Focal Concerns and Police Decision Making in Sexual Assault Cases: A Systematic Review and Meta-Analysis - ProHIC
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Review Manuscript TRAUMA, VIOLENCE, & ABUSE 1-15 Focal Concerns and Police Decision Making ª The Author(s) 2021 Article reuse guidelines: sagepub.com/journals-permissions in Sexual Assault Cases: A Systematic DOI: 10.1177/1524838021991285 journals.sagepub.com/home/tva Review and Meta-Analysis David S. Lapsey Jr.1 , Bradley A. Campbell1, and Bryant T. Plumlee1 Abstract Sexual assault and case attrition at the arrest stage are serious problems in the United States. Focal concerns have increasingly been used to explain police decision making in sexual assault cases. Because of the popularity of the focal concerns perspective and potential to inform evidence-based training, a systematic review and meta-analysis are needed to condense the literature. In this study, we assess the overall strength of the relationship between focal concerns variables and police decisions to arrest in cases of sexual assault. Our assessment of the effects of focal concerns variables on arrest decision making in sexual assault cases followed the systematic review protocols provided by the Campbell Collaboration of Systematic Reviews. Specifically, we used the Campbell Collaboration recommendations to search empirical literature and used meta-analysis to evaluate the size, direction, and strength of the impact of focal concerns variables on arrest decisions. Our search strategy detected 14 eligible studies and 79 effect sizes. The meta-analysis found several robust and statistically significant correlates of arrest. In fact, each focal concerns concept produced at least one robust arrest correlate. Overall, focal concerns offers a strong approach for explaining police decisions in sexual assault cases. Although practical concerns and resource constraints produced the strongest arrest correlates, results show the importance of additional case characteristics in officers’ decision to arrest. Keywords focal concerns, police decision making, arrest predictors, sexual assault investigation, meta-analysis The decision to arrest has been considered the most important similar to sentencing decisions—police decisions in sexual discretionary phase in the criminal justice response to sexual assault cases are dependent on three conditions, including the assault cases (LaFree, 1981). Indeed, the police decision to (a) culpability or blameworthiness of the offender, (b) the per- arrest has demonstrated law enforcement’s role as ceived dangerousness of the offender, and (c) practical resource “gatekeepers” to the criminal justice system, highlighting the constraints faced by the criminal justice system. Focal concerns wide discretion to determine which cases do—and do not— research has shown that, when determining the blameworthiness move forward (Kerstetter, 1990; Spohn & Tellis, 2014). Police of an offender and a suspect’s perceived dangerousness, detailed discretion in cases of sexual violence is of particular impor- information on case and/or suspect characteristics and forensic tance because research has found the majority of sexual assault evidence are often unavailable to officers (B. Campbell et al., cases do not result in an arrest (Alderden & Ullman, 2012a). In 2015). In the absence of this evidence, officers rely on a fact, prior work has shown that less than 20% of sexual assaults “perceptual shorthand” of case characteristics based on extrale- are cleared by arrest in some jurisdictions (R. Campbell et al., gal variables, including myths and stereotypes connected to 2014; Morabito, Williams, & Pattavina, 2019). Low rates of offender, victim, and case attributes that practitioners incorrectly arrest are problematic as they have allowed large numbers of view as characteristic of a “real rape” (Estrich, 1987; O’Neal serial offenders to continue perpetrating crimes (Lovell et al., et al., 2019; O’Neal & Spohn, 2017; Venema et al., 2019). For 2017, 2020) and have resulted in denied justice for sexual example, police investigators have explained that stranger cases assault survivors (Kerstetter, 1990; Spohn et al., 2001). involving a victim who sustains injury and reports the crime to For decades, scholars have identified key suspect, victim, and case characteristics that impact the decision to arrest in cases of 1 sexual assault. While much of this work has been atheoretical, University of Louisville, KY, USA recent studies have applied the focal concerns framework to Corresponding Author: make sense of decisions made by sex crimes investigators (see David S. Lapsey Jr, University of Louisville, 2311 S 3rd St., Louisville, KY 40208, Kaiser et al., 2017; Spohn & Tellis, 2019; Venema et al., 2019; USA. Wentz, 2019). Focal concerns scholars have posited that— Email: david.lapsey@louisville.edu
2 TRAUMA, VIOLENCE, & ABUSE XX(X) police in a timely manner are more likely to move forward in the 2004), and undisclosed urban/suburban/rural locations criminal justice process (B. Campbell et al., 2015). Making (Bouffard, 2000; Morabito et al, 2019b). Studies also differed decisions based on these perceptual shorthand variables is based on sample age, where some studies used juvenile and adult problematic, as research has shown that victims often do not cases (Kaiser et al., 2017; O’Neal et al., 2019; Scott & Beaman, experience physical injury during a sexual assault and that many 2004; Smith, 2005; Spohn & Tellis, 2019) and others only ana- victims may not report promptly to law enforcement (see R. lyzed data derived from cases involving adult victims (Alderden Campbell, 2012). & Ullman, 2012a, 2012b; Wentz, 2019). Although several studies have examined police decisions to Findings from sexual assault arrest decision-making studies arrest in sexual assault cases, to date, no systematic are rather mixed. For instance, some studies have found that meta-analytic review of this literature has been conducted to cases involving intimate partner assaults, nonstranger assaults, understand which correlates—or focal concerns—are most non-White victims, victim resistance, victim injury, victim important to police decisions. Thus, it is critical to summarize cooperation, and the availability of witnesses or physical findings from prior work to identify strongest predictors of the evidence significantly increased the odds of arrest (Alderden decision to arrest in sexual assault cases. To fill this gap in the & Ullman, 2012a, 2012b; Kaiser et al., 2017; Morabito, literature, this study synthesized the empirical findings across Williams, & Pattavina, 2019b, O’Neal et al., 2019; Spohn & studies “take stock” (Cullen et al., 2006) in the applicability of Tellis, 2019; Tasca et al., 2013; Venema et al., 2019). Even so, focal concerns to police decision making in sexual assault other studies have reported either a negative or nonsignificant investigations. Specifically, this study contributes to the liter- relationship between law enforcement’s decision to arrest and ature by (a) providing a systematic review of research on police nonstranger assaults, victim age, non-White victims, decision making in sexual assault cases; (b) performing a non-White suspects, victim injury, victim resistance, offender meta-analysis to assess the magnitude and direction of the weapon use, and the availability of witnesses and physical relationships between case, victim, and suspect characteristics evidence (Alderden & Ullman, 2012a; Bouffard, 2000; and police decision to arrest; and (c) assessing the applicability Morabito, Williams, & Pattavina, 2019b; O’Neal et al., 2019; of the focal concerns framework to police decision making in Spohn & Tellis, 2019; Wentz, 2019). Based on these mixed these cases. findings, it is important to assess which variables are the stron- gest predictors of arrest and to place these findings in the context of the focal concerns framework to understand why Police Decision Making in Sexual these relationships arise in the literature. Assault Cases Practitioners and advocates have urged for improved responses to sexual assault for decades (Field, 1978). Because of such Focal Concerns Perspective efforts, researchers have since conducted studies directed at Although the majority of scholarship examining police arrest identifying correlates of law enforcement’s decision to arrest decisions in sexual assault cases has been atheoretical, in the and reduce sexual assault case attrition (Bouffard, 2000; Kaiser last decade, scholars have begun to apply focal concerns to et al., 2017; Lafree, 1981). To study this topic, scholars have understand police decision making. In its original form, focal relied primarily on samples of sexual assault reports collected concerns was developed by Steffensmeier and colleagues from police departments. Using these sexual assault reports, (1998) to make sense of judges’ sentencing decisions; however, researchers collect data by reviewing case files and coding for the framework has been adapted to explain decision making at relevant variables linked to decision-making outcomes (e.g., several stages of the criminal justice process (Hartley et al., victim and suspect demographics, victim–offender relation- 2007). The focal concerns framework borrowed from ship, criminal histories, drug/substance use, evidence availabil- Albonneti’s (1986, 1987, 1991) finding that prosecutorial ity, weapon used, victim report time, victim injury, victim decisions were guided by a “bounded rationality” that aimed verbal and physical resistance, suspect use of force). Of such to reduce uncertainty in obtaining a guilty verdict or plea bar- studies, most sampled cases occurred prior to 2010 (Alderden gain in criminal cases. Based on this idea, focal concerns & Ullman, 2012a, 2012b; Bouffard, 2000; Kaiser et al., 2017; initially incorporated three concepts to explain judicial O’Neal et al., 2019; Scott & Beaman, 2004; Smith, 2005; decision making, including the blameworthiness of the offen- Spohn & Tellis, 2019; Tasca et al., 2013), whereas fewer sam- der, the perceived dangerousness of an offender, and practical ples cases (see Morabito, Williams, & Pattavina, 2019; constraints faced by decision makers in the criminal justice Venema et al., 2019; Wentz, 2019; Wentz & Keimig, 2019; system. In addition to these three concepts, focal concerns Ylang & Holtfreter, 2019). Additionally, a large portion of explains that judges make decisions with limited knowledge studies used samples from Los Angeles (Kaiser et al., 2017; regarding an offender’s history and character. When this infor- O’Neal et al., 2019; Spohn & Tellis, 2019; Ylang & Holtfreter, mation is unavailable to fully inform these decisions, judges 2019) and Midwestern cities (Alderden & Ullman, 2012a, often make decisions using a “perceptual shorthand” that relies 2012b; Venema et al., 2019; Wentz, 2019; Wentz & Keimig, on an assessment of extralegal variables (e.g., offender gender, 2019), while others sampled cases from Arizona (Tasca et al., race, age) to determine an offender’s perceived culpability and 2013), Maryland (Smith, 2005), Canada (Scott & Beaman, dangerousness (Steffensmeier et al., 1998).
Lapsey et al. 3 Focal concerns scholars have expanded the framework to 2019; Smith, 2005) were not significantly correlated with explain prosecutorial (Beichner & Spohn, 2012; Spohn et al., arrest. 2001) and police decision making (Crow & Adrion, 2011; Johnson et al., 2015; Spohn et al., 2014; Tillyer & Hartley, 2010). Unlike prosecutors, who are guided primarily by eva- Protection of the Community luations of “convictability” (e.g., likelihood of conviction), O’Neal and Spohn measured the need for protection of the police make decisions to arrest largely based on their percep- community from sexual assault offenders by indicating whether tions of the case characteristics needed for prosecutors to the suspect used—or threatened to use—a weapon to during the obtain a plea bargain or guilty verdict at trial (B. Campbell, attack. Several studies found weapon use significantly 2015; Spohn et al., 2014; Tillyer & Hartley, 2010). It follows increased the odds of arrest (Morabito, Williams, & Pattavina, that police officers are more likely to make an arrest when 2019b; Spohn & Tellis, 2019; Venema et al., 2019). Conver- they believe a prosecutor will accept charges. Conversely, sely, other studies did not detect a significant relationship when officers believe a prosecutor will decline charges, the between weapon use and arrest decisions (Kaiser et al., 2017; likelihood of arrest is reduced, and case attrition is likely. Scott & Beaman, 2004; Ylang & Holtfreter, 2019). Over the past 5 years, scholars have begun to operationalize common correlates of police decision making in sexual assault cases into clearer measures of the focal concerns framework Practical Constraints (Kaiser et al., 2017; Spohn & Tellis, 2019; Venema et al., 2019; Practical constraints were operationalized by O’Neal and Wentz, 2019, Ylang & Holtfreter, 2019). Specifically, in stud- Spohn by coding cases to indicate whether (a) physical ies of sexual assault investigations, researchers have used focal evidence was available, (b) the suspect was interviewed by the concerns to understand how suspect blameworthiness (e.g., police, (c) law enforcement believed the victim was cooperat- victim injury), the need for protection of the community ing with police, (d) a witness to the assault was available, and (e.g., suspect weapon use), practical constraints (e.g., physical (e) the victim reported promptly. Some studies have shown that evidence available), and perceptual shorthand (e.g., victim– witness availability (Kaiser et al., 2017; Morabito, Williams, & offender relationship, victim credibility) work to shape police Pattavina, 2019; O’Neal et al., 2019; Spohn & Tellis, 2019), decision making (O’Neal & Spohn, 2017; Spohn & Tellis, physical evidence availability (Kaiser et al., 2017; Morabito, 2019). Although some prior research used focal concerns to Williams, & Pattavina, 2019), prompt reporting (Kaiser et al., explain police decisions (see Crow & Adrion, 2011; Higgins 2017; Morabito, Williams, & Pattavina, 2019; O’Neal et al., et al., 2012), O’Neal and Spohn (2017) were the first to expli- 2019; Spohn & Tellis, 2019), and perceived victim cooperation citly operationalize and measure focal concerns variables in (Alderden & Ullman, 2012a; Morabito, Williams, & Pattavina, regard to police decisions to arrest in sexual assault cases. 2019; O’Neal et al., 2019) were significantly correlated with However, studies have examined the impact of several arrest. Conversely, others find a nonsignificant correlation variables included in O’Neal and Spohn’s operationalization between witness availability (Alderden & Ullman, 2012a; of focal concerns on police decisions to arrest. Wentz, 2019), prompt reporting (Tasca et al., 2013), and phys- ical evidence availability (O’Neal et al., 2019) and arrest. Suspect Blameworthiness In their study of intimate partner sexual assaults, O’Neal and Perceptual Shorthand Spohn measured suspect blameworthiness using indicators Finally, O’Neal and Spohn’s perceptual shorthand measures gleaned from case files regarding (a) the suspect’s history of included (a) victim and suspect race, (b) victim and suspect sexually and physically assaulting a victim, (b) whether the consumption of alcohol before or during the assault, (c) victim– victim sustained injury from the sexual assault, (c) whether the offender relationship, (d) victim credibility (e.g., discrepancy suspect physically assaulted the victim during the incident, and noted in statement, criminal history, question about character/ (d) whether the victim reported physically or verbally resisting reputation), and (e) whether police believed the victim had a the attack. Some prior studies have found that when victims reason to fabricate the assault. Indeed, studies have found a resist their assailant (Alderden & Ullman, 2012ab; Morabito, significant correlation between arrest and victim alcohol Williams, & Pattavina, 2019), the suspect physically assaults a consumption (O’Neal et al., 2019; Venema et al., 2019), victim victim (O’Neal et al., 2019; Spohn & Tellis, 2019), and when credibility (Alderden & Ullman, 2012b; Kaiser et al., 2017; the victim sustains an injury from an assault (Morabito, Wil- Morabito, Williams, & Pattavina, 2019), intimate partner rela- liams, & Pattavina, 2019b; Spohn & Tellis, 2019; Venema tionships (Alderden & Ullman, 2012b; Kaiser et al., 2017; et al., 2019), the odds of arrest are significantly increased. Morabito, Williams, & Pattavina, 2019), nonstranger relation- Others, however, have found that victim injury (Alderden & ships (Alderden & Ullman, 2012b; Morabito, Williams, & Pat- Ullman, 2012; Scott & Beaman, 2004) and physical or verbal tavina, 2019), and victim race (Morabito, Williams, & resistance (Kaiser et al., 2017; O’Neal et al., 2019; Scott & Pattavina, 2019; Venema et al., 2019). Nonetheless, other stud- Beaman, 2004; Spohn & Tellis, 2019) and suspects who phy- ies found victim credibility measures (Kaiser et al., 2017; Mor- sically assault of a victim (Morabito, Williams, & Pattavina, abito, Williams, & Pattavina, 2019; O’Neal et al., 2019; Spohn
4 TRAUMA, VIOLENCE, & ABUSE XX(X) & Tellis, 2019), victim alcohol consumption (Scott & Beaman, Sample of Studies 2004; Wentz & Keimig, 2019), suspect race (Alderden & Ull- Our sample included studies that examined law enforcement man, 2012a; Kaiser et al., 2017), victim race (Kaiser et al., officers’ decision to arrest in sexual assault cases. It is impor- 2017; Spohn & Tellis, 2019), intimate partner relationships tant to note that the review focused solely on officers’ decision (O’Neal et al., 2019; Spohn & Tellis, 2019), and nonstranger to arrest. Because law enforcement must establish probable relationships (Kaiser et al., 2017; O’Neal et al., 2019; Spohn & cause for arrest, the decision to arrest has a higher legal stan- Tellis, 2019) were not significantly correlated with arrest. dard than other relevant police decisions (e.g., present case to Given these divergent findings across studies, a prosecutor, interview suspect). Thus, arrest is categorically meta-analysis is needed to evaluate the current state of the different than other police decisions in sex crimes. As such, police sexual assault decision-making literature. The current studies examining case clearance status and law enforcement’s study is the first attempt to assess the strength and direction decision to present a case to prosecutors prior to making an of the relationships between focal concerns variables and arrest were excluded from our review. To complete our review, police decisions to arrest across studies and to determine what we included studies that: focal concerns are most important to arrest decisions in sexual assault investigations. Indeed, discerning the most important 1. included a general sample of sexual assault cases and correlates of arrest is needed to inform evidence-based policy. not only a subsample (e.g., intimate partner sexual Findings from this review will be well suited to inform law assaults); enforcement training and educational topics (e.g., rape myths 2. assessed law enforcement officers’ decision to arrest; and misconceptions, evidence processing, maintaining victim 3. quantitatively assessed arrest decision making as an engagement, interview techniques). These contributions may outcome variable; help reduce case attrition and improve the police response to 4. measured the effects of predictor variables on the deci- victims of sexual assault. sion to arrest using data derived from law enforcement case files; and 5. provided adequate statistical information to compute at Current Study least one effect size from a multivariate model. To our knowledge, the literature on police decision making in sexual assault cases has not been subjected to systematic meta-analytic review. As such, using the operationalizations Search Strategies of focal concerns proposed by O’Neal and Spohn (2017), our In April 2019, we employed an exhaustive search strategy review assessed the impact of each focal concerns variable on using the following four online databases: (1) ProQuest arrest decisions to achieve the following three goals: Dissertation and Theses, (2) Criminal Justice Abstracts, (3) Psy- 1. systematically review the existing literature on officers’ chInfo, and (4) Sociological Abstracts. Because our initial decisions to arrest in sexual assault cases; search produced a high number of irrelevant studies, we devel- 2. calculate the combined effects of case, victim, and sus- oped a more focused search command—AB (sexual assault OR pect characteristics on the police decision to arrest rape) AND (arrest) AND (deci* OR discretion) AND (law across studies; and enforcement OR police OR investigator)–that yielded articles 3. assess the applicability of the focal concerns framework specific to the review. After completing the search, we to police decision making in cases of sexual assault. reviewed references from articles and dissertations and included any relevant studies that were not detected by our key word search. Finally, an additional search of the literature was conducted in December 2019 to identify studies that were pub- Method lished after our initial search in April 2019. Our systematic meta-analytic review followed guidelines Once our sample of articles was identified, we created a established by the Campbell Collaboration of Systematic coding sheet adapted from Higginson and colleagues (2019). Reviews (Steering Group of the Campbell Collaboration, The coding sheet comprised 25 unique items and was compiled 2019). The meta-analysis summarized results across quantita- in an excel file to code each study included in the analysis. The tive studies to produce a single effect size for each focal 25-item coding sheet included 12 methodological items (e.g., concerns variable. This effect size represents an average of data source, sample location, sample frame), nine statistical effect sizes across studies, which allowed us to present the analysis items (e.g., sample size, outcome variables, predictor overall strength and direction of the average effects for each variables), and the variables measuring the four focal concerns measure of focal concerns (Borenstein et al., 2009; Haidich, categories in each study. For coding consistency during data 2010; Lipsey & Wilson, 2001; Sullivan & Feinn, 2012). As collection and to calculate interrater reliability, the decision to such, this review enabled us to “take stock” of a large body accept/reject an article was assessed by our research team. of literature and improve the precision of estimated effects Next, a random sample of five studies was independently coded (Cullen et al., 2006; Haidich, 2010; Turanovic & Pratt, 2020). by two research team members and then compared for
Lapsey et al. 5 consistency. The number of agreements was divided by the effect sizes should be derived from a single data source. When total number of coding items, which resulted in an interrater a variable was included across two (or more) studies sharing a agreement of 87%. Where discrepancies occurred, the two common data source, we only included one estimate for that team members discussed means to resolve such coding variable (Lipsey & Wilson, 2001). Our selection of variables in disagreements (Yeaton & Wortman, 1993). these cases was guided by Kochel and colleagues’ (2011) strat- egy for selecting the most robust effect size from each data set. Thus, when studies were derived from a common data sources, Meta-Analytic Review of Focal Concerns Measures we selected variables based on the following criteria: (1) The The review analyzed variables measuring the four aspects of variable was included in a logistic regression model, (2) the focal concerns, including suspect blameworthiness, community standard error was directly reported, (3) the effects were based protection, practical constraints, and perceptual shorthand on the full sample size, (4) models clearly operationalized the variables. To examine these concepts, we adapted O’Neal and dependent variable as arrest versus no arrest, (5) the variable Spohn’s (2017) operationalization of each focal concern. First, was included in a model with the largest number of indepen- suspect blameworthiness was operationalized using victim dent variables, and (6) the statistical power of the model. resistance (i.e., victim physically or verbally resisted the assault) and victim injury (i.e., physical injuries were noted on the victim). Second, protection of the community included Statistical Procedures and Effect Size Estimates whether the suspect used or attempted to use a weapon during To account for methodological and measurement differences the assault. Third, physical evidence, prompt report, wit- across studies, our meta-analysis used followed recently ness(es), and victim cooperation were included as measures published studies and used random effects modeling (Fitton of practical constraints. Physical evidence included any vari- et al., 2020; Spencer & Stith, 2020). Rather than assume effect able measuring indicating the presence of physical evidence in size estimates between studies were attributed to random error, the case (DNA, fingerprints, and clothing). Prompt report a random-effects model anticipates methodological differences included any case in which the victim reported the incident between studies that might influence effect sizes. to an officer or medical professional within 72 hr of the assault, A random-effects model, unlike a fixed-effects model, does and victim cooperation indicated that law enforcement not produce a true effect size. Instead, the model produces an perceived the victim was engaged with the investigation. estimate of the effect size distribution across studies. In this Finally, perceptual shorthand was measured using victim– way, we ensured all studies were represented in the combined offender relationship (nonstranger and intimate partner rela- estimate. In addition, using random-effects models allowed for tionships), victim and suspect age, victim and suspect race, and generalizability of our findings (Card, 2012). We used the victim credibility. Because a large number of studies measured inverse-variance weight method (see Fleiss & Berlin, 2009; credibility uniquely, a composite measure was created to Hedges & Olkin, 1985; Lipsey & Wilson, 2001; Shadish & analyze the combined effects of victim credibility on arrest Haddock, 2009). This approach gives greater weight to larger decision making. Specifically, we measured victim credibility samples and takes the size of standard errors into account, using widely accepted definitions for credibility issues, includ- which helps produce more precise effect sizes. ing victim drug/alcohol use before or during the incident, Analyses were conducted in RStudio’s metafor package history of drug/alcohol use, history of prostitution, engagement using logged odds ratios (ORs). Logged ORs were selected to in behaviors police believe put victims at risk, police percep- reduce partial spuriousness and to avoid overestimating effect tions of moral character, and inconsistent victim statements to sizes. Eligible studies required a dichotomous outcome; thus, criminal justice personnel. Some studies included two different we analyzed multivariate coefficients, which were largely measures for credibility. For each of these studies, models were binary logistic coefficients. Two studies, however, used either used to combine measures into a single effect size for credibil- probit regression models (Bouffard, 2000)1 or weighted ordi- ity rather than introducing potential bias into the analyses by nary least squares (OLS; LaFree, 1981). Probit and weighted selecting one effect size from the study (see Pratt & Cullen, OLS produce estimates that substantially differ from logit 2000). We also used a single measure for victim alcohol use estimates, thus we needed to convert all estimates to logit prior to or during the assault. Enough studies consistently mea- estimates. Probit coefficients were transformed into an esti- sured victim alcohol use, which allowed us to analyze this mated logged OR through multiplying the coefficients by variable within the aggregate credibility measure and as a p/3. The standard errors were transformed by converting to its stand-alone variable. variance, multiplying by p2/3 and finding its square root (see Hasselblad & Hedges, 1995). Unfortunately, we failed to locate an equation that would perform accurate effect size transfor- Study Dependence mations for weighted OLS; therefore, the LaFree (1981) study To ensure effect size independence and that no variables in our was ultimately excluded from our sample. We analyzed the analyses were drawn from the same sample, we closely mon- remaining 14 studies using meta-analysis for each variable by itored where and when study data collection occurred (Lipsey calculating an overall average effect size statistic for arrest. In & Wilson, 2001). In a meta-analysis—to further prevent bias— other words, we ran an analysis for each variable to determine
6 TRAUMA, VIOLENCE, & ABUSE XX(X) Identification Records identified through database searching (n = 950) Records after duplicates removed Records excluded (n = 937) (n = 13) Found in reference search Eligibility Screening Records screened by (n = 16) title and abstract (n = 49) Found in database search (n = 33) Records screened by full text Full-text articles (n = 29) excluded (n = 15) Studies included in the review Found in reference search Included (n = 14) (n = 3) Found in database search (n=11) Figure 1. Search strategy and results. the average effect for each predictor across all eligible studies they analyzed case clearance status or the decision to present a on the decision to arrest a suspect. case to prosecutors prior to an arrest.3 Another three studies To address the issue of publication bias (Hunter & Schmidt, were excluded because we were unable to accurately transform 2004), we used two fail-safe Ns. First, we calculated their coefficients into logit estimates, and one study that Rosenthal’s classic fail-safe N to estimate the number of poten- focused solely on police decisions in intimate partner cases was tially missing nonsignificant studies required to increase the also removed from the sample. Finally, two studies were mean effect size significance level above p < .05 (Rosenthal, removed from the analysis because their samples were drawn 1979). Second, we used Orwin’s fail-safe N to calculate the from a shared data source and contained no effect sizes repre- number of potentially missing studies with null effects required senting unique variables. In the end, our sample included to reduce the mean effect size to OR ¼ 1.0 (Orwin, 1983). 14 studies representing 79 total estimates that were retained In this way, we can assess each predictors stability against in our meta-analysis.4 publication biases. Study Characteristics Results Table 1 displays a list of the 14 studies included in our analy- Our search strategy produced several potentially relevant stud- ses, as well as the independent predictor variables, sample size, ies for screening eligibility (see Figure 1). The process located data collection location, and the time period covered by each 937 total studies, 49 likely relevant studies, and 29 eligible study. Table 2 lists the variables included in our analysis from studies,2 of which 14 studies were included in our analyses. each study along with corresponding log ORs and ORs. Of We excluded nine eligible studies from the analysis because these studies, 12 were peer-reviewed academic journals, one
Lapsey et al. 7 a Table 1. Study Characteristics. Study N Data Location Data Frame Data Sourceb Alderden & Ullman (2012a) 399 Large midwestern police department 2003 Shared Alderden & Ullman (2012b) 328 Large midwestern police department 2003 Shared Bouffard (2000) 326 Urban/suburban 1995 Independent Kaiser et al. (2017) 770 Los Angeles Police Department (LAPD)/Los Angeles 2008 Shared Sheriff’s Department (LASD) Morabito, Williams, & 2,732 Urban, suburban, and rural 2008–2010 Independent Pattavina (2019) O’Neal et al. (2019) 655 LAPD/LASD 2008 Shared Scott & Beaman (2004) 87 Western Canada 1996 Independent Smith (2005) 121 Maryland 2002–2003 Independent Spohn & Tellis (2019) 491 LAPD/LASD 2008 Shared Tasca et al. (2013) 115 Arizona City 2003 Independent Venema et al. (2019) 22,348 Midwestern police department 1999–20014 Independent Wentz (2019) 231 Midwestern police department 2000–2010 Shared Wentz & Keimig (2019) 418 Midwestern police department 2000–2010 Shared Ylang & Holtfreter (2019) 310 LAPD/LASD 1982–2012 Independent a The rows only include studies that were selected for the analysis of the specified variable. bData source denoted whether the data were shared with another study. To ensure unbiased results, we were careful to select only one unique estimate was derived from the study. was a dissertation (Smith, 2005), and one was a final report logged OR estimates, OR estimates, and 95% confidence inter- from a National Institute of Justice funded study (Morabito, vals. For suspect blameworthiness, both victim injury Williams, & Pattavina, 2019b). One study was completed in (OR ¼ 1.53, p < .01) and victim resistance (OR ¼ 1.48, Canada (Scott & Beaman, 2004), with the remaining 13 taking p < .05) were statistically significant and robustly impacted the place in the United States. Canada and the United States share likelihood for arrest. Results indicated that when a victim was similar correlates of arrest in sexual assault investigations, and injured, the likelihood of arrest was 53% higher, and when the thus, Data from studies were largely collected from locations victim physically or verbally resisted, the odds of arrest classified as urban or suburban, with only Morabito, Williams, increased by 48%. Offender weapon use was the only measure & Pattavina (2019) analyzing data from multiple sites includ- for the protection of the community and was statically signifi- ing rural jurisdictions. Importantly, several studies shared data cant and robust (OR ¼ 1.49, p < .001), indicating a 49% higher sources from the Los Angeles Police Department and Sheriff’s likelihood of arrest when a weapon was used or present during Department (Kaiser et al., 2017; O’Neal et al., 2019; Spohn & an assault. Tellis, 2019), a large Midwestern police department (Alderden Practical constraint variables were the most robust arrest & Ullman, 2012a, 2012b), and a midwestern police department predictors. Cases involving a victim believed to be cooperating (Wentz, 2019; Wentz & Keimig, 2019). with the investigation by law enforcement had the highest odds of arrest (OR ¼ 7.46, p < .001). Similarly, when physical evi- Analyses of Publication Bias dence was collected, the odds of arrest more than doubled (OR ¼ 2.55, p > .05), while cases involving a witness to the To safeguard against “the file drawer problem” (Hunter & assault increased the odds of arrest by 96% (OR ¼ 1.96, Schmidt, 2004), we relied on the classic fail-safe N test p < .01). Additionally, when a victim reported an assault to the (Rosenthal, 1979) and Orwin’s fail-safe N test (Orwin, 1983) police within 72 hr, the odds of arrest increased by 29% to estimate the possibility of publication biases impacting p values and mean effect size results. Results show that all (OR ¼ 1.29, p > .05). predictors were robust against publication bias. All significant Finally, several perceptual shorthand variables were also predictors require large numbers of insignificant studies to found to substantially impact the odds of arrest across studies. increase their p value above .05. Likewise, robust mean effect Nonstranger cases were 38% more likely to result in an arrest sizes are protected against publication biases producing null (OR ¼ 1.38, p > .05), and cases involving intimate partner mean effects. assaults significantly increased the odds of an arrest by 98% (OR ¼ 1.98, p < .05). Victim age (OR ¼ 1.01, p > .05) and suspect age (OR ¼ .99, p > .05) appeared to exert minimal Meta-Analysis Results effects on arrest. Cases involving non-White victims Table 3 displays the average effect size estimates for all pre- (OR ¼ 1.49, p < .001) significantly increased the odds for arrest dictor variables organized by the focal concerns concepts they by 49%, whereas cases involving non-White suspects failed to measured. Table 3 also lists the number of studies that included significantly predict arrest (OR ¼ 0.99, p > .05). Our composite each variable, the number of unique cases for each variable, measure of victim credibility failed to reach statistical
8 TRAUMA, VIOLENCE, & ABUSE XX(X) Table 2. Individual Study Characteristics. Predictor Study N Log. Odds Ratios Odds Ratios Suspect blame Resisted Alderden & Ullman (2012a) 399 0.81 2.25 Kaiser et al. (2017) 770 0.01 1.01 Morabito, Williams, & Pattavina (2019) 2,732 0.45 1.57 Scott & Beaman (2004) 87 0.56 1.75 Smith (2005) 121 0.45 1.57 Wentz (2019) 231 0.23 1.26 Injured Alderden & Ullman (2012a) 399 0.23 1.26 Kaiser et al. (2017) 770 0.02 1.02 Morabito, Williams, & Pattavina (2019) 2732 0.32 1.37 Scott & Beaman (2004) 87 0.56 1.75 Venema et al. (2019) 22,348 0.46 1.58 Wentz & Keimig (2019) 418 0.14 1.15 Ylang & Holtfreter (2019) 310 0.61 1.84 Community protection Weapon Kaiser et al. (2017) 770 0.29 1.34 Morabito, Williams, & Pattavina (2019) 2,732 0.72 2.05 Scott & Beaman (2004) 87 0.98 2.66 Venema et al. (2019) 22,348 0.37 1.45 Wentz & Keimig (2019) 418 0.21 1.23 Ylang & Holtfreter (2019) 310 0.01 1.01 Practical constraints Physical evidence Kaiser et al. (2017) 770 0.98 2.66 Morabito, Williams, & Pattavina (2019) 2,732 0.65 1.92 Smith (2005) 121 0.36 1.43 Tasca et al. (2013) 115 3.3 27.11 Wentz & Keimig (2019) 418 2.2 9.03 Report time Kaiser et al. (2017) 770 0.64 1.90 Morabito, Williams, & Pattavina (2019) 2,732 0.54 1.72 Tasca et al. (2013) 115 2.3 0.100 Venema et al. (2019) 22,348 0.22 1.25 Wentz & Keimig (2019) 418 0.48 1.62 Witness Alderden & Ullman (2012a) 399 0.40 1.49 Kaiser et al. (2017) 770 0.17 1.19 Morabito, Williams, & Pattavina (2019) 2,732 0.16 1.17 Venema et al. (2019) 22,348 0.76 2.14 Wentz (2019) 231 0.70 2.01 Cooperated Alderden & Ullman (2012a) 399 2.11 8.25 Morabito, Williams, & Pattavina (2019) 2,732 1.98 7.21 O’Neal et al. (2019) 655 1.75 5.75 Wentz & Keimig (2019) 418 2.48 11.94 Perceptual shorthand Nonstranger Kaiser et al. (2017) 770 0.34 1.40 Morabito, Williams, & Pattavina (2019) 2,732 0.89 2.43 Scott & Beaman (2004) 87 0.79 2.20 Smith (2005) 121 0.31 1.36 Tasca et al. (2013) 115 2.21 0.11 Venema et al. (2019) 22,348 0.25 1.28 Wentz & Keimig (2019) 418 0.18 1.20 Ylang & Holtfreter (2019) 310 1.49 4.44 Intimate partner Alderden & Ullman (2012b) 328 1.19 3.29 Kaiser et al. (2017) 770 0.47 1.60 Morabito, Williams, & Pattavina (2019) 2,732 1.34 3.83 Scott & Beaman (2004) 87 1.96 7.10 Venema et al. (2019) 22,348 0.61 1.84 Wentz (2019) 231 0.02 1.02 (continued)
Lapsey et al. 9 Table 2. (continued) Predictor Study N Log. Odds Ratios Odds Ratios Victim age Bouffard (2000) 326 0.02 0.98 Kaiser et al. (2017) 770 0.00 1.00 Morabito, Williams, & Pattavina (2019) 2,732 0.01 1.01 Scott & Beaman (2004) 87 0.02 1.02 Smith (2005) 121 0.00 1.00 Wentz (2019) 231 0.00 1.00 Ylang & Holtfreter (2019) 310 0.01 1.01 Suspect age Kaiser et al. (2017) 770 0.00 1.00 Scott & Beaman (2004) 87 0.03 1.03 Smith (2005) 121 0.00 1.00 Wentz (2019) 231 0.03 1.03 Ylang & Holtfreter (2019) 310 0.06 0.94 Victim race Kaiser et al. (2017) 770 0.10 1.11 Morabito, Williams, & Pattavina (2019) 2,732 0.22 1.25 Scott & Beaman (2004) 87 0.34 1.40 Venema et al. (2019) 22,348 0.43 1.54 Ylang & Holtfreter (2019) 310 0.53 1.70 Suspect race Alderden & Ullman (2012a) 399 0.56 1.75 Kaiser et al. (2017) 770 0.22 0.80 Wentz (2019) 231 0.40 1.50 Ylang & Holtfreter (2019) 310 0.54 0.58 Credibility Alderden & Ullman (2012)a 399 0.69 1.99 Alderden & Ullman (2012)b 328 0.80 2.25 Kaiser et al. (2017) 770 0.81 2.25 Morabito, Williams, & Pattavina (2019) 2732 0.43 0.65 Smith (2005) 121 0.21 0.81 Spohn & Tellis (2019) 491 0.32 0.73 Tasca et al. (2013) 115 0.42 1.13 Venema et al. (2019) 22,348 0.53 0.59 Wentz & Keimig (2019) 418 0.48 0.62 Ylang & Holtfreter (2019) 310 0.44 0.64 Victim alcohol use O’Neal et al. (2019) 655 0.63 0.53 Smith (2005) 121 0.21 0.81 Venema et al. (2019) 22,348 0.53 0.59 Wentz & Keimig (2019) 418 0.48 0.62 Ylang & Holtfreter (2019) 310 0.71 0.49 significance (OR ¼ 0.65, p > .05). Even so, the measure had a Discussion robust effect and decreased the odds of arrest by 35% when a The focal concerns framework has been a salient perspective report mentioned at least one variable known to impact percep- within criminal justice for decades (Crow & Adrion, 2011; tions of victim credibility. Finally, victims’ alcohol use prior to Hartley et al., 2007; Spohn et al., 2001; Steffensmeier et al., or during the assault significantly and substantially reduced the 1998; Tillyer & Hartley, 2010; Ulmer & Johnson, 2004). Yet, likelihood of arrest by 41% (OR ¼ 0.59, p < .001). until recently, focal concerns has rarely been applied to the Using the logged OR estimates for each predictor across all literature on police decision making in sexual assault cases. 14 studies, we created forest plots (see Figure 2). The forest However, because of somewhat consistent measurement of plots provide a visual display of the odds of arrest surrounding case, victim, and suspect characteristics across studies, several each variable. Included in Figure 2 are the effect size estimates, scholars have included variables proposed by O’Neal and standard errors, p values, and 95% confidence intervals for Spohn’s (2017) application and operationalization of focal con- each focal concerns variable. In Figure 2, points plotted to the cerns measures. Although measurement across studies is sim- right of zero indicate increased odds of arrest, whereas plots ilar, no systematic meta-analytic review to determine the pointed to the left of zero indicate decreased odds of arrest. framework’s empirical status for police decisions in sexual Figure 2 helps visually demonstrate the overall magnitude and assault cases exists in the published literature. Thus, to the direction for each predictor, as well as display the precision of authors’ knowledge, this study is the first to conduct a systema- the estimates by plotting the variables around 95% confidence tic meta-analysis on the subject. intervals.
10 TRAUMA, VIOLENCE, & ABUSE XX(X) Table 3. Mean Effect Size Estimates for Focal Concerns Predictors. Classic Orwin Predictor k N N N Log. Odds Ratio (OR) OR Ci.Lb (95%) Ci.Ub (95%) Suspect blame Resisted 6 4,340 28 245 .394* 1.48 0.0185 0.7688 Injureda 7 27,064 78 232 .427** 1.53 0.1138 0.7407 Protection of the community Weapon 6 26,665 45 252 .399*** 1.49 0.2687 0.5288 Practical constraints Physical evidencea 5 4,156 748 .935 2.55 0.4901 2.361 Report timea 5 26,383 37 .255 1.29 0.75 1.2604 Witnessa 5 26,480 340 214 .675** 1.96 0.1721 1.1778 Cooperated 4 4,204 436 829 2.01*** 7.46 1.7731 2.2401 Perceptual shorthand Nonstrangera 8 26,901 197 .322 1.38 0.7437 1.3883 Intimate partnera 6 26,168 214 549 .684* 1.98 0.0831 1.284 Victim age 7 4,577 0 .01 1.01 0.0201 0.0315 Suspect age 5 1,519 0 .002 0.99 0.0561 0.0518 Victim race 5 26,247 55 157 .400*** 1.49 0.1834 0.6153 Suspect race 4 1,710 17 .012 0.99 1.32 0.4416 Credibilitya 10 28,032 21 .439 0.65 1.32 0.4416 Victim alcohol use 5 23,852 84 252 .533*** 0.59 0.7063 0.3589 a Effect sizes were significantly different according to (p < .05) Q-statistic measuring heterogeneity. Effect size estimated are weighted by the inverse of the variance. k denotes the total number of studies, while N signifies the total number of cases. *Statistically significant at p < .05. **Statistically significant at p < .01. ***Statistically significant at p < .001. Log Odds Ratio and 95% CI Grouped by Statistics for each study Less likely to arrest More likely to arrest Predictor LogOR SE P-value Resisted .394 .191 .039 Injury .427 .160 .008 Weapon .399 .066 .000 Evidence .935 .727 .198 Report time .255 .513 .619 Witness .675 .257 .009 Cooperated 2.010 .119 .000 Non-stranger .322 .544 .554 IP .684 .306 .026 Victim age .006 .013 .664 Suspect age -.001 .032 .798 Victim race .399 .110 .000 Suspect race -.077 .248 .757 Credibility -.439 .449 .328 Alcohol use -.533 -.706 -.359 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Arrest Odds Meta-Analysis Random Effects Models Figure 2. Forrest plot and log odds ratio for each focal concerns variable.
Lapsey et al. 11 The primary goal of meta-analysis is to help understand the suspect demographic variable impacted the magnitude of magnitude and direction of effect sizes across individual effects. Based on the focal concerns framework, the impact studies and not to determine the statistical significance (see of non-White victims on arrest is opposite of the anticipated Haidich, 2010; Hedges & Olkin, 1980; Sullivan & Feinn, direction (see Steffensmeier et al., 1998). Cases involving vari- 2012). That said, although we report statistical significance, ables known to victim credibility correlated with a decrease of our goal was not to calculate p values but to summarize effect .35 in the odds of arrest, meaning, officers’ perceptions of a size estimates for predictors of arrest included in published victim had a robust effect on the decision to arrest and alcohol studies of police sexual assault investigations using focal con- use by the victim had a strong negative correlation with odds of cerns concepts. According to Steffensmeier et al. (1998), each arrest. If the victim reportedly used alcohol prior to or during focal concern interacts to influence criminal justice practitioner the incident, the arrest odds decreased by .41. Findings for decision making. Thus, we adapted the work of O’Neal and credibility and alcohol consumption are expected given that Spohn (2017) to operationalize focal concerns concepts and police often endorse rape myths and believe “true” victims combine variable estimates across studies. We classified com- behave in a specific manner prior to, during, and after the monly used focal concerns variables into suspect blameworthi- assault. In sum, our results indicated that focal concerns is a ness, protection of the community, practical constraints, and suitable framework for making sense of police decisions in perceptual shorthand and created a single comprehensive mea- sexual assault cases. sure for victim credibility. The work here is intended to explore these concepts, “take stock” of the published literature, and identify the empirical status of focal concerns within the con- Policy Implications text of sexual assault investigative decision making. Based on While our primary goal was to “take stock” of focal concerns our analysis, our findings revealed overall empirical support and its application to police decisions in sexual assault cases, for the focal concerns perspective as applied to police deci- our findings are well suited to inform policy aimed at improv- sions to arrest in sexual cases. ing police response to victims. Past research has shown that Our analysis demonstrated that the majority of predictors sexual assault training courses can be effective at producing produced robust effect size estimates. First, variables operatio- positive outcomes among law enforcement, such as reduced nalized as practical constraints had the greatest order of attributions of blame toward victims (Darwinkel et al., 2013; magnitude. In fact, the three most robust estimates were Tidmarsh et al., 2020), reduced rape myth acceptance produced by practical constraints variables. Specifically, (B. Campbell et al., 2019; Murphy & Hine, 2019; Rich & victim cooperation during an investigation produced the stron- Seffrin, 2012), and improved knowledge of trauma-informed gest effect (OR ¼ 7.46), followed by the availability of physical investigative techniques (B. Campbell et al., 2019; Franklin evidence (OR ¼ 2.55), and the presence of a witness to the et al., 2019; Lonsway et al., 2001; Rich & Seffrin, 2012). assault (OR ¼ 1.96). These findings are likely due to witnesses Research has also suggested that training may help improve being readily available immediately after the incident, the officers’ evaluation of victim credibility and increase their importance of corroborating evidence, and victim testimony knowledge about the utility of forensic evidence, which may to establishing probable cause for arrest. Second, suspect in turn reduce reliance on victim credibility during the blameworthiness variables produced moderate effect size esti- decision-making process (B. Campbell et al., 2015). Thus, mates. Cases involving injured victims had the greatest odds of based on findings from our meta-analytic review, there are arrest (OR ¼ 1.53) and cases where the victim either verbally or three primary areas of focus that could enhance police sexual physically resisted had 1.48 greater odds of arrest. These results assault training. are consistent with myths and stereotypes surrounding rape, First, training should focus on facilitating victim coopera- whereby officers have historically assumed that “real” rapes tion with investigators and the criminal justice system. To involve a victim fighting back against their assailant and being improve victim cooperation, training programs should focus physically harmed during the struggle. Third, regarding the on educating officers about victim-centered and trauma- need for protection of the community, when an offender used informed techniques during initial and follow-up interviews. a weapon during the assault, the cases had a 49% greater odds Using these techniques may help officers maintain victim for arrest. This is unsurprising given that police generally view engagement and reduce the likelihood of secondary trauma. offenders who use weapons during a crime as more violent Second, given the importance of physical evidence as a pre- individuals with a greater propensity for future criminal acts. dictor of arrest, officers should be offered additional training Finally, for perceptual shorthand variables, intimate partner regarding the collection and processing of physical evidence at sexual assaults had 1.98 higher odds of arrest, whereas non- crimes scenes. Educating officers about proper evidence col- stranger assaults produced 1.38 greater odds of arrest. This is lection may assist them in identifying and securing critical expected given the importance of identifying potential suspects evidence needed for prosecution. Related, training programs interacting with the heightened ability to locate the suspect in could also include a discussion of recent work on the impor- intimate partner and nonstranger assaults (Spencer & Stith, tance of collecting and testing sexual assault kits. Some juris- 2020). Except for victim race, which increased the arrest odds dictions have found that testing sexual assault kits is a critical by 1.49 when the victim was non-White, no other victim or investigative and crime prevention tool. Not only have sexual
12 TRAUMA, VIOLENCE, & ABUSE XX(X) assault kits linked multiple offenders to unsolved sexual statistical power, and (d) improve the ability to detect effects assaults, the testing of these kits has exposed the multitude of (Borenstein et al., 2011; Rosenthal & DiMattateo, 2001). violent and property crimes committed by sexual offenders Indeed, these advantages assisted this study in identifying the who have avoided arrest. Specifically, in their analysis of strongest correlates of arrest based on findings from existing cases associated with untested sexual assault kits, Lovell and research. colleagues found that undetected offenders committed an aver- To address these limitations and advance this line of age of 7.4 felonies before being arrested for their latest sexual research, further meta-analyses on this topic should be con- offense (Lovell et al., 2020). Thus, training officers about the ducted once more studies on the police decision to arrest in importance of processing physical evidence may reduce other sexual assault investigations become available. These future property and violent crimes in addition to facilitating the arrest studies should follow O’Neal and Spohn’s (2017) operationa- of sexual offenders. lization of focal concerns variables to promote consistency of Finally, our analysis revealed that arrest decisions were measures across studies. Doing so will assist in providing impacted by rape myths across studies. As such, more training standardized measures of focal concerns variables that can is needed to dispel misconceptions about sexual assault cases, improve study replication and generalizability, as well as which is evident by our detected effects demonstrating the increase the precision of effect size estimates, and improve impact of victim injury, victim resistance, suspect weapon use, an understanding of the relationship between focal concerns victim credibility, and victim alcohol use on officers’ decision and the decision to arrest in sexual assault cases. To this end, to arrest. Training programs that fit these three recommenda- our systematic review and meta-analysis is a critical first step tions have recently been implemented in some jurisdictions. in assessing the effects of focal concerns variables on arrest in For example, in the State of Kentucky, each police department sexual assault cases across studies. Indeed, our findings can sends at least one officer to attend a comprehensive 40-hr speak to the strongest correlates of case attrition and advance- sexual assault investigation course. The program covers topics ment at this critical decision point in the criminal justice such as rape myth acceptance, the use of DNA evidence in process. investigations, victim interviewing, and trauma-informed investigations. Importantly, a recent evaluation found that Ken- tucky’s training program improved officers’ short- and Implications for Practice, Policy, and Research long-term knowledge of trauma-informed practices, knowl- Training should focus on improving lpolice responses to edge of state laws, and perceptions of victims (B. Campbell victims of sexual assault, including reducing rape myths et al., 2019). and misconceptions of rape. Training should focus on police educating officers on victim-centered, trauama informed interview Limitations techniques. While our meta-analysis has theoretical and policy implica- Officers should be offered additional training on the tions, this review is not without limitation. Although our study collection and processing of physical evidence, specifi- help synthesize the literature on police decision making in cally the utility of sexaul assault kits. sexual assault cases, our sample size was relatively small. As Future research should continue to use the focal con- a result, we were unable to perform moderator analyses, assess- cerns framework to explain officer decision-making in ments of interaction effects, or statistical tests concerning pub- sexual assault cases. lication bias. Because of selective reporting and publication bias, the results from the current study could potentially over- estimate effect size estimates, although results from Orwin’s Declaration of Conflicting Interests Fail-safe N test were found all predictors to be robust against The author(s) declared no potential conflicts of interest with respect to potential publication bias. However, based on our thorough the research, authorship, and/or publication of this article. search effort, it seems unlikely that enough studies with con- flicting findings exist to meaningfully alter the strength or sig- Funding nificance focal concerns variables detected by our analysis. Additionally, although our analysis contains a small number The author(s) received no financial support for the research, author- of studies, prior work has suggested that meta-analyses—even ship, and/or publication of this article. when conducted using small samples—can use existing data to identify strengths and gaps across studies (Rosenthal & DiMat- ORCID iD tateo, 2001) and produce relevant findings to inform future David S. Lapsey Jr. https://orcid.org/0000-0001-8031-2407 research (Borenstein et al., 2011). In fact, the ability to weight studies by sample size and accuracy across studies has several advantages over drawing conclusions from a single study. Notes These advantages include the ability to (a) improve validity of 1. Although transformations were completed for Bouffard (2000), the findings, (b) avoid an overreliance on p values, (c) increase estimates contained extreme statistical outliers and a credibility
You can also read