Visiting Parents in Times of COVID-19: the Impact of Parent-adult Child Contacts on the Psychological Health of the Elderly


 Using the 8th wave of the SHARE and the SHARE Corona Survey, we investigated whether the disruption of parent–adult child contacts due to social distancing restrictions increased the symptoms of depression among old age individuals during the first wave of the COVID-19 pandemic. We model the relationship between the disruption of parent–adult child contacts and the mental health of the elderly using a recursive simultaneous equation model for binary variables. Our findings show that the likelihood of disruption of parent–adult child contacts was higher with adult children who do not live with or close to their parents (i.e., in the same household or in the same building) for whom contact disruption increases about 15%. The duration of restrictions to movement and lockdowns also has a positive and significant effect on parent-child contact disruption: an additional week of lockdown significantly increases the probability of parent-child contact disruption, by about 1.5%. The interventions deemed essential to reduce the spread of the pandemic, such as the “stay-at-home” order, necessarily disrupted personal parent–child contacts and the social processes that facilitate psychological well-being, increasing the probability of suffering from a deepening depressed mood by about 17% for elderly parents.


Prior research on parent-child relationships and the well-being of older parents has mostly focused on the frequency of contacts, assuming that the more numerous the contacts, the greater the parent's wellbeing (Mancini & Blieszner, 1989;Tommasini et al., 2004). However, the COVID-19 pandemic and the consequent self-isolation, has led to a disruption of parent-adult child contacts. Indeed, in the recent literature on symptoms of depression in older persons related to the COVID-19 pandemic, one neglected topic is the influence of the disruption of the parent-adult child relationship on the psychological and emotional well-being of the elderly parent. Using data from the 8th wave of the Survey of Health, Ageing and Retirement in Europe (SHARE) until its suspension in March 2020 and the SHARE Corona Survey fielded from June to August 2020, this study aims to fill this gap providing additional insights into the psychological status of, and strain on, older parents during the COVID-19 outbreak and contributing to the very limited, but growing body of research on the negative association between social isolation and the psychological wellbeing of the elderly.
The remainder of the paper is organized as follows. Section 2 sets out the data, Section 3 describes the empirical strategy, while the results are presented and discussed in Section 4. The concluding remarks are in Section 5.

Data
This study makes use of individual-level data drawn from the 8th wave of the Survey of SHARE and the SHARE Corona Survey. The 8th wave of SHARE is a regular wave collecting information on the health, demographic and socio-economic status of individuals who are 50 years old or over. The interviews took place between October 2019 and March 2020. A sub-sample of SHARE panel respondents was interviewed from June to August 2020, via a Computer Assisted Telephone Interview (CATI), partly to collect a set of basic information as in the regular SHARE questionnaire, and partly to elicit information on life circumstances in the presence of COVID-19. The data collected with the latter questionnaire provide a detailed picture of how older adults were coping with the health-related and socio-economic impact of COVID-19 (Scherpenzeel et al., 2020). It also included the most important life domains for the target population and specific questions about the COVID-19 infection and life changes during the lockdown i.e. physical health (general health before and after the COVID-19 outbreak, infections and COVID-19 related symptoms); mental health (anxiety, depression, sleeping problems and loneliness before and after the COVID-19 outbreak); health behavior (social distancing, mask wearing etc.); SARS-CoV-2 testing and hospitalization; any medical treatment missed; satisfaction with treatment; changes in work and the patient's J o u r n a l P r e -p r o o f economic situation; social networks (changes in personal contacts with family and friends, help given and received, personal care given and received).
Our sample consists of 15,508 individuals living in 24 European countries plus Israel. This paper focuses specifically on individuals aged 65 and over. The COVID-19 pandemic has taken a heavy toll on their physical as well as mental health. The restrictive measures taken by governments (social distancing and isolation) to prevent the spread of the infection, have often resulted in social isolation and loneliness, to which older adults are likely more vulnerable because of their functional dependency, which in turn may have increased their depression and cognitive dysfunction with significant consequences for their psychological well-being (Banerjee, 2020).

Main Variables
The aim of our paper is to investigate the effect of an unexpected disruption of parent-child contacts due to COVID-19 in terms of the mental health of elderly parents. We constructed a binary indicator of a worsened depressed mood based on the SHARE Corona Survey. Respondents were asked to report whether, in the month before the interview, they felt sad or depressed. If the answer was -yes‖, they were also asked to report whether they felt sad or depressed -more‖, -less‖, or -about the same‖ compared to the period before the COVID-19 outbreak. Based on the answers, we created a dummy variable with value 1 if respondents, who said they felt sad or depressed, also said they were more depressed than before the COVID-19 pandemic outbreak, and 0 otherwise (less or about the same). 1,2 In order to measure the potential disruption of the parent-child relationship due to the pandemic, we created a binary variable taking into account the variation in the frequency of parent-child contacts in the periods before and during the first wave of the COVID-19 outbreak. Specifically, the 8th wave of SHARE includes a module on respondents' personal social networks. Each respondent can name up to seven people considered confidants. The social network module also gathers information on a respondent's relationship with these confidants (children, relatives, friends and neighbors) as well as additional characteristics for each social network member (gender, degree of kinship, network proximity). Using information drawn from the social network module, we were able to identify children considered by respondents to be -confidants‖. In addition, we established whether respondents had regular contacts with them before the COVID-19 outbreak (either in-person, by phone, email or any other electronic means). By -regular contacts‖ we mean either (i) daily; (ii) several times a week, (iii) about once a week (against about every two weeks, less than once a month and never). Using the information in the SHARE Corona Survey, we were also able to elicit the frequency of respondents' contacts with their -confidant‖ children during the COVID-19 outbreak.
In order to generate a variable to accurately measure the disruption of parent-children contacts, we compared the respondents' answers to the questions concerning the frequency of in-person and electronic contacts, as in the social network module of the 8th wave of SHARE with the answers reported in the SHARE Corona Survey. We constructed a binary variable with value one if those reporting regular contacts with confidant children living outside the household (in-person, by phone, email or any other electronic means) before the outbreak also reported contacts with them -less often‖ or -never‖ during the outbreak.
Unfortunately, while the SHARE Covid Survey enables different forms of contact with social network members to be distinguished-e.g., in-person, by phone or mail, email or any other electronic means, the 8th wave of SHARE does not distinguish between in-person and electronic contacts with people who live outside the household (children, parents, relatives, friends and neighbors). We assumed that the negative trend in the frequency of contacts with children was mainly due to the disruption of face-to-face contacts. Indeed, the restrictions many countries adopted during the first wave of the pandemic meant that people had to stay home and avoid all contacts with non-household members. This is confirmed by our data. In Appendix, figure 1 shows the fraction of individuals aged 65+ in each country who reported personal or electronic contacts with offspring -less often‖ or -never‖ since the outbreak. The majority of respondents who stated -less often‖ or -never‖ indicated face-to-face contacts. On average, about 50.7% of our sample had personal contacts with children -less often‖ or -never‖ since the outbreak, against about 10.2% for electronic contacts. 3

Econometric Model: the Recursive Bivariate Probit
From the methodological point of view, it should be noted that associating the disruption of parentchildren contacts with the psychological well-being of the elderly may be complicated by the presence of endogeneity. The disruption of parent-children contacts associated with the pandemic might have undermined older adults' mental health which, in turn, might have simultaneously influenced parental access to the informal support of their offspring (Cacioppo, Hughes, Waite, Hawkley, & Thisted, 2006;Holt-Lunstad, Smith, & Layton, 2010). In order to take this potential simultaneous relationship into account, we adopted a recursive bivariate probit model. The recursive structure of the bivariate probit model is built on a first structural form equation determining the probability of a deterioration in mental health conditions (y 1i in the equation (1)) and a second reduced form equation for the potentially endogenous dummy measuring the disruption of the parent-adult child relationship (y 2i in the equation (1)). In the probit model used to predict a deterioration of mental health, among the dependent variables, we included the indicator of disruption of the parent-adult child relationship (y 2i ).  on offdiagonal elements. In the above setting, the exogeneity condition is stated in terms of the correlation coefficient, which can be interpreted as the correlation between the unobservable explanatory variables of the two equations. The equations in (1) can be estimated separately as single probit models only in the case of independent error terms, i.e., the correlation coefficient is not significantly different from zero. 4

Identification Strategy
Conventionally, the identification of a recursive bivariate probit model has been based on exclusion restrictions to obtain a more robust identification of the parameters. Maddala (1983, p. 123) suggests that at least one of the reduced-form exogenous variables be not included in the structural equations as explanatory 4 The STATA software provides the statistic z= ̂ ⁄ to test the hypothesis H 0 : . If the error terms are independent, the maximum simulated likelihood estimation is equivalent to the separate maximum likelihood probit estimation.

Additional Independent Variables
In our model, we also control for a rich set of individuals' demographic and socio-economic characteristics and health variables. For demographics, we included the respondent's sex (0: male, 1: female) and age. The International standard classification of education (Isced) was used to classify the education variable. Three levels of education were considered: (1) low education (no educational certificates or primary school certificate or lower secondary education); (2) medium education (upper secondary education or high school graduation); (3) high education (university degree or postgraduate). Marital status was categorized as ‗living with a spouse or a partner in the same household' and ‗living as single'.
We also included an indicator of current financial distress to proxy the household's ability to make ends meet. Participants were asked to think about the household's total monthly income and rate the degree to which they felt able to make ends meet: with great difficulty, with some difficulty, fairly easily or easily.
This information was treated as a dummy variable with value one if respondents reported -with great difficulty‖ or -some difficulty‖ and zero otherwise. While in a long-term view one may argue that financial distress is also endogenous, in the short time period the survey captures it is likely that the economic impact of the lockdown measures did have an immediate effect on the main variables of interest.
In order to capture -needs‖ unrelated to the pandemic itself and the associated lockdown, we also included information on respondents' health status before the outbreak (between October 2019 and February 2020  In addition to social isolation, the local virus spread might also be a key factor in determining mental health issues and the disruption of social network contacts during a lockdown. Therefore, we considered a set of variables related to the COVID-19 experience, including a variable that provides information on the spread of COVID-19 among respondents' contacts. This dummy indicator has value one if anyone close to a respondent (i) had suffered from the Coronavirus; (ii) was hospitalized due to the infection; (iii) died after being affected by the Coronavirus, and 0 otherwise. We also introduced a measure that indicates whether the individual was directly affected by COVID-19, using a set of questions to establish if a respondent i) had experienced symptoms, ii) had been tested for COVID-19, and/or iii) had been hospitalized (Bergmann and Wagner, 2021). According to recent studies, symptoms indicating COVID-19 are associated with higher rates of anxiety and depression (Rajkumar, 2020; Le &Nguyen, 2021). We also included an indicator of personal distance behavior, expected to influence the disruption of contacts with social networks (including adult children) and mental health.
Finally, in the empirical model, we included an indicator of regular contacts (either in-person or by phone, email, or any other electronic means) with a relative/non-relative during the outbreak and country dummies to control for country fixed effect differences. Table 1 sets out a full description of the variables used in the model.

J o u r n a l P r e -p r o o f
[ Table 1 about here] [ Table 2 about here] [ Table 3 about here]

Results and Discussion
With specific reference to the reduced-form equation As mentioned above, we estimated the two equations for the probability of a disruption in parentchildren contacts and mental health issues, using the recursive bivariate probit specification. This allowed us to test for unobserved heterogeneity, the effect of which was captured by the correlation between the error terms from the single equation models. By simultaneously estimating the two equations and considering the correlation in the error terms, we controlled for the effect of unobserved factors. Tables 3 and 4  The results are shown in Table 4.
[ Table 4 about here] Table 4 shows that the results are consistent with those obtained from the baseline model.
Interestingly, the marginal effect of the disruption of parent-child contacts remains highly significant (at 99%) but greater than in the baseline model, increasing from 0.173 to 0.187. Our findings revealed that the higher the frequency of contacts between parents and children before the Covid outbreak, the greater the impact of the disruption on the psychological well-being of the elderly parents. 9 As a further sensitivity analysis, the baseline model was also tested considering two other symptoms of mental health: greater anxiety and greater loneliness. We considered anxiety since, like a depressed mood, it is a key symptom of mental health disorder (see Kalin, 2020). Moreover, we also considered feeling lonely. In relation to anxiety, respondents were asked the following questions: -In the last month, have you felt nervous, anxious, or on edge?‖, with yes or no answer options. If the answer was -yes‖, respondents were also asked -Has that been more so, less so, or about the same as before the outbreak of Corona?‖. Based on their answers, it was possible to create a dummy indicator that captures greater anxiety symptoms, considered to be the case if respondents reported -more so‖ and not the case otherwise (-less so or about the same‖).
For loneliness, respondents were asked -How much of the time do you feel lonely?‖, with response options being often, some of the time, hardly ever or never. Greater loneliness was assessed among those responding -often or some of the time‖ to the first question, and they were also asked -Has that been more so, less so, or about the same as before the outbreak of Corona?‖. We constructed a binary variable with value one if they reported -more so‖ and zero otherwise (-less so or about the same‖). Again, the outcomes [ Table 5 about here] [ Table 6 about here] The marginal effect of the disruption of parent-children contacts remains highly significant (once again at 99%). The factors that significantly influence greater anxiety and loneliness again include the 9 According to the previous literature strong family ties are negatively correlated with individuals generalized trust (Ermisch & Gambetta, 2010) and a higher intensity of the family relationship may also reduce -the capacity of the family members to interact in the outside world‖ (Herreros, 2015). Arguably, this may be the explanation why the effect of the parent-child contacts disruption on the psychological well-being of the elderly parents tend to be greater when the frequency of contact before the COVID-19 outbreak were higher (and family ties were stronger). 10 Table 4A in the Appendix shows the correlation matrix for the three self-reported psychological worsening distress symptoms, namely a deepening depressed mood, greater anxiety and greater loneliness.
worsening economic situation of respondents, their gender (female), suffering from mobility limitations, multiple chronic conditions and being directly affected by Covid-19. The probability of suffering from greater loneliness in particular also increases with age. 11 Finally, we test whether the marginal effect of the disruption of parent-children contacts on a deepening depressed mood varies by gender, age groups and respondents' education. Specifically, concerning age groups we considered first people aged 70 and over and then people aged 75s and over; for respondents' education we re-run the model dividing the sample into two groups: poorly educated against medium and highly educated. These analyses show that the estimated marginal effect is present in all groups of individuals considered: the disruption of parent-child contacts due COVID-19 pandemic and related lockdown policies appears to be a significant element for understanding the deterioration of mental health among older Europeans in times of COVID-19, irrespective of gender, age and level of education. Our results, included in Tables 7, 8 and 9 show that the sign of the estimated marginal effects is always positive and statistically significant. Our findings suggest that especially men, older people (aged 70 and over), as well as individuals with a lower level of education are at higher risk for mental health problems related to disrupted personal parent-child contacts during this challenging time.

Conclusions
The

1.A Wilde's (2000) Identification Approach (without exclusion restrictions)
For our estimation we employed a recursive bivariate probit model since we assumed that the disruption of parent-child contacts due to COVID-19 and the probability of suffering from a deepening depressed mood were determined simultaneously since we cannot exclude a priori that common  Table 1.A. We find only a marginal variation J o u r n a l P r e -p r o o f in the coefficients moving from one specification to another; hence the results are robust to this different identification assumption.
It is worth to notice that the variables that we employed as exogenous instruments in the model with exclusion restrictions (Maddala, 1983) do not significantly predict the probability of suffering from a deepening depressed mood.  Figure 1: Individuals aged 65+ reporting personal or electronic contacts with offspring -less often‖ or -never‖ since the outbreak  We constructed a joint model of parent-child contacts disruption and mental health issues that we estimated by using a recursive bivariate probit model that also takes into account the individuals unobserved heterogeneity that may characterize this relationship.  Our evidence reflects that the COVID-19's home confinement, which have also translated to a disruption of interpersonal contacts between older parents and their children, might have created a new set of challenges including mental health consequences that tend to be even more important when the family tie is stronger.  Our study provides additional insights into the psychological status and strain on older during the COVID-19 outbreak and contributing to the very limited, but growing body of research on the negative association between social isolation and the psychological wellbeing of the elderly.