**2. Methods**

#### **2.1 Participants and study setting**

Details of the survey method were shown in previous studies [25, 26]. Here, only the main survey methods are shown.

In September 2001, we surveyed all elderly people (65 years and older) living in their homes in a municipality near Tokyo. Of the 16,462 participants, 13,195 (80.2% of respondents) gave informed consent to participate in the study and returned a self-administered questionnaire by mail. A second questionnaire was mailed to respondents in September 2004. As a result, 8,558 people responded. There were 505 relocations, 914 deaths, and 3218 no responses. The survey respondents were all elderly people living in their homes over 65. We tracked all participants until August 31, 2007, and obtained personal data using IDs for 8,162 deaths and survivals through the City Hall Resident Registry.

The municipalities studied are urban suburbs that partially developed as suburban cities from the 1970s to the 1990s, including the period of high economic growth, to accommodate the increase in workers and their families in the metropolitan area. The total population of the cities surveyed was about 140,000 as of 2001, and 9.6% of the population was over 65 years old, almost half of the national average (17.3%) in 2000.

#### **2.2 Observed measurement variables**

Many observed variables and latent variables were used to clarify the causal structure using different factors in the survey year. The latent variables were determined by exploratory factor analysis, as shown in section 3.2. The question methods, options, and survey methods of the observed variables used in the survey are shown in the literature of previous studies [25, 26].

Socioeconomic status in 2001 was assessed via educational attainment and annual income. Levels of educational attainment were categorized into four groups; graduation from junior high school, graduation from high school, those achieving a higher academic level than junior college, and those who did not want to respond. Annual income levels fell into four categories; less than one million Japanese yen(equivalent to less than US \$7,400), less than three million yen, less than five million yen, and more than seven million yen in 2001.

Healthy dietary and lifestyle habits in 2004 were assessed by the dietary health score and the healthy lifestyle score, respectively [25, 26]. The three health-related dimensions examined in our study consisted of physical, mental, and social health components [25, 26].

#### **2.3 Analyses**

Detailed analysis of analytical methods, methods, and fitness was described in previous studies [25, 26]. Respondents aged 80 and older and those with more than a moderate degree of long-term care in the 2001 survey were excluded from the analyses due to an increased and indispensable deviation in their measurement variables. All data obtained were evaluated by degree to examine differences between males and females using Kendall's tau rank correlation coefficient. Exploratory factor analysis was used to fit all observed variables to corresponding latent variables.

The most important research method of this research is structural equation modeling (SEM). The analysis software used AMOS ver.28 for Windows (IBM). The analysis data used longitudinal follow-up data over seven years. Based on data from the 2001 survey, the 2004 survey investigated the three factors of a healthy diet, lifestyle, and health, as well as the degree of care required. We also clarified the number of days lived from 2004 to 2007. For causal structures, we used all combinations of four latent variables to find the model with the best fit.

Estimating the best-fitting model was carried out by the maximum likelihood method of SEM. The optimization algorithm was implemented with no-missing-data parameters. The direct, indirect, and total standardized effects of different latent variables on the endogenous health and life condition variable were measured by gender. The models employed indices criteria for assessing model fitness. Goodnessof-fit was approved by the chi-square goodness-of-fit test (χ2, degree of freedom, P value), NFI (Normed fit index), 1FI (Incremental fit index), RMSEA (Root mean square error of approximation) for the structural relationship model. Results were considered statistically significant if the p*-*value was less than 0.05.

A mutual agreement was signed between the city and local governments and Tokyo Metropolitan University regarding protecting individual privacy during the entire survey process. In this agreement, the confidentiality obligation was confirmed, and personal information handled by the university was limited to IDs. In September 2004 and September 2007, consent was obtained from the Tokyo Metropolitan University Graduate School Ethics Committee for a follow-up survey on manufacturing.
