Effect of Health Comparisons on Functional Health and Depressive Symptoms - Results of a Population-Based Longitudinal Study of Older Adults in Germany

Objective To investigate the effect of health comparisons on functional health and depressive symptoms in a longitudinal approach. Gender differences were examined. Methods The German Ageing Survey (DEAS) is a nationwide, representative longitudinal study of community dwelling individuals living in Germany aged 40 and older. The surveys in 2008 and 2011 were used, with n = 3,983 respondents taking part in both waves. Health comparisons were quantified by the question “How would you rate your health compared with other people your age” (Much better; somewhat better; the same; somewhat worse, much worse). Functional health was assessed by the subscale “physical functioning” of the 36-Item Short Form Health Survey (SF-36) and depressive symptoms were measured by the Center for Epidemiologic Studies Depression Scale (CES-D). Results Adjusting for sociodemographic factors, self-assessed health, social network, self-efficacy and optimism, and morbidity, fixed effects regressions revealed that functional health decreased significantly and considerably with negative health comparisons in the total sample (transitions from ‘the same’ to ‘much worse’: β = -11.8), predominantly in men. The effects of negative health comparisons (transitions from ‘the same’ to ‘much worse’: β = 4.8) on depressive symptoms were comparable (in terms of significance) to the effects on functional health, with stronger effects in women. Positive comparisons did not affect functional health and depressive symptoms. Conclusion Our findings underline the relevance of negative health comparisons on functional health (men) and depressive symptoms (women). Comparison effects are asymmetric and mostly upwards.

Introduction affect the outcome variables differently (asymmetric effects). Additionally, (3) gender-specific effects were estimated.
In total, our hypotheses are as follows: (1) Health comparisons affect (i) functional health as well as (ii) depressive symptoms. (2) Negative health comparisons affect the outcome variables more strongly than positive health comparisons. (3) The effect of health comparisons on the outcome variables is more pronounced in men.
Fixed effects (FE) regressions were used to estimate the effect of health comparisons on functional health and depressive symptoms, adjusting for numerous regressors often found to be relevant for our outcome variables, such as sociodemographic factors [24,25], social network, self-efficacy and optimism [26], as well as morbidity [27]. By using panel-econometric techniques, unobserved heterogeneity (e. g. personality or genetic disposition) can be taken into account.

Methods Sample
Data were derived from the third and fourth wave from the public release of the German Ageing Survey (DEAS), provided by the Research Data Centre of the German Centre of Gerontology (DZA). It is a population-based, representative survey of the German population aged 40 and older where individuals were interviewed in their homes by trained staff using a standardized questionnaire. The sampling procedure reflects a national probability sampling. While 8,200 individuals participated in the third wave, 4,855 individuals participated in the fourth wave, thereof 3,983 respondents took part in both waves and filled out the functional health and depressive symptoms questions. These 3,983 individuals were included in our analysis.
Discrepancies regarding sample sizes can mainly be explained by the collection of new samples in the third wave. Thereby, 6,205 community-dwelling individuals (birth cohorts 1923-1968) participated for the first time in the third wave, whereas 1,995 individuals had already been interviewed in the former waves. The fourth wave is a "pure" panel survey. Most of the individuals included in the fourth wave belong to the panel sample 2008. More details regarding the sampling frame and the sample composition have been reported by Engstler and Motel-Klingebiel [28]. Written informed consent was given prior to the interview.
Functional health and depressive symptoms. Functional health was assessed by the subscale "physical functioning" of the 36-Item Short Form Health Survey (SF-36) [29]. Impairments in ten activities of daily living such as climbing stairs, bathing or lifting or carrying groceries were rated on a three-point scale [30]. Consequently, we focus on rather simple activities of daily living. The items were transformed into a scale ranging from 0 (worst score) to 100 (best score).
Depressive symptoms was quantified by the Center for Epidemiologic Studies Depression Scale (CES-D) [31], consisting of 15 items. The scale represents the sum of all 15 items (0-45), with high values indicating more depressive symptoms.
Health comparison. Health comparisons were quantified by the question "How would you rate your health compared with other people your age" (Much better; somewhat better; the same; somewhat worse; much worse). The category 'the same' was considered as reference category.
Potential confounders. Furthermore, we controlled for time-dependent regressors which might be relevant for functional health and depressive symptoms. Thus, we controlled for age, family status (Ref.: married, living together with spouse; married, living separated from spouse; divorced; widowed; never married), monthly household net income in Euro (logarithmized) and morbidity (total number of physical diseases, ranging from 0 to 10, e. g. cardiovascular diseases, diabetes, cancer, respiratory diseases, eye diseases, hearing problems). These conditions were informed by the Charlson Comorbidity Index [32]. Moreover, we controlled for the number of important people in regular contact (0-9) and self-efficacy and optimism (HOPE scale [33]), ranging from 1-4 (high values indicate great self-efficacy) since it has been shown that self-efficacy is a predictor of our outcome measures [34,35]. Additionally, we included dummy coded variables for region and employment status (Ref.: working; retired; other: not employed) (both not shown, but available upon request).
In order to rule out the possibility that our health comparisons only reflect deteriorations in own subjective health, we additionally controlled for subjective health (1 = "very good" to 5 = "very bad").

Statistical analyses
Compared with cross sectional regression techniques, longitudinal regression techniques offer the advantage of controlling for unobserved heterogeneity, such as genetic disposition. This is important in well-being research as unobserved heterogeneity is in most cases systematically correlated with the regressors. In such a case, random effects (RE) regressions are inconsistent. Thus, FE regressions which leads to consistent estimates are the method of choice since unobserved heterogeneity can be taken into account [37]. It is worth mentioning that in FE regressions, solely within-variations are used (hence, also called 'within-estimator'). Thus, only timedependent variables can be included in FE regression models. Standard errors that cluster errors at the individual level were reported in order to account for heteroscedasticity and serial correlation [38].
A panel regression model can be denoted as follows [39]: . ., N: units (persons); t = 1,. . ., T: time Factors changing over time (but constant across individuals) are denoted as λ t . Constant observed characteristics of individuals are denoted as W i . Time-varying idiosyncratic errors are denoted as ε it and time-dependent predictors are denoted as X it . Contrary to cross-sectional regressions, an individual specific intercept α i is included in this model. This factor captures the impact of unobserved factors constant over time of an individual I on outcomes Y it . This is especially important when a correlation between observed predictors and α i is allowed (addressing the endogenous selection into treatment-based on time-constant unobserved factors). This is achieved by the FE-estimator.
The FE-estimator used within-transformed data to estimate the equation mentioned above from variation in observed independent variables and outcome variables (intraindividual changes over time): By differencing the data, the impact of time-constant factors (both unobserved and observed) is removed. Hence, changes in the dependent variable (Y it − Y i ) merely depend on changes in time-varying predictors X it as well as time-varying idiosyncratic errors ε it .
Furthermore, it is worth mentioning that the Stata command for FE regression analysis include individuals with only one observation in calculating the number of observation as they provide information about the constant and the variance components and so forth. Nevertheless, it does not affect the standard errors and the beta-coefficients.

Descriptive analysis
For individuals included in FE regressions, our variables are described over time (waves 3-4) in Table 1. As for FE regressions with functional health as outcome variable, at wave 3, mean age was 62.1 years (±10.9 years), ranging from 40 to 93 years. The majority was male (52.3%), was married, living together with spouse (74.7%), had medium education (51.1%), and was retired (50.7%). The mean monthly household net income was €2,703.7 (±€2,978.9). The mean Descriptive statistics for individuals included in FE regressions with depressive symptoms as outcome variable were almost the same. Besides, mean depressive symptoms score was 6.0 (±5.8) at wave 3.
In men, the pooled mean depressive symptoms score was 5.5 (±5.1) and the pooled functional health score was 86.2 (±19.7), whereas in women, the pooled mean depressive symptoms score was 6.9 (±6.5) and the pooled functional health score was 82.9 (±22.0).

Regression analysis
Findings of the longitudinal regressions are depicted in Table 2 (functional health) and Table 3 (depressive symptoms). Since we were interested in the intraindividual changes, individuals were only included in the sample if they had changes in the outcome variable between the third and fourth wave. Consequently, 4,184 participants were interviewed in the third and fourth wave. Thereof, 2,240 individuals were included in FE regressions with functional health as outcome variable. Moreover, 2,201 individuals were included in FE regressions with depressive symptoms as outcome variable. The slight differences between the different regressions can be explained by missing values in depressive symptoms.
In the total sample, the occurrence of negative health comparisons was significantly associated with substantial losses in functional health (transitions to 'much worse': β = -11.8), while the occurrence of positive health comparisons was not significantly associated with improvements in functional health. Gender-specific analysis showed that negative health comparisons were associated with considerable losses in functional health in men and in women (men: transitions to 'somewhat worse': β = -3.4; transitions to 'much worse': β = -12.9; women: transitions to 'much worse': β = -10.0). Furthermore, in the total sample and in both sexes, functional health decreased with age and decreases in self-assessed health. Moreover, functional health decreased with less self-efficacy (HOPE scale) in the total sample and in both sexes. As for marital status, only the transition from 'married, living together with spouse' to 'married, living separated from spouse' affected functional health in men significantly. Other potential confounders did not achieve statistical significance.
In the total sample and in both sexes, the occurrence of negative health comparisons was significantly associated with more depressive symptoms (transitions to 'much worse': β = 4.8; men: β = 4.0; women: β = 5.4), while the occurrence of positive health comparisons was not significantly associated with less depressive symptoms.
Furthermore, in the total sample, depressive symptoms increased with a decrease in selfassessed health. In the total sample and in men, depressive symptoms increased with age and less self-efficacy (HOPE scale). In total sample and in both sexes, all other potential confounders did not affect depressive symptoms significantly.

Additional analysis
We also estimated the effect of transitions from 'the same or better (= somewhat or much better)' (0) to 'somewhat worse' (1) or 'much worse' (2). Moreover, we estimated the effect of transitions from 'the same or worse' (= somewhat or much worse) (0) to 'somewhat better' (1) or  'much better' (2). In terms of effect size and significance, findings were mostly the same. However, in this model positive comparisons slightly increased functional health in women (transitions to 'somewhat better': β = 1.9) and the number of important people in regular contact was significantly associated with depressive symptoms in women (results of additional analysis are not shown, but are available upon request).

Main findings
Longitudinal regressions revealed that functional health decreased significantly and considerably with negative health comparisons in the total sample (transitions from 'the same' to 'much worse': β = -11.8), predominantly in men.
In total sample and in both sexes, the occurrence of negative health comparisons was significantly associated with increased depressive symptoms (transitions from 'the same' to 'much worse': β = 4.8), with stronger effects in women. The occurrence of positive health comparisons was not significantly associated with improvements in both outcome measures.

Previous research
It is difficult to compare our study with previous studies, since this is-to our knowledge-the first study examining the effect of health comparisons on functional health and depressive symptoms. Nevertheless, our finding that individuals weight upward comparisons more heavily than downward comparisons is filled with evidence from cross-sectional [14,40] and longitudinal studies [7,9] using micro data sets. It is worth repeating that these studies referred to income comparisons and measures of happiness.
The slight gender differences in our study might be explained by the fact that women might more heavily rely on mental rather than functional aspects by taking health comparisons, whereas men might heavily rely on functional aspects in the process of making health comparisons in mid and late life. This is supported by a study detecting associations between selfassessed health and depressive symptoms in old age women in Germany [41]. Moreover, evidence is present concerning the predictive value of global health on functional decline in old men [42], mostly underlining our findings.

Strengths and limitations
To our knowledge, this is the first study analyzing the effect of health comparisons on functional health as well as depressive symptoms in Germany. Moreover, asymmetries between upwards and downwards health comparisons were examined.
Additionally, by using longitudinal data, insights into the mechanisms can be derived and unobserved heterogeneity can be taken into account by using FE regressions. Furthermore, a key strength is that we draw on a representative sample of individuals aged 40 and older living in Germany.
One limitation might be that the corresponding reference group for health comparisons is explicitly listed (age-bracket). Therefore, we cannot rule out the possibility that other reference groups (e. g. colleagues or individuals having the same sex)-as indicated in income comparisons [43,44]-might be more relevant for health comparisons or at least might also enter the comparison process [45]. However, we strongly assume that the age bracket is the most salient dimension for health comparisons. Moreover, even though the face validity of this item seems to be acceptable, future research is required to ensure its validity. Another limitation is that our estimates might be slightly biased downwards due to panel mortality (endogeneity selection bias) in the German Ageing Survey between the third and fourth wave [46]. For example, the probability of participation in the fourth wave is affected by sex, education or self-assessed health.

Conclusion
Our findings underline the relevance of negative health comparisons for functional health (men) and depressive symptoms (women). Comparison effects are asymmetric and primarily upwards. Furthermore, the concept of comparisons might be more general and can be extended to the area of health.
More research is required in this area. For instance, personality [47,48] might moderate the relationship of comparisons and functional health as well as depressive symptoms. Even adaptation might play a role in this area [49][50][51].