Sibship size, family structure and educational outcomes

Tak Wing Chan, Morag Henderson and Rachel Stuchbury, UCL

(Project no. 0301702, previously 30170)

The aim is to estimate sibship size (from household composition) and whether both parents were present at two successive censuses, when the LS member is aged 4-7 and 14-17. S/he will be followed up at age 24-27 to see whether broad educational achievements are associated with those characteristics from earlier life.  Three successive cohorts will be studied.

We plan to use ONS LS data from 1971 through 2011 to explore whether the sibship size of LS members has a causal effect on their educational outcome. Our strategy is to extract household information of LS members (number of siblings, whether they lived with both parents up to mid-teens, parental class and education, housing tenure, etc) when they were aged 4-7 and 14-17, and link this to their educational outcome when they were aged 24-27.  By using twin births as an instrumental variable, we seek to determine whether number of siblings really leads to lower educational outcome. And by comparing LS members who were 4-7 in 1971, 1981 and 1991 respectively, we seek to determine whether any causal effect of family size has changed over time.

We hope that permission will be granted to use the exact date of birth and sex of other household members. We intend to write STATA syntax (to be run by ONS staff) to derive three non-disclosive indicators (one for each cohort). Each indicator will show whether a non-member at the first census (when LSM is aged 4-7) can be matched with a non-member at the second census, and vice versa. Values will be along the lines of:

1 Sex and complete dob match

2 Dob matches, sex differs

3 Sex matches, day or month of birth differs

4 Sex matches, year of birth differs

5 Other form of match or none 

The aim of the exercise is to estimate complete sibship size for the LS member (including siblings not present, or not born, at one census); and also to see whether any parents identified at the first census are still present at the second. The process will generate a number of false positives which we will have to clean up.

Another variable referred to above will indicate whether the LS member has a twin or triplet in the household; this will be estimated by comparing year of birth and age at census for the LS member with those of other household members at the same census, and will therefore not require restricted data.

Page last modified on 28 jun 16 10:44 by Joanne Tomlinson