Loneliness, Social Isolation, and Thwarted Belongingness
Loneliness and social isolation have been identified as increasingly significant issues worldwide, and there have been recent calls for their public health prioritisation (Holt-Lunstad
2018; Holt-Lunstad et al.
2017). Conservative estimates suggest that approximately three out of ten people experience loneliness in Australia (Baker
2012), and nearly half of adults aged 18 years and older in the United States report sometimes or always feeling alone or left out (Cigna
2018). Both loneliness and social isolation have been found to be associated with a number of physical and psychological health issues including depression, cognitive decline and dementia (Cacioppo and Cacioppo
2014), and increased risk of early mortality comparable to many leading health determinants (Holt-Lunstad et al.
2015).
Suicide research is an area that has done well in recognising the impact of loneliness and social isolation on suicide risk. According to the Interpersonal Psychological Theory of Suicide (IPTS; Joiner
2005; Van Orden et al.
2010), the need to form and maintain strong, stable interpersonal relationships is considered a fundamental psychological need that when unmet results in a state of thwarted belongingness. Thwarted belongingness (TB) is said to comprise two facets: (1) loneliness, an affectively laden cognition that one has too few social connections, and (2) the absence of reciprocal caring relationships (i.e., where individuals feel cared about and demonstrate care of another). It is viewed as a dynamic cognitive-affective state that is influenced by inter- and intra-personal factors such as experiencing family conflict, living alone, possessing few social supports, and being prone to interpret others’ behaviour as rejection.
According to the IPTS, the presence of either TB or perceived burdensomeness (PB; the view that one’s existence is a burden on friends, family members, and/or society) are causal, proximal risk factors for the development of passive suicide desire (i.e., “I wish I was dead”). However, active suicide desire (i.e., “I want to kill myself”) emerges only when both TB and PB and a sense of hopelessness about these interpersonal states is experienced (i.e., interaction between TB, PB, and hopelessness). Additionally, the IPTS states that for an individual to enact a lethal suicide attempt, they have to both actively desire suicide (i.e., interaction between TB, PB, and hopelessness) and possess the capability for suicide (CS; one’s ability to overcome the inherent drive for self-preservation and engage in lethal self-injury through repeated exposure and habituation to physically painful and/or fear-inducing experiences) (Van Orden et al.
2010). Thus, individuals who have high levels of all three interpersonal risk factors (TB, PB, and CS) combined with a sense of hopelessness, are said to be at highest risk for enacting a lethal suicide attempt.
The Need for Additional Measures
Available measures for screening thwarted belongingness are currently limited to one self-report assessment: the Interpersonal Needs Questionnaire thwarted belongingness subscale (INQ TB; Van Orden et al.
2012). The 25-item INQ was developed in 2009 as part of a doctoral thesis to investigate the aetiology of suicidal desire/behaviour and provide part of a risk assessment framework grounded in the IPTS (Van Orden
2009). It aims to measure beliefs about the extent to which individuals believe their need to belong is met or unmet (i.e., thwarted belongingness) and the extent to which they perceive themselves to be a burden on the people in their lives (i.e. perceived burdensomeness). There are currently six versions of the INQ (5, 10, 12, 15, 18, and 25-item). All six versions have been used in studies of the IPTS since 2009, despite psychometric validation of the 25-item scale only being conducted 3 years after its development (Van Orden et al.
2012).
Research using the INQ has shown thwarted belongingness to be linked, in conjunction with other risk factors, to elevated suicidal thoughts and behaviours (Van Orden et al.
2010). However, findings for the relationship between thwarted belongingness and suicidal thoughts/behaviours have generally been weaker or less supported in comparison to those found for perceived burdensomeness (Chu et al.
2017; Ma et al.
2016). In particular, weaker findings regarding the hypothesised relationship between thwarted belongingness and suicide ideation have raised questions around whether perceived burdensomeness is a more robust interpersonal risk factor that plays a larger role in the development of suicidal thoughts. On the other hand, recent research has also indicated that the different versions of the INQ (10, 12, 15, 18, and 25-item) are not equivalent and that differences across the versions may influence associations found between perceived burdensomeness, thwarted belongingness and suicide ideation in studies of the IPTS (Hill et al.
2015). The possibility of the INQ TB subscale not adequately capturing the thwarted belongingness construct has also been raised to account for this discrepancy (Cero et al.
2015; Ma et al.
2016). As such, the question of whether the comparatively weaker relationship identified between thwarted belongingness and suicide ideation is attributable to construct related or to measurement related issues remains an important area of investigation.
In order to expand the availability of valid measurement approaches for interpersonal risk and promote better identification of thwarted belongingness, the present study aimed to:
1)
Develop a new self-report scale for thwarted belongingness (TB)
2)
Test the psychometric properties of this newly developed scale, including establishing convergent validity with the INQ TB subscale (Van Orden et al.
2012) in a community-based sample, and,
3)
Provide a comparative test of the IPTS (Joiner
2005; Van Orden et al.
2010) hypotheses around suicide ideation and attempt using the newly developed TB self-report scale and the original INQ TB subscale.
Method
A pool of 42 candidate items was selected for potential inclusion in the Thwarted Belongingness Scale (TBS) (Appendix 2). Items were derived and adapted from existing belonging, loneliness, and social support scales identified in a systematic literature search. These existing scales included the Interpersonal Needs Questionnaire (INQ; Van Orden et al.
2012), UCLA loneliness scale (Russell et al.
1980; Russell et al.
1978), De Jong Gierveld Loneliness Scale (de Jong-Gierveld and Kamphuls
1985), Family subscale of the SELSA (DiTommaso and Spinner
1993), General Mattering Scale (Marcus
1991), and Self-efficacy subscales of the Spirituality Index of Wellbeing (Daaleman and Frey
2004). The selection of items into the pool was based on, and expanding upon, the definition of thwarted belongingness provided by the Interpersonal Psychological Theory of Suicide (IPTS; Joiner
2005; Van Orden et al.
2012), which highlights the role of loneliness, disconnection, meaning/mattering, contribution, additive risk factors (e.g., abuse), and social entrapment in contributing to thwarted belongingness (TB). This 42-item pool underwent item refinement via three consecutive stages: (1) expert feedback to revise and remove items, (2) item selection study of the revised item pool in a sample of community-dwelling Australian adults, with further refinement (Study 1), and (3) validation of the final scale and test of the IPTS hypotheses in a large sample of community-dwelling Australian adults (Study 2).
Expert Panel
Email invitations were sent out to 30 Australian and international researchers and clinicians, identified by their contribution to suicide-research and/or clinical experience with suicidal behaviour, to participate in a study to develop a self-report measurement for thwarted belongingness (TB). Seven experts consented to participate and were sent an online survey to evaluate a pool of 42 items. Participants were asked to rate each item for its relevance on a scale from 1 (irrelevant) to 5 (highly relevant). They were also asked to provide comments about each item and its wording, to rate whether the items taken as a whole adequately covered the construct of TB, and provide suggestions as to whether any other items or concepts could be included in the item pool. The study received ethics approval from the relevant institutional review board.
After receiving expert feedback, items were systematically selected or eliminated from the 42-item pool based on whether a majority of experts (4 or more) rated the item as being ‘quite’ (4) or ‘highly (5) relevant, and whether a majority of experts (4 or more) rated the item as being ‘irrelevant’ (1). Several items were also reworded in line with expert feedback to promote item clarity. This resulted in a 22-item TBS pool.
Study 1
Participants and Procedure
Australian adults (N = 284; 85% female) aged 18 years and over were recruited from the online social media website Facebook. A series of paid advertisements were placed on the website between September 2016 and January 2017, targeting Australians aged 18 years or older fluent in English. The advertisements read: “Social Support & Mental Health: Complete a 10 min survey for a PhD project on relationships, suicide, and mental health,” and linked to the study’s Facebook page and the survey. The Facebook page enabled participants to interact (share links, comment, like the page) and provided links to the survey and occasional messages to encourage study participation. The survey was administered online via Qualtrics. Participants were provided with a comprehensive information screen prior to commencing the survey, with informed consent and a list of mental health resources provided online. Participants were not compensated monetarily. The study received ethics approval from the relevant institutional review board.
Analysis
The item pool selected after expert feedback consisted of 22 items. The psychometric properties of these items were initially established in a sample of community-dwelling Australian adults using Exploratory Factor Analysis (EFA, principal axis) alongside the INQ-15 9-item TB subscale to explore factor structure and identify items loading most strongly on the TB factor. Parallel Analyses with 1000 datasets specified on a permutation of the original raw data set using O'Connor (
2000) SPSS syntax for parallel analysis was conducted to determine the number of factors selected. Inter-item correlations between the top TB items (≥0.78 loading) were inspected for item redundancy. Items that displayed a significant correlation of ≥0.70 with another item that measured the same sub-theme of TB (e.g., closeness to others) were systematically compared, based on their conceptual relatedness to the TB sub-theme of interest and item clarity/understandability, and removed from the final scale by the authors. The eight items that remained after these analyses formed the Thwarted Belongingness Scale (TBS). Descriptive analysis and EFA were conducted using SPSS v21 (IBM Corp
2012).
Study 2
Participants and Procedure
Analysis
Comparisons between individuals with and without suicidal thoughts/behaviours were analysed using chi-square statistics for dichotomous variables, and independent-samples t-tests for continuous variables. ‘Prefer not to answer’ responses were treated as missing.
Uni-dimensional Confirmatory Factor Analyses (CFA) was conducted to obtain fit statistics for the previously identified EFA one-factor TB model. To ascertain how the TBS compared to the INQ TB subscale, three competing TB models were tested: the INQ TB subscale (9 items), the 8-item TBS scale, and both TB scales combined (17 items). Weighted Least Squares with Mean and Variance adjustment (WLSMV) estimation was used, with items treated as categorical given their Likert scale format. Bi-factor exploratory analyses (EFA) on the competing TB models were conducted to complement the CFA and explore whether the dataset was sufficiently uni-dimensional for Item Response Theory (IRT) analysis as recommended by Reise et al. (
2007). Two-factor against three-factor, and three-factor against four-factor models were compared for the INQ TB subscale (9 items), the 8-item TBS scale, and both TB scales combined (17 items). WLSMV estimation and Bi-Geomin Orthogonal rotation were used, with items treated as categorical given their Likert scale format. Uni-dimensionality of the TBS and combined scales were computed using Explained Common Variance (ECV) to determine the proportion of common variance across items explained by the TB general dimension.
Model based reliability for the TBS was calculated using the Omega Hierarchical for the total score (ωH), which reflects the proportion of total score variance that can be attributed to the general factor (i.e., TB) after accounting for all additional first order factors (i.e., group factors) that may share variance. The Comparative Fit Index (CFI: >.90 acceptable, >.95 excellent; Bentler
1990), Tucker Lewis Index (TLI: >.90 acceptable, >.95 excellent; Tucker and Lewis
1973), Root Mean Square Error of Approximation (RMSEA: <.08 acceptable, <.05 excellent; Browne and Cudeck
1993), and Standardised Root Mean Square Residual (SRMR: <.08 acceptable, <.05 good; Hu and Bentler
1999; Kline
1998) goodness-of-fit indices were used in the CFA and EFA to assess degree of fit between the models and sample.
IRT analysis was conducted to compare measurement precision across the 8-item TBS and INQ TB subscale. IRT is a model-based method for describing the relationship between individual items on a scale to the construct being measured, the individual’s levels on the latent trait (i.e., TB) and their response to the scale items. IRT is known for addressing practical measurement problems characteristic of classical test theory methods, providing richer and more accurate descriptions of item- and scale-level performance (Hambleton and Jones
1993). The graded response model was used to calibrate item parameter estimates for the TBS and INQ TB subscale given their ordered polytomous response format. Item fit was evaluated using polytomous extensions of the
S-χ2 (Pearson’s chi-square; Orlando and Thissen
2003). Individual information function curves of all the items for each scale were summed separately to create test information function curves for the two TB scales. To test the reading grade of the TBS compared to the INQ TB, The Flesch Kincaid Reading Ease test was used (Flesch
1948; score = 0–100, higher scores indicate text is easier to read). The CFA, bi-factor EFA, and IRT analyses used all available participant data on the thwarted belongingness items (pairwise deletion).
Lastly, due to the over-dispersion and the presence of excess zeros for the suicide ideation outcome (INQ TB: LR χ
2 = 934.75,
df = 1,
p < 0.01; TBS: LR χ
2 = 927.10,
df = 1,
p < 0.01), zero inflated negative binomial regression models were used to test the IPTS hypotheses regarding suicide ideation (past month). Logistic regression models were used to test the IPTS hypotheses regarding suicide attempt (past 3 months). Based on IPTS hypotheses, the ideation model included the main effects of TB (differentially assessed by INQ TB subscale or TBS), PB, and their two-way interaction. The suicide attempt model included the main effects of TB (differentially assessed by INQ TB subscale or TBS), PB, CS, and their two and three-way interactions. IPTS variables were standardised to have a mean of 0 and
SD of 1 to aid interpretation. The zero inflated negative binomial and logistic regression models were conducted on participants with complete responses across the suicide and interpersonal risk factor outcomes (
n = 561; listwise deletion). Descriptive analysis and logistic regressions were conducted using SPSS v21 (IBM Corp
2012). Zero inflated negative binomial regression models were conducted using STATA v14 (StataCorp
2015). CFA and IRT analyses were conducted using MPlus v8 (Muthén and Muthén
1998–2017) and R v2.15.2 (R Core Team
2012).
Discussion
There is currently a need to expand the availability of valid measurement approaches for assessing interpersonal suicide risk. Theoretically, better measurement can help to inform current models of suicide risk by addressing questions regarding the conceptualisation, role, and possible prioritisation of some interpersonal risk factors over others. Additional measures can also help provide enhanced identification of interpersonal risk and may aid suicide screening and prevention efforts. The present study aimed to develop and validate a new self-report scale for the interpersonal risk factor of thwarted belongingness (TB) in a large community population, and provide a comparative test of the Interpersonal Psychological Theory of Suicide using this scale (IPTS; Joiner
2005; Van Orden et al.
2010). From an initial pool of 42 TB items, an 8-item scale (TBS) was developed through consecutive stages of refinement via expert feedback and validation studies in Australian community-based adult samples.
Confirmatory (CFA) and bi-factor exploratory analysis (EFA) supported the uni-dimensionality of the 8-item TBS, where it was found to measure a similar underlying latent construct (i.e., TB) as the Interpersonal Needs Questionnaire thwarted belongingness subscale (INQ TB; Van Orden et al.
2012). Model fit across the CFA and EFA TB models was difficult to discern as inconsistency was observed across fit indices. In the CFA, the Comparative Fit Index (CFI) and Tucker Lewis Index (TLI) for the TBS suggested excellent fit compared to the INQ TB and combined INQ TB and TBS scales. In the bi-factor EFA, CFI and TLI suggested excellent fit for the TBS, the INQ TB, and combined TB scales. In addition, the Standardised Root Mean Square Residual (SRMR) measure of absolute fit was good across all the TB models in the bi-factor EFA. However, the Root Mean Square Error of Approximation (RMSEA) parsimony corrected fit index across all TB models indicated poor fit. One explanation for this inconsistency may be that the RMSEA is more sensitive to the presence of secondary dimensions, model complexity (e.g., number of items/estimated parameters) and data distribution compared to the CFI and TLI (Cook et al.
2009). As such, depending on the interpretational weight placed on the different indices, it could be concluded that the TBS either displays excellent fit in both uni-dimensional CFA and bi-factor EFA based on CFI and TLI indices, or similarly poor fit alongside all other TB models based on the RMSEA.
In regard to the range and level of information captured by the 8-item TBS compared to the INQ TB subscale, Item Response Theory (IRT) analysis indicated that the TBS captured approximately double the amount of information across moderate to high levels of TB compared to the INQ TB. However, this was at the expense of a slightly narrower range, where the TBS was found to provide marginally less information in the extreme TB trait regions. This finding is particularly interesting as the TBS consisted of one less item and was approximately half the length of the INQ TB subscale, with a Flesch Kincaid Reading Ease grade indicating that the scale could be easily understood by eight to nine year olds. This finding suggests that the TBS may be a more efficient scale for assessing TB in populations experiencing moderate to high levels of TB, with greater applicability in low literacy populations compared to the INQ TB. Future studies exploring interpersonal suicide risk may benefit from employing the TBS and INQ TB as complementary assessments to capture TB range (very low or very high levels) via the INQ TB and depth (moderate levels) via the TBS in order to better tailor assessments across different populations.
The IRT findings for the two different TB scales also suggest that in order to retain uni-dimensionality as well as capture a high amount of information, TB may require individualised items/subscales for low, moderate, and high levels of the construct. This would have implications for screening individuals on their interpersonal suicide risk, as TB measures may not be sufficiently sensitive to detect TB in the extreme ranges. Here, developing a computerised adaptive version of the larger TB item bank may be a fruitful way to capture all levels of severity with sufficient precision whilst maintaining efficiency.
Tests of the IPTS hypotheses around suicide ideation provided support for the main effects of TB, perceived burdensomeness (PB) and their two-way interaction on suicide ideation (past month) when using the INQ TB subscale and TBS. Both TB models displayed similar beta-coefficients and significance levels across variables. Additionally, both significant two-way interaction effects showed that participants who experienced high levels of TB and PB had more severe levels of ideation compared to those with low levels of TB and PB. The two-way interaction effects also indicated that participants with high levels of PB but low levels of TB had similar levels of ideation severity compared to those with high levels of TB and PB.
Tests of the IPTS hypotheses around suicide attempt provided support for the main effects of PB and capability for suicide (CS) when using the INQ TB subscale and TBS. Both models explained similar levels of variance in the suicide attempt outcome and displayed similar beta-coefficients, odds ratios, and significance levels across significant variables. Participants experiencing PB were three and a half times more likely to report a suicide attempt in the past 3 months, and those experiencing CS were over one and half times more likely to report a suicide attempt. The lack of a significant three-way interaction effect found for both the TBS and INQ models may be attributable to limitations in power. However, the effect size for the interaction was also negligible, suggesting that participants in the thousands would be required to detect such an effect (Ma et al.
2016). Taken together, these findings provide support for the role of PB as a particularly pernicious interpersonal risk factor contributing to suicide ideation and attempt risk. When experienced at high levels, PB may confer equivalent levels of ideation risk irrespective of TB levels, and contribute double the risk to suicide attempt compared to CS. Given that PB and TB are considered amenable to change, future studies comparing the weight of risk attributed to PB and TB are needed as this could have implications on the way interpersonal suicide risk is screened and targeted for intervention (e.g., targeting PB may be given prominence over TB in high risk populations). The findings also lend support to the validity of INQ TB, although it was found to be a longer scale that captured less information than the TBS, with one item of the INQ TB identified as redundant.
Overall, findings from this study indicate that even with the development of an alternative measure of TB that includes independent items which capture the same underlying construct measured by the INQ TB, some questions remain regarding the significance and role of TB in relation to interpersonal suicide risk and how to best approach its measurement. While complementary use of the INQ TB and TBS may aid in capturing information about TB across different risk regions in the population, it may still be the case that neither measure adequately assesses TB in its entirety as outlined by the IPTS. Perceptions of the intractability of TB, hypothesised in the IPTS to predict the progression from passive to active suicide ideation, were not measured by the INQ TB or the TBS (despite being included in the initial item pool; Appendix 2: items 40, 41 and 42) and have generally been excluded from tests of the IPTS as there have been no sufficient measures of interpersonal hopelessness to date. Given the above-mentioned difficulties in developing a uni-dimensional measure of TB that captures a high amount of information across all regions of TB, the development of a separate measure for intractability of the interpersonal risk factors (i.e., hopelessness about interpersonal challenges relating to both TB and PB) may help to provide the missing link in this process. For example, the interaction between TB and intractability may be more predictive of and comparative in strength to the main effect of PB on suicide ideation as evidenced in the literature.
On the other hand, the findings could also lend support to the literature regarding the stronger role of PB in contributing to interpersonal suicide risk. Here, despite using multiple measures of TB to test the IPTS hypotheses for suicide ideation and attempt outcomes in an online sample, PB was still found to be a more important interpersonal risk factor in both the ideation and attempt models. It may then be that TB as a construct (rather than how it is measured) has limited predictive value for suicide ideation, or at least less of an influence than initially theorised by the IPTS. Another possible avenue to investigate is whether TB functions as a categorical rather than dimensional risk factor that only contributes to suicide risk when experienced at a certain threshold (see Witte et al.
2017). With the anticipation of additional measures for TB and PB being developed in the near future, continuing these lines of enquiry, and in particular comparing new measures for TB and PB against the INQ, may help in terms of examining where gaps remain and whether differences in scale performance can be attributed to construct, measurement, or both construct and measurement related issues. This can, in turn, support the refinement and predictive abilities of theoretical models of suicide risk. In relation to clinical practice, following from this point, though that we cannot unequivocally recommend for the exclusion of TB assessment in resource-drained environments as the full extent of TB’s clinical utility remains unknown, it may be that PB assessments serve as a more robust indicator of interpersonal suicide risk and should be prioritised during the initial risk assessment process at this point in time, whereas it may be more advantageous to incorporate the secondary targeting of TB alongside PB in follow-up/intervention sessions.
Strengths and Limitations
To our knowledge, this is the first study to provide an alternative self-report measure of TB outlined by the IPTS. As such, this study fills a much-needed gap in the IPTS and suicide literature base by providing an additional interpersonal suicide risk screening option and in doing so, contributing to discussion regarding the conceptualisation and role of TB and some next steps for furthering its measurement. Nevertheless, the findings also suggest that limitations in the measurement of TB may not be the only reason why it explains less variance in SI than PB, given the relative consistency of outcomes between the INQ and TBS. Additional strengths of the study include the recruitment of two independent community-based samples during the item refinement and validation process, as well as the use of bi-factor EFA and IRT analysis to provide a more robust assessment of uni-dimensionality and richer description of the TBS’ performance compared to the INQ TB. However, the study also had several limitations. Despite recruiting community-based samples, there was an overrepresentation of females in both studies. Future studies evaluating the TBS may benefit from testing measurement invariance across relevant characteristics such as age group and gender. In addition, suicide outcomes in tests of the IPTS had relatively short time frames (past month and past 3 months). However, given their proximal nature, these outcomes may exhibit less recall bias, and time frames may be better aligned to the IPTS. Another limitation of the current study was that the suicide attempt outcome measure used included elements of suicide preparations and may not represent a pure suicide attempt measure. As such, tests of the IPTS’ three-way interaction effects should be interpreted with this limitation in mind. Further validation of the TBS in other sub-samples utilising validated and conceptually consistent suicide outcome measures within longitudinal/prospective study designs are needed to further explore and support the performance of the TBS. Lastly, it is important to note that though an aim in developing the TBS was to conceptually extend items to better capture TB, perceptions of intractability were not included in the final scale. Given that intractability of interpersonal states is viewed by the IPTS as being predictive of active ideation, the development and inclusion of a measure of hopelessness about interpersonal challenges is needed in order to test the theory’s predictions in more detail.
Conclusions
The TBS has the potential of providing enhanced identification of the interpersonal suicide risk factor of TB, particularly in individuals who display moderate to high levels of TB. The TBS may aid in forming a robust assessment of suicide risk in conjunction with other validated interpersonal measures, with applicability in low literacy populations. However, more research is needed to build upon the findings of this study. In particular, the development of additional interpersonal measures can provide further construct and measurement points of comparison to shed more light on the role of TB in relation to interpersonal suicide risk and how to best approach its conceptualisation and measurement.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.