In the pooled OLS regression model, socioeconomic variables, such as marital status, housing type, educational level, living expenses, and economic activity, as well as behavioral variables, such as walking and management of depression, were not significant in explaining SRH. In the FE panel regression model, only the behavioral variables maintained the power to explain SRH. This difference in the result between the FE panel regression and pooled OLS regression models could be attributable to the stability of the variables. Specifically, the difference could have been due to the fact that the general characteristics, such as marital status, educational level, housing type, living expenses, and economic activity, do not undergo significant intra-individual variability during the survey period. Pheiffer [19] observed that many risk factors for NCDs were not captured in FE regression analysis and noted that important variables can be excluded from analysis by using certain panel analysis methods, leading to distorted results [20, 21].
The implication is that even if an individual cannot alter the socioeconomic status, he can still improve his health by changing the lifestyle. Many studies verified walking, depression, and age as variables affecting SRH [11, 17, 22, 23]. Depression has a negative effect on SRH [6, 14], and is an important predictor of SRH [15, 24, 25]. In our study as well, depression was found to have negatively affected SRH in both men and women. Whereas depression was found to be more prevalent in women than in men, men with depression tended to have poorer SRH than women with depression. On a related note, depression was found to lead to poorer SRH by reducing physical activity [14]. Depression is associated with socioeconomic features and lifestyle [6], and tends to cause SRH to drop through correlation with other variables [15, 24, 26]. In South Korea, 17.8% of older adults suffer from depression [25], older adults with depression have poor SRH and show high incidence of NCDs [27]. SRH is inversely proportionally related to depression [15], and regular walking mitigates against the severity of depression [6, 25, 28].
Walking has an overall positive effect; and increased walking was shown to contribute to enhanced self-perceived health [17]. In this study, we included all walking with a minimum duration of 10 min, regardless of speed, method, and purpose; for example, walking while commuting, walking at normal speed to and from a bus stop or subway station, and walking rapidly for exercise. This suggests that health can be improved just by walking and that regular walking is important to health [17, 19]. Walking improves health and reduces the risk of NCDs [1, 10, 29, 30]. Walking for 30 minutes every day was shown to improve health [28, 31]. Regular walking for more than 6 weeks improved SRH [17]. An intervention program of walking 30–60 minutes/week to change the sedentary lifestyle contributed to the narrowing of the health inequality gap [32] and the reduction of healthcare cost [9, 27].
The UK adopted new public health policies intended to promote walking as a commute mode [11, 33], and switching the commute mode from car to public transportation and walking was found to promote health by increasing physical activity [24, 34, 35]. It has been reported that people who commute on foot every weekday showed 45% higher physical activity, lower obesity rate [35, 36], and reduced NCDs risk [37].
In a study using British national panel data, Flint et al. [38] found that active commuting and walking to and from public transportation increases physical activity, which in turn reduces the healthcare burden of NCDs [39]. Even walking a short distance was found to positively influence SRH, BMI, and systolic and diastolic blood pressure [19]. Some longitudinal panel studies also verified the positive effects of regular walking on SRH [2, 17, 19, 38]. Given that 60.8% of the respondents of the present study were economically active, many of them can change their commute mode to public transport [11, 38], which can efficiently improve their SRH by offering various walking routes.
In the FE panel regression, BMI was not identified as an influential factor for SRH. During the study period, neither the intra-individual variations in obesity nor the effect of such variations on SRH were significant [20]. However, there is a need to analyze BMI in a longitudinal panel study, given that is an important influential factor for SRH [9, 11, 22] and that it has been shown to affect SRH in separate analysis for men and women.
For comparison purposes, both pooled OLS and FE panel regressions were run in the current study. In general, panel data are analyzed with an FE regression which addresses the limitations of cross-sectional and time-series analyses and considers attributes inherent to survey units [18]. However, since unobserved endogenous attributes are not always associated with observed attributes, a pooled OLS regression should be run on adult population data [20, 21, 24]. This is in line with the aforementioned finding that important variables that may influence a dependent variable cannot be identified if only an FE panel regression is run on panel data that undergo little change, leading to distorted results [20, 21]. Furthermore, since health-related variables such as SRH are affected by socioeconomic characteristics, a pooled OLS regression is often preferred when analyzing health-related panel data [20–22].
In the present study as well, socioeconomic variables were found to have no effects on SRH in the FE panel regression, whereas the pooled OLS regression, which included the socioeconomic variables identified marital status, housing type, educational level, living expenses, and economic activity as influential variables for SRH. Since these important variables affecting SRH, rarely show significant intra-individual variability, care should be taken to identify them by varying the analysis method [20, 21, 40, 41].
Limitations of this study included the fact that although four-waves of the KHPS data were used as the sampling frame representing the population average, analysis was performed only with variables presented in the raw data. The variable “walking” was analyzed based on the daily mean time spent walking for lack of the weekly mean. Additionally, whereas obesity is known to be an influential factor for SRH [10, 26, 38], no significant association was found in this study. This may be ascribed to the short data collection period (four waves of the KHPS data 2010–2013).