Investigating Correlates of Suicide Among Male Youth: Questioning the Close Affinity Between Suicide Attempts and Deaths

Abstract

This study was based on a sample of male high school students who completed National Longitudinal Adolescent Health Surveys in 1994, 1995, and 2001. We studied these students prospectively, comparing those who later died by suicide (n = 21) with those who were still living (n = 10,101). We employed chi-square and analysis of variance tests for statistical significance between suicide decedents and living respondents. Results showed suicide decedents were more likely to have experienced the suicide loss of another family member, to have been expelled from school, to have engaged in more delinquent actions including fighting, and to have greater involvement with the criminal justice system. Although one might have expected suicide casualties to have exhibited a greater amount of suicidal thoughts, attempts, and higher incidences of suicidality among their friends, our analyses did not find that these factors were associated with actual suicides. Should these findings be replicated, this would point to a need to refine youth suicide risk assessments. Collecting life histories, as well as identifying patterns of delinquency and fighting, may serve as more potentially fruitful means for assessing genuine suicide risk than some traditional risk assessment methods.z

Previous suicide attempts are often thought to be strong predictors of future attempts or eventual suicide death (Coryell et al., 2002; Coryell & Young, 2005; Joiner et al., 2005; Sokero et al., 2005). Compared to those without a prior attempt, individuals with a previous attempt have been found in one study to be at a 773% greater risk of suicide (Ziherl & Zalar, 2006). A meta-analysis that followed suicide attempters (SA) for between 5 and 37 years reported an absolute risk of death by suicide of 7%–13%, which roughly corresponds to a 30–40 times increased death risk from suicide compared to the general population (Harris & Barraclough, 1997). Variabilities between those attempting suicide and those dying by suicide may be affected by many factors: demographic differences within samples, sociocultural differences and contexts, and whether samples were composed primarily of clinical patients or general population samples, among other factors. Higher mortality from nonsuicide-related causes also may occur for those with a history of at least one suicide attempt (Ostamo & Lӧnnqvist, 2001).

Yet, recent research has challenged the assumption of a close affinity between prior attempts and death by suicide. The studies challenging this conventional wisdom can be categorized into one of two types of research: those contrasting suicide decedents with suicide attempters (SD/SA), with the bulk of the research falling into this subgroup; or suicide decedents as contrasted with those dying from other causes (SD/OD), with only one study (DeJong, Overholser, & Stockmeier, 2010) falling into this subgroup. Diverging personality traits have been indicated between those who attempt and those who actually die by suicide, in addition to findings of gender differences and variations in the violent method chosen (Hirvikoski & Jokinen, 2012). Although there is some overlap between those who attempt and actual suicide deaths, important differences between these groups appear to exist (DeJong et al., 2010; Fushimi, Sugawara, & Saito, 2006; Gagnon, Davidson, Cheifetz, Martineau, & Beauchamp, 2009; Uribe et al., 2013).

One study comparing selected subgroups matched for age and depression from coroner’s records found that suicide decedents were more likely than other decedents to use alcohol or drugs prior to death, more likely to leave suicide notes, and more likely to have previously encountered significant job stress and financial problems (DeJong et al., 2010). Another study, conducted using emergency room visitors to a Spanish hospital and coroner’s records in that locality, found that those who died by suicide were more likely to be male, older, living alone, with more somatic problems, more depressed, and had chosen more lethal means (Uribe et al., 2013). Importantly, suicide decedents were less likely to have received support from mental health caregivers, in comparison with attempters. A Japanese study confirmed most of these results and found significantly more suicide attempts among SA, compared with those who had actually died by suicide (Fushimi et al., 2006).

Of particular interest is a study that examined suicide decedents aged between 14 and 25, with an age- and gender-matched sample of Canadian youth (Gagnon et al., 2009). Friends and families of the suicide decedent were interviewed on several occasions and contrasted with similar interviews with demographically matched peers who had attempted suicide. Decedents were less likely to have received previous mental health care compared to controls and tended to internalize their conflicts and emotions. Furthermore, a high percentage of suicide decedents (72%) had taken their lives on their first attempt (Gagnon et al., 2009). This finding appears to be consistent with some past research, including a study of White American males over age 45, which found 88% taking their lives on their first attempt (Maris, 1981); a Finnish psychological autopsy study, which found a 56% rate (Isometsӓ & Lӧnnqvist, 1998); and recent study, which found a 75% rate, based on bereaved parents’ information (Feigelman, Jordan, McIntosh, & Feigelman, 2012). Notably, however, these studies are limited by the often sparse and/or second-hand information usually available on suicide decedents as compared to attempters.

Findings demonstrate that attempters appear to be more inclined to seek mental health services compared to those who actually complete suicide (Fushimi et al., 2006; Gagnon et al., 2009) should have great import for suicide prevention. A number of studies have demonstrated that many problem-prone youth, who also stand at a higher risk for suicide, often remain disinclined to seek mental health treatment and services (Clement et al., 2015; Colognori et al., 2012; O’Connor, Martin, Weeks, & Ong, 2014; Rickwood, Deane, Wilson, & Ciarro-chi, 2005; Vanheusden et al., 2008). Some youth, who would otherwise seek care, do not do so for fear of being stigmatized (Clement et al., 2015). Given these findings, it would be helpful to have longitudinal measurements of both suicide attempters and completers in seeking counseling over the passage of time. This would enable us to see whether emerging youth problems bring more youth into counseling and whether there may be differing rates for obtaining counseling among suicide completers and attempters.

Recently, a new opportunity to longitudinally observe both suicide attempters and decedents has become available. In March 2014, the National Longitudinal Survey of Adolescent and Adult Health (Add Health) matched survey decedents to the death records in the National Death Index (NDI), revealing that approximately 10% of deceased survey participants died from self-inflicted causes.

The Add Health Survey began in 1995 with a nationally representative sample of more than 20,000 high school students, first interviewed in 1995; then again, approximately a year later (at Wave 2); once more, 6 years afterward (at Wave 3); and finally again, 7 years later, at Wave 4. Each successive survey wave asked respondents at least seven questions about their suicidal thoughts, attempts, and friends’ and family members’ recent suicides and attempts and included a variation of the Center for Epidemiologic Depression Scale (CES-D; Radloff, 1977).

Focusing on suicide decedents, we examined hypotheses about youth suicide based on previous studies investigating suicide attempt behavior and its correlates. As prior studies have found associations between attempted suicide and depression (Nanayakkara, Misch, Chang, & Henry, 2013), self-esteem differences (Maraš et al., 2013), parental closeness (Piña-Watson, Castillo, Rodriguez, & Ray, 2014), homo-sexual identification (Russell & Joyner, 2001), delinquent and/or criminal conduct (Langhinrichsen-Rohling & Lamis, 2008), substance use (Hallfors et al., 2004), and suicidal contagion (Abrutyn & Mueller, 2014), we tested the predictive value of these factors for predicting youth suicide deaths. We also examined one additional hypothesis, not linked to suicide ideation and attempts; namely, that suicides are more common among those with greater access to firearms (Miller, Lippmann, Azrael, & Hemenway, 2007). But, given the paucity of completed suicides in our sample (n = 21), we offer our findings provisionally, claiming our data as more tentative and exploratory than predictive.

METHOD

Upon identifying 126 Add Health respondents deceased by Wave 4, the Add Health team matched these cases to NDI death records, which include cause-of-death information. In addition, they also presented more than 1,500 potentially missing cases to NDI with the expectation of uncovering additional death cases. Ultimately, they verified 227 respondent deaths. In an effort to protect respondent confidentiality, the Add Health staff created a broader categorization of death causes than the more specific ICD-9 and ICD-10 death cause codes (Add Health, 2014). They grouped cause-of-death into six categories: (1) accidental deaths (n = 66; 29%), including motor vehicle accidents, drug poisonings, drownings, pedestrian accidents, and similar events; (2) intentional self-injury (n = 22; 10%), including suicides, intentional self-injury, and terrorist suicide cases; (3) assaults (n = 19; 18%), including homicides, terrorism-homicides, and other assaults; (4) all other natural death causes (n = 72; 32%); and (5) cause not available, but death confirmed (n = 17; 7%). An additional 31 unmatched cases comprised the remaining 14%.

In our analyses, we initially regrouped these categories into three groups: suicide decedents, n = 22; all other decedents, n = 205; and living respondents, n = 20,547. Upon discovering that 21 of the 22 suicide decedents were males, we thought it would be more appropriate if we contrasted the male suicide decedents (n = 21) against living males (n = 10,101) for our analysis. We then conducted bivariate tests of each hypothesis with chi-square tests, contrasting the male suicide decedent with those males who were still living. Given the extremely modest number of suicide decedents in the sample, we conducted all tests using Fisher’s exact test statistic. In cases that we had scalar variables available, we applied one-way analysis of variance (ANOVA) tests. We also computed tests for effect size utilizing Cramer’s V for cross-tabular associations and Cohen’s d for mean comparison relationships.

Add Health sampling procedures and survey details are available in detail at http://www.cpc.unc.edu/projects/addhealth. The Add Health survey had a high response rate, with 20,745 respondents participating in the Wave 1 interview. Approximately 5,000 high school seniors were intentionally omitted from re-interview at Wave 2, yielding a sample n of 14,738. At Wave 3, 7 years after the first survey, 15,170 respondents were re-interviewed, representing approximately 73% from the original Wave 1 sample. For the male suicide sample, data from 21 were available at Wave 1, 17 remained at Wave 2, and 11 were still alive at Wave 3. No presently known Add Health suicide decedent lived long enough to complete the Wave 4 survey. Virtually all survey respondents were between ages 13 and 19 at Wave 1. Each survey wave took place within a 12- to 15-month period, primarily in 1995 for Wave 1, 1996 for Wave 2, and 2001–2002 for Wave 3. All respondent deaths occurred anywhere from 4 to 12 years after the first survey.

Each cross-tabulation shows the available Ns in the analysis and whether there were any missing cases in particular cross-tabulations. Missing cases were simply omitted from analyses, unless their numbers were exceptionally high, whereupon this was discussed in the text; this occurred in a single instance.

Measures

Depression.

Symptoms of depression were assessed via a 19-item variation of the CES-D scale (Radloff, 1977) at Wave 1, assessing depression experienced during the past 12 months. Responses ranged from 0 (Rarely or none of the time) to 3 (Most or all of the time), with higher total scores indicating more prevalent depressive symptoms. Sample items included “felt you could not shake off the blues” and “felt depressed.” Alpha coefficient was .86.

Self-Esteem.

A self-esteem scale was created from nine items, offered at Wave 1, on a 5-point Likert-type scale. Sample items include “You have a lot of good qualities” and “You have a lot to be proud of.” Alpha coefficient was .86.

Parental Closeness.

We created scales of parental closeness from three questions that inquired about parental relationships at Wave 1. The questions employed a 5-point Likert agree–disagree scale and included items such as “Most of the time your mother is warm and loving to you,” “You are satisfied with the way your mother and you communicate with each other,” and “Overall you are satisfied with your relationship with your mother.” The same set of questions was asked about fathers. Alpha coefficient was .84 for the mother-related questions and .88 for the father-related questions.

Delinquency.

At Wave 1, all respondents were given a list of 15 different delinquent acts and asked whether they had engaged in any and if so, how frequently in the past 12 months. The list included such behaviors as painting graffiti, deliberately damaging property, stealing things, getting into a physical fight, going into a house to steal things, selling illegal drugs, and taking a car without its owner’s permission. Alpha coefficient was .83.

Violence.

We created a 9-item scale of fighting and violent actions at Wave 1. Respondents were asked if, in the last 12 months, they had witnessed someone shoot or stab someone else; if someone had pulled a knife or gun on them; if someone had shot or knifed them; if they were jumped; if they had shot or knifed someone else; and if they had been seriously injured in a fight that required medical care. Alpha coefficient was .72.

RESULTS

Demographic Contrasts

Suicide decedents were predominately White (76%) as compared to only slightly over half (53%) for the living (Fisher’s exact p < .05; Table 1); they were also somewhat younger when they completed their surveys, although this difference was nonsignificant. Household incomes did not differ significantly between suicide decedents (SD) and the living. Slightly greater numbers of the suicide casualties came from ruptured families as compared to living respondents, although these differences too were nonsignificant. There were no differences between SD and living respondents in being self-identified as a gay/lesbian or bisexual.

Table 1.

Demographic and Other Important Distinguishing Characteristics of Living Compared to Suicide-Deceased Male Respondents

Characteristic	Percent Still Living	Percent Suicide	χ2 (df)	χ2 p Value	Fisher’s Exact p Value	Cramer’s V
Race (Wave 1)
White	53.0	76.2	4.51 (1)	.034	.047	.021
Non-White	47.0	23.8
Total sample size (N)	10,022	21
Age When Entered Sample (Wave 1)
<14	18.8	33.3	3.61 (2)	.165	.174	.019
15–17	54.2	52.4
18+	27.0	14.3
Total sample size (N)	10,093	21
Age at Time of Death (National Death Index)
17 or under	-	9.5	-	-	-	-
18–20	-	9.5
21–25	-	76.2
26+	-	4.8
Total sample size (N)	-	21
Household Income (Wave 1)
<$20,000	24.4	33.3	1.32 (3)	.724	.716	.013
$21,000-$40,000	31.1	22.2
$41,000-$60,000	23.7	27.8
$61,000+	20.8	16.7
Total sample size (N)	7,553	18
Parentage (Wave 1)
Lived with both bio-parents	42.1	28.6	1.64 (2)	.44	.39	.013
One or both parents died	4.9	4.8
Step/one-parent/adoption family	53.0	66.7
Total sample size (N)	10,101	21
Identified as Gay/Lesbian (Wave 1)
Yes	10.6	9.5	0.02 (1)	.875	1.00	.002
No	89.4	90.5
Total sample size (N)	10,101	21
Ever Repeat a Grade (Wave 1)
Yes	27.2	23.8	0.12 (1)	.727	1.00	.004
No	72.8	76.2
Total sample size (N)	10,078	21
Ever Expelled from School (Wave 1)
Yes	6.6	19.1	5.19 (1)	.023	.05	.023
No	93.4	81.0
Total sample size (N)	10,052	21
Received Counseling (Wave 1 or 2)
Yes	10.6	19.1	1.55 (1)	.21	.27	.012
No	89.4	80.9
Total sample size (N)	10,071	21
Drug or Alcohol Abuse Treatment (Wave 1 or Wave 2)
Yes	4.9	9.5	0.95 (1)	.33	.27	.010
No	95.0	90.5
Total sample size (N)	10,101	21

Further, there was no greater tendency for SD to repeat a grade, as compared to living respondents. Yet, SD had a nearly three times higher rate of school expulsions, 19% as compared to 7% (Fisher’s exact p < .05). Results showed no variations between SD and living respondents in ever receiving psychological counseling or drug/alcohol abuse treatment at Wave 1 or Wave 2.

Depression, Suicide Thoughts, Attempts, and Self-Esteem Differences

No significant differences were found on depression when SD were compared with the living on the CES-D (Table 2) at Wave 1 or Wave 2. Suicide decedents also reported similar levels of suicidal thoughts, attempts, and serious attempts (those requiring emergency room or doctor’s care) as compared to living respondents at both Wave 1 and Wave 2. In addition, both SD and the living reported similar rates of friends’ suicide attempts, friends’ suicide deaths, and family members’ suicide attempts at Wave 1 and Wave 2. We also examined suicide thoughts, attempts, friend’s attempts, and friend’s deaths at Wave 3, none of which showed any associations with actual suicide deaths (these findings were not included in the tables). Nevertheless, SD reported significantly higher rates of a family member’s suicide within the past year at Wave 1 or Wave 2, 10% compared to 1.2% among the living (Fisher’s exact p < .03). There were no significant differences in self-esteem between SD and the living.

Table 2.

Depression, Suicidalities, Suicide Contagion, Self-esteem, and Parental Closeness Among Suicide Decedents Compared to Living Male Respondents

Characteristic	Percent Still Living	Percent Suicide	χ2 (df)	χ2 p Value	Fisher’s Exact p Value	Cramer’s V
Any Suicidal Thoughts (Wave 1 or Wave 2)
Yes	13.4	19.1	0.59 (1)	.44	.51	.008
No	86.6	81.0
Total sample size (N)	10,101	21
Any Suicide Attempt (Wave 1 or Wave 2)
Yes	3.4	4.8	0.12 (1)	.73	.51	.004
No	96.6	95.2
Total sample size (N)	10,101	21
Serious Suicide Attempt Requiring Medical Attention (Wave 1 or Wave 2)
Yes	1.0	0.0	0.21 (1)	.65	1.00	.005
No	99.0	100.0
Total sample size (N)	10,132	21
Family Member’s Suicide Attempt (Wave 1 or Wave 2)
Yes	4.8	9.5	1.03 (1)	.31	.27	.010
No	95.2	90.5
Total sample size (N)	10,101	21
Family Member’s Suicide (Wave 1 or Wave 2)
Yes	1.2	9.5	12.62 (1)	.0001	.025	.035
No	98.8	90.5
Total sample size (N)	10,101	21
Friend’s Attempted Suicide (Wave 1 or Wave 2)
Yes	17.0	19.1	0.063 (1)	.801	.771	.003
No	83.0	80.9
Total sample size (N)	10,101	21
Friend’s Suicide (Wave 1 or Wave 2)
Yes	4.1	9.5	1.58 (1)	.21	.21	.013
No	95.9	90.5
Total sample size (N)	10,101	21

Characteristic	Still Living	Suicide	F Test	p Value	Cohen’s d (95% CI)
Mean (and SD) score on Depression Scale (Wave 1)	10.4 (6.8)	8.6 (6.8)	1.48	.22	.26 (−0.16, 0.69)
Total sample size (N)	10,025	21
Mean (and SD) score on Depression Scale (Wave 2)	11.0 (7.2)	8.5 (6.5)	2.00	.16	.34 (−0.13, 0.82)
Total sample size (N)	7,043	17
Mean (and SD) score on Self-Esteem Scale (Wave 1)	16.3 (4.7)	15.3 (4.5)	0.94	.33	.21 (−0.22, 0.64)
Total sample size (N)	10,023	21
Mean (and SD) closeness to mother (Wave 1)	5.1 (2.0)	4.3 (1.5)	2.49	.11	.37 (−0.09, 0.83)
Total sample size (N)	9,428	18
Mean (and SD) closeness to father (Wave 1)	5.6 (2.3)	4.8 (2.0)	1.69	.19	.38 (−0.19, 0.94)
Total sample size (N)	7,253	12
Mean (and SD) feeling of being loved & wanted (Wave 1)	1.7 (0.7)	1.5 (0.5)	1.40	.24	.26 (−0.17, 0.69)
Total sample size (N)	10,066	21

Parental Closeness

In regard to parental closeness, neither scale (i.e., mother’s or father’s) showed any associations with death by suicide compared to living respondents. Father’s closeness scale scores and missing data patterns indicated a high percentage (43%; n = 9) of SD were not living with their biological fathers at the time of the Wave 1 interview and, hence, did not report on their closeness to fathers. For the remaining 12 cases, SD did not differ in their reports of closeness to fathers as compared to living respondents. A one-way ANOVA also showed no differences between SD and the living in feeling close to one’s nonresidential father.

Delinquency, Violence, and Encounters with the Criminal Justice System

Suicide decedents had significantly higher self-reported delinquency rates than living respondents in one-way ANOVA tests, p < .0003 (Table 3). A similar delinquency scale, offered to respondents at Wave 2, yielded similar results, with SD scoring significantly higher on delinquency items than living respondents (this result is not displayed in our tables). SD also scored significantly higher on the fighting scale than living respondents. To ascertain whether these ANOVA associations reflected homogeneity of variances, equality of variance tests were conducted, and all tests showed approximately equal variances across compared groups.

Table 3.

Delinquency, Gun Access, Exposure to Abuse, and Substance Use Among Suicide Decedents, Com- pared to Living Male Respondents

Characteristic	Still Living	Suicide	F Test	p Value	Cohen’s d (95% CI)
Mean (and SD) score on Delinquency Scale (Wave 1)	3.3 (3.1)	5.3 (4.3)	9.11	.0025	.66 (0.23, 1.09)
Total sample size (N)	9,893	21
Mean (and SD) score on Fighting and Violence Scale (Wave 1)	1.3 (1.7)	2.0 (1.8)	4.60	.032	.47 (0.04, 0.90)
Total sample size (N)	9,891	21

Characteristic	Percent Still Living	Percent Suicide	χ2 (df)	χ2 p Value	Fisher’s Exact p Value	Cramer’s V
Physical Fight in Past 12 Months (Wave 1)
Yes	23.5	52.4	9.7 (1)	.002	.004	.031
No	76.5	47.6
Total sample size (N)	10,085	21
Serious Injury from a Fight (Wave 1)
Yes	12.2	23.8	2.6 (1)	.11	.17	.016
No	87.8	76.2
Total sample size (N)	10,015	21
Used a Weapon in a Fight (Wave 1)
Yes	9.6	23.8	4.9 (1)	.028	.045	.022
No	90.4	76.2
Total sample size (N)	10,004	21
Ever Been Arrested (Wave 3)
Yes	18.7	36.4	2.3 (1)	.133	.134	.018
No	81.3	63.7
Total sample size (N)	7,032	11
Number of Times Arrested Before Age 18 (Wave 3)
None	92.3	72.7	16.6 (2)	.0001	.012	.048
One	3.9	0.0
Two or more	3.8	27.3
Total sample size (N)	7,080	11
Ever Been Convicted or Plead Guilty to a Crime (Wave 3)
Yes	3.7	18.2	6.5 (1)	.011	.060	.030
No	96.3	81.8
Total sample size (N)	7,085	11
Sentenced to Probation or Jail/Prison (Wave 3)
Not arrested	91.7	81.8	4.5 (3)	.212	.126	.025
Probation	4.9	9.1
Jail/prison	1.6	9.1
Other	1.8	0.0
Total sample size (N)	7,077	11
Ran Away from Home (Wave 1 or 2)
Yes	7.4	23.8	8.3 (1)	.004	.016	.029
No	92.7	76.2
Total sample size (N)	10,002	21
Has Easy Access to a Gun (Wave 1 or 2)
Yes	30.8	42.9	1.4 (1)	.23	.24	.012
No	69.3	57.1
Total sample size (N)	10,101	21
Smoking Days in Past 30 Days (Wave 1)
None	73.9	52.4	10.1 (3)	.018	.018	.032
5 or less days	9.1	4.8
6 to 29 days	7.8	19.1
Daily	9.2	23.8
Total sample size (N)	10,018	21
Got Drunk in Past Year (Wave 1)
Once a week or more	7.2	0.0	1.6 (2)	.44	.611	.013
Monthly	22.6	23.8
Never	70.2	76.2
Total sample size (N)	10,076	21
Problems with Parent(s) Because of Drinking (Wave 1)
Never	89.7	81.0	2.2 (2)	.339	.138	.015
Once	6.4	9.5
Two times or more	3.9	9.5
Total sample size (N)	10,086	21
Problem at School Because of Drinking (Wave 1)
Never	96.7	95.2	1.4 (2)	.505	.509	.012
Once	1.8	4.7
Two times or more	1.5	0.0
Total sample size (N)	10,090	21
Got into Fights Due to Alcohol (Wave 1)
Yes	8.6	14.3	0.9 (1)	.353	.419	.009
No	91.4	85.7
Total sample size (N)	10,092	21
Times Used Marijuana (Wave 1)
Never	71.1	63.2	1.4 (2)	.508	.532	.012
1–10 times	16.4	26.3
More than 10 times	12.4	10.5
Total sample size (N)	9,654	19
Tried Using Cocaine (Wave 1)
Yes	3.7	0.0	0.7 (1)	.391	1.00	.009
No	96.3	100.0
Total sample size (N)	9,886	19
Tried Using Inhalants (Wave 1)
Yes	6.3	15.0	2.6 (1)	.109	.128	.016
No	93.7	85.0
Total sample size (N)	9,892	20
Tried Using Other Illegal Drugs (Wave 1)
Yes	7.8	10.0	0.13 (1)	.716	.667	.004
No	92.2	90.0
Total sample size (N)	9,856	20

In addition, over half of SD reported getting into a fight in the past year, compared to only 23% for living respondents, p < .004. Twice as many SD reported having a serious injury from a fight than the living sample (24% vs. 12%, an association that fell below the .05 significance thresh-old), and significantly more reported using a weapon in a fight (24% vs. 10%, p < .05).

Although by the time Wave 3 arrived only 11 SD remained alive, we examined questions presented specifically in that survey regarding contact with the criminal justice system. A total of 36% of SD indicated having been arrested, as compared to only 19% for living respondents, an association that fell below the .05 significance thresh-old, with p = .13. Significantly more SD reported being arrested two or more times as compared to living respondents, 27% as compared to 4%, p < .01. Finally, more SD reported being convicted or pleading guilty to a crime than living respondents, 18% as compared to 4% (Fisher’s exact p < .06).

Suicide decedents were also three times more likely to report running away from home compared to living respondents (24% vs. 7%), p < .02 (Table 3). Nearly half (43%) reported having easy access to a gun in their homes, at Wave 1 or Wave 2, compared to 31% for living respondents, differences that fell within the range of chance variabilities.

Drug Use

An uneven pattern of differences in drug-taking behavior was observed comparing SD to living respondents (Table 3). Suicide decedents were significantly more likely to report heavier cigarette smoking compared to living respondents, p < .02; however, SD did not report heavier drinking as compared to living respondents. Suicide decedents reported somewhat more problems with parents because of their alcohol use, but these differences were nonsignificant. Neither of the subgroups differed on school problems or fights because of heavy drinking.

In terms of illegal drug use, SD were no more inclined to smoke marijuana compared to living respondents. There were no significant differences in inhalant or cocaine use among the two subgroups.

Suicide Attempt History

As previous research has found drastically heightened risks of suicide (Ziherl & Zalar, 2006) and higher mortality risks from all other causes among previous attempters (Ostamo & Lӧnnqvist, 2001), we investigated this possibility within our sample. Note that it is possible that former suicide attempters could die from a suicide that was not correctly classified, as well as succumbing to other causes that may have been compounded by their depression and/or other psychiatric difficulties (Timmermans, 2006), so we offer our results with caution.

The Add Health study collected information on respondents’ suicide attempts occurring at either Wave 1, Wave 2, or Wave 3. Although 1,343 respondents indicated a previous suicide attempt at any wave, in only one instance did a previous attempter later die by suicide. Thus, only 0.07% of all attempters actually died by suicide in the Add Health sample. There were 16 deaths from other causes among those who previously attempted suicide, yielding an overall death rate among attempters of 1.19%. There were no suicide deaths among those who made two or more previous attempts, and of the 367 who had a previous serious suicide attempt (requiring medical care), only four died from any cause, at a rate 1.09%. Correspondingly, for the Add Health respondents who had not reported a previous suicide attempt, the suicide death rate was 1.08% and their overall death rate was .97%. Thus, those who had not previously attempted suicide had a slightly higher suicide death rate than those who had, 1.08% versus 0.07%. The overall death rate for previous attempters slightly exceeded the death rate for nonattempters, at 1.19% versus 0.97%. In conclusion, we found no increased risk of suicide death among previous attempters compared to nonattempters and little differences in overall death rates comparing attempters to completers.

Suicide Attempts Versus Completions and Counseling Experiences

We have already noted that those who complete suicide showed no higher use of mental health counseling services than the living, 19% compared to 11% (at Wave 1 or Wave 2). Despite having mounting problems over the years from school expulsions, running away from home, delinquent behavior, and criminal involvements, none of the original 17 respondents who had not already received mental health counseling at Wave 1 sought it in later years at Wave 2 or at Wave 3 (it should be noted that, at Wave 1, of the 21 males who eventually died by suicide, 4 reported receiving mental health counseling at Wave 1 and 17 reported not having it). Thus, these suicide decedents seem to fit the pattern of a group who, in their early twenties, were averse to gaining counseling support.

Among suicide attempters, an entirely different pattern was noted. A total of 38% of male SA saw a counselor either at Wave 1, Wave 2, or Wave 3, compared to 12% for those who had not made a suicide attempt (this association yielded a significant chi-square p value of .001 and a Cramer’s V of .15). For those who made more than one suicide attempt at any survey wave, 55% had sought counseling compared to only 13% for all others (chi-square p = .001, V = .09). Thus, with greater numbers of problems, especially from attempting suicide, these vulnerable young males sought help from mental health professionals.

For the most part, our effect size comparisons showed almost uniformly higher magnitude among statistically significant associations, as compared to nonsignificant associations. Yet, we must acknowledge that all effect sizes reported were relatively small. These results suggest that a combination of a broad array of causal forces is associated with suicide and that a single “smoking gun” correlate does not exist.

DISCUSSION

The annals of suicide research rarely provide opportunities for longitudinally studying youth who ultimately complete suicide. Even more unusual are studies that include a wealth of information on potential suicide risks within a youth sample, where data are obtained before their deaths and collected over time in successive survey waves. Such are the unique benefits available from the Add Health-NDI data set. In this investigation, we have taken a number of familiar hypotheses about suicide risk, derived primarily from clinical populations, and explored their veracity in a general population sample of males, contrasting the suicide casualties with those who remained alive.

The findings showing elevated suicide risk for young adults who report another family member’s recent suicide should come as no surprise. Previous research has confirmed higher suicide rates among the first-degree relatives of suicide decedents (Agerbo, Nordentoft, & Mortensen, 2002; Qin, Agerbo, & Mortensen, 2002). Yet, we were surprised to find no associations between a friend’s suicide and their attempt. Further, our finding that previous suicide attempts, thoughts, or attempts that required medical attention had no association with actual death by suicide was also unexpected. In fact, we saw no instance of a youth who reported needing medical attention for a suicide attempt ultimately become a suicide casualty. While these results might seem unusual and surprising, we are not the first researchers to uncover such findings. Similar results were found with data from the National Violent Death Reporting System (Kaplan et al., 2014), where a sizable fraction of suicide decedents did not report histories of substance abuse, mental health disorders, suicide ideation, or attempts prior to their deaths. We also found no clear evidence of heightened depression, substance abuse, or utilization of mental health services among our sample of male teenagers dying by suicide.

While our findings may seem to be at odds with fundamental precepts of a close affinity between suicide ideation, attempts, and deaths, evidence is beginning to mount showing a lack of close convergence between suicide attempts and death. This was true in some of the studies we cited at the outset of this report and seems especially true in cases of youth suicide. Other studies have confirmed this lack of a close affinity between attempts and death among youth. Goldsmith, Pellmar, Kleinmen, and Bunney (2002) suggested a ratio of 200:1 attempts for every suicide among young females. A more recent longitudinal study of a French Canadian youth sample found that of 9% who reported a prior suicide attempt, very few of them (1 in 500) had actually died by suicide (Brezo et al., 2007). Thus, our findings are consistent with others showing a lack of a close correspondence between attempts and youth suicides.

Detecting differences between suicide attempters and completers assumes a central place in suicide prevention, with efforts designed to engage more problem-prone youth and bring them into seeking mental health counseling. Suicide attempters appear more likely to seek to resolve their life adjustment difficulties with help from mental health professionals. Suicide completers, by contrast, appear less inclined to place their faith in these agents. More averse to getting care, they may view mental health professionals with suspicion and distrust. They also may want to avoid becoming stigmatized as “patients.” Based on these longitudinal findings, it would appear that early interventions, when problem-prone youth are in their high school years, may be the best time to engage them. After leaving high school, none of our suicide respondents started therapy who had not previously seen a counselor. Early suicide prevention interventions appear supported by these findings.

Findings on the family lives of these suicide casualties were paradoxical. While results showed no differences between suicide decedents and the living in feelings of parental closeness, more suicide casualties reported running away from home. Another study based on the Add Health data set did not find a close mesh between parental closeness and running away from home, suggesting some variability in this association as it may be affected by factors such as gender differences, or differences in sexual orientation, as well as other elements of family dynamics (Pearson & Wilkinson, 2013). Such findings seem counterintuitive unless they occurred at different times in the lives of these respondents. It will be an important task for future research to more fully investigate the overall quality of the family lives of suicide casualties in contrast with those who did not take their lives.

Other important findings we obtained show the importance of being expelled from school and of engaging in violent or delinquent actions as precursors to young adult suicide. A portion of these respondents ended up in jails and juvenile detention facilities, where the risks of dying by suicide accelerate dramatically. One earlier study found a 9 to 15 times higher suicide rate for incarcerated young men as compared to their nonincarcerated counterparts (Hayes, 1989). It was also interesting to note that gay, lesbian, and bisexually identified youth were no more likely to die by suicide, although many studies, including this data set, have established associations between sexual identity and suicidal thoughts and attempts.

The Add Health study was conceived before researchers began to focus specifically on bullying as an instigator to suicide within youth communities. Yet, participation in a violent subculture, where getting beaten and beating others are part of the everyday cycle of events, may lead some to take their lives (Matza, 1964). Future research may find it fruitful to investigate the school and friendship relationships that suicide decedents had prior to their deaths and other unique features of their lives in contrast to living respondents.

This study has a number of limitations. Owing to the lack of completed suicides among Add Health females, we confined this study to males exclusively. The overwhelming majority of respondents, 76%, were between ages 21 and 25 when they died. Therefore, we cannot extend our findings beyond males within this narrow age range. Our small sample of suicide casualties resulted in low power to detect statistically significant differences. For example, when examining the criminal justice system exposure items from Survey Wave 3, where male suicide case numbers shrank down to a sample of 11, differences of nearly 20% or more between subcategories hardly showed up as statistically significant, despite obvious differences in arrest rates. Despite these limitations, this report offers some new and illuminating information on the shared characteristics of youth suicide casualties. Should these findings be replicated in future studies, this would point to a need to refine traditional youth suicidalities risk assessments. Confirmation of these findings would suggest that the collection of life histories, and identifying patterns of delinquency, fighting, and running away from home, as important events for detecting genuine suicide risk. When young males begin to experience expulsions from school, violent confrontations with peers, and/or getting into trouble with the police and the courts, these and other similar events may serve as a more clear warning of suicide risk than some traditional indicators used for assessing youth suicidality.