BCS70 response and missingness

Find out more about the level of response for every sweep of the 1970 British Cohort Study (BCS70), important predictors of non-response, and how to handle missing data in the study.

As of August 2026:

• 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).

BCS70 response (1970-2021)

Fig 1: BCS70 response over time

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	16,959a	567	0	16,392	13,135	80.1
Age 10 – 1980	17,799a	589	0	17,210	14,870	86.4
Age 16 – 1986	18,038a	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.

Missingness in BCS70

Missing data are common in surveys and cohort studies. To tackle this in BCS70, we have used a systematic data-driven approach to identify variables that are important predictors of non-response at each sweep.

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.

Find a list of these variables for BCS70 sweeps between age 5 and age 46 in the appendix of the Handling missing data in the CLS cohort studies user guide.

Example: restoring the composition of BCS70 at age 46

We were 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

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

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.

Resources