ArticlePDF Available

Improving victim engagement and officer response in rape investigations: A longitudinal assessment of a brief training



The initial interaction between rape victims and police officers affects how cases progress through the criminal justice system. In one US state capitol, the police agency determined its initial response to rape victims was sub-par. Victim engagement was low, and officer-written reports often endorsed negative stereotypes about rape victims. A four-hour training to enhance police response was developed and implemented. Within four months, all sworn officers (n ~ 600) completed an in-person, four-hour training. We first test the effects of training on the percentage of rape victims who stay engaged in the investigative process following their initial contact with officers. We then use a machine-learning-based text analysis of all written reports (77 pre-training, and 55 post-training cases) of initial contacts between officers and victims. Compared to the six months before training, victim engagement improved 32% in the post-training period. Written reports by officers also improved, with increased victim-supportive language and improved focus on victim services.
Improving Victim Engagement and Ocer Response in Rape
Investigations: A Longitudinal Assessment of a Brief Training
SCOTT M. MOURTGOS, University of Utah
IAN T. ADAMS, University of Utah
SHARON H. MASTRACCI, University of Utah
e initial interaction between rape victims and police aects how cases progress through the criminal justice system. In
one US state capitol, the police agency determined its initial response to rape victims was sub-par. Victim engagement
was low, and ocer-written reports often endorsed negative stereotypes about rape victims. A four-hour training to
enhance police response was developed and implemented. Within four months, all sworn ocers (n~600) completed an
in-person, four-hour training. We rst test the eects of training on the percentage of rape victims who stay engaged in the
investigative process following their initial contact with ocers. We then use a machine-learning-based text analysis of all
written reports (77 pre-training, and 55 post-training cases) of initial contacts between ocers and victims. Compared to
the six months before training, victim engagement improved 32% in the post-training period. Written reports by ocers
also improved, with increased victim-supportive language and improved focus on victim services.
Keywords: police training, victim engagement, rape investigation, text analysis
is is a pre-print of a manuscript that is forthcoming in the Journal of Criminal Justice. ere may be slight
changes between this version and the version of record.
Police response to sexual assault victims has received public scrutiny and scholarly attention for decades
(Franklin et al., 2020). is attention and scrutiny are not unwarranted. Sexual violence happens frequently
(Franklin et al., 2020), yet when victims1report occurrences to police, they are frequently met with skepticism,
dismissiveness, questions probing the validity of their statement, and inquiries regarding extralegal topics
such as previous sexual history (R. Campbell, 2008; Martin, 2005). Police-victim interactions such as this often
leave victims reluctant to have further contact with the police and discontinue their cooperation in the justice
process (Ahrens, 2006). Seventy-ve percent of case attrition occurs during the investigation phase of a sexual
assault complaint (Barrett & Hamilton‐Giachritsis, 2013), with one of the most common reasons for attrition
being a victim withdrawing their2participation (Feist et al., 2007).
Police agencies attempt to address victim-engagement concerns through specialized sexual assault training
for their ocers. However, few jurisdictions have implemented mandatory training of this nature (B. Campbell
et al., 2019), and even fewer have been evaluated for their eectiveness (Franklin et al., 2020). Moreover, studies
tend to evaluate individual ocers’ attitudes rather than examining ocers’ behavioral performance (Lonsway
Forthcoming at Journal of Criminal Justice.Current version: April 28, 2021; Corresponding
1ere are active terminology debates among academics, practitioners, and advocates concerning descriptive language, including
“victim, and “survivor,” with the former more consistently used within the criminal justice practioner context. Given the focus of
this study in policing, we opt to use “victim” throughout the paper, though dierent audiences may nd alternate language choices
appropriate for their use.
2We recognize that the vast majority of sexual assaults and other forms of sexual violence are perpetrated against women. However,
sexual violence is also committed against males, thus, we seek gender neutrality throughout the manuscript by using plural nouns.
et al., 2001). e study of police attitudes through surveys and scales may not detect implicit attitudes and
preclude assessing whether such attitudes aect sexual assault investigations (Shaw et al., 2017).
e attitudinal focus arises from a belief that ocers’ attitudes towards sexual assault and assault victims
are the “primary block” to an eective police response (Lonsway et al., 2001, p. 697). We submit that police
attitudes regarding sexual assault are a secondary concern compared to improved police behaviors during their
interactions with sexual assault victims. While a decrease in ocer ascriptions to rape myths or other think-
ing errors regarding sexual assault would ideally coincide with improved behaviors, behaviors should be the
primary concern (Go, 2016; Lonsway et al., 2001). Research suggests that “change eorts are increasingly
successful when they target behavioral change as opposed to trying to force value change from an external
agent (Shaw et al., 2017, p. 611). Whether words spoken or actions taken, behaviors are what victims tangibly
experience, regardless of whether those words or actions align with unseen police attitudes.
In the current study, we evaluate a police sexual assault training program’s ecacy on police behavior when
engaging with sexual assault victims. We accomplish this with a longitudinal study examining two outcome
measures. First, we test victim engagement: Using a Bayesian paradigm, we nd an average expected improve-
ment in victim engagement post-training of 32%. Second, we examine the content of written police reports doc-
umenting rape complaints before and after the provided training. By examining police behavior in this manner,
we move beyond measuring attitudes with survey methods. How an ocer ocially documents a rape allega-
tion directly impacts the case’s future processing (Venema, 2016) and provides insight into the police-victim
interaction itself (Shaw et al., 2016, 2017). Our algorithmic analysis of reports nds that post-training, o-
cers’ written reports are more concerned with victim well-being, use more appropriate phrasing, and generally
reect a positive change in how ocers interact with victims.
Police Response to Sexual Assault and Victim Engagement
While not all police-victim interactions with sexual assault victims are negative, bad experiences are not un-
common. Negative interactions are attributed primarily to police ocers holding problematic attitudes about
sexual assault victims, such as victim-blaming and rape-myth acceptance (R. Campbell et al., 1999; Ullman &
Filipas, 2001). By rape myths, we mean extralegal factors associated with stereotypes about ‘real rape’ (Estrich,
1987; e.g., “It can’t really be rape if …” victim drug or alcohol use, sex worker status, and antecedent victim ‘risk-
taking’ behavior) that may bias ocers’ perceptions about a sexual assault report and their decision-making
regarding it. Further, police ocers have been found to discount cases based on assessments of a victim’s cred-
ibility, determined by alcohol or drug consumption on the victim’s part, aectual responses by victims that do
not align with what an ocer thinks a victim should act like, and having a current or past relationship with the
suspect (R. Campbell & Raja, 2005; Patterson, 2011; Shaw et al., 2017; Sleath & Bull, 2017). Indeed, ocers who
more readily accept rape myths tend to rate victims as having some responsibility in the sexual assault, hold
perpetrators less responsible, and deem sexual assault cases less authentic (Hine & Murphy, 2019).
However, it remains unclear whether police attitudes are the proximal cause of negative police-victim in-
teractions upon examining the relevant literature. In a study conducted by Campbell and Raja (2005), victims
were frequently asked how they were dressed when the assault occurred and about their prior sexual history.
Accordingly, most of the study victims reported that their interactions with the police made them feel guilty,
distrustful, depressed, and anxious. In research conducted by Patterson (2011), a substantial portion of victims
reported being told their story was unbelievable, being asked about prior sexual history, being asked about
how they were dressed leading up to the assault, and being told they would be charged with a crime if they
did not provide an accurate story. Unsurprisingly, many of these victims reported that they would not have
reported the sexual assault if they had known beforehand what they would experience. Further, Temkin (1999)
relates that victims who reported a negative experience with the police described ocers as having disbelieving
attitudes, expecting ‘genuine’ victims to act in a particular manner, and lacking sympathy.
All of these studies have in common overt ocer behavior that resulted in negative perceptions on the vic-
tims’ part. Victim blaming and endorsing rape myths are presumed to be the motivations behind these actions.
However, behavior and motivations are separate phenomena. at is, while it is hard to imagine some of the
actions taken by ocers during negative police-victim interactions are not driven by internal attitudes (espe-
cially when one reads rst-hand accounts from victims of sexual assault, see Spohn & Tellis, 2010 for example),
the causal link between behavior and motivation has not been clearly established (Sleath & Bull, 2017).
Some argue that it is not personal attitudes that drive negative interactions with sexual assault victims;
rather, the problem’s root lies within police systems. Ocers are trained to be skeptical, unemotional, and to
challenge and validate stories in an attempt to reveal the truth. ey are not typically trained to be supportive
and empathetic. Processing cases in the manner described above benets the police system, as victims can be
managed with less eort. Ocers are free to respond to additional calls for service more quickly if the patience
and time it takes to respond to a sexual assault victim appropriately are not required (Martin, 2005). “us,
the priorities of the organization take precedence over the needs of the victim” (R. Campbell & Raja, 2005,
p. 98). Barrett et al. (2013) nd support for this idea in an experimental paradigm studying police sexual assault
investigators. ey nd that investigators primarily perceive rape victims as a source of information to further
case progress. e welfare of victims was of secondary importance.
Viewing negative police-victim interactions in sexual assault cases through this lens may also help explain
the oft-reported victims’ experiences of police persuasion to pursue their case or not. As Kerstetter (1990) and
Kerstetter and Van Winkle (1990) highlight, police ocers have a substantial amount of inuence over whether
victims cooperate with an investigation or not. Ocers can sway victims to withdraw from participation if de-
tails of the case do not align with ideas about ‘real rape. Conversely, ocers can encourage victims to continue
participating in the justice process if they perceive the case to be ‘legitimate. Indeed, Konradi (2007) found that
sexual assault victims often felt that ocers were trying to persuade them to either pursue or not pursue their
ese experiences could arise for a couple of reasons. First, ocers often become accustomed to factors
within cases that determine whether prosecutors will le charges. Sometimes referred to as a ‘downstream
orientation’ (Alderden & Ullman, 2012; Frohmann, 1991; Pattavina et al., 2016; Spohn & Tellis, 2019; Ylang &
Holtfreter, 2019), if the circumstances of a particular case do not meet those standards (i.e., a ‘trial suciency’
standard), ocers may attempt to dissuade a victim from pursuing the case to conserve agency resources. Sec-
ond, it is also possible that ocers attempt to dissuade victims from continuing in the justice process because
they do not assess the victim as credible (Kelly et al., 2005; Patterson, 2011; Venema, 2016). Regardless of which
explanation, or a combination thereof, is accurate, a victim’s continued engagement with the justice process
represents more than a simple statement by a victim of their own volition (Kerstetter, 1990).
Police ocers have a sizeable eect on a victim’s decision to cooperate and engage in the legal process (Ker-
stetter & Van Winkle, 1990). Negative interactions with police ocers can serve a silencing function, resulting
in the victim’s withdrawal of cooperation. at is, if a victim perceives that the police do not care about what
happened to them, convey blame, or express doubt in their narrative, a victim may be reluctant to have further
contact with the police (Ahrens, 2006; Alderden & Ullman, 2012; Kelly et al., 2005).
Most sexual assault victims’ rst contact will be with the responding patrol ocer to make an initial report
(Patterson, 2011; Spohn & Tellis, 2010). is initial interaction is crucial to retaining victims’ engagement in the
justice process3. Suppose the responding ocer discourages a victim’s participation by framing the interview
3is study explicitly examines an intervention aimed at increasing victims’ engagement in the justice process. However, victim
engagement may not be the primary goal when exploring ways to improve society’s response to sexual assaults. Involvement with
with repeated questions to evaluate consistency or giving the impression of questioning the victim’s credibility.
In that case, the victim will be less likely to engage in the justice process further. Inquiries regarding extralegal
factors, such as prior sexual history or whether the victim sexually responded to the assault, can also discourage
continued engagement with the justice process (R. Campbell, 2008).
As Spohn and Tellis (2010) emphasize, inappropriate treatment of victims by initial interviewing ocers is
the primary cause of case closures due to ‘uncooperative’ victims. ey provide numerous rst-hand accounts
from victims describing their treatment by initial responding ocers. e rst excerpt from their study illus-
trates what occurs when a responding ocer focuses on extralegal factors, leaving the victim to conclude that
the ocer blames them for the assault.
I was harshly interrogated…e police ocer was incredibly rude and harsh; well, not rude, harsh.
eir main focus was that I was drunk and how drunk was I but they never considered if I was
too drunk to consent…I gave a statement and again they xated on how much I had drank and
moved towards blaming me because the rapist was someone I knew…At that point they believed
him because I was drinking a lot and they made the assumption it was consensual (pp. 1412-1413).
e second excerpt illustrates an explicit accusation of dishonesty.
Immediately they told me I was lying and on drugs. ‘Straight up! You’re on drugs…You’re lying,
you’re lying! Stand up, close your eyes, and count to thirty. Can you count to thirty?…Tell me the
truth or you will personally go to prison for lying to a police ocer’ (p. 1414).
Given this type of initial interaction with ocers, a victim may choose to withdraw from the justice process.
As will be seen below, documentation of police-victim interactions similar to the two excerpts from Spohn and
Tellis (2010) leads us to the present study.
e Current Study
Context of the Study
e setting for the study is one major metropolitan police agency. e subject agency provides full-spectrum
policing services to a mixed urban core and suburban service area that includes the state capital, a population
of more than 200,000 nighttime residents, and a wider metro area population of over 1.1 million. e agency is
one of only two in the state that has attained accreditation through the Commission on Accreditation for Law
Enforcement Agencies (CALEA). e agency employs approximately 600 sworn ocers and has a specied unit
within its investigation division responsible for investigating all sexual assault allegations.
In 2014, the department came under public criticism regarding its handling of sexual assault investigations.
During the resulting review of past police reports documenting sexual assault investigations, numerous in-
stances were discovered of less-than-ideal interactions between victims and the initial responding ocer. e
below excerpt is only one example and echoes the Spohn and Tellis’ (2010) examples listed above.
My entire interview with [] was done alone, without her boyfriend or family present. I did nd it
the criminal justice system can take an extreme toll on victims’ mental and emotional health (Adams & Mastracci, 2017). Accordingly,
stepping away from the justice process may be the healthiest decision for some victims. While recognizing this, a primary goal of
police involvement in sexual assault investigations is to hold oenders accountable. is cannot be accomplished without the victim’s
engagement in the justice process.
odd, however, that [] stated she did not ght to get away from [] during the incident. [] stated he
had a hold of her hands so she couldn’t get away, but she stated she didn’t use her legs as a weapon,
or call out at any time for help. []’s story isn’t consistent with being able to ght back since she
would have had a free hand while he was pulling down her shorts and then again when he inserted
his penis into her vagina…[] could not explain any of these things to me. I did not press [] for
explanations about these oddities.
While the ocer conducting this initial investigation may have been well-meaning (Martin, 2005), concerns
abound in their interview with the victim: Formally documented skepticism, unrealistic expectations of a vic-
tim during a violent assault, and inappropriate judgment. A victim could easily interpret active disbelief on the
ocer’s part and that they were being ascribed responsibility for the incident. In this case, the victim refused
further contact with detectives for continued engagement with the justice process. is example was just one
of many similarly styled police-victim interactions, resulting in a loss of victim contact or victim withdrawal
from the justice process. While there are likely many reasons why victims choose to withdraw from the justice
process, previous research suggests problematic interactions with the initial responding ocer are responsible
for a substantial portion of victim withdrawal.
Hypotheses: Victim Engagement and “What police actually do”
Prior studies on the ecacy of sexual assault training focus on ocer attitudes. e ndings of these studies
are mixed (Sleath & Bull, 2017), with several resulting in decreased problematic attitudes (B. Campbell et al.,
2019; Darwinkel et al., 2013; Franklin et al., 2020) and several others nding no eect (Goodman‐Delahunty &
Graham, 2011; Sleath & Bull, 2012).
While an improvement in ocer attitudes is a desirable goal in and of itself, we are concerned with the
eect of training on victim engagement. Specically, we test whether focused training leads to improved victim
engagement in the justice process. Past research indicates that police’s improved response may increase victim
engagement (Ahrens, 2006; Alderden & Ullman, 2012; Bouard, 2000; R. Campbell & Fehler‐Cabral, 2018; Kelly
et al., 2005; Kerstetter, 1990; Spohn & Tellis, 2010). us, we oer the following hypothesis:
Hypothesis 1: Specialized sexual assault training for ocers will increase victim engagement.
But what is it about the training that keeps victims engaged with the process? How does training increase
engagement? e only study that we are aware of that examines the eect of sexual assault training on police
behavior suggests that sexual assault-specic training improves police-victim interactions. In an experiment,
Lonsway et al. (2001) provided specialized sexual assault training to new police recruits, measuring the eects of
training on attitudinal outcomes and behavioral performance through simulated victim interviews. ey found
no signicant dierence between recruits in the experimental group and control group when it came to attitudes
regarding sexual assault and sexual assault victims. However, when behavior was examined in simulated victim
interviews, recruits in the experimental group were more likely to allow victims to control the pace and tone
of the interview, address the needs and concerns of the victim, provide empathy and reassurance, and question
victims less about their use of alcohol.
We build upon Lonsway et al.’s (2001) ndings and examine the eect of sexual assault training on how police
actually describe sexual assaults in their case reports, rather than in follow-up surveys of attitudes about sexual
assault or in interview simulations. As Lonsway and colleagues indicate, “police attitudes are ultimately of lesser
concern than their actual behavioral performance when responding to sexual assault cases and victims…[I]t is
particularly unfortunate that…research has not systematically documented what police actually do when they
respond to a sexual assault case and interview a victim” (p. 698, emphasis supplied). With this in mind, we
provide the following hypothesis:
Hypothesis 2: Specialized sexual assault training will lead to improved police-victim interactions during
the initial response.
Data and Method
To test these hypotheses, we use ocial police report data from the subject department. e training inter-
vention was provided to all sworn members of the police department between September 2015 and December
2015. All rape reports for six months prior (March 2015 through August 2015) and six months after (January
2016 through June 2016) were collected to allow for a pre- and post-training comparison. Cases were identi-
ed through evidence reports documenting the collection of a Sexual Assault Kit (SAK). Typically performed
by sexual assault nurse examiners (SANEs), SAK examinations include collecting biological evidence, such as
semen and saliva, which may be present on victims’ bodies post-assault. After collection, police take custody of
the SAKs and are tasked with transporting them to a forensic laboratory for analysis, after which their contents
may be used in the investigatory and prosecutorial processes (Mourtgos et al., 2021; Valentine et al., 2019).
is department’s policy is to obtain a SAK whenever possible in all reported incidents of rape when the
rape is reported within 120 hours of the occurrence4. While a victim may refuse to undergo a SAK examination,
it is rare5. us, identifying cases in this manner is nearly (if not entirely) comprehensive for the crime of
rape. After removing cases where a SAK was collected for a crime against a juvenile, the crime occurred in
another jurisdiction, or an ocer conducted no initial interview, 77 pre-training cases were retained, and 55
post-training cases were retained for analysis6.
Training Intervention
Based on an internal review, a training program was developed within the subject department to better prepare
rst responders for handling the initial investigation of sexual assaults. e principal investigator is a subject
matter expert in sexual assault investigations and was asked to help develop a training course for initial respond-
ing ocers. e class was designed as a 4-hour training block consisting of the following modules, all based on
contemporary research:
Understanding sexual assault oenders.
Victim vulnerabilities that oenders are looking to exploit.
Perceptual, behavioral, and cognitive distortions experienced by victims of sexual assault.
Victim neurobiology.
Explaining paradoxical behaviors sometimes reported by victims during sexual assault investigations.
e current state of research regarding false allegations of sexual assault.
How to properly interview sexual assault victims.
Instructions on recently updated department policy requiring: 1) the primary consideration in sexual
4is timeframe is set by the local SANE organization that conducts the exams for the studied department.
5Due to data limitations, we are unable to provide an exact percentage of victims that choose to not undergo a SAK examination.
However, the sergeant overseeing the Special Victims Unit at the time of this study estimates that only approximately 5% of victims
6Cases with a juvenile victim were removed prior to analysis. At the studied agency, policy dictates that child sexual abuse victims
are not to be interviewed in the eld by the responding patrol ocer. Rather, the patrol ocer collects limited information from the
reporting parent or guardian and the child is interviewed at a later time by a detective trained in child forensic interviewing approaches.
assault investigations to be the health and safety of the victim; 2) patient, objective, and non-judgmental
interviews; and 3) no opinion of whether the case is founded or unfounded to be included in the initial
Notably, the training was developed so that police ocers could relate their own experiences to those of
the victims they were interviewing (see Di Nota et al., 2020, for a synthesis of research). For example, the
perceptual (and other) distortions often experienced by police ocers during violent encounters were paralleled
with those of sexual assault victims (e.g., impaired memory, auditory blunting, visual distortions, disassociation,
etc.). Further, two case studies were included during the training to showcase sexual assault cases that appeared
unlikely upon the initial report but were found to be accurate upon investigation. ese case studies included
reviews of the initial report (specically noting paradoxical behaviors that the responding ocers expressed
doubt about), subsequent investigation, and case outcomes. e inclusion of these training methods was deemed
necessary by those developing the training program to get buy-in from ocers within the department.
Longitudinal Design
In the current study, we test the ecacy of a sexual assault training program. Specically, we examine if the
program resulted in more positive police-victim interactions during ocers’ initial response to sexual assault
reports. e training was conducted during in-service sessions from September 2015 through December 2015.
All sworn employees received the training. All training sessions were taught by the principal investigator and
the then-current supervisor of the department’s Special Victims Unit.
While a randomized control trial (RCT) design would be ideal, the associated legal, ethical, and practical
considerations often make RCTs impossible to employ (Singleton Jr. & Straits, 2018). Such was the case here.
Public and political pressure to improve the department’s response to sexual assault victims would not allow
RCTs. Further, conducting RCTs in this particular circumstance raises ethical concerns. Assuming the training
was benecial and resulted in ocers better serving sexual assault victims, by conducting an RCT the agency
would provide this enhanced response to some victims and withhold it from others.
Accordingly, a longitudinal analysis was employed to test the training program’s eect on victim engage-
ment. is type of design is proper when an entire sample receives the same treatment. In this case, the entire
population of sworn ocers in the agency received the same training intervention. By extension, all rape cases
were handled (in the post-training context) by ocers who had undergone the training. By leveraging available
data in both the pre- and post-training period, we build a more robust causal inference than is possible with
a single-shot cross-sectional design. Notably, the pre-training phase includes reports in the period following
the initial intense public and political attention, such that we are more condent that any eect detected in the
post-intervention period is not simply a short-term reaction to the outcry. A further strength of using available
data is that we avoid reactive measurement problems, as neither the ocers nor the sexual assault victims were
aware of being studied or observed, nor was a study planned at the time (Singleton Jr. & Straits, 2018). ere-
fore, there is no reasonable connection between the results we document in this study due to behavioral change
in response to being studied.
Victim Engagement
To measure victim engagement, case records were examined to determine if a victim withdrew from the justice
process after the initial police report was made. A victim was determined to have withdrawn from the justice
process if 1) they failed to make contact with the follow-up detective after the initial report, or 2) the victim
explicitly told the assigned detective that they no longer wanted to move forward with their case. To clarify the
point regarding a victim failing to make contact with a follow-up detective after the initial report, the following
is provided. At the studied agency, detectives investigating sexual assault cases must adhere to a three-step pro-
cess regarding victim contact. First, they have to exhaust all telephone options, including leaving voicemails. If
contact cannot be made over the telephone, detectives must respond to the victim’s residence. If contact cannot
be made at the residence, detectives must leave their business card. Finally, if a victim still does not contact the
detective, the detective must send a letter from the police department requesting contact and explaining that
the case will be inactivated if the victim does not contact them. If victims fail to make contact after this three-
step process is taken, the case is inactivated and would be considered meeting the rst condition listed above
for withdrawing from the justice process.
Ocer Behavior
How an ocer judges a victim’s rape report is reected in the ocer’s written report (Venema, 2016). Of course,
an ocer’s written report cannot capture the full context of a police-victim interaction. However, based on
Shaw et al.’s (2017) research, ocers’ written reports provide a workable proxy for measuring an ocer’s be-
havior during a police-victim interaction. Indeed, a “specic behavior to target for change” in sexual assault
investigations is police report writing (p. 611).
While one may question whether an ocer would document their own problematic behavior in a police
report, the examples above show that this does occur (along with numerous other examples from the studied
department that we do not have space to provide). Indeed, Shaw et al. (2017) recognized that ocers frequently
provide evidence of how police-victim interactions unfold in their written reports. Examining 248 sexual as-
sault reports, they found ocers routinely invoke rape myths in the ocial documentation. In more than half
of the cases, at least one statement was made regarding what ‘real rape’ looks like.
Shaw et al. (2017) support the view that systemic, rather than individual, failures are at play. ese interac-
tions’ root problem is organizational, and ocers only conduct themselves in the expected manner (Barrett &
Hamilton‐Giachritsis, 2013; R. Campbell & Raja, 2005; Martin, 2005). As such, ocers likely do not view these
less-than-ideal interactions as problematic. Instead, they are documenting the ‘proper’ way of conducting an
investigation. is is especially problematic when considering Shaw and colleagues’ (2016) ndings that as rape
myths within a report increase, investigative eorts decrease. e number of investigative steps taken during
a sexual assault investigation signicantly predicts the likelihood of an arrest being made and the case being
referred for prosecution. Accordingly, the behavior of police report writing in sexual assault cases “illuminates
an invaluable opportunity for intervention (Shaw et al., 2017, p. 611).
Measuring behavior in policing is critical. While many studies concentrate on attitude alone, the results
are subject to important limitations, including the Hawthorne eect, demand bias, and even overt faking by
respondents. Further, even where attitudinal changes are real, they are subject to quick change. Studies with a
sole focus on attitudinal change can lead to a false sense of ecacy, as demonstrated in a recent review of the
literature on the impact of de-escalation training in policing (Engel, McManus, & Herold, 2020).
Analytic Plan
is study uses a two-part analysis, and two dierent modeling techniques are used. First, the eect of o-
cer behavior on victim engagement is examined with a Bayesian two-sample hypothesis test. Second, ocer
behaviors are examined by using machine learning-based text analysis.
Bayesian Hypothesis Testing
We employ a Bayesian two-sample hypothesis test to examine the change in victim engagement between pre-
and post-training. We accomplish this by treating the pre- and post-training cases as separate parameters. We
estimate each parameter with beta distributions and then conduct Monte Carlo simulations to estimate the
amount of change, if any, and the probability of observing that change due to chance.
Machine Learning-Based Text Analysis
e historical problems with prohibitive time allocation and protracted human coding eorts are no longer
impediments to evaluating large corpora of word data. With the advent of machine learning, natural language
processing, and widespread digitalization of text, including police reports, scholars can obtain a richer under-
standing of police behavior that was not possible even a decade ago (Mourtgos & Adams, 2019). We utilize
machine learning-based text analysis to assess changes in ocial police reports documenting initial interviews
of rape reports. Doing so allows us better to understand ocers’ actions through their own words and at a
larger scale than usually allowed by more conventional qualitative analysis. Recalling that ocer reports are
necessarily only representative of the police’s view of the case, these reports should not be considered a full
accounting of the event. is limitation is well-known in policing research that can be over-reliant on agency-
generated records. However, for this study’s purposes, the initial written reports function as a record of ocer
behavior, rather than an attempt by the researchers to derive a nding about the victim, the victim’s perceptions,
or the case details.
Analysis and Results
Victim Engagement: Bayesian Hypothesis Testing
Conventional (i.e., frequentist) statistics assume there is only one true population parameter within the pop-
ulation (i.e., there is one true parameter coecient that is xed but unknown). However, within a Bayesian
framework, all unknown parameters incorporate uncertainty dened as a probability distribution. As a result,
Bayesian methods do not provide a single outcome value but rather a distribution with a probability that the
distribution contains the given parameter coecient (Van de Schoot & Depaoli, 2014). Barnes, TenEyck, Pratt,
and Cullen (2019) recently called for a shift to Bayesian statistical analysis in criminological research. A failure
to acknowledge and leverage uncertainty through Bayesian inferential methods can lead to an “increased rate
of false-negative results, inated false-discovery rates, and over-estimates of eect size,” and thus, a crisis of
condence in criminological research (Barnes et al., 2019, p. 1).
e uncertainty around parameter values is captured by a distribution dened before observing the data
called a prior distribution. Van de Schoot and Depaoli (2014) identify three main classes of priors. ese include
non-informative priors, weakly-informative priors, and informative priors. e variance, or precision, of the
prior reects a researcher’s level of certainty about the parameter of interest. is prior knowledge typically
stems from previous studies with similar data or expert knowledge.
e ‘principle of indierence’ advises researchers to assign equal probabilities to all events unless known a
priori that some events are more probable than others (Golan, 2018). Uninformative (or “at”) priors specify
equal plausibility for every possible value of the modeled parameters. Some researchers argue that uninforma-
tive priors should be the default because they maximize entropy (Golan, 2018) and reduce error probabilities
(Mayo, 2018).
Weakly-informative priors reect more certainty about population parameters than do non-informative
priors. Although weakly-informative priors typically have little inuence on the nal parameter estimate (McEl-
reath, 2020; Van de Schoot & Depaoli, 2014), they help constrain parameters to reasonable ranges and retain a
conservative estimate of any prior knowledge available about the parameter. Accordingly, it is often argued that
unless one truly believes that all possible events have an equal probability, weakly-informative priors should be
used in Bayesian analysis (McElreath, 2020).
When examining victim engagement in the present study, it is clear that a non-informative prior is not an ac-
curate reection of reality. Pre-training, approximately 77% of all victims were classied as withdrawing from
the justice process. Post-training, approximately 67% of victims were classied as withdrawing from the justice
process. With this in mind, a weakly-informative prior is more appropriate than an assumption that engage-
ment or a failure to retain have equal probabilities. We are then left with determining the prior distribution for
the analysis. To do this, we consult three dierent sources of prior information. First, the sergeant overseeing
the Special Victims Unit at the studied department was consulted (i.e., expert opinion). Based on seven years of
experience overseeing all sexual assault investigations for the studied department, the sergeant estimated that
approximately 30% of victims do not withdraw from the justice process. Second, internal records management
system data was examined for the two years before the training intervention. According to case closure codes
in the records management system, approximately 68% of victims did not withdraw from the justice process.
Finally, a recent publication using National Incident-Based Reporting System (NIBRS) data from 2011 to ex-
amine variables that aect sexual assault case closures was consulted. is study reported that, per NIBRS data,
only 8.2% of sexual assault cases were closed due to a victim ‘refusing to cooperate’ (in other words, 91.8% of
victims did not withdraw from the justice process).
ese three sources provide a wide range of estimates for victim engagement priors, even given what is
known about misused clearance codes (Spohn & Tellis, 2010). Discrepancies are found across the US with dif-
ferent agencies’ use of clearance codes, and discrepancies exist across detectives within the same police agency.
Accordingly, the prior for victim engagement based on expert opinion (30%) was retained. Given the uncer-
tainty, we used a wide variance (i.e., weak prior) for our analysis. ese conservative choices allow us to repre-
sent a distribution where 0.3 is the mean but allow for a wide range of possible alternative rates. Like any other
statistical assumption, the use of priors in Bayesian analysis should be interrogated. is maxim is especially
true when there is no strong argument for a particular prior over another, and there are several reasonable
choices, as in this instance. Accordingly, in Appendix A, we analyze the data with each of the three described
priors to test how sensitive inference is to the prior’s specication. We demonstrate that regardless of which
of the three priors are used, our results do not change. is outcome is expected as the likelihood function
(i.e., observed data) quickly overtakes the prior assumption with even moderate size samples, producing simi-
lar inferences (McElreath, 2020). In other words, just as in many other Bayesian applications, in large enough
samples, the data overwhelm the prior, regardless of the form that prior takes.
e second statistic necessary in a Bayesian method is the likelihood function. e likelihood function, in
its simplest sense, is the observed data. In this analysis, the dependent variable of interest is victim engagement.
We model the dependent variable with the probability density function (PDF) of the beta distribution. e PDF
of the beta distribution is dened as:
        
  
where represents the probability of an event, represents how many times we observe the ‘positive’ event
(i.e., a victim does not withdraw from the justice process), and represents how many times we observe the
positive event did not occur (i.e., a victim withdraws from the justice process). e equation’s denominator is
the beta function, which allows us to normalize the posterior distribution such that our distribution sums to 1,
giving us a workable probability (Kurt, 2019).
Each victim has a unique probability of staying engaged in the justice process or withdrawing. ese proba-
bilities of individual victim engagement also have a common distribution, which is the beta distribution. us,
the beta distribution is a probability distribution for probabilities (McElreath, 2020). By taking these unique
probabilities into account when estimating the overall probability distribution, we include more information
and more appropriately account for uncertainty in the model.
We have two sample parameters (pre- and post-training) to estimate. We arrive at a posterior distribution
for each parameter by estimating the PDF for each. Figure 1 shows the estimates for the pre- and post-training
parameters side by side. e data suggests that there was an increase in victim engagement post-training. How-
ever, it is also clear that there is overlap between the two parameters’ distributions. e distribution overlap
leaves open the possibility that post-training gains are possibly due to chance or that pre-training victim en-
gagement could be better than post-training. We use Monte Carlo simulation to interrogate further whether
post-training gains in victim engagement are due to chance or not.
Based on Figure 1, we should expect more possibilities where post-training victim engagement is better
than pre-training victim engagement. e more frequently we sample, the more precisely we can observe how
much more frequently post-training victim engagement is better than pre-training victim engagement. Once
all of the samples are drawn, the ratio between when post-training victim engagement is superior, and the total
number of possibilities can be calculated to arrive at an exact probability (Kurt, 2019).
One million samples were drawn from each distribution using a Monte Carlo simulation to establish an
exact probability. Next, the number of times post-training victim engagement was greater than pre-training
victim engagement was divided by the total number of samples. e results indicate that in 86% of the one
million samples, post-training victim engagement was better than pre-training victim engagement. us, based
on the distribution of all possible scenarios, in 86% of those scenarios, post-training victim engagement was
better than pre-training victim engagement. is nding indicates that even with a relatively small number
of observed sexual assault cases, we can have a relatively strong belief that post-training victim engagement is
better than pre-training victim engagement.
Returning to the analysis of victim engagement, we want to know not only if post-training victim engage-
ment is better, but also how much better. We can accomplish this by calculating how many times greater post-
training victim engagement is than pre-training victim engagement. We divide the post-training samples by
the pre-training samples and plot the results. Figure 2 shows the resulting density plot, which indicates that the
average expected improvement in victim engagement post-training is 32% (ratio of 1.32). is nding supports
our rst hypothesis: An improvement in police-victim interactions will lead to an increase in victim engage-
Sexual Assault Case Reports: Text Analysis
Before employing machine learning techniques on the text data in the police reports, the text required trans-
formation into a format a computer can ‘read. First, the text is tokenized. Tokenization entails separating the
text into pieces by separating each word from other words. In practice, this is accomplished by treating white
spaces and punctuation as word boundaries. Next, punctuation and numerical values are removed, followed
by the conversion of all letters to lowercase. Finally, stopwords are removed. Stopwords are ubiquitous words
(e.g., “the,” “of,” “to”) that are not useful for analysis. A custom stopword list of all names and locations was then
generated, and all instances were removed from the corpus to protect condentiality.
Once pre-processing was completed, the resulting text data were grouped into either the pre-training or
post-training period. Next, the frequency with which each word was used between periods was calculated and
plotted. In Figure 3, words near the red line were used with equal frequency pre- and post-training, while words
further away from the line were used more frequently in one period. Note that pre-training, words such as “bar”
and “club” (as in nightclub) were used more often by ocers compared to the post-training period. is change
is especially signicant, as the internal evaluation of how ocers had been handling sexual assault cases indi-
cated that ocers had been over-focusing their reports describing alcohol-use and intoxication levels of victims
(recall, Lonsway et al., 2001 observed the same change in behavior following their training intervention with
police recruits). Post-training, words indicating increased attention to victim well-being such as “advocate” (as
in victim advocate) and “drugged” were used more frequently by ocers. Further, it appears that post-training,
ocers were more actively suspect-focused. For example, the words “charges,” “aggravated,” and “arrests” are
used more frequently.
While this single-word analysis provides some insight, consecutive words, called n-grams, can also be ex-
amined. By seeing how word X followed by word Y is followed by word Z, greater insight into the context of
ocers’ writing can be obtained. e dierence in reports pre- and post-training is even starker when examin-
ing trigrams in Figure 4.
Post-training, ocers’ reports emphasize victims’ well-being: “rape recovery center” and “family justice
center” are phrases directly related to providing victim services to sexual assault victims. In the pre-training
period, the focus was more on the system’s needs: “SANE nurse responded” and “crime lab responded” (i.e., the
system needs evidence). Of further note are the increased use of “Gold Cross ambulance” and “public safety
building” post-training. During the review of past handling of sexual assault cases, one problem that was iden-
tied was that some ocers required rape victims to secure their own transportation to the hospital for medical
attention. In the provided training, an emphasis was placed on ocers facilitating this transport through the
local ambulance service and not burdening victims with that detail.
Further, very rarely are victims taken to the public safety building immediately after an initial report at
the subject department. e increase in the public safety building mentions indicates that suspects were being
taken to the police department for questioning by detectives more frequently than being questioned by patrol
ocers on scene and then released.
Finally, especially interesting is the choice of words used by ocers during the pre-training period: Specif-
ically, the more frequent use of “alleged sexual assault. e inclusion of the adjective “alleged” modies the
phrase, shifting meaning in a way that does not place a high level of veracity in victims’ statements. A shift in
this phrasing suggests that a better explanation of victim responses to a traumatic experience during training,
in a way that parallels an ocer’s own experience with traumatic experiences, pays dividends.
In sum, this algorithmic analysis of reports indicates that following training, ocers’ written reports were
more concerned with victim well-being, used more appropriate phrasing, and generally reected a positive
change in how ocers interacted with victims. ese ndings support our second hypothesis: e training
intervention lead led to improved police-victim interaction during initial sexual assault investigations and in-
creased victim engagement. ese eects are not likely to be the result of chance, but rather, the training inter-
Sexual violence receives signicant attention from the public, policymakers, and researchers (Franklin et al.,
2020). Much of the attention surrounding policing and sexual assault encompasses how to improve police
response to sexual assault victims and sexual assault investigations. While assessing police attitudes regarding
rape myths and attitudes towards victims is important, aecting and assessing what police actually do should
be of primary importance (Go, 2016; Lonsway et al., 2001).
Police departments typically provide specialized training to improve police response to sexual assault vic-
tims and sexual assault investigations. Most specialized training curricula tend to cover the same domain con-
tent, including the neurobiological response to trauma, interviewing techniques, legal and procedural issues,
crisis intervention, evidence collection, and attitudinal orientation toward the topic of rape (Lonsway et al.,
2001). e training program examined in this study was similar. However, this study sidesteps the traditional
focus on attitudinal outcomes and concentrates on assessing behavioral outcomes. We rst tested victim en-
gagement post-training and then discussed reasons why the attendant positive changes might occur. We found
evidence that victims were 32% more likely, on average, to remain engaged in the justice process post-training
and that there is a .86 probability that this increase in victim engagement is not due to chance. Examining o-
cer reports of sexual assault, we found that the training brought about positive behavioral change manifested in
ocers’ written reports exhibiting more concern for victim well-being and less focus on the extralegal factor
of the victim’s alcohol consumption, and less language questioning victim credibility.
We agree with Lonsway and colleagues (2001) that it is unfortunate that research in this particular domain
has not better documented what police actually do when responding to sexual assault investigations. Rather
than relying solely on attitudinal scales and surveys (Shaw et al., 2017), the eld should attempt to think more
creatively about measures that will more accurately assess desired changes in performance (Lonsway et al., 2001;
Shaw et al., 2016). We accomplished this through the use of ocial police reports written by initial responding
ocers. We are not the rst to use written reports to evaluate police response to sexual assault (Shaw et al.,
2016, 2017). We build upon previous work through the use of machine learning text analysis, which allows
for a more objective and reproducible evaluation of word data than conventional human coding (Mourtgos
& Adams, 2019). ese techniques reduce the cost and personnel required to analyze large volumes of ocer
reports and provide an opportunity to recover potentially valuable patterns that have been mainly inaccessible
to researchers and practitioners.
Our results indicate that focusing on police behavior rather than police attitudes may be a fruitful path for
policing scholars. e mixed ndings of previous research on police sexual assault training may not accurately
reect training ecacy but instead related to what is chosen to be measured. Lonsway et al.’s (2001) study lends
credence to this possibility as their results found a positive change in police behaviors post-training, without a
corresponding change in police attitudes.
One limitation of our study again points to what is chosen to be measured. Based on prior research, we used
ocers’ written reports as a proxy for ocer behavior during police-victim interactions (Shaw et al., 2016,
2017). While we believe our measure is a workable proxy for ocer behavior during a police-victim inter-
action, as stated previously, it clearly cannot capture the full context of an interaction. is study’s design
assumes that ocers’ increased documentation of referrals to victim services and victim-centered language in
their reports reects an improved police-victim interaction. Of course, it is possible that the training changed
the organizational demands perceived by ocers to note these activities in their reports, rather than an actual
improvement in interactions. Without observations of the police-victim interactions themselves, it cannot be
denitively concluded that interactions improved—a combination of improved interactions and a change in
perceived organizational demands is likely. A research design that includes observational data or interviews
with victims should be considered in combination with the text analysis approach we used in future research
assessing the ecacy of similar training programs.
Second, it remains unclear if simply providing specialized training to ocers improves police-victim in-
teractions or if there is something specic about the training developed at the subject police department that
resulted in the observed positive outcomes. Our research design is not able to parse the relationship between
specic training and outcomes at that level. Regardless, practitioner inclusion in the development and evalua-
tion of the training is critical. It provides a more nuanced view of the complex issues involved when working
with police ocers and sexual assault investigations (Huey & Mitchell, 2016). Similar to the causal questions
addressed below, we should be careful not to overstate the evidence of a link between ocer report-writing be-
havior and subsequent victim decisions to interact (or not) with the police investigation. is limitation could
begin to be addressed with qualitative interviews of ocers, with a focus on uncovering the link between the
two outcomes studied.
irdly, the ndings here should be considered within the limits of causal identication. In short, the design
here is not a true experiment, and so the detected eects could have been confounded by unmeasured eects.
For example, there was prior negative publicity surrounding how police responded to sexual assault cases and
victims. ere is a possibility that ocers were individually responsive to the negative attention, and changed
their behavior. is design limitation is fundamental but should also be contextualized by time precedence, in
that in the present study, the pre-training phase included reports that followed the public outcry, and yet the
reports from that period were not as victim-centered as the reports in the post-training period. e lack of a
true experimental design means the detected eects could also theoretically be confounded through turnover of
the existent front-line ocer cohort and replacement with newly hired ocers. Because we specically study
ocer turnover in this agency in parallel research, we are condent this vulnerability is not seen in the currently
studied agency (blinded, under review). However, causal identication is vulnerable to this class of confounding
eects generally, and so generalizing our ndings to other agency contexts need to consider how agency and
personnel responses could aect similar training. We recommend that researchers attempt to replicate the
results reported here using a true experimental design to confront the causal identication problem inherent
to non-experimental designs.
Fourth, it should also be recognized that while the observed improvements are important, the majority of
victims still did not stay engaged in the justice process, even after the training was provided (approximately
67%). ere are other factors at play when it comes to victim engagement in sexual assault investigations. One
such likely factor is interactions with, and actions taken by, follow-up investigators. While our study explicitly
examined police-victim interactions during the initial report and their eect on victim engagement, further
research should assess the impact of police-victim interactions with follow-up investigators.
Finally, it is currently unknown if the training eects were durable. ere is always a possibility that the
training eect wore o over time. Expanded longitudinal analyses could assess this possibility. Further, victim
engagement in this study was measured by continued engagement with the police. Due to data limitations, we
could not examine if the increase in victim engagement further increased prosecutions or guilty verdicts.
It is incumbent upon police departments to engage in data-driven policing by scientically testing interven-
tions to determine if those interventions had the desired eect (Engel, McManus, & Isaza, 2020). Despite its
limitations, the current study’s resulting increase in victim engagement, alongside a high probability that that
increase resulted from the department’s training, suggests that the training holds promise. Further, the addi-
tional evidence provided by the text analysis oers support that the training did indeed cause a positive change
in police-victim interactions, as reected in ocers’ written reports. In sum, we nd that a relatively small
training intervention was associated with better police-victim interactions and a subsequent increase in victim
engagement in the justice process.
Adams, I., & Mastracci, S. (2017). Visibility is a trap: e ethics of police body-worn cameras and control.
Administrative eory & Praxis, 39(4), 313–328.
Ahrens, C. E. (2006). Being silenced: e impact of negative social reactions on the disclosure of rape. Amer-
ican Journal of Community Psychology, 38(3–4), 31–34.
Alderden, M. A., & Ullman, S. E. (2012). Creating a more complete and current picture: Examining police
and prosecutor decision-making when processing sexual assault cases. Violence Against Women, 18(5), 525–
Barnes, J. C., TenEyck, M. F., Pratt, T. C., & Cullen, F. T. (2019). How powerful is the evidence in criminology?
On whether we should fear a coming crisis of condence. Justice Quarterly.
Barrett, E. C., & Hamilton‐Giachritsis, C. (2013). e victim as a means to an end: Detective decision making
in a simulated investigation of attempted rape. Journal of Investigative Psychology and Oender Proling, 10(2),
Bouard, J. A. (2000). Predicting type of sexual assault case closure from victim, suspect, and case charac-
teristics. Journal of Criminal Justice, 28(6), 527–542.
Campbell, B., Lapsey Jr, D., & Wells, W. (2019). An evaluation of Kentucky’s sexual assault investigator
training: Results from a randomized three-group experiment. Journal of Experimental Criminology. https:
Campbell, R. (2008). e psychological impact of rape victims’ experiences with the legal, medical, and
mental health systems. American Psychologist, November, 702–717.
Campbell, R., & Fehler‐Cabral, G. (2018). Why police “couldn’t or wouldn’t” submit sexual assault kits for
forensic DNA testing: A focal concerns theory analysis of untested rape kits. Law & Society Review, 52(1),
Campbell, R., & Raja, S. (2005). e sexual assault and secondary victimization of female veterans: Help-
seeking experiences with military and civilian social systems. Psychology of Women Quarterly, 29(1), 97–106.
Campbell, R., Se, T., Barnes, H. E., Ahrens, C. E., Wasco, S. M., & Zaragoza-Diesfeld, Y.(1999). Community
services for rape survivors: Enhancing psychological well-being or increasing trauma? Journal of Consulting
and Clinical Psychology, 67(6), 847–858.
Darwinkel, E., Powell, M., & Tidmarsh, P. (2013). Improving Police Ocers’ Perceptions of Sexual Oending
rough Intensive Training. Criminal Justice and Behavior, 40(8), 895–908.
Di Nota, P. M., Stoliker, B. E., Vaughan, A. D., Andersen, J. P., & Anderson, G. S. (2020). Stress and mem-
ory: A systematic state-of-the-art review with evidence-gathering recommendations for police. Policing: An
International Journal, 44(1), 1–17.
Engel, R. S., McManus, H. D., & Herold, T. D. (2020). Does de-escalation training work? Criminology &
Public Policy, 19(3), 721–759.
Engel, R. S., McManus, H. D., & Isaza, G. T. (2020). Moving beyond “Best Practice”: Experiences in police
reform and a call for evidence to reduce ocer-involved shootings. e ANNALS of the American Academy of
Political and Social Science, 687(1), 146–165.
Estrich, S. (1987). Real rape. Harvard University Press.
Feist, A., Ashe, J., Lawrence, J., McPhee, D., & Wilson, R. (2007). Investigating and detecting recorded of-
fences of rape [Data set]. Home Oce.
Franklin, C. A., Garza, A. D., Goodson, A., & Bouard, L. A. (2020). Police perceptions of crime victim behav-
iors: A trend analysis exploring mandatory training and knowledge of sexual and domestic violence survivors’
trauma responses. Crime & Delinquency, 66(8), 1055–1086.
Frohmann, L. (1991). Discrediting victims’ allegations of sexual assault: Prosecutorial accounts of case
rejections. Social Problems, 38(2), 213–226.
Go, P. A. (2016). Identity traps: How to think about race & policing. Behavioral Science & Policy, 2(2),
Golan, A. (2018). Foundations of info-metrics: Modeling, inference, and imperfect information. Oxford
University Press.
Goodman‐Delahunty, J., & Graham, K. (2011). e inuence of victim intoxication and victim attire on
police responses to sexual assault. Journal of Investigative Psychology and Oender Proling, 8(1), 22–40. https:
Hine, B., & Murphy, A. (2019). e inuence of ‘high’ vs. ‘Low’ rape myth acceptance on police ocers’
judgements of victim and perpetrator responsibility, and rape authenticity. Journal of Criminal Justice, 60,
Huey, L., & Mitchell, R. J. (2016). Unearthing Hidden Keys: Why Pracademics Are an Invaluable (If Under-
utilized) Resource in Policing Research. Policing, 10(3), 300–307.
Kelly, L., Lovett, J., & Regan, L. (2005). A gap or a chasm?: Attrition in reported rape cases (research study
293) [Data set]. Home Oce.
Kerstetter, W. A. (1990). Gateway to justice: Police and prosecutorial response to sexual assaults against
women. e Journal of Criminal Law and Criminology, 81(2), 267–313.
Kerstetter, W. A., & Van Winkle, B. (1990). Who decides?: A study of the complainant’s decision to prosecute
in rape cases. Criminal Justice and Behavior, 17(3), 268–283.
Konradi, A. (2007). Taking the stand: Rape survivors and the prosecution of rapists. Praeger.
Kurt, W. (2019). Bayesian statistics the fun way: Understanding statistics and probability with Star Wars,
Lego, and rubber ducks. No Starch Press, Inc.
Lonsway, K. A., Welch, S., & Fitzgerald, L. F. (2001). Police Training in Sexual Assault Response: Process,
Outcomes, and Elements of Change. Criminal Justice and Behavior, 28(6), 695–730.
Martin, P. Y. (2005). Rape work: Victims, gender, and emotions in organization and community context.
Mayo, D. G. (2018). Statistical inference as severe testing: How to get beyond the statistics wars. Cambridge
University Press.
McElreath, R. (2020). Statistical rethinking: A Bayesian course with examples in R and Stan (2nd ed.). CRC
Mourtgos, S. M., & Adams, I. T. (2019). e rhetoric of de-policing: Evaluating open-ended survey responses
from police ocers with machine learning-based structural topic modeling. Journal of Criminal Justice. https:
Mourtgos, S. M., Adams, I. T., Nix, J., & Richards, T. N. (2021). Mandatory sexual assault kit testing policies
and arrest trends: A natural experiment. Justice Evaluation Journal.
Pattavina, A., Morabito, M., & Williams, L. M. (2016). Examining connections between the police and pros-
ecution in sexual assault case processing: Does the use of exceptional clearance facilitate a downstream orien-
tation? Victims & Oenders, 11(2), 315–334.
Patterson, D. (2011). e linkage between secondary victimization by law enforcement and rape case out-
comes. Journal of Interpersonal Violence, 26(2), 328–347.
Shaw, J., Campbell, R., & Cain, D. (2016). e view from Inside the system: How police explain their response
to sexual assault. American Journal of Community Psychology, 58(3–4), 446–462.
Shaw, J., Campbell, R., Cain, D., & Feeney, H. (2017). Beyond surveys and scales: How rape myths manifest
in sexual assault police records. Psychology of Violence, 7(4), 602–614.
Singleton Jr., R. A., & Straits, B. C. (2018). Approaches to social research (6th ed.). Oxford University Press.
Sleath, E., & Bull, R. (2012). Comparing Rape Victim and Perpetrator Blaming in a Police Ocer Sample:
Dierences Between Police Ocers With and Without Special Training. Criminal Justice and Behavior, 39(5),
Sleath, E., & Bull, R. (2017). Police perceptions of rape victims and the impact on case decision making: A
systematic review. Aggression and Violent Behavior, 34, 102–112.
Spohn, C., & Tellis, K. (2010). Justice denied? e exceptional clearance of rape cases in Los Angeles. Albany
Law Review, 74(3), 1379–1421.
Spohn, C., & Tellis, K. (2019). Sexual assault case outcomes: Disentangling the overlapping decisions of
police and prosecutors. Justice Quarterly, 36(3), 383–411.
Temkin, J. (1999). Reporting rape in London: A qualitative study. e Howard Journal of Criminal Justice,
38(1), 17–41.
Ullman, S. E., & Filipas, H. H. (2001). Correlates of formal and informal support seeking in sexual assault
victims. Journal of Interpersonal Violence, 16(10), 1028–1047.
Valentine, J. L., Sekula, L. K., Cook, L. J., Campbell, R., Colbert, A., & Weedn, V. W. (2019). Justice Denied:
Low Submission Rates of Sexual Assault Kits and the Predicting Variables. Journal of Interpersonal Violence,
34(17), 3547–3573.
Van de Schoot, R., & Depaoli, S. (2014). Bayesian analyses: Where to start and what to report. e European
Health Psychologist, 16(2), 75–84.
Venema, R. M. (2016). Police Ocer Schema of Sexual Assault Reports: Real Rape, Ambiguous Cases, and
False Reports. Journal of Interpersonal Violence, 31(5), 872–899.
Ylang, N., & Holtfreter, K. (2019). e Decision to arrest in sexual assault case processing: A test of Black’s
theory of the behavior of law. Violence Against Women.
... Outcomes from existing evaluations are mixed. Some found training increased knowledge of investigative practices (Lonsway et al., 2001), reduced victim blame (Darwinkel et al., 2013), improved officer behavior in hypothetical (Tidmarsh et al., 2021) and actual cases (Mourtgos et al., 2021), and decreased rape myth acceptance (B. A. Campbell et al., 2020). Others found training did not reduce officers' adherence to rape myths (Lee et al., 2012;Lonsway et al., 2001). ...
... In this way, researchers can evaluate the impact of training, impulsivity, and education on behavior in addition to cognitive and attitudinal outcomes. Recent studies of officer behavior before and after traumainformed training found that a short (4-hr) program increased victim engagement with investigations (Mourtgos et al., 2021) and a longer (96-hr) training improved officer performance in mock interviews (Tidmarsh et al., 2021). Third, our sample was relatively homogeneous in regard to race/ethnicity and largely comprised White officers (95%). ...
Full-text available
Few studies have evaluated the effectiveness of sexual assault investigations training using experimental designs. Existing studies have not examined the impact of officers’ levels of impulsivity and education on training effectiveness. Using a Solomon four-group quasi-experimental design to assess pretesting effects, we examined the impact of training, impulsivity, and education on officers’ ( N = 432) adherence to rape myths and knowledge of victim reporting behaviors. Ordinary least squares (OLS) models were estimated to examine main effects of training, and moderating effects of impulsivity and education on training for our outcome variables. Results demonstrated that training, impulsivity, and education predicted improvements in attitudinal and cognitive outcomes. However, neither impulsivity nor education moderated—or changed—the effectiveness of training. In addition, training effects held over time, and we did not detect evidence of pretesting effects. Findings from this study improve our understanding of police sexual assault investigations training and provide methodological advancements for police training evaluations.
... For example, several recent studies have demonstrated that training can improve officers' short-and long-term knowledge of trauma-informed investigative techniques and perceptions of victims (B. Campbell, Lapsey Jr., & Wells, 2020Franklin, Garza, Goodson, & Bouffard, 2019;Lathan, Langhinrichsen-Rohling, Duncan, & Stefurak, 2019;Tidmarsh, Hamilton, & Sharman, 2020), behavior in mock interviews (Tidmarsh, Sharman, & Hamilton, 2021), and victim engagement with investigators in actual cases (Mourtgos, Adams, & Mastracci, 2021). Understanding case attrition in sexual assault investigations can help identify the policy-and practicerelevant factors associated with increased victim participation. ...
Purpose Using data collected from a sample of cases associated with untested sexual assault kits (SAKs), we used focal concerns theory – a framework that identifies themes relevant to criminal justice practitioners' decision-making processes – to assess the correlates of suspect identification and officers' pre-arrest decision to present a case to prosecutors. Methods We estimated two logistic regression models. The first model examined the ability of investigators to identify a suspect in the full sample (N = 444). The second model examined officers' decision to present cases to the prosecutor in a subsample of cases in which a suspect was identified (N = 293). Results The odds of suspect identification increased when victims cooperated with investigators and witnesses were available, and decreased in stranger cases and when victim credibility concerns were present in police reports. Cases were more likely to be presented to a prosecutor when victims cooperated with police, witnesses were available, suspects confessed to the crime, and suspects had a history of arrest. Conclusions Researchers should continue to examine pre-arrest investigative actions to inform policy and training aimed at enhancing victim engagement and reducing case attrition. Additionally, SAKs should be tested to aid investigators in identifying suspects in stranger cases.
Full-text available
Purpose: To synthesize recent empirical research investigating memory of stressful critical incidents (both simulated and occurring in the field) among law enforcement officers. Design: Systematic state-of-the-art review. Findings: 20 studies of police and military officers show reduced detail and accuracy of high-versus low-stress incidents, especially for peripheral versus target information. Decrements in memory performance were mediated by the extent of physiological stress responses. Delayed recall accuracy was improved among officers that engaged in immediate post-incident rehearsal, including independent debriefing or reviewing body-worn camera footage. Research limitations/implications: Most studies were not found through systematic database searches, highlighting a need for broader indexing and/or open access publishing to make research more accessible. Practical implications: By understanding how stress physiology enhances or interferes with memory encoding, consolidation, and recall, evidence-based practices surrounding post-incident evidence gathering are recommended. Social implications: The current review addresses common public misconceptions of enhanced cognitive performance among police relative to the average citizen. Originality/value: The current work draws from scientific knowledge about the pervasive influence of stress physiology on memory to inform existing practices surrounding post-incident evidence gathering among police.
Full-text available
In post-Ferguson America, police departments are being challenged to implement evidence-based changes in policies and training to reduce fatal police-citizen encounters. Of the litany of recommendations believed to reduce police shootings, five have garnered widespread support: body-worn cameras, de-escalation training, implicit bias training, early intervention systems, and civilian oversight. These highly endorsed interventions, however, are not supported by a strong body of empirical evidence that demonstrates their effectiveness. Guided by the available research on evidence-based policing and informed by the firsthand experience of one of the authors in implementing departmental reforms that followed the fatal shooting of a civilian by an officer, this article highlights promising reform strategies and opportunities to build the evidence base for effective use-of-force reforms. We call upon police executives to engage in evidence-based policing by scientifically testing interventions, and we call on academics to engage in rapid research responses for critical issues in policing.
Full-text available
Objectives This study evaluated a 40-h statewide sexual assault investigations training program implemented in Kentucky in 2017. The study examined short- and long-term effects of training on police officers’ self-reported rape myth acceptance, knowledge of state laws, and knowledge of trauma-informed practices. Methods Using a randomized design to control for pre-test sensitization, we assigned training courses to three groups: Group A (pre- and post-training assessment), Group B (post-training only assessment), and Group C (pre-training only assessment). Mean scores for rape myth acceptance, knowledge of state laws, and trauma-informed practices were compared using t tests to assess differences between Group A’s pre- and post-test scores, as well as differences between Groups B’s post-test and C’s pre-test scores. OLS models were estimated to assess the effects of training between treatment (Groups A and B) and control (Group C) groups. Follow-up assessments were sent to participants to examine long-term training effects on all outcome variables. Results Findings indicated that training was effective in reducing rape myth acceptance and increasing knowledge of state laws and knowledge of trauma-informed practices. These effects remained stable over time, evidenced by analyses of data from follow-up assessments. Finally, the pre-test did not have a priming effect on any outcome variables. Conclusions This study found that sexual assault training can be effective at improving police perceptions of victims, and knowledge of laws and trauma-informed practices for sexual assault investigations. Additionally, this study provided evidence that three-group experimental designs are feasible for evaluating law enforcement training programs.
Full-text available
Highlights Machine learning-based textual analysis is a viable tool for police survey research Analyzing large numbers of police free-text responses provides more nuanced understanding of police perceptions of the public Officers' attention to professionalism guards against de-policing, while attention to perceived unfair criticism increases it The public's integrity has a stronger effect on propensity to de-police than the public's knowledge about police work
Full-text available
Police perceptions of a victim’s self-presentation style can have an impact on secondary victimization, case processing, and public safety. Trauma survivors may present to police with flat or restricted affect, emotional numbing, and disjointed recollections. Often, police personnel have misperceived manifestations of trauma as indicators of reliability and credibility. Using a trend design, this study employed a sample of 979 police from one of the five largest U.S. cities to examine the relation between trauma-informed training and endorsement of trauma misperceptions. Multivariate ordinary least squares (OLS) regression models were estimated and revealed mean adherence to trauma misperceptions was significantly lower among participants who had completed training, controlling for demographic, occupational, and attitudinal variables. Implications and future research are discussed.
The present study employs a quasi-experimental design to evaluate the effects of a mandatory sexual assault kit (SAK) testing policy on rape arrests in a large western US jurisdiction. We use a Bayesian structural time-series model and monthly data on arrests for rape from 2010 through 2019. In the post-implementation period, we observed a downward trend in the arrest rate for rape. Based on the results, the most conservative interpretation of our findings is that the policy implementation did not affect rape arrest rates. While mandatory SAK testing policies are often advocated for based on the belief that they will increase arrest rates for sexual assault (among other proposed benefits), we add to growing empirical evidence that policy interventions beyond mandatory SAK testing are needed to increase arrest rates for sexual assault. Jurisdictions that currently use mandatory SAK testing policies are encouraged to assess stakeholders' experiences to proactively address resource allocation, consider other policies that may increase accountability for sexual assault offenders, and utilize victim service providers to support other measures of success with victims in instances where no arrest is made.
De‐escalation training has been widely implemented by U.S. police agencies in the wake of adverse public reaction to recent controversial police use‐of‐force incidents. Despite vast promotion from politicians, academics, expert panels, and the public, we know little about the effects of de‐escalation training on officers and police–citizen interactions. In this article, we offer findings from a multidisciplinary systematic literature review that demonstrate limited knowledge concerning the impact of de‐escalation training across all professions. We identified 64 de‐escalation training evaluations conducted over a 40‐year period, primarily in the fields of nursing and psychiatry. Although assessment outcomes reveal few adverse consequences and provide some confidence that de‐escalation trainings lead to slight‐to‐moderate individual and organizational improvements, conclusions concerning the effectiveness of de‐escalation training are limited by the questionable quality of almost all evaluation research designs. As such, important questions regarding the impact of de‐escalation training for police remain. Given the critical impact that de‐escalation training could have on officers and the public they serve, we conclude with a direct call to academics, practitioners, and funders across the field of policing to prioritize as soon as possible the testing of de‐escalation and other police use‐of‐force policies, tactics, and training.
Past research applying Donald Black’s theory of the behavior of law to sexual assault case processing has focused on victim decisions to report the crime to the police. This study builds on and extends prior research by examining the next stage of legal mobilization (i.e., arrest). Using secondary data on 310 cases from the 1982-2012 Sexual Assault Kit Backlog Study in Los Angeles, California, the current study explores the effects of victim, offender, and case characteristics on arrest. The results suggest limited support for the theory in this victimization context. Implications for theory, research, and criminal justice practice are discussed.
A crisis of confidence has struck the behavioral and social sciences. A key factor driving the crisis is the low levels of statistical power in many studies. Low power is problematic because it leads to increased rates of false-negative results, inflated false-discovery rates, and over-estimates of effect sizes. To determine whether these issues impact criminology, we computed estimates of statistical power by drawing 322 mean effect sizes and 271 average sample sizes from 81 meta-analyses. The results indicated criminological studies, on average, have a moderate level of power (mean = 0.605), but there is variability. This variability is observed across general studies as well as those designed to test interventions. Studies using macro-level data tend to have lower power than studies using individual-level data. To avoid a crisis of confidence, criminologists must not ignore statistical power and should be skeptical of large effects found in studies with small samples.
This book provides a framework for info-metrics-the science of modeling, inference, and reasoning under conditions of noisy and insufficient information. Info-metrics is an inherently interdisciplinary framework that emerged from the intersection of information theory, statistical inference, and decision-making under uncertainty. It allows us to process the available information with minimal reliance on assumptions that cannot be validated. This book focuses on unifying all information processing and model building within a single constrained optimization framework. It provides a complete framework for modeling and inference, rather than a problem-specific model. The framework evolves from the simple premise that our available information is often insufficient to provide a unique answer for decisions we wish to make. Each decision, or solution, is derived from the available input information along with a choice of inferential procedure. The book contains many multidisciplinary applications that demonstrate the simplicity and generality of the framework in real-world settings: These include initial diagnosis at an emergency room, optimal dose decisions, election forecasting, network and information aggregation, weather pattern analyses, portfolio allocation, inference of strategic behavior, incorporation of prior information, option pricing, and modeling an interacting social system. This book presents simple derivations of the key results that are necessary to understand and apply the fundamental concepts to a variety of problems. Derivations are often supported by graphical illustrations. The book is designed to be accessible for graduate students, researchers, and practitioners across the disciplines, requiring only basic quantitative skills and a little persistence.