When Are Bystanders More Likely and Confident to Act: Validation of the Bystanders to Alcohol Risk Scales - Willingness and Self-Efficacy

Abstract

Objective:

This study evaluated the psychometrics of two new measures of willingness and self-efficacy to intervene in alcohol-related risk situations involving others. Items assess the constructs across bystander scenarios that vary by situation (before, during, after risk), one’s relation to the target (friend/stranger), and the gender of the target (man/woman).

Method:

Young adults (N = 1,011, M_age 22.0 years, 46.0% women, 70.7% White, 23.3% Latinx), completed the new measures and related measures to assess validity, and a subsample was re-tested two weeks later. Multivariate variance decomposition was used to determine how the scenario components influenced variance; bivariate correlations evaluated validity and intraclass correlations evaluated test-retest reliability.

Results:

Models for willingness and self-efficacy constructs fit the data well. The largest component contributing to both measures was the general component, reflecting overall willingness/self-efficacy regardless of situation, relation, and gender components (alphas >.90). For willingness, the next largest component was situation, with “after” drinking showing the greatest variability but the lowest mean. For self-efficacy, the second most important component was relation to the target, with friend showing higher self-efficacy than stranger. Participants indicated higher willingness and self-efficacy when the target was a woman, and women participants reported higher willingness than men. Validity was supported by significant correlations with bystander behaviors and alcohol-related constructs. Test-retest reliability indicated moderate stability for both measures (ICCs: 0.50–0.64).

Conclusions:

These measures provide valid and reliable tools for measuring bystander intervention willingness and self-efficacy in alcohol-related risk situations and offer utility for the evaluation of bystander intervention programs.

Rates of alcohol use, heavy drinking and high intensity drinking are highest in young adulthood in the US (Linden-Carmichael et al., 2017; SAMHSA, 2025). The range and volume of alcohol-related consequences experienced by emerging adults have been summarized in multiple reports (Barnett et al., 2014; Hingson et al., 2016; Patrick et al., 2020), and most serious alcohol-related consequences (alcohol poisoning, assault, injury) are characterized as acute (Centers for Disease Control and Prevention, 2024), and could be avoided or minimized with intervention. The vast majority of alcohol use at this age occurs in social settings with same-age peers (Beck et al., 2008; Lau-Barraco & Linden, 2014), and young adults are commonly around someone who has consumed too much alcohol or is showing signs of risk (Christiansen et al., 2002; Oster-Aaland et al., 2009), regardless of whether they are drinkers themselves. Young adults observing alcohol-related behavioral risk therefore have the opportunity and influence to act to minimize acute risk in their peers.

Bystander intervention (BI), which involves intervening to help others in hazardous situations, has been identified as a promising approach to address various community problems, including drunk driving, suicidality, and bullying (Cimini et al., 2014; Guerette et al., 2013; Polanin et al., 2012), with research and intervention development most rapidly advancing in the area of dating and sexual violence prevention (Coker et al., 2016; Mainwaring et al., 2023). BI training programs in dating and sexual violence prevention, for example, encourage community members to challenge the social norms that sustain and condone violence against women, be willing to take action to interrupt inappropriate social and dating situations, and be proactive in providing support (Mujal et al., 2021; Pfaff et al., 2024). These programs show efficacy for increasing bystander intentions, actions, and attitudes (Kettrey & Marx, 2019) and researchers have found lower violence victimization and perpetration rates following BI training compared to comparison groups (Coker et al., 2016; Coker et al., 2015).

Bystander intervention training programs targeting alcohol use have been less developed than those targeting sexual violence, despite the high rates of hazardous alcohol use in young adults (Patrick et al., 2024) and the likelihood of peer presence during alcohol use (Lau-Barraco & Linden, 2014). Indeed, recent research has emphasized the potential success of training peers in best practices for helping their friends (Read et al., 2024). However, to adequately assess these training programs and the efficacy of bystander intervention, valid measures are needed (Mennicke et al., 2022). Our work in this area has consisted of thoroughly investigating alcohol bystander constructs using mixed methods (Haikalis et al., 2025), and subsequently validating measures of exposure to alcohol risk and opportunity for alcohol bystander intervention (Barnett et al., 2024), barriers and facilitators to intervening (Haikalis et al., in preparation), use of bystander intervention strategies in alcohol risk situations (Meisel et al., 2024), and consequences of helping (Meisel et al., 2025). Here, we describe our evaluation of measures of willingness to change and self-efficacy for acting in alcohol risk situations, which are important antecedents to behavioral action.

Willingness to Help

The Prototype Willingness Model (PWM) (Gerrard et al., 2008) is applicable to socially based health behaviors like bystander intervention (BI), as it invokes dual processes that reflect a reactive process of willingness or openness to act given the opportunity, and more conscious, deliberative intention or plans (Gerrard et al., 2008; Lewis et al., 2020). In bystander intervention for the prevention of sexual assault, the construct of intent to help is more often conceptualized and measured than willingness (Labhardt et al., 2017), but we prefer the term willingness for several reasons. First, as framed by the PWM, behavior may not be reasoned or intentional, and indeed may occur in reaction to circumstances, often social (Gerrard et al., 2008; Pagani et al., 2022; Todd et al., 2016), which is fundamentally how bystander behavior occurs. Relatedly, behavioral willingness does not require instrumental action (Gibbons, Gerrard, Ouellette, et al., 1998), but reflects the degree to which an individual thinks they would react to a particular risk situation, whether or not they have experienced that situation (Gerrard et al., 2008). Consistent with this conceptualization, measures of willingness commonly present a specific hypothetical situation and ask the responder how likely they are to engage in bystander behavior (Gibbons, Gerrard, Blanton, et al., 1998). We conceptualize bystander intervention willingness, then, as reflecting the degree to which an individual thinks they would act in response to the exposure of risk in others, i.e., the openness to react to an opportunity (Boyle et al., 2023). Moreover, a component of the PWM is a hypothesized process whereby the individual may anticipate social consequences for their behavior (Pomery et al., 2009), a process that is consistent with the experience of bystanders who perceive barriers and possible consequences to intervening on the risk behavior of others (Bennett et al., 2014; Haikalis et al., 2026; Meisel et al., 2025). No validated measures exist that assess willingness to help a peer who is engaging in risky alcohol use.

Self-Efficacy to Help

Self-efficacy is a situation-specific self-perception of one’s capability to be successful at accomplishing a goal (Bandura, 1977). In BI for sexual assault, the self-perception that one is capable of successfully intervening is directly related to whether the person will act (Latané & Darley, 1970). Items measuring self-efficacy for BI for sexual assault (e.g., asking how confident a person is to perform bystander behaviors to prevent or intervene in sexual assault; Banyard, 2008) can be used for initial item development for a measure addressing self-efficacy in engaging in alcohol-related BI, as can measures of self-efficacy for changing one’s own drinking behavior (Breslin et al., 2000; DiClemente et al., 1994). However, these measures require adaptation to ensure that they address alcohol-related BI specifically.

Measurement Considerations

Whereas existing assessments of BI provide a good starting point for assessment of alcohol BI, current measures are not adequate, nor are they easily revised to assess alcohol-related BI through wording changes. Moreover, bystander scholars highlight the challenges in assessing bystander outcomes and the need for measures to incorporate the nuanced contexts in which there is opportunity for bystander action (Bush et al., 2019). For example, whether the behavior is observed before a negative outcome is evident (e.g., drinking quickly), during an episode when an outcome is evident (e.g., vomiting), or is reported later (e.g., got drunk the night before), may affect willingness and self-efficacy. In our formative qualitative work (Haikalis et al., 2025), which drew from an established framework for bystander intervention for sexual violence (McMahon & Banyard, 2012), participants reported that their willingness and self-efficacy to intervene in alcohol-risk situations commonly depended on the characteristics of the person showing risk, and specifically whether that person was a friend (vs. stranger; consistent with other bystander intervention research; Bennett et al., 2014; Walker et al., 2024) and whether the person presented as a man or woman. As willingness and self-efficacy will be situation-specific, it is necessary to consider situations that vary in circumstances and characteristics of the person needing help to accurately measure these constructs. Incorporating these dimensions will allow for more comprehensive measurement of bystander willingness and self-efficacy.

Study Objective and Hypotheses

The purpose of this study is to present the development and evaluation of complementary measures of Willingness (i.e., one’s self-rating of the likelihood that one would intervene) and Self-Efficacy (i.e., one’s self-rating of the belief that one would be effective in such intervention) for situations in which drinking-related risk or harm is observed in others. In these measures we include items that reflect risks observed before, during and after actual drinking situations, as well as derivations varying the gender and familiarity with the person showing the signs of hazardous drinking. We conducted psychometric evaluation in five broad steps. First, we conducted a multivariate variance decomposition analysis to evaluate the importance of the components (i.e., situation, target relationship, target gender) contained in the item sets. In the second step, we conducted construct validity of the new measures, in which we compared the Willingness and Self-Efficacy measures to three groups of related measures:

• (a)Measures of bystander intervention in alcohol-related situations, including exposure to alcohol-related risk in others, perceived opportunity to intervene, barriers and facilitators of bystander action, and use of bystander strategies, as well as bystander self-efficacy in sexual violence risk situations (for our self-efficacy measure only). We expected exposure and perceived opportunity to be positively related to willingness to intervene because bystander theory emphasizes that observing risky behavior is a prerequisite for taking action (Latané & Darley, 1970). In our prior work, we found that exposure was related to bystander strategy use (Barnett et al., 2024), and so anticipate exposure will also be related to willingness. We expect that exposure and opportunity will also be positively related to self-efficacy based on the assumption that familiarity with alcohol-related risk will increase confidence to act; this is supported by bystander research as well (Krieger et al., 2017).

By their nature, perceived barriers to bystander intervention should inhibit willingness to help, and facilitators should be positively related to bystander willingness (Bennett et al., 2014; Park et al., 2023). Self-efficacy has been framed as a facilitator of intervention in itself (Latané & Darley, 1970), and lack of self-efficacy has been identified as a barrier (Park et al., 2023), and there is evidence that individuals higher in bystander self-efficacy are more likely to intervene to prevent sexual assault (Yule & Grych, 2020) and peer victimization (Sjögren et al., 2020). Thus, we expect willingness and self-efficacy will be positively related to perceived facilitators and negatively related to perceived barriers to intervention.

For bystander strategies, there should be positive correlations with willingness/self-efficacy, consistent with the PWM (Pomery et al., 2009) and other bystander research (Banyard et al., 2014; Boekeloo & Griffin, 2009; Krieger et al., 2017). Finally, we expected our new measure of self-efficacy to be positively related to self-efficacy to engage in bystander behavior in response to sexual violence risk.

• (b)Measures of drinking-related constructs, including use of protective behavioral strategies (PBS), descriptive and injunctive drinking norms, and participants’ own alcohol use and level of alcohol risk. Protecting oneself by using protective behavioral strategies in drinking situations intersects with one’s use of bystander strategies to help others (Kenney et al., 2024), and in our recent work, we found a strong correlation between personal PBS use and adoption of bystander strategies (Meisel et al., 2024), so predict that willingness and self-efficacy to help others would also be related to the use of PBS.

We can think of drinking norms as the context in which bystander exposure occurs. Based on the logic that the more a person thinks others drink and approve of drinking, the less willing and confident they will be in interrupting others’ drinking-related risk behavior, we hypothesized a negative relationship between drinking norms and willingness/self-efficacy to help.

For alcohol use and problems, it is plausible that drinking a higher quantity or showing higher risk would make a person more comfortable acting to support others (supported by Boekeloo & Griffin, 2009); alternatively, given familiarity with (and possible minimization of) risk, one might be less willing to intervene (Nelson et al., 2021) and/or less confident in acting. For these competing reasons, we pose no directional hypotheses for the relationship between willingness/self-efficacy and alcohol use/problems.

• (c)General measures of social assertiveness and social desirability; we expect willingness and self-efficacy to be positively related to both. Assertive individuals are more likely to intervene when they are exposed to bystander risk (Jenkins & Nickerson, 2017), and assertiveness training is a type of bystander intervention training (Yosep et al., 2024). Social desirability (i.e., bias that emerges when there is a clear socially acceptable response) has been found to relate to intent to act and to taking action in bystander situations (Banyard et al., 2014; Brown & Messman-Moore, 2010).

Third, we analyzed the association between the new willingness and self-efficacy measures, expecting a positive correlation, consistent with findings in prior bystander research (Banyard et al., 2014; Pagani et al., 2022). Fourth, we evaluated test-retest reliability of both measures. Finally, informed by observations of gender differences in willingness to engage in bystander intervention (Hackman et al., 2022; Walker et al., 2024), we investigated gender differences and gender invariance for both measures.

Method

Procedure

This work is one component of a broader initiative aimed at developing a battery of bystander-related measures in alcohol-risk scenarios. The procedures of our iterative qualitative approach to item formulation and refinement can be found in Haikalis et al. (2025). To ensure a sample reflective of national demographic diversity in terms of race, ethnicity, and gender within our target age range, we contracted with the Qualtrics Panel team (Qualtrics, 2022). Participants were 18 to 25 years old who answered “Yes” to the item “In the past 3 months, have you been around someone who showed signs of alcohol intoxication (e.g., slurring words, vomiting, stumbling)?” The parent project focuses on bystander-related attitudes and behaviors, so a baseline level of risk exposure was necessary (see Barnett et al., 2024; Meisel et al., 2024 for more information).

Since the parent study involved evaluating several new measures, it necessitated administering numerous validation measures. To reduce burden, participants completed demographic items and were randomly assigned to answer subsets of other measures. Two-thirds of participants completed items for each measure. For a more detailed explanation of how the full battery was administered to subsets of participants, please contact the authors. The battery took approximately 20 minutes to complete and contained checks for speed, attention, and consistent answering. A random subset of participants completed a two-week retest survey containing only the new measures. All procedures were approved by the university’s Institutional Review Board.

Participants

Participants (N = 1011) included in the current study were on average 22.0 years old (SD = 2.3). Gender identity was reported as cis/trans-man: 51.2%; cis/trans-woman: 46.0%, and 2.8% non-binary, gender non-conforming, or genderqueer. Racial identity was predominantly White (70.7%), followed by Black (13.3%), Asian (4.7%), Multiracial (4.5%), American Indian or Alaskan Native (1.4%), and Native Hawaiian or other Pacific Islander (0.6%). Latinx ethnicity was reported by 23.3%. Any drinking in the past three months was reported by 81.1% of the sample. A subset of 345 participants completed the retest survey approximately two weeks after the first observation (M_days = 13.33; SD = 4.13).

Measures

Bystander willingness and bystander self-efficacy.

These two newly designed measures (items for both are in Table 1) contain three hypothetical prompts reflecting examples of situations that might occur prior to an occasion of hazardous drinking (before, prompt A), within an occasion of drinking (during, prompt B), or after drinking (after, prompt C). For each of these situational prompts respondents are asked to indicate their willingness to intervene (“How likely are you to do something to try to reduce risk or danger in this situation”) and (in a separate set of questions) their perception that they will be efficacious in their intervention efforts (“How confident are you that you would be effective in reducing risk or danger in this situation”). In addition, each of the three situational prompts is answered relative to four hypothetical persons that are the cross products of the two sets friend vs stranger and men vs women (i.e., if the person is “a friend who is a woman,” “a stranger who is a woman,” “a friend who is a man,” “a stranger who is a man”). The three situational prompts (before, during, after) by four hypothetical person response categories (friend/stranger, man/woman) result in 12 possible ratings in each of the Willingness and Self-Efficacy instruments. However, given prompt C (after) implies some prior knowledge of the person, and this prior knowledge would not be available for strangers, we leave off the stranger response person type to prompt C questions, resulting in 10 items for each measure. Responses are collected on a five-point response scale; for the Willingness instrument, from 0 (not at all likely) to 4 (extremely likely) and for the Self-Efficacy instrument, from 0 (not at all confident) to 4 (extremely confident). Items are summed.

Table 1.

Alcohol Bystander Willingness to Help and Alcohol Bystander Self-Efficacy Items

Alcohol Bystander Willingness to Help
Instructions: These questions ask about how likely you are to do something to help someone in different situations.
	Not at all likely	Slightly likely	Somewhat likely	Very likely	Extremely likely
AYou hear that someone is planning to drink to a level that concerns you (e.g., talking about getting drunk, wasted, or black out). How likely are you to do something to try to reduce risk or danger in this situation if the person is…
1. A friend who is a woman	0	1	2	3	4
2. A stranger who is a woman	0	1	2	3	4
3. A friend who is a man	0	1	2	3	4
4. A stranger who is a man	0	1	2	3	4
BYou see someone showing signs that they drank too much during a drinking event (e.g., slurring, stumbling, aggressive, dilated pupils) and you are concerned. How likely are you to do something to try to reduce risk or danger in this situation if the person is…
5. A friend who is a woman	0	1	2	3	4
6. A stranger who is a woman	0	1	2	3	4
7. A friend who is a man	0	1	2	3	4
8. A stranger who is a man	0	1	2	3	4
CYou know that someone got dangerously drunk the night before and you are concerned because this has happened more than once. How likely are you to do or say something to help (e.g., express concern, offer resources) if the person is…
9. A friend who is a woman	0	1	2	3	4
10. A friend who is a man	0	1	2	3	4

Alcohol Bystander Self-Efficacy
Instructions: The following questions will ask about your confidence in your ability to be effective when your goal is to reduce harm to others when they are drinking.
	Not at all confident	Slightly confident	Somewhat confident	Very confident	Extremely confident
AYou hear that someone is planning to drink to a level that concerns you (e.g., talking about getting drunk, wasted, or black out). How confident are you that you would be effective in reducing risk or danger in this situation if the person is…
1. A friend who is a woman	0	1	2	3	4
2. A stranger who is a woman	0	1	2	3	4
3. A friend who is a man	0	1	2	3	4
4. A stranger who is a man	0	1	2	3	4
BYou see someone showing signs that they drank too much during a drinking event (e.g., slurring, stumbling, aggressive, dilated pupils) and you are concerned. How confident are you that you would be effective in reducing risk or danger in this situation if the person is…
5. A friend who is a woman	0	1	2	3	4
6. A stranger who is a woman	0	1	2	3	4
7. A friend who is a man	0	1	2	3	4
8. A stranger who is a man	0	1	2	3	4
CYou know that someone got dangerously drunk the night before and you are concerned because this has happened more than once. How confident are you that you would be effective in doing or saying something to help (e.g., express concern, offer resources) if the person is…
9. A friend who is a woman	0	1	2	3	4
10. A friend who is a man	0	1	2	3	4

^APrompt A: before;

^BPrompt B: during;

^CPrompt C: after. Note that prompt C only has two response options because knowledge of prior/patterns of behavior is not possible for strangers.

Footnote information should not be included in the administered measures.

Exposure and opportunity.

Exposure to alcohol risk in others was measured using the Exposure to the Hazardous Drinking of Others scale (EHDO; Barnett et al., 2024). This measure contains 21 items presented in three sections; the first section starts with the prompt “In the past 3 months, I saw someone who had been drinking alcohol and they were…” and contains 12 items that reflect in-the-moment observed risks or consequences (e.g., “slurring their words”). The second section starts with the prompt “In the past 3 months someone told me that they…” and contains 4 items that reflect risks or consequences reported to the participant by others (e.g., “wanted to drink a lot to cope with something”). The final section starts with the prompt “In the past 3 months I have noticed someone…” and contains 5 items that reflect participant observation of a concerning behavior or pattern (e.g., “drinking more than they used to”). Answer options are: 0 (not in past three months), 1 (once), 2 (twice), 3 (three or more times). The measure has two factors: Situational Risk Signs (omega = .93) and Problematic Pattern (omega = .88).

If participants endorsed one or more items on the EHDO the Perceived Alcohol-Related Bystander Opportunity item was administered: “In the past 3 months, how often do you feel like you had the opportunity to do something to help someone who was showing a sign of possibly drinking too much?” Answer options range from 0 (never) to 11 (every day). The validation for this item is also in Barnett et al. (2024).

Barriers and facilitators.

The Bystanders to Alcohol Risk Scales - Barriers and Facilitators (BARS-BF) is a 19-item measure of circumstances that could interfere with or enhance a person’s intervening on an observed alcohol risk (Haikalis et al., in preparation). The instruction is “How much would the following circumstances make it more or less likely that you would help another person who has been drinking too much?” Barriers (11 items, e.g., “If I feel like it might not be safe to intervene”) and Facilitators (8 items, e.g., “If the person asks for help”) subscales were derived from exploratory and confirmatory factor analysis, with alphas of .91 and .88, respectively. Items are answered on a five-point scale ranging from 0 (I would be much less likely to help) to 4 (I would be much more likely to help). Subscale items are summed, with barriers items reverse scored for analytic interpretability.

Alcohol bystander strategies.

The Bystanders to Alcohol Risk Scales–Strategies measure (Meisel et al., 2024) consists of 17 items that assess frequency of helping another person in a drinking event in the past three months. It contains two subscales; Level One (9 items) reflects strategies used before or during drinking events (e.g., “I brought water or other nonalcoholic beverages to someone”), and Level Two (8 items) contains strategies more likely to be used during or after an event or when a problematic pattern has emerged (e.g., “I suggested someone talk to a professional counselor”). Response options range from 0 (not in the past 3 months) to 3 (three or more times), with item responses summed by subscale. Alphas in this sample were .88 for Level One and .89 for Level Two.

Self-efficacy in bystander sexual assault situations.

The Bystander Efficacy Scale (Banyard, 2008) contains four items that assess a person’s confidence in performing bystander behaviors when a person is at risk of sexual assault (e.g., “Do something if I see a woman surrounded by a group of men at a party who looks very uncomfortable”). Participants indicate how confident they are that they could perform these behaviors on a scale from 0% (can’t do) to 100% (very certain). Alpha was .92 in this sample.

Protective behavioral strategies.

The Protective Behavioral Strategies Scale-20 (PBS-20; Treloar et al., 2015) measures how often people employ different strategies to minimize the adverse effects of alcohol. The PBS-20 contains three subscales: Limiting/stopping drinking (7 items; e.g., “Alternate alcoholic and nonalcoholic drinks”; α =.80), Serious harm reduction (8 items; “Know where your drink has been at all times”; α =.78), and Manner of drinking (5 items; “Drink slowly, rather than gulp or chug”; α =.76). Response options range from 1 (never) to 6 (always), with items on each subscale summed.

Descriptive and injunctive drinking norms.

Two items from the Drinking Norms Rating Form (Baer et al., 1991) asked participants to estimate for a same-age and same-gender individual the frequency of heavy drinking from 0 (never) to 11 (every day) and the maximum number of drinks (from 0 to 25+). For injunctive norms, participants were similarly asked to estimate the frequency of heavy drinking and the maximum number of drinks someone of their same age and gender would consider acceptable, with the same wording and response options of descriptive norms items. The four items were analyzed separately.

Alcohol use and alcohol use severity.

For alcohol use we measured alcohol quantity, “In the past 3 months, on the days when you drank, how many drinks did you drink on average?” answered on a scale from 0 to 30 drinks, and heavy drinking frequency, “Considering all types of alcoholic beverages, how many times in the past 3 months did you have 4/5 or more drinks in one occasion?” (4+ for women, 5+ for men) answered on a 12-point scale ranging from 0 (never) to 11 (every day). Both alcohol use items were adapted from the Behavioral Risk Factor Surveillance System Questionnaire (Centers for Disease Control and Prevention, 2021). The Alcohol Use Disorders Identification Test (Saunders et al., 1993) is a 10-item measure used to screen for alcohol use severity in the past year; alpha was .86.

Assertiveness.

The Assertiveness Inventory (Wills et al., 1989) has two subscales: general assertiveness contains 7 items that include assertiveness in everyday situations such as requesting service in consumer situations or disagreeing with others (e.g., “Tell people when you feel they have done something that is unfair“). The social assertiveness subscale contains 4 items that include engaging someone in conversation, making compliments, and inquiring about a possible offense (e.g., “Compliment a person you are going out with”). Items are scored from 1 (never do this) to 5 (always do this) and summed; alphas for the general and social subscales were .81 and .60, respectively.

Social desirability.

Social desirability was measured using the Socially Desirable Response Set-5 (Hays et al., 1989), a 5-item scale (e.g., “I am always courteous even to people who are disagreeable”) with response options from 1 (definitely true) to 5 (definitely false). This measure is scored according to the number of items on which the extreme socially desirable response was endorsed. Alpha was .60.

Data Analysis

Typically, in measures development, there is a common sequence of steps, starting with items being subjected to a factor analysis to evaluate the extent to which the responses can be attributed to one or more underlying common latent factors. Not uncommonly, it is hypothesized or desired to claim that a single common factor is sufficient to account for a supermajority of the variance among the test item responses, and that secondary factors, if present, capture a minority of variance related to subsets of items sharing specific content or methods variance (Widaman, 2007). In this study, the underlying structure of the items in both instruments is apparent. We can expect shared variance among items referring to strangers, to friends, to men, to women, and to items having the same situational prompt. For use as a pair of overall summary measures of willingness and self-efficacy, then, our main goal is to evaluate the extent to which a single factor loading in all 10 items (separately by measure) accounts for the majority of shared variance in item responses, while controlling for situational prompt and target-related variability apparent in the construction of the instrument. An illustration of our general modeling approach is shown in Figure 1, using willingness items as an example.

Figure 1.

Modeling Approach Reflecting Situational, Relationship, and Gender Components. The Stem Reflects the Willingness Measure, but Components are the Same for the Self-Efficacy Measure

Multivariate variance decomposition.

We estimate our models using Mplus software (version 8.10, Muthén & Muthén, Los Angeles, CA) as highly constrained confirmatory factor analysis (CFA) models, to estimate a multivariate variance decomposition CFA model (Widaman, 2007). Item responses are treated as categorical indicators and related to one another via polychoric correlation coefficients. A set of latent variables are introduced as illustrated in Figure 1, corresponding to a general or common variance component (G) and specific components corresponding to methods-related variance given the item content (S, strangers; F, friends; M, men; W, women; A, before; B, during; C, after for prompts). All loadings are fixed to one, and latent variable variances are freely estimated. Covariances among the latent variables (variance components) are assumed to be 0. We model the mean structure part of the model with a set of four thresholds separating the (five) observed response categories. One of the thresholds was fixed to 0 to identify the underlying response variable mean. All other thresholds were constrained to be equal for all items. Underlying response variable means were freely estimated. We estimate means for variance components, fixing one in a contrast set to zero for identification. We summarize models with omnibus fit statistics, and the magnitude of the variance components. The variances of components are equivalent to the square root of the standardized factor loading for the item in the component. We used the Mplus weighted least squares estimator. Willingness and self-efficacy measures were analyzed separately.

Validation, test-retest reliability, gender differences.

Bivariate correlations were used to evaluate validity and intraclass correlations (ICCs) were calculated for test-retest reliability. Analysis of variance compared scores for the scales among men, women, and nonbinary participants. Gender differences observed in total scores were probed with analyses directed at testing the assumption of measurement invariance.

Transparency and Openness

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. The sample size was determined a priori based on guidelines suggesting 500 participants is adequate for confirmatory factor analysis (Comrey & Lee, 1992). We conducted a simulation study to evaluate the probability that an item set yielded a Cronbach’s alpha > 0.67 if the population alpha was 0.70, which is 89% with 600 participants. For reliability tests (i.e., test-retest) and calculation of correlations to demonstrate construct validity (De Vet et al., 2011), smaller sample sizes (n=50) are required. Descriptive data were analyzed using SPSS, version 29 and models were estimated using Mplus software. This study was not registered. Materials are available by emailing the corresponding author.

Results

There were no item-level missing data in the 677 respondents to the willingness measure, and no item-level missing data for the 671 respondents to the self-efficacy measure. The distribution of responses for the two measures is shown in Figure 2.

Figure 2.

Distribution Of Responses for the Willingness and Self-Efficacy Measures, With Situational (A = Before, B = During, C = After), Relationship (F = Friend, S = Stranger), and Gender (W = Woman, M = Man) Elements Combined

Variance decomposition: Willingness.

Results for willingness are summarized in Table 2. This model fit the data reasonably well, with a confirmatory fit index (CFI) of 0.966, a root mean squared error of approximation (RMSEA) of 0.087, and a standardized root mean squared residual (SRMR) of 0.042. Targets for good fitting models for these statistics are 0.95 or higher for CFI, less than 0.05 for RMSEA (MacCallum et al., 1996), and less than 0.08 for SRMR (Hu & Bentler, 1999). The SRMR is the most directly interpretable, as it is a summary of the average residual between the sample data (i.e., correlations among response items) and the model-implied values for these estimates.¹ The good fit on SRMR and CFI are notable given the model estimates only 8 variance components and there are 45 (((10*10)-10)/2) correlations among the response items to account for. Cronbach’s alpha was .90.

Table 2.

Variance Components and Means for the Willingness Construct by Situation, Target Relationship, and Target Gender

Component	Variance (SE)	Mean (SE)
G (general)	0.38 (0.03)	−0.33 (0.04)
Situation
A (before drinking)	0.16 (0.01)	−0.22 (0.03)
B (during drinking)	0.11 (0.01)	0 (Ref)
C (after drinking)	0.25 (0.02)	−0.31 (0.03)
Relationship to target
F (friend)	0.18 (0.01)	0.72 (0.03)
S (stranger)	0.20 (0.02)	0 (Ref)
Gender of target
M (man)	0.13 (0.01)	0 (Ref)
W (woman)	0.16 (0.01)	0.41 (0.03)

Note: The variance component estimates are equal to squared standardized factor loadings for items in components so they reflect the proportion of the variance in the item response that can be attributed to the component.

Ref = Reference

For willingness, the component contributing the most variance to the item responses overall is the general component (G in Table 2). Because of how the model is parameterized and categorical response variables are handled, this variance component can be interpreted as the proportion of variance in item responses (averaged across all response items) that can be attributed to the G component (i.e., 38%). We can interpret G as capturing overall willingness regardless of the situational (i.e., situation, relationship, gender) factors. The next-most important willingness component is labeled C, which is the shared variance among the items that utilize prompt C (the situation referring to “after” drinking). This component explains 25% of the variance in the items that refer to “after” situations. Thus, for willingness, we have confirmed that the most important variance component is that due to the general factor, with the second most important being the situational element of “after” drinking. Adding the variances from the relationship and the gender components (friend + stranger, man + woman in Table 2), it appears that the relationship component carries more variance than the gender component.

The means of the components in Table 2 are on an arbitrary scale, but for the situation prompt (A, B, C), they indicate the respondents endorse significantly greater willingness for prompt B (during), than A (before) or C (after), and significantly greater willingness on prompt A than C, holding constant the effect of relationship to target (friend/stranger) and gender of the target (man/woman). Comparing the relationship and gender component means and standard errors in Table 2, we can see that respondents indicate significantly higher willingness when the target is a friend or women, relative to strangers and men, respectively.

Variance decomposition: Self-Efficacy.

Results for the self-efficacy measure are summarized in Table 3. This model fit the data reasonably well, with CFI of 0.968, RMSEA of 0.09, and SRMR of 0.042. Cronbach’s alpha was .91. As with willingness, the component contributing the most to the item responses as a whole is the general component (G in Table 3), that is, 34% of the variance in item responses that can be attributed to the G component. Thus, for self-efficacy, as with willingness, we have confirmed that the most important variance component is that due to the general factor, reflecting overall confidence for intervening regardless of the other (i.e., situation, relationship, gender) factors. The next most important component is the relationship component, with the stranger relationship showing the highest proportion (26%), indicating that after the general component, the next most important variance component is whether the person needing help is a stranger.

Table 3.

Variance Components and Means for the Self-Efficacy Construct by Situation, Target Relationship, and Target Gender

Component	Variance (SE)	Mean (SE)
G (general)	0.34 (0.03)	−0.38 (0.03)
A (before drinking)	0.08 (0.01)	−0.02 (0.01)
B (during drinking)	0.09 (0.01)	0 (Ref)
C (after drinking)	0.20 (0.02)	−0.07 (0.02)
Relationship to target
F (friend)	0.21 (0.02)	0.36 (0.02)
S (stranger)	0.26 (0.02)	0 (Ref)
Gender of target
M (man)	0.19 (0.01)	0 (Ref)
W (woman)	0.20 (0.01)	0.16 (0.01)

Note: The variance component estimates are equal to squared standardized factor loadings for items in components so they reflect the proportion of the variance in the item response that can be attributed to the component.

Ref = Reference

Similar to willingness, the means and standard errors of the self-efficacy components show respondents have significantly lower self-efficacy when responding to “before” or “after” situations than they do for “during” situations, and they are significantly more self-efficacious when considering helping a friend or a woman, relative to a stranger or a man, respectively.

Construct Validity.

Bivariate correlations between the willingness and self-efficacy measures and validity measures are in Table 4. Alcohol bystander measures, including exposure to hazardous drinking in others, opportunity to intervene, barriers and facilitators to intervention (with barriers being negatively related), and recent use of alcohol bystander strategies were all significantly related to willingness and self-efficacy, and in the expected directions. Also as expected, we found a positive association between alcohol bystander self-efficacy and bystander self-efficacy for sexual assault situations.

Table 4.

Validation Measure Descriptives and Correlations with Willingness and Self-Efficacy Measures

Validation Measure		Willingness				Self-Efficacy
	M (SD)	Hyp	r	p	n	Hyp	r	p	n
Exposure to Hazardous Drinking
Situational Risk Signs	24.8 (10.2)	pos	.202	0.000	677	pos	.192	0.000	671
Problematic Pattern	9.8 (5.8)	pos	.165	0.000	677	pos	.191	0.000	671
Perceived Alcohol Related Bystander Opportunity	3.2 (3.1)	pos	.168	0.000	676	pos	.236	0.000	671
Barriers and Facilitators to Bystander Intervention
Barriersa	1.7 (0.8)	neg	−.334	0.000	340	neg	−.357	0.000	334
Facilitators	3.0 (0.8)	pos	.344	0.000	340	pos	.252	0.000	334
Alcohol Bystander Strategies
Level 1	13.0 (7.3)	pos	.286	0.000	677	pos	.309	0.000	671
Level 2	7.7 (6.5)	pos	.286	0.000	677	pos	.301	0.000	671
Bystander Efficacy Scale (sexual assault situations)	74.2 (26.3)	N/A	N/A	N/A	N/A	pos	.233	0.004 †	149
Protective Behavioral Strategies b
Serious harm	31.8 (6.3)	pos	.407	0.000	269	pos	.244	0.000	279
Limiting/stopping drinking	21.9 (6.6)	pos	.396	0.000	269	pos	.307	0.000	279
Manner of drinking	16.4 (4.8)	pos	.352	0.000	269	pos	.182	0.002	279
Perceived Peer Norms
Descriptive norm for HED frequency	4.3 (2.4)	neg	.005	0.955	142	neg	.204	0.018 †	134
Descriptive norm for max drinks	7.8 (4.9)	neg	−.086	0.308	142	neg	.068	0.437	134
Injunctive norm for HED frequency	3.6 (2.5)	neg	−.025	0.764	142	neg	.085	0.328	134
Injunctive norm for max drinks	6.6 (4.3)	neg	−.046	0.584	142	neg	.053	0.547	134
Alcohol use and Problems
Past 3 months typical drinks per drinking dayc	3.5 (3.0)	None	−.104	0.014 †	552	None	.042	0.324	544
Past 3 months HED frequency	1.9 (2.3)	None	−.117	0.002	675	None	.077	0.047 †	670
AUDIT total	7.1 (6.9)	None	−.099	0.010 †	677	None	.079	0.040 †	671
Assertiveness
General	23.3 (5.4)	pos	.148	0.080	142	pos	.246	0.004 †	134
Social	13.0 (3.1)	pos	.124	0.141	142	pos	.152	0.080	134
Social Desirability	1.5 (1.4)	pos	.305	0.000	677	pos	.258	0.000	671
Alpha			.90				.91
Correlation between Willingness and Self-Efficacy						pos	.626	0.000	337

Notes: Hyp = Hypothesis; HED = heavy episodic drinking, defined as four or more drinks for women and five or more drinks for men; AUDIT = Alcohol Use Disorders Identification Test. Ns for each analysis vary because a subset of participants were administered each measure.

^aFor the barriers subscale the mean and standard deviation are the original scores (indicating lower likelihood of helping in the situations on the barriers subscale relative to the facilitators scale). For the correlation, the barriers items were reverse scored, allowing for easier interpretation (i.e., higher barriers were related to lower willingness and self-efficacy).

^badministered to a random half of those who reported drinking.

^cnot administered to participants who denied drinking.

^†raw value p < .05 but Holm-Bonferroni p > .05.

Bold values are significant at p < .05

pos = positive; neg = negative

In the set of drinking-related constructs the validity findings were mixed, with high positive correlations with the PBS-20 subscales as expected. Inconsistent with expectations, we found mostly null relationships with descriptive and injunctive norms. There was only one significant finding among all the correlations in Table 4 that was opposite to expectations – a positive relationship between self-efficacy and descriptive norms for heavy drinking frequency, reflecting that the more a person believes peers drink heavily, the more confident they feel to intervene as a bystander. We did not pose directional hypotheses for associations between the new measures and alcohol use, but found most were significant, with negative relationships between willingness and alcohol quantity, heavy drinking and AUDIT score, but positive relationships between self-efficacy and heavy drinking and AUDIT score.

As expected, we found significant positive associations with social desirability for both measures, but mostly nonsignificant (although positive in direction as expected) associations between the new measures and assertiveness. Finally, the two measures of willingness and self-efficacy were highly positively correlated with each other. Most differences that were statistically significant in Table 4 remained significant when the Holm-Bonferroni correction was applied (Holm, 1979).

In the interest of determining whether the two new measures account for unique variability in bystander behavior beyond the other relevant correlates, we ran four regression models with willingness and self-efficacy predicting alcohol bystander strategies (Level 1 and Level 2). We included as covariates variables that correlated with willingness and self-efficacy and in previous work correlated with Level 1 and Level 2 bystander strategies. Covariates included were exposure and opportunities (Barnett et al., 2024), barriers and facilitators (Haikalis et al., in preparation), and AUDIT score and social desirability (Meisel et al., 2024). Both willingness and self-efficacy significantly related to Level 1 and Level 2 bystander strategies after accounting for these covariates (Supplemental Tables 1 and 2). Moreover, in all four models when all covariates were entered in one step in the regression and willingness/self efficacy entered on the next, the explained variance in bystander strategies increased significantly (ps ≤ .01).

Test-retest reliability.

At the two-week re-test, the mean for willingness (n = 231) was 24.54 (SD = 7.88) – very close to the baseline mean of 25.37 (SD = 8.71), and the retest alpha was .89 (baseline was .90). The ICC for test-retest reliability was .50 (95% CI = .392–.587). For self-efficacy (n = 226) the mean at retest was 21.80 (SD = 8.38) – very close to the baseline mean of 21.99 (SD = 9.00), and the retest alpha was .89 (baseline was .91). The ICC for test-retest reliability for self-efficacy was .64 (95% CI = .555–710); the ICCs for both measures are considered moderate (Koo & Li, 2016).

Gender differences.

When comparing men, women and nonbinary gender groups, there was a significant gender difference in willingness (F[2, 674] = 4.734, p = .009). Post-hoc tests indicated that women (M = 26.27, SD = 8.64) reported significantly higher willingness than men (M = 24.34, SD = 8.70), with an effect size for the difference of Cohen’s d = .22. There were no differences between the nonbinary group and men or women on willingness. No gender differences in self-efficacy were found.

Measurement invariance by gender.

To further evaluate whether mean differences by participant gender reflected true differences in willingness, we conducted a multiple indicator, multiple cause (MIMIC) analysis (cf. Jones, 2003) using model modification indices. The analysis indicated that men showed a lower score on the general willingness factor (−0.60 standardized units) relative to women. However, this effect was offset when the hypothetical target was a male (+0.56 standardized units). Smaller offsets were also observed for prompt A (before) and prompt B (during), particularly when the target was a male stranger. Taken together, these findings suggest that men’s lower scores on the general willingness scale may underestimate their willingness to intervene in contexts where the target of intervention is another man. This evidence of partial noninvariance underscores the importance of considering both general and context-specific factors when interpreting gender differences in willingness. Figure 3 presents these findings.

Figure 3.

Measurement Invariance by Gender for Willingness

We also considered measurement invariance by gender in self-efficacy. Although there was a small (and not significantly different from null) difference in the general self-efficacy variance component in men compared to women (−0.10 standardized units), men did report greater self-efficacy to assist other men (+0.41 standardized units) and a slightly lower (but significantly non-null) self-efficacy for prompt C (after). Therefore, for both willingness and self-efficacy, we see that men are more willing and feel more capable of helping other men, and complementarily women are more willing and feel more capable of helping other women. For self-efficacy, we see little other evidence for measurement non-invariance.

Discussion

The purpose of this research was to create a measure of willingness to intervene in hazardous drinking situations and a companion measure of self-efficacy in those same situations. We derived the items from rigorous qualitative work (Haikalis et al., 2025), and created a distinctive structure consistent with measures in bystander intervention for violence research (Banyard et al., 2014). We intentionally included different situational (before, during, after drinking), relationship (friend/stranger) and gender (man/woman) item stems to provide an instrument that acknowledges and measures responses to distinctly different hypothetical situations. The panel of participants used to validate the measures was a large and nationally representative sample of young adults, and we used standard measures development methods for evaluating the measures.

The multivariate variance decomposition method determined that, given a novel set of items assessing willingness and self-efficacy in intervening in risky drinking behavior on the part of friends, strangers, men and women, the largest share of the variance in the item responses for both measures was due to general (G) responsiveness. Two points can be made about this finding. First, it supports the use of a single summary measure of willingness to help (i.e., the self-rating of the likelihood one would intervene) and self-efficacy (i.e., the self-rating of the likelihood that one’s intervention efforts would be effective in reducing harm) for hypothetical persons engaged in drinking behavior. Second, G explaining the largest share of variance in both measures means that the general tendency is the most important driver of how people respond and matters more than who the risky drinker is (friend vs. stranger, man vs. woman) or the specific scenario (e.g., before, during, or after risk detection).

After the general factor, the next most important source of shared variance in the willingness measure was the situation component, with items framed around the after scenario showing the strongest residual correlation, followed by the relationship component, with items referencing strangers showing strong coherence. For the self-efficacy measure, this pattern was reversed: the strongest residual correlation was among items referencing strangers, followed by items referencing the after scenario. These components reflect consistent patterns in how individuals differentiate their responses based on the context of the risky drinking situation, beyond their general tendency to intervene. Importantly, variance components summarize how strongly responses to items with similar contextual features cluster together, not whether people tend to respond with higher or lower willingness or confidence in those situations. That information comes from the estimated component means, which are comparable across dimensions. The component means in Tables 2 and 3 indicate that, on average, willingness and self-efficacy were lower in after situations and for strangers, compared to during situations and for friends, respectively. This suggests that while people’s responses were most coherent (i.e., most strongly clustered) for after and stranger scenarios, they also endorsed lower response categories on those items. This finding aligns with prior work showing that relational closeness and immediacy shape the perceived feasibility and urgency of bystander intervention (Palmer et al., 2016).

Validity and Reliability

When the new Willingness and Self-Efficacy measures were compared to other bystander intervention measures, all the hypothesized correlations were supported. Specifically, the more one was exposed to the hazardous drinking of others, the higher their willingness to intervene and the higher their self-efficacy. Positive associations also were found for perceived opportunity and the use of bystander strategies with both willingness and self-efficacy, suggesting that the more opportunity individuals have to intervene and the more they report using bystander strategies, the more willing and confident they feel about acting. For barriers and facilitators, as expected, the more a person perceived barriers to intervention, the lower their willingness and self-efficacy, whereas the more they perceived facilitators to intervention, the higher their willingness and self-efficacy. Both the pattern of expected findings and the differential directionality for barriers and facilitators provide strong evidence that the bystander measures (i.e., Willingness/Self-Efficacy and Barriers/Facilitators) relate to one another in theoretically predicted ways. We also showed that bystander efficacy in sexual assault situations was positively related to self-efficacy in alcohol situations, also supporting the construct validity of the new measures.

We found mixed support for hypotheses when we correlated the Willingness and Self-Efficacy measures with measures of alcohol use-related behavior. As expected, the more individuals used protective behavioral strategies for their own alcohol use, the more willing and self-efficacious they felt to intervene with others. We expected that (descriptive and injunctive) normative perceptions would be related to willingness and self-efficacy but found nonsignificant correlations for most of these comparisons. In other words, the perceptions of heavy drinking and perceived approval of heavy drinking among peers are not meaningfully related to whether a person is willing or confident to act as a bystander in alcohol risk situations. It’s likely that the norms for alcohol use are not as relevant for our purposes as perceived norms of bystander intervention (not measured here), which we would expect would be positively related to willingness and self-efficacy. One finding in the set of norms measures was contrary to expectations: the perception of heavy drinking frequency among peers was positively related to alcohol bystander self-efficacy. It appears that the more common one perceives heavy drinking to be among peers, the more capable one feels in doing something to help others in these situations, even though willingness to intervene was unrelated to those perceptions. This finding may be explained by our sample inclusion criteria; participants were required to have some exposure/experience to seeing others intoxicated, and so may have more experience helping others, which supports greater confidence helping in the future (indeed, we found a positive relationship between exposure and self-efficacy). We might see different relationship between perception of peer drinking and self-efficacy among individuals who have less exposure to risky drinking among others.

We posed no directional hypotheses as to the relationship between alcohol use/problems and willingness/self-efficacy but found consistent negative relationships for willingness and positive relationships with self-efficacy. Thus, the higher a person’s drinking quantity, heavy drinking frequency, and AUDIT score, the lower their willingness to intervene, but the higher their self-efficacy (for heavy drinking and AUDIT). In other words, being more experienced with alcohol may make a person less inclined to act to help others but also feel more capable of doing so. This interpretation that one’s own experience with alcohol may make one more confident to help others is consistent with our finding of the positive relationship between self-efficacy and perceived heavy drinking in peers; that is, the more a person drinks and the more they think others drink, the more confident they are that they will be effective helping others.

General and social assertiveness showed correlations in the hypothesized direction, but only the relationship between general assertiveness and self-efficacy was significant. It is possible that the lower number of observations available for these analyses (ns = 134 to 142) is a reason the correlation did not reach significance. Consistent with other bystander work (Banyard et al., 2014; Brown & Messman-Moore, 2010), social desirability was significantly positively related to both willingness and self-efficacy. Given the highly significant positive correlation, future research on alcohol bystander willingness and self-efficacy should consider controlling for social desirability.

Additional regression analyses in which significant correlates were included together also support that willingness and self-efficacy contribute uniquely to the prediction of bystander behavior used in different types of alcohol risk situations. When prior exposure, barriers and facilitators, AUDIT score, and social desirability were included, willingness and self-efficacy remained significantly related to the two types of bystander behavior measured, with the explained variance in bystander strategies increasing significantly once willingness and self-efficacy were included. The consistency of these findings is a strength of this research and, while not providing proof of predictive validity, are compelling that willingness and self-efficacy are highly relevant predictors of behavior.

Finally, test/retest reliability statistics showed stability in the means and alphas and moderate ICCs for both measures, indicating acceptable across-time stability of the two measures.

Gender

Two findings of note related to gender are that, consistent with other research, for both measures, respondents indicated they were more willing and more confident in their ability to intervene when the target was a woman. We also established that compared to men, women demonstrated a greater willingness to help others, but no gender differences were found on self-efficacy. Prior research has shown similar gender differences on bystander intention (Hackman et al., 2022) and willingness (Walker et al., 2024), although this finding is not ubiquitous (Park et al., 2023). However, our measurement invariance analysis suggests that observed gender differences on the general willingness factor should be interpreted with caution. Men appear less willing on the general scale, but the measurement invariance analysis indicates that this is partly explained by context-specific differences that increase their willingness when the target is another man.

Limitations

We used a contractor (Qualtrics) to collect the panel data which can result in low quality data (Litman et al., 2014). We were attentive to this possibility and utilized multiple methods, including filtering based on speed of completion and missing data to exclude problem responders. We administered measures to subsets of participants to keep burden reasonable and acquire the necessary number of observations for our primary comparisons, so in some cases, this meant approximately one in six participants were administered some measures. For some comparisons, the number of observations may have affected results. Items referring to the gender of the target individual state “if the person is…a man” or “if the person is…a woman.” Using a binary gender for the target is a limitation, but the alternative of adding an item for nonbinary gender identity would increase the length of the measures for what is a small proportion of the population. We also considered using the phrase “presents as a man/woman” but the text is cumbersome, and we were concerned that this language would not be well understood by all respondents. We did not consider the possible interactions between the characteristics of the bystander, the measured circumstances and the target person. For example, one might predict that a bystander’s level of prior exposure to hazardous drinking in others would be associated with their willingness to act in specific situations, or that one’s willingness to act would depend on the interaction between their gender, the target person’s gender, and their friend status. Similarly, we didn’t measure other contextual factors that could be important, including the social context (e.g., location/number of people), and we did not measure willingness/self-efficacy to use different types of bystander strategies in different situations, all of which would make for considerably more complex and potentially burdensome measurement. Future research could consider these possible conditions as well as study person x situation differences, possibly using prospective and in-the-moment observations such as ecological momentary assessment. Finally, we did not measure the predictive validity of these new measures.

Conclusion and Future Directions

It is clear from this and other work (McMahon et al., 2017; Woods et al., 2020) that individuals are more or less likely to feel ready and efficacious depending on the situational circumstances, defined herein as the type of the observed risk and the characteristics of the person showing the risk. These new measures of willingness and self-efficacy should assist in improving our understanding of potential bystander opportunities, naturalistic bystander intervention, and interactions between bystander characteristics and intervention scenarios. The findings of our analyses indicate that the items can be summed for a total score to reflect readiness and self-efficacy and could be used as well to understand and potentially target specific areas for change consistent with the structure of the measures (e.g., willingness or self-efficacy in “after” situations).

Advancing the science and practice of alcohol-related BI training programs necessitates the availability of valid assessments that are well-matched to the putative mechanisms and outcomes of program effects. These new measures can support the further development of bystander intervention training for alcohol risk by helping determine if bystander interventions increase willingness and self-efficacy and whether changes in these constructs translate into bystander behavior. Program outcomes will be more accurately evaluated, instilling greater confidence in findings, ensuring greater replicability and generalizability, and resulting in more rapid dissemination of empirically supported programming.

Public Health Significance.

This study developed and tested two new tools to measure young adults’ willingness and confidence (self-efficacy) to step in during alcohol-related risk situations involving others. Results showed the tools are reliable and valid, with overall willingness and self-efficacy being the strongest factor, and that these tools can be useful for evaluating bystander intervention programs.