Study design, setting and participant recruitment

We conducted a two-wave panel survey (with computer-assisted telephone interview methodology) of a representative sample of 302 of the estimated 500 000 Syrian refugees in Istanbul.¹¹ We worked with the polling firm Frekans Research, which has extensive experience conducting research with Syrian refugees. Frekans Research has a database that includes the contact information for a representative sample of 1200 Syrian refugees in Istanbul. This sample was selected randomly based on government statistics about the residential area Syrian refugees live in. The sample for the current study was randomly selected from that representative sample of 1200 refugees in Istanbul who completed face-to-face interviews and gave their consent to participate in future studies.

The participants included adult refugees (aged 18 year or older) living under temporary protection in Turkey, but not in camps. Frekans research has validated our questionnaire for Syrian refugees. Professional translators at Frekans Research translated our questionnaire in parallel into both Turkish and Arabic. The questionnaire was back translated into English as well. We conducted the first wave of data collection between 9 and 15 July 2020 (n = 302), and the follow-up 2 months later between 11 and 14 September 2020 (n = 210); the response rate for the follow-up was 70%. Of our 302 Syrian refugees, 50% are females, mean age is 32 (range 18–69), 34% have primary school, 39% middle school, 20% high school and 7% master/doctorate education. The mean total household income per month was 2536 Turkish Liras, 17% were unemployed, 28% had reported to have experienced discrimination in the past 12 months, 47% have reported being very or extremely concerned about the stability of their residential status in Turkey and the average number of years living in Istanbul was 5.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. More specifically, Frekans Research is a member of European Society for Opinion and Marketing Research and the Turkish Research Association and was awarded the Trustworthy Research Certificate, and so they abide with the code of conduct of these committees. All procedures involving human participants were approved by the ethics committee of the School of Histories, Languages and Cultures of the University of Liverpool, approval number 7866.

Before starting the survey, the interviewers asked participants to give their informed consent by reading the consent wording to each participant, briefly explaining the nature of the study, the approximate duration of the survey, and informing the individuals that participation in the survey was voluntary and that their confidentiality was protected. The data were fully anonymised after the fieldwork and individual ID numbers were created that were used to link respondents’ answers in the follow-up wave. All analyses were conducted with Stata 14.

Measures

Our questionnaire integrates a selection of questions about life changes because of COVID-19 from the CoRonavIruS Health Impact Survey (CRISIS) developed by the National Institute of Mental Health (http://www.crisissurvey.org). More specifically, we used seven items from the CRISIS questionnaire section on ‘Life changes due to Coronavirus/COVID-19 crisis in the last two weeks’ integrated by one context-specific item about the quality of relationships with the Turkish community (see Appendix).

The CRISIS questionnaire asks about respondents’ feelings concerning a number of COVID-19-related changes in their daily life, including stress about restrictions of leaving home, change in the quality of social relationships, concerns about the financial situation and feelings of hope towards the end of the pandemic. The rationale of the item selection meets three criteria: budget constraints, survey timing and context. As our surveys were conducted in the middle of the pandemic and not at the beginning and as some of those items were US-specific, our team has selected a subset of items capturing COVID-19 stressors that could be used in later phases of the pandemic and outside the USA. The Appendix reports the full list of COVID-19 questions used in this research along with their response options and scoring range. In the analyses, responses for questions on the quality of the relationships and hopefulness for the end of the pandemic were reversed so that higher values denote more negative or problematic responses.

We used three validated measures of symptoms of mental disorders to assess Syrian refugees’ mental health. All measures were coded so that higher values reflect higher symptoms. In Supplementary Appendix A available at https://doi.org/10.1192/bjo.2021.1009 we present the full question wordings and response options of these questionnaires. For symptoms of depression we used the 10-item Center for Epidemiologic Study Depression Scale (CESD-10).^{Reference Radloff29} The depression scale is a shorter 10-item form of the CES-D. Respondents were asked to indicate how often or how consistently they have felt or behaved this way during the past 2 weeks: I was bothered by things that usually don't bother me; I had trouble keeping my mind on what I was doing; depressed; I felt that everything I did was an effort; I felt hopeful about the future; I felt fearful; My sleep was restless; I was happy; I felt lonely; I could not get ‘going’. Response options were: 0, rarely or none of the time (less than 1 day); 1, some or a little of the time (1–2 days); 2, occasionally or a moderate amount of time (3–4 days); 3, most or all of the time (5–7 days).

For symptoms of anxiety, we used the 6-item State-Trait Anxiety Inventory (STAI-6).^{Reference Marteau and Bekker30} Respondents were asked to state how they have been feeling in the past 2 weeks on the following items: I have been feeling calm; I have been feeling tense; I have been feeling relaxed; I have been feeling upset; I have been feeling content; I have been feeling worried. Response options were: 0, never; 1, sometimes; 2, often; 3, almost always.

For symptoms of stress, we used the 4-item Perceived Stress Scale (PSS-4).^{Reference Cohen, Kamarck and Mermelstein31} Respondents were asked about their feelings and thoughts in the past 2 weeks: How often have you felt that you were unable to control the important things in your life? How often have you felt confident about your ability to handle your personal problems? How often have you felt that things were going your way? How often have you felt difficulties were piling up so high that you could not overcome them? Response options were: 0, never; 1, almost never; 2, sometimes; 3, fairly often; 4, very often.

In the analyses we included a number of pre-migration and post-migration factors for which we asked for information in our surveys: gender, age, education, employment status, total monthly household income, number of years living in Istanbul, whether experienced discrimination, perception of Turkish society's negative attitude towards Syrian refugees, and concern about stability of residential status in Turkey. Descriptive statistics for all variables are presented in Supplementary Appendix B. To maximise comparability of questions across Syrian refugee surveys, we asked many of the above questions in the same way as they were asked in the Syrian Refugee Panel Study,^{Reference Lasala-Blanco, Chattaraj and Morales32} which examines the integration of Syrian refugees into host European societies.

Statistical analysis

We first conducted a principal components factor analysis with varimax orthogonal rotation to derive COVID-19 factor scores. Rotation is a common procedure in that, by forcing factors to be uncorrelated, it reduces ambiguity about which item belongs to which factor. Next, we took advantage of the longitudinal nature of our data and conducted a series of cross-lagged panel analyses based on linear regression models with maximum likelihood random-effects estimator to explore the causal relationship between COVID-19 factors and mental health.

Thus, in the models that estimate the effect of COVID-19 factors on measures of mental health at time (t – 1), we include levels of COVID-19 factors and mental health at time (t). An example of such models is presented in Equation (1), where CESD-10 (t) is the dependent variable. We estimate the same model for STAI-6 and PSS-4:

(1)$$\eqalign{ {\rm CESD}\hbox{-} 10( t ) &= {\rm \alpha } + {\rm CESD}\hbox{-}10 ( {t\ndash 1} )\cr & \quad + {\rm COVID}\hbox{-} 19\;{\rm Social}\;{\rm Relationships}( t ) \cr & \quad + {\rm COVID}\hbox{-} 19\;{\rm Social}\;{\rm Relationships}( {t\ndash 1} ) \cr & \quad + {\rm COVID}\hbox{-} 19\;{\rm Stress}( t ) + {\rm COVID}\hbox{-} 19\;{\rm Stress}( {t\ndash 1} ) \cr & \quad + {\rm COVID}\hbox{-} 19\;{\rm Hope}( t ) + {\rm COVID}\hbox{-} 19\;{\rm Hope}( {t\ndash 1} ) + {\rm \varepsilon }} $$

Thus, Equation (1) models depressive symptoms at time (t) as a function of depressive symptoms at time (t – 1), and COVID-19 factors (i.e. social relationships, stress and hope) at both time (t) and (t – 1), where α is the intercept and ε the error term.

Similarly, in the models that estimate the effects of mental health on COVID-19 factors at time (t – 1), we include COVID-19 factors along with mental health at time (t). Thus, in these analyses COVID-19 factors are the dependent variables and we estimate these models separately for each of our three COVID-19 factors. In these analyses, we also control for the two other two COVID-19 factors at both time (t) and (t – 1). An example of such models is presented in Equation (2), where COVID-19 social relationships (t) is the dependent variable:

(2)$$\eqalign{& {\rm COVID}\hbox{-} 19\;{\rm Social}\; {\rm Relationships}( t )\cr &\quad = {\rm \alpha } + {\rm COVID}\hbox{-} 19\;{\rm Social}\; {\rm Relationships}( {t\ndash 1} ) \cr & \qquad + {\rm COVID}\hbox{-} 19\;{\rm Stress}( t ) + {\rm COVID}\hbox{-} 19\;{\rm Stress}( {t\ndash 1} ) \cr & \qquad + {\rm COVID}\hbox{-} 19\;{\rm Hope}( t ) + {\rm COVID}\hbox{-} 19\;{\rm Hope} ( {t\ndash 1} ) \cr & \qquad + {\rm CESD}\hbox{-} 10( t ) + {\rm CESD}\hbox{-} 10( {t\ndash 1} ) \cr & \qquad + {\rm STAI}\hbox{-} 6( t ) + {\rm STAI}\hbox{-} 6( {t\ndash 1} ) \cr & \qquad + {\rm PSS}\hbox{-} 4( t ) + {\rm PSS}\hbox{-} 4( {t\ndash 1} ) + {\rm \varepsilon }} $$

Thus, Equation (2) models COVID-19 social relationships at time (t) as a function of COVID-19 social relationships at time (t – 1), the other two COVID-19 factors at both time (t) and (t – 1), and our three mental health measures at both time (t) and (t – 1). For simplicity, the basic models reported in Equations (1) and (2) omit all sociodemographic variables included in the analyses.

The longitudinal nature of our data allowed us to examine the bidirectional nature of the association between COVID-19 factors and symptoms of mental illness. That is, COVID-19 factors could adversely affect mental health, and poorer mental health could sustain or increase pandemic-related worries and stresses. Akaike's Information Criterion (AIC) is a goodness of fit measure that can be used to examine this question^{Reference Akaike33,Reference Vrieze34} by comparing alternative models. The model with the smallest AIC value fits the data better and, therefore, represents the best balance of goodness of fit and complexity. Finally, in supplementary analyses we test for an interaction between gender and COVID-19 factors to see whether the pandemic has affected Syrian refugees’ mental health differently for males and females.