The identifiable victim effect: a meta-analytic review

Abstract

The identifiable victim effect (IVE) refers to individuals’ tendency to offer greater help to specific, identifiable victims than to anonymous, statistical victims. A random-effects meta-analysis was conducted to determine the overall weighted effect of IVE. Overall, 41 studies were included. Results indicated an overall significant yet modest IVE (r = .05). In addition, findings showed that IVE appears reliable mainly when there is a single identified or a single unidentified victim, and/or when study characteristics include elements of the following: a photographed child suffering from poverty, bearing little responsibility for the need, and/or associated with monetary requests. The implications of the findings and directions for future research are discussed.

Keywords:

• Identifiable victim effect
• willingness to contribute
• helping
• meta-analysis

View correction statement:

Corrigendum

The identifiable victim effect (IVE) refers to individuals’ tendency to offer greater aid to specific, identifiable victims than to anonymous, statistical victims (Jenni & Loewenstein, 1997; Small, Loewenstein, & Slovic, 2007). Schelling (1968) first discussed this effect in his seminal economic analysis of the valuation of human life, noting that in almost all cases an individual life described in detail elicits more emotional reactions and aid than an equivalent life described as a statistic. In one famous real-world example, when 18-month-old Jessica McClure (“Baby Jessica”) fell down a narrow well in Texas in 1987 and her plight was famously reported by the media until her rescue 2 days after her fall, people donated hundreds and thousands of dollars to her family for the rescue effort (Variety, 1989). Although the story of Baby Jessica was heart wrenching, the reality is that millions of other less publicized children under the age of 5 die every year from hunger or disease (World of Children Award, 2015). Clearly, the outpouring of generosity to Baby Jessica seems to be the exception rather than the rule.

Studies over the past decade have supported Schelling’s notion of the different reactions to identified vs. unidentified lives, demonstrating that people are more willing to help an identifiable victim than an anonymous or a statistical victim. This effect holds even when the identification occurs via minimal or trivial information, such as age, name, or picture (Friedrich & McGuire, 2010; Kogut & Ritov, 2005a, 2005b; Small et al., 2007). For example, Kogut and Ritov (2005a) found that when asked to help sick children who need a costly life-saving treatment, participants were more willing to contribute to a single child identified by age, name, and picture than to a single unidentified child or a group of unidentified children.

Although a considerable body of research has shown evidence supporting IVE, the results have not always been consistent when IVE tests were conducted in different contexts. Some studies have found little or no support for IVE, and some have even found effects in the opposite direction (e.g., Dickert, Kleber, Peters, & Slovic, 2011; Study 2, group; Ein-Gar & Levontin, 2013; Study 2, high temporal distance; Kogut, 2011, Study 1, responsible). The reason for the heterogeneity among studies remains unclear, but several potential factors might explain some of the variability. For example, studies often vary depending on the number of identified victims (a single victim vs. a group of victims), the information used to identify victims (e.g., age, name, picture), the cause of victims’ plight (poverty vs. disease/injury), or whether the victims belonged to the same social group as the respondents (in-group vs. out-group).

In addition, studies have considered the different psychological mechanisms such as sympathy, distress, perceived impact, or perceived responsibility that may cause IVE (e.g., Erlandsson, Björklund, & Bäckström, 2015; Friedrich & McGuire, 2010; Ritov & Kogut, 2011). Some studies found that sympathy, distress, perceived impact, or perceived responsibility causes IVE (e.g., Erlandsson et al., 2015; Study 1), while other studies failed to find support these same factors as IVE determinants (e.g., Ritov & Kogut, 2011). Thus, there is no consensus on the psychological mechanisms that bring about IVE.

In an attempt to account for the varied results in relation to studying IVE, a meta-analysis was conducted of the research literature in this area of inquiry. Specifically, the purpose of the current analysis is threefold: (a) to assess the average weighted effect size of IVE and (b) to examine the boundary conditions under which IVE works, and (c) to identify the psychological mechanisms that cause IVE. Before describing the meta-analytic procedures to address the study purposes, a brief overview of explanation for IVE is required.

Theoretical accounts of the IVE

Emotional reactions

Emotional reactions to identified victims have been suggested to explain why IVE works in increasing compliance (Slovic, 2007; Small et al., 2007). According to the affect heuristic proposed by Slovic and colleagues, individuals tend to depend on affective feelings rather than analytical considerations in many judgments and decisions (Slovic, Finucane, Peters, & MacGregor, 2002). In particular, affective feelings are germane to explaining decision-making in helping situations such as those relevant to IVE (Batson, 2011; Slovic, 2007). Identified victims may evoke more powerful emotional responses than do statistical victims and these responses engender greater likeliness to help (Small & Loewenstein, 2003). Schelling (1968) suggested that the death of a particular person evokes “anxiety and sentiment, guilt and awe, responsibility and religion, [but] … most of this awesomeness disappears when we deal with statistical death” (p. 142). Schelling’s suggestion was supported by several studies showing that identified victims elicit stronger sympathy and distress than do statistical victims, and these feelings significantly predict helping intentions, suggesting that feelings about the identified victims are critical determinants of IVE (Erlandsson et al., 2015; Studies 1 & 4; Kogut & Ritov, 2005a; Study 2; 2005b, Experiment 3).

Slovic (2007) described a collapse model of people’s affective feelings with respect to the number of victims. Slovic’s model indicates that the level of attention or imagery victims evoke drives affective feelings, and that these feelings are greatest in the case of one identified victim but diminish for two identified victims and collapse for a large number of victims who become a statistic. Identified victims, by drawing attention and generating imagery, produce emotional reactions that make people less likely to experience the collapses of emotion or feeling that occur for statistical victims.

Perceived impact

Perceived impact also has been proposed as an underlying mechanism for IVE (Duncan, 2004; Jenni & Loewenstein, 1997). People are more inclined to help victims when the rescue proportion is higher. For example, one might be more willing to help when the victims are part of a small (e.g., 10 victims out of 100 in need) rather than a large (e.g., 10 victims out of 1000 in need) reference group because the former is perceived as having a greater impact than the latter, even though the actual number of victims to be saved is equal. This has also been described as a kind of proportion dominance effect (Baron, 1997; Bartels, 2006; Fetherstonhaugh, Slovic, Johnson, & Friedrich, 1997; Friedrich et al., 1999). Jenni and Loewenstein (1997) first tested this notion in an effort to explain IVE and found that compared with larger numbers of unidentified victims, identified victims elicited more helping motivation, suggesting that one cause for IVE is the reference group’s relative size. Because the identified victims constitute their own reference group, individuals may perceive helping them as helping 100% of the relevant group; that is, people may believe their contribution will have a greater impact on an identified victim than on a large group of unidentified victims.

Perceived responsibility

Perceived responsibility has also been discussed in relation to IVE (Basil, Ridgway, & Basil, 2006) and is thought to be affected by situational circumstances (Brewer & Gardner, 1996; Dovidio, Piliavin, Gaertner, Schroeder, & Clark, 1991; Dovidio et al., 1997). Individuals tend to feel a greater responsibility to help others who are psychologically closer to them (Kogut & Ritov, 2011). Small (2015) noted that one reason for IVE is that identification with the victim reduces psychological distance (i.e., the connection that people feel with identified victims). An identified victim may be perceived as closer and more relatable than a statistical victim at the same physical distance, thus making people feel more responsibility and help more. Supporting this notion, Erlandsson and colleagues (2015) found that identified victims evoked more perceived responsibility and greater helping motivation than do statistical victims.

The current study

The current study examines the average weighted effect size of IVE across studies in the scientific literature and investigates several potential moderators to account for variation in effect sizes via meta-analysis. The present research promises to identify the degree of heterogeneity around the mean estimate of IVE in an effort to identify study factors that may account for the variability in IVE findings and help explain why IVE works to increase helping among targeted individuals. The next section will describe methods used to conduct the analyses.

Method

Search strategies and inclusion criteria

Two methods were used to search the research literature on IVE with journal articles, book chapters, dissertations, and conference papers reviewed for study inclusion. First, we conducted a search in the databases Communication & Mass Media Complete, PsycINFO, PubMed, Social Science Citation Index, Google Scholar, and ProQuest Dissertations and Theses for the 1968–2015 time period using the search terms “IVE,” “effect of identifying victims,” “identifiable and statistical victims,” “identified and unidentified victims,” and “identifiability.” Second, the present study performed an ancestry search of relevant articles by reviewing reference sections of articles found through the first method (see Cooper, 1989 for procedures).

All studies identified were carefully considered for this review. For a study to be included, it had to meet the following inclusion criteria: each study had to (1) have a treatment condition that presented an identified victim through personally identifiable information such as age, name, picture, or personal narrative (i.e., an identified condition), (2) have a control condition that presented anonymous or statistical victims (i.e., an unidentified condition), (3) have the target presented in the conditions to be a victim who had suffered from poverty, disease, disaster, or other serious harm, (4) utilize a between-subjects design having two or more groups of subjects, (5) employ a measure of intention to contribute money or time, or a measure of actual contribution of money as the dependent variable, and (6) report sufficient statistical information to calculate effect sizes across relevant studies.

Of the 21 relevant articles (19 published articles and 2 unpublished dissertations or theses), three articles were excluded because the target presented was not a victim (e.g., charitable organization or estimators seeking professional advice, Cryder, Loewenstein, & Scheines, 2013; Sah & Loewenstein, 2012; Small & Loewenstein, 2003). Two articles and one unpublished master’s thesis were excluded because they used an alternate dependent variable (e.g., frequency of compliance, Genevsky, Västfjäll, Slovic, & Knutson, 2013; Jenni & Loewenstein, 1997; Willis, 2008); one article was excluded because it utilized a within-subjects design (Kogut & Beyth-Marom, 2008); and one unpublished doctoral dissertation (Small, 2004) was excluded because its results were identical to those reported in one published article (Small et al., 2007). In total, there were 41 studies from 13 articles included in this meta-analysis.

Coding for moderators

Systematic differences in contextual and methodological characteristics of studies may result in variation in reported effects (Bijmolt & Pieters, 2001; Sutton, Abrams, Jones, Sheldon, & Song, 2000). Therefore, the current study examined 15 potential moderators related to contextual and methodological factors. The contextual factors are related mainly to the characteristics of victims and request messages, and the methodological factors are related primarily to the characteristics of respondents and dependent variables. The moderators are reviewed along with the rationale for their inclusion (see Table 1).

Table 1. Effect sizes and moderators associated with the IVE.

Study	r	N	a	b	c	d	e	f	g	h	i	j	k	l	m	n	o
Cryder and Loewenstein (2012; Study 2, determined-pictures)	–.003	169	1	2	1	2	1	2	1	2	2	2	3	2	1	1	1
Dickert et al. (2011, Study 2, group)	–.006	84	1	4	2	2	1	2	1	2	2	2	3	2	1	1	2
Dickert et al. (2011, Study 2, single)	.290	84	1	4	1	1	1	2	1	2	2	2	3	2	1	1	2
Ein-Gar and Levontin (2013, Study 2, high temporal distance)	–.013	136	1	1	1	2	2	1	1	2	2	2	2	2	2	1	2
Ein-Gar and Levontin (2013, Study 2, low temporal distance)	.167	136	1	1	1	2	2	1	1	2	1	2	2	2	2	1	2
Erlandsson et al. (2015, Study 4, IVE-ad)	.027	432	1	2	1	2	1	1	2	2	2	2	1	1	1	1	2
Friedrich and McGuire (2010; high rational scorers)	–.021	44	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Friedrich and McGuire (2010; low rational scorers)	.372	36	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Kogut (2011, Study 1, not responsible)	.179	62	2	1	1	2	1	1	2	1	2	2	3	1	1	1	2
Kogut (2011, Study 1, responsible)	–.198	51	2	1	1	2	1	1	2	1	2	2	3	1	1	1	2
Kogut (2011, Study 2)	–.091	52	2	1	1	2	1	1	2	1	2	2	3	1	1	1	2
Kogut (2011, Study 4, just world for others)	–.302	47	2	2	1	2	2	1	1	1	1	2	1	1	1	1	2
Kogut (2011, Study 4, unjust world for others)	.257	47	2	2	1	2	2	1	1	1	1	2	1	1	1	1	2
Kogut and Kogut (2013, Study 1)	.175	79	1	1	1	1	1	1	2	2	1	2	3	1	1	1	2
Kogut and Kogut (2013, Study 2)	.193	112	1	1	1	1	1	1	2	2	1	2	3	1	1	1	2
Kogut and Ritov (2005a, Study 1, group; age, name, & picture)	–.007	62	1	3	2	2	1	1	2	2	1	1	1	1	1	1	2
Kogut and Ritov (2005a, Study 1, single; age, name, & picture)	.263	62	1	3	1	1	1	1	2	2	1	1	1	1	1	1	2
Kogut and Ritov (2005a, Study 2, group)	–.209	56	1	3	2	2	1	1	2	2	1	1	1	1	1	1	2
Kogut and Ritov (2005a, Study 2, single)	.129	56	1	3	1	1	1	1	2	2	1	1	1	1	1	1	2
Kogut and Ritov (2005b, Experiment 1, group)	–.236	78	1	3	2	2	1	1	2	2	1	1	3	1	1	1	2
Kogut and Ritov (2005b, Experiment 1, single)	.270	75	1	3	1	1	1	1	2	2	1	1	3	1	1	1	2
Kogut and Ritov (2005b, Experiment 3, group; needy)	–.190	56	1	3	2	2	1	1	1	2	2	1	1	1	1	1	2
Kogut and Ritov (2005b, Experiment 3, single; needy)	.255	56	1	3	1	1	1	1	1	2	2	1	1	1	1	1	2
Kogut and Ritov (2007, Experiment 1, group; Indian names)	–.258	58	2	4	2	2	2	2	2	2	2	2	3	1	1	1	2
Kogut and Ritov (2007, Experiment 1, group; Israeli names)	–.071	56	2	4	2	2	2	1	2	2	2	2	3	1	1	1	2
Kogut and Ritov (2007, Experiment 1, single; Indian names)	–.036	61	2	4	1	1	2	2	2	2	2	2	3	1	1	1	2
Kogut and Ritov (2007, Experiment 1, single; Israeli names)	.349	60	2	4	1	1	2	1	2	2	2	2	3	1	1	1	2
Lesner and Rasmussen (2014)	.004	12,802	1	3	2	2	1	2	1	2	1	2	3	2	2	2	2
Ritov and Kogut (2011, Experiment 1, different political view)	.100	90	1	1	1	1	1	1	2	2	1	2	1	1	1	1	2
Ritov and Kogut (2011, Experiment 1, same political view)	–.193	109	1	1	1	1	1	1	2	2	1	2	1	1	1	1	2
Ritov and Kogut (2011, Experiment 3, non-rival; rival team)	–.240	42	2	1	1	1	2	1	2	2	2	2	1	1	1	1	2
Ritov and Kogut (2011, Experiment 3, non-rival; same team)	.270	32	2	1	1	1	2	1	2	2	2	2	1	1	1	1	2
Ritov and Kogut (2011, Experiment 3, rival; rival team)	.211	46	2	1	1	1	2	1	2	2	2	2	1	1	1	1	2
Ritov and Kogut (2011, Experiment 3, rival; same team)	–.179	51	2	1	1	1	2	1	2	2	2	2	1	1	1	1	2
Small et al. (2007, Study 1, intervention)	.027	59	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Small et al. (2007, Study 1, no intervention)	.406	60	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Small et al. (2007, Study 2, less frame)	–.016	50	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Small et al. (2007, Study 2, more frame)	–.040	49	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Small et al. (2007, Study 3)	.292	106	1	2	1	2	1	2	1	2	2	2	3	1	1	2	1
Small et al. (2007, Study 4, calculate)	–.097	82	1	2	1	2	1	2	1	2	2	2	3	1	1	1	1
Small et al. (2007, Study 4, feel)	.231	82	1	2	1	2	1	2	1	2	2	2	3	1	1	1	1

Note: a, Age of victims (1 = Children, 2 = Adults); b, Gender of identified victims (1 = Male only, 2 = Female only, 3 = Both male and female, 4 = NA); c, Number of identified victims (1 = A single identified victim, 2 = A group of identified victims); d, Number of unidentified victims (1 = A single unidentified victim, 2 = A group of unidentified victims); e, Type of identifying information (1 = Picture, 2 = No Picture); f, Perceived group belonging (1 = In-group, 2 = Out-group); g, Cause of plight (1 = Poverty, 2 = Disease/injury); h, Responsibility for plight (1 = Responsibility, 2 = No responsibility); i, Information about urgency information (1 = Yes, 2 = No); j, Fundraising monetary end goal (1 = Yes, 2 = No); k, Dependent variable (1 = Intention to donate money, 2 = Intention to donate time, 3 = Actual contribution of money); l, Type of Sample (1 = Students, 2 = Non-students); m, Approach to sample (1 = Offline, 2 = Online); n, Type of study (1 = Laboratory experiment, 2 = Field study), o, County (1 = United States, 2 = Outside of the United States).

Age of victims

Studies were coded to indicate whether victims described in each study were children or adults.

Gender of identified victims

Studies were coded to determine whether the gender of identified victims was male only, female only, or both male and female.

Number of identified victims

Studies were coded to indicate whether the number of victims described by the identifying information in the identified condition was a single victim or a group of victims (i.e., two or more). For example, Kogut and Ritov (2005a, Study 1, single) depicted one sick child by using his or her age, name, and picture, while Dickert et al. (2011, Study 2, group) illustrated a group of five children in danger of starvation by using their pictures.

Number of unidentified victims

Studies were coded to designate whether the number of victims described in the unidentified condition was a single victim or a group of victims (i.e., two or more). For example, Ritov and Kogut (2011, Experiment 1) described one sick child without providing any identifying information, whereas Small et al. (2007, Study 1) presented the statistical information on the number of victims (i.e., three million) suffering from starvation.

Type of identifying information

Studies were coded to determine whether a picture was used to identify victims along with other identifying information (e.g., age or name) in the identified condition. If a study presented a picture of victims in its identified condition, it was coded as “picture”; otherwise, it was coded as “no picture.”

Perceived group belonging

Studies were coded to indicate whether the victims belonged to the same nationality as the respondents. If the victims and the respondents had the same nationality in each study, it was coded as “in-group”; otherwise, it was coded as “out-group.”

Cause of plight

Studies were coded to indicate whether the cause of victims’ plight was poverty or disease/injury. If the victims in a given study were underprivileged or hungry children (e.g., Lesner & Rasmussen, 2014) or homeless people (e.g., Kogut, 2011; Study 4), it was coded as “poverty,” and if the victims were people suffering from an illness such as HIV/AIDS or cancer (e.g., Erlandsson et al., 2015) or injured in car accidents or natural disasters such as a tsunami (e.g., Kogut & Ritov, 2007, Experiment 1), it was coded as “disease/injury.”

Responsibility for plight

Studies were coded to determine whether the victim was perceived as responsible for his or her plight. Specifically, each study was coded as belonging to one of two categories: “responsibility” (e.g., people with HIV/AIDS; Kogut, 2011; Study 1, responsible) or “no responsibility” (e.g., underprivileged children; Ein-Gar & Levontin, 2013, Study 2).

Information about urgency situation

Studies were coded to indicate whether the urgency situation information was included in the request message. Each study was coded as belonging to one of two categories: if the request message contained information about the urgency situation, it was coded as “yes”; otherwise, it was coded as “No.” For example, Kogut and Kogut (2013) included the following information about the urgency situation in the request message: “It was imperative to collect the money as soon as possible to ensure that a bone marrow donor would be found in time to save Tom’s life (p. 654).”

Fundraising monetary end goal

Studies were coded to indicate whether the fundraising monetary end goal was included in the request messages. Each study was coded as belonging to one of two categories: if the request message contained the fundraising monetary end goal, it was coded as “yes”; otherwise, it was coded as “No.” For example, Kogut and Ritov (2005b, Experiment 1) included the following fundraising monetary end goal in the request message: “Unfortunately, this drug is extremely expensive, and unless a sum of 1500,000 shekels (about $300,000) is raised soon, it will no longer be possible to save the lives of the sick child [children] (p. 109).”

Dependent variable

Studies were coded for whether the type of dependent variable was intention to contribute money, intention to contribute time, or actual contribution of money.

Type of sample

Studies were coded for whether respondents were students recruited from a college campus or nonstudents recruited from the general population.

Approach to sample

Studies were coded based on whether experimenters approached respondents offline or online. If an experimenter approached respondents individually in public places, it was coded as “offline,” while if the experimenter approached respondents via an online survey website, it was coded as “online.”

Type of study

Studies were coded based on whether they were conducted in a laboratory or in the field.

Country

Studies were coded to indicate whether the country in which experiments were conducted was United States or outside of the United States.

Coding reliability

For all of the studies included in this review, coding for the relevant study moderators was completed by two independent coders. Perfect reliability between the two coders was achieved for all of the relevant study moderators.

Calculation of effect sizes

Effect sizes for all studies were calculated in the metric of r for ease of interpretation using formulas provided by Lipsey and Wilson (2001) all in relation to the linear model.¹ The effect size indicates the difference between the treatment condition (i.e., the identified condition) and the control condition (i.e., the unidentified condition); positive effects represent an advantage for the identified condition. All of the effect sizes were calculated based on means and standard deviations of the amount of contribution for treatment and control conditions. With respect to sample size, it was assumed that an equal number existed in each condition in cases where authors neglected to report counts per condition.

To derive effects from each study, this meta-analysis applied two general rules. First, when a study employed an experimental design comparing an identified condition with a corresponding control condition with multiple treatment arms, each effect contributed uniquely to the meta-analysis rather than collapsing the treatment arms together. For example, Small and colleagues (2007, Study 1) employed a 2 (identifiable victim vs. statistical victim) by 2 (no intervention vs. intervention) between-subjects design. Thus, two independent effects (i.e., comparisons between identifiable victim condition and statistical victim condition in both no-intervention and intervention conditions) were included for analysis. Second, when a study employed multiple identified conditions and only a single control condition, the condition representing the victims who were described with more identifying information was compared with the control condition. For example, Kogut and Ritov (2005a, Study 1) employed three identified conditions using age, name, or picture of the victims (i.e., age only; age and name; age, name, and picture). Thus, the effect representing the comparison between the identified condition in which the victims were described by age, name, and picture and the control condition was included for analysis as this was the most conventional illustration of IVE.

Analysis

Analyses were conducted using a random-effects model based on DerSimonian and Laird (1986). Each analysis was performed in the metric of Fisher’s z, with weighted averages of effect sizes and 95% confidence intervals (CIs) transformed back to the metric of r to facilitate interpretation. Heterogeneity tests were conducted using I² and Q statistics. The I² test addresses the degree of variance between studies, and indicates of 25, 50, and 75 correspond to low, moderate, and high between-group variance, respectively (Higgins, Thompson, Deeks, & Altman, 2003).

Moderator analyses were conducted with mixed-effects models. Q_B, a measure of significance for the moderators, was computed under the mixed-effects model. The effect size, variance, and 95% CI for each subgroup were also calculated under the mixed-effects model. Computations were conducted using the statistical software package Comprehensive Meta-Analysis 2.2 (Borenstein, Hedges, Higgins, & Rothstein, 2006).

Results

The results are represented in two main sections. The first section presents overall results for IVE. The second section presents results for IVE moderator. Before conducting the analyses, a funnel plot was constructed of the effect size on the x-axis vs. the standard error on the y-axis to examine the possibility of publication bias in data-sets (Card, 2012). A serious asymmetry would be expected in the funnel plot when publication bias may exist. As seen in Figure 1, the funnel plot did not show serious asymmetry. Therefore, there is little concern for publication bias in the data.

Figure 1. Funnel plot standard error by Fisher’s.

Overall effect

The overall weighted average correlation under the random-effect model for the 41 effects associated with IVE was .05, z = 2.07, p < .05, 95% CI = .003, .10. This finding indicates that the identified conditions elicited a significantly greater helping than did the control conditions. The homogeneity test (Q_T [40] = 104.65, p < .001, I² = 61.8%) represented a moderate amount of heterogeneity. Thus, moderators were investigated in an attempt to account for the variation in effects across studies.

Moderator effects

For all the study moderators, Table 2 provides a list of weighted mean effects, 95% CIs, and Q statistics for each subgroup. The results indicate that the number of identified victims was a significant moderator (Q_B [1] = 10.72, p < .001). Studies that depicted a single victim in the identified conditions resulted in a weighted mean effect size of .10 (k = 33, 95% CI = .03, .16), while studies that described a group of victims in the identified conditions resulted in a weighted mean effect size of –.09 (k = 8, 95% CI = –.19, .001). Thus, the identification of victims resulted in gains in helping when a single victim was described in the identified conditions and decreased helping when a group of victims was presented in the identified conditions. The other study moderators considered in the current meta-analysis did not moderate the relationship between the identification of victims and helping.

Table 2. Moderator analyses for the IVE.

Moderator [Q Between]	k	r	95% CI	z	p	Q Within
Overall	41	.052	[.003, .10]	2.07	.038	104.65^***
Age of victims [Q (1) = 1.38, p = .241]
Children	28	.073	[.02, .13]	2.59	.010	72.91^***
Adults	13	–.010	[–.14, .12]	–.15	.880	31.37^**
Gender of identified victims [Q (3) = .78, p = .854]
Male only	13	.034	[–.06, .13]	.69	.491	26.06^*
Female only	13	.088	[–.01, .19]	1.70	.088	31.74^**
Both male and female	9	.029	[–.08, .14]	.51	.611	22.84^**
NA	6	.052	[–.14, .24]	.54	.589	17.46^**
Number of identified victims [Q (1) = 10.72, p = .001]
A single identified victim	33	.095	[.03, .16]	3.00	.003	75.32^***
A group of identified victims	8	–.093	[–.19, .001]	–1.94	.052	12.99
Number of unidentified victims [Q (1) = 3.76, p = .053]
A single unidentified victim	15	.127	[.03, .22]	2.52	.012	33.60^**
A group of unidentified victims	26	.014	[–.04, .07]	.50	.620	58.28^***
Type of identifying information [Q (1) = .52, p = .470]
Picture	29	.065	[.01, .12]	2.25	.024	74.03^***
No picture	12	.014	[–.11, .14]	.23	.821	30.53^**
Perceived group belonging [Q (1) = .42, p = .515]
In-group	26	.038	[–.03, .11]	1.04	.300	62.82^***
Out-group	15	.075	[–.01, .16]	1.76	.079	40.21^***
Cause of plight [Q (1) = 1.25, p = .264]
Poverty	19	.084	[.01, .16]	2.24	.025	52.17^***
Disease/injury	22	.022	[–.06, .10]	.55	.580	52.22^***
Responsibility for plight [Q (1) = .65, p = .419]
Responsibility	5	–.029	[–.24, .18]	–.27	.788	11.46^*
No responsibility	36	.061	[.01, .11]	2.31	.021	92.85^***
Information about urgency situation [Q (1) = .04, p = .851]
Yes	14	.046	[–.04, .13]	1.06	.291	39.35^***
No	27	.056	[–.01, .13]	1.57	.116	61.61^***
Fundraising monetary end goal [Q (1) = .06, p = .812]
Yes	8	.036	[–.13, .19]	.43	.665	22.40^**
No	33	.056	[.003, .11]	2.06	.040	82.02^***
Dependent variable [Q (2) = 1.26, p = .532]
Intention to contribute money	15	.009	[–.09, .11]	.18	.858	33.68^**
Intention to contribute time	2	.078	[–.10, .25]	.86	.392	2.19
Actual contribution of money	24	.075	[.01, .14]	2.15	.031	67.65^***
Type of sample [Q (1) = .004, p = .952]
Students	35	.049	[–.02, .12]	1.43	.152	90.30^***
Non-students	6	.052	[–.03, .13]	1.26	.208	10.56
Approach to sample [Q (1) = .16, p = .693]
Offline	38	.053	[–.01, .12]	1.67	.096	96.01^***
Online	3	.032	[–.06, .12]	.71	.478	3.61
Type of study [Q (1) = 1.43, p = .232]
Laboratory experiment	33	.035	[–.03, .10]	1.11	.268	78.62^***
Field study	8	.126	[–.01, .26]	1.83	.068	24.23^**
County [Q (1) = 1.55, p = .213]
United States	10	.116	[–.002, .23]	1.92	.054	21.84^**
Non-United States	31	.033	[–.02, .09]	1.18	.238	76.12^***

* p < .05;

** p < .01;

*** p < .001.

Discussion

The purpose of this study was to provide a quantitative estimate of the strength of IVE and to investigate if study moderators account for the heterogeneity surrounding IVE studies. The following sections discuss analysis results and possible directions for future IVE research.

Overall effect of the IVE

The current meta-analysis summarized the effect sizes from 41 studies on IVE and found a significant overall mean effect size of r = .05. Although it is not realistic to expect a large mean effect in social influence research, this mean effect size appears to be relatively small given the enthusiasm of the field for IVE, shown through the consistent research attention paid to IVE and the many practical applications of IVE assumptions in charitable giving campaigns.

However, the observed mean effect is not trivial in practical importance (for a general discussion of effect sizes, see Rosenthal, 1994; for some specific discussion regarding the understanding of effect size magnitudes, see Abelson, 1985; Cooper, 1981; Haase, Ellis, & Ladany, 1989). From a pragmatic standpoint, the results of the current meta-analysis suggest that charitable organizations may benefit from using the strategy of identifying victims with their personal information when eliciting charitable giving. It is conceivable that identifiable victims, on average, elicit about a .10 SD more support than anonymous or statistical victims (r = .05 ~= d = .10).

Moderators of the IVE

The effect sizes across studies on IVE were heterogeneous. Thus, the current meta-analysis examined potential IVE moderators to account for this heterogeneity. Several key findings warrant discussion with respect to IVE moderators.

With respect to the number of identified victims, the current study indicates that the mean effect size is larger for studies that described a single victim in the identified conditions than for studies that depicted a group of victims in the identified conditions. In addition, the current meta-analysis demonstrates that IVE works only for a single victim, indicating that the mean effect for a single victim is significantly positive, whereas the effect for a group of victims (i.e., two or more) does not differ significantly from zero. This finding is consistent with Kogut and Ritov’s (2005a, 2005b) argument that IVE is largely confined to situations involving a single victim, which is dubbed the singularity effect. Kogut and Ritov (2005a, 2005b) claimed that the cause of the singularity effect lies in emotions evoked by observing victims under hardship, demonstrating that a single identified victim evokes stronger sympathy or distress than a group of victims, regardless of whether the group has been identified. Furthermore, it may be possible to explain the finding based on the collapse model of people’s emotional responses to the number of victims (Slovic, 2007), which indicates that people’s emotions in relation to helping victims are greatest for a single victim but begin to fade with two victims and collapse with a larger number of victims that simply becomes a statistic. Thus, the increase in emotions that observing a single victim evokes might enhance helping, whereas the decrease in emotions people feel when confronted with a group of victims might undermine helping.

Moreover, the present research implies that the number of unidentified victims was a significant moderator to explain, in part, the variation in effect sizes across studies on IVE (p = .053). Specifically, the mean effect size is larger for studies that presented a single victim in the unidentified conditions (r = .13) than for studies that presented a group of victims in the unidentified conditions (r = .01). One possible explanation for this finding concerns the proportion dominance effect (Baron, 1997; Bartels, 2006; Fetherstonhaugh et al., 1997; Friedrich et al., 1999). Unidentified victims might be seen as part of the relevant group in need (i.e., the large statistical group of victims), whereas identified victims constitute their own reference group. Specifically, a single unidentified victim or a group of unidentified victims might be perceived as members of the large statistical group of victims, which might be regarded as the reference group of the unidentified victims. People might perceive that helping a single unidentified victim has a lesser impact than helping a group of unidentified victims because a single unidentified victim constitutes a lower proportion of the reference group than does a group of unidentified victims. Thus, it seems that IVE is greater when identified victims are contrasted with a single unidentified victim than with a group of unidentified victims.

Meanwhile, there are several cases in which the moderation effect was not significant, but there appear to be clear patterns in the homogeneous subgroups to show IVE boundary conditions. For example, the moderator analysis for the type of identifying information yielded p = .470—a moderation effect that is far from significant, suggesting that the type of identifying information might be a non-issue. However, when one looks at the simple effects for the homogeneous groups, it is clear that the mean effect is significant when studies present the picture of victims in the identified condition with their ages or names (r = .07, p < .05), whereas the mean effect is far from significant when studies do not present the picture of victims in the identified condition but reveal their ages or names (r = .01, p = .821). Therefore, it seems worth highlighting that the only robust effect appears to involve studies in which a picture is used to identify victims. This finding is consistent with Kogut and Ritov’s (2005a) argument that presenting the picture of victims in addition to their names or ages further reinforces the perception of identified victims and in turn evokes greater emotional reactions and likelihood of helping.

In addition, a similar pattern of a dominant subgroup of studies appears for several other moderator variables. Specifically, the mean effect size approaches statistical significance (p < .05) only when victims were children (r = .07), only when the victims’ source of need was poverty (r = .08), only when the victims were seen as not responsible for their plight (r = .06), only when monetary end goals were not included (r = .06), and only when actual monetary contributions were included (r = .08). Based on these homogeneous subgroups of studies that did yield appreciable mean effect sizes, one could argue that IVE appears reliable mainly when there is a single identified or a single unidentified victim, and/or when study characteristics include elements of the following: a photographed child suffering from poverty, bearing little responsibility for the need, and/or associated with monetary requests. This argument for IVE boundary conditions would seem to hint at directions for future theorizing and empirical work.

Meanwhile, another notable finding in the current meta-analysis that warrants discussion is that the perceived group belonging was not a significant moderator of IVE. Specifically, the mean effects did not differ between the in-group condition in which the victims belonged to the same nationality as the respondents and the out-group condition in which the victims belonged to a nationality different from that of the respondents. People are more likely to help others who belong to their own social group because they feel greater closeness and responsibility and perceive stronger emotions (Brewer & Gardner, 1996; Dovidio et al., 1991, 1997). However, the results from the current meta-analysis indicate that this preference might not apply to the identified victims in a situation that distinguishes social groups by nationality. That is, people seem to consistently help the identified victims regardless of whether the identified victims are of their nationality.

Finally, a remarkable finding from the current meta-analysis is IVE’s magnitude. The present research indicates that even under the most favorable conditions, the mean effect size is rarely much larger. In fact, only four conditions of any size yielded mean effect sizes near or greater than .10: a single identified victim (r = .10), a single unidentified victim (r = .13), field study (r = .13), and United States (r = .12). That is to say that even under the most favorable of moderator variable conditions, the mean IVE effect sizes appear to remain quite modest. This is one of the more interesting lessons of this study and an interesting consideration for researchers in this area. Stated plainly, the IVE effect is small.

Directions of future research

Although the current study examined a number of IVE moderators, future research must consider other factors that would broaden our understanding of IVE boundary conditions and causes. For example, given that people recognize the emotions a victim’s face displayed in pictures (Small & Verrochi, 2009), the victim’s facial expression of emotion (e.g., happy, sad, or neutral emotional expression) may influence people’s emotional reactions and desire to help. In addition, the extent of a victim’s plight described in his or her personal narrative may be the cause of people’s emotional reactions and desire to help. Thus, it would be worthwhile to test these possible moderators related to emotional reactions in future research.

Finally, the present research used nationality as a determinant of perceived group belonging. However, individuals have multiple social identities derived from different perceived memberships in social groups, which might create different levels of group belonging (Kogut & Ritov, 2011). The determinants that distinguish between social categories may include various factors such as ethnicity, religion, ideology, or experimentally manipulated criteria. Because factors that determine social categories may affect IVE, future research should examine them as IVE moderators.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. To compute the r value for each study, the d value for each study was first calculated using the following formula:

and then the d value was converted to the r value using the following formula:

where a is a correction factor for cases where n₁ ≠ n₂,