Perceived Posttraumatic Growth and Depreciation After Spinal Cord Injury: Actual or Illusory?

Objective: This study examined whether retrospective reports of posttraumatic growth (PTG) and depreciation (PTD) of individuals recently diagnosed with a spinal cord injury (SCI) coincide with prospectively measured changes in the conceptually close domains of general self-efficacy (SE) and purpose in life (PIL). The study also tested whether PTG/D and changes in SE and PIL independently predict psychological adjustment to the injury (depressive symptoms, anxiety, life satisfaction). Method: Adopting a longitudinal design, a sample of 206 newly injured patients admitted to one of the four Swiss SCI rehabilitation centers was analyzed. SE and PIL were assessed one month after injury diagnosis and at rehabilitation discharge, PTG/D and the adjustment indicators only at discharge. Structural equation modeling was used to calculate latent change scores for SE and PIL, to correlate these scores to PTG/D scores, and to regress the adjustment indicators on both of them. Results: PTG/D scores were weakly (rmax = .20, p = .033) correlated to changes in SE and PIL. In the multivariate analyses, positive changes in SE and PIL and PTG scores were all associated with better adjustment (e.g., fewer depressive symptoms). In contrast, PTD scores were related to lower adjustment. Conclusions: These results suggest that PTG/D in the initial time after a potentially traumatic medical event seem to be illusory to some degree, as indicated by their weak association with “actual” (i.e., longitudinally measured) changes. Nevertheless, both, PTG/D and actual changes, need to be considered by researchers and clinicians, as they seem to be independently related to psychological adjustment.

changes in PTG domains. In one of these studies, Frazier et al. (2009) followed a sample of students across eight weeks. In those who experienced a potentially traumatic event during that time, pre-to post-event changes in measures identified as corresponding to PTGI domains were only weakly related to perceived changes assessed with the PTGI. Importantly, perceived PTG was found to be positively, and actual changes to be negatively, related to psychological distress. Following another student sample, Yanez, Stanton, Hoyt, Tennen, and Lechner (2011) showed that the positive association between perceived PTG and distress was partially mediated by denial coping, whereas actual changes had a direct, negative effect on distress. Together, these results suggest that perceived PTG may not reflect actual positive changes and that both may be independently, but differently related to psychological adjustment.
Such findings fuel a theoretical debate about the conceptualization of perceived PTG and whether it is a counterfactual positive illusion (e.g., Taylor, 1983), and if so, whether this is an adaptive coping strategy or a defensive, denial-like strategy that interferes with psychological adjustment (e.g., Ford, Tennen, & Albert, 2008;Zoellner & Maercker, 2006).
Thus, a clarification of whether perceived PTG represents actual changes is of high clinical significance.
However, it is uncertain whether results from student samples can be transferred to individuals coping with health-related potentially traumatic events. Responses could differ depending on the type of event and students are not at the peak age for the onset of a chronic health condition (Aspinwall & Tedeschi, 2010). Studies testing the association of perceived PTG with adjustment to chronic health conditions revealed positive or non-significant results (Barskova & Oesterreich, 2009), but to date no study examined whether the association is independent of the effects of actual changes. As initial evidence from a sample of individuals with cancer indicates that perceived PTG is weakly related to actual changes in such a population too (Ransom, Sheldon, & Jacobsen, 2008), this is an important research gap.
Moreover, more research on the association between perceived PTG and actual changes is warranted because the usual measurement of changes in previous research raises several issues: First, the studies that examined actual changes have typically relied on raw change scores, that is, subtracting the baseline score from the one of a second measurement of the same construct later on, to operationalize actual changes. Yet, the reliability of such change scores is questionable because measurement error of the two combined measures may accumulate (e.g., Gollwitzer, Christ, & Lemmer, 2014;McArdle & Nesselroade, 2014).
Second, none of the studies tested for longitudinal measurement invariance of the constructs across time. This means that they did not test whether the measure of the same construct at the two measurement occasions had equivalent properties (see e.g., Little, Preacher, Selig, & Card, 2007). In this case, changes in self-reports cannot be unambiguously interpreted as quantitative differences in the construct itself. They may also reflect a change in the reference point of the respondent or even a change in his or her subjective definition of the underlying construct (i.e., response shift; Gollwitzer et al., 2014;Tennen & Affleck, 2009).
Third, whereas measures of actual changes can show positive and negative changes over time, questionnaires such as the PTGI allow respondents to report only positive changes. This might have led to a positivity bias explaining the difference between perceived and actual changes (Park & Lechner, 2006). Perceived negative changes in the typical PTG domains are termed posttraumatic depreciation (PTD) and were shown to co-occur with PTG (e.g., Baker, Kelly, Calhoun, Cann, & Tedeschi, 2008;Kunz et al., 2017). As such, it is necessary to assess both, perceived PTG and PTD, and their associations with actual changes.

The Present Study
In order to fill the identified research gaps, the present study had two aims. The first aim was to compare perceived PTG and PTD in individuals with SCI at discharge from their clinical rehabilitation assessed with an expanded version of the short form of the PTGI (Kunz et al., 2017) with actual changes in general self-efficacy (SE) and purpose in life (PIL) across rehabilitation. The reasoning to focus on changes in SE and PIL for such a comparison was that they reflect two broader domains which Janoff-Bulman (2006)  Because perceived PTD covers the same domains as perceived PTG, the same reasoning was applied to relate it to negative changes in SE and PIL. Previous research (Blackie, Jayawickreme, Helzer, Forgeard, & Roepke, 2015;Helgeson, 2010) suggested that perceived PTD may be reported more accurately compared to perceived PTG, as indicated by a stronger corroboration by significant others. Hence, we expected a moderate to strong association of perceived PTD with actual changes in SE and PIL.
The second aim of the study was to examine whether perceived PTG and PTD predict psychological adjustment (i.e., fewer symptoms of depression and anxiety, and greater life satisfaction) to SCI at discharge from clinical rehabilitation independent of actual changes in the conceptually close domains of SE and PIL. Both, actual positive changes in SE and PIL, were shown to be related to improved mental health in a previous study examining individuals with SCI (van Leeuwen, Edelaar-Peeters, Peter, Stiggelbout, & Post, 2015). Accordingly, we expected actual changes in SE and PIL to be positively related to psychological adjustment.
Regarding perceived changes, analyses of a smaller subsample of the same cohort (see method section) indicated that perceived PTG was negatively related to symptoms of depression, positively related to life satisfaction and unrelated to anxiety, while perceived PTD was generally related to lower adjustment scores (Kunz et al., 2017). We expected these associations of perceived PTG to remain even when controlling for actual changes. As we expected perceived PTD to represent actual changes more accurately than perceived PTG, we hypothesized that perceived PTD does not independently predict psychological adjustment.

Participants and Design
The present study used data from Pathway 3 of the Swiss Spinal Cord Injury Cohort Study (SwiSCI PW3; see Post et al., 2011). SwiSCI PW3 is an extensive inception cohort study following individuals recently diagnosed with an SCI across their clinical rehabilitation and beyond. Data is collected by clinical assessments and self-report surveys. Included are all Swiss residents newly diagnosed with an SCI, aged 16 years or older, and admitted for clinical rehabilitation to one of the four Swiss rehabilitation centers (Spinal Cord Injury Center of the Balgrist University Hospital, Zürich; Centre for Spinal Cord Injury and Severe Head Injury, REHAB Basel; Clinique Romande de Réadaptation, Sion; and the Swiss Paraplegic Centre, Nottwil). Exclusion criteria are: congenital conditions leading to paraplegia or tetraplegia, new SCI in the context of palliative care, and neurodegenerative disorders such as multiple sclerosis. In addition, trained research assistants invite the eligible patients to participate in the study only upon approval of the responsible physician. The SwiSCI study was approved by the principal ethics committee on research involving humans of Northwest and Central Switzerland (covering the collaboration centers in Nottwil and Basel), the Ethics Committee Vaud (covering the center in Sion), and the Ethics Committee Zürich (covering the center in Zürich). All participants gave written informed consent.
During clinical rehabilitation, data collection in SwiSCI PW3 takes place at one, three, and six months after SCI diagnosis, and at rehabilitation discharge (Post et al., 2011). The present study focused only on data collected at one month post injury (T1; baseline measurement of SE and PIL) and at rehabilitation discharge (T2; follow up measurement of SE and PIL, measurement of perceived PTG, PTD, and the adjustment indicators).
As SwiSCI PW3 data collection is ongoing, the 371 participants who completed rehabilitation until July 1, 2017, were considered for the study. Of these, 123 participants were excluded because they did not complete T1, because complete scales of interest were missing at T1, or because they had a very short rehabilitation duration leading to collapsed measurement occasions. Of the remaining participants, 42 did not complete T2 or did not answer complete scales at T2 and were excluded (see Supplemental Figure S1). Thus, the attrition rate is 16.9%. In terms of selection effects, participants who completed T1 and T2 (continuers: n = 206) and participants who completed only T1 (dropouts: n = 42) did not differ significantly regarding language of the questionnaire, age at injury, duration of clinical rehabilitation, type of injury, SE at T1, and PIL at T1. However, continuers were more likely to be male, χ 2 (1) = 4.18, p = .029, V = .14, and married or widowed, χ 2 (4) = 14.04, p = .007, V = .24. Furthermore, continuers did also not differ significantly from those participants who did not complete T1 (n = 100) regarding sex, language of the questionnaire, and age at injury. The scale demonstrated satisfactory reliability (person reliability index = .82) in an SCI sample (Peter et al., 2014). In terms of validity, GSES scores and the ones of similar measures of general self-efficacy have been shown to be moderately to strongly associated with a broad range of psycho-social outcomes (e.g., symptoms of depression) in the SCI context (Peter, Müller, Cieza, & Geyh, 2012;van Diemen, Crul, van Nes, Geertzen, & Post, 2017). These effects are comparable to the ones observed when SCI-or other context-specific measures of self-efficacy were used (Peter et al., 2012;van Diemen et al., 2017).

Data Analyses
Stata, version 14, was used to clean the data set and to calculate descriptive statistics.
The amount of participants having missing values varied between 0% and 12.6% per variable (Table 1). To deal with the missing data, multiple imputation with chained equations (MICE) was conducted using the mice package in R (van Buuren & Groothuis-Oudshoorn, 2011). As recommended by Enders (2010), data was imputed at the item level and 20 imputed data sets were created. Besides all variables used in subsequent analyses, three auxiliary variables (sex, marital status, rehabilitation duration) were included in the imputation model.
To examine the research questions, latent change score models (McArdle & Nesselroade, 2014) were used to estimate changes in SE and PIL from T1 to T2. These latent change score models were then integrated into broader structural equation models in which changes in SE and PIL were correlated to the different PTG and PTD scores and in which the psychological adjustment indicators measured at T2 were regressed on changes in SE and PIL and the PTG and PTD scores, while controlling for T1 levels in SE and PIL. Separate models were tested for each combination of the latent change score model of SE or PIL with one of the three psychological adjustment indicators. All of these six models included PTG and PTD Total scores. In another six models, the PTG and PTD Strength Through Suffering scores combined with the latent change score model of SE and the PTG and PTD Existential Reevaluation scores combined with the latent change score model of PIL were included instead of the PTG and PTD Total scores. As an illustrative example, a simplified version of the model analyzing the associations between changes in SE, PTG and PTD Total scores, and symptoms of Depression is depicted in Supplemental Figure S2. All of these analyses were performed in the R software package lavaan (Rosseel, 2012).
In building these models, a stepwise approach was followed (see McArdle & Nesselroade, 2014;Weston & Gore, 2006). First, longitudinal measurement invariance of the measures of SE and PIL was tested to ensure that corresponding changes can be meaningfully interpreted (e.g., Gollwitzer et al., 2014). To do so, configural, loading, and intercept invariance was tested (e.g., Gollwitzer et al., 2014;Little et al., 2007). Regarding both SE and PIL, a first model representing a longitudinal confirmatory factor analysis without any constraints besides those for scale setting of the latent factor at T1 and T2 (configural invariance) was fitted. In the second and third model, equality constraints on the factor loadings of corresponding items (loading invariance) and additionally on the intercepts (intercept invariance) were imposed. The loading and the intercept invariant model were both compared to the less restrictive one by using a Chi-square (χ 2 ) difference test. A nonsignificant difference indicates that the corresponding level of measurement invariance holds.
Second, the latent change score models of SE and PIL were created by adding a latent change score factor to the T1 and the T2 intercept invariant measurement models. The T2 factor was regressed on the T1 factor and the latent change score factor using structural weights set equal to 1.0. Together with fixing the variance of the T2 factor at 0, this mimics a subtraction. In these latent change score models, mean changes (i.e., intercept of the latent change score factor, μ∆) and interindividual differences in changes in SE and PIL over time (i.e., variance of the latent change score factor, σ∆ 2 ) are included as model parameters. The main advantage of this way to calculate change is that it is measured without measurement error (Gollwitzer et al., 2014;McArdle & Nesselroade, 2014).
Third, the latent change score models were expanded by adding the PTG and PTD scores and the adjustment indicators to the models. Although guidelines consider a sample size of 200 to be sufficient for structural equation models, each of them was included as single indicator construct (i.e., observed variable) to ensure parsimony and an adequate ratio of the number of variables included in a model and the sample size (see e.g., Weston & Gore, 2006).
All models were estimated using robust maximum likelihood to adjust for non-normal distribution (Finney & DiStefano, 2006). Goodness of fit was judged by χ 2 , comparative fit index (CFI), and root mean square error of approximation (RMSEA) including the 90% confidence interval (CI). Good model fit is indicated by a nonsignificant χ 2 , a CFI value above .95, and an RMSEA value below .06 (Hu & Bentler, 1998).

Participant Characteristics and Descriptive Statistics
Demographic and injury-related characteristics of the sample (n = 206) are shown in Table 2. Descriptive statistics of all study variables are depicted in Table 1.

Measurement Invariance and Actual Changes in SE and PIL
Model fit and χ 2 comparisons of the nested models testing longitudinal measurement invariance of the measures of SE and PIL are presented in Table 3. For both constructs, the configural invariant model demonstrated good fit statistics. Standardized factor loadings (p < .05) ranged between .58 and .82 and between .68 and .87 for the measure of SE or PIL, respectively. The equality constraints in the loading invariant models and in the intercept invariant models did not result in a significant increase in χ 2 (Table 3). Hence, intercept invariance was adopted as a basis for the latent change score models of SE and PIL.
The fit of both latent change score models exactly mirrored the one in the corresponding intercept invariant model. In the latent change score model of SE, there was no significant mean change from T1 to T2 (μ∆ = 0.01, p = .746). However, there were significant interindividual differences in intraindividual change (σ∆ 2 = 0.25, p < .001). In the latent change score model of PIL, there was a significant decrease in the latent mean from T1 to T2 (μ∆ = -0.21, p = .005). Again, there was also significant interindividual variability in intraindividual changes (σ∆ 2 = 0.94, p < .001). Altogether, the significant variability of individual patterns of change from T1 to T2 regarding both, SE and PIL, allowed for probing associations of these changes with other constructs.

Correlations of Actual Changes in SE and PIL with PTG and PTD Scores
All twelve models still achieved good fit (i.e., CFI > .95 and RMSEA < .06) when the different PTG and PTD scores and the adjustment indicators were added to the latent change score models, except that the χ 2 value was significant (p < .05) for most of them. In these models, PTG (r = .15, p = .079) and PTD Total scores (r = .09, p = .298) were both not significantly correlated to changes in SE 1 nor changes in PIL 1 (r = .08, p = .282; r = .03, p = .731, respectively).
Similarly, the PTG (r = .20, p = .033) and the PTD (r = .12, p = .155) Strength Through Suffering scores were weakly or not significantly correlated to changes in SE 1 and the PTG (r = .03, p = .714) and PTD (r = .00, p = .975) Existential Reevaluation scores were not significantly correlated to changes in PIL 1 . In sum, perceived PTG and PTD scores were at best weakly related to actual changes in the conceptually close domains of SE and PIL.

Predicting Psychological Adjustment from Actual and Perceived Changes
Analyses with PTG/PTD Total scores. The standardized beta coefficients of interest in the six models in which changes in SE or PIL and the PTG and PTD Total scores were regressed on the adjustment indicators are presented in Table 4. Positive changes in SE and PIL from T1 to T2 were related to better psychological adjustment (i.e., lower symptoms of Depression, lower Anxiety, and higher Life Satisfaction) at T2. Except for the association 1 The correlation of each type of PTG and PTD score with the two latent change score factors was estimated three times (i.e., once in each model combining the latent change score model of SE or PIL with one of the three adjustment indicators). We only report the correlations estimated in the model combining the latent change score model of SE or PIL with symptoms of Depression as these correlations were virtually the same when Anxiety or Life Satisfaction was included instead (i.e., ∆r < .01). between changes in SE and Anxiety (β = -.15, p = .071), which only approached borderline statistical significance, all of these associations were statistically significant. Effect sizes were stronger for changes in PIL (β ranging between -.45 with symptoms of depression and -.30 with anxiety) than for changes in SE (β ranging between -.15 with Anxiety and .24 with Life Satisfaction).
In the models including changes in SE, higher PTG Total scores were significantly related to fewer symptoms of Depression (β = -.26, p < .001) and to higher Life Satisfaction (β = .15, p = .032). These effect sizes are weaker than those of changes in SE and PIL. Higher PTD Total scores were related to more symptoms of Depression (β = .48), higher Anxiety (β = .37), and lower Life Satisfaction (β = -.33) . In the three models in which changes in PIL were controlled for, the effects of both PTG and PTD Total scores were slightly weaker, but followed the same pattern (Table 4).

Analyses with PTG/PTD Strength Through Suffering and Existential
Reevaluation scores. Most notably, the PTG Strength Through suffering scores were not significantly related to any of the adjustment indicators (Table 5). Although still significant, also the effects of the PTD Strength Through Suffering scores (β ranging between .29 with symptoms of Depression and -.20 with Life Satisfaction) were substantially weaker than the ones of the corresponding Total scores. In contrast, the effects of PTG and PTD Existential Reevaluation scores did not differ substantially from the ones of the Total scores (Table 5).

Sensitivity Analysis
Multiple imputation is based on an iterative process that can go wrong. Thus, the fit of the imputation model needs to be checked by comparing distributional properties of the imputed and observed data as well as the parameter estimates that result in complete and imputed data sets (Lee, Roberts, Doyle, Anderson, & Carlin, 2016; van Buuren & Groothuis-Oudshoorn, 2011). In the present study, graphical analyses yielded no remarkable distributional differences between imputed and observed data which indicates a good fit of the imputation model. This was further supported as no discrepancies in parameter estimates (i.e., ∆ in relevant coefficients was on average .03) were detected when the main analyses were rerun with complete cases only (n = 162).. Furthermore, to test the power of the significant regression paths in all 12 models (Tables 4 and 5), post hoc Monte Carlo power analyses were performed using the simsem package in R (Pornprasertmanit, Miller, & Schoemann, 2013). Average effect power was .82.

Discussion
The present study examined the association of perceived PTG and PTD at discharge from clinical rehabilitation with actual (i.e., longitudinally measured) changes in the conceptually close domains of SE and PIL across rehabilitation in individuals with SCI. Our hypotheses were partially supported. First, perceived PTG, but unexpectedly also PTD scores, were at best weakly related to actual changes in SE and PIL. This can be interpreted as suggesting that both perceived PTG and PTD are illusory to some degree. Second, the study also investigated whether actual and perceived changes independently predict psychological adjustment to the injury at discharge from rehabilitation. As expected, individuals showing more increases in SE and PIL reported fewer symptoms of depression and anxiety and higher life satisfaction. Partially supporting our hypothesis, perceived PTG in domains covering PIL (i.e., existential reevaluation) was related to fewer symptoms of depression and higher life satisfaction, whereas perceived PTG in domains covering SE (i.e., strength through suffering) was unrelated to the adjustment indicators. Contrary to our expectations, perceived PTD was negatively related to all adjustment indicators in these multivariate analyses.

Perceived PTG, Actual Changes, and Psychological Adjustment
The finding that perceived PTG may be illusory to some degree, as indicated by weak associations with actual positive changes, was also observed in studies focusing on individuals coping with cancer (Ransom et al., 2008) or other types of potentially traumatic events (e.g., Frazier et al., 2009). In contrast to these studies, longitudinal measurement invariance for the measures of actual changes was established and latent change score models were used to calculate actual changes in the present study. In doing so, confidence was increased that unreliability in the measures of actual changes or response shift cannot serve as alternative explanations for this finding (Gollwitzer et al., 2014). Moreover, the present study used an expanded version of the PTGI allowing respondents not only to report perceived PTG, but also PTD. Thus, it seems unlikely that positivity bias, which may result in PTG measures allowing respondents to report only positive changes, can explain these results.
Even so, perceived PTG and actual changes may both be relevant in the psychological adjustment process to SCI, as indicated by their independent associations with the adjustment indicators. The result that positive changes in SE and PIL were related to better adjustment scores confirms and expands the mostly cross-sectional previous research showing that SE and PIL are important resources in the adjustment process to SCI (Peter et al., 2012;van Diemen et al., 2017). This result also conforms to the longitudinal study in individuals with SCI showing that increases in SE and PIL were related to improved mental health (van Leeuwen et al., 2015).
Perceived PTG in domains covering meaning in life (i.e., existential reevaluation) was related to better psychological adjustment to the injury even though we controlled for actual changes in PIL. Although these effects were weaker than those of the actual changes, this result supports the idea that perceived PTG could represent a positive illusion which has an adaptive function in the adjustment process (Taylor, 1983). At the same time, this finding contrasts with studies showing that perceived, potentially illusory PTG is related to more psychological distress in students coping with other types of potentially traumatic events (Frazier et al., 2009;Yanez et al., 2011). There are several possible explanations for this discrepancy. First, it may be explained by the different populations examined or by the different methods used in the other studies. Second, perceived PTG was assessed considerably earlier after the event in these studies (i.e., maximally two months compared to five months on average in the present study). In this respect, the Janus face model of perceived PTG (Zoellner & Maercker, 2006) posits that perceived PTG undergoes a process in which one of two co-existing components dominates at different points in time. A dysfunctional component (e.g., denial), which is associated with lower adjustment, is predominant in the very initial time after the event. However, with time, a constructive component (e.g., positive reappraisal, cognitive processing), which is associated with better adjustment, prevails. Taken

Perceived PTD, Actual Changes, and Psychological Adjustment
The present study is the first to test how perceived PTD is related to actual negative changes after a potentially traumatic event such as SCI. Hence, we based our hypothesis on previous studies (Blackie et al., 2015;Helgeson, 2010) showing that perceived PTD received stronger corroboration by close others than perceived PTG. The result that perceived PTD was unrelated to actual negative changes may seem contradictory. Yet, it has been argued that corroboration could either indicate that actual changes took place or simply that individuals informed close others about perceived changes (Park & Lechner, 2006). As perceived negative changes are likely to attract more attention than positive ones (Helgeson, 2010), they may be more often reported to close others. Thus, an individual's perceived negative changes may resemble more the ones of his or her close others than perceived positive changes, albeit both may not represent actual changes, as suggested by the results of the present study.
Similar to perceived PTG, perceived PTD was also significantly related to lower adjustment scores, although actual negative changes in SE and PIL were controlled for. Therefore, perceived PTD may represent an illusory, that is, an overly pessimistic view on posttraumatic life changes that generally signals maladjustment to SCI.

Limitations and Future Research
The present study is subject to several limitations. First, the baseline measurement of SE and PIL was scheduled one month after injury diagnosis. As a result, the time span for actual changes in these constructs does not cover the complete time span (i.e., time since SCI) asked for in the measures of perceived PTG and PTD. It is possible that actual changes occurring in the first month after the injury affected the comparisons of perceived and actual changes. Moreover, we were not able to include pre SCI measures of SE and PIL. Hence, actual changes in these constructs cannot be unambiguously interpreted as actual PTG or PTD (see e.g., Ford et al., 2008). Panel studies, which repeatedly assess SE and PIL and whether there was an onset of a chronic health condition like SCI, are needed to better understand the impact of such an event on SE and PIL. This type of study design would allow for determining pre-to post-event change in these constructs (i.e., actual PTG and PTD).
Second, the associations of perceived PTG and PTD with actual changes in SE and PIL and psychological adjustment were examined only at one point after SCI onset. However, perceived PTG may reflect a process in which its function differs depending on time since the event (Zoellner & Maercker, 2006). Accordingly, an important avenue for future research is to assess perceived and actual changes and their independent association with adjustment at various points after SCI or other potentially traumatic events.
Third, perceived PTG and PTD as well as the adjustment indicators were included as observed variables in the structural equation models to ensure an adequate ratio of variables included in the models and the sample size. However, this means that measurement error affects correlations with these constructs. Thus, future studies having larger sample sizes could replicate the present findings by integrating all variables as latent constructs.

Conclusion and Clinical Implication
The present study contributes to a better understanding of perceived PTG and PTD and their role in the psychological adjustment process to health-related potentially traumatic events such as SCI. In sum, the results indicate that both seem to represent a distorted or illusory view on personal changes, at least in the initial time after the event. Nonetheless, the present results suggest that researchers and clinicians should consider perceived PTG and PTD besides actual changes as both were independently related to adjustment.
As such, perceived PTG and actual positive changes in the conceptually close domains of SE and PIL seem to be both suitable, but distinct targets for intervention to improve psychological adjustment to SCI in the clinical rehabilitation context. Regarding perceived PTG, although research in the SCI context is lacking, results of randomized controlled trials examining individuals with cancer or after other types of potentially traumatic events suggest that, for example, mindfulness exercises and written or spoken self-disclosure can effectively increase levels of perceived PTG (Roepke, 2015;Shiyko, Hallinan, & Naito, 2017). With respect to actual changes in SE and PIL across time, several cognitive-behavioral interventions have been shown to effectively increase levels of SE and PIL in individuals with SCI or after other health-related potentially traumatic events (e.g., Dorstyn, Mathias, & Denson, 2011;Hart, Fonareva, Merluzzi, & Mohr, 2005;Jonkers, Lamers, Bosma, Metsemakers, & van Eijk, 2012). Future research is now needed to examine the independent long term effects of perceived and actual changes in typical PTG domains on adjustment, and for clinicians to consider the use of the PTG construct in assessing patients with SCI and developing therapeutic interventions to promote their psychological adjustment.
* p < 0.05. Note. Different models were estimated for each combination of Self-efficacy or Purpose in Life with one of the adjustment indicators. Results are presented as standardized beta coefficients. ∆ = change from T1 to T2. PTG = posttraumatic growth. PTD = posttraumatic depreciation.
* p < 0.05. ** p < 0.01. *** p < 0.01. Note. Different models were estimated for each combination of Self-efficacy or Purpose in Life with one of the adjustment indicators. Results are presented as standardized beta coefficients. ∆ = change from T1 to T2. PTG = posttraumatic growth. PTD = posttraumatic depreciation.

Supplemental Material
-Informed consent after T1 (n = 65) -Consent withdrawal (n = 6) -Sudden discharge (n = 6) -Back to acute care (n = 3) -Language problems (n = 3) -Other (n = 6) Complete scales at T1 not answered (n = 11) Eligible (N = 748) Refusal of consent SwiSCI PW3 (n = 377) Figure S2. Latent change score model analyzing the associations between changes in Self-Efficacy, posttraumatic growth (PTG), posttraumatic depreciation (PTD), and symptoms of Depression. I1-I5 = Item1-Item5. ∆ = change from T1 to T2. Single headed arrows represent regressions. Double headed arrows represent correlations. To simplify presentation, constraints on intercepts (including the one to set the scale of the latent factors), constraints on factor loadings, constraints on regression paths to create the latent change score factor, correlations of the same item across time, and correlations between PTG and PTD Total scores on one side and T1 Self-Efficacy scores on the other side are not depicted.