The results of the present study indicated a substantial relationship between health risk behaviour and lower socioeconomic status at the level of the individual in Japanese men and women. The regional variance and the influence of regional socioeconomic indicators on risk behaviour were marked in women, but small in men.
Among men, those in the lower occupational classes showed a higher likelihood of risk behaviours, except for stress, as compared to "professionals". Especially, "service work," "transport" and "labour" showed significantly higher likelihood of current smoking, excessive alcohol consumption, physical inactivity and poor dietary habits. These observations suggest a plausible explanation for the higher mortality rates among these occupational classes noted in the national data .
Individual income was significantly related to risk behaviour of smoking, exercise, diet and health check-ups in both men and women, and lower income increased the likelihood of these behaviours. In men, a clear gradient of OR was found only for non-attendance of health check-ups and the gradient of OR for risk behaviours was not clearer than that of women. The model without occupation showed a clear gradient of OR in men  indicating a substantial degree of income-related inequality in health behaviour and its interaction with occupational class in Japanese men.
For women, no occupation showed a significantly lower OR as compared to "housework", with the exception of current smoking for "agriculture" and non-attendance of health check-ups for all occupational categories. This suggests that employment is associated with risk behaviour in women. Previous studies demonstrated that women's participation in society was related to a higher prevalence of smoking in accordance with the reduced intolerance toward this habit in women [40, 41]. In addition, the tendency for higher OR of current smoking, excessive alcohol consumption and poor dietary habits in "sales workers" and "service workers" among the occupational categories implies the accumulation of risk behaviours in Japanese women in lower occupational classes.
It is interesting that excessive alcohol consumption did not show an income-related gradient in either men or women: the second highest quintile in men showed a significantly lower OR and the lowest quintile in women showed a significantly higher OR as compared to the highest quintile. For women, the difference in excessive alcohol consumption by occupation was greater than those in the other health risk behaviours. A previous study confirmed that participation in the workforce increases women's drinking habit in Japan .
The relationship between individual socioeconomic status and non-attendance of health check-ups showed a different pattern from other behaviours. Women in the "housework" category and men in the "agriculture" category were less likely to attend health check-ups. Health check-ups in people of working age are strongly dependent on the workplace . Employees, particularly in large companies, have greater benefits of not only occupational health services but also preventive health services [44, 45]. The steepest gradient of OR for non-attendance of health check-ups among all health risk behaviours suggests substantial inequality in receiving preventive health services according to socioeconomic status in the Japanese population.
There was a clear gender difference in the influence of regional socioeconomic indicators on health risk behaviour. All risk behaviours showed higher intra-regional correlations in women than in men, and marked differences were found for current smoking and excessive alcohol consumption. Area socioeconomic conditions have been shown to influence health-related behaviour, and in general those living in socioeconomically disadvantaged areas have a higher likelihood of health risk behaviour [7, 13, 23, 46]. However, similarly to previous studies in France showing correlations between higher gross domestic product per capita in residential areas and both smoking and alcohol consumption [47, 48], the results of the present study indicated that women living in areas with higher per capita income had higher likelihood of current smoking, excessive alcohol consumption, stress and non-attendance of health check-ups. This may be explained by the following two points.
First, the regional differences in socioeconomic conditions in Japan are relatively small, and thus, regional disadvantage and deprivation would have little influence on individual health-related behaviour in the Japanese population. The data indicated that the degree of income inequality in Japan is smaller than in other industrial countries [49, 50], and the regional inequality in per capita income has decreased over the past several decades . As a previous study demonstrated that national financial adjustment policy contributed to a reduction of regional disparity in health levels , a national minimum across the country was achieved by egalitarian social policies in Japan.
A second explanation is related to the linkage between per capita income and urban-rural differences. In Japan, indicators reflecting urbanisation, such as population size and population density, are strongly correlated with higher income, and therefore income-related indicators represent not only socioeconomic conditions but also aspects of urban-rural differences – higher per capita income indicates an urban context [28, 53, 54]. Therefore, the results of the present study imply a relationship between a higher likelihood of health risk behaviour and urbanisation. Urbanisation accompanied by social participation of women is likely to increase the likelihood of smoking and alcohol drinking in Japan [40–42].
One notable feature of the geographical variation in Japan is the deteriorating relative health levels of urban populations, especially for women. Osaka Prefecture, which is the second largest metropolis after the Tokyo Metropolitan Area, had the second shortest life expectancy for women among the 47 prefectures in Japan . In addition, life expectancy in the Tokyo Metropolis, which had the longest life expectancy before 1965, ranked 15th for men and 37th for women in 2000 . Urban areas showed higher mortality rates from cancer and ischemic heart disease than rural areas , and individual health-related behaviour contributes strongly to these diseases [23, 57, 58].
For men, higher mortality rates are found in areas with lower income- and education-related indicators . As shown in the present study, regional socioeconomic indicators had little influence on health risk behaviour in men taking individual socioeconomic indicators into consideration. As mentioned above, a previous study indicated a gender-related difference in the relationship between mortality and area socioeconomic status: higher mortality rates in areas with lower per capita income were seen only in men . The higher likelihood of health risk behaviour in men on the lower socioeconomic scale suggests one plausible explanation for the higher mortality in areas with lower per capita income, where lower-income individuals are more likely to live. In contrast, the relationship between health risk behaviour and higher per capita income can explain the marked deterioration of health level in women in urban areas with higher per capita income.
Finally, it is necessary to discuss the limitations of this study, as well as its strengths. The present study was performed in a large nationally representative sample with stratified random sampling, although potential differences in response rates based on socioeconomic status and region may have caused selection bias [59, 60]. The questionnaire regarding health risk behaviour was comprehensive, although it was self-reported and a few behaviours (e.g., poor dietary habits and physical inactivity) are quite subjective. The items of dietary habits were based on the national guidelines for a healthy diet , and they (or the index formulated using these items) have been shown to be related to some aspects of physical health status [62, 63]. Nevertheless, the validity and reliability of these questions have not been examined in detail, and those of single-item questions about physical activity were verified but only moderately so [64, 65]. These issues probably induced misclassification bias [59, 60].
In this study, we effectively applied multilevel analysis to elucidate the influence of socioeconomic factors of two levels and to demonstrate regional variances. However, our models did not consider random effects of the variables or interactions between the variables. Supplemental analysis using random slope models for household income did not show statistically significant regional variance in the effects of household income on any health risk behaviour in either men or women (data not shown).
Other limitations are related to the measurements of socioeconomic status. In the present study, household income was used as the main measure of individual socioeconomic status, which was estimated from the details of income-related information and adjusted for family size and composition. Educational attainment is also commonly used as another major measure of socioeconomic status [66, 67]. As the survey used in the present study did not include educational information, no education-related variable could be introduced into the analyses. Previous studies have shown that differences in health-related behaviour among groups stratified according to educational attainment in Japan are substantial [29, 30, 68]. Further studies are required to clarify the independent and interactive influences of different socioeconomic measures on health risk behaviour.