The relationship between Korean university students’ suicidal ideation and risk factors: a meta-analysis

ABSTRACT This study examines risk factors for university students’ suicidal ideation based on a meta-analysis of research published in South Korea from 2004 to 2019. Risk factors included depression, life stress, parents’ critical communication, and job stress. Publication bias and study-specific moderator variables were also examined. Depression had the largest effect size, followed by life stress, parents’ critical communication, and job stress. Meta-regression analysis showed differences in the effect sizes for depression, parents’ critical communication, and job stress according to publication year. The probability of publication bias was low. These findings can help to develop effective suicidal ideation prevention programs for university students. In particular, counselling centres should help intensively prevent and manage depression among university students. Furthermore, mental health services, such as education and counselling programs that address suicidal ideation, must be easily accessible by integrating the students’ homes, schools, and communities.


Introduction
Suicide rate has recently been increasing in South Korea. Korea National Statistical Office (2019) reports that more than 13,000 South Koreans commit suicide each year, the equivalent of approximately 36 people per day, and that South Korea's suicide rate is more than twice the average for OECD countries. In particular, suicide is emerging as a new and serious social problem among young Koreans. The suicide rate among Korean university students has increased by 60% over the past six years, and the number of university students who experience suicidal ideation is also increasing (Kim & Cha, 2018).
Currently, university students in South Korea suffer from job stress due to low employment rate and from economic stress due to economic polarization. They are also subject to interpersonal and academic stress (Jo et al., 2011). This leads many Korean university students to experience negative emotions such as anxiety, depression, and frustration. Persistent negative emotions may increase the likelihood of increasing lethargy, reduced enjoyment of life, and the development of suicidal ideation (Cukrowicz et al., 2011;Schwartz, 2011).
Suicidal ideation is a precursor to suicide and includes thinking about or actually planning one's death. Research on suicidal ideation in university students has continued to accumulate until the present day, and most previous studies have verified risk factors that increase suicidal ideation and protective factors that reduce it. However, it is difficult to assess the effectiveness of these factors because previous studies have differed in terms of measurement tool, study year, sample size, and so

Factors affecting suicidal ideation in university students
The variables that affect suicidal ideation in Korean university students can be distinguished into individual, family, and school factors, and these factors work in combination (Lockman & Servaty-Seib, 2016). Therefore, diverse factors must be considered to understand the phenomenon of suicidal ideation. Furthermore, factors that affect suicidal ideation can be either risk factors, which increase the risk of suicide, or protective factors, which reduce the risk. This study examines ways to reduce suicidal ideation among Korean university students by analysing risk factors. Based on previous studies of Korean university students, the major individual risk factors include depression, stress, impulsiveness, and low self-esteem; the major family risk factors include critical communication by parents, family discord, and childhood abuse by parents, and the major school factors include employment stress and interpersonal relationships at university Jeong et al., 2015).
Depression is one of the most powerful factors affecting suicidal ideation according to many prior studies (Chen et al., 2017;Pace & Zappulla, 2010;Tsujimoto et al., 2015). Pace and Zappulla (2010) found that higher levels of depression lead to more suicidal ideation because higher levels of depression resulted in less willingness to live, more negative thoughts, and more negative feelings. Furthermore, depression has important effects on suicidal ideation both directly and indirectly through other variables (Farabaugh et al., 2012;Lester, 2014).
The life stress experienced by Korean university students is also a major risk factor for suicidal ideation (Kim & Hong, 2016;Jin et al., 2015;Noh et al., 2014). According to previous research, students with higher life stress have higher levels of suicidal ideation. The stress they experience reduces their desire to live through negative emotions such as anxiety, depression, and frustration; they thus grow lethargic and lose a sense of value in their lives. If this continues, they may begin to want to end their lives. Additionally, excessive life stress can increase suicidal ideation because it weakens the selfdefence mechanisms needed to face challenges (Kelliher Rabon et al., 2018). A study by Kim and Hong (2016) found that the number of university students in South Korea who report serious life stress is increasing, as is the number of students who experience suicidal ideation as a result. Excessive life stress is thus a serious risk factor for suicidal ideation among university students.
Some previous studies have suggested that critical communication by parents could be an important family-related risk factor for suicidal ideation in Korean university students . Parents are the first humans with which children form relationships, and they serve as a human psychological and emotional base of safety. As such, they can be the most important influences on a person's personality, values, behavioural development, interpersonal relationships, and so on (Moretti et al., 2015). Parents' influence remains important even after their children grow up, particularly in South Korea, as Korean parents often identify with their children and maintain mutually dependent relationships with them (Kwon et al., 2016). Therefore, many studies of Korean university students' suicidal ideation have examined parents-related influencing factors. One of these, critical communication by parents, causes children to form negative egos and patterns of thinking, which can lead to more negative feelings and more suicidal ideation (Choi, 2012). Choi (2012) found that university students who experience considerable critical communication from their parents experience more suicidal ideation, indicating that the more parents criticize their children and communicate in a way that reduces their self-esteem, the more negatively they perceive their life, and this results in less self-confidence, less motivation for living, and greater likelihood of suicide.
Finally, many previous studies on suicidal ideation in Korean university students have identified school-related influencing factors among which job stress has recently emerged as important Seo & Lee, 2014).  found that of all causes of suicidal ideation among Korean university students, job stress accounts for the highest percentage. Excessive job competition due to job losses, increasing numbers of irregular workers, and the growing gap between the rich and the poor leads students to experience great stress related to finding a job, and even to consider suicide if the stress persists (Zhang et al., 2012). Specifically, excessive job stress reduces university students' self-esteem and self-efficacy, ultimately making them feel worthless and purposeless in life.
Previous research on Korean university students' suicidal ideation differs in terms of study design, measurement tool, sample size, and year of study. Thus, the different research environments and designs of each study may lead to different outcomes regarding the factors that affect university students' suicidal ideation . The most commonly used measurement tools in previous studies are the Scale for Suicide Ideation (SSI), developed by Beck et al. (1979), and the Suicidal Ideation Questionnaire (SIQ), developed by Reynolds (1987). The SSI includes 19 questions on specific attitudes, plans, and intensity of actions regarding suicide, all rated on a 3-point Likert scale; the SIQ measures the severity of suicidal ideation through 15 or 30 questions depending on age, all rated on a 7-point Likert scale. The present study examines whether the magnitude of each variable's impact on Korean university students' suicidal ideation differs depending on the measurement tool used.
Thus far, no meta-analysis has been conducted to account for the characteristics of various studies (e.g. sample size, publication year, publication type, etc.) on Korean university students' suicidal ideation. Therefore, the present study, which aggregates risk factors affecting Korean university students' suicidal ideation and analyzes the effects of different study characteristics on the results, is very meaningful in filling this gap.

Meta-analysis of risk factors for suicidal ideation
There are two main methods for aggregating the various results of reports on the same subject. Until the 1990s, narrative reviews were most commonly used. In a narrative review, experts on the subject summarize and draw conclusions regarding the various research findings. However, the strong subjective characteristics of this method have the disadvantage of producing potentially conflicting results from the same data. Research can also be synthesized by counting significant numbers of studies and deriving final results, but this method's results are also less verifiable. In contrast, the meta-analysis method determines whether to include or exclude studies based on clear criteria, and allows for comparison between the results of various studies by deriving an effect size. This has the advantage of validating causes of heterogeneity through sub-group analysis and meta-regression analysis based on study-specific characteristics (Borenstein et al., 2009).
The present study employs a meta-analysis to aggregate the factors affecting suicidal ideation in Korean university students by deriving their effect sizes. To do this, the average effect size of the major risk factors is first estimated. Then, to determine the reasons for heterogeneity of effect sizes among the results of each study, meta-regression analysis is performed with publication type, publication year, sample size, and measurement tool as moderator variables. Using this method, the present study aims to suggest effective ways to reduce suicidal ideation and prevent suicide in Korean university students.

Literature search
We collected studies published from 2004 (when studies on suicidal ideation in Korean university students first appeared) to 2019 using the keywords 'suicidal ideation,' 'suicidal thought,' and 'suicidal impulse' from databases such as the Research Information Sharing Service (RISS), the National Assembly Library, and Google Scholar. Articles were included based on the following criteria. First, because this study aimed to integrate research on suicidal ideation and related risk factors in university students, study participants were limited to university students. Second, to calculate the magnitude of effects, studies must have shown a correlation value between suicidal ideation and the risk factors; papers with no correlation values were excluded from the analysis. Third, in cases of overlap between theses and journal articles, the journal articles were included in the analysis. Figure 1 presents the paper selection per PRISMA guidelines (Moher et al., 2009). A total of 136 studies were identified through the computerized search. a total of 43 articles were excluded because they represented duplicates. An additional 19 were excluded because the titles/abstracts were not relevant. This process resulted in 74 studies to potentially include. The same authors then independently reviewed the full-length version of each study. Studies presenting redundant samples were identified and triaged based on the following parameters, with preference for inclusion presented in sequential order; (a) includes correlations examining suicidal ideation and risk factors, and (b) has a larger sample size. In cases where more than one article utilized the same sample, the articles were considered nonredundant if they reported on different criterion variables. The inclusion of a duplicate sample within these parameters would likely not artificially inflate or deflate individual effect sizes because distinct domains of suicidal ideation were examined. When one article presented two samples and one sample was redundant with the sample of another article, the article was retained for analysis of the nonredundant sample. At this stage, an additional 17 articles were excluded because they represented a redundant sample. Thus, we included a total of 57 articles. At this stage, an additional 17 articles were excluded because they represented a redundant sample (k = 11) or did not examine suicide ideation (k = 6). Thus, we included a total of 57 articles.

Study coding
Papers that met the inclusion criteria of the present study were selected and the corresponding data were entered. Each study was organized based on the author's name and research subjects and categorized according to publication type (thesis, journal article, research report, and other), publication year, sample size, and measurement tool (SSI, SIQ, and other).

Effect sizes
Effect sizes in the meta-analysis were calculated based on the correlation value r reported by each study included in the data. Because effect size is a standardized value, it allows comparison between studies. It also conveys information about direction along with the size of the values. First, effect size was calculated by converting the correlation value r computed by each individual study to Cohen's d. If the sample size was small, effect size was corrected using Hedges's g to account for bias in the estimate of the d-value (Higgins et al., 2003). An effect size of 0.2 is interpreted as small, 0.5 as medium, and 0.8 as large (Cohen, 1998).

Publication bias
The fact that the statistical significance or effect size depending on the sample size is referred to as 'publication bias'. In general, when the sample is of very large size, it is likely to appear statistically significant regardless of its actual statistical significance. That is, because the sample is large size, the results of the statistics may appear to be more exaggerated than they really are. On the other hand, small sample are likely to appear statistically significant only when effect size is large. If studies with publication bias are included in a meta-analysis, the results of the metaanalysis also be systematically biased. Therefore, it is necessary for meta-analyses to confirm whether publication bias is present in the included papers. The present study checked for potential publication bias using a funnel plot, Orwin's fail-safe N, and Egger's test of intercept. A funnel plot is a visual representation of studies' results with effect size on the x-axis, standard error on the y-axis, and a standard error graph to determine whether effect size tends to grow as sample size shrinks. Publication bias is considered present when the graph is bilaterally asymmetrical or the difference between arbitrarily added estimated effect sizes and original effect sizes are substantial, such that the shape of the graphs are symmetrical. Orwin's fail-safe N measures how many additional studies must be included to reduce the current effect size to a very small, meaningless effect size. Finally, Egger's test of intercept sets standardized effect size as a dependent variable and accuracy as an independent variable (Egger et al., 2001). If the intercept appears statistically significant and not equal to zero, publication bias is likely present.

Data analysis
All analyses in this study were conducted using Comprehensive Meta-Analysis (CMA) 3.0, and metaregression analyses to verify the effectiveness of moderator variables were performed using SPSS macro 23.0. Meta-analysis models can be either fixed-effects models, which assume that the collected studies have the same actual effect size, or random-effects models, which assume that the actual effect size varies depending on the study. Because a sufficient number of studies were available for the present meta-analysis, a random-effects model was adopted to confirm heterogeneity, as effect sizes were expected to differ by study according to each study's unique characteristics.
A meta-regression analysis was performed to identify heterogeneity of study-specific effect sizes. In a meta-regression analysis, individual studies are the units of analysis and the study characteristic variables input to an independent variable are moderator variables, as effect sizes depend on these characteristics. In the present study, each influence was examined according to the moderator variables of publication type (thesis, journal article, research report, or other), publication year, sample size and measurement tool (SSI, SIQ, or other). Table 1 shows the effect sizes for each study on the relationship between suicidal ideation and depression. The average effect size, 1.908, was statistically significant. The Q-value for determining the degree of heterogeneity among each study, 891.314, was also statistically significant. To account for the heterogeneity of effect sizes, moderator variables were included in meta-regression analysis.

Relationship between suicidal ideation and depression
The results of meta-regression analysis are shown in Table 2. Only publication year was found to be statistically significant. Specifically, more recently published studies were more likely to associate suicidal ideation with depression. No other study characteristics showed a significant impact.
To identify publication bias, average effect size was assumed to be 0.1 and the cut-off value for a meaningless effect size was set to 0.15 to calculate Orwin's fail-safe N. The results indicated that 489 more studies were needed to reduce the effect size below 0.15. As shown in Figure 2, the difference between the corrected and original effect sizes was very small (0.01) after adding studies to ensure the graph was symmetrical using the trim-and-fill method. Finally, Egger's test of intercept revealed that the intercept value was significant (p < .001), but when combined with the overall results publication bias was unlikely to be present. Table 3 shows the effect sizes for each study on the relationship between suicidal ideation and life stress. The average effect size, 0.959, was statistically significant. The Q-value for heterogeneity, 701.034, was also statistically significant. To account for the heterogeneity of effect sizes, moderator variables were included in meta-regression analysis.

Relationship between suicidal ideation and life stress
The results of meta-regression analysis are shown in Table 4. None of the moderator variables were found to be statistically significant.
The results for Orwin's fail-safe N showed that 104 more studies were needed to reduce the effect size below 0.15. As shown in Figure 3, the difference between the corrected and original effect sizes was very small (0.01) after implementing the trim-and-fill method. Finally, Egger's test of intercept revealed that the intercept value was significant (p < .001), but when combined with the overall results publication bias was unlikely to be present. Table 5 shows the effect sizes for each study on the relationship between suicidal ideation and critical communication by parents. The average effect size, 0.550, was smaller than the effect sizes for depression and life stress, but still statistically significant. The Q-value for heterogeneity among each study, 128.031, was also statistically significant. To account for the heterogeneity of effect sizes, moderator variables were included in meta-regression analysis.

Relationship between suicidal ideation and critical communication by parents
The results of meta-regression analysis are shown in Table 6. Only publication year was found to be statistically significant. Specifically, more recently published studies were more likely to associate suicidal ideation with critical communication by parents. No other moderator variables showed a significant impact.
The results for Orwin's fail-safe N showed that 102 more studies were needed to reduce the effect sizes below 0.15. As shown in Figure 4, the difference between the corrected and original effect sizes was very small (0.01) after implementing the trim-and-fill method. Finally, Egger's test of intercept revealed that the intercept value was significant (p < .001), but when combined with the overall results publication bias was unlikely present.    Table 7 shows the effect sizes for each study on the relationship between suicidal ideation and job stress. The average effect size, 0.704, was statistically significant. The Q-value for heterogeneity, 129.702, was also statistically significant. To account for the heterogeneity of effect sizes, moderator variables were included in meta-regression analysis.  The results of meta-regression analysis are shown in Table 8. Only publication year was found to be statistically significant. Specifically, more recently published studies were more likely to associate suicidal ideation with job stress. No other moderator variables showed a significant impact.

Relationship between suicidal ideation and job stress
The results for Orwin's fail-safe N showed that 72 more studies were needed to reduce the effect size below 0.15. As shown in Figure 5, the difference between the corrected and original effect sizes was very small (0.01) after implementing the trim-and-fill method. Finally, Egger's test of intercept revealed that the intercept value was significant (p < .001), but when combined with the overall results publication bias was unlikely to be present.

Discussion
The purpose of this study is to explore ways to reduce Korean university students' suicidal ideation by integrating prior studies on the influence of risk factors that affect their suicidal ideation. Metaanalysis was conducted to determine the effect sizes of these risk factors. The analysis results indicated that depression, life stress, parents' critical communication, and job stress have statistically significant effects on suicidal ideation. In particular, based on Cohen's (1998) criteria, depression showed a large effect size, and the other factors a medium effect size. This result is consistent with other studies that have identified depression as the strongest predictor of suicidal ideation (Chen et al., 2017;Garlow et al., 2008;Jeong, 2016;Pace & Zappulla, 2010;Tsujimoto et al., 2015). In other words, the more depressed university students felt, the more they thought of suicide. Therefore, university students need help in controlling their feelings of depression to ensure that they think less about suicide. Furthermore, they need access to specific information on how to receive counselling, treatment, etc. when considering suicide (Shtayermman et al., 2012). Critical communication by parents is identified as a variable that affects suicidal ideation of Korean university students. This indicates that parents still influence their adult children. This result is consistent with previous findings suggesting that children of parents who negatively communicate are likely to think of suicide (Choi, 2012;Kwon et al., 2016;Shtayermman et al., 2012). How parents communicate with their children significantly impacts their children's psychological states, as they directly or indirectly express emotions, thoughts, beliefs, etc. (Satir et al., 1991). Therefore, the more negative messages parents send to their children, the more their children's psychology is adversely affected, the more they negatively perceive themselves, and the more their inner strength weakens, rendering them vulnerable to suicidal ideation upon encountering difficult life experiences (Gibb et al., 2001). The children raised by parents who maintained close relationship with their children and communicated positively with their children are more likely to have the stable and  healthy psychological states by expressing their emotion, thoughts, beliefs, etc. Thus, parents must positively communicate with their children to reduce Korean university students' suicidal ideation and to empower them to develop self-protection skills. Satir et al. (1991) suggested that internal growth of self-esteem is crucial in addressing the negative effects parents' critical communication might have had on their children. In other words, university students must develop inner strength to judge and resolve the negative words or emotions received from their parents. Therefore, appropriate and easily accessible educational programs and counselling are needed to help university students achieve inner growth.
Stress is another factor that affects Korean university students' suicidal ideation. High levels of stress lead to suicidal ideation. In particular, Korean university students experience much stress while preparing for employment, leading to negative emotions such as anxiety, fear, and lethargy, as well as low self-esteem (Shtayermman et al., 2012). Previous studies have indicated that life and job search-related stress are major factors that lead university students to consider suicide (Seo & Lee, 2014;Van Orden et al., 2008;Zhang et al., 2012). High levels of stress lead to lethargy, diminished selfworth, depression, etc., as life vitality and meaning decrease, which is likely to lead to suicidal ideation (Jin et al., 2015). Moreover, high levels of job stress cause despair, frustration, thoughts about the futility of the future, etc., as one's expectations from life decrease, which is also likely to lead to suicidal ideation (Seo & Lee, 2014). Therefore, Korean university students must be educated on managing life and job search-related stress.
Studies on the relationship among suicidal ideation, depression, parents' critical communication, and job stress reveal that the effect size of these variables differ by year; recent studies demonstrated that depression, parents' critical communication, and job stress have an increasingly greater impact on suicidal ideation. Thus, reducing the influence of these factors is crucial for preventing Korean students' suicidal ideation.
Based on the present results the following measures are proposed to prevent suicidal ideation in Korean university students. First, counselling and treatment must reduce depression in university students by detecting it early. To achieve this, families, schools, and the local community must recognize depression as an important issue and help university students address it before it becomes serious. Students must also be taught how to manage and relieve their stress. Garlow et al. (2008) indicated that university students often miss the critical time for dealing with their depression as they do not have access to good consultation or interventions. Therefore, families and universities should cooperate to reduce or treat depression and suicidal ideation by providing quality mental health services for university students. Furthermore, special lectures or programs on managing depression and stress at university should be run to help students self-manage their emotions and stress. Universities should include mental health content in liberal arts lectures to promote students' understanding of mental health. Finally, communities should raise awareness of depression by alerting the public to its risks and prevention methods.
Second, a national system must be established to organically link homes, schools, communities, and mental health services. This study's results show that factors in various areas have a more than moderate effect on suicidal ideation in university students. Thus, an integrated response must be provided. Currently, the various systems for teenagers (elementary, middle, and high school students) are consolidated, whereas those for university students are independent. The various systems for teenagers are also well integrated, allowing active implementation of suicidal ideation prevention education and counselling programs. In contrast, suicide prevention education and counselling programs for university students are highly insufficient, and those that are implemented are ineffective because the systems are disconnected. Therefore, students' access to mental health services such as suicidal ideation prevention education and counselling programs should be increased by integrating home, school and community.
This study makes several meaningful contributions to the literature. First, it systematically organizes and aggregates effect sizes for the results of existing studies on the link between university students ' suicidal ideation and various risk factors. Second, by comparing the effect sizes among risk factors, it provides substantial data for developing and implementing future suicide prevention education programs for university students. Third, unlike many previous meta-analytic studies, this study used a random-effects model to estimate average effect sizes, examined publication bias, and conducted meta-regression analysis of heterogeneity of effect sizes. These research methodologies could benefit future researchers seeking to apply a meta-analytical approach.

Disclosure statement
No potential conflict of interest was reported by the author(s).