Main results
Overall, our results suggest that the household CTVB assigned to each individual is a useful proxy for socio-economic status. Residents of lower value properties were more likely to be in the lower NS-SEC categories and live in more deprived wards. We found a clear trend of association between CTVBs and each of the lifestyle and health outcomes that we investigated. The trends were similar for CTVBs and NS-SEC and steeper than for the Townsend score, suggesting the CTVB is a satisfactory classification of socio-economic status in relation to the outcomes assessed in this study.
Classification of socio-economic status
The theory of the classification of socio-economic status has been widely debated [22, 23]. Different domains of socio-economic status, such as income, education and occupation, relate to different pathways between socio-economic status and health and each classification should be theoretically based and capable of being measured with readily available valid and reliable data [23]. The three methods of classifying socio-economic status used in this study all have some drawbacks. Although the new NS-SEC is based on an underlying theory of occupational stratification, the classification is less satisfactory for women than men [24] and it has problems classifying people not in the labour market, such as the permanently sick or disabled [15]. Thus although the classification is intended for people who are economically active, the eighth category of 'never worked and long-term unemployed' may include people who are economically inactive. This may explain the particularly large odds ratio for limiting long-term illness and the low MCS and PCS scores for the never worked and long-term unemployed, as these poor outcomes are associated with economic inactivity [25, 26]. The NS-SEC was not designed as an ordinal scale, although studies investigating NS-SEC and health outcomes have found evidence for significant trends across the classification [24, 25, 27]. Analyses reporting associations between area-based deprivation scores and individual health outcome are prone to the ecological fallacy [6], and the use of area methods is less justifiable where individual measures of socio-economic status are available [28].
CTVBs are defined in a walk-past assessment of the house and represent a relative scale of property value within an area, but the absolute value depends on the estimated amount the property might have been sold for on the open market, making the classification locally context-specific. As a household measure, assigning the CTVB classification to individuals may be prone to the ecological fallacy, but the household unit of aggregation is small and recent studies have demonstrated the importance of the homogeneity at household level in explaining variation in health status [29, 30].
Using CTVBs as a measure of socio-economic status is most likely to misclassify students and elderly people on low incomes who live alone in large old houses, or who move in with relatives in a more affluent household. The elderly living alone qualify for a 25% rebate, which effectively lowers their CTVB by one letter, but the implications of this re-categorisation for studies of health outcome has not yet been investigated. Further research is required to distinguish between the CTVB as a measure of income or wealth, since the value of a property may only be weakly associated with income or wealth, particularly in the rented sector.
A further disadvantage of using CTVBs is the differences in the systems between England, Wales and Scotland, including different time intervals between revaluations, and the changes that might result from the recent political debate on the funding and functions of local government. In England, the government has established an independent Inquiry into local government finance led by Sir Michael Lyons, and no decision on the postponed revaluation exercise in England will be taken until the Inquiry reports in late 2006 [31]. In Wales, however, CTVBs were revalued and rebanded in 2005, within a system of bandwidths which aimed to keep 59% of properties in Caerphilly borough in the same band, 25% to move down, 14% to move up one band and 2% to move up two or more bands [32]. Scotland has no plans to revalue CTVBs in the foreseeable future [33]. Although the valuation bands vary between England, Wales and Scotland, the relative position of properties might remain comparable. This needs to be assessed in further research.
CTVBs have one particular advantage over social class when used as a marker of socio-economic status in population surveys. Since the CTVB is known for both the sampling frame and respondents, this allows an assessment of non-response bias to be made by comparing the CTVB distribution between the sample and responders. Social class is only known for responders. Assessment of non-response can also be made by area deprivation score, but will suffer from a greater degree of ecological bias [6].
Comparison with previous studies
Three previous studies have investigated associations between CTVBs and a population health or lifestyle outcome [5, 9, 10]. Based on an analysis of 856 deaths occurring in one general medical practice population in England over a nine year study period, standardised death rates were shown to be significantly higher than average in bands A and B, and significantly lower than average in bands C to H [9]. We did not find consistent evidence of such a threshold effect, with significant trends in all outcomes investigated in this study across the categories of CTVB. A study of 1390 mothers sampled from the ALSPAC study found a significant trend in the risk of smoking cigarettes in pregnancy and up to eight weeks after delivery, and in breast feeding rates at up to four weeks with nearly twice as many mothers breastfeeding in CTVB D to H (57%) as in CTVB A (31%) [5]. A study to ascertain whether CTVBs are associated with household smoking rates found that 55% of CTVB A households included at least one smoker, compared to 22% in CTVB E and above [10]. Our current study corroborates the findings from this research group and extends the range of outcomes assessed to include other lifestyle factors, limiting long-term illness and physical and mental health status.
Methodological issues
One possible limitation of the comparisons made in this study is that each category of the three different measures was based on different proportions of the underlying population and hence the comparisons were not 'like for like'. However the NS-SEC and CTVB categories were fixed and used pragmatically in the form that they are available, although some merging was required to avoid small numbers in some categories of CTVB. The Townsend score categories were researcher defined, but we found little difference in the results between Townsend categories using centile cut-points and Townsend categories defined by approximately the same distribution of the population as the CTVB categories. Thus we have some confidence that the comparisons between the three measures were meaningful. A further limitation is that the analysis did not distinguish between owner-occupation and the rented sector and further work should assess the impact of housing tenure on the utility of CTVBs as a measure of socio-economic status.
Although CTVBs can be assigned to each resident in the sampling frame using the household address and the public domain website [2], this is clearly a laborious process, and was not possible given the size of this study. Using the household address for matching the council tax register to our sampling frame, we were unable to assign a CTVB to 10.2% of the 12,092 respondents. This proportion of missing CTVBs is similar to the 9.9% of respondents missing NS-SEC data. The trend of a higher degree of non-matching in more affluent areas is probably explained by the greater preponderance of house names in affluent areas, which due to spelling inconsistencies are harder to match than a street address. It is hoped that in future the National Land and Property Gazeteer (NLPG) will improve address matching. This will be achieved by linking disparate datasets using the Unique Property Reference Number. Once NLPG data are of a consistently high standard and in use across local government, it is expected that the anonymised matching of address to assign the CTVB to individuals will approach 100%.
Our study was faced with the usual limitations of postal questionnaire health surveys that result ultimately in non-response bias [34–36]. We made strenuous efforts to maximise the response, including an evidence-based approach to question wording, order and format (consistent question layout and fonts, paper colour, A5 booklet size, contact details on front page), a suitably worded covering letter signed, where possible, by the recipients GP, pre-paid return envelopes, three waves of questionnaires and multiple contacts to collect them by electoral canvassers. In order to raise awareness of the survey in the borough we used publicity posters in GP surgeries, wrote articles in local newspapers and publicised the survey on local radio. We cannot be certain that the use of the canvassers increased the survey response, but because they were local residents it is likely that they were able to facilitate the collection of completed questionnaires through personal contact with recipients. In addition they kept an accurate record of reasons for non-completion, distinguishing between people moved away and people who chose not to complete a questionnaire. This was important in accurately determining the second and third mailing list and compiling an accurate adjusted denominator.
The 62.7% overall response to the survey was satisfactory considering the socio-economic characteristics of the borough [34], and in comparison to the 57.7% response achieved in Caerphilly borough by the Welsh Health Survey in 1998 [16]. Non-response bias resulting from CTVB and NS-SEC item non-response rates of around 10% should be considered further. However, since the purpose of the analyses was to compare trends between CTVBs, NS-SEC and the Townsend score, rather than make inferences about the population, item non-response is unlikely to be a threat to the validity of the study.
As expected from the literature on non-response [34–36], we found the lowest response to the survey was from young males. However we found a similar distribution of CTVB categories in respondents compared to the adjusted denominator, but with a small under-representation of respondents in band A. Overall, we achieved a reasonably representative response by housing value.