Method

2.2.1 Study settings and its aging population

Tama city is a suburban municipality located in Tokyo, Japan. Because of the rapid development of economy after World War Ⅱ, the city was reclassified as Tama Town from Tama Village on April 1, 1964, after which the construction of Tama New Town was started in 1966. Finally, on November 1, 1971, Tama Town was reclassified as Tama City (Hoshi et al., 2011).

In 2001, the total number of the population aged 65 years old and above in Tama city was 16,164, accounting for 11.42% of the city’s total population, which is less than the average percentage in Tokyo (16.11%) (Kawachi & Kennedy, 1997). Of the 16,164, 3,089 were admitted to use the LTC (Tokyo Metropolitan Government, 2001). In 2004, this total number of elderly people increased to 20,040 (14.16% of the total population) which is still less than the average in Tokyo in terms of percentage (17.52%) (Kagamimori et al., 2009). The total number of elderly admitted to use the LTC during that time increased to 4,370 (Crowther et al., 2002).

2.2.2 Study design and research subjects

We conducted the follow-up study in Tama City, Tokyo, and the baseline of the longitudinal study was September 2001. All elderly aged 65 years or above residing in Tama city were invited to participate in our study. Among the 16,164 elderly in Tama city in 2001, 13,195 respondents (response rate= 80.2%) have completed and submitted their self-reported questionnaire. In September 2004, a follow-up survey was conducted by sending the same questionnaire to the respondents who participated in the first survey. Those individuals who relocated outside Tama City (n=505), and those who died (n=914), during the three years interval were excluded, and 2,642 of the

59

respondents did not submit their questionnaire. In total, 9,134 elderly have responded.

Questionnaires that were not completely filled (n=257), as well as responses from elderly aged 85 years and older (n=972) were removed. Finally, the data from 7,905 respondents were analyzed.

Table 2.1 shows the basic socio-demographic characteristics of the participants by gender and age. Of the 7,905 participants, 4,151 of them are female, accounting for 52.5 percent. As for the age, 3,589 of the subjects belong to the age group of 65-69 years old and constitute 45.4 percent (3,589/7,905) of all participants.

Table 2.1 Study subjects by gender and age

Male Female Total

n % n % n %

65-69 years old 1,814 50.5 1,775 49.5 3,589 100.0

70-74 years old 1,074 48.5 1,141 51.5 2,215 100.0

75-79 years old 585 41.2 835 58.8 1,420 100.0

80-84 years old 281 41.3 400 58.7 681 100.0

Total 3,754 47.5 4,151 52.5 7,905 100.0

2.2.3 Data collection procedure

Firstly, we requested and obtained the approval of the Department of Health and Welfare of Tama city government. The agreement was accompanied with cooperation of some LTC facilities within the city. Then, a self-reported questionnaire was sent to the elderly in Tama city, attached with a recommendation letter from the mayor of Tama city and a return mail envelope. The elderly respondents can complete the questionnaire anonymously by themselves or with someone’s help by following the instructions in the questionnaire. The completed questionnaires were sent to us via post mail. Finally, the respondents can send the questionnaire back to the city office in the provided envelope by mail once they finished it. The same procedure was carried in both the 2001 and 2004 surveys.

2.2.4 Variable measurement

The two measures of SES were education and annual income in this study. Education was measured by using a single-item. The subjects were asked the question, “What level

60

of education did you finish?” of which the options include three categories: (1) Graduate from Junior Middle School or lower; (2) Graduate from Senior Middle School;

(3) Graduate from College or higher. The equivalent annual income was measured by the following categories: (1) less than 1 million yen (USD=8,229), (2) from 1 to 3 million yen (USD=24,685), (3) from 3 to 5 million yen (USD=41,142), (4) from 5 to 9 million yen (USD=74,056) and (5) more than 9 million yen (USD=74,056). In this study, the equivalent income is calculated by the following method: 1) when the elderly lives alone without wife or husband, the income equals the equivalent annual income and calculation will not be processed; 2) when the elderly lives with the spouse, the equivalent annual income will be computed from dividing the household annual income by square root of two. Such a calculation is needed because there is big difference on the income between the Japanese male and female elderly.

Mental health was assessed by using the Three Health Factors Scale (Chan et al., 2011; Chiang et al., 2013; Hoshi & Sakurai, 2012; Roos & Havens, 1991), which is composed of nine questions that measure mental health, physical health and social health, respectively. In this study, we used the answers in the mental health section, which include three questions: (1) “How is your health status this year?”, (2) “Is your health status as good as last year?” and (3) “Are you satisfied with your life now?”. The Three health Factors Scale was also found to be suitable and easy to administer to Japanese elderly in Hanno city, Saitama Prefecture (Phelan et al., 2004).

NLTC was measured by two variables: LTC level 2001 and LTC level 2004. LTC level means the extent to which the elderly need for LTC and health care services which certified by the Japanese Ministry of Health, Labor and Welfare. In 2000, Japan implemented the LTCI and classified each applicant into one of six levels (or to reject—about 3 percent in the first round). The lowest level, called “assistance required”

(yôshien), is intended for preventive services; the other five levels are collectively called “care required” (yôkaigo) (Campbell & Ikegami, 2000; Fan et al., 1999). In our analysis, the subjects who were not eligible to use LTC were scored 0, while the subjects who were eligible to use the LTC were scored 1 to 6, which stood for the LTC levels range from the lowest to the highest, respectively. Finally, all of the subjects were categorized into two groups: “No NLTC” (score 0) and “NLTC” (score 1-6).

The socio-demographic data, including gender and age, were also collected. The

61

participants were classified into five categories: 65-69 years old, 70-74 years old, 75-79 years old,80-84 years old and 85 years old and above. Based on descriptive analysis, we found that there are substantial missing values in the participants aged 85 years and above. Furthermore, the standard deviations for many variables were too big. This may due to the subjects’ reduced ability of writing and reading, or visual and hearing impairments. Therefore, this age group was excluded in the current study.

2.2.5 Analytical approach

Descriptive statistics were carried out to demonstrate the socio-demographic characteristics of the respondents. The Chi-square Test was carried out to determine whether the difference exist on the distribution of the corresponding variables between the male and female elderly. The p value for the statistical significance test was set at 0.001 for Chi-square Test. Bivariate correlation analysis was employed to illustrate the correlation between the observed variables used in the SEM, and its p value for the statistical significance test was set at 0.001 or 0.05. Factor analysis was also conducted to identify the underlying main factors from a large initial set of the corresponding observed variables. All the analyses above were performed using the Statistical Package for Social Science for Windows (SPSS, version 17.0; SPSS Inc., Chicago, IL, USA).

In the present study, SEM was used to examine the structural relationship between SES, mental health and NLTC among the Japanese elderly. The model in the SEM consists of two kinds of variables: exogenous variable and endogenous variable. In the study, the endogenous variables were SES, and the exogenous variables were mental health and NLTC.

Assessment of the model fitness calculates how the proposed model might be consistent with the empirical data. Maximum-likelihood estimation is used to estimate the best-fitting model in this study. Chi-square test was the commonly used analysis of model fitness. However, when sample size is large (as in the present study), a non-significant chi-square is rarely obtained (Bentler & Bonett, 1980; Gerbing &

Anderson, 1993) and it is highly depend on the sample size (Jöreskog et al., 1981). The chi-square test was thus not used to assess the model fitness in this study, although it was shown in the figures. In SEM with maximum likelihood estimation and independent latent variables, when samples size is larger than 250, IFI is often reported

62

and used as the fitness statistics. Meanwhile, IFI are found relatively unaffected by sample size (Bråne et al., 1989; Hu & Bentler, 1995). Thus, IFI were used for the refinement of the model in the present study. In addition, CFI is not too sensitive to sample size(Jöreskog et al., 1981) and could adjust the value of IFI(Schilling, 2006).

Thus, CFI was also used as the fitness indices in the study. At last, RMSEA was also reported and used as the fitness statistics. The models are regarded as a good fit when p > 0.05; IFI value close to 1 (Bentler & Bonett, 1980); CFI value > 0.90 (Schilling, 2006) and RMSEA ≤ 0.05 (Tanaka, 1993).

AMOS version 17.0 statistical software package for Windows was used to conduct the SEM to obtain maximum-likelihood estimates of model parameters and calculate the goodness-of-fit indices.