Survey response rates and consent rates
At the 23rd sweep of data collection on the NSHD, 85 per cent (N = 2,661) of the eligible sample provided information. The majority (N = 2,229) were interviewed and examined at a clinical research facility or in their own homes by research nurses, with others completing only the postal questionnaire (N = 432) [
11]. Out of the 2,661 who either had a clinic/home visit or completed a postal questionnaire, 2,505 (94 per cent) agreed to provide consent to GP/hospital records; 141 respondents (5 per cent) did not sign the consent form and 15 (<1 per cent) refused to give their consent. Out of the 2,229 cohort members who either had a clinic/home visit, only 1 person refused to allow access to GP/hospital records, representing a near 100 per cent consent rate. By contrast, among the 432 cohort members who only supplied a postal questionnaire, the consent rate was 64.1 per cent. Seeing as non-consent and non-participation in the health assessment is perfectly collinear, it is not possible to analyse selectivity in consent in our comparative analysis framework. Therefore, the NSHD is dropped from the analyses in this paper.
At the 8th sweep of data collection on the NCDS, 11,461 cohort members were eligible for interview. Of these, 9,758 (85.1 per cent) provided a fully productive interview. All interviewees were asked for consent to link to health records. From them 7,681 (79 per cent) consented to link to health records; of the non-consenters, 1,148 respondents (12 per cent) did not return the consent form and 873 (9 per cent) returned the form but did not give their consent.
At the 18th sweep of data collection on the BHPS, 5,483 households were eligible for interview. Of these, 4,509 (84.2 per cent) participated in the study, providing 11,272 fully productive interviews with adults. All interviewees were asked for consent to link to health records, and 4,568 (40.5 per cent) agreed to data linkage.
The UKHLS started with 55,436 households that were eligible for interview in the 1st sweep of data collection. Of these, 31,346 (56.5 per cent) participated in the study, providing 45,735 fully productive interviews with adults. All interviewees were asked for consent to link to health records, and 32,618 (68.3 per cent) gave consent.
In summary, participation rates (i.e., survey response rates) in the three long-running studies NSHD, NCDS and BHPS were at the same level, amounting to around 85 per cent. At 56.5 per cent the rate was considerable lower in the UKHLS. However, this was the first wave of the UKHLS and it is a common feature of longitudinal samples that participation rates at earlier waves of the study are considerably lower than later on in the life of the panel [
12,
13]
f. For survey methodologists this raises interesting questions over when in the life of the panel it is best to ask for consent [
14].
In our classification the NSHD is the most health-focussed, with health and development explicitly part of the title of the study, and funded directly by the MRC. The NCDS is the second most health-focused study, having “child development” in its title and having collected, particularly at the early sweeps, a great deal of medical and health information from the family of the cohort member. The UKHLS and BHPS, funded by the Economic and Social Research Council (ESRC), are ‘branded’ as social science studies and have emphasised the broad range of topics from the beginning of the studies. The UKHLS, of the two household panel surveys, is branded as a “bio-social” survey, with the aim of adding detailed biomarker data to the broad social and economic data collected in the survey. In line with our hypothesis, the overall consent rates in the four studies suggest that the more health-focused cohort studies have higher consent rates than the two general population-based studies. Moreover, within study types it is the more health-focused NSHD (99.9 per cent) that achieved higher consent rate than the less health-focused NCDS (79 per cent), and the same pattern holds for the household panel studies with the more health-focused UKHLS (69.0 per cent) achieving a higher consent rate than the BHPS (41 per cent). Note that we restrict analysis of the UKHLS and BHPS to GB to make a fair comparison between the studies; preliminary results for the UK suggest that consent is much lower in Northern Ireland.
For a description of the samples with respect to all variables used in the analysis, see Additional file
1. Note that we did not include the NSHD in this overview; the data are considered representative of the population aged 60 living in Britain [
11], and everybody consented. In other words, there are no health biases in consent to analyse.
Bivariate associations with consent
For reports of bivariate associations with consent for our independent variables, see Additional file
2. These are split into three blocks, (1) socio-demographic, (2) socio-economic characteristics, and (3) markers of health and related behaviours such as smoking and using health services. Results for the household panel studies consider the complex survey design (which includes oversampling, clustering and stratification) and are calibrated to the population living in Britain in 1991 (BHPS) and 2009 (UKHLS) using the appropriate population weights provided in the studies.
The results suggest that there are some statistically significant associations within all three sets of characteristics in all three studies but there is little evidence for a systematic pattern. This finding itself is interesting since it echoes the general lack of consistent socio-demographic effects found in previous consent research. For example, among the socio-demographic characteristics, only being a member of the British/Irish White population is positively associated with consent in all studies with minority ethnic group members being less likely to consent. Given the larger sample size and the incorporated ethnic minority boost sample, the UKHLS afforded the opportunity to investigate whether the effect is driven by any of the minority ethnic groups in particular. The results suggest that all ethnic groups except White Irish and Mixed groups were less likely to consent than British/Irish White (see Additional file
3). Consent was particularly low among Pakistani and Bangladeshi. This result warrants further investigation in a separate analysis outside our comparative analysis framework.
Results for the household panel studies suggest that those aged 16–24 and those aged 50–52 are more likely to consent. Smaller households, i.e., people living by themselves or with no children have lower consent rates on the NCDS but higher consent rates on the UKHLS. Whilst education is associated with consent at all education levels in the UKHLS except A-Level, there is no association in the NCDS (except no qualifications) and in the BHPS, only those with A-level as their highest qualification are less, and those with or a degree or higher are more, likely to consent; ‘having no qualifications’ is associated with lower consent across all three studies.
A similar (non-)pattern is shown with socio-economic characteristics. Individual gross earnings are associated with consent in the NCDS, with the bottom two quartiles being less likely to consent and the top two quartiles more likely to consent, and there is an association in the bottom quartile (less likely to consent) and third quartile (more likely to consent) on the UKHLS; on the BHPS, there is no association.
Last, but not least, there is only one health-related characteristic that is associated with consent across all three studies, and that is being overweight (or, using our alternative marker, being in the top quartile of the Body Mass Index (BMI)). Having reported a health problem is also positively associated with consent in two studies (NCDS and UKHLS). As to all other characteristics, the associations are either not statistically significant in at least one of the studies, or the direction of the effect differs across studies.
Multivariate regression analysis
Table
2 reports conventional coefficients from multivariate logistic regression models which include explanatory variables available across the NCDS, BHPS and UKHLS
g.
Table 2
Logistic regressions on consent to health data linkage: Beta-coefficients
England | 0.01 | 0.08 | 0.22 | 0.16 | -0.11 | 0.07 |
London/SE | -0.14* | 0.06 | 0.16 | 0.11 | -0.10* | 0.05 |
Male | -0.02 | 0.06 | 0.13* | 0.05 | 0.09*** | 0.03 |
British/Irish White | 0.33* | 0.13 | 0.74*** | 0.13 | 0.47*** | 0.05 |
Aged 50-52 | | | 0.22 | 0.14 | 0.10 | 0.06 |
Number of own children in the household (ref: none) | | | | | | |
1
| -0.10 | 0.07 | -0.02 | 0.12 | 0.11* | 0.05 |
2
| -0.04 | 0.07 | -0.12 | 0.12 | 0.08 | 0.05 |
3 or more
| -0.09 | 0.10 | 0.24 | 0.18 | 0.05 | 0.08 |
Lives alone | 0.19 | 0.10 | 0.12 | 0.10 | -0.13** | 0.04 |
Highest degree (ref: higher degree) | | | | | | |
First degree
| -0.13 | 0.16 | -0.41* | 0.19 | 0.06 | 0.06 |
Diploma
| -0.19 | 0.19 | -0.56** | 0.19 | 0.15* | 0.06 |
A-level
| -0.20 | 0.17 | -0.45* | 0.21 | 0.19** | 0.07 |
Other qualification
| -0.14 | 0.15 | -0.48* | 0.19 | 0.15** | 0.05 |
No educational qualification
| -0.31 | 0.16 | -0.65** | 0.20 | -0.02 | 0.06 |
Unemployed | 0.16 | 0.18 | -0.05 | 0.22 | 0.11 | 0.06 |
Socio-economic status (ref = managerial/professional) | | | | | | |
Intermediate
| -0.05 | 0.10 | -0.08 | 0.11 | -0.05 | 0.06 |
Employers
| 0.34** | 0.11 | -0.50* | 0.20 | -0.11 | 0.09 |
Supervisory
| 0.11 | 0.11 | 0.05 | 0.15 | 0.08 | 0.07 |
Routine
| 0.14 | 0.09 | -0.02 | 0.11 | 0.11* | 0.05 |
Other status
| 0.20 | 0.11 | -0.13 | 0.18 | -0.09 | 0.07 |
Monthly gross earnings (ref: bottom quartile)
| | | | | | |
2nd quartile
| 0.18* | 0.08 | -0.10 | 0.09 | 0.00 | 0.04 |
3rd quartile
| 0.67*** | 0.09 | -0.12 | 0.19 | -0.15* | 0.07 |
4th quartile
| 0.66*** | 0.09 | -0.18 | 0.19 | -0.16* | 0.08 |
Subjective health (ref: excellent) | | | | | | |
Good
| 0.00 | 0.08 | -0.13 | 0.09 | 0.02 | 0.04 |
Fair
| -0.28*** | 0.08 | -0.30** | 0.11 | -0.02 | 0.05 |
Poor
| 0.05 | 0.11 | -0.21 | 0.17 | -0.06 | 0.06 |
Very poor
| 0.12 | 0.16 | 0.04 | 0.23 | -0.02 | 0.07 |
Body Mass Index (ref: bottom quartile) | | | | | | |
2nd quartile
| 0.53* | 0.25 | -0.12 | 0.19 | 0.05 | 0.09 |
3rd quartile
| 0.54* | 0.26 | -0.03 | 0.20 | 0.01 | 0.09 |
4th quartile
| 0.72** | 0.26 | 0.21 | 0.21 | 0.12 | 0.10 |
Health limits daily activities | -0.03 | 0.09 | 0.02 | 0.12 | 0.06 | 0.04 |
Suffering from an illness | 0.17 | 0.15 | 0.23 | 0.16 | 0.12** | 0.05 |
Reported health problem | | | | | | |
Diabetes
| -0.16 | 0.13 | 0.24 | 0.13 | 0.06 | 0.07 |
Relating to stomach problems
| 0.21* | 0.09 | 0.05 | 0.11 | -0.04 | 0.07 |
Cancer
| 0.11 | 0.28 | 0.35 | 0.25 | -0.08 | 0.13 |
Epilepsy
| -0.13 | 0.28 | -0.12 | 0.31 | 0.05 | 0.17 |
Relating to chest problems
| 0.16 | 0.08 | -0.01 | 0.10 | 0.01 | 0.05 |
Other health problem
| -0.00 | 0.13 | 0.10 | 0.18 | -0.02 | 0.04 |
Constant | 0.11 | 0.34 | -0.58 | 0.38 | 0.67*** | 0.14 |
Number of observations | 9,264 | | 5,881 | | 35,536 | |
The results suggest that a number of socio-demographic characteristics are statistically significant factors, in particular in the household panel studies. Those living in London or the South East are less likely to consent (true for both NCDS and UKHLS) and there is a positive association with belonging to the UK White population. The results are inconsistent across studies for the effect of the number of children in the household and whether or not the respondent lives alone – significant only in the UKHLS. If anything, results for the UKHLS suggest that the propensity to consent is higher in multi-person households.
There is some empirical evidence that people’s socio-economic position is associated with the propensity to consent. However, the results go in opposite directions in the different studies. Results on the long-running NCDS and BHPS suggest that those with generally higher levels of education are more likely to consent (note that this is statistically significant only for the BHPS sample) and the opposite is true for the UKHLS sample. Similarly, the association with consent and the respondent’s socio-economic status appears to be idiosyncratic to each study. Last but not least, whilst not being in the bottom quintile of the gross earnings distribution is strongly associated with a higher propensity to consent in the NCDS, the opposite is true for the household panel studies (albeit, this is statistically significant only for the UKHLS).
With respect to markers of health and use of health services, the results suggest that those with fair health are less likely to consent than those with excellent health (true for all studies but statistically significant only for the long-running BHPS and NCDS). Whilst none of the health conditions reported in the household panel studies are associated with the propensity to consent, being in a higher quartile of the BMI and suffering from stomach-related health conditions are associated with a higher propensity to consent on the NCDS.
Regression coefficients in non-linear probability models do not lend themselves to easy interpretation and so we report the corresponding marginal effects in Table
3. The reported marginal effect for, say London/SE, tells us that for two hypothetical individuals with average characteristics, the probability of giving consent is two percentage points lower if the person lives in London/SE than if the person lives elsewhere. Note that some of the effects that were statistically significant in the model reporting beta coefficients may not be statistically significant when expressed as Marginal Effects (ME). This is due to a non-linear transformation of the estimates.
Table 3
Logistic regressions on consent to health data linkage: Marginal effects
England | 0.00 | 0.01 | 0.06 | 0.04 | -0.02 | 0.01 |
London/SE | -0.02* | 0.01 | 0.03 | 0.03 | -0.02* | 0.01 |
Male | -0.00 | 0.01 | 0.03* | 0.01 | 0.02*** | 0.00 |
British/Irish White | 0.05* | 0.02 | 0.18*** | 0.03 | 0.09*** | 0.01 |
Aged 50-52 | | | 0.02 | 0.03 | 0.02 | 0.01 |
Number of own children in the household (ref: none) | | | | | | |
1
| -0.02 | 0.01 | 0.00 | 0.03 | 0.02* | 0.01 |
2
| -0.01 | 0.01 | -0.01 | 0.03 | 0.01 | 0.01 |
3 or more
| -0.02 | 0.02 | 0.07 | 0.04 | 0.01 | 0.01 |
Lives alone | 0.03 | 0.02 | 0.04 | 0.02 | -0.02** | 0.01 |
Highest degree (ref: higher degree) | | | | | | |
First degree
| -0.02 | 0.03 | -0.08 | 0.05 | 0.01 | 0.01 |
Diploma
| -0.03 | 0.03 | -0.12** | 0.05 | 0.03* | 0.01 |
A-level
| -0.03 | 0.03 | -0.08 | 0.05 | 0.03** | 0.01 |
Other qualification
| -0.02 | 0.02 | -0.10* | 0.05 | 0.03** | 0.01 |
No educational qualification
| -0.05 | 0.03 | -0.14** | 0.05 | -0.00 | 0.01 |
Unemployed | 0.03 | 0.03 | 0.02 | 0.05 | 0.02 | 0.01 |
Socio-economic status (ref = managerial/professional) | | | | | | |
Intermediate
| -0.01 | 0.02 | -0.01 | 0.03 | -0.01 | 0.01 |
Employers
| 0.06** | 0.02 | -0.11* | 0.05 | -0.02 | 0.02 |
Supervisory
| 0.02 | 0.02 | 0.02 | 0.04 | 0.02 | 0.01 |
Routine
| 0.02 | 0.01 | 0.00 | 0.03 | 0.02* | 0.01 |
Other status
| 0.03 | 0.02 | -0.03 | 0.04 | -0.02 | 0.01 |
Monthly gross earnings (ref: bottom quartile)
| | | | | | |
2nd quartile
| 0.03* | 0.01 | -0.02 | 0.02 | 0.00 | 0.01 |
3rd quartile
| 0.11*** | 0.01 | -0.01 | 0.05 | -0.03* | 0.01 |
4th quartile
| 0.11*** | 0.01 | -0.04 | 0.05 | -0.03* | 0.01 |
Subjective health (ref: excellent) | | | | | | |
Good
| 0.00 | 0.01 | -0.03 | 0.02 | 0.00 | 0.01 |
Fair
| -0.04*** | 0.01 | -0.07** | 0.03 | -0.00 | 0.01 |
Poor
| 0.01 | 0.02 | -0.04 | 0.04 | -0.01 | 0.01 |
Very poor
| 0.02 | 0.03 | 0.01 | 0.06 | -0.00 | 0.01 |
Body mass index (ref: bottom quartile) | | | | | | |
2nd quartile
| 0.09* | 0.04 | -0.05 | 0.05 | 0.01 | 0.02 |
3rd quartile
| 0.09* | 0.04 | -0.04 | 0.05 | 0.00 | 0.02 |
4th quartile
| 0.12** | 0.04 | 0.02 | 0.05 | 0.02 | 0.02 |
Health limits daily activities | -0.00 | 0.02 | 0.00 | 0.03 | 0.01 | 0.01 |
Suffering from an illness | 0.03 | 0.02 | 0.05 | 0.04 | 0.02** | 0.01 |
Reported health problem | | | | | | |
Diabetes
| -0.03 | 0.02 | 0.05 | 0.03 | 0.01 | 0.01 |
Relating to stomach problems
| 0.03* | 0.02 | 0.02 | 0.03 | -0.01 | 0.01 |
Cancer
| 0.02 | 0.05 | 0.10 | 0.06 | -0.02 | 0.02 |
Epilepsy
| -0.02 | 0.05 | -0.01 | 0.08 | 0.01 | 0.03 |
Relating to chest problems
| 0.03 | 0.01 | 0.00 | 0.02 | 0.00 | 0.01 |
Other health problem
| -0.00 | 0.02 | -0.01 | 0.03 | -0.00 | 0.01 |
As can be seen in Table
3, overall the effects are rather small amounting to around 3 percentage points. However, there are a number of greater effects. For instance, if a (hypothetical) person had no educational qualification rather than a higher degree, this would be associated with a 14 percentage point lower probability of consent. Whilst the majority of the larger effects are found on socio-economic and demographic characteristics in the BHPS, the same is true for health related factors in the NCDS. For instance, compared to being in the bottom quartile of the BMI, a person in the top quartile has a 12 percentage point higher probability to consent.