Exposure to the troubles in Northern Ireland, memory functioning, and social activity engagement: results from NICOLA

In order to create a measure of social engagement, a range of social activities were recorded from 5,965 participants. Overall, 38% (n = 2292) said that they participated in at least one social activity. Of these 2292, 30.5% (929) participated in hobbies and social clubs; 31% (712) participated in sports/exercise groups; 13.5% (309) participated in local community or neighbourhood groups; 5% (110) participated in groups for children or young people; 7.6% (175) participated in adult education groups; 12.8% (293) participated in groups for older people; 1.6% (36) participated in environmental groups; 6.8% (157) participated in health, disability, or welfare groups; 0.9% (22) participated in political groups; 0.8% (19) participated in trade union groups; and 34% (783) participated in religious groups. A count of involvement in all groups was created so that participants could have participated in a minimum of 0 and a maximum of 11 social activities (see Table 2, below).

Questions assessing participants’ experiences of the Troubles were adapted from the NI Poverty and Social Exclusion (PSE) study (more information at http://​www.​poverty.​ac.​uk/​pse-research/​legacies-troubles), 2012 version (see Table 1). Since associated subjective trauma was not reported, these items could instead be described as querying potentially traumatic events. Data were available for 3275 individuals. Questions on experiences during the Troubles were grouped in the SCQ into those about “death” (question 69; whether an individual experienced a close friend, relative, or other individual dying as a result of the Troubles), “injury” (question 70; whether an individual experienced a personal injury or the injury of a close friend, relative, or other individual as a result of the Troubles), “witness” (question 71; whether the individual witnessed a bomb explosion, murder, gunfire, rioting, or another individual being assaulted during the Troubles). These three groupings were specified as latent variables in an initial measurement model. A fourth grouping, “authorities” (questions 72–77; whether the individual had interactions with the authorities such that they were either sent to prison for Troubles-related activity, or had their home searched by police or army) was also specified in the model, but it was later removed because of poor loading. The measurement model (and all subsequent analyses) were conducted using the lavaan package in R Studio (Rosseel 2012). All questions in Table 1 were coded such that 0 indicated that this experience had not been reported by the respondent, and 1 indicated that it had been reported by the respondent. The model converged normally after 61 iterations. Fit was acceptable, χ²₅₁ = 1412, CFI = 0.90, TLI = 0.87, RMSEA = 0.084 (CI₉₀ = 0.080, 0.088), SRMR = 0.046. Factor loadings were all acceptable and above 0.4. There were significant covariances among the latent factors (0.881 between Death and Injury; 0.593 between Death and Witness; 0.689 between Injury and Witness, all significant at p < 0.001). A second measurement model with a higher-order latent factor (Troubles Exposure) was also tested but this did not converge. A third measurement model was conducted which included Memory as an additional latent factor, and this model is described in Table 2, below.

Immediate and Delayed Word Recall are tasks evaluating episodic memory adapted from the version used in the Health and Retirement Study. For the immediate recall task, one (of four) randomly selected list of ten words was read to the participants and participants were tasked with repeating these ten words, yielding a score out of 10. Immediately after this, participants were given a second opportunity to recall the same ten words, yielding a second score of Immediate Word Recall. After a delay, during which the participants answered the CAPI section on health, they then recalled the words for a third time, and their score out of 10 was taken as the Delayed Word Recall measure.

An “Animal Naming” task of verbal fluency was used to evaluate semantic memory. This task involves instructing participants to name as many animals as they can in a sixty-second timeframe. This task is also used in the Health and Retirement Study and other harmonised cohort studies of ageing. A latent “Memory” factor with four indicators (Immediate word Recall 1 and 2; Delayed Word Recall; Animal Naming) was tested using a second measurement model that also included the latent Troubles measures described above. A similar approach was previously taken by the authors using data from the Irish Longitudinal Study on Ageing (McHugh Power et al. 2019).

This model converged normally after 72 iterations and model fit was good, χ²₉₈ = 1589, CFI = 0.939, TLI = 0.925, RMSEA = 0.049 (CI₉₀ = 0.047, 0.052), SRMR = 0.046. All memory items loaded well (minimum factor loading of 0.510; see Table 2). There were significant correlations between Memory and Injury, r = 0.059, p = 0.011; and between Memory and Witness, r = 0.153, p < 0.001, but not between Memory and Death, r = 0.021, p = 0.400.

Choice of covariates was based on the following principles (VanDerWeele 2019): control for covariates that are causes of the exposure, or of the outcome, or both (including proxies for unmeasured variables that cause both), except for those that qualify as instrumental variables. We could not find theoretical justification to expect predictors of traumatic exposure to the Troubles. We included, as known causes of memory decline, age in years, gender (male or female), education level (operationalized as an ordinal variable with three levels as per TILDA: no qualifications, second level qualification, third level or higher qualification), and depressive symptoms (a single item about frequency of feelings of depression, responded as a five point scale from none to extreme). For a subset of participants (n = 1713), information was available on a question about time of the worst events of the Troubles (“Thinking of the worst thing that happened to you because of the Troubles, when was this?”) with categorical responses (1969–1973; 1974–1978; 1979–1983; 1984–1988; 1989–1993; 1994–1998, after 1998). The middle year of each category was used to calculate an approximate age of each of 1713 participants during the worst thing that had happened to them because of the Troubles (see Table 3), and this was also included in the model as a covariate.

An initial model was performed with main effects only and excluding the moderation effect to provide a baseline comparator. Bonferroni corrections for multiple comparisons, therefore, set alpha at 0.025. The model converged normally after 79 iterations and demonstrated acceptable fit, χ²₆₉ = 1084, CFI = 0.94, TLI = 0.92, RMSEA = 0.061 (CI₉₀ = 0.058, 0.064), SRMR = 0.077. Sample-size-adjusted Bayesian Information Criterion (BIC) was 202480, and Akaike Information Criterion (AIC) was 202310. Being younger, β = − 0.419, p < 0.001, male, β = 0.114, p < 0.001, having lower depression, β = − 0.119, p < 0.001, higher education, β = 0.254, p < 0.001, higher levels of Troubles Exposure, β = 0.048, p = 0.01, and higher levels of Social Activity Engagement, β = 0.102, p < 0.001, were all associated with higher levels of the latent Memory variable.

The model converged normally after 78 iterations and demonstrated acceptable fit, χ²₇₇ = 1272, CFI = 0.94, TLI = 0.92, RMSEA = 0.049 (CI₉₀ = 0.046, 0.051), SRMR = 0.066. AIC was 232493 and sample-size-adjusted BIC was 232702. In the measurement component of the model, memory was included as a latent variable with four indicators, Immediate Recall 1 and 2, Delayed Recall, and Animal Recall, which had factor loadings of 0.826, 0.876, 0.858, and 0.530 respectively. Troubles Exposure was included as a latent variable with three indicators, Death, Witness, and Injury, all mean-centred prior to inclusion, and with factor loadings of 0.702, 0.873, and 0.638 respectively. An interaction product term was built as a latent factor, Interaction, with three indicators, Death, Injury, and Witness, each separately multiplied by Social Activity Engagement (mean-centred), which achieved factor loadings of 0.756, 0.658, and 0.816 respectively. Comparing AIC and sample size adjusted BIC between the model with moderation and the model without, fit was better in the version without (χ² difference = 187.3, p < 0.001, df difference = 34). Memory was associated with being younger, β = − 0.42, p < 0.001, being male, β = 0.113, p < 0.001, greater education level, β = 0.253, p < 0.001, and lower depression, β = − 0.131, p < 0.001, higher Troubles Exposure, β = 0.053, p = 0.007, higher levels of Social Activity Engagement, β = 0.106, p < 0.001. The interaction term between Troubles Exposure and Social Activity Engagement was significant at β = − 0.054, p = 0.01 (see Table 4). A negative interaction term can be interpreted to mean that at higher levels of Social Activity Engagement, there is less of an effect of Troubles Exposure on Memory (see Fig. 1).