Intention to Act: Predicting Bystander Intervention in Violent Situations in South Korea

Abstract

The purpose of this study is to explore the predictors of bystanders’ intentions to intervene and types of intended intervention in domestic violence (DV) and sexual violence (SV) situations in South Korea. Using nationally representative data from the 2016 National Fact-Finding Survey on Gender Equality (N = 7,399) and logistic regression-based models, this study focused on two key predictors, bystander gender and attitudes about gender equality, while controlling for age, educational level, and employment status. Our findings indicated that males and females were equally likely to state intentions to intervene in both DV and SV situations. While both males and females were more likely to state that they would intervene indirectly rather than directly, females were even less likely to intervene directly than were males for both situations. Attitudes indicative of gender equality further increased the intention to intervene for DV and SV situations in multivariate models. Gender equality attitudes, however, decreased the intentions to intervene directly in DV but not SV situations. Limitations and implications are discussed.

Domestic violence (DV) and sexual violence (SV) are serious and prevalent social problems. Although prevalence estimates vary widely, it is generally suggested that over one-third of women have experienced DV or SV by an intimate partner in their lifetime (World Health Organization, 2017). Several studies suggest, however, that the presence of bystanders can reduce the incidence of different forms of interpersonal violence, including DV and SV (Banyard et al., 2004; Pradipto et al., 2016; Safta et al., 2010; Zain, 2012). These studies also indicate that a bystander’s intentions to intervene are frequently influenced by individual and situational characteristics, including the gender and age of the actors, bystander personality, the number of bystanders, and the type of offense (Banyard et al., 2004; Banyard, 2008; Cismaru et al., 2010; Weitzman et al., 2020).

Many of these studies have been conducted in the United States and other Western countries, while less attention has been paid to Eastern countries such as South Korea, where reports of DV to the Korean National Police Agency (2021) number more than 200,000 incidents each year. Although some studies suggest that findings about bystanders and their intention to intervene in Western countries likely apply to Eastern countries (Emery et al., 2017; La Ferle et al., 2019; Pradipto et al., 2016; Zain, 2012), the unique factors associated with Korean culture and history might be explored more carefully as cultural differences may differentially affect a Korean bystander’s response to observing violent situations. For example, the legacy of Confucianism, one of the traditional Korean values that remains in many parts of Korean society, is patriarchal and favors masculine over feminine traits (Kim et al., 2010). The emphasis on patriarchy and traditional gender norms has the potential to make violent behavior against females more acceptable in South Korea than in countries with non-Confucian aspects because many Confucian cultural elements incorporate and empower violence-tolerant attitudes toward women (Han, 2017).

In contrast, another traditional aspect of Korean society is a form of collectivism (Yoon, 2010) that may increase people’s willingness to intervene in violent situations. If most Koreans consider society to be one big family, it is possible that even traditional Korean families consider it permissible to accept the interference of a neighbor, or even a stranger, in a family issue (Kim & Sung, 2000; Morgan & Rindfuss, 1984).

Background

The concept of bystander intervention was originally introduced by Darley and Latane (1968) after the 1964 murder of Kitty Genovese in the middle of a residential area in New York City. According to some reports, 38 people saw or heard the attack but failed to come to her aid. While many of the event’s details have been questioned over the years, what remains certain is that this event initiated a sizeable amount of research into the behavior of bystanders (Amar et al., 2014; Shotland & Goodstein, 1984; Song & Oh, 2018), who are generally defined as third parties who observe dangerous behavior and have the potential to escalate or deescalate a situation by encouraging or discouraging a perpetrator and/or to do nothing (Hart & Miethe, 2008; McMahon & Banyard, 2012).

The Bystander’s Intention to Intervene

Since the time of Darley and Latane’s early research, contemporary scholars have identified several factors that affect bystander behavior including the following: whether a bystander has previously witnessed others offering help in violent situations; how bystanders interpret the event; and the interactions between the bystander, the victim, and the offender (Banyard, 2008; Shotland & Straw, 1976). Researchers have also found that a bystander’s intentions to intervene are influenced by the gender of the bystander, victim and offender as well as other bystander characteristics such as age, personality, level of self-esteem, and religious faith (Banyard, 2008).

Among bystander characteristics, gender has been examined most frequently, although often with inconsistent results (Banyard et al., 2004; Banyard, 2008; Buckler et al., 2019; Cismaru et al., 2010; Griffith et al., 2006; Nicksa, 2014; Pozzoli et al., 2012). For example, while one meta-analysis has found that males are more willing to intervene than females (Eagly & Crowley, 1986), another study has argued that females have less violence-tolerant attitudes, which may lead females to intervene more often (Griffith et al., 2006). A recent systematic review (Mainwaring et al., 2023) has observed a pattern where most prior research (e.g., Franklin et al., 2020; Savage et al., 2017) has found that females are more willing to intervene than males in SV contexts. This review also notes, however, that other research suggests that males and females are willing to intervene at a similar rate (e.g., Banyard et al., 2020; Galdi et al., 2017), although the individuals’ intervention type depends on their gender (Chabot et al., 2009; Mainwaring et al., 2023), a matter we address more fully in the next section.

Closely related to gender are matters associated with a bystander’s attitude toward gender equality and the subsequent influence of this attitude on the bystander’s behavior. In a recent study, Heine (2020) has found that participants with higher levels of hostile sexism are more likely to approve of aggression toward victims and less likely to state an intention to intervene. Similarly, Jin (2017) discovered that bystander sexism and the victim’s sexuality affect bystander behavior. Jin’s findings were complicated, however, by the nuance that bystander sexism affects intentions to directly intervene, although it does not have measurable influences on intentions to indirectly intervene.

Several studies have identified bystander age as another important predictor of bystander intention to intervene (Griffith et al., 2006; Mainwaring et al., 2023; Weitzman et al., 2020), although these results have been inconsistent. For example, some researchers have found that older participants are more likely to believe that intervention in sexual assault situations is beneficial in ways that outweigh the potential costs (Diamond-Welch et al., 2016). In contrast, some researchers have found that younger participants are more likely to intervene than older counterparts in intimate partner violence situations (Beeble et al., 2008). Others have noted complexities where the effect of age on intervention varies across different cultures. For instance, Griffith et al., (2006) compared American and Trinidadian bystander behavior in DV situations and found that bystander age is the only variable that accounts for intervention differences between these cultures.

While the relationship between education and bystander behavior has not been thoroughly explored in the DV and SV context, there are studies suggesting that education is associated with liberal values and norms, which presumably should lead to less tolerance for attitudes that support violence against women (Boyle et al., 2009) and subsequently support a greater willingness of bystanders to intervene in these situations. When examining the empirical research on this matter, however, one study suggests that education does not affect bystander behavior in sexual assault and intimate partner violence situations (Weitzman et al., 2020), while another study indicates that a 1-year increase in formal education predicts a 7.2% increased odds of bystander intervention in child neglect situations (Fledderjohann & Johnson, 2012).

In terms of the effect of employment status on intervention, one recent study has examined the relationship between bystander employment status and intervention behavior in sexual assault and intimate partner violence situations (Weitzman et al., 2020). This study found that self-employed participants have 192% greater odds of intervening than those who are unemployed.

The contradictions in past research on this matter may be due to the type of offense observed as research that comes to different conclusions has often examined different offense types. Only a few studies have examined different offense types within the same study. Consequently, more precise conclusions about how offense type may alter bystander behavior have been tentative. In addition, the vast majority of this research has been conducted only on convenience samples involving university students, and often only in Western countries. These facts have left gaps in our knowledge about how other populations might behave (Bennett et al., 2014; Burn, 2009; Coker et al., 2011; Weitzman et al., 2020).

Bystander intervention research using samples more representative of the general public in the study of DV and SV, however, has recently accelerated as scholars have come to realize the potential of bystanders in preventing DV and SV (Banyard et al., 2004; Banyard, 2008; Barna & Barna, 2014; Chabot et al., 2009; Safta et al., 2010; Weitzman et al., 2020). For example, contemporary DV prevention approaches targeting victim or offender behavior may not necessarily be the most effective approaches (Pradipto et al., 2016). For this reason, some authors have argued that focusing on enhancing bystander-oriented intervention may be more effective in decreasing DV (Pradipto et al., 2006).

Types of Bystander Intervention

Scholars have often classified the tactics bystanders use to intervene in violent situations into either direct or indirect methods (Chabot et al., 2018; Weitzman et al., 2020). Direct interventions include physically interfering and/or talking to the victim or offender. Indirect interventions include causing distractions, asking for help from other bystanders and/or calling the police. One detail that readers should recognize, however, is that the prior literature has not always differentiated direct from indirect intervention methods. For example, Mainwaring and colleagues (2023) contrast “direct” interventions, defined as behaviors enacted to stop an incident presently occurring, with “primary,” “secondary,” and “tertiary” interventions designed to alter the attitudes that encourage SV occurrence, to address a risky situation or to assist victims in the aftermath of an event (also see Powell, 2014). These definitions contrast with the more common and original Latane and Darley (1970) definitions that describe “indirect” methods as those involving getting help from someone who knows the parties, checking with the victim later and notifying the police; “direct” methods involve directly stopping the incident or taking the victim away from the situation (also see Chabot et al., 2018).

Many factors affect bystanders’ decisions about how to intervene, including characteristics of the bystander and situational factors. For example, as mentioned earlier, prior research suggests gender differences as a major factor affecting the type of intervention where males may be more likely than females to intervene directly through physical means or by intimidating the offender, while females may be more likely than males to intervene indirectly by asking others for help or calling the police (Chabot et al., 2009).

Although prior research has identified gender differences in intervention type, the existing literature has not yet definitively explained why gender differences exist. It is certainly possible that even if females intervene to the same degree as males, they may choose less direct forms of intervention due to a heightened concern for their own safety, given a typically lower degree of physical strength and lesser ability to defend themselves during direct confrontation with an offender.

In addition, research has not fully examined the mediating mechanism(s) for observed gender differences and how these differences may tie into related matters such as attitudes regarding gender equality or other characteristics potentially associated with gender, such as employment and education. It is well-established, for example, that females are more likely to hold attitudes in favor of greater gender equality and that females are less likely to be employed, particularly in more traditional cultures (Eun, 2007; World Bank Group, 2022). These facts make it possible that the effect of gender on intervention type may work through an intervening attitudinal mechanism, while also being potentially confounded by matters associated with traditional gender roles, such as employment outside the home. For example, females may hold more pro-gender equality attitudes than males, which, in turn, causes pro-equality females to intervene more often or in a different manner.

Current Study

Our research team had access to Korean survey data that the first author translated to English. These data permitted us to address four closely related research questions intended to replicate certain aspects of the existing literature and do so in a non-Western context. Because the dataset also had information about each respondent’s attitudes about gender-equality as well as common demographic variables regarding gender, age, education, and employment, we were able to control for these measures in our models.

We first wanted to know about the main factors that might predict whether Korean survey participants would state an intention to intervene in: (a) DV and (b) SV situations. Next, among those who would state an intention to intervene, we wanted to know about the factors that might predict whether these participants would state an intention to intervene directly or indirectly in: (a) DV and (b) SV situations. Because attitudes regarding gender equality are known correlates of gender, as are some of our demographic variables, we ran preliminary bivariate models to first determine whether any association between gender and any of our outcome measures existed. If so, we intended to examine the change in the regression coefficients and odds ratios (ORs) to determine whether any associations between gender and intervention intentions or intervention type might be explained by gender’s associations with attitudes about gender equality and the control variables in our multivariate models.

Based on the literature, we hypothesized that even if males and females would state equal intentions to intervene, males would be more likely to intervene directly, while females would be more likely to intervene in an indirect manner as suggested by some of the past literature. Although we anticipated that attitudes about gender equality would be associated with gender, the lack of prior research regarding the association between these attitudes and intervention intentions or intervention type did not permit us to hypothesize any direction for these effects. We did anticipate, however, that attitudes regarding gender equality may mediate relationships between gender and our dependent variables.

Method

Sample

The data for this study come from the South Korean National Fact-Finding Survey on Gender Equality (NFFSGE), conducted in 2016 by the Ministry of Gender Equality and Family. The South Korean Minister’s office selected participants using a stratified cluster sampling method, where the sampling frame was based on the 2010 Population and Housing Census and the 2010 to 2015 Newly-Constructed Apartments List. South Korea consists of nine provinces, one special city (the capital, Seoul), one special self-governing city (Sejong), and six metropolitan cities. Each of these nine provinces and Sejong City were further stratified into two smaller regions for a total of 27 regional clusters. Households from the 27 regional clusters were selected proportional to the population size of that region. The households in each region were chosen through random selection and all family members in each household who were born in 1997 and earlier were asked to complete the survey. We weighted our analyses using the weights available in the dataset to account for the complex sampling structure to the degree possible.

The response rate for included households was reported to be 100%, while the response rate for individuals within households was 96.6%. A total of 7,660 people were contacted and all but 3.4% agreed to participate. The final dataset had 7,399 adults from 4,004 households. Although a response rate this high might be considered improbable in many countries, participation was more likely to occur because the survey was administered by government employees and specified by Korean law through the passage of the Framework Act on Gender Equality and Statistics.

Measures

Dependent variables

We used two survey questions asked under the “Human Rights and Safety” section of the questionnaire to construct four dependent variables. The two questions were: “How would you react if you witnessed domestic violence in your neighbourhood?” and “How would you react if you witnessed SV?” The survey presented hypothetical situations, asking participants about their intention to intervene and the type of intervention they would attempt if they witnessed the described situation.

The first set of dependent variables was dichotomous variables measuring participants’ intention to intervene in DV and SV (one dependent variable for each type of violence). The original survey questions had four categories of responses: (1) directly intervene; (2) call the police; (3) ask for help; and (4) stay out of it. For both DV and SV, a dichotomous variable was created by combining the first three categories as “intent to intervene” (coded as 1) and the last category as “no intent to intervene” (coded as 0). For the subset of participants who indicated an intention to intervene, the second set of dichotomous dependent variables for both DV and SV was created. These variables indicated whether the respondents supported direct or indirect intervention. Of the participants who originally chose one of three intervention responses, those who said they would directly intervene were recoded as “directly intervene” (1) and those who said they would call the police or ask for help were recoded as “indirectly intervene” (0).

Independent variables

The independent variables for all research questions were demographic characteristics of the participants: gender, gender equality attitudes, age, education level, and employment status. As outlined previously, these variables were identified in the literature as potential predictors of bystander intervention and were measures available in the NFFSGE data. They are described in detail below.

Gender

Bystander gender has been a robust correlate shown to matter in research about bystander decisions to intervene, although the research has differed with regard to how gender is related to bystander intervention intention and type of intervention. In this study, bystander gender was measured by asking participants about their biological sex and coded as either male (1) or female (0). Table 1 indicates that the number of male participants was 3,684 (49.80%) while the number of female participants was 3,715 (also noted in Table 3).

Table 1.

Frequencies for Independent Variables^*.

Variables	Frequency	Percent	Mean (SD)
Male	3,684	49.80
60 and older	3204	43.31
High school graduate+	5,723	77.34
Employed	4,792	64.76
Gender equality attitude			2.74 (0.38)
Total	7,399	100.00

^*Weighted frequencies.

Table 3.

Frequencies for Dependent Variables^*.

	Male			Female
Dependent Variables	n	Frequency/Yes	Percent	n	Frequency/Yes	Percent
Intention to intervene in DV	3,684	3,172	86.10	3,715	3,172	85.39
Direct intervention in DV	3,172	301	9.50	3,172	139	4.39
Intention to intervene in SV	3,684	3,363	91.27	3,715	3,373	90.80
Direct intervention in SV	3,363	678	20.15	3,373	163	4.82

Note. DV = domestic violence; SV = sexual violence.

^*Weighted frequencies.

Gender equality index

There were 13 questions that asked participants about their attitudes toward gender equality and gender roles. Participants answered each question on a four-point Likert response in categories from 1 (strongly disagree) to 4 (strongly agree). Nine out of 13 questions were recoded so that a higher numeric value consistently indicated a higher level of gender equality. The sum across all 13 questions ranged from 13 (low gender equality) to 52 (high gender equality). Table 2 indicates the average of this sum so that the final values ranged from 1 (low gender equality) to 4 (high gender equality) with a mean of 2.74 (SD = 0.38). The averaging does not affect the regression coefficients in the model but permits an easier interpretation of the mean across the participants. Although no psychometric analyses were conducted to assess criterion-related forms of validity for this index by the Korean scholars who developed these questions, we have been informed by these scholars that the questions were designed to be culturally appropriate (Korean Ministry of Gender Equality and Family, personal communication with the first author, February 6, 2024). The questions are also quite like many used in other surveys of this type (e.g., Jha et al., 2020; Research New Zealand, 2023) and appear to have considerable face validity (e.g., “It is the husband’s responsibility to make the family living”). We computed the internal reliability of the index with Cronbach’s alpha as .80.

Table 2.

Descriptive Statistics for Attitude Toward Gender Equality and Its Original Items.

Items	Mean	SD
Gender equality index	2.74	0.38
If wife’s income exceeds husband’s, husband feels discouraged	2.63	0.78
It is husband’s responsibility to make family living	2.63	0.73
Even if wife makes a living, important family decisions should be decided by her husband	2.71	0.72
It is a shame to be a house husband	2.82	0.74
Women should prioritize taking care of young children over her work	2.48	0.74
It is unpleasant if men work for women	2.81	0.70
Men should not disclose their weaknesses	2.67	0.74
Men should lead a date	2.79	0.67
It is important for women to be financially independent	2.93	0.66
Men should be able to take care of his children alone without getting help from others	2.94	0.62
Men should try to be in careers that are mainly dominated by women	2.55	0.66
Women should try to be in careers that are mainly dominated by men	2.56	0.65
It is reasonable to dismiss women when staff reduction is required for managerial reasons	3.13	0.71

Age

Although some studies have examined the effects of bystander age on intervention intention or type of intervention, research samples have often been limited to university students responding to questions about SV. Because university student samples are limited in age, they might not be reflective of general trends across society. The NFFSGE was not limited in this way. Age was originally reported in the five categories. Preliminary analyses, however, revealed that age was not linearly related to the dependent variables and that participants who were 60 years old and older substantially differed from the other age groups, while the other age groups did not substantially differ from each other. For this reason, we converted the five original categories into a dichotomous variable by coding adults 60 and older as one (1) and adults younger than 60 as zero (0). Table 1 indicates that 3,204 participants were 60 years old and older (43.31%).

Education level

The publicly available version of the NFFSGE data provided educational information in only three categories. Because preliminary analyses indicated that intervention differences were affected primarily by whether the participant graduated high school or not, we coded education as either high school graduation and beyond (1) or lower than a high school education (0). Of the total 7,399 participants, 5,723 had completed a high school or higher education (77.34%).

Employment status

Participants were asked whether they participated in paid work for more than 1 hour and/or worked for more than 18 hours on a family-run farm or business (without pay) in the week previous to taking the survey. This measure was included as a dichotomous variable coded “Yes/employed” as 1 and “No/unemployed” as 0. The frequency of those who were employed was 4,792 (64.76%).

Data Analysis

All dependent variables were dichotomous, and for this reason, we used logistic regression in all models. We began with a series of bivariate models used primarily to examine the bivariate relationships between gender and the different outcomes of interest, but also to examine the bivariate relationships between our other independent variables and these outcomes.

We next turned to our multivariate models. The two main multivariate models were fit to the full sample and related to whether the participants stated an intention to intervene or not in either: (a) DV and (b) SV. Finally, two secondary models restricted the sample to a subset of participants who had previously stated that they would intervene and examined whether this subset of participants would intervene directly or indirectly in either: (a) DV and (b) SV.

Results

Descriptive Statistics

Table 3 indicates that after our weighting adjustments due to the sampling procedure, there were 3,684 males and 3,715 females included in our analyses for a total of 7,399 participants. Among the participants, an equal number of male and females (3,172; about 86% for each gender) expressed an intention to intervene in DV. Among the subset of those who intended to intervene in DV, 301 males (9.50%) and 139 females (4.39%) indicated an intent to intervene directly. In addition, 3,363 males (91.27%) and 3,373 females (90.80%) expressed their intention to intervene in SV and among those who intended to intervene in SV, 678 males (20.15%) and 163 females (4.82%) opted for direct intervention.

Bivariate Statistics

Table 4 displays the results of five bivariate models that examined the relationships between independent variables and the intention to intervene in DV and SV situations. For both DV and SV, all independent variables except for gender showed statistically significant relationships at the p < .001 level with the intention to intervene. Gender was not found to be statistically significant (p > .05) for either DV or SV, which indicates that we had little to no evidence that males in our population of interest would state an intention to intervene in either scenario to a degree greater than females.

Table 4.

Bivariate Models of Predictors of Intention to Intervene in Domestic and Sexual Violence (n = 7,399).

	Domestic Violence		Sexual Violence
Variables	b (SE)	OR	b (SE)	OR
Male	0.06 (0.07)	1.06	0.06 (0.08)	1.06
60 years or older	−0.63 (0.07)***	0.53	−0.43 (0.08)***	0.65
High school graduate+	0.93 (0.07)***	2.52	0.85 (0.09)***	2.34
Employed	0.27 (0.07)***	1.31	0.23 (0.08)***	1.25
Gender equality attitude	0.97 (0.09)***	2.64	0.97 (0.11)***	2.65

Note. In these bivariate models, the dependent variable is intention to intervene (vs. no intervention). OR = odds ratio.

^***p < .001.

Table 5 shows five similar bivariate models illustrating the relationships between each independent variable and the type of intervention in DV and SV situations among the subset of individuals who indicated an intention to intervene in the earlier models. In the case of DV, all variables were statistically significant except for employment status. For SV, all variables were statistically significant except for age. Unlike the models in Table 4, gender was statistically significant (p < .001) and indicated that males were much more likely than females to state an intention to intervene directly.

Table 5.

Bivariate Models of Predictors of Direct Intervention in Domestic and Sexual Violence.

	Domestic Violence (n = 6,344)		Sexual Violence (n = 6,736)
Variables	b (SE)	OR	b (SE)	OR
Male	0.83 (0.11)***	2.28	1.61 (0.09)***	4.98
60 years or older	1.25 (0.11)***	3.49	−0.08 (0.08)	0.92
High school graduate+	−1.13 (0.10)***	0.32	0.20 (0.09)*	1.22
Employed	−0.05 (0.10)	0.95	0.63 (0.09)***	1.87
Gender equality attitude	−1.43 (0.14)***	0.24	−0.30 (0.10)***	0.74

Note. In these bivariate models, the dependent variable is direct intervention (vs. indirect intervention) among the subset of participants who stated intentions to intervene in any way. OR = odds ratio.

^*p < .05. ***p < .001.

Multivariate Model 1A—Intention to Intervene in DV

The first multivariate logistic regression model examined all independent variables simultaneously to determine their effects on bystanders’ intention to intervene in DV while accounting for all other variables in the model (Model 1A, Table 6). Like the corresponding bivariate model, gender was not significantly related to intervention intentions (p > .05). The very small bivariate OR for gender in Table 4 dropped even further for the multivariate model in Table 6 (OR bivariate = 1.06; OR multivariate = 1.01). Similarly, the effects for the other variables dropped in Table 6 from their bivariate values in Table 4, with the exception of age, although they remained in the same direction as in the preliminary bivariate models. These facts together with auxiliary analyses not shown, indicate a moderate amount of intercorrelation across the independent variables in the models in expected ways. For example, in analyses not shown, high school graduates are more likely to be employed while those over the age of 60 are more likely to be unemployed (potentially retired). In addition, attitudes regarding gender equality are modestly associated with age, education, employment, and gender. Although the degree of intercorrelation was not enough to exceed standard indicators of multicollinearity, our results still need to be interpreted with this understanding.

Table 6.

Multivariate Predictors of Intention to Intervene in Domestic and Sexual Violence (n = 7,399).

	Model 1A: Domestic Violence			Model 2A: Sexual Violence
Variables	b (SE)	β+	OR	b (SE)	β+	OR
Male	0.01 (0.07)	.01	1.01	0.03 (0.09)	.03	1.03
60 years or older	−0.10 (0.09)	−.10	0.91	0.20 (0.11)	.19	1.22
High school graduate+	0.68 (0.09)***	.57***	1.98	0.76 (0.11)***	.63***	2.13
Employed	0.03 (0.07)	.03	1.03	0.02 (0.09)	.02	1.03
Gender equality attitude	0.64 (0.10)***	.48***	1.90	0.75 (0.12)***	.57***	2.13
Constant	−0.39 (0.30)	−.39	0.68	−0.37 (0.36)	−.37	0.69
Nagelkerke R2	.05			.04

Note. In Model 1A and 2A, the dependent variable is intention to intervene (vs. no intervention). β⁺ indicates semistandardized coefficient as per Gelman, 2008. OR = odds ratio.

^***p < .001.

For example, it is meaningful to discuss the effect of attitudes about gender equality on intervention intentions controlling for the other variables in the model (increasing one unit on the gender-equality score results in a 90% greater odds of intervening in a DV event after accounting for the other variables in the model). It is unrealistic, however, to think that the OR for the effect of high school graduation on intentions to intervene in DV will remain exactly the 1.98 listed in Table 6 for alternative model specifications, since this effect will vary depending on which of the other moderately-correlated variables are included or excluded from the analysis.

Due to a desire to determine the relative importance of the predictor variables in our multivariate models and the fact that the predictors were measured on different scales where gender equality was on a scale from 1 to 4 while the remainder of the variables were dichotomies, we followed a procedure recommended by Gelman (2008) to create semistandardized logistic regression coefficients. This approach divides each of the independent variables’ values by twice the variable’s standard deviation. In this way, the resulting coefficients are more directly comparable, particularly when binary and interval-level predictors are present. Although this method does not fully compensate for the different metrics of our predictor variables (no method could possibly do this completely for conceptual reasons), it potentially provides a better conceptual basis on which to compare the relative magnitude of gender and gender-equality attitudes with the other predictor variables. We placed the semistandardized coefficients under a column labeled β⁺ in the multivariate models in Tables 6 and 7. By examining this column’s coefficients, we can confirm that high school graduation is the strongest predictor of intervention intentions for both DV and SV while gender equality attitudes are a close second.

Table 7.

Multivariate Predictors of Direct Intervention in Domestic and Sexual Violence.

	Model 1B: Domestic Violence (n = 6,344)			Model 2B: Sexual Violence (n = 6,736)
Variables	b (SE)	β+	OR	b (SE)	β+	OR
Male	0.94 (0.11)***	.94***	2.55	1.55 (0.09)***	1.55***	4.71
60 years or older	0.86 (0.13)***	.85***	2.36	−0.02 (0.09)	−.02	0.98
High school graduate+	−0.70 (0.12)***	−.59***	0.50	−0.09 (0.11)	−.08	0.91
Employed	0.08 (0.11)	.07	1.08	0.29 (0.09)**	.28**	1.34
Gender equality attitude	−0.63 (0.16)***	−.47***	0.54	−0.04 (0.11)	−.03	0.96
Constant	−1.49 (0.46)***	−1.49**	0.23	−2.97 (0.34)***	−2.97***	0.05
Nagelkerke R2	.11			.11

Note. In Model 1B and 2B, the dependent variable is direct intervention (vs. indirect intervention) among the subset of participants who stated intentions to intervene in any way. β⁺ indicates semistandardized coefficient as per Gelman, 2008. OR = odds ratio.

^**p < .01. ***p < .001.

Multivariate Model 2A—Intention to Intervene in SV

In the second multivariate logistic regression model, all independent variables were examined to ascertain their collective effects on bystanders’ intention to intervene in SV (Model 2A, Table 6). Of most interest, gender was once again not a statistically significant predictor of SV intervention, although attitudes about gender equality remained a statistically significant predictor (p < .001) even after accounting for the other variables in model. As in Model 1A, the effect of gender equality attitudes on SV intervention dropped moderately as compared to the bivariate model (OR bivariate = 2.65; OR multivariate = 2.13) as expected. As mentioned above, the results for the control variables should be interpreted with care given the nature of their mild intercorrelations. Similar to Model 1A, when examining the semistandardized coefficients, we can see that the effect of high school graduation had the strongest effect on intervention intentions with gender equality attitudes as a close second.

Mediation and Explained Variance for Models 1A and 2A

A pseudo-R² statistic, the Nagelkerke R², indicated that the combined ability of the five independent and control variables in models 1A and 2A was quite modest (only .05–.04 for each model). Of greater interest, however, was the relative ability of the model’s variables to correctly predict intervention intentions as mentioned in prior sections. While we had intended to determine whether attitudes about gender equality may mediate a relationship between gender and intentions to intervene, there was no statistically significant relationship of this type to mediate.

Multivariate Model 1B—Intention for Direct Versus Indirect Intervention in DV

Table 7 Model 1B included only the subset of participants who reported an intention to intervene in DV (n = 6,344). The results indicated that all independent variables except for employment status were statistically significant at the traditional p < .05 level. Male participants had two and a half times the odds of direct intervention in DV than females when all other variables were held constant (OR = 2.55, p < .001). Participants 60 years old and older had an OR of 2.36 (p < .001), indicating that the odds of direct intervention in DV situations was more than twice as high for participants who were older than 60 than those who were younger than 60 when all other variables were held constant. Participants who finished high school (and beyond) had an OR of .50 (p < .001), which indicates that the odds of direct intervention in DV was 50% less for participants who completed high school compared to those who did not complete high school when all other variables in the model were held constant. The OR was 0.54 (p < .001) for attitudes toward gender equality, indicating that a one-unit increase on the gender equality scale predicted almost half the odds of direct (vs. indirect) intervention in DV when all other variables in the model were held constant. When examining the semistandardized coefficients, we can see that gender is the strongest predictor of direct intervention followed by age, education, and gender equality attitudes.

Multivariate Model 2B—Intention for Direct Versus Indirect Intervention in SV

As in model 1B, model 2B in Table 7 again only included the subset of participants who stated an intention to intervene in SV (n = 6,736). Only two variables, gender and employment status, were statistically significant for this model at the traditional p < .05 level. The OR for gender was 4.71 (p < .001), indicating that male participants had almost five times the odds of direct intervention in SV than females when all other variables were held constant. The OR for employment status was 1.34 (p < .01), indicating that the odds of direct intervention for SV among those who were employed was 34% more than those who were unemployed when all other variables were held constant. When examining the semistandardized coefficients, we see that gender is a stronger predictor than employment status.

Mediation and Explained Variance for Models 1B and 2B

The Nagelkerke R² values for the models in Table 7 were slightly higher than those of the previous table (.11 in both models) while remaining relatively modest. As above, however, we remained more interested in whether we might find statistically significant gender differences in the stated type of intervention and whether gender equality attitudes might mediate any relationship between gender and these intervention types.

In contrast to our models predicting intervention versus no intervention, our models addressing the type of intervention do indicate a statistically significant effect for gender for both DV and SV. While a substantial decrease in the ORs for gender between the multivariate model and bivariate model would indicate the presence of a potential mediation effect involving one or more of the other variables, we found no such meaningful change. The OR for gender in the bivariate DV model was 2.28 while the corresponding OR in the multivariate DV model was 2.55, a modest increase rather than decrease. Similarly, the gender OR for the bivariate SV model was 4.98 while in the multivariate model it dropped only to 4.71. For these reasons, we can conclude that our models were unable to detect the presence of a meaningful mediation mechanism that might explain why the gender and intervention type relationship exists.

Discussion

To briefly summarize our more important findings: (1) most people in our sample said that they would intervene in some way for both the DV and SV situations; (2) among the subset of those who said they would intervene, most said that their intervention would be in an indirect way; (3) the stated tendency to intervene indirectly rather than directly applies to both males and females; (4) males who say that they would intervene, however, are more likely than are females to say that their intervention would be direct, but since the percentage of males stating a direct intervention intention is less than 50% of the males, it is still a minority of the males stating that they would intervene directly; and (5) the female minority who stated that they would intervene directly is even a smaller minority than the male minority who would intervene directly.

An important finding in our study that has often been overlooked in the literature is the relatively high level of intervention intentions among study participants. Our preliminary univariate statistics indicated that about 86% of our participants stated that they would intervene in both DV and SV situations in some way. This finding is at odds with Darley and Latane’s original claim of low rates of bystander intervention. Also contrary to some of the prior literature, we found no meaningful evidence of gender differences in the high rate of intervention intentions.

The actual degree to which research participants state an intention to intervene directly versus indirectly has often been overlooked in the prior research that has instead examined only the gender differences in intervention type. The univariate statistics in Table 3 indicate that among the subset of those who state that they would intervene in any way for either type of violence, both males and females were unlikely to state an intention to intervene directly. Instead, both male and females stated that they would either “call the police” or “ask for help.” Males stated an intention for a direct intervention method in only 9.50% of the DV cases and 20.15% of the SV cases, while females preferred a direct intervention method only 4.39% and 4.82% of the time in DV and SV, respectively. Furthermore, this pattern held in our multivariate model that controlled for age, education, employment status, and gender equality attitudes where we found an OR for the gender variable and direct intervention of 2.55 for DV and 4.71 for SV. This low preference for direct intervention among both genders can be masked in models that only examine gender differences without any emphasis in their underlying frequencies.

Our findings also demonstrate that many of our other variables describing bystander characteristics have substantial effects on both the intention to intervene and the type of intervention for both DV and SV. Similar patterns regarding the direction of the effects for these variables are found for both the DV and SV situations. Although there are similarities in the magnitude of the effects of these variables for DV versus SV (and corresponding similarities in whether these effects can be considered statistically significant at the traditional p < .05 level), there are several important differences across the two offense types that we discuss in more detail below.

Intention to Intervene in DV and SV

As noted above, we found no effect for gender in our models regarding an intention to intervene in either DV or SV situations. Our findings about an equal intention to intervene for males and females are consistent with the more recent work of Chabot and colleagues (2009), but do not support the findings of earlier scholarship that has sometimes shown males to be more willing to intervene (Eagly & Crowley, 1986) and at other times has shown females to be more willing to intervene (Griffith et al., 2006). The fact that we found no gender differences for intentions to intervene for either DV or SV within the same study and using the same methods and sample does not support the possibility that some of the conflicting findings within prior research may be due to the different types of violence examined in different studies.

Our findings also show statistically significant effects for education and attitudes about gender equality for intention to intervene for both DV and SV. In both situations, high school graduates and those with more egalitarian attitudes regarding gender equality are more likely to state an intention to intervene. Although these data do not permit us to positively determine more about the reasons for these patterns, it is likely that bystanders with higher education have learned that DV is not a private matter but a crime they can prevent. Alternatively, or in addition, perhaps their better knowledge of and ability to identify violent situations outweighs the barriers of intervention, including potential fears of misinterpreting the situation, physical harm, and public embarrassment, all of which have been identified as common among bystanders (Darley & Latane, 1968; Weitzman et al., 2020).

Although the research sample does not permit cross-national comparisons of attitudes about gender equality, it is clear from the results that, in Korea, gender equality attitude is among the most important of the variables that predict intention to intervene. In addition, while gender differences did not significantly predict intentions to intervene, positive attitudes about gender equality did. From these findings, we can likely assume that the patriarchal social norms that are still strong in Korea are likely causing bystanders with less positive attitudes toward gender equality to agree with traditional ideas that “DV is a family matter” and SV may not be a serious crime and subsequently be less likely to intervene in these offenses.

Type of Intervention in DV and SV

Turning toward the models examining the type of intervention, the findings in Table 7 show that among the subset of people who state an intention to intervene in some way, males are more likely to intervene directly than are females for both DV and SV situations. These findings are consistent with much of the prior research and may result from a male’s perception that they have the physical ability to stop the violence or, alternatively, males may have a desire to be seen as a masculine and powerful hero (Banyard, 2008; Eagly & Crowley, 1986). The gender difference is much more pronounced for SV than DV, where the OR for gender is 2.55 for DV and 4.71 for SV.

The effects of the other variables are also quite different in the DV versus SV situations. Those who were 60 years of age or older are more likely to intervene directly in DV situations but not in SV situations. We remain uncertain exactly why the DV and SV results differ for age.

Although the coefficient for being a high school graduate is negative for both situations, the magnitude is statistically significant only in the DV model. Perhaps bystanders with lower education have less ability to distinguish the risk factors of direct intervention in DV such as an increased risk of homicide of the victim (Dugan et al., 2003). Again, however, we remain uncertain why the DV and SV results differ for this factor.

Employed individuals are more likely than the unemployed to directly intervene in SV situations, while the effect of employment for DV situations is near zero. Although this finding is consistent with prior literature, the underlying reasons for this relationship also remain unclear. Perhaps, however, motivational differences or differences regarding one’s perceptions about the locus of control might be responsible.

Gender equality attitude is associated with indirect rather than direct intervention for DV, while having a near-zero effect in the SV model. It is somewhat difficult to speculate further about the reasons underlying this finding, and for this reason, future research may be required to better understand this trend. Subsequently, one of the areas that merits particular attention in future research is the in-depth examination of personality traits such as intelligence and corresponding differences in the decision-making processes of bystanders. For example, higher levels of intelligence may lead individuals to consider the need of others and favor indirect intervention strategies, aligning with the findings of increased indirect intervention among those with higher education and more positive gender equality attitudes. Researchers may need to employ comprehensive measures of these traits and examine their interactions with demographic and attitudinal variables. In addition, qualitative research could offer more valuable insights into the underlying motivations and thought processes of bystanders.

In addition, while the current study focused on the main effects of individual predictor variables on bystander intervention intentions and types, it is important to acknowledge the potential for more in-depth exploration of interactions between these factors. Specifically, future research could examine whether gender interacts with gender equality attitudes in shaping bystander intervention behaviors. Also, exploring interactions between other individual and situational variables, such as age, education, and cultural factors, could provide a more nuanced understanding of the complex dynamics involved in bystander intervention. Understanding how various factors intersect to influence bystander behavior can contribute significantly to the development of more targeted and effective intervention methods in the future.

Limitations

The NFFSGE survey data have limitations that constrain the analysis and substantive findings. First, these data do not provide information on the gender of the victim or offender. The participants, however, are likely to imagine the victim as female and the offender as male, especially given the patriarchal nature of Korean society and the likelihood that many Koreans know that the majority of DV and SV offenses reported to the police have female victims (Korean Statistical Information Service, 2019). In addition, these data do not provide the exact age and education level of the participants, but only include broad categories. This was a decision apparently made by the data stewards to ensure privacy for respondents. These data also do not include information about situational factors in detail (e.g., time of the day, place, the relationship between the bystander and victim or offender, and the number of other bystanders).

A question that applies to the current study as well as much of the existing literature based on survey methods is whether it is realistic to constrain participants to select only one possible intervention type. For example, the question regarding direct versus indirect intervention in the NFFSGE survey data constrains participants to only one of the following options: (1) directly intervene; (2) call the police; (3) ask for help; and (4) stay out of it. It is certainly possible, however, for someone to call the police before asking another bystander to help with intervening directly.

Another potential limitation is a known issue of self-report surveys, whereby participants’ responses to the survey questions may be shaped by social desirability biases. This becomes more likely with surveys such as the NFFSGE, which are conducted with an investigator present (Mayhew et al., 2018).

In addition, these data did not contain cross-national information and we could not include different cultural aspects in our analysis. Consequently, we cannot ascertain whether some of our unique findings are due to particular cultural elements of Korea.

Finally, this study is based on hypothetical situations, measuring only the participants’ intention to intervene and not actual behavior. Previous research, however, has compared bystanders’ intention to intervene and actual intervention and found that intentions are moderately predictive of actual behavior (Chabot et al., 2009; Mayhew et al., 2018; Nicksa, 2014).

Conclusion

A key aim of this study was to close several gaps in bystander research in the Korean DV and SV context. Although most of this study’s findings are similar to what has been found in previous Western studies, Korean cultural aspects may have resulted in unique outcomes, including the fact that older bystanders are more likely to intervene directly in DV situations and that a bystanders’ employment status is a significant predictor of a bystander’s type of intervention in SV situations in addition to the fact that positive attitudes about gender equality are associated with increased intentions to intervene in both DV and SV and with direct intervention with DV (but not SV) situations.

Despite the limitations, this study provides a substantial contribution to the bystander literature by highlighting new information about bystander behaviors in a Korean sample. The findings demonstrate the effects of bystander gender, gender equality attitudes, age, education level, and employment status. Furthermore, this study highlights the uniqueness of Korean culture and social norms and the importance to further explore how these unique aspects may affect the efficacy of bystander intervention as a prevention approach to both DV and SV.

These findings also suggest the need for public education campaigns that can allow the general public to have a clear understanding about the severity of DV and SV (Banyard et al., 2004; Lee, 2012). Future research is, of course, required to clarify these results through a more detailed examination of bystander characteristics and situational aspects.

Author Biographies

Woojung Aria Ahn is a Criminology and Criminal Justice doctoral student, who has recently finished her bachelor degree with honours, at Griffith University in Queensland, Australia. Her research interests include both online and offline domestic violence, child abuse, and bystander intervention.

Jeffrey Ackerman, PhD, is a senior lecturer in the School of Criminology and Criminal Justice at Griffith University. He has studied intimate partner violence, juvenile delinquency, victimization, and related matters.

Nadine Connell, PhD, is an associate professor in the School of Criminology and Criminal justice at Griffith University. She studies juvenile delinquency and school safety, with a focus on program and policy evaluation.