Scottish Longitudinal Study
Development & Support Unit

Completed Projects

Project Title:

Healthy life expectancy in Scotland

Project Number:



David Bell (University of Stirling)
Elizabeth Roberts (University of Stirling)

Start Date:

Approved on 04-09-2008


In his Independent Review of Free Personal and Nursing Care in Scotland, Lord Sutherland (2008) argues that the Scottish Government should:

“Establish long-term vision. Government at all levels should seek to establish a new vision for dealing with the challenge of demographic change, not just looking at long-term care, but also pensions, housing, transport, etc.”

This research is intended to help focus this long-term vision by testing an hypothesis that is vital to the prediction of health care costs. This hypothesis has never been tested in Scotland before and the Scottish Longitudinal Study is an ideal dataset on which to carry out this test.

Health care costs £10bn in Scotland: around one third of all spending authorised by the Scottish government is on health care. With an ageing population, it has been generally accepted that this share will have to rise due to the additional costs of ill health associated with age. If this assumption is correct, less will be available to spend on roads, schools, universities etc.

Estimates of how much health care is likely to cost in the future could be based on establishing the relationship between health care expenditures (HCE) and age. A simple regression model which estimates parameters for HCE (health care expenditure) and AGE (calendar age) could be used. One might include a range of other control variables, but they are omitted here to simplify the exposition. Note that the request for data includes some additional controls, such as gender, ethnicity, rural/urban indicator, qualifications etc.

The AGE variable(s) may be specified in different ways, such as a spline or categorical dummies and the observations may be indexed on both time (t) and individuals (i), so that one could make use of the longitudinal characteristics of the dataset, which would give added precision to the estimates, since one could eliminate individual-specific fixed effects. This is not being contemplated at this stage of the study, however. The intention is to work only with those aged 50 and above at the 2001 Census.

One can then use estimates of the parameters to forecast HCE when the age structure of the population changes. In particular, one could use this approach to forecast HCE as the population ages over the next few decades. This might seem a sensible approach to planning future resource allocation in the health service. However there is a body of evidence which suggests otherwise including Stearns and Norton (2004), Werblow, Felder and Zweifel (2007) Seshami and Gray (2004) O’Neill, Groom, Avery, Boot and Thornhill (2000). These studies augment the first regression model with an additional variable, time to death (TTD).

Stearns and Norton (2007) argue that the failure to include TTD in the HCE specification is a classic omitted variable problems, whereby the coefficient ?1 is biased due to the correlation of AGE with TTD, leading to overprediction of future HCE. Their estimate is that this over-prediction of individuals’ lifetime HCE could be as much as 9 per cent.

The estimate of HCE will require access to the hospital admissions and discharges data. The intention is to use time spent in hospital as a proxy for health care costs. While this is a fairly crude approximation, it is unlikely that data on the costs of individual health care episodes will become available.

As well as standard linear models.we shall experiment with count-based estimation models (Poisson or negative binomial) to allow for the expected skewness of the distribution of days in hospital.


Independent Review of Free Personal and Nursing Care in Scotland, A Report By Lord Sutherland (2008, April), ISBN 978 0 7559 5710 1

Hanlon, P., Sutton, M., Walsh, D., Lawder. R., Elders, A., Clark, D. and Whyte, B. (2007), Using the Linked Scottish Health Survey to Predict Hospitalisation & Death, Scottish Public Health Observatory

O'Neill C, Groom L, Avery AJ, Boot D, Thornhill K, "Variations in GP nursing home patient workload: results of a multivariate analysis." Public Health 2000; 114:446-50.

Seshamani, M. and Gray, A. (2004), 'Ageing and health care expenditures: the red herring argument revisited', Health Economics, 13(4): 303-314.

Stearns, S.C. and Norton, E.C. (2004), 'Time to include time to death? The future of health care expenditure predictions', Health Economics, 13(4): 315-327.

Werblow, A. & Stefan Felder & Peter Zweifel, 2007. "Population ageing and health care expenditure: a school of 'red herrings'?," Health Economics, 16(10): 1109-1126.

Related Outputs (viewable on CALLS Hub):

Explore the variables held in the SLS data dictionary.

Recent News

Upcoming Events

Latest SLS Tweets