taken, exclusive breast-feeding, age when weaned, the age at starting complementary feeding, quality of complementary feeding); Mother’s characteristics (i.e., height of mother, mother’s education), and the socioeconomic status of the household (household assets, expenditures, hygiene and sanitation of health and environment, rural-urban location). Morbidity was defined as whether or not the children were sick during the month before the interviews.
Breastfeeding initiation was defined as the baby being breast-fed less than or equal to three hours after birth. Exclusive breast-feeding means that no other drinks or food is given to the infant: the infant should feed frequently and for restricted periods. According to the policy during this period all women should be enabled to practice exclusive breastfeeding and all infants should be fed exclusively on breast milk at least 4 months [28]. Complementary feeding refers to the process of the baby consuming other food regularly as well as continuing to be breast-fed. The quality of complementary feeding was defined as the type of food or drinks given to the baby in the past 24 hours, which was converted into food groups including carbohydrates, protein and vitamins.
Mother’s education was defined as formal education completed by mothers. With regard to socio-economic status, variable of household assets was defined as the possession of valuable assets such as land, building, mobile phone, savings, gold and jewels, converted into rupiahs during the field interview. Household expenditure was defined as percent of food expenditure over total household expenditures. The cut-off point of poor and not poor was taken from the Indonesian Central Bureau of Statistics that was 56.86; 55.34 and 65.81 in 1993, 1997 and the 2000. Household hygiene and sanitation was defined as the access of the households to clean water, source of drinking water, availability of toilet, materials of household floor, roof, wall, garbage bins, sewerage and the availability of electricity.
Statistical analysis
Data analysis was undertaken using SPSS version 13.0. At the beginning, the unit of analysis was households with children age of 0-2 years in 1993, and these children were followed up at the age of 4-6 and 7-9 years old. The original sample was 312 children. After the data was cleaned, and missing or extreme values were removed, the number remaining was 301 children. Before the removal of cases with missing or extreme data, an analysis of comparability between these and remaining cases was undertaken to ensure that removal would not affect the outcome variable.
Normality of the distribution of numerical variables was tested using the Kolmogorov-Smirnov test. As the distribution was not normal, a binary binomial categorical variable was constructed by classifying stunted growth if the children had a height for age Z-score of less than -2 standard deviations (HAZ <-2 SD) and as normal growth if HAZ ≥-2 SD. Univariate analysis was used to understand the distribution of the growth-related variables at the age of 0-2, 4-6 and 7-9 years old. Cross-tabulations (Chi-square tests) were used to analyze the correlation of growth at early age (0-2 years old) consecutively to the following age (4-6 and 7-9 years old). Finally, stepwise logistic regressions were run by selecting the final list of variables used among all candidates based on whether they were significant. All variables with p <0.20 were included in the multiple logistic regression using forward selection.