IMPUTATION
In the 2014-15 NHS, voluntary measurements for height, weight and waist circumference were taken of respondents aged 2 years and over, while blood pressure was also measured for adult respondents (aged 18 years and over). Physical measurements not taken were imputed.
Non-response rates for physical measurements were higher in 2014-15 than in the 2011-12 NHS; for example, the non-response for BMI for adults in 2014-15 was 26.8% compared with 16.5% in 2011-12.
An investigation was undertaken to determine whether the characteristics of the people who were measured differed from those who were not measured. This investigation looked at variables such as smoking status, self-assessed health, employment status, marital status, country of birth, self perceived body mass, level of exercise and whether or not has high cholesterol (as a long-term health condition) and found no differences. While there were some differences in age, sex and part of state, these were taken account of in the weighting process.
As the sample weights have been calculated to apply to the whole fully responding sample, use of these weights would produce a correct estimate of the proportion of people with these characteristics (for example, overweight or obesity) but would not produce a correct estimate of the number of people who are overweight or obese. The use of a second weight, to be applied to the measured population only, was considered too confusing for use in microdata products such as TableBuilder and therefore imputation was used to obtain values for respondents for whom physical measurements were not taken and therefore allow the calculation of correct estimates for the number of people who are overweight or obese.
After investigations using 2011-12 NHS data and simulating lower response rates, the 'hot decking' imputation method was chosen for the 2014-15 NHS. In this method, a record with a missing response (the 'recipient') receives the response of another similar record (the 'donor'). A number of characteristics with which to match recipients to donors were used; for adults they were:
- age group
- sex
- part of state (capital city and balance of state)
- self-perceived body mass (underweight, acceptable, or overweight)
- level of exercise (sedentary, low, moderate or high)
- whether or not has high cholesterol (as a long-term health condition).
For example, a female recipient aged 35-39 years who lives in a capital city, has a self-perceived body mass of overweight, has high cholesterol and lives a sedentary lifestyle will match to a donor record who has the same profile (female, 35-39, self-reports as overweight, etc).
For adult BMI, around 92% of imputed records used all 6 variables in identifying donor records. For the remaining 8%, donor records could not be found using all 6 variables therefore fewer variables were used. For example, around 2% of recipients were matched to donors according to self-perceived body mass, level of exercise and cholesterol but not part of state.
For children 2-14 years, age group, sex and part of state were used as imputation variables while for 15-17 year olds, level of exercise was also used as an imputation variable, due to the other variables not being collected for children aged 2-17 years.
As a result of this investigation, results for all physical measurement data (BMI, waist circumference and blood pressure) from 2014-15 are of suitable quality and are directly comparable to 2011-12 and earlier years.
Print Page
Print All