1. IntroductionEver since Rowe and Kahn proposed “successful aging” [1,2] to counteract the long-standing tendency of gerontology, that is, emphasizing only a distinction between older people with diseases or disabilities and those suffering from neither, “New Gerontology”, the study deviating from the negative perspective on aging and accepting the diversity of the elderly, has received much attention . Similarly, Bowling and colleagues [4,5] reviewed 170 studies about successful aging and classified them into biomedical theories [1,6] and psycho-social approaches [7-9].Although Rowe and Kahn’s model was the most widely accepted approach, one of the limitations of the model was that it failed to address the implications of the fact that a disease-free older age was unrealistic for most people . In addition, psycho-social viewpoints criticized the relatively low factors of social network (SN) and psychological trait (PT) in the model [10,11]. In this context, Kahn proposed a theoretical integration  of Rowe and Kahn’s model with studies of Riley  and Baltes , which has not been studied yet. In spite of the criticisms and shortcomings, we still believe in Rowe and Kahn’s model [1,2,6] and the later-added social and psychological aspects to the model were rather inclusive and more comprehensive than others [7-9,13].After all, Rowe and Kahn concluded that successful aging was achieved in the following hierarchical order: “minimize risk and disability”, “maximize physical and cognitive ability”, “engage in activities”. In addition, SN and PT had effects on physical and cognitive function (PCF), and productive activity (PA) respectively, thus supporting this hierarchical order [1,2,9,14-21].So far we could find consensus that the model was truly multidimensional [1,4,5,10]. However, like the definition of successful aging still being contentious, there has been some debate about the appropriate hierarchical order of components in Rowe and Kahn’s successful aging as well [10,12,22,23]. Specifically, Rowe and Kahn clarified self-reported health (SRH) and SN were antecedents of PA (H3: SRH → PA and H6: SN → PA), which were potential inclusion errors in their modeling. Also, we speculated that there was a possible omission error in the path between SRH and PT (H2: SRH → PT). In addition, the path of H4: SN → PCF had discrepancies among various studies [10,14,19] and the path of H8: PCF → PA has suspected potential inclusion errors (Figure 1).[10,20] (Figure 1). As a result, we propose a modified version of Rowe and Kahn’s successful aging model for older people residing in a medium-sized Korean city.
2.1. ParticipantsThe survey was conducted on 600 seniors aged 65 and over, about 1% of the aging population living in the medium-sized city of Jeonju. Stratified sampling was used to allocate the number of samples into two administrative districts and purposively extracted samples of the male/female aging population ratio in the city. Trained surveyors visited the sampled elderly population at welfare centers, senior citizen centers, apartments, and houses in Jeonju from July 20 to August 31, 2011. Before the survey, the questionnaire was verified by five experts in health science, family medicine, and preventive medicine. Some items were added to/deleted from the basic model of Rowe and Kahn. Furthermore, a preliminary survey was carried out on 30 senior citizens in Seoul from May 16 to 20, 2011. The result of this study led to the modification of certain phrases in the previous questionnaire. Eventually, 600 questionnaires were distributed and 100% collection was achieved. Excluding the 44 incomplete or incorrectly answered questionnaires, 556 cases (92.6%) were included in the study.
2.2. SRHRowe and Kahn (1998) did not operationalize their definition but later on Strawbridge and colleagues , and Bowling and Dieppe  did so, which was followed in most of the follow-up studies [3,10].This study measured SRH with three indicators: healthy habits, subjective health, and number of chronic diseases. First, for healthy habits, smoking and drinking were measured with the items used in the 2008 2nd Korean Longitudinal Study of Aging. Regular dietary habits and exercise were measured using a 5-point Likert scale based on the suggestions of a gerontology specialist. Second, for subjective health, the Health Rating Scale developed in Northern Illinois University and used by Lee and Park  was used. This 5-point scale was composed of two items: one item on current health status perceived by the participant, and one item comparing current health with the health status of the previous year. The Cronbach α value was reported as 0.87 in the previous study , whereas it was 0.721 in our study. Third, to measure the number of diseases, we selected a total of 12 chronic diseases including diabetes, cataract (glaucoma), systolic hypertension, stroke, asthma, cancer, heart disease, arthritis, disk (backache), depression, gastritis (peptic ulcers), and prostate disease (urinary incontinence) based on the 2001 Korean National Health and Nutrition Examination Survey of the Ministry of Health and Welfare, and on the study conducted by Andrews and colleagues .
2.3. SNCharacteristics of SN have two dimensions: size of SN and social support . However, this study did not include the size of SN based on the contact theory which asserts that quality and not frequency is important in social relationships . In spite of the various forms, social support behavior could be largely classified into two categories: emotional support and instrumental support [10,16]. This study used eight items for emotional support and for instrumental support, respectively. Each item used a 5-point Likert scale. In the study conducted by Sung , the Cronbach α value of emotional support was 0.80, whereas the value of instrumental support was 0.83. In this study, the Cronbach α value of emotional support was 0.863, whereas the value of instrumental support was 0.878.
2.4. PCFPhysical function may be defined as the ability to perform physical acts required in daily life. To measure the degree in which the elderly maintained physical function, this study used the following items: “Can you walk 400 m without stopping?”, “Can you climb 30 stairs without stopping?”, “Can you stand up without feeling dizzy?”. These items were used by Strawbridge and colleagues  to operationalize Rowe and Kahn’s model.Cognitive function could be defined as the ability to perform mental acts required in daily life. To measure cognitive function, this study used the following items adapted by Strawbridge and colleagues : “Can you remember past incidents?”, “Can you remember where you placed an object?”, “Can you easily remember words when you speak?”. In addition, a total of six items, including home address, current date, and recollection of people, were used to find out the ability of the elderly to recognize a certain time, place, situation, or environment correctly . Sung’s  Cronbach ga certain time, place, situation, or environmethe value of cognitive function was 0.70. In this study, the Cronbach value of physical function was 0.784, whereas the value of cognitive function was 0.874.
2.5. PTIn MacArthur studies, PT was measured by factors such as self-efficacy, life satisfaction, and depression . Because self-efficacy and life satisfaction embrace positive aspects of PT, and on the contrary, depression encompasses a negative aspect, these constructs were not interchangeable PTs; however, depression was already included as an item of chronic diseases in SRH. Thus, to avoid conflicts and measure a purer construct of positive PT, we excluded depression.Self-efficacy could be a belief that one could solve unique problems, confront unique challenges, and influence the affairs occurring in everyday life . This study deployed a total of eight items , six items on self-efficacy and two items on interpersonal selfefficacy. Life satisfaction could refer to the subjective satisfaction level for one’s present life. This study measured a total of three items, which was the satisfaction with life scale .In the previous study , the Cronbach udyudy measured a total of tisfaction could the life satisfaction value was 0.789. In this study, the Cronbach . In this study, easured a total of tisfaction could refer to value was 0.873.
2.6. PAMost people would easily perceive PA, the final domain of the successful aging model, as a concept of paid labor. However, Rowe and Kahn included managing one’s household, taking care of family or friends, volunteering at a church or civic group, and looking after grandchildren in PA. In this study, we used five items—paid labor, volunteer work, group activity, household chores, and taking care of grandchildren—to measure PA according to the scale used by Herzog and colleagues , and other MacArthur studies. The participation number numerically expressed the participation in the five domains, whereas the participation time combined the hours participated in each activity variable.
2.7. Statistical analysesInitially, the cases which have multiple missing values in exogenous and endogenous variables were discarded. Each item in the case with only one missing value at random was replaced with a number computed
|x2||DF||SRMR||AGFI||TLI (NNFI)||CFI||RMSEA (LO 90-HI 90)|
3.1. Demographic characteristicsDemographic characteristics included gender, age, subjective economic status, religion, and marital status.Among the 556 participants, 219 (39.4%) were men and 337 (60.6%) were women. This was approximately the same as the gender ratio of the aging population in the city studied. Age wise, 12.8% of participants were 65–69 years old, 27.3% were 70–74 years old, 32.0% were 75–79 years old, and 27.9% were >80 years old. Regarding educational levels, 38.5% did not have any level of education, 38.8% were elementary school
3.2. Confirmatory factor analysisModel fitness for measurement was examined as shown in Table 1. First, as a generalized likelihood ratio, model x2 = 109.105 [degreesoffreedom (d.f.) = 34] and normalized x2 = 2.784 were appropriate. Second, as absolute fitness indices, standardized root mean residual (SRMR) = 0.027 which was less than 0.050 and root mean square error of approximation (RMSEA) = 0.057 (low: 0.044; high: 0.071), whose high value did not exceed 0.08 and the average value was close to 0.05, which is enough to be appropriate. Also, as adjusted goodness-of-fit index (AGFI) = 0.940, Tucker Lewis index (non-normalized fit index) [TLI (NNFI)] = 0.975, comparative fit index (CFI) = 0.985, the relative fitness indices implied that the measurement model was appropriate.
3.3. Structural regression analysisStructural regression analysis was performed to verify the hierarchical order of constructs in successful aging. Model fitness for structural regression analysis was confirmed (Table 4). First, as a generalized likelihood ratio, model x2 = 95.254 (d.f. = 36) and normalized x2 = 2.646, which were appropriate. Second, as absolute fitness indices, SRMR = 0.027, which was less than 0.050 and RMSEA = 0.055 (low: 0.042; high: 0.068), whose high value did not exceed 0.08 and the average value was close to 0.05, which is enough to be appropriate. Also, as AGFI = 0.944, TLI (NNFI) = 0.977, CFI = 0.985, the relative fitness
|x2||DF||SRMR||AGFI||TLI (ZNNFI)||CFI||RMSEA (LO 90-HI 90)|
3.4. Mediating effectThe paths rejected because of inclusion error and/or discrepancy among studies could have had an indirect effect. Therefore, we analyzed the mediating effects. As presented in Table 6, the paths rejected because of inclusion errors and/or discrepancies among studies did not have a direct but had an indirect effect, which meant that there were mediating effects in those paths. Thus, PCF and PT positively and fully mediated the relationships between SRH and PA (H3) and between SN and PA (H6). Also, PT positively and fully mediated the relationship between PCF and PA (H8).[3,10,14,16,21].Second, SRH, presumed to be an omission error and not covered in previous studies, had an influence on PT. Thus, we included the path from SRH to PT to a modified Rowe and Kahn’s successful aging model.Third, SRH did not directly affect PA, which confirmed an inclusion error. However, this did not correspond with the results of the previous studies [2,18]. This causality was supported by indirect effect that PCF and PT mediated between SRH and PA. Thus, those works [2,18] were partially supported.Fourth, SN had a positive effect on PCF. This corresponded with the reports that receiving high social support improved physical function and prevented decrease in cognitive function [22,29].
|Relevant factors||Standardized weights|
|Total effects||Direct effect||Indirect effect|
|H3: SRH -> PA||0.411*||0.000||0.411*|
|H6: SN -> PA||0.235*||0.000||0.235*|
|H8: PCF -> PA||0.255*||0.000||0.255*|