BCS70 response and missingness

BCS70 response

The table below presents statistics about response in BCS70 at every major sweep from birth.

Of the 16,589 cohort members who participated in the first sweep, 2,763 (16.7%) have participated in all 11 major sweeps.

Of all 18,038 cohort members, 11,300 (62.6%) have taken part in at least half the sweeps (ie six or more sweeps).

Table: Participation in BCS70 from birth to 51 years

	Total cohort	Dead	Emigrants	Eligible sample	Participants	(% of eligible sample)
Birth – 1970	16,606	0	0	16,606	16,589	99.9
Age 5 – 1975	16,959^a	567	0	16,392	13,135	80.1
Age 10 – 1980	17,799^a	589	0	17,210	14,870	86.4
Age 16 – 1986	18,038^a	622	0	17,416	11,615	66.7
Age 26 – 1996	18,038	716	35	17,287	9,003	52.1
Age 30 – 2000	18,038	766	235	17,037	11,261	66.1
Age 34 – 2004	18,038	819	432	16,787	9,665	57.6
Age 38 – 2008	18,038	881	456	16,701	8,874	53.1
Age 42 – 2012	18,038	965	433	16,640	9,841	59.1
Age 46 – 2016	18,038	985	466	16,587	8,581	51.7
Age 51 – 2021	18,038	1,047	242	16,749	8,016	47.9

^a The original sample was supplemented by migrants born in 1970.

Figure 1: BCS70 response (as % of eligible sample) over time

Graph showing BCS70 response as percentage of eligible sample over time

Missingness in BCS70

BCS70 variables

We have used a systematic data-driven approach to identify variables that are important predictors of non-response at each BCS70 sweep between age 5 and age 46. Future work will extend this analysis to include non-response at the age 51 sweep.

A list of these variables is provided in the appendix of the Handling missing data in the CLS cohort studies user guide.

These variables can then be considered for inclusion in analyses – for example as auxiliary variable when using multiple imputation – in order to maximise the plausibility of the missing at random (MAR) assumption.

Read more about the process in the preprint: A data-driven approach to address missing data in the 1970 British birth cohort.

Example: restoring the composition of BCS70 at age 46

We have been able to restore the composition of the BCS70 sample at age 46 to be more representative of the study’s target population. We did this by including predictors of non-response at age 46 as auxiliary variables in multiple imputation analyses.

For example, we were able to replicate:

the original distribution of paternal social class observed at the birth survey
the distribution of cognitive ability at age five

Figure 2: Social class of mother’s husband at birth before and after adjustment for missing data

Graph showing social class of mother’s husband at birth before and after adjustment for missing data.

The imputation phase of this analysis included predictors of non-response at age 46 and social class at birth only for cohort members that participated at age 46.

Figure 3: Mean cognitive ability at age 5 before and after adjustment for missing data

Graph showing mean cognitive ability at age 5 before and after adjustment for missing data

The imputation phase of this analysis included predictors of non-response at age 46 and cognitive ability at age 5 only for cohort members that participated at age 46.

BCS70 response and missingness

BCS70 response

Table: Participation in BCS70 from birth to 51 years

Figure 1: BCS70 response (as % of eligible sample) over time

Missingness in BCS70

BCS70 variables

Example: restoring the composition of BCS70 at age 46

Figure 2: Social class of mother’s husband at birth before and after adjustment for missing data

Figure 3: Mean cognitive ability at age 5 before and after adjustment for missing data

Useful documents

Handling missing data

News

CLS Bibliography

Data access & training

Contact us

Follow us