Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Dec 1.
Published in final edited form as: Suicide Life Threat Behav. 2021 Jul 26;51(6):1106–1116. doi: 10.1111/sltb.12793

Being Certain that Negative Events Will Happen or that Positive Events Will Not Happen: Depressive Predictive Certainty and Change in Suicide Ideation over Time

Beverlin Rosario-Williams 1, Christina Rombola 2, Regina Miranda 3
PMCID: PMC8678144  NIHMSID: NIHMS1723602  PMID: 34309062

Abstract

Introduction:

Relatively little research has examined the precise components of hopelessness that increase vulnerability to suicidal thinking. We examined whether certainty about an absence of positive future outcomes (Certainty-AP) would more strongly predict suicide ideation over time than certainty about negative future outcomes (Certainty-N).

Method:

Young adults (N = 208), ages 18–34 (M = 19.08, SD = 2.22), with either recent suicide ideation, suicide attempt history, or past-year psychiatric diagnosis were assessed 4 times over 18 months.

Results:

We used multilevel modeling to assess within-participant differences in suicide ideation over time. Both Certainty-AP and Certainty-N predicted later suicide ideation above and beyond generalized hopelessness and depressive symptoms, when examined in separate models. However, Certainty-AP emerged as a stronger predictor of suicide ideation than Certainty-N when examined in the same model.

Discussion:

These findings suggest that certainty about an absence of positive future outcomes may have a more unique prospective relationship to SI than certainty about the presence of negative future outcomes. We discuss clinical and theoretical implications of these findings.

Keywords: depressive predictive certainty, depression, hopelessness, future events, suicide ideation

Suicide is the second leading cause of death in the US among youth aged 10–34 (CDC, 2018), and young adults aged 18–25 are at elevated risk for suicide ideation (SI), relative to older adults (Piscopo et al., 2016). A recent meta-analysis suggested that up to 1 in 4 college students reported lifetime SI (Mortier et al., 2018b), and a study of students in 19 colleges across 8 countries found a 33% lifetime prevalence of SI (Mortier et al., 2018a). Cognitive factors such as hopelessness, have been identified as key predictors of SI over time (Qiu et al., 2017; Smith et al., 2006). However, the specific cognitions involved in increasing risk have received less attention. Our goal was to identify how specific hopelessness-related cognitions differentially predict SI over time, using four waves of data collected over an 18-month period.

At the core of hopelessness are negative expectations about the future (Beck et al., 1974; O’Connor et al., 2000; Marchetti, 2019). Early research by Andersen (1990) conceptualized hopelessness as depressive predictive certainty, the extent to which an individual is convinced that positive future outcomes will not occur and that negative future outcomes will occur. This conceptualization assumes that people become hopeless once they reach the point of 100% certainty about an absence of positive future outcomes (Certainty-AP) and the occurrence of negative future outcomes (Certainty-N), and upon feeling powerless to alter these events in a favorable direction (Andersen & Lyon, 1987). In this view, individuals become hopeless only once they reach the point where unfavorable future outcomes are perceived as inevitable, and not just likely (if negative) or unlikely (if positive) to occur. This formulation is meant as a more precise index of hopelessness and draws from the hopelessness theory of depression, which posits that hopelessness is a “proximal sufficient cause” that increases the likelihood of depressive symptoms, rather than a symptom of depression (Abramson et al., 1989, p. 358). Considering cognitive theories of hopelessness alongside work suggesting that fewer positive future expectancies are more important than greater negative future expectancies in predicting depression and SI or suicidal behavior (MacLeod et al., 1993; MacLeod et al., 2005; O’Connor et al., 2008)—a person’s appraisal of desired or undesired future events and the certainty that they will occur would appear to carry strong implications for suicide risk.

Research on the relation between depressive predictive certainty and SI remains limited, and research examining the differential impact of Certainty-AP versus Certainty-N on SI is even more scarce and mixed. While some evidence suggests that Certainty-AP is more strongly associated with SI than Certainty-N, independently of pessimism (both in expecting negative outcomes and an absence of positive outcomes), generalized hopelessness, and depressive symptoms (Sargalska et al., 2011), other findings show that Certainty-N, but not Certainty-AP, mediates the relationship between suicide attempt history and future SI, even after adjusting for hopelessness and depressive symptoms (Krajniak et al., 2013, footnote). These findings show that both elements of depressive predictive certainty are associated with SI above and beyond depressive symptoms and generalized hopelessness. However, no studies, to our knowledge have examined which subtype of depressive predictive certainty is a better predictor of SI over multiple time points. The current investigation sought to further clarify these mixed findings to determine which subtype of depressive predictive certainty is most strongly associated with SI over time in a high-risk sample of young adults.

Hopelessness, Depressive Symptoms, and SI

Early research also found that depressive predictive certainty was associated with increased depressive symptoms (Andersen, 1990). In nonclinical samples, young adults with higher levels of depressive symptoms anticipated negative events as more likely and positive events as less likely and with greater degrees of certainty, after adjusting for perceived event likelihood and generalized hopelessness (Andersen, 1990, Study 1), and when accounting for negative life events and attributional style (Andersen, 1990, Study 2). Such findings implied that high negative and low positive outcome certainty played a significant role in the development of depressive symptoms, above and beyond Beck and Steer’s (1988) conventional formulation of hopelessness.

Research on the relationship between hopelessness and depression bears further mentioning, particularly in how these constructs may interact to predict SI. Generalized hopelessness, as contrasted with depressive predictive certainty, refers to hopelessness as measured by the Beck Hopelessness Scale (BHS; Beck & Steer, 1988), which does not incorporate a “certainty” construct. A study that examined participant scores on the positive and negative expectation subscales of the BHS separately found that low positive future expectancies more strongly predicted SI and attempts (Horwitz et al., 2017). Similarly, a cross-sectional study of adolescent inpatients found that a lack of positive future expectations was associated with SI, but the presence of negative expectations was not (Elledge et al., 2021). A study of college students found that generalized hopelessness fully mediated the relationship between Certainty-AP and depressive symptoms, while partially mediating the relationship between Certainty-N and depressive symptoms (Miranda et al., 2008). Further, Certainty-AP was significantly associated with an increase in depressive symptoms, but not symptoms of generalized anxiety disorder (GAD). By contrast, Certainty-N was more strongly associated with combined depression and GAD, a finding that supported previous work by Miranda and Mennin (2007) and underscored the functional distinction between these certainty constructs.

We note that most studies examining the mechanisms linking depressive predictive certainty, hopelessness, and depressive symptoms to SI have generally used simple regression models, both in cross-sectional and longitudinal designs. Our research design used four waves of data collection and examined all constructs at each time point; thus, we opted to use multilevel modeling to examine differences within participants in levels of Certainty-AP, Certainty-N, hopelessness, and depressive symptoms in relation to SI over time. This analytic method enabled us to partial out differences from the within and between-person levels of analysis.

The Present Study

We know that previous SI is a strong predictor of future SI and that both hopelessness and depressive symptoms predict future SI (Miranda-Mendizabal et al., 2019; Ribeiro et al., 2018). Previous research has established an association between Certainty-AP and Certainty-N in relation to SI and has suggested that Certainty-AP might be more uniquely related to SI than Certainty-N after adjusting for hopelessness and depressive symptoms, albeit cross-sectionally (Sargalska et al., 2011). Given that previous research was cross-sectional, the present study sought to examine Certainty-AP and Certainty-N as time-varying predictors of SI, accounting for hopelessness and depressive symptoms as time-varying covariates. In other words, we examined within-person changes in Certainty-AP and Certainty-N in predicting change in SI over time. We predicted that both Certainty-AP and Certainty-N would predict SI, independently of levels of hopelessness and depressive symptoms; further, we explored which subtype would more strongly predict SI over time, beyond the effects of generalized hopelessness and depressive symptoms.

Method

Participants

Two hundred eight first- or second-year college students, ages 18–34, who had a lifetime history of a suicide attempt, recent SI, or who screened positive for a mood, anxiety, or substance-related diagnosis took part in an 18-month longitudinal study. Participants were recruited from a sample of 2,054 individuals screened for a history of recent SI or a lifetime suicide attempt between 2010 and 2015. Individuals were recruited from a large public commuter college in the New York City Metropolitan area (either as part of an introductory psychology course or via e-mails sent to first- and second-year students), or from the general community local to this area via newspaper or web-based advertisements. At initial screening, participants were administered measures assessing SI and lifetime suicide attempt history. Of the 2,054 participants who completed the screener, 354 participants completed the remainder of the study, including a second baseline assessment that included a computerized diagnostic interview. Participants were followed up 6-, 12- and 18-months post baseline, with 278 participants (79% retention) completing all sessions and measures. The present analyses only include 208 young adults who reported either SI in the preceding 6 months (at baseline), a suicide attempt history, or who screened positive for a mood, anxiety, or substance-related diagnosis in the previous year (at the second baseline assessment) on the Computerized Diagnostic Interview Schedule for Children (C-DISC-IV, young adult version) (Shaffer et al., 2000), in order to more closely approximate a clinical or high-risk sample.

Participants were predominantly female (n = 149, 72%) and from racially and ethnically diverse backgrounds: 32% Asian (n = 66), 30% White (n = 63), 19% Hispanic (n = 40), 10% Black (n = 20), and 9% mixed/other race/ethnicity (n = 19). Most participants (n = 170, 82%) identified as heterosexual, 5% as gay/lesbian (n = 10), 7% as bisexual (n = 15), and 6% (n = 12) identified as either unknown, other, or preferred not to report. One person left the item blank. Eighty-seven (42%) participants reported a lifetime suicide attempt at baseline. Further, 179 (76%) participants screened positive for a mood, anxiety, or substance use diagnosis in the previous year, 87 (42%) participants reported a lifetime suicide attempt, and 81 (39%) reported SI within the preceding 6 months.

Measures

Future Events Questionnaire (FEQ; Miranda & Mennin, 2007; see also Andersen, 1990).

The FEQ is a 34-item measure used to assess depressive predictive certainty. The FEQ consists of 17 positive and 17 negative future events, for which participants indicate “yes” or “no” as to whether each event is likely to happen to them and rate their degrees of certainty about these predictions on a 5-point Likert scale from 1 (“not at all certain”) to 5 (“as certain as one can be”). Items from this scale were originally derived so that events had a perceived moderate likelihood of occurrence to the average college student, based on pilot testing with undergraduates (Miranda & Andersen, 2008). For instance, positive items include “Have a successful career” and “Have many long-lasting friendships.” Negative items include “Get the blame for things going wrong,” and “Be stuck in an unfulfilling job.” Two total scores are calculated to reflect two types of predictive certainty: the number of “no” responses to positive future events with certainty ratings of “5” (Certainty-AP), and the number of “yes” responses to negative future events with certainty ratings of “5” (Certainty-N). This scale demonstrated high internal consistency reliability for certainty ratings in our study, both for the Certainty-AP (α = .87) and Certainty-N subscales (α = .91), consistent with prior research (Miranda & Mennin, 2007).

Beck Hopelessness Scale (BHS; Beck & Steer, 1988).

The BHS is a 20-item self-report measure administered to assess level of overall hopelessness about the future. Each item consists of a statement addressing hopeless expectations about the future, to which participants indicate their agreement with a response of “true” or “false”. The scale includes 11 negatively-phrased (e.g. “My future seems dark to me”) and 9 positively-phrased (e.g. “I look forward to the future with hope and enthusiasm”) items, with positively-phrased items reverse-scored. Each response of “true” to a negative and “false” to a positive item is counted towards a total score ranging from 0–20, such that a total score reflects generalized negative expectations about the future. High internal consistency as well as convergent and discriminant validity have been reported for the BHS (Steed, 2001), and this scale demonstrated high internal consistency reliability in the present study (α = .97).

Beck Depression Inventory, Second Edition (BDI-II; Beck et al., 1996).

The BDI-II is a widely used self-report measure assessing clinical features of depression over the last two weeks. The BDI has demonstrated high internal consistency reliability and convergent validity in clinical and nonclinical samples (Beck et al., 1996; Whisman et al., 2000; Storch et al., 2004). In the present study, total score was computed without item 9, which assesses SI, to reduce overlap with our measure of suicide ideation. The BDI demonstrated high internal consistency reliability in the present study (α = .92), with and without the inclusion of item 9.

Beck Scale for Suicide Ideation (BSS; Beck & Steer, 1991).

The BSS is a 21-item self-report measure used to assess active and passive suicidal thoughts experienced within the previous week. Each item is coded on a 0–2 scale. Items 20 and 21 address the occurrence of one or more past suicide attempts and the severity of one’s wish to die during the last attempt but are not included in the total score. The total score is calculated from responses to questions 1–19 and can range from 0 to 38. A score of 3 or above is considered indicative of higher suicide risk (Beck & Steer, 1991). The BSS has demonstrated good internal consistency as well as construct and concurrent validity among adolescent populations (Holi et al., 2005). High internal consistency reliability was found for the BSS in our study (α = .87).

Procedure

Participants completed self-report questionnaires, including the FEQ, BHS, BDI-II, and BSS, either individually or in small groups, as part of an initial screening. A subsample of 354 participants, stratified by SI and attempt history, were enrolled in the study, based on whether they had a lifetime history of a suicide attempt or SI in the previous 6 months. The sample was stratified so that about one-third of the sample would have a history of SI or a suicide attempt. About one month from the initial screening, enrolled participants completed the C-DISC-IV to determine if they met criteria for a mood, anxiety, or substance-related diagnosis in the previous year. At each follow-up point (6 months, 12 months, and 18 months from baseline), participants completed the FEQ, BHS, BDI-II, and BSS. At each phase of the study, additional measures assessing constructs outside of the scope of the present paper were also administered. The final sample for analysis (N = 208) consists of participants who either reported recent SI or a lifetime suicide attempt, or who screened positive for a mood, anxiety, or substance-related diagnosis at baseline.

Analytic plan

We first computed Pearson correlations among Certainty-AP, Certainty-N, hopelessness, depressive symptoms, and SI at each time point. However, for ease of interpretation, we also computed ecological correlations across the 4 waves of data collection (Table 1). Ecological correlations provide correlations between group averages rather than individual scores. Given that these data are longitudinal, each participant’s score was averaged across the four measurement occasions to reflect between group variability in the correlations. We conducted independent samples t-tests and one-way analysis of variance to examine sociodemographic differences in study variables at all measurement occasions. To test our main hypotheses, we conducted longitudinal multilevel regression analyses using the nlme package (Pinheiro et al., 2021) in RStudio Team (2020). The longitudinal nature of our study produces a nested structure in which measurement occasions are nested within individuals. For nested data, ordinary least squares regression models ambiguate changes that may occur within people over time from average group differences across people (Snijders & Bosker, 2011). With multilevel modeling, we can model the effects within people and those between (across) people. We used a stepwise process to identify which unconditional means model best fit the data. First, we modeled a random intercept model, followed by a random intercept model with a fixed effect of time. We then modeled a random intercept and random time slope model to disaggregate the variance in SI across the within- and between-person levels. Next, we modeled the covariance structure of the data by fitting a first-order autoregressive covariance structure, because each measurement occasion was equally spaced (see Table 2 for the fit indices of each model). Next, we modeled the time trend to identify whether a linear, quadratic, or cubic trend best fit the data. We then tested four two-level regression models with SI as the outcome variable in all models. In the first model, we included time-varying Certainty-AP and time-varying Certainty-N without covariates. In the second model, time-varying Certainty-AP predicted SI, adjusting for sex, and time-varying hopelessness and depressive symptoms. In the third model, time-varying Certainty-N predicted SI, adjusting for covariates. The final model included both Certainty-AP and Certainty-N as predictors and adjusted for covariates. Since this study focused on the effect of Certainty-AP and Certainty-N on SI over time, we did not assess between-level average differences. Time-varying covariates were not group-mean centered, as we were not interested in understanding how changes in Certainty-AP or Certainty-N relative to a person’s average score predicted changes in SI. Rather, we were interested in identifying how within-person changes in Certainty-AP and Certainty-N over time (irrespective of their average levels) predict changes in SI. We rescaled time such that the first measurement occasion was 0; this allows for the intercept to indicate levels of SI at baseline, or initial levels of SI (Biesanz et al., 2004; Finch & Bolin, 2017). We used full information maximum likelihood to handle missing data, given that it is more robust, relative to traditional techniques used for handling missing data (Baraldi & Enders, 2010). We established the level of significance at p < .05.1

Table 1.

Ecological Correlations Averaged within Individuals Over Time

1 2 3 4 5
1. Certainty-AP -
2. Certainty-N .49** -
3. Hopelessness .51** .43** -
4. Depressive symptoms .48** .48** .70** -
5. Suicide Ideation .41** .29** .57** .56** -
Open in a new tab

Note:

**

p <.01

Table 2.

Model Fit Indices for Random Components of Unconditional Means Models

Model AIC BIC Log likelihood Test Log likelihood ratio p
Random intercept Model 1 4411.36 4425.28 −2202.68
Random intercept + time Model 2 4353.23 4371.78 −2172.61 1 vs 2 60.14 <.01
Random slope Model 3 4279.52 4307.35 −2133.76 2 vs 3 77.71 <.01
Autoregressive Model 4 4280.34 4312.81 −2133.17 3 vs 4 1.18 0.28
Autoregressive & variance Model 5 4243.22 4280.33 −2113.61 4 vs 5 39.11 <.01
3 vs 5 40.30 <.01
Open in a new tab

Note: Each model builds upon the previous model. For instance, the random intercept model allows for the intercept to vary, whereas the random intercept + time model includes a fixed effect of time. The random slope model includes a random intercept and a random slope. The autoregressive model accounts for the autocorrelation between two consecutive measurement occasions while including the components from the previous models. The autoregressive and variance model includes the random components from the previous models and models the decrease in variance from one measurement occasion to the next.

Results

Preliminary findings

There were sex differences in SI at T1 and T4, Certainty-AP at T1-T3, Certainty-N at T1-T2, and depressive symptoms at T2-T3 (results available upon request). Overall, females scored higher than males, except for SI at T1, where males scored higher. Therefore, we included sex as a covariate in all multilevel analyses. Age was positively and weakly correlated with SI at T3 and Certainty-N at T2-T3. Including age in the multilevel analyses did not alter the results. Therefore, we did not include age in the final models. The bivariate correlations were consistent with the ecological correlations. Ecological correlations are presented in Table 1.

Multilevel Model Analyses

We first estimated unconditional means models to identify which model best fit the data. The first model was a random intercept only model, and the second model was a random intercept model with a fixed effect of time. The third model included both the random intercept and a random slope. We then modeled the autoregression structure of the data. This model did not improve the fit of the data compared to the random slope and random intercept model. Thus, we modeled the change in variance along with the autoregression structure and the random slope and random intercept. This model produced the best fit in the data (see Table 2 for the fit indices of each model and the model comparisons). The intraclass correlation indicated that 66% of the variance in SI occurred within participants, indicating substantial variation within participants over time. On average, severity of SI was low (M = 3.59), and the slope was negative (b = −0.93), indicating that on average, the rate of change in SI decreased by 0.93 units from one measurement occasion to the next. Next, we modeled the time trend of the data to ascertain whether a quadratic or cubic trend improved model fit. We created orthogonal polynomials of time to eliminate the correlation between the quadratic and cubic time trend predictors. The results suggested that both the quadratic, b = 14.83 (8.79, 20.86), t(553) = 4.81, p < .01, and cubic trends, b = −7.60 (−13.63, −1.57), t(553) = −2.47, p < .05, significantly described the pattern of the data over time. Thus, we included the quadratic and cubic trends of time in the multilevel models.

We then estimated four conditional multilevel models to examine the effects of Certainty-AP and Certainty-N on SI. The first model included no covariates, while models 2–4 adjusted for sex, time-varying hopelessness, and time-varying depressive symptoms. In the unadjusted model, time-varying Certainty-AP (b = 1.04, p < .01) and Certainty-N (b = 0.40, p < .01) predicted changes in SI over time. In the second model, time-varying Certainty-AP predicted greater severity of SI over time (b = 0.61, p <.01) after adjusting for sex and time-varying hopelessness and depressive symptoms. Further, both time-varying hopelessness (b = 0.13, p <.01) and depressive symptoms (b = 0.15, p <.01) predicted greater SI over time. In the third model, time-varying Certainty-N significantly predicted greater severity of SI over time (b = 0.15, p = .03), when adjusting for sex and time-varying hopelessness and depressive symptoms. However, time-varying hopelessness (b = 0.14, p < .01) and depressive symptoms (b = 0.15, p < .01) also predicted greater severity SI over time.

In the final model, in which both Certainty-AP and Certainty-N were predictors, Certainty-AP (b = 0.56, p <.01), but not Certainty-N (b = 0.08, p = .26), significantly predicted change in SI over time, even after adjusting for sex and time-varying levels of hopelessness and depressive symptoms. Further, both hopelessness (b = 0.12, p < .01) and depressive symptoms (b = 0.14, p < .01) predicted greater levels of SI over time. See Table 3 for additional details.

Table 3.

Multilevel Model Examining Within- and Between-Level Effects of Certainty-AP and Certainty-N on Suicide Ideation

Model 1 Model 2
Variable b se t 95% CI b se t 95% CI
Fixed effects
Level 1
 Certainty-AP 1.04** 0.17 6.06 0.70, 1.37 0.61** 0.16 3.90 0.30, 0.91
 Certainty-N 0.40** 0.08 5.30 0.25, 0.55 - - - -
 Hopelessness - - - - 0.13** 0.04 3.32 0.05, 0.20
 Dep. Symptoms - - - - 0.15** 0.02 8.94 0.11, 0.18
Level 2
 Sex - - - - −0.77 0.35 −2.20 −1.46, -0.08
Random Effects
 Intercept 3.97 3.32, 4.75 2.93 2.21, 3.89
Slope 1.17 0.89, 1.52 0.85 0.54, 1.34
 Level 1 residual 14.42 10.83, 19.19 15.13 11.26, 20.33
Model 3 Model 4
b se t 95% CI b se t 95% CI
Level 1
 Certainty-AP - - - - 0.56** 0.16 3.44 0.24, 0.88
 Certainty-N 0.15* 0.07 2.16 0.01, 0.29 0.08 0.07 1.14 −0.06, 0.23
 Hopelessness 0.14** 0.04 3.60 0.06, 0.22 0.12* 0.04 3.16 0.05, 0.20
 Dep. Symptoms 0.15** 0.02 8.92 0.12, 0.18 0.14** 0.02 8.51 0.11, 0.17
Level 2
 Sex −0.72* 0.36 −2.04 −1.42, -0.03 −0.77* 0.35 −2.19 −1.46, -0.08
Random Effects
 Intercept 3.01 2.27, 4.00 2.98 2.03, 4.36
 Slope 0.88 0.57, 1.36 0.88 0.46, 1.66
 Level 1 residual 15.40 11.33, 20.94 14.81 10.18, 21.56
Open in a new tab

Note: Both quadratic and cubic time trends are included in the models. Level 1 includes time-varying predictors and covariates; level 2 includes sex as a time-invariant covariate. Dep. Symptoms = depressive symptoms. Model 1 includes Certainty-AP and Certainty-N as predictors and excludes covariates; model 2 includes Certainty-AP and covariates; model 3 includes Certainty-N and covariates; and model 4 includes both predictors and covariates; 95% confidence intervals are for unstandardized estimates.

*

p < .05

**

p < .01

Discussion

The present study examined which subtype of depressive predictive certainty more strongly predicted SI over an 18-month period. Overall, findings from four multilevel regression models revealed that Certainty-AP was a stronger predictor of SI over time than Certainty-N, when adjusting for sex, and time-varying hopelessness and depressive symptoms. This suggests that subtypes of depressive predictive certainty have a differential effect on the severity of SI among young adults over time. Although Certainty-N predicted change in SI when adjusting for sex, time-varying hopelessness, and depressive symptoms, it was no longer a significant predictor when including Certainty-AP in the model. Thus, it appears that the certainty that positive events will not occur more strongly maintains SI than the certainty that negative events will occur. Further, ecological correlations revealed a moderately positive correlation between Certainty-AP and Certainty-N (r = .49), highlighting that though these constructs are related, they are conceptually distinct and not redundant. These findings have important clinical and research implications and advance our understanding of how nuanced cognitive processes may be implicated in the development and maintenance of SI.

Substantial evidence demonstrates that both hopelessness and depressive symptoms predict SI (Miranda-Mendizabal et al., 2019; Ribeiro et al., 2018). However, hopelessness is often conceptualized as a general, global index of pessimistic future predictions (e.g., “the future seems bleak to me”). This general conceptualization of hopelessness does not explain what about the future seems bleak. Is it that nothing good will happen or that only bad things will happen? By parsing out the specific expectations that people have about the future, we may be able to better examine how these expectations and the types of expectations are related to the maintenance of SI over time. Indeed, previous studies have suggested that expecting an absence of positive versus the presence of negative outcomes be examined separately in measures of hopelessness (Elledge et al., 2021). Our research extends such findings by suggesting a potential role for certainty about those future-event anticipations, as well.

Our findings demonstrate that not only do Certainty-AP and Certainty-N have differential effects on SI, but that these constructs have effects that are independent of hopelessness and depressive symptoms. In all multilevel models, both hopelessness and depressive symptoms significantly predicted change in SI, and this is consistent with previous research suggesting that both hopelessness and depressive symptoms predict SI (Gilmour, 2016; Miranda-Mendizabal et al., 2019; Ribeiro et al., 2018). However, Certainty-AP, but not Certainty-N, predicted change in SI over time when adjusting for both time-varying hopelessness and depressive symptoms (and sex). These findings suggest that the variance in SI accounted for by Certainty-AP, hopelessness, and depressive symptoms may overlap but are sufficiently distinct from each other. Further, these factors maintain suicidal thinking independently of each other. We note, however, that although these findings were significant in the statistical models, they may have limited utility in clinical settings (given the limitations discussed below). Suicidal thoughts and behaviors are very difficult to predict (Franklin et al., 2017). Thus, although our work shows that certainty about future outcomes predicts changes in SI, further work with more fine-grained measurement (e.g., ecological momentary assessments) is necessary to demonstrate the extent to which these factors can help accurately predict subsequent SI. Still, these findings may help clinicians with the conceptualization of a patient’s risk level.

Other maladaptive future-oriented cognitions may also contribute to the development and maintenance of SI over time. For instance, the powerlessness aspect of depressive predictive certainty may be similar to the feelings of defeat and entrapment proposed in the Integrated Motivational-Volitational (IMV) model (O’Connor & Kirtley, 2018). The IMV model suggests that defeat and entrapment are key proximal predictors that contribute to the development of SI. Defeat involves emotional sensitivity to stressors the individual experiences, while entrapment involves feeling unable to escape a stressful situation. According to the model, defeat may lead to entrapment when the individual experiences social and environmental pressures that heighten their sensitivity to negative feedback from the environment. Entrapment then leads to SI. We also note that entrapment involves some element of future-oriented cognition, as do the subtypes of depressive predictive certainty. Given these similarities in the constructs, we suggest that future work examine their unique contributions to SI. On the other hand, research has suggested that rumination about the future may lead to greater automaticity in making pessimistic future-event predictions, which can then lead to certainty about those predictions (Andersen & Limpert, 2001; Andersen et al., 1992). Further research is necessary to identify the conditions under which people, particularly vulnerable youth, develop certainty about future outcomes.

More broadly, recent suicide research suggests that suicidal cognitions (e.g., hopelessness, entrapment) may be influenced by an underlying suicidal belief system, such that as an individual’s suicide risk increases, these specific types of cognitions can be more distinguishable (Bryan & Harris, 2019). Subtypes of depressive predictive certainty may be one of these types of suicidal cognitions that are influenced by an underlying suicidal belief system or, more generally, a hopeless belief system that includes feelings of defeat, entrapment, and ruminative thinking. Again, the unique value of depressive predictive certainty subtypes is that these factors account for the degree of certainty with which individuals make future predictions. Further work is needed to clarify how these suicide-related cognitions relate to each other and increase risk for SI and suicide attempts.

Hope is another factor worth investigating in relation to depressive predictive certainty and SI. One study using the IMV as a framework found that hope weakened the relation between entrapment and SI (Tucker et al., 2016), suggesting that hope may have a protective effect. Theoretically, hope refers to future-oriented goals, the motivation to achieve the goals, and the specific action plan that individuals use to achieve their goals (Davidson et al., 2009; Snyder et al., 1991). Hope stands in stark contrast to both Certainty-AP and Certainty-N, because hope incorporates an element of agency and determination. Given that suicide research has focused primarily on identifying risk factors, particularly hopelessness, future work may benefit from examining how hope and hopelessness-related cognitions interact to predict (or reduce) risk for SI and suicide attempts.

Finally, these findings can be considered in light of the interpersonal-psychological theory of suicide (IPTS; Joiner, 2005; Van Orden et al., 2010) which is a widely used framework to understand the transition from suicidal thoughts to suicidal behaviors. The IPTS proposes that people will think about suicide when they perceive themselves as alone in the world (thwarted belonging) and perceive that they are burdens to others (perceived burdensomeness). The theory suggests that these thoughts, coupled with hopelessness, will lead a person to think about suicide, and that people will transition to attempt suicide when they have also acquired the capability to engage in suicidal behaviors (e.g., through non-suicidal self-injury, risk-taking behaviors, previous exposures to suicide deaths). Given the specific predictions that the IPTS makes about socio-cognitive factors and hopelessness, future work incorporating predictions of the IPTS along with Certainty-AP and Certainty-N may clarify how these cognitive processes increase risk for future SI.

We highlight the complexity in predicting suicide ideation (and behaviors). Given our longitudinal design of four equally spaced measurement occasions, we were able to model up to a cubic trend of time. The model fit indices suggested that a quadratic trend of time improved the data relative to a linear time trend; likewise, a cubic time trend significantly improved model fit compared to a quadratic time trend. These patterns demonstrated that suicidal cognitions are not stagnant but fluctuate across time. Specifically, participants who presented with high levels of SI reported lower levels of SI at later times, and some reported an increase in SI. Importantly, individuals differ in how their suicide ideation fluctuates over time. Thus, appropriate nonlinear models are necessary to capture these changes over time. Further, SI fluctuates daily and even multiple times per day (Kleiman et al., 2017). Thus, more research incorporating intensive longitudinal designs is needed to capture this variability in SI and better account for proximal predictors of SI.

The findings in this study should be interpreted in light of its limitations. First, participants in this sample reported low levels of Certainty-AP, Certainty-N, and SI at all measurement occasions in the study. These scores are expected to be significantly lower than scores found among clinical samples, and thus may be of limited generalizability to such populations. Second, Certainty-AP, by definition, seems to suggest that an individual is convinced about an absolute absence of desirable future outcomes, which would presumably necessitate an inability to generate thoughts of any positive future events whatsoever, and therefore consider them as certain not to occur. However, it is unclear whether this certainty is state-dependent or a stable characteristic. Further, the FEQ presents a list of preformulated future events that are meant to be generalizable to college students and young adults, given the inclusion of items that may apply more to this age group than to older adults (e.g., items addressing future job satisfaction) (Miranda & Mennin, 2007). However, the items may be irrelevant or unimportant to some participants and may not capture the idiographic prediction that individual participants make about their future. Indeed, previous research suggested that more specific (low) positive future expectations were better predictors of SI than global hopelessness among individuals with a history of repeat self-harm (O’Connor et al., 2008). Further, our sample was predominantly female (72%), limiting our inferences primarily to female young adult populations. The general limitations of using self-report measures also bear mentioning. Some often-cited criticisms of such measures include the problem of demand characteristics, especially if a measure is re-administered in the same study (Gemar et al., 2001). Lastly, we note that at the time of data collection, most of the existing work followed participants over a year later; ecological momentary assessment studies examining changes in SI over time were not published at the time. Thus, we selected a 6-month follow-up to mitigate the long follow-up periods in other studies. However, we acknowledge that these cognitive processes may occur within minutes or hours, and our design was unable to capture those changes.

Despite these limitations, the present study included several strengths. First, the sample consisted of high-risk young adults in a college population. Participants reported either a suicide attempt, recent SI (past 6 months), or met criteria for a mood, anxiety, or substance-related diagnosis in the preceding 12 months. Further, the sample was racially and ethnically diverse, facilitating generalizability across racial and ethnic groups—a prevalent limitation in suicide research (Cha et al., 2018). Finally, this study used four waves of data collection and retained almost 80% of the sample over an 18-month period.

Further research on depressive predictive certainty employing a similar longitudinal design and high-risk clinical samples may be warranted to substantiate the findings in this study. Researchers should also use more fine-grained methodologies, including ecological momentary assessments to identify how real-time changes in certainty about future predictions are associated with SI and suicidal behaviors. Future studies may also consider examining depressive predictive certainty alongside processes such as rumination and future-event fluency at multiple measurement occasions to identify how these factors interact to predict and maintain SI. It may also be important to include a shorter time lapse between measurements. Six months between assessments may be too long to best capture how SI changes over time.

Conclusion

Contemporary research suggests that future-oriented cognitions involving expecting an absence of positive future outcomes are more strongly associated with SI than expecting negative future outcomes. The present study extends previous findings by suggesting that certainty that positive events will not occur more strongly predicts change in SI over time than certainty that negative events will occur, independently of generalized hopelessness and depressive symptoms. Measurement of depressive predictive certainty in future studies may provide fine-grained insights into the kinds of hopelessness-related cognitions experienced by at-risk young adults that may be crucial to better streamlining prevention and treatment targeted to this age group, and to broadening awareness of how these specific cognitive processes may confer risk for SI and suicidal behavior.

Acknowledgments

This research was funded by NIH Grant MH 091873 and GM 060665. Thanks to Valerie Khait, Kaerensa Craft, Dalia Gefen, Amy Kephart, Justyna Jurska, Soumia Cheref, Eileen Fener, Nargus Harounzadeh, Matthew Kaplowitz, Giulia Landi, Robert Lane, Lisa Lerner, Wendy Linda, Lillian Polanco-Roman, Victoria Quiñones, Jessica Silver, Lauren Uss, and Jorge Valderrama for their assistance with data collection. Thanks to Dr. Evelyn Behar and Dr. Ana Ortin Peralta for comments on a previous version of this manuscript, and thanks to Thomas Corbeil for comments on our statistical analyses.

Footnotes

1

Although not the goal of this study, we also computed interactions between Certainty-AP and Certainty-N given the theoretical conceptualization that high levels of both subtypes may predict elevated levels of SI. We examined interactions predicting SI at baseline and at 6-month follow-up, adjusting for baseline hopelessness and depressive symptoms. However, none of the models were statistically significant.

Contributor Information

Beverlin Rosario-Williams, Hunter College and The Graduate Center, City University of New York.

Christina Rombola, Hunter College, City University of New York.

Regina Miranda, Hunter College and The Graduate Center, City University of New York.

References

  1. Abramson LY, Metalsky GI, & Alloy LB (1989). Hopelessness depression: A theory-based subtype of depression. Psychological Review, 96(2), 358–372. [Google Scholar]
  2. Andersen SM (1990). The inevitability of future suffering: The role of depressive predictive certainty in depression. Social Cognition, 8(2), 203–228. [Google Scholar]
  3. Andersen SM, & Limpert C (2001). Future-event schemas: Automaticity and rumination in major depression. Cognitive Therapy and Research, 25(3), 311–333. [Google Scholar]
  4. Andersen SM, & Lyon JE (1987). Anticipating undesired outcomes: The role of outcome certainty in the onset of depressive affect. Journal of Experimental Social Psychology, 23(5), 428–443. [Google Scholar]
  5. Andersen SM, Spielman LA, & Bargh JA (1992). Future-event schemas and certainty about the future: Automaticity in depressives’ future-event predictions. Journal of Personality and Social Psychology, 63(5), 711–723. [DOI] [PubMed] [Google Scholar]
  6. Baraldi AN, & Enders CK (2010). An introduction to modern missing data analyses. Journal of school psychology, 48(1), 5–37. [DOI] [PubMed] [Google Scholar]
  7. Beck AT, & Steer RA (1988). Manual for the Beck Hopelessness Scale (BHS). San Antonio, TX: Psychological Corporation. [Google Scholar]
  8. Beck AT, & Steer RA (1991). Manual for the Beck Scale for Suicide Ideation (BSS). San Antonio, TX: Psychological Corporation. [Google Scholar]
  9. Beck AT, Steer RA, & Brown GK (1996). Manual for the Beck Depression Inventory-II (BDI-II). San Antonio, TX: Psychological Corporation. [Google Scholar]
  10. Beck AT, Weissman A, Lester D, & Trexler L (1974). The measurement of pessimism: The hopelessness scale. Journal of Consulting and Clinical Psychology, 42(6), 861–865. [DOI] [PubMed] [Google Scholar]
  11. Biesanz JC, Deeb-Sossa N, Papadakis AA, Bollen KA, & Curran PJ (2004). The role of coding time in estimating and interpreting growth curve models. Psychological Methods, 9(1), 30–52. [DOI] [PubMed] [Google Scholar]
  12. Bryan CJ, & Harris JA (2019). The structure of suicidal beliefs: A bifactor analysis of the Suicide Cognitions Scale. Cognitive Therapy and Research, 43(2), 335–344. [Google Scholar]
  13. Centers for Disease Control and Prevention. (2018). 10 Leading Causes of Death, United States – 2018. WISQARS: National Vital Statistics System National Center for Health Statistics, CDC. Retrieved on April 22, 2020 from https://www.cdc.gov/injury/wisqars/pdf/leading_causes_of_death_by_age_group_2018-508.pdf. [PubMed] [Google Scholar]
  14. Cha CB, Tezanos KM, Peros OM, Ng MY, Ribeiro JD, Nock MK, & Franklin JC (2018). Accounting for diversity in suicide research: Sampling and sample reporting practices in the United States. Suicide and Life-Threatening Behavior, 48(2), 131–139. [DOI] [PubMed] [Google Scholar]
  15. Davidson CL, Wingate LR, Rasmussen KA, & Slish ML (2009). Hope as a predictor of interpersonal suicide risk. Suicide and Life-Threatening Behavior, 39(5), 499–507. [DOI] [PubMed] [Google Scholar]
  16. Elledge D, Zullo L, Kennard B, Diedrich A, Emslie G, & Stewart S (2021). Refinement of the role of hopelessness in the Interpersonal Theory of Suicide: An exploration in an inpatient adolescent sample. Archives of Suicide Research, 25(1), 141–155. [DOI] [PubMed] [Google Scholar]
  17. Finch H, & Bolin J (2017). Multilevel modeling using Mplus. CRC Press. [Google Scholar]
  18. Franklin JC, Ribeiro JD, Fox KR, Bentley KH, Kleiman EM, Huang X, Musacchio KM, Jaroszewski AC, Chang BP, & Nock MK (2017). Risk factors for suicidal thoughts and behaviors: a meta-analysis of 50 years of research. Psychological bulletin, 143(2), 187–232. [DOI] [PubMed] [Google Scholar]
  19. Gemar MC, Segal ZV, Sagrati S, & Kennedy SJ (2001). Mood-induced changes on the Implicit Association Test in recovered depressed patients. Journal of Abnormal Psychology, 110(2), 282–289. [DOI] [PubMed] [Google Scholar]
  20. Gilmour H (2016). Threshold and subthreshold Generalized Anxiety Disorder (GAD) and suicide ideation. Health Reports, 27(11), 13–21. [PubMed] [Google Scholar]
  21. Holi MM, Pelkonen M, Karlsson L, Kiviruusu O, Ruuttu T, Heilä H, Tuisku V, & Marttunen M (2005). Psychometric properties and clinical utility of the Scale for Suicidal Ideation (SSI) in adolescents. BMC Psychiatry, 5(8), 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Horwitz AG, Berona J, Czyz EK, Yeguez CE, & King CA (2017). Positive and negative expectations of hopelessness as longitudinal predictors of depression, suicidal ideation, and suicidal behavior in high-risk adolescents. Suicide and Life-Threatening Behavior, 47(2), 168–176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Joiner TE Jr. (2005). Why People Die by Suicide. Cambridge, MA: Harvard University Press. [Google Scholar]
  24. Kleiman EM, Turner BJ, Fedor S, Beale EE, Huffman JC, & Nock MK (2017). Examination of real-time fluctuations in suicidal ideation and its risk factors: Results from two ecological momentary assessment studies. Journal of abnormal psychology, 126(6), 726–738. [DOI] [PubMed] [Google Scholar]
  25. Krajniak M, Miranda R, & Wheeler A (2013). Rumination and pessimistic certainty as mediators of the relation between lifetime suicide attempt history and future suicidal ideation. Archives of Suicide Research, 17(3), 196–211. [DOI] [PubMed] [Google Scholar]
  26. MacLeod AK, Rose GS, & Williams JMG (1993). Components of hopelessness about the future in parasuicide. Cognitive Therapy and Research, 17(5), 441–455. [Google Scholar]
  27. MacLeod AK, Tata P, Tyrer P Schmidt U, Davidson K, & Thompson S (2005). Hopelessness and positive and negative future thinking in parasuicide. British Journal of Clinical Psychology, 44(4), 495–504. [DOI] [PubMed] [Google Scholar]
  28. Marchetti I (2019). Hopelessness: A network analysis. Cognitive Therapy and Research, 43(3), 611–619. [Google Scholar]
  29. Miranda R, & Andersen S (2008). Mental rehearsal in making pessimistic future-event predictions: Modeling the acquisition of depressive schemas. Unpublished manuscript, New York University. [Google Scholar]
  30. Miranda R, Fontes M, & Marroquín B (2008). Cognitive content-specificity in future expectancies: Role of hopelessness and intolerance of uncertainty in depression and GAD symptoms. Behaviour Research and Therapy, 46(10), 1151–1159. [DOI] [PubMed] [Google Scholar]
  31. Miranda R, & Mennin DS (2007). Depression, generalized anxiety disorder, and certainty in pessimistic predictions about the future. Cognitive Therapy and Research, 31(1), 71–82. [Google Scholar]
  32. Miranda-Mendizabal A, Castellví P, Parés-Badell O, Alayo I, Almenara J, Alonso I, Blasco MJ, Cebrià A, Gabilondo A, Gili M, Lagares C, Piqueras JA, Rodríguez-Jiménez T, Rodríguez-Marín J, Roca M, Soto-Sanz V, Vilagut G, & Alonso J (2019). Gender differences in suicidal behavior in adolescents and young adults: systematic review and meta-analysis of longitudinal studies. International Journal of Public Health, 64(2), 265–283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Mortier P, Auerbach RP, Alonso J, Bantjes J, Benjet C, Cuijpers P, Ebert DD, Green JG, Hasking P, Nock MK, O’Neill S, Pinder-Amaker S, Sampson NA, Vilagut G, Zaslavsky AM, Bruffaerts R, & Kessler RC (2018a). Suicidal thoughts and behaviors among first-year college students: Results from the WMH-ICS Project. Journal of the American Academy of Child and Adolescent Psychiatry, 57(4), 263–273. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Mortier P, Cuijpers P, Kiekens G, Auerbach RP, Demyttenaere K, Green JG, Kessler RC, Nock MK, & Bruffaerts R (2018b). The prevalence of suicidal thoughts and behaviours among college students: A meta-analysis. Psychological Medicine, 48(4), 554–565. [DOI] [PubMed] [Google Scholar]
  35. O’Connor RC, Connery H, & Cheyne WM (2000). Hopelessness: The role of depression, future directed thinking and cognitive vulnerability. Psychology, Health, & Medicine, 5(2), 155–161. [DOI] [PubMed] [Google Scholar]
  36. O’Connor RC, Fraser L, Whyte M-C, MacHale S, & Masterson G (2008). A comparison of specific positive future expectancies and global hopelessness as predictors of suicidal ideation in a prospective study of repeat self-harmers. Journal of Affective Disorders, 110(3), 207–214. [DOI] [PubMed] [Google Scholar]
  37. O’Connor RC, & Kirtley OJ (2018). The Integrated Motivational-Volitional model of suicidal behaviour. Philosophical Transactions of the Royal Society B, 373(1754), 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Piscopo K, Lipari RN, Cooney J, & Glasheen C (2016). Suicidal thoughts and behavior among adults: Results from the 2015 National Survey on Drug Use and Health. NSDUH Data Review. Retrieved from http://www.samhsa.gov/data/.
  39. Pinheiro J, Bates D, DebRoy S, Sarkar D, R Core Team (2021). nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1–152, https://CRAN.R-project.org/package=nlme.
  40. Qiu T, Klonsky ED, & Klein DN (2017). Hopelessness predicts suicide ideation but not attempts: A 10-year longitudinal study. Suicide and Life-Threatening Behavior, 47(6), 718–722. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Ribeiro JD, Huang X, Fox KR, & Franklin JC (2018). Depression and hopelessness as risk factors for suicide ideation, attempts and death: meta-analysis of longitudinal studies. The British Journal of Psychiatry, 212(5), 279–286. [DOI] [PubMed] [Google Scholar]
  42. RStudio Team (2020). RStudio: Integrated Development for R. RStudio, PBC, Boston, MA: URL http://www.rstudio.com/. [Google Scholar]
  43. Sargalska J, Miranda R, & Marroquin B (2011). Being certain about an absence of the positive: Specificity in relation to hopelessness and suicidal ideation. International Journal of Cognitive Therapy, 4(1), 104–116. [Google Scholar]
  44. Shaffer D, Fisher P, Lucas CP, Dulcan M, & Schwab-Stone M (2000). NIMH Diagnostic Interview Schedule for Children Version IV (NIMH DISC-IV): Description, differences from previous versions, and reliability of some common diagnoses. Journal of the American Academy of Child and Adolescent Psychiatry, 39(1), 28–38. [DOI] [PubMed] [Google Scholar]
  45. Smith JM, Alloy LB, & Abramson LY (2006). Cognitive vulnerability to depression, rumination, hopelessness, and suicidal ideation: Multiple pathways to self-injurious thinking. Suicide and Life-Threatening Behavior, 36(4), 443–454. [DOI] [PubMed] [Google Scholar]
  46. Snijders TA, & Bosker RJ (2011). Multilevel analysis: An introduction to basic and advanced multilevel modeling. 2nd Ed, London: Sage. [Google Scholar]
  47. Snyder CR, Harris C, Anderson JR, Holleran SA, Irving LM, Sigmon ST, Yoshinobu L, Gibb J, Langelle C, & Harney P (1991). The will and the ways: Development and validation of an individual-differences measure of hope. Journal of Personality and Social Psychology, 60(4), 570–585. [DOI] [PubMed] [Google Scholar]
  48. Steed L (2001). Further validity and reliability evidence for Beck Hopelessness Scale scores in a nonclinical sample. Educational and Psychological Measurement, 61(2), 303–316. [Google Scholar]
  49. Storch EA, Roberti JW, & Roth DA (2004). Factor structure, concurrent validity, and internal consistency of the Beck Depression Inventory – Second Edition in a sample of college students. Depression and Anxiety, 19(3), 187–189. [DOI] [PubMed] [Google Scholar]
  50. Tucker RP, O’Connor RC, & Wingate LR (2016). An investigation of the relationship between rumination styles, hope, and suicide ideation through the lens of the Integrated Motivational-Volitional model of suicidal behavior. Archives of Suicide Research, 20(4), 553–566. [DOI] [PubMed] [Google Scholar]
  51. Van Orden KA, Witte TK, Cukrowicz KC, Braithwaite SR, Selby EA, & Joiner TE Jr. (2010). The interpersonal theory of suicide. Psychological Review, 117(2), 575–600. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Whisman MA, Perez JE, & Ramel W (2000). Factor structure of the Beck Depression Inventory - Second Edition (BDI-II) in a student sample. Journal of Clinical Psychology, 56(4), 545–551. [DOI] [PubMed] [Google Scholar]

RESOURCES