Evidence on birth spacing and child cognition from Indonesia

ABSTRACT

Research from the medical and social science literature has found that spacing births closely together leads to adverse health and socioeconomic outcomes for the younger child. More recent evidence using inter-family comparisons is challenging the validity of those results. However, while inter-family comparisons allow for time-invariant unobservable heterogeneity across families, they cannot fully control for changes in families that occur over time and may affect both decisions about birth intervals as well as child outcomes. To our knowledge, this paper provides the first examination of the effects of birth spacing on later childhood cognitive outcomes that accounts for both time-invariant and time-varying heterogeneity in the same models. Instrumental variables estimations on data from Indonesia indicate that younger siblings perform significantly better on maths assessment exams when they are spaced farther apart from their immediately older sibling. This effect appears to be concentrated among the poorest and least educated households. The data thus suggest that close spacing may be detrimental to the socioeconomic outcomes of the younger sibling.

KEYWORDS:

• Birth spacing
• cognition
• later childhood
• family structure

JEL CLASSIFICATION:

• J13
• J12
• O15

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Month and day of birth are missing for a portion of birth outcomes; therefore, we are limited to using birth year and pregnancy order to define birth order and spacing variables.

² Specifically, we drop any mother who has more than three miscarriage or stillbirth events combined. We further allow a maximum of only two miscarriages or two stillbirths. For example, our sample would include a mother with two miscarriages and one stillbirth, but not a mother with three miscarriages.

³ The spatial assessment questions are similar in format to those used in typical tests of spatial intelligence. The actual assessment instrument can be downloaded at http://www.rand.org/labor/FLS/IFLS/ifls4.html.

⁴ In previous survey waves, 15–24 year olds refused to take the assessments at higher rates. In order to lower those refusal rates, the spatial assessment was shortened for that age group starting with the 2000 wave.

⁵ The numbers of older and younger siblings do not precisely match. While we know birth order and spacing for all individuals, not all siblings took cognitive assessments. For example, we may know older sibling A and younger sibling B’s spacing, but only have cognitive test scores for A. Sibling A would thus be included in regressions for older siblings, but sibling B could not be included in regressions for younger siblings.

⁶ For example, if the family has a second child with serious health issues, they may choose to delay having a third child longer than anticipated. Once the third child is finally born, the continuing health needs of the second child may also prevent the parents from devoting many resources to the third child, thereby negatively impacting their cognitive development. This change in birth intervals due to the health endowment of the second child will not be captured by the fixed effects and would lead to an underestimate of the spacing effect.

⁷ It is important to note that most miscarriages occur so early in pregnancy that they occur before a woman would choose to have an abortion (Hotz et al. 2005).

⁸ For purposes of estimation, these miscarriages include all miscarriages and stillbirths occurring to mothers of the siblings in our sample.

⁹ Due to the extensive set of control variables included, the sample size for Table 2 is smaller than the samples reported in Table 3. Mother-year fixed effects in Table 3 removes the need to explicitly control for those variables.

¹⁰ Demographic characteristics for households in our fixed-effects sample (households with more than two children) are similar to those for households in the full dataset. Comparison summary statistics are available upon request of the authors.

¹¹ Household assets are valued at the initial survey wave in which they are present. In our sample, the median household reports an initial total asset value of 60,700,000 rupiah.

¹² If we instead define highly educated households as those in which the mother attends junior high and above, the magnitude and significance of our effect for less educated households becomes even larger. Our results in Table 6 can thus be considered a conservative estimate. Results available upon request.