Procedures and participants
CentERdata (Tilburg, the Netherlands) operates the Longitudinal Internet Studies for the Social Sciences-panel (LISS-panel) starting in 2007 . This panel, based on a traditional random sample drawn from the population register by Statistics Netherlands, consists of about 7000 individuals with a yearly attrition rate of about 10%. They are frequently invited to complete online surveys (more information about the LISS-panel and how to gain free access can be found in English at: http://www.lissdata.nl). There are several so-called core studies that are repeated each year in the same period (with a few exceptions), such as the core studies on health that are conducted each year November–December.
Proposals for studies, such as the studies on trauma we included in the present study (see below), are first evaluated by the Board of Overseers, the Internal Review Board. During the recruitment of the panel, respondents who agreed to participate in the panel received a confirmation e-mail, and a letter with login code. With the login code provided, they could confirm their willingness to participate and immediately start the first interview (i.e. double consent was ensured, for details see ).
Using this panel, in 2012 three surveys on trauma were conducted (T2a, Aprilresponse = 78.4%; T2b, Augustresponse = 83.5%, T2c, Decemberresponse = 86.3%) [14, 43]. In total, 4857 respondents participated in all three surveys on trauma and their data was extracted for this study. We also extracted data from studies on Social Integration and Leisure to obtain data on pre- and post-event loneliness, that were conducted in the spring of 2011 (T1a, February–Marchresponse = 63.1%), 2013 (T3, February–Marchresponse = 86.0%), and of 2014 (T4, February–Marchresponse = 86.0%). To obtain data on pre-event mental health problems, we extracted data from a survey on health conducted at the end of 2011 (T1b, November–December 2011, response = 77.2%). In total, 3244 respondents participated in these seven surveys.
We next made a distinction between by PTE’s affected respondents and comparison respondents not confronted with PTE’s or other life events in the 2 years before T2a and not between T2a and T2c (Nno PTE = 1513; questions on PTE see below). With respect to the affected respondents, we also excluded respondents confronted with PTE’s in the 2 years before T2a to ascertain that they (also) were not confronted with PTE’s about a year before T1, or between T1 and T2a. We finally selected respondents confronted with a PTE between T2a and T2b, and/or between T2b and T2c (ntotal PTE = 285).
Pre- and post-event loneliness was assessed at T1a, T3 and T4 using the six-item De Jong Gierveld Loneliness Scale . Respondents are asked to rate items such as ‘I often feel deserted’ and ‘there are enough people I can count on in case of a misfortune’ on three-point Likert scales (1 = yes, 2 = more or less, 3 = no). For the present study, we calculated the total score after recoding the three negative formulated items (0.76 ≤ Cronbach’s Alpha’s ≤ 0.81 across 3 loneliness assessments). For the present study, we calculated the total score after recoding the three negative formulated items, but we did not use the prior dichotomization of the answer categories . The reason was that the total score scale of loneliness without prior dichotomization enabled the categorization of sub groups with different levels of loneliness with almost equal cell counts somewhat better [Spearman’s rank (rho) correlation between scales with and without prior dichotomization is 0.99]. According to the cut-offs of the manualFootnote 1 about 10% (n = 176) had severe loneliness scores.
Mental health problems
Pre-event mental health problems were examined at T1b using the Mental Health Index or Inventory (5-item sub scale of the MOS 36-item short-form health survey) [46, 47]. Respondents were asked to rate their mental health during the past month on six-point Likert scales, such as ‘This past month I felt very anxious’ and ‘I felt depressed and gloomy’ (1 = never, 2 = seldom, 3 = sometimes, 4 = often, 5 = mostly, 6 = continuously). After recoding the third and fifth item, the total scores were computed (Cronbach’s Alpha = 0.84).
Potentially traumatic events
Confrontations with potentially traumatic events were examined at T2a, T2b and T2c. Respondents were administered a list of ten potentially traumatic events [i.e.: serious threat(s); physical violence; robbery; traffic accident; airplane incident; fire; burglary; serious infection/contamination such as HIV, legionella, poison; sexual violence/abuse; death of significant other/colleague] and an open question with regard to having experienced another potentially traumatic event in the past 2 years not listed that were recoded afterwards [14, 43]. In case respondents were confronted with more than one event in the past 2 years, they were asked to focus on the most drastic event (1 = 1 week ago, 2 = 2 weeks ago, 3 = 3 weeks ago, 4 = 4 weeks ago, 5 = 1–2 months ago, 6 = 3–4 months ago, 7 = 5–6 months ago, 8 = 7–12 months ago, 9 = more than 1 year ago). In case respondents reported life events that cannot be considered PTE (such as conflicts in the family, job loss) and were not confronted with PTE, they were excluded from the analyses.
Posttraumatic stress symptoms
To assess PTSD-symptom severity (DSM-IV), the 15-item impact of event scale (IES) [48, 49] assessing intrusion and avoidance reactions during the past week was administered at T2c. In addition, the seven hyper arousal items from the IES-R  were added while the original scoring system of the IES was retained. This approach has been used in previous research  and has the benefit of comparability with results obtained using the original IES, while still allowing for the measurement of all three symptom clusters (avoidance, re-experiencing, hyper arousal) of PTSD (DSM-IV). Factor analysis confirmed the general three-factor structure . Affected respondents were asked, while taking (the most recent) potential traumatic event in mind, to rate items such as: during the past week ‘I thought about it when I did not mean to’ or ‘I stayed away from reminders of it’ and ‘I had trouble concentrating’ on four-point Likert sales (0 = not at all, 1 = rarely, 3 = sometimes, 5 = often). We call this instrument the IESplus (Cronbach’s Alpha = 0.94).
In Fig. 1, we have presented a graphical overview of the surveys and the nature of the assessments (main topics).
For the present study, several total scores were recoded. Because of the skewness loneliness scores at T1a, T3 and T4 (for instance SkewnessT1a = 1.494, SE Skewness = 0.058), scores were recoded (based on T1a data) into relatively high (6 thru 15; 30.5%), medium (16, 17; 30.6%) and low levels of loneliness (18; 38.9%). With respect to PTE and posttraumatic stress reactions five PTE groups were composed:
Respondents without PTE (not in 2 years before T2a, and not between T2a and T2c);
Affected respondents with very low PTSD-symptom levels (PTE between T2a and T2c, and IESplus ≤ 5);
Affected respondents with low PTSD-symptom levels (PTE between T2a and T2c, and 6 ≤ IEplus ≤ 17);
Affected respondents with medium PTSD-symptom levels (PTE between T2a and T2c, and 18 ≤ IEplus ≤ 40); and
Affected respondents with high PTSD-symptom levels (PTE between T2a and T2c, and IEplus ≥ 41).
The range of IESplus scores in subgroups 2–5 were chosen to obtain more or less equal numbers of respondents in each subgroup as much as possible, although this categorization is somewhat arbitrary.
The total scores of mental health problems (MHP) were divided into three levels with a more or less equal numbers of respondents: i.e. relatively low (5–8; 32.7%), medium (9–11; 34.3%), and high levels of MHP (12–hi; 33.0%) to be able to present results in the tables in a uniform way.
Because the dependent variables had three values (low, medium, high), multivariate multinomial logistic regression (MMLR) analyses were conducted with pre-event loneliness/satisfaction at T1a (2011), mental health problems at T1b (2011), PTE group membership (T2c, 2012) and as predictors. Age, gender, having a partner and education level were also added as predictors. As such, the MMLR analyses assessed to what extent the aforementioned variables predict (a) medium levels of loneliness/satisfaction at T3 (medium versus low) and (b) predict high levels of loneliness/satisfaction at T3 (high versus low).