- Email: [email protected]
Evidence of Big-Five personality changes following acquired brain injury from a prospective longitudinal investigation
Journal of Psychosomatic Research 82 (2016) 17–23
Contents lists available at ScienceDirect
Journal of Psychosomatic Research
Evidence of Big-Five personality changes following acquired brain injury from a prospective longitudinal investigation Anne Leonhardt ⁎, Stefan C. Schmukle, Cornelia Exner Department of Psychology, University of Leipzig, Neumarkt 9-19, D-04081 Leipzig, Germany
a r t i c l e
i n f o
Article history: Received 25 June 2015 Received in revised form 22 December 2015 Accepted 13 January 2016 Available online xxxx Keywords: Acquired brain injury Big Five HILDA Panel data Personality change
a b s t r a c t Objective: Many studies using different assessment methods have reported personality changes after acquired brain injury (ABI). However, to our knowledge, no prospective study has yet been conducted to examine whether previous cross-sectional and retrospective results can be replicated in a longitudinal prospective design. Further, because clinical control groups were only rarely used, it remains debatable if the personality changes found are unique to patients with ABI or if they also affect patients with other disabilities. Methods: This study examined personality change in 114 participants with different kinds of ABI, 1321 matched controls (general control, GC), and 746 matched participants with restrictive impairments other than brain injury (clinical control, CC) in a prospective longitudinal design using data from the panel survey Household, Income and Labour Dynamics in Australia (HILDA). Results: Participants with ABI showed significantly larger declines in Extraversion and Conscientiousness compared with the GC group. When the ABI participants were compared with the CC group, only the difference in Conscientiousness remained significant. Conclusion: Our prospective data corroborate evidence from previous cross-sectional studies that patients with ABI experience larger declines in Extraversion and Conscientiousness than the general population. Whereas the effect on Conscientiousness was unique to patients with ABI, the decline in Extraversion was also observed in participants with other impairments. © 2016 Elsevier Inc. All rights reserved.
1. Introduction Many studies have reported that acquired brain injury (ABI) can lead to personality changes [1–5]. These changes are typically associated with a worse psychosocial outcome [6,7], especially in terms of heightened depression and anxiety [8,9], reduced activities [10], increased social difficulties [11], and worse health-related quality of life [5]. Previous studies have assessed changes in personality after ABI in very different ways. Investigations focusing on changes in personality structure, as assessed by the Big Five dimensions (Neuroticism or Emotional Stability, Extraversion, Openness to Experience, Conscientiousness, and Agreeableness) [12], have reported a decline in Extraversion, Conscientiousness, or Emotional Stability [13,14,7,2,15] with a decrease in Extraversion being the most consistent finding. However, these previous studies have suffered from two major methodological flaws: The main limitation of all previous approaches has been the retrospective nature of the designs. Retrospective data acquisition is supposed to be less accurate than prospective data
⁎ Corresponding author. E-mail address: [email protected] (A. Leonhardt).
http://dx.doi.org/10.1016/j.jpsychores.2016.01.005 0022-3999/© 2016 Elsevier Inc. All rights reserved.
acquisition because memory is affected by several biases including the current mood bias [16], the self-serving bias [17] or the fading affect bias [18]. Previous studies [14,7,2,15] have tried to limit these effects by assessing pre- and post-injury personality at separate points in time after injury. Even though the bias was not completely removed that way, this still led to reductions in reports of personality changes, underlining the concern that at least some of the effect might be due to a retrospective bias. The second major methodological limitation of previous designs is related to the fact that most studies have lacked a clinical control group. This issue is especially important as a clinical control group may help to generate hypotheses about possible underlying causes of personality changes after brain damage. So far, most authors have suggested a neurological basis such as that damage to the frontotemporal or more specifically the ventromedial or orbital prefrontal cortex has caused personality change [19–21]. However, if personality changes found after ABI closely resemble those found in other medical conditions that do not involve brain damage, other causes should be discussed: Problems in psychosocial adjustment such as changes in social roles, a loss of personal goals, values that became unreachable, anxiety, and withdrawal from social activity might also be responsible for personality changes [22]. Lannoo et al.'s [13] and Rush et al.'s [7]
18
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
studies, which found no differences in personality changes between individuals with ABI and those with other disabilities, suggest that personality changes are related to difficulties dealing with the disability rather than neurological causes. To the best of our knowledge, the current study is the first to investigate the effects of ABI on personality using a prospective longitudinal design. Our aims were, first, to determine whether people with ABI experience stronger personality changes than the general population and whether we can replicate the results of previous retrospective studies in a prospective design. Second, we wanted to clarify whether personality changes are unique to patients with ABI or also affect patients with other restrictive impairments. In addition, we aimed to investigate whether the association between ABI and personality change would be mediated by a reduction in health-related quality of life (HRQoL). We expected to find no mediation effect for physical HRQoL but for mental HRQoL which is assumed to resemble problems in psychosocial adjustment. This finding would support the hypothesis that problems in psychosocial adjustment should be considered as one underlying reason for personality changes besides neurological causes.
2. Method 2.1. Participants We used data from the Household, Income, and Labour Dynamics in Australia survey (HILDA). The HILDA panel study was initiated and is funded by the Australian Government Department of Social Services. It is subject to oversight and approval by the University's Office for Research Ethics and Integrity. With regard to the use of the data, all HILDA data users are required to sign the license agreement, which legally binds them to use the data for bona fide research purposes. In this nationally representative panel survey, which began in 2001, individuals from more than 6500 Australian households are interviewed every year and fill out several questionnaires [23]. In 2005 (t1) and 2009 (t2), the interview included a Big Five questionnaire. Therefore, we used data from 2005 to 2009. Of the 16,373 people who participated in the survey between t1 and t2, we excluded people who had already reported ABI in t1 or who had missing data on their t1 health status, their education, or their personality at t1 or t2. Therefore, 8322 participants between the ages of 15 and 92 (3801 men and 4521 women) were included in the study.
2.2. Main outcome variable: personality All participants filled out a 36-item scale based on Saucier's Mini Markers Scale [24] to assess their Big Five personality dimensions at t 1 and t2 . Each item consisted of a single adjective (e.g. “orderly”, “shy”), and the respondents had to rate how well each adjective described them using a Likert scale that ranged from 1 (does not describe me at all) to 7 (describes me very well). On the basis of previous factor analyses and reliability analyses, eight adjectives had to be excluded due to simultaneous loadings on several factors [25]. The other adjectives were assigned to one of the Big Five dimensions. Participants who had answered less than 75% of the items for one dimension were excluded from the analyses (i.e. at most one item could be missing in each dimension). The participants had up to five missing values which were replaced by the mean of the other items. Over 95% of the participants had no missing values. We calculated T-scores using the data of all participants at t1 (N = 8322) and used the difference between the t 2 personality T-score and the t1 personality T-score as an indicator of personality change. Therefore, a negative change score represented a decline in the respective personality variable.
2.3. Main predictor variable: participants with acquired brain injury (ABI) and control groups The respondents were asked every year for 5 years (i.e. 2005 = t1, 2006, 2007, 2008, and 2009 = t2) if they had any long-term health condition, impairment, or disability that restricted their everyday activities and that had lasted or was likely to last for 6 months or more. Respondents answering this question in the affirmative were then shown a list of medical conditions from which they had to name one or more, e.g. “hearing problems”, “a nervous or emotional condition which requires treatment”, or “chronic or recurring pain”. One of the items from the list was “long term effects as a result of a head injury, stroke or other brain damage”. This item was used to divide the dataset into a group with ABI and control participants. As already mentioned above, we excluded people who had already affirmed this item at t1, because we were interested in analyzing personality change after having acquired a brain injury in a prospective design. Next we describe how participants were divided into three groups based on whether they had acquired a brain injury or another impairment between t1 and t2: 2.3.1. Patients with acquired brain injury (ABI) The group of patients with an acquired brain injury (ABI, n = 114) contained all respondents who had at least once given an affirmative answer to the question about “long term effects as a result of a head injury, stroke, or other brain damage” at any of the four time-points after t1 (but not at t1). As the question did not differentiate between the various forms of brain injury, it was impossible to tell how many of these participants suffered a stroke, a traumatic brain injury, or another form of brain injury. These participants could also have other disabilities in addition to ABI. 2.3.2. Global control (GC) group Our global control (GC) group (n = 8208) contained all other participants and therefore included healthy people as well as people with impairments other than ABI at any time-point. This group can be considered as representative of the population of Australia with the difference that people with ABI are excluded here. 2.3.3. Clinical control (CC) group We additionally created a subgroup of the GC group that included only those participants who had acquired a disability after t1 (clinical control (CC) group, n = 3154). These participants had indicated a newly occurring disability other than ABI from the list of medical conditions at any time-point after t1, which had not already existed at t1. They could however have had other kinds of pre-existing impairments at t1. 2.4. Control variables and matching procedure In a second step, we performed a matching procedure to create groups that are equal on several variables that could influence personality change, such as age, sex, education, preexisting impairments, baseline personality, and health-related quality of life (HRQoL). 2.4.1. Health-related quality of life (HRQoL) The SF-36 Health Survey was used to assess health-related quality of life (HRQoL) on two global outcome scales: physical and mental HRQoL [26]. Both scales were transformed into T-scores using data from the general U.S. population [26] (with M = 50, SD = 10) with higher scores indicating better HRQoL. The respondents filled out the questionnaire annually. We used the SF-36 scores of 2005 (= t1) and 2009 (= t2). We calculated the difference between the t2 SF-36 scores and the t1 SF-36 scores to estimate changes in physical and mental HRQoL with negative values indicating a decline in HRQoL. Missing values were replaced by the mean of the other items from the subscale, only if more than 75% of the items were filled out. Over 94% of the participants had
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
no missing values. The maximum number of missing values was six. HRQoL was used in the analyses as a matching variable but also as a potential mediating variable. In the matching procedure, missing scale scores for physical and mental HRQoL were estimated by single imputation. In the mediation model, participants with missing scale scores for physical and mental HRQoL were excluded from the analysis. 2.4.2. Age, sex, education Participants indicated their age and sex in the questionnaires as well as their highest level of education obtained. These levels were then transformed into levels of the International Standard Classification of Education (ISCED-levels). The ISCED differentiates between eight levels of education ranging from 1 (early childhood education) to 8 (doctoral or equivalent level) [27]. 2.4.3. Propensity score matching The groups differed on several variables that could influence personality change (age, sex, education, pre-existing impairments, HRQoL at t1, personality at t1). Therefore, we used propensity score matching to create control groups that would be comparable to the ABI group on these potentially confounding variables. Propensity score matching is a key tool for estimating causal effects in a quasi-experimental design by accounting for the covariates that differ between the groups [28]. Here, a participant's propensity score reflects the predicted probability of belonging to the ABI group [29]. To match the ABI and GC groups, we estimated the propensity scores by regressing the group membership on the potentially confounding variables, personality scores at t 1, pre-existing impairments (i.e., when the respondent named a medical condition other than a brain injury at t1), HRQoL scores at t1, age, sex and education. As we used single imputations to replace missing values in t1 HRQoL, we also included a dummy variable in the matching procedure to indicate missing data as suggested by Stuart [28]. We used a 1:k nearest neighbor matching scheme to match each participant from the ABI group with a larger number of participants from the GC group using the MatchIt package in R [30,31]. The participants in the GC group were drawn with replacement and were given a weight to indicate how often they were drawn for different participants from the ABI group. We tested different ratios and examined the quality of the matching by checking the standardized mean difference (standardized by the standard deviation of the ABI group) between the ABI and GC groups on each covariate. The best fit between the ABI and GC groups was found for a 1:16 matching with a resulting matched GC group consisting of 1321 participants. After matching, propensity scores were nearly identical for the ABI group with .048 and for the GC group with .048 (standardized mean difference b 0.001). Most
19
importantly, for all covariates standardized mean differences between the groups were negligible with values smaller than .04 after matching (see Table 1). This indicates that the matching procedure was successful and that the matched control group was comparable to the ABI group on all potentially confounding variables. To match the ABI and CC groups, we used the same covariates for matching as before, but additionally included change in HRQoL between t1 and t2 as a further matching variable. The intention was to create two groups of participants who had not only been similar at t1 but had also acquired a disability after t1 that had a similar impact on the change in HRQoL between t1 and t2. As we used single imputations for missing values in HRQoL, we also included an additional dummy variable to indicate missing data in the change in HRQoL. The best fit occurred for a 1:9 matching, leading to a matched CC group comprising 746 participants. The matching was again successful, as shown by nearly identical propensity scores of .060 for the ABI group and of .060 for the CC group (standardized mean difference: 0.007). Additionally, for all covariates standardized mean differences between the groups were smaller than .07 after matching (see Table 1). Due to the matching procedure, participants of the matched control groups obtained different weights. All further analyses that were conducted on these matched GC and CC groups thus considered these individual weights. 2.5. Statistical analysis To investigate group differences in personality change, we computed separate weighted t-tests on change for each personality dimension. The effect size Cohen's d was calculated for each personality dimension. To verify the results of the t-tests in matched groups, we computed ANCOVAs using the original samples that included the potentially confounding variables, personality scores at t1, pre-existing impairments, HRQoL scores at t1, age, sex and education and for the comparison with the CC group the change in HRQoL as covariates. In a mediation model using weighted linear regression, we explored the indirect effect of ABI on personality change through changes in physical and mental HRQoL as mediators. The significance of the regression coefficients was determined by t-tests. To assess the significance of the mediation effect, we tested whether the product of the paths (ab = c-c′) was significantly different from zero as suggested by Preacher and Hayes [32]. As the distribution of ab was not assumed to be normal, we calculated 95% nonparametric bias-corrected and accelerated bootstrap confidence intervals based on 10,000 bootstrap samples. The effect of the mediator is significant at the 5% level if the 95% bootstrap confidence interval does not include zero. We used two-tailed tests with α = .05 as the level of significance for all tests. All statistical analyses were conducted using the statistics program R [33].
Table 1 Sample demographics for matched groups. ABI group
GC group
n (%)
n (%)
Total Sex (female) Pre-existing impairments
114 58 (50.9) 89 (78.1)
1321 674 (51.0)a 1012 (76.6)a
Age ISCED-level Physical HRQoL t1 Change in physical HRQoL t2 − t1 Mental HRQoL t1 Change in mental HRQoL t2 − t1
M (SD) 57.44 (15.51) 3.18 (1.54) 37.91 (11.74) −1.73 (8.87) 46.70 (12.48) −1.49 (9.42)
M (SD) 57.71 (16.35)a 3.13 (1.46)a 37.74 (11.95)a 0.84 (9.12)a 46.78 (11.79)a 0.88 (10.73)a
s. m. d. (ABI/GC)
−0.002a −0.034a
−0.018a 0.027a 0.015a −0.007a
CC group n (%)
s. m. d. (ABI/CC)
746 382 (51.2)a 577 (77.4)a
−0.006a −0.016a
M (SD) 57.60 (15.99)a 3.16 (1.49)a 37.54 (11.61)a −1.54 (9.47)a 46.93 (11.61)a −2.22 (10.95)a
−0.011a 0.010a 0.032a −0.021a −0.019a 0.025a
Note. ABI = acquired brain injury, GC = global control, CC = clinical control, ISCED = International Standard Classification of Education, HRQoL = health-related quality of life, s. m. d. = standardized mean difference = difference between the means of both groups standardized by the standard deviation of the ABI group, t1 = 2005, t2 = 2009. a Based on matched samples with weighted individuals.
20
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Table 2 Personality T-scores in the study groups. ABI group (n = 114)
E
C
ES
O
A
GC group (n = 1321) M
Group difference (ABI/GC)
M
SD
SD
t
t1 t2 t2 − t1
47.25 44.83 −2.41
14.02 13.81 10.81
47.38a 47.57a 0.18a
14.80a 14.68a 11.22a
t1 t2 t2 − t1
49.48 46.37 −3.11
14.99 14.77 10.67
49.20a 48.85a −0.35a
15.47a 15.15a 11.97a
a
a
t1 t2 t2 − t1
50.12 49.93 −0.19
15.09 15.34 12.42
49.98 51.97a 1.99a
16.25 14.94a 13.02a
t1 t2 t2 − t1
50.54 48.27 −2.27
16.97 16.93 13.36
51.15a 48.30a −2.86a
15.30a 16.61a 13.11a
13.64 16.00 14.82
a
a
t1 t2 t2 − t1
50.13 48.52 −1.60
50.23 48.50a −1.73a
15.57 47.20a 16.30a
df
−2.38
−2.38
−1.73
0.46
0.08
1433
1433
1433
1433
1433
CC group (n = 746) p
.018
.017
.085
.646
.937
Cohen's d [95% CI]
−0.23 [−0.42, −0.04]
−0.23 [−0.42, −0.04]
−0.17 [−0.36, 0.02]
0.04 [−0.15, 0.24]
0.008 [−0.18, 0.20]
Group difference (ABI/CC)
M
SD
t
df
p
Cohen's d
47.89a 46.59a −1.30a
14.99a 14.90a 11.65a
−0.96
858
.339
−0.10 [−0.29, 0.10]
48.95a 48.68a −0.27a
15.56a 15.49a 12.36a
−2.32
858
.021
−0.23 [−0.43, −0.04]
a
a
49.96 50.99a 1.03a
15.33 15.78a 13.74a
−0.89
858
.373
−0.09 [−0.29, 0.11]
50.82a 48.78a −2.04a
15.61a 16.20a 12.44a
−0.18
858
.858
−0.02 [−0.22, 0.18]
a
a
−0.87
858
.385
−0.09 [−0.28, 0.11]
49.18 48.82a −0.36a
16.03 16.77a 14.09a
Note. E = Extraversion, C = Conscientiousness, ES = Emotional Stability, O = Openness, A = Agreeableness, ABI = acquired brain injury, GC = global control, CC = clinical control, Cohen's d = difference between the means of both groups standardized by the pooled variance, t1 = 2005, t2 = 2009. a Based on matched samples with weighted individuals.
3. Results 3.1. Personality changes in the ABI and GC groups The mean scores and standard deviations of the five personality domains at t1 and t2 for the ABI group and the GC group and the statistical characteristics of the t tests are presented in Table 2. The courses of personality change for the ABI and GC groups are shown in the left column of Fig. 1. The groups differed significantly in their change in Extraversion (Cohen's d = −0.23, p = .018). Whereas the GC group did not change, t(1320) = 0.59, p = .552, the ABI group showed a significant decline in Extraversion of 2.41 T-scores, t(113) = −2.38, p = .019. The two groups also differed in their change in Conscientiousness (Cohen's d = −0.23, p = .017). Again, there was no change in the GC group, t(1320) = −1.06, p = .29, whereas in the ABI group Conscientiousness declined 3.11 T-scores, t(113) = −3.11, p = .002. In addition, there was a trend towards a group difference in the change in Emotional Stability, which was, however, not significant (Cohen's d = −0.17, p = .085). Here, the GC group showed an increase of 1.99 T-scores over time, t(1320) = 5.56, p b .001, a change that was not apparent in the ABI group, t(113) = − 0.16, p = .870. The two groups did not differ in their change in Openness to Experience and Agreeableness. Openness to Experience, t(1320) = − 7.92, p b .001, and Agreeableness, t(1320) = − 3.85, p b .001, declined in the GC group, whereas the ABI group did not change in Openness to Experience, t(113) = −1.81, p = .073, and Agreeableness, t(113) = −1.16, p = .250. However, these differences between the groups were not significant.1 3.2. Personality changes in the ABI and CC groups The mean scores and standard deviations of the five personality domains at t1 and t2 for the CC group and the statistical characteristics of 1 The ANCOVAs revealed the same results showing a greater decline in Extraversion, t(8309) = −2.82, p = .005, and Conscientiousness, t(8309) = −2.77, p = .006, in the ABI group compared to the GC group, and no group difference in the change in Emotional Stability, t(8309) = −1.60, p = .109, Openness to Experience, t(8309) = −0.43, p = .666, and Agreeableness, t(8309) = −0.87, p = .385.
the t tests are reported in Table 2. The course of personality change in the CC group is also shown in the right column of Fig. 1. The previously found group difference in Extraversion was not significant in this sample. Similar to the ABI group, which experienced a decline of 2.41 T-scores, the CC group also showed a decline in Extraversion that was, however, much smaller (1.30 T-scores), t(745) = − 3.05, p = .002. The size of Cohen's d indicates a smaller group difference between the ABI and CC groups (Cohen's d = − 0.10, p = .339) than between the ABI and GC groups (Cohen's d = −0.23). By contrast, the ABI and CC groups showed significant differences in the change in Conscientiousness. The significant decline of 3.11 T-scores in the ABI group was not apparent in the CC group, t(745) = − 0.60, p = .549. Here, the effect size for the group difference between the ABI and CC groups (Cohen's d = − 0.23, p = .021) was the same as for the analysis with the GC group (Cohen's d = −0.23). There were no group differences in the changes in Emotional Stability, Openness to Experience or Agreeableness.2 Similar to the GC group, the CC group showed a significant increase in Emotional Stability, t(745) = 2.04, p = .042, and a significant decline in Openness, t(745) = −4.48, p N .001. The CC group did not change in Agreeableness, t(745) = −0.71, p = .480. 3.3. Influence of reduced physical and mental HRQoL on personality change In the mediation model, we used changes in physical and mental HRQoL as mediators. Participants with missing scale scores for physical and mental HRQoL were excluded from the mediation analysis. Therefore, 1183 people of the GC group and 100 people of the ABI group were included. Changes in HRQoL were estimated by calculating the difference between the t2 SF-36 scores and the t1 SF-36 scores, with negative values indicating a reduction in HRQoL. The mediation model shown in Fig. 2 illustrates the influence of group (GC = 0, ABI = 1) on 2 The results of the ANCOVAs were again similar to the t-tests in matched groups with a greater decline in Conscientiousness, t(3253) = −2.06, p = .039 in the ABI group and no group differences in the change in Emotional Stability, t(3253) = −0.74, p = .456, Openness to Experience, t(3253) = −0.17, p = .868, and Agreeableness, t(3253) = −0.38, p = .700. However, in contrast to the t-test with matched groups, the ANCOVA also revealed a greater decline in Extraversion for the ABI group compared to the CC group, t(3253) = −2.01, p = .044.
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
21
effect of ABI on the reduction in Extraversion after controlling for reductions in physical and mental HRQoL was lower, but still marginally significant (c′ = −2.14, t(1279) = −1.84, p = .066), indicating that the decline in Extraversion for participants with ABI was partly mediated by the reduction in mental HRQoL. The mediation model for Conscientiousness revealed similar results. There was no significant relationship between the change in physical HRQoL and the change in Conscientiousness (b1 = 0.03, t(1279) = 0.73, p = .465), and the bootstrap confidence interval for the indirect effect of ABI on the change in Conscientiousness through the change in physical HRQoL (a1b1 = − 0.08) was not significant (95% CI [−.38, .16]). A reduction in mental HRQoL was significantly related to a reduction in Conscientiousness (b2 = 0.12, t(1279) = 3.86, p b .001) with a bootstrap confidence interval for the indirect effect of ABI on the reduction in Conscientiousness through the reduction in mental HRQoL (a2b2 = −0.34) that was also significant (95% CI [−.96, −.21]). A decrease in Conscientiousness for participants with ABI was only partly mediated by reductions in physical and mental HRQoL (c′ = − 3.06, t(1279) = −2.48, p = .013). 4. Discussion
Fig. 1. Mean personality scores at t1 and t2 with 95% confidence intervals for the acquired brain injury (ABI), global control (GC), and clinical control (CC) groups. t1 = 2005, t2 = 2009.
the changes in Extraversion and Conscientiousness mediated by the change in HRQoL. The ABI group showed a larger decline in physical (a1 = − 3.17, t(1281) = − 3.33, p b .001) and mental HRQoL (a2 = − 2.81, t(1281) = −2.53, p = .011) than the GC group did. There was only a marginally significant relationship between the change in physical HRQoL and the change in Extraversion (b1 = 0.05, t(1279) = 1.61, p = .108), and the bootstrap confidence interval for the indirect effect of ABI on the change in Extraversion through the change in physical HRQoL (a1b1 = −0.17) was not significantly different from zero (95% CI [−.26, .18]). A reduction in mental HRQoL was significantly related to a reduction in Extraversion (b2 = 0.11, t(1279) = 3.78, p b .001), and the bootstrap confidence interval for the indirect effect of ABI on the reduction in Extraversion through the reduction in mental HRQoL (a2b2 = −0.31) was also significant (95% CI [−.72, −.12]). The direct
To our knowledge, this is the first study to investigate personality changes after ABI using a prospective longitudinal design. The main finding is that personality changed differently in the ABI group than in the two control groups. Similar to the results of previous studies that used retrospective designs [13,15,14], Extraversion and Conscientiousness scores declined significantly more in the ABI group than in the GC group. The previously found decrease in Emotional stability for patients with ABI [15] was not found in our samples although there was a non-significant trend towards a greater decline in the ABI group. The results were irrespective of the statistical method (t-tests in matched groups vs. ANCOVA). It is important to note that we were able to replicate most of the findings from previous studies although the design (longitudinal vs. retrospective) and the rater (most previous studies used significant other ratings, whereas we used self-ratings) differed. However, the use of different raters could explain why we could not replicate the decrease in Emotional Stability. Altered emotional reactivity might be experienced differently from the first-person vs. thirdperson perspective. Our results confirm that personality changes in patients with ABI are a robust finding that does not only appear in retrospective designs. To investigate whether personality changes are unique to patients with ABI or if they also affect people with other impairments as previous studies have suggested [13,7], we additionally compared the ABI group with a subgroup of the GC group containing only those participants who had sustained a similarly restrictive disability other than a brain injury. In comparison with this CC group, the difference in the change in Extraversion disappeared, a result that is consistent with previous findings [13,7]. Although the effect size showed a larger decline in patients with ABI than in the clinical and global control groups, the change in Extraversion did not seem to be unique to patients with ABI which suggests that psychosocial adaptation processes following a serious disability have to be considered as potential underlying reasons for personality changes in addition to structural and functional brain damage. By contrast, the difference in the change in Conscientiousness remained significant. Conscientiousness declined more in patients with ABI than in patients with other disabilities and therefore seems to be unique to patients with ABI in contrast to the change in Extraversion. As the Conscientiousness items used in this study assess orderliness and efficiency traits which describe intact executive functions, it can be assumed that the decline in Conscientiousness is also related to neuropsychological impairments, which are often reported after ABI [34,35]. The results of the ANCOVA differed from that of the t-tests in matched groups in one point. Whereas the ANCOVA revealed a significant group difference in Extraversion, this was not true for the t-test in matched groups. This
22
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Fig. 2. Mediation model for changes in Extraversion and Conscientiousness. ABI = acquired brain injury, GC = global control, HRQoL = health-related quality of life. Unstandardized regression coefficients are reported. *p b .05, **p b .01, ***p b .001.
additional significant result might be due to the greater test power of the ANCOVAS, which used larger sample sizes. However, the matchingprocedure gives the more reliable results as it allows for the selection of a control group that is similar to the treatment group concerning possible confounding variables. According to Miller and Chapman [36] ANCOVA does not adequately control for group differences. Previous studies that compared ABI patients with other patient groups found no differences in personality changes [13,7]. One explanation for this difference could be the longer time frame in our study. Whereas previous studies assessed post-ABI personality one to two years after the injury, the time span could be up to 4 years in our study and some changes might take time to develop. As expected, the changes in Extraversion and Conscientiousness in the ABI group were mediated by mental but not by physical HRQoL which means that personality changes are more related to mental health problems than to physical limitations. This supports the importance of psychological influences including changed social roles, social withdrawal, or mental disorders, that lead to a changed selfperception and a changed evaluation by significant others [22]. Beyond the mediation effect, however, ABI still showed a significant relationship with the changes in Conscientiousness and Extraversion. This may be seen as support of the biopsychosocial model of personality changes which proposes that both neurobiological factors and psychosocial adaptation processes affect personality change in people with ABI [22]. 4.1. Limitations The major limitation of the study is that the indication of ABI could not be validated by a medical examination. As is true of all large panel studies, we had to rely on the statements made by the participants. Nevertheless, the similarity of our results to those of previous studies that used retrospective designs and full clinical diagnostics including neuroimaging in medical facilities supports the validity of the participants' statements. As the question about brain injury on which we built our ABI group did not differentiate between various etiologies of brain injury, we did not know how many of these participants suffered a stroke, a traumatic brain injury, or another form of brain injury. Accordingly, we could not determine whether the personality changes we found depended on the etiology of brain damage. Lastly, although our longitudinal data allowed us to compute the prospective effect of ABI on personality, this effect may only be interpreted as a causal effect after controlling for all confounding
variables. Although we are confident that we indeed considered the most important confounds in our matching analyses by conditioning on sex, age, education, and both personality and health status variables before brain injury, this is no guarantee that the groups are equal concerning all potentially confounding variables. 4.2. Conclusions and directions for future research The findings of this study indicate that people with ABI experience larger declines in Extraversion and Conscientiousness compared with the general population over a time period of 4 years. Compared with patients with other impairments, there was no significant difference in the reduction in Extraversion, whereas the reduction in Conscientiousness seemed to be unique to patients with ABI. Future studies are needed to examine whether the decline in Conscientiousness reflects neuropsychological impairments. Studies with full clinical ABI diagnostics and further specifications of clinical features should be conducted to investigate which moderators and clinical mediators influence personality change in patients with ABI. According to our findings, it seems reasonable to conclude that neurological causes as well as psychosocial adaptation processes affect personality change in people with ABI. As the ABI group comprised individuals with different types of brain injury and diverse severity, the changes should be seen as a general trend for people with different kinds of ABI. This trend is apparently different from that in the general population. Therefore, personality change should be considered in every neurological rehabilitation program although it will not affect every patient. Furthermore, the fact that psychosocial adaptation processes seem to affect personality changes stresses the importance of psychotherapy for people with ABI to help them adapt to the changed circumstances. Most importantly, our study shows that personality changes following ABI are not only observable when assessing personality retrospectively, but are also observable in a prospective longitudinal design, underlining the robustness of this effect. Conflict of interest No conflicts of interest declared. Funding sources No external funding.
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Acknowledgments This paper used unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Social Services (DSS) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either DSS or the Melbourne Institute. References [1] M. Milders, M. Ietswaart, J.R. Crawford, D. Currie, R. Adolphs, S. Baron-Cohen, D. Tranel, H. Bibby, S. McDonald, N. Bogod, Social behavior following traumatic brain injury and its association with emotion recognition, understanding of intentions, and cognitive flexibility, J. Int. Neuropsychol. Soc. 14 (2008) 318–326. [2] R.L. Tate, Impact of pre-injury factors on outcome after severe traumatic brain injury: does post-traumatic personality change represent an exacerbation of premorbid traits? Neuropsychol. Rehabil. 13 (2003) 43–64. [3] J.E. Max, Predictors of personality change due to traumatic brain injury in children and adolescents six to twenty-four months after injury, J. Neuropsychiatr. 18 (2006) 21–32. [4] M.R. Hibbard, J. Bogdany, S. Uysal, K. Kepler, J.M. Silver, W.A. Gordon, L. Haddad, Axis II psychopathology in individuals with traumatic brain injury, Brain Inj. 14 (2000) 45–61. [5] A.P. Diaz, M.L. Schwarzbold, M.E. Thais, A. Hohl, M.M. Bertotti, R. Schmoeller, J.C. Nunes, R. Prediger, M.N. Linhares, R. Guarnieri, R. Walz, Psychiatric disorders and health-related quality of life after severe traumatic brain injury: a prospective study, J. Neurotrauma 29 (2012) 1029–1037. [6] C.J. Cattran, M. Oddy, R.L. Wood, J.F. Moir, Post-injury personality in the prediction of outcome following severe acquired brain injury, Brain Inj. 25 (2011) 1035–1046. [7] B.K. Rush, J.F. Malec, A.W. Brown, A.M. Moessner, Personality and functional outcome following traumatic brain injury, Rehabil. Psychol. 51 (2006) 257–264. [8] J.B. Cantor, T.A. Ashman, M.E. Schwartz, W.A. Gordon, M.R. Hibbard, M. Brown, L. Spielman, H.J. Charatz, Z. Cheng, The role of self-discrepancy theory in understanding post-traumatic brain injury affective disorders: a pilot study, J. Head Trauma Rehabil. 20 (2005) 527–543. [9] I. Aben, J. Denollet, R. Lousberg, F. Verhey, F. Wojciechowski, A. Honig, Personality and vulnerability to depression in stroke patients: a 1-year prospective follow-up study, Stroke 33 (2002) 2391–2395. [10] J. Stone, E. Townend, J. Kwan, K. Hoga, M.S. Dennis, M. Sharpe, Personality change after stroke: some preliminary observations, J. Neurol. Neurosurg. Psychiatry 75 (2004) 1708–1713. [11] M. Hommel, S. Trabucco-Miguel, S. Joray, B. Naegele, N. Gonnet, A. Jaillard, Social dysfunctioning after mild to moderate first-ever stroke at vocational age, J. Neurol. Neurosurg. Psychiatry 80 (2008) 371–375. [12] G. LR, Language and individual differences: the search for universals in personality lexicons, in: W. L. (Ed.), Review of Personality and Social Psychology, Vol. 2, Sage Publications, Beverly Hills, CA 1981, pp. 141–165.
23
[13] E. Lannoo, C. de Deyne, F. Colardyn, G. de Soete, C. Jannes, Personality change following head injury: assessment with the NEO Five-Factor Inventory, J. Psychosom. Res. 43 (1997) 505–511. [14] J.E. Kurtz, S.H. Putnam, C. Stone, Stability of normal personality traits after traumatic brain injury, J. Head Trauma Rehabil. 13 (1998) 1–14. [15] A. Norup, E.L. Mortensen, Prevalence and predictors of personality change after severe brain injury, Arch. Phys. Med. Rehabil. 96 (2015) 56–62. [16] G.H. Bower, Mood and memory, Am. Psychol. 36 (1981) 129. [17] F. Heider, The Psychology of Interpersonal Relations, John Wiley & Sons, New York, 1958. [18] T. Ritchie, J.J. Skowronski, J. Hartnett, B. Wells, W.R. Walker, The fading affect bias in the context of emotion activation level, mood, and personal theories of emotion change, Memory 17 (2009) 428–444. [19] J. Barrash, D. Tranel, S.W. Anderson, Acquired personality disturbances associated with bilateral damage to the ventromedial prefrontal region, Dev. Neuropsychol. 18 (2000) 355–381. [20] J.D. Duffy, J.J. Campbell, The regional prefrontal syndromes: a theoretical and clinical overview, J. Neuropsychiatr. Clin. Neurosci. 6 (1994) 379–387. [21] M.S. Mega, J.L. Cummings, Frontal-subcortical circuits and neuropsychiatric disorders, J. Neuropsychiatr. Clin. Neurosci. 6 (1994) 358–370. [22] G.N. Yeates, F. Gracey, J.C. Mcgrath, A biopsychosocial deconstruction of “personality change” following acquired brain injury, Neuropsychol. Rehabil. 18 (2008) 566–589. [23] M. Summerfield, S. Freidin, M. Hahn, P. Ittak, N. Li, N. Macalalad, N. Watson, R. Wilkins, M. Wooden, HILDA User Manual — Release 12, Melbourne Institute of Applied Economic and Social Research, University of Melbourne, 2013. [24] G. Saucier, Mini-markers: a brief version of Goldberg's unipolar big-five markers, J. Pers. Assess. 63 (1994) 506–516. [25] B. Headey, D. Warren, Personality traits measured in HILDA: the ‘Big Five’. Families, incomes and jobs, Volume 3: A Statistical Report on Waves 1 to 5 of the HILDA Survey, 3 2008, pp. 118–122. [26] J.E. Ware, SF-36 health survey update, Spine 25 (2000) 3130–3139. [27] UNESCO Institute for Statistics. International Standard Classification of Education ISCED, Montreal, UNESCO Institute for Statistics, 20122011. [28] E.A. Stuart, Matching methods for causal inference: a review and a look forward, Stat. Sci. 25 (2010) 1–21. [29] P.R. Rosenbaum, D.B. Rubin, The central role of the propensity score in observational studies for causal effects, Biometrika 70 (1983) 41–55. [30] D.E. Ho, K. Imai, G. King, E.A. Stuart, Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference, Polit. Anal. 15 (2007) 199–236. [31] D.E. Ho, K. Imai, G. King, E.A. Stuart, MatchIt: nonparametric preprocessing for parametric causal inference, J. Stat. Softw. 42 (2011) 1–28. [32] K.J. Preacher, A.F. Hayes, Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models, Behav. Res. Methods 40 (2008) 879–891. [33] R Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, 2014 Available from: URL http://www.R-project.org/. [34] T.B. Cumming, R.S. Marshall, R.M. Lazar, Stroke, cognitive deficits, and rehabilitation: still an incomplete picture, Int. J. Stroke 8 (2013) 38–45. [35] D.J. Schretlen, A.M. Shapiro, A quantitative review of the effects of traumatic brain injury on cognitive functioning, Int. Rev. Psychiatry 15 (2003) 341–349. [36] G.A. Miller, J.P. Chapman, Misunderstanding analysis of covariance, J. Abnorm. Psychol. 110 (2001) 40–48.
Contents lists available at ScienceDirect
Journal of Psychosomatic Research
Evidence of Big-Five personality changes following acquired brain injury from a prospective longitudinal investigation Anne Leonhardt ⁎, Stefan C. Schmukle, Cornelia Exner Department of Psychology, University of Leipzig, Neumarkt 9-19, D-04081 Leipzig, Germany
a r t i c l e
i n f o
Article history: Received 25 June 2015 Received in revised form 22 December 2015 Accepted 13 January 2016 Available online xxxx Keywords: Acquired brain injury Big Five HILDA Panel data Personality change
a b s t r a c t Objective: Many studies using different assessment methods have reported personality changes after acquired brain injury (ABI). However, to our knowledge, no prospective study has yet been conducted to examine whether previous cross-sectional and retrospective results can be replicated in a longitudinal prospective design. Further, because clinical control groups were only rarely used, it remains debatable if the personality changes found are unique to patients with ABI or if they also affect patients with other disabilities. Methods: This study examined personality change in 114 participants with different kinds of ABI, 1321 matched controls (general control, GC), and 746 matched participants with restrictive impairments other than brain injury (clinical control, CC) in a prospective longitudinal design using data from the panel survey Household, Income and Labour Dynamics in Australia (HILDA). Results: Participants with ABI showed significantly larger declines in Extraversion and Conscientiousness compared with the GC group. When the ABI participants were compared with the CC group, only the difference in Conscientiousness remained significant. Conclusion: Our prospective data corroborate evidence from previous cross-sectional studies that patients with ABI experience larger declines in Extraversion and Conscientiousness than the general population. Whereas the effect on Conscientiousness was unique to patients with ABI, the decline in Extraversion was also observed in participants with other impairments. © 2016 Elsevier Inc. All rights reserved.
1. Introduction Many studies have reported that acquired brain injury (ABI) can lead to personality changes [1–5]. These changes are typically associated with a worse psychosocial outcome [6,7], especially in terms of heightened depression and anxiety [8,9], reduced activities [10], increased social difficulties [11], and worse health-related quality of life [5]. Previous studies have assessed changes in personality after ABI in very different ways. Investigations focusing on changes in personality structure, as assessed by the Big Five dimensions (Neuroticism or Emotional Stability, Extraversion, Openness to Experience, Conscientiousness, and Agreeableness) [12], have reported a decline in Extraversion, Conscientiousness, or Emotional Stability [13,14,7,2,15] with a decrease in Extraversion being the most consistent finding. However, these previous studies have suffered from two major methodological flaws: The main limitation of all previous approaches has been the retrospective nature of the designs. Retrospective data acquisition is supposed to be less accurate than prospective data
⁎ Corresponding author. E-mail address: [email protected] (A. Leonhardt).
http://dx.doi.org/10.1016/j.jpsychores.2016.01.005 0022-3999/© 2016 Elsevier Inc. All rights reserved.
acquisition because memory is affected by several biases including the current mood bias [16], the self-serving bias [17] or the fading affect bias [18]. Previous studies [14,7,2,15] have tried to limit these effects by assessing pre- and post-injury personality at separate points in time after injury. Even though the bias was not completely removed that way, this still led to reductions in reports of personality changes, underlining the concern that at least some of the effect might be due to a retrospective bias. The second major methodological limitation of previous designs is related to the fact that most studies have lacked a clinical control group. This issue is especially important as a clinical control group may help to generate hypotheses about possible underlying causes of personality changes after brain damage. So far, most authors have suggested a neurological basis such as that damage to the frontotemporal or more specifically the ventromedial or orbital prefrontal cortex has caused personality change [19–21]. However, if personality changes found after ABI closely resemble those found in other medical conditions that do not involve brain damage, other causes should be discussed: Problems in psychosocial adjustment such as changes in social roles, a loss of personal goals, values that became unreachable, anxiety, and withdrawal from social activity might also be responsible for personality changes [22]. Lannoo et al.'s [13] and Rush et al.'s [7]
18
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
studies, which found no differences in personality changes between individuals with ABI and those with other disabilities, suggest that personality changes are related to difficulties dealing with the disability rather than neurological causes. To the best of our knowledge, the current study is the first to investigate the effects of ABI on personality using a prospective longitudinal design. Our aims were, first, to determine whether people with ABI experience stronger personality changes than the general population and whether we can replicate the results of previous retrospective studies in a prospective design. Second, we wanted to clarify whether personality changes are unique to patients with ABI or also affect patients with other restrictive impairments. In addition, we aimed to investigate whether the association between ABI and personality change would be mediated by a reduction in health-related quality of life (HRQoL). We expected to find no mediation effect for physical HRQoL but for mental HRQoL which is assumed to resemble problems in psychosocial adjustment. This finding would support the hypothesis that problems in psychosocial adjustment should be considered as one underlying reason for personality changes besides neurological causes.
2. Method 2.1. Participants We used data from the Household, Income, and Labour Dynamics in Australia survey (HILDA). The HILDA panel study was initiated and is funded by the Australian Government Department of Social Services. It is subject to oversight and approval by the University's Office for Research Ethics and Integrity. With regard to the use of the data, all HILDA data users are required to sign the license agreement, which legally binds them to use the data for bona fide research purposes. In this nationally representative panel survey, which began in 2001, individuals from more than 6500 Australian households are interviewed every year and fill out several questionnaires [23]. In 2005 (t1) and 2009 (t2), the interview included a Big Five questionnaire. Therefore, we used data from 2005 to 2009. Of the 16,373 people who participated in the survey between t1 and t2, we excluded people who had already reported ABI in t1 or who had missing data on their t1 health status, their education, or their personality at t1 or t2. Therefore, 8322 participants between the ages of 15 and 92 (3801 men and 4521 women) were included in the study.
2.2. Main outcome variable: personality All participants filled out a 36-item scale based on Saucier's Mini Markers Scale [24] to assess their Big Five personality dimensions at t 1 and t2 . Each item consisted of a single adjective (e.g. “orderly”, “shy”), and the respondents had to rate how well each adjective described them using a Likert scale that ranged from 1 (does not describe me at all) to 7 (describes me very well). On the basis of previous factor analyses and reliability analyses, eight adjectives had to be excluded due to simultaneous loadings on several factors [25]. The other adjectives were assigned to one of the Big Five dimensions. Participants who had answered less than 75% of the items for one dimension were excluded from the analyses (i.e. at most one item could be missing in each dimension). The participants had up to five missing values which were replaced by the mean of the other items. Over 95% of the participants had no missing values. We calculated T-scores using the data of all participants at t1 (N = 8322) and used the difference between the t 2 personality T-score and the t1 personality T-score as an indicator of personality change. Therefore, a negative change score represented a decline in the respective personality variable.
2.3. Main predictor variable: participants with acquired brain injury (ABI) and control groups The respondents were asked every year for 5 years (i.e. 2005 = t1, 2006, 2007, 2008, and 2009 = t2) if they had any long-term health condition, impairment, or disability that restricted their everyday activities and that had lasted or was likely to last for 6 months or more. Respondents answering this question in the affirmative were then shown a list of medical conditions from which they had to name one or more, e.g. “hearing problems”, “a nervous or emotional condition which requires treatment”, or “chronic or recurring pain”. One of the items from the list was “long term effects as a result of a head injury, stroke or other brain damage”. This item was used to divide the dataset into a group with ABI and control participants. As already mentioned above, we excluded people who had already affirmed this item at t1, because we were interested in analyzing personality change after having acquired a brain injury in a prospective design. Next we describe how participants were divided into three groups based on whether they had acquired a brain injury or another impairment between t1 and t2: 2.3.1. Patients with acquired brain injury (ABI) The group of patients with an acquired brain injury (ABI, n = 114) contained all respondents who had at least once given an affirmative answer to the question about “long term effects as a result of a head injury, stroke, or other brain damage” at any of the four time-points after t1 (but not at t1). As the question did not differentiate between the various forms of brain injury, it was impossible to tell how many of these participants suffered a stroke, a traumatic brain injury, or another form of brain injury. These participants could also have other disabilities in addition to ABI. 2.3.2. Global control (GC) group Our global control (GC) group (n = 8208) contained all other participants and therefore included healthy people as well as people with impairments other than ABI at any time-point. This group can be considered as representative of the population of Australia with the difference that people with ABI are excluded here. 2.3.3. Clinical control (CC) group We additionally created a subgroup of the GC group that included only those participants who had acquired a disability after t1 (clinical control (CC) group, n = 3154). These participants had indicated a newly occurring disability other than ABI from the list of medical conditions at any time-point after t1, which had not already existed at t1. They could however have had other kinds of pre-existing impairments at t1. 2.4. Control variables and matching procedure In a second step, we performed a matching procedure to create groups that are equal on several variables that could influence personality change, such as age, sex, education, preexisting impairments, baseline personality, and health-related quality of life (HRQoL). 2.4.1. Health-related quality of life (HRQoL) The SF-36 Health Survey was used to assess health-related quality of life (HRQoL) on two global outcome scales: physical and mental HRQoL [26]. Both scales were transformed into T-scores using data from the general U.S. population [26] (with M = 50, SD = 10) with higher scores indicating better HRQoL. The respondents filled out the questionnaire annually. We used the SF-36 scores of 2005 (= t1) and 2009 (= t2). We calculated the difference between the t2 SF-36 scores and the t1 SF-36 scores to estimate changes in physical and mental HRQoL with negative values indicating a decline in HRQoL. Missing values were replaced by the mean of the other items from the subscale, only if more than 75% of the items were filled out. Over 94% of the participants had
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
no missing values. The maximum number of missing values was six. HRQoL was used in the analyses as a matching variable but also as a potential mediating variable. In the matching procedure, missing scale scores for physical and mental HRQoL were estimated by single imputation. In the mediation model, participants with missing scale scores for physical and mental HRQoL were excluded from the analysis. 2.4.2. Age, sex, education Participants indicated their age and sex in the questionnaires as well as their highest level of education obtained. These levels were then transformed into levels of the International Standard Classification of Education (ISCED-levels). The ISCED differentiates between eight levels of education ranging from 1 (early childhood education) to 8 (doctoral or equivalent level) [27]. 2.4.3. Propensity score matching The groups differed on several variables that could influence personality change (age, sex, education, pre-existing impairments, HRQoL at t1, personality at t1). Therefore, we used propensity score matching to create control groups that would be comparable to the ABI group on these potentially confounding variables. Propensity score matching is a key tool for estimating causal effects in a quasi-experimental design by accounting for the covariates that differ between the groups [28]. Here, a participant's propensity score reflects the predicted probability of belonging to the ABI group [29]. To match the ABI and GC groups, we estimated the propensity scores by regressing the group membership on the potentially confounding variables, personality scores at t 1, pre-existing impairments (i.e., when the respondent named a medical condition other than a brain injury at t1), HRQoL scores at t1, age, sex and education. As we used single imputations to replace missing values in t1 HRQoL, we also included a dummy variable in the matching procedure to indicate missing data as suggested by Stuart [28]. We used a 1:k nearest neighbor matching scheme to match each participant from the ABI group with a larger number of participants from the GC group using the MatchIt package in R [30,31]. The participants in the GC group were drawn with replacement and were given a weight to indicate how often they were drawn for different participants from the ABI group. We tested different ratios and examined the quality of the matching by checking the standardized mean difference (standardized by the standard deviation of the ABI group) between the ABI and GC groups on each covariate. The best fit between the ABI and GC groups was found for a 1:16 matching with a resulting matched GC group consisting of 1321 participants. After matching, propensity scores were nearly identical for the ABI group with .048 and for the GC group with .048 (standardized mean difference b 0.001). Most
19
importantly, for all covariates standardized mean differences between the groups were negligible with values smaller than .04 after matching (see Table 1). This indicates that the matching procedure was successful and that the matched control group was comparable to the ABI group on all potentially confounding variables. To match the ABI and CC groups, we used the same covariates for matching as before, but additionally included change in HRQoL between t1 and t2 as a further matching variable. The intention was to create two groups of participants who had not only been similar at t1 but had also acquired a disability after t1 that had a similar impact on the change in HRQoL between t1 and t2. As we used single imputations for missing values in HRQoL, we also included an additional dummy variable to indicate missing data in the change in HRQoL. The best fit occurred for a 1:9 matching, leading to a matched CC group comprising 746 participants. The matching was again successful, as shown by nearly identical propensity scores of .060 for the ABI group and of .060 for the CC group (standardized mean difference: 0.007). Additionally, for all covariates standardized mean differences between the groups were smaller than .07 after matching (see Table 1). Due to the matching procedure, participants of the matched control groups obtained different weights. All further analyses that were conducted on these matched GC and CC groups thus considered these individual weights. 2.5. Statistical analysis To investigate group differences in personality change, we computed separate weighted t-tests on change for each personality dimension. The effect size Cohen's d was calculated for each personality dimension. To verify the results of the t-tests in matched groups, we computed ANCOVAs using the original samples that included the potentially confounding variables, personality scores at t1, pre-existing impairments, HRQoL scores at t1, age, sex and education and for the comparison with the CC group the change in HRQoL as covariates. In a mediation model using weighted linear regression, we explored the indirect effect of ABI on personality change through changes in physical and mental HRQoL as mediators. The significance of the regression coefficients was determined by t-tests. To assess the significance of the mediation effect, we tested whether the product of the paths (ab = c-c′) was significantly different from zero as suggested by Preacher and Hayes [32]. As the distribution of ab was not assumed to be normal, we calculated 95% nonparametric bias-corrected and accelerated bootstrap confidence intervals based on 10,000 bootstrap samples. The effect of the mediator is significant at the 5% level if the 95% bootstrap confidence interval does not include zero. We used two-tailed tests with α = .05 as the level of significance for all tests. All statistical analyses were conducted using the statistics program R [33].
Table 1 Sample demographics for matched groups. ABI group
GC group
n (%)
n (%)
Total Sex (female) Pre-existing impairments
114 58 (50.9) 89 (78.1)
1321 674 (51.0)a 1012 (76.6)a
Age ISCED-level Physical HRQoL t1 Change in physical HRQoL t2 − t1 Mental HRQoL t1 Change in mental HRQoL t2 − t1
M (SD) 57.44 (15.51) 3.18 (1.54) 37.91 (11.74) −1.73 (8.87) 46.70 (12.48) −1.49 (9.42)
M (SD) 57.71 (16.35)a 3.13 (1.46)a 37.74 (11.95)a 0.84 (9.12)a 46.78 (11.79)a 0.88 (10.73)a
s. m. d. (ABI/GC)
−0.002a −0.034a
−0.018a 0.027a 0.015a −0.007a
CC group n (%)
s. m. d. (ABI/CC)
746 382 (51.2)a 577 (77.4)a
−0.006a −0.016a
M (SD) 57.60 (15.99)a 3.16 (1.49)a 37.54 (11.61)a −1.54 (9.47)a 46.93 (11.61)a −2.22 (10.95)a
−0.011a 0.010a 0.032a −0.021a −0.019a 0.025a
Note. ABI = acquired brain injury, GC = global control, CC = clinical control, ISCED = International Standard Classification of Education, HRQoL = health-related quality of life, s. m. d. = standardized mean difference = difference between the means of both groups standardized by the standard deviation of the ABI group, t1 = 2005, t2 = 2009. a Based on matched samples with weighted individuals.
20
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Table 2 Personality T-scores in the study groups. ABI group (n = 114)
E
C
ES
O
A
GC group (n = 1321) M
Group difference (ABI/GC)
M
SD
SD
t
t1 t2 t2 − t1
47.25 44.83 −2.41
14.02 13.81 10.81
47.38a 47.57a 0.18a
14.80a 14.68a 11.22a
t1 t2 t2 − t1
49.48 46.37 −3.11
14.99 14.77 10.67
49.20a 48.85a −0.35a
15.47a 15.15a 11.97a
a
a
t1 t2 t2 − t1
50.12 49.93 −0.19
15.09 15.34 12.42
49.98 51.97a 1.99a
16.25 14.94a 13.02a
t1 t2 t2 − t1
50.54 48.27 −2.27
16.97 16.93 13.36
51.15a 48.30a −2.86a
15.30a 16.61a 13.11a
13.64 16.00 14.82
a
a
t1 t2 t2 − t1
50.13 48.52 −1.60
50.23 48.50a −1.73a
15.57 47.20a 16.30a
df
−2.38
−2.38
−1.73
0.46
0.08
1433
1433
1433
1433
1433
CC group (n = 746) p
.018
.017
.085
.646
.937
Cohen's d [95% CI]
−0.23 [−0.42, −0.04]
−0.23 [−0.42, −0.04]
−0.17 [−0.36, 0.02]
0.04 [−0.15, 0.24]
0.008 [−0.18, 0.20]
Group difference (ABI/CC)
M
SD
t
df
p
Cohen's d
47.89a 46.59a −1.30a
14.99a 14.90a 11.65a
−0.96
858
.339
−0.10 [−0.29, 0.10]
48.95a 48.68a −0.27a
15.56a 15.49a 12.36a
−2.32
858
.021
−0.23 [−0.43, −0.04]
a
a
49.96 50.99a 1.03a
15.33 15.78a 13.74a
−0.89
858
.373
−0.09 [−0.29, 0.11]
50.82a 48.78a −2.04a
15.61a 16.20a 12.44a
−0.18
858
.858
−0.02 [−0.22, 0.18]
a
a
−0.87
858
.385
−0.09 [−0.28, 0.11]
49.18 48.82a −0.36a
16.03 16.77a 14.09a
Note. E = Extraversion, C = Conscientiousness, ES = Emotional Stability, O = Openness, A = Agreeableness, ABI = acquired brain injury, GC = global control, CC = clinical control, Cohen's d = difference between the means of both groups standardized by the pooled variance, t1 = 2005, t2 = 2009. a Based on matched samples with weighted individuals.
3. Results 3.1. Personality changes in the ABI and GC groups The mean scores and standard deviations of the five personality domains at t1 and t2 for the ABI group and the GC group and the statistical characteristics of the t tests are presented in Table 2. The courses of personality change for the ABI and GC groups are shown in the left column of Fig. 1. The groups differed significantly in their change in Extraversion (Cohen's d = −0.23, p = .018). Whereas the GC group did not change, t(1320) = 0.59, p = .552, the ABI group showed a significant decline in Extraversion of 2.41 T-scores, t(113) = −2.38, p = .019. The two groups also differed in their change in Conscientiousness (Cohen's d = −0.23, p = .017). Again, there was no change in the GC group, t(1320) = −1.06, p = .29, whereas in the ABI group Conscientiousness declined 3.11 T-scores, t(113) = −3.11, p = .002. In addition, there was a trend towards a group difference in the change in Emotional Stability, which was, however, not significant (Cohen's d = −0.17, p = .085). Here, the GC group showed an increase of 1.99 T-scores over time, t(1320) = 5.56, p b .001, a change that was not apparent in the ABI group, t(113) = − 0.16, p = .870. The two groups did not differ in their change in Openness to Experience and Agreeableness. Openness to Experience, t(1320) = − 7.92, p b .001, and Agreeableness, t(1320) = − 3.85, p b .001, declined in the GC group, whereas the ABI group did not change in Openness to Experience, t(113) = −1.81, p = .073, and Agreeableness, t(113) = −1.16, p = .250. However, these differences between the groups were not significant.1 3.2. Personality changes in the ABI and CC groups The mean scores and standard deviations of the five personality domains at t1 and t2 for the CC group and the statistical characteristics of 1 The ANCOVAs revealed the same results showing a greater decline in Extraversion, t(8309) = −2.82, p = .005, and Conscientiousness, t(8309) = −2.77, p = .006, in the ABI group compared to the GC group, and no group difference in the change in Emotional Stability, t(8309) = −1.60, p = .109, Openness to Experience, t(8309) = −0.43, p = .666, and Agreeableness, t(8309) = −0.87, p = .385.
the t tests are reported in Table 2. The course of personality change in the CC group is also shown in the right column of Fig. 1. The previously found group difference in Extraversion was not significant in this sample. Similar to the ABI group, which experienced a decline of 2.41 T-scores, the CC group also showed a decline in Extraversion that was, however, much smaller (1.30 T-scores), t(745) = − 3.05, p = .002. The size of Cohen's d indicates a smaller group difference between the ABI and CC groups (Cohen's d = − 0.10, p = .339) than between the ABI and GC groups (Cohen's d = −0.23). By contrast, the ABI and CC groups showed significant differences in the change in Conscientiousness. The significant decline of 3.11 T-scores in the ABI group was not apparent in the CC group, t(745) = − 0.60, p = .549. Here, the effect size for the group difference between the ABI and CC groups (Cohen's d = − 0.23, p = .021) was the same as for the analysis with the GC group (Cohen's d = −0.23). There were no group differences in the changes in Emotional Stability, Openness to Experience or Agreeableness.2 Similar to the GC group, the CC group showed a significant increase in Emotional Stability, t(745) = 2.04, p = .042, and a significant decline in Openness, t(745) = −4.48, p N .001. The CC group did not change in Agreeableness, t(745) = −0.71, p = .480. 3.3. Influence of reduced physical and mental HRQoL on personality change In the mediation model, we used changes in physical and mental HRQoL as mediators. Participants with missing scale scores for physical and mental HRQoL were excluded from the mediation analysis. Therefore, 1183 people of the GC group and 100 people of the ABI group were included. Changes in HRQoL were estimated by calculating the difference between the t2 SF-36 scores and the t1 SF-36 scores, with negative values indicating a reduction in HRQoL. The mediation model shown in Fig. 2 illustrates the influence of group (GC = 0, ABI = 1) on 2 The results of the ANCOVAs were again similar to the t-tests in matched groups with a greater decline in Conscientiousness, t(3253) = −2.06, p = .039 in the ABI group and no group differences in the change in Emotional Stability, t(3253) = −0.74, p = .456, Openness to Experience, t(3253) = −0.17, p = .868, and Agreeableness, t(3253) = −0.38, p = .700. However, in contrast to the t-test with matched groups, the ANCOVA also revealed a greater decline in Extraversion for the ABI group compared to the CC group, t(3253) = −2.01, p = .044.
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
21
effect of ABI on the reduction in Extraversion after controlling for reductions in physical and mental HRQoL was lower, but still marginally significant (c′ = −2.14, t(1279) = −1.84, p = .066), indicating that the decline in Extraversion for participants with ABI was partly mediated by the reduction in mental HRQoL. The mediation model for Conscientiousness revealed similar results. There was no significant relationship between the change in physical HRQoL and the change in Conscientiousness (b1 = 0.03, t(1279) = 0.73, p = .465), and the bootstrap confidence interval for the indirect effect of ABI on the change in Conscientiousness through the change in physical HRQoL (a1b1 = − 0.08) was not significant (95% CI [−.38, .16]). A reduction in mental HRQoL was significantly related to a reduction in Conscientiousness (b2 = 0.12, t(1279) = 3.86, p b .001) with a bootstrap confidence interval for the indirect effect of ABI on the reduction in Conscientiousness through the reduction in mental HRQoL (a2b2 = −0.34) that was also significant (95% CI [−.96, −.21]). A decrease in Conscientiousness for participants with ABI was only partly mediated by reductions in physical and mental HRQoL (c′ = − 3.06, t(1279) = −2.48, p = .013). 4. Discussion
Fig. 1. Mean personality scores at t1 and t2 with 95% confidence intervals for the acquired brain injury (ABI), global control (GC), and clinical control (CC) groups. t1 = 2005, t2 = 2009.
the changes in Extraversion and Conscientiousness mediated by the change in HRQoL. The ABI group showed a larger decline in physical (a1 = − 3.17, t(1281) = − 3.33, p b .001) and mental HRQoL (a2 = − 2.81, t(1281) = −2.53, p = .011) than the GC group did. There was only a marginally significant relationship between the change in physical HRQoL and the change in Extraversion (b1 = 0.05, t(1279) = 1.61, p = .108), and the bootstrap confidence interval for the indirect effect of ABI on the change in Extraversion through the change in physical HRQoL (a1b1 = −0.17) was not significantly different from zero (95% CI [−.26, .18]). A reduction in mental HRQoL was significantly related to a reduction in Extraversion (b2 = 0.11, t(1279) = 3.78, p b .001), and the bootstrap confidence interval for the indirect effect of ABI on the reduction in Extraversion through the reduction in mental HRQoL (a2b2 = −0.31) was also significant (95% CI [−.72, −.12]). The direct
To our knowledge, this is the first study to investigate personality changes after ABI using a prospective longitudinal design. The main finding is that personality changed differently in the ABI group than in the two control groups. Similar to the results of previous studies that used retrospective designs [13,15,14], Extraversion and Conscientiousness scores declined significantly more in the ABI group than in the GC group. The previously found decrease in Emotional stability for patients with ABI [15] was not found in our samples although there was a non-significant trend towards a greater decline in the ABI group. The results were irrespective of the statistical method (t-tests in matched groups vs. ANCOVA). It is important to note that we were able to replicate most of the findings from previous studies although the design (longitudinal vs. retrospective) and the rater (most previous studies used significant other ratings, whereas we used self-ratings) differed. However, the use of different raters could explain why we could not replicate the decrease in Emotional Stability. Altered emotional reactivity might be experienced differently from the first-person vs. thirdperson perspective. Our results confirm that personality changes in patients with ABI are a robust finding that does not only appear in retrospective designs. To investigate whether personality changes are unique to patients with ABI or if they also affect people with other impairments as previous studies have suggested [13,7], we additionally compared the ABI group with a subgroup of the GC group containing only those participants who had sustained a similarly restrictive disability other than a brain injury. In comparison with this CC group, the difference in the change in Extraversion disappeared, a result that is consistent with previous findings [13,7]. Although the effect size showed a larger decline in patients with ABI than in the clinical and global control groups, the change in Extraversion did not seem to be unique to patients with ABI which suggests that psychosocial adaptation processes following a serious disability have to be considered as potential underlying reasons for personality changes in addition to structural and functional brain damage. By contrast, the difference in the change in Conscientiousness remained significant. Conscientiousness declined more in patients with ABI than in patients with other disabilities and therefore seems to be unique to patients with ABI in contrast to the change in Extraversion. As the Conscientiousness items used in this study assess orderliness and efficiency traits which describe intact executive functions, it can be assumed that the decline in Conscientiousness is also related to neuropsychological impairments, which are often reported after ABI [34,35]. The results of the ANCOVA differed from that of the t-tests in matched groups in one point. Whereas the ANCOVA revealed a significant group difference in Extraversion, this was not true for the t-test in matched groups. This
22
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Fig. 2. Mediation model for changes in Extraversion and Conscientiousness. ABI = acquired brain injury, GC = global control, HRQoL = health-related quality of life. Unstandardized regression coefficients are reported. *p b .05, **p b .01, ***p b .001.
additional significant result might be due to the greater test power of the ANCOVAS, which used larger sample sizes. However, the matchingprocedure gives the more reliable results as it allows for the selection of a control group that is similar to the treatment group concerning possible confounding variables. According to Miller and Chapman [36] ANCOVA does not adequately control for group differences. Previous studies that compared ABI patients with other patient groups found no differences in personality changes [13,7]. One explanation for this difference could be the longer time frame in our study. Whereas previous studies assessed post-ABI personality one to two years after the injury, the time span could be up to 4 years in our study and some changes might take time to develop. As expected, the changes in Extraversion and Conscientiousness in the ABI group were mediated by mental but not by physical HRQoL which means that personality changes are more related to mental health problems than to physical limitations. This supports the importance of psychological influences including changed social roles, social withdrawal, or mental disorders, that lead to a changed selfperception and a changed evaluation by significant others [22]. Beyond the mediation effect, however, ABI still showed a significant relationship with the changes in Conscientiousness and Extraversion. This may be seen as support of the biopsychosocial model of personality changes which proposes that both neurobiological factors and psychosocial adaptation processes affect personality change in people with ABI [22]. 4.1. Limitations The major limitation of the study is that the indication of ABI could not be validated by a medical examination. As is true of all large panel studies, we had to rely on the statements made by the participants. Nevertheless, the similarity of our results to those of previous studies that used retrospective designs and full clinical diagnostics including neuroimaging in medical facilities supports the validity of the participants' statements. As the question about brain injury on which we built our ABI group did not differentiate between various etiologies of brain injury, we did not know how many of these participants suffered a stroke, a traumatic brain injury, or another form of brain injury. Accordingly, we could not determine whether the personality changes we found depended on the etiology of brain damage. Lastly, although our longitudinal data allowed us to compute the prospective effect of ABI on personality, this effect may only be interpreted as a causal effect after controlling for all confounding
variables. Although we are confident that we indeed considered the most important confounds in our matching analyses by conditioning on sex, age, education, and both personality and health status variables before brain injury, this is no guarantee that the groups are equal concerning all potentially confounding variables. 4.2. Conclusions and directions for future research The findings of this study indicate that people with ABI experience larger declines in Extraversion and Conscientiousness compared with the general population over a time period of 4 years. Compared with patients with other impairments, there was no significant difference in the reduction in Extraversion, whereas the reduction in Conscientiousness seemed to be unique to patients with ABI. Future studies are needed to examine whether the decline in Conscientiousness reflects neuropsychological impairments. Studies with full clinical ABI diagnostics and further specifications of clinical features should be conducted to investigate which moderators and clinical mediators influence personality change in patients with ABI. According to our findings, it seems reasonable to conclude that neurological causes as well as psychosocial adaptation processes affect personality change in people with ABI. As the ABI group comprised individuals with different types of brain injury and diverse severity, the changes should be seen as a general trend for people with different kinds of ABI. This trend is apparently different from that in the general population. Therefore, personality change should be considered in every neurological rehabilitation program although it will not affect every patient. Furthermore, the fact that psychosocial adaptation processes seem to affect personality changes stresses the importance of psychotherapy for people with ABI to help them adapt to the changed circumstances. Most importantly, our study shows that personality changes following ABI are not only observable when assessing personality retrospectively, but are also observable in a prospective longitudinal design, underlining the robustness of this effect. Conflict of interest No conflicts of interest declared. Funding sources No external funding.
A. Leonhardt et al. / Journal of Psychosomatic Research 82 (2016) 17–23
Acknowledgments This paper used unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Social Services (DSS) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either DSS or the Melbourne Institute. References [1] M. Milders, M. Ietswaart, J.R. Crawford, D. Currie, R. Adolphs, S. Baron-Cohen, D. Tranel, H. Bibby, S. McDonald, N. Bogod, Social behavior following traumatic brain injury and its association with emotion recognition, understanding of intentions, and cognitive flexibility, J. Int. Neuropsychol. Soc. 14 (2008) 318–326. [2] R.L. Tate, Impact of pre-injury factors on outcome after severe traumatic brain injury: does post-traumatic personality change represent an exacerbation of premorbid traits? Neuropsychol. Rehabil. 13 (2003) 43–64. [3] J.E. Max, Predictors of personality change due to traumatic brain injury in children and adolescents six to twenty-four months after injury, J. Neuropsychiatr. 18 (2006) 21–32. [4] M.R. Hibbard, J. Bogdany, S. Uysal, K. Kepler, J.M. Silver, W.A. Gordon, L. Haddad, Axis II psychopathology in individuals with traumatic brain injury, Brain Inj. 14 (2000) 45–61. [5] A.P. Diaz, M.L. Schwarzbold, M.E. Thais, A. Hohl, M.M. Bertotti, R. Schmoeller, J.C. Nunes, R. Prediger, M.N. Linhares, R. Guarnieri, R. Walz, Psychiatric disorders and health-related quality of life after severe traumatic brain injury: a prospective study, J. Neurotrauma 29 (2012) 1029–1037. [6] C.J. Cattran, M. Oddy, R.L. Wood, J.F. Moir, Post-injury personality in the prediction of outcome following severe acquired brain injury, Brain Inj. 25 (2011) 1035–1046. [7] B.K. Rush, J.F. Malec, A.W. Brown, A.M. Moessner, Personality and functional outcome following traumatic brain injury, Rehabil. Psychol. 51 (2006) 257–264. [8] J.B. Cantor, T.A. Ashman, M.E. Schwartz, W.A. Gordon, M.R. Hibbard, M. Brown, L. Spielman, H.J. Charatz, Z. Cheng, The role of self-discrepancy theory in understanding post-traumatic brain injury affective disorders: a pilot study, J. Head Trauma Rehabil. 20 (2005) 527–543. [9] I. Aben, J. Denollet, R. Lousberg, F. Verhey, F. Wojciechowski, A. Honig, Personality and vulnerability to depression in stroke patients: a 1-year prospective follow-up study, Stroke 33 (2002) 2391–2395. [10] J. Stone, E. Townend, J. Kwan, K. Hoga, M.S. Dennis, M. Sharpe, Personality change after stroke: some preliminary observations, J. Neurol. Neurosurg. Psychiatry 75 (2004) 1708–1713. [11] M. Hommel, S. Trabucco-Miguel, S. Joray, B. Naegele, N. Gonnet, A. Jaillard, Social dysfunctioning after mild to moderate first-ever stroke at vocational age, J. Neurol. Neurosurg. Psychiatry 80 (2008) 371–375. [12] G. LR, Language and individual differences: the search for universals in personality lexicons, in: W. L. (Ed.), Review of Personality and Social Psychology, Vol. 2, Sage Publications, Beverly Hills, CA 1981, pp. 141–165.
23
[13] E. Lannoo, C. de Deyne, F. Colardyn, G. de Soete, C. Jannes, Personality change following head injury: assessment with the NEO Five-Factor Inventory, J. Psychosom. Res. 43 (1997) 505–511. [14] J.E. Kurtz, S.H. Putnam, C. Stone, Stability of normal personality traits after traumatic brain injury, J. Head Trauma Rehabil. 13 (1998) 1–14. [15] A. Norup, E.L. Mortensen, Prevalence and predictors of personality change after severe brain injury, Arch. Phys. Med. Rehabil. 96 (2015) 56–62. [16] G.H. Bower, Mood and memory, Am. Psychol. 36 (1981) 129. [17] F. Heider, The Psychology of Interpersonal Relations, John Wiley & Sons, New York, 1958. [18] T. Ritchie, J.J. Skowronski, J. Hartnett, B. Wells, W.R. Walker, The fading affect bias in the context of emotion activation level, mood, and personal theories of emotion change, Memory 17 (2009) 428–444. [19] J. Barrash, D. Tranel, S.W. Anderson, Acquired personality disturbances associated with bilateral damage to the ventromedial prefrontal region, Dev. Neuropsychol. 18 (2000) 355–381. [20] J.D. Duffy, J.J. Campbell, The regional prefrontal syndromes: a theoretical and clinical overview, J. Neuropsychiatr. Clin. Neurosci. 6 (1994) 379–387. [21] M.S. Mega, J.L. Cummings, Frontal-subcortical circuits and neuropsychiatric disorders, J. Neuropsychiatr. Clin. Neurosci. 6 (1994) 358–370. [22] G.N. Yeates, F. Gracey, J.C. Mcgrath, A biopsychosocial deconstruction of “personality change” following acquired brain injury, Neuropsychol. Rehabil. 18 (2008) 566–589. [23] M. Summerfield, S. Freidin, M. Hahn, P. Ittak, N. Li, N. Macalalad, N. Watson, R. Wilkins, M. Wooden, HILDA User Manual — Release 12, Melbourne Institute of Applied Economic and Social Research, University of Melbourne, 2013. [24] G. Saucier, Mini-markers: a brief version of Goldberg's unipolar big-five markers, J. Pers. Assess. 63 (1994) 506–516. [25] B. Headey, D. Warren, Personality traits measured in HILDA: the ‘Big Five’. Families, incomes and jobs, Volume 3: A Statistical Report on Waves 1 to 5 of the HILDA Survey, 3 2008, pp. 118–122. [26] J.E. Ware, SF-36 health survey update, Spine 25 (2000) 3130–3139. [27] UNESCO Institute for Statistics. International Standard Classification of Education ISCED, Montreal, UNESCO Institute for Statistics, 20122011. [28] E.A. Stuart, Matching methods for causal inference: a review and a look forward, Stat. Sci. 25 (2010) 1–21. [29] P.R. Rosenbaum, D.B. Rubin, The central role of the propensity score in observational studies for causal effects, Biometrika 70 (1983) 41–55. [30] D.E. Ho, K. Imai, G. King, E.A. Stuart, Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference, Polit. Anal. 15 (2007) 199–236. [31] D.E. Ho, K. Imai, G. King, E.A. Stuart, MatchIt: nonparametric preprocessing for parametric causal inference, J. Stat. Softw. 42 (2011) 1–28. [32] K.J. Preacher, A.F. Hayes, Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models, Behav. Res. Methods 40 (2008) 879–891. [33] R Core Team, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, 2014 Available from: URL http://www.R-project.org/. [34] T.B. Cumming, R.S. Marshall, R.M. Lazar, Stroke, cognitive deficits, and rehabilitation: still an incomplete picture, Int. J. Stroke 8 (2013) 38–45. [35] D.J. Schretlen, A.M. Shapiro, A quantitative review of the effects of traumatic brain injury on cognitive functioning, Int. Rev. Psychiatry 15 (2003) 341–349. [36] G.A. Miller, J.P. Chapman, Misunderstanding analysis of covariance, J. Abnorm. Psychol. 110 (2001) 40–48.