From starvation to depression: unveiling the link between the great famine and late-life depression | BMC Public Health

From starvation to depression: unveiling the link between the great famine and late-life depression | BMC Public Health

Placebo examination

When employing a difference-in-differences (DID) model, it is essential to satisfy the parallel trends assumption. The placebo test is a crucial method for verifying this assumption. The primary objective of this test is to ensure that the estimated treatment effect genuinely reflects a causal relationship, rather than being a spurious effect resulting from model misspecification, random noise, or specific data characteristics.

To conduct the placebo test, we randomly assigned individuals to placebo treatment and control groups and repeated the simulated regressions 500 times. This process yielded 500 estimates of placebo treatment effects. If the distribution of these placebo effects is centered around zero and does not exhibit systematic deviations, it indicates that the model effectively identifies the true causal effect, rather than mistakenly attributing the influence of other factors to the treatment effect. This also implies that there are no inherent systematic differences between the treatment and control groups, thereby satisfying the parallel trends assumption.

As illustrated in Fig. 1, the 500 placebo regression coefficients are concentrated around zero and are distinctly separate from the actual estimated coefficient (represented by the red dashed line). This finding confirms that the baseline results are indeed driven by the experience of the Great Famine, rather than by other unobserved factors.

Fig. 1
figure 1

In addition, we followed the approach of Chen and Zhou (2007) to conduct a parallel trend test: using the cohort born from 1963 to 1967 as a subsample, with 1967 as the control group and other years as treatment groups for regression analysis. The coefficients were not statistically significant, supporting the parallel trends hypothesis. Specific results can be found in Appendix D.

Baseline results and robustness checks

Next, we employ the econometric model (1) to rigorously identify the causal relationship between exposure to the Great Famine and depression among the elderly. The specific results are presented in Table 2:

Table 2 Basic regression and robust tests

Column (1) reports the baseline regression results. The coefficient of the interaction term. EDR×I(Birth ≤ 1961) is 0.00158, which is statistically significant at the 1% level, indicating that experiencing the Great Famine significantly increased the level of depression among the elderly by 0.158%. Column (2) addresses the robustness of the baseline results by transforming the dependent variable, depression level, into a binary variable based on the median and conducting a logit regression. The results indicate that the odds ratio for EDR×I(Birth ≤ 1961) is 1.017, which is statistically significant at the 1% level. This finding suggests that individuals who experienced the Great Famine face a higher risk of depression. Column (3) considers the potential impact of widespread panic and population outflow caused by the famine during 1959–1961, which might have attenuated the effect of famine exposure on depression. Following the approach of Chen and Zhou (2007) [1], we exclude samples with population mobility to minimize bias from selective migration. The regression results show that the coefficient of EDR×I(Birth ≤ 1961) is 0.00138, still significant at the 1% level, indicating that famine exposure increased depression levels among the elderly by 0.138%. The slight decrease in the coefficient aligns with expectations.

Column (4) employs an alternative measure of famine severity based on the method proposed by Shi (2011) [42], using abnormal death rates to gauge the impact of the famine across provinces. Specifically, abnormal death rates are calculated as the difference between actual mortality rates and estimated normal mortality rates, with the latter derived using finite differences from mortality rates in the five years before and after the famine for each province (or municipality). The results show that the coefficient of EDR×I(Birth ≤ 1961) is 0.00220, statistically significant at the 1% level, indicating that experiencing the Great Famine increased depression levels among the elderly by 0.220%, further confirming the robustness of the results.

Column (5) excludes individuals who participated in the “Up to the Mountains and Down to the Countryside” movement, as previous studies have shown that early experiences in this movement significantly affect depression levels in old age [43, 44]. The re-estimated regression results show that the coefficient of EDR×I(Birth ≤ 1961) is 0.00139, significant at the 1% level, indicating that famine exposure increased depression levels among the elderly by 0.139%. This finding further supports the robustness of our results.

Additionally, we conducted other robustness checks, such as expanding the treatment group to include cohorts born in 1962 and earlier, as well as incorporating EDR and birth date variables, all of which remained robust. The results are presented in Columns (1) and (3) of Appendix C.

Mechanism test

This section will test the three theoretical mechanism hypotheses proposed in the previous article. The specific results are shown in Table 3 below.

  1. (1)

    Social Support.

In examining social support, this study uses two indicators: the time spent on caregiving by family members [45] and the frequency of participation in social activities by the elderly [46]. For caregiving time, we calculated the total monthly caregiving hours (hours/month) provided by parents, spouses, children, and grandchildren over the past month. The regression results are shown in column (1): the coefficient of the interaction term \(EDR_p \times I\left( Birth \le 1961 \right)\) is -0.0831, which is statistically significant at the 5% level, indicating that the experience of the Great Famine significantly reduced the time spent on family caregiving.

Regarding the frequency of participation in social activities, we measured it by summing up the number of activities such as “interacting with friends, attending community or recreational events, assisting solitary individuals, participating in clubs, volunteering or engaging in charitable activities, caring for sick or disabled people, using the internet, and investing in stocks”. The regression results are shown in column (2): the coefficient of the interaction term \(EDR_p \times I\left( Birth \le 1961 \right)\) is -0.00641, which is statistically significant at the 1% level, indicating that the experience of the Great Famine significantly decreased the frequency of elderly participation in social activities. In summary, the experience of the Great Famine indeed significantly reduced social support for the elderly, confirming Hypothesis 1.

  1. (2)

    Socioeconomic Status.

To assess socioeconomic status, this study employs two indicators: education level [47, 48] and wage income [49, 50]. Education level is measured according to Zhu and He (2021)’s [51] approach: less than elementary school = 0, elementary school = 6, middle school = 9, high school/technical school = 12, associate degree = 15, bachelor’s degree = 16, master’s degree = 19, and doctoral degree = 22. The regression results are shown in column (3): the coefficient of the interaction term \(\rmEDR \times I(Birth \le \rm1961)\) is -0.0389, which is statistically significant at the 1% level, indicating that the experience of the Great Famine significantly lowered the education level of the elderly.

For wage income, it is measured by whether the individual received wages or bonuses in the past year, coded as 1 for receiving and 0 for not receiving. Given that the dependent variable is binary, a Logit regression was used. The results are presented in column (4): the odds ratio of the interaction term \(\rmEDR \times I(Birth \le \rm1961)\) is 0.946, which is statistically significant at the 1% level. Due to the challenges in interpreting this result using relative risk, we calculated the average marginal effect from the Logit model, which is -0.0055 and statistically significant at the 1% level. This indicates that experiencing the Great Famine reduces the likelihood of receiving a wage by 0.55%. In summary, the experience of the Great Famine significantly reduced the socioeconomic status of the elderly, confirming Hypothesis 2. We also re-regressed using the logarithm of the number of wages or bonuses received, and found that the results remained robust. Please refer to Column (2) in Appendix C for details.

  1. (3)

    Intergenerational Support.

Intergenerational support is assessed using two indicators: material support and emotional support. Material support is measured by the total amount of material goods and cash received from children over the past year. The regression results are shown in column (5): the coefficient of the interaction term \(EDR_p \times I\left( Birth \le 1961 \right)\) is -0.00573, which is statistically significant at the 1% level, indicating that the experience of the Great Famine significantly reduced the material support provided by children to the elderly.

Emotional support is measured by the total frequency of contact with all children through phone calls, messages, WeChat, mail, or email. The regression results are shown in column (6): the coefficient of the interaction term \(EDR_p \times I\left( Birth \le 1961 \right)\) is -0.0116, which is statistically significant at the 5% level, indicating that the experience of the Great Famine significantly reduced the emotional support provided by children to the elderly. In conclusion, the experience of the Great Famine indeed significantly reduced intergenerational support for the elderly, confirming Hypothesis 3.

Table 3 Mechanism testing

Heterogeneity analysis

  1. (1)

    Household Registration Type

Urban-rural differences are a key factor when examining the long-term psychological health impacts of the Chinese Great Famine on the elderly. Chen and Zhou (2007) [1] revealed that the famine had different effects on urban and rural populations, showing that the impact on urban residents’ height was significantly weaker, as they were less affected by the famine during that time. This result contrasts sharply with findings related to rural populations. Due to limited mobility, rural residents were often forced to live in resource-scarce environments, leading to more severe damage to their health and psychological well-being. In contrast, urban residents, with better living conditions and higher social mobility, were able to access relatively adequate medical resources and social support, which helped mitigate the negative effects of the famine to some extent. Fan and Qian (2015) [22] further supported this view, noting that the long-term effects of the famine on middle-aged health were more significant in rural areas, especially for individuals born into rural families. Hence, it is expected that the famine experience has a more pronounced impact on depression among the elderly in rural areas. The results of the subgroup regression based on household registration (hukou) are presented in columns (1)-(2) of Table 4: the coefficient of \(EDR_p \times I\left( Birth \le 1961 \right)\) is significantly higher for the rural sample than for the urban sample (0.00237 > 0.00157), which is consistent with expectations.

  1. (2)

    Growth Period

Adolescence is a critical period for psychological and physical development, and experiences during this stage can have long-term impacts on mental health. Zhang (2002) [52] pointed out that adolescence is a key stage for individuals to understand and interpret the world, form lasting memories, and shape personality. Major societal upheavals like the Great Famine during this period could result in long-lasting negative effects on mental health in adulthood. Developmental psychology further emphasizes that adolescents are highly sensitive to environmental changes and lack mature strategies to cope with significant societal disruptions. Consequently, such experiences may damage the social support systems during adolescence, weaken positive social interactions, and increase the risk of depression in adulthood [19]. Therefore, it is anticipated that the impact of experiencing the famine during adolescence on depression will be greater. Following the approach of Cheng and Zhang (2011) [53], and Lin and Zhou (2019) [54], the sample born between 1943 and 1952 is defined as the adolescent cohort, which is then compared with other cohorts who experienced the famine during different life stages. The subgroup regression results are presented in columns (3)-(4) of Table 4: the coefficient of \(EDR_p \times I\left( Birth \le 1961 \right)\) is significantly higher for the adolescent sample compared to other periods (0.00182 > 0.00142), which aligns with expectations.

  1. (3)

    Confucian Culture

Religious beliefs have been shown to buffer the negative impact of life stressors, including major events like the Chinese Great Famine, on psychological health. Religion typically provides individuals with a sense of meaning, belonging, and emotional support, all of which can mitigate the adverse effects of traumatic experiences on mental health. For instance, the stress-buffering hypothesis suggests that religious beliefs, by offering emotional support and social networks, can help individuals cope with life’s stresses, thus reducing the incidence of mental health issues. Confucian culture plays a key role in this process. Confucianism emphasizes family, society, and interpersonal relationships, which can strengthen individuals’ social support networks and further buffer the negative impact of significant life stressors like the famine on mental health. This suggests that in regions where Confucian culture is stronger, individuals may be able to draw more psychological support from religious beliefs and cultural values, helping them cope better with stress and reducing the incidence of depression [27]. In this paper, the number of Confucian temples is used to measure the strength of Confucian culture in different regions, as these temples symbolize Confucianism. Li and Zhu (2021) [55] noted that the higher the density of Confucian temples in a region, the greater the influence of Confucian culture on the region’s cultural and economic development, as well as residents’ behavior. Based on the presence or absence of Confucian temples, the sample is divided into regions with strong and weak Confucian cultural influences. The subgroup regression results are presented in columns (5)-(6) of Table 4: the coefficient of \(EDR_p \times I\left( Birth \le 1961 \right)\) is significantly lower in regions with strong Confucian culture compared to regions with weak Confucian culture (0.00139 < 0.00175), which is in line with expectations.

Table 4 Heterogeneity regressions

link

Leave a Reply

Your email address will not be published. Required fields are marked *