Skip to main content
SearchLoginLogin or Signup

Sexual homicide and the forensic process: The decision-making process of collecting and analyzing traces and its implication for crime solving

Published onSep 24, 2022
Sexual homicide and the forensic process: The decision-making process of collecting and analyzing traces and its implication for crime solving
·

Introduction

The crime scene is the starting point of any traditional criminal investigation (Crispino, 2008; De Forest, 1999; Martin et al., 2010; Schuliar, 2009). The crime scene and the traces left are especially crucial in cases of sexual homicide as both are necessary to identify them as such. As suggested by the Federal Bureau of Investigation (FBI) (Ressler et al., 1988), for a homicide to be considered as sexual, it has to present at least one of the following characteristics at the crime scene: victim’s attire or lack of attire, exposure of the sexual parts of the victim’s body, sexual positioning of the victim’s body, insertion of foreign objects into the victim’s body cavities, evidence of sexual intercourse, and evidence of substitute sexual activity, interest, or sadistic fantasy.

Important pitfalls impede the investigation of sexual homicide. For instance, sexual homicide presents a low base rate – between 1% and 4% of all homicides (Beauregard & Martineau, 2016) – which makes gaining valuable and reliable knowledge about these crimes and offenders challenging. More specifically, accumulating knowledge that can inform effective investigative practices has been problematic, as many investigators cannot rely upon investigative experience to approach these crimes. In addition, while the behavioral elements of sexual homicide crimes differ from other types of violent crime, they are still an inherently heterogenous group. Each case may vary in terms of the modus operandi and ritualistic behavior exhibited, causing each case to appear unique. The differential characteristics as well as the possible presence of psychopathology, which may be identified through a careful study of the crime scene, may appear atypical for the detective involved in such a case for the first time. Due to these various investigative challenges, traces are critical to the criminal investigation and its outcome. The focus of the current study is to examine the collection and analysis of traces that are related to crime scene behaviors in sexual homicide cases as well as the factors influencing the solving of these crimes.

Crime scene investigation and the use of traces

The use of traces in a criminal investigation follows a three-step decision process (Bitzer et al., 2015; Delémont et al., 2018): (1) the decision to attend a scene, (2) the decision to search for and collect traces, and (3) the triaging of traces for forensic analysis. The gradational flow of this process limits the impact traces can have on the course of investigation, as well as on the resolution of the case. For example, traces left at a crime scene that is not attended by the police and therefore not investigated, cannot be used in the later stages of the judicial process.

Once the police have decided to attend a crime scene and proceed to the investigation stage, the criminal investigation faces the issue of to ‘find’. As suggested by Kind (1994), to ‘find’ is to search for traces in the context of crime scene investigation. Fundamentally, to ‘find’ includes a chance component. However, the investigation is active in its nature, and its outcome depends on what one is looking for. At best, the search for traces results in their detection, which implies the crucial capability of recognizing traces as such. While this is easily conceivable for visible traces (e.g., fingermarks, shoe marks), biological traces – especially contact traces – are often not visible at the crime scene. This is why the concept of trace detection needs to be understood in a wider sense. Specifically, adopting the hypothetical-deductive reasoning process allows for the “imagination” of possible traces at specific locations according to the case scenario or the presence of other traces (Fann, 1970). The crime scene investigator must thus “imagine” or infer which traces may have been exchanged during the criminal event, which ones may remain, and which ones could be relevant to the case at hand and provide answers to the investigative questions (Delémont et al., 2012).

To find the most relevant traces at a crime scene, the search for traces must be systematic and based on cognitive and observational skills. A broad understanding of the criminal and immediate environments as well as the range and properties of traces is also required (Delémont et al., 2012, 2014; Resnikoff et al., 2015; Ribaux et al., 2010) . The factors influencing the search for traces can be separated into three categories (Hazard and Margot, 2014, Hazard, 2014): (1) human (i.e., the knowledge), (2) structural (i.e., the organization, strategy, resources), and (3) contextual (i.e., the type of encountered situations). Moreover, finding traces may also be influenced by the behavior of offenders, such as their ability to limit the traces they leave and their forensic awareness (James & Beauregard, 2020).

Once traces have been identified and recognized as such, a decision regarding their collection is made. Material considerations, as well as cognitive aspects, need to be taken into account in this decision. The material aspect refers to components such as the quality of the trace and the availability of technological resources. The cognitive aspect includes a retrospective dimension (i.e., the knowledge of traces, their properties, and their collection techniques) as well as a “prospective orientation of the scene examiner to a set of likely investigative and evidential trajectories to which these artefacts may become decisive – or at least relevant” (Williams, 2007, p.763).

In parallel with these studies on the decision-making process underlying the search and the collection of traces, researchers worked for a better understanding of the true contribution of forensic science to security and justice. Rather than focusing on the factors influencing the early steps of the investigation process, end-to-end studies (ANZPAA NIFS, 2012, 2016; Bradbury & Feist, 2005; Brown et al., 2015; Bruenisholz et al., 2019; Home Office, 2007; Mapes et al., 2015) consider the processing of traces through the entirety of the judicial process. This method allows for trace evidence to be used at different steps. Metrics include the timeliness and number of crime scene investigations, analysis submissions, and reports. For burglary cases, Bruenisholz et al. (2019) determined that between 2011 and 2015, 10 to 40% of cases resulted in the collection and submission of traces for analysis. Furthermore, the authors found that traces were increasingly used in 2015 compared to 2011, with more traces submitted and analyzed yearly, thus positively impacting the arrest rate. Mapes et al. (2015) determined that biological traces were collected and analyzed in roughly 60% of serious crime cases but only in 6% of other high-volume crimes (e.g., burglaries, theft, vandalism, etc.). For homicide, it was found that in 97% of cases, physical traces were collected, and in 88% of cases, these traces were submitted for analysis (Baskin & Sommers, 2010).

Interestingly, research has suggested that the collection and analysis of these traces does not guarantee a proportional influence on the outcome of the investigation. For instance, Brodeur (2005) found that traces played a role in the identification of the suspect in only 1% of homicide cases. Ten years later, Mapes et al. (2015) have shown that a suspect was identified through a hit in the DNA database in only 1% of high-volume crimes and in merely 3% of serious crimes. Similarly, the presence of traces was not predictive of any of the judicial stages (i.e., arrest, charging, conviction) in homicide cases (Baskin and Sommers, 2010). However, using the same dataset but a different modeling approach, Peterson et al. (2013) were able to determine a link between the presence of traces and the arrest of a suspect.

While the importance of the early steps of the crime scene investigation has been recognized (Julian et al., 2012; Ludwig et al., 2012; Wyatt, 2014), little empirical research has been conducted on the identification of factors involved in the search, detection, and collection of traces at the crime scene of sexual homicide cases (Delémont et al., 2018) as well as their impact on the outcome of the case. Thus, the aim of this study is twofold: (1) to determine the factors influencing trace collection and analysis from offenders’ crime scene behaviors, and (2) to study the impact trace collection and analysis has on the resolution of the sexual homicide cases.

Methodology

Data and Sample

Data used in this study come from the Sexual Homicide International Database (SHIelD). This database provides information on 762 solved (i.e., cases in which police investigators identified and charged a suspect) and unsolved (i.e., where an offender has not been identified by investigators) cases of extrafamilial (i.e., stranger or acquaintance relationships) sexual homicides that occurred in France and Canada between 1948 and 2018. Data was collected in Canada and France with the same tool created to reference information on victims, offenders, and the crime-commission process with a total of 126 variables (see Chopin & Beauregard, 2019a). Data included in the database comes from various sources of information (e.g., investigative reports, offender interview reports, autopsy reports provided by pathologists, psychological reports provided by a team of forensic psychologists, and reports provided by forensic experts) and were compiled by crime analyst experts in violent crimes who used the same standardized grid of data collection in both Canada and France. Sexual homicide cases were identified using the FBI definition from Ressler et al. (1988) which states that for a homicide to be considered as sexual, it has to present at least one of the following characteristics at the crime scene: victim’s attire or lack of attire; exposure of the sexual parts of the victim’s body; sexual positioning of the victim’s body; insertion of foreign objects into the victim’s body cavities; evidence of sexual intercourse; evidence of substitute sexual activity, interest, or sadistic fantasy. As the FBI definition has been criticized for potentially presenting false positives (see Beauregard & Martineau, 2017; Chopin & Beauregard, 2021; Kerr et al., 2013; Stefanska et al., 2016), all cases included in SHIelD present at least two criteria of the sexual homicide definition. This methodological decision was made to avoid the inclusion of cases that were only characterized by the victim being found without or lack of attire.

For the current study, we made several choices to select a sample of 230 sexual homicide cases (see Figure 1). First, we decided to include only cases that occurred in Canada in order to avoid differences in practices for collecting forensic samples between the two countries. Second, we included only cases that occurred after 1990 to ensure that DNA sampling was practiced by forensic services in Canada1. We tested the association between the continuous temporal distribution of the collection and analysis of traces. Results showed no significant difference over time (β = -0.002; p = 0.951).

[Insert Figure 1 here]

Measures

Dependent variables. In this study, we computed two different models to predict two dependent variables. The first dichotomous dependent variable identified whether traces were collected and analyzed (0 = no traces were collected and analyzed; 1 = traces were collected and analyzed). The traces collected and analyzed included the following: hair, saliva, blood, semen, and fingerprints. This dependent variable was included as an independent variable in the second model computed in this study. The second dichotomous dependent variable identified the status of the cases at the time of the data collection (0 = unsolved, 1= solved)2.

Independent variables. The selection of independent variables was guided by previous studies focusing on crime scene behaviors predicting the collection and analysis of traces (Ribaux et al., 2010), as well as research examining crime scene behaviors in sexual homicide cases (Balemba et al., 2014; Beauregard & Martineau, 2014; 2016). We selected a total of 12 variables (i.e., 11 dichotomous, and 1 continuous) describing crime scene characteristics observable at both solved and unsolved cases. This information comes from reports provided by coroners, investigators, and forensic experts which included the following: (1) evidence of sexual penetration (i.e., there is evidence that the offender perpetrated vaginal and/or anal penetration of the victim), (2) victim’s body was burned, (3) victim sustained gunshot, (4) victim’s body was dismembered (i.e., the entire removal, by any means, of a large section of the body of a living or dead person, specifically, the head, arms, hands, torso, pelvic area, legs, or feet, see Stone & Brucato, 2019, p. 83), (5) victim was beaten, (6) victim was stabbed/cut, (7) victim was strangled/asphyxiated, (8) evidence of overkill (i.e., infliction of more injury than is necessary to kill a person, see Ressler et al., 1988), (9) evidence that the crime scene was cleared/modified (e.g., offender removed or destroyed forensic evidence, offender set fire to scene, offender cleaned the scene, see Chopin et al., 2020), (10) evidence that the victim’s body was moved from the crime location to another, (11) number of days before body recovery (i.e., measured by subtracting the date the victim was last seen alive from the date the body was recovered), and (12) body recovery location was outdoors (i.e., in contrast to an indoor location).

Analytical Strategy

We computed two neural network models based on the multi-layer perceptron algorithm. Model 1 predicted the trace collection in sexual homicide cases, whereas Model 2 predicted crime solving results. Neural network models are mostly used in data mining and computational science. One of the main objectives of neural network analysis is to identify complex patterns and relationships between several inputs that cannot be identified by the human brain (Bigi et al., 2005). This technique is used in a large number of scientific disciplines (e.g., medicine, neuroscience, computer sciences, etc.) and is proving to be particularly promising in the field of social and human sciences given the complexity of predicting human behavior (Liu et al., 2011). In comparison with other classical methods of prediction such as logistic regression or classification and regression trees, neural networks present several advantages. Logistic regressions are known to be robust however, this method ignores the possibility that different variables predict the same outcome for different subgroups of individuals (see Steadman et al., 2000). Classification and regression trees provide a better representation of the different possibilities of predicting the same outcome for different groups but they are limited in terms of stability and present a risk of overfitting (Colombet et al., 2000; Dillard et al., 2007). Neural network models avoid the weaknesses of conventional methods. They are particularly effective when the primary goal is outcome prediction and important interactions, or when complex non-linearity exist within the data (Liu et al., 2011; Tu, 1996). The multi-layer perceptron neural network algorithm is the most commonly used network architecture consisting of inputs (i.e., independent variables), hidden layers (i.e., nodes) and output layers (i.e., dependent variable). These different layers are connected, and the force of the association is identified through the synaptic weights (i.e., the closer to zero, the weaker the relationship; the farther from zero, the stronger is the relationship). In order to test the quality of the predictive model, the multi-layer perceptron neural network algorithm consists of training and testing processes (Price et al., 2000). The percentage of correct predictions, as well as the area under the curve (AUC) value from the receiver operating characteristics (ROC) analysis, allowed assessment of the model quality (Liu et al., 2011).

The neural network analysis was calculated with the Statistical Package for the Social Sciences (SPSS) 27.0 software package. In Model 1, the input layer consisted of the crime scene variables, and the output layer contained two units of the two categories of the dependent variable (no collection of traces = 0; collection of traces = 1). To create the neural network model, 73.47% of cases (n = 169) were randomly sampled, while 26.53% of the cases (n = 61) were used to test the model. In Model 2, the input layer consisted of the crime scene variables as well as the collection of traces, and the output layer contained two units of the two categories (case is unsolved = 0; case is solved = 1) of the dependent variable.

Results

Descriptive analysis

Table 1 shows the descriptive statistics of variables used in this study. Traces were collected and analyzed in approximately half of the cases (46.09%) and 75.22% of the cases included in our sample were solved by the police. As to violence observed in these cases, almost half involved evidence of sexual penetration (48.26%) and beatings (46.96%). Victims were strangled/asphyxiated in 45.65% of the cases, and evidence of overkill was identified in 41.30% of the cases. Victims were stabbed/cut in 25.22% of the cases, and their bodies were dismembered in 6.09% of the cases. Finally, victims’ bodies were burned in 5.65% of the cases and gunshot was sustained in 5.22% of the cases. As to the use of forensic awareness strategies by the offender, the victim’s body was moved from the crime location to another location in 36.96% of the cases, while 29.57% of offenders used strategies to clean/modify the crime scene.

[Insert Table 1 here]

Table 2 presents findings of the neural network analysis predicting the collection and analysis of traces. Results showed that the model presented a good level of prediction with an Area Under the Curve (AUC) of 0.81 (see Appendix 1). The model correctly classified approximately 81.3% of the training sample and 82.1% of the testing sample. This model was based on five nodes (i.e., hidden layers) which includes three nodes to predict the collection and analysis of forensic evidence and two nodes to predict the absence of trace collection and analysis. Specifically, nodes 2, 4, and 5 predicted the collection of traces while nodes 2 and 4 predicted the absence of trace collection and analysis. Figure 2 summarizes the best factors of the model predicting the collection and analysis of traces according to the variable normalized importance indicators. Node 2 showed that cases with evidence of sexual penetration (0.73), where the victim was beaten (0.61), strangled/asphyxiated (1.14), stabbed/cut (0.33), burned (1.17), where there was evidence of overkill (0.88) were more likely to have traces collected and analyzed. Moreover, cases in which the victim’s body was moved from the crime location to another (0.47), where body recovery location was indoor (0.29), where offenders used strategies to clean/modify the crime scene (0.16), and for which the number of days before body recovery was low (-0.37) were more likely to have traces collected and analyzed. Node 4 showed that cases with evidence of sexual penetration (0.02), where victims were strangled/asphyxiated (0.11), victim sustained gunshot (0.77), where strategies were used by offenders to clean/modify the crime scene (0.37), where body recovery location was indoor (0.45), with a low number of days before body recovery (-0.20) were more likely to have traces collected. Node 5 showed that cases with evidence of sexual penetration (0.49), overkill (0.64), dismemberment (0.84), and for which the number of days before body recovery was low (-0.01) were more likely to have traces collected and analyzed.

[Insert Table 2 here]

[Insert Figure 2 here]

Table 3 summarizes and classifies the normalized importance of each variable predicting the collection and analysis of traces. The variables with the greatest predictive weight, in descending order, were evidence of sexual penetration, number of days before body recovery, crime scene was cleaned/modified, evidence of overkill, victims’ body was dismembered, victim’s body was burned, victim was strangled/asphyxiated, body recovery location was outdoor. The following variables were not significant in predicting the collection and analysis of forensic evidence: victim was beaten, victim was stabbed/cut, victim sustained gunshot, the victim’s body was moved from the crime location to another location.

[Insert Table 3]

Table 4 presents findings of the neural network analysis predicting the crime-solving outcome. Results showed a high level of prediction for the model, with an Area Under the Curve (AUC) of 0.85 (see Appendix 2). The model correctly classified approximately 83.3% of the training sample and 83.9% of the testing sample. This model was based on six nodes (i.e., hidden layers) including three nodes to predict the crime solving and three other nodes to predict the absence of case solving. Figure 3 summarizes the results of the model predicting the crime solving of sexual homicide cases. Node 1 showed that cases with evidence of sexual penetration (0.45), where victims were beaten (0.27), dismembered (0.12), and moved to another location than the crime location (0.27) were more likely to be solved. Moreover, cases with a low number of days before body recovery (-0.46), where the body recovery location was indoor (0.02), and for which traces were collected and analyzed (0.17), were more likely to be solved. Node 4 showed that cases with evidence of sexual penetration (0.26), where victims were strangled/asphyxiated (0.36), where victims’ bodies were moved to a different location than the crime location (0.22), and when offenders cleaned/modified the crime scene (0.52) were more likely to be solved. Moreover, cases with a low number of days before body recovery (-0.07), where the body recovery location was indoor (0.38), and for which forensic evidence were collected and analyzed (0.46), were more likely to be solved. Finally, Node 5 showed that cases with evidence of sexual penetration (0.83), where victims were beaten (0.14), strangled/asphyxiated (0.01), stabbed/cut (0.33), burned (0.67), with evidence of overkill (0.38), and who sustained gunshot (0.65), were more likely to be solved. Moreover, cases with a higher number of days before body recovery (0.68) were more likely to be solved.

[Insert Table 4 here]

[Insert Figure 3 here]

Table 5 summarizes and classifies the normalized importance of each variable predicting crime solving. The variables with the greatest predictive weight, in descending order, were evidence of sexual penetration, body recovery location was indoors, crime scene was cleared/modified, number of days before body recovery, victim was gunshot, victim’s body was burned, collection and analysis of traces, and victim was strangled/asphyxiated. The following variables were not significant in predicting the crime solving: victim’s body was dismembered, the victim’s body was moved to a different location from the crime location, victim was beaten, evidence of overkill, and victim was stabbed/cut.

[Insert Table 5]

Discussion

Over the past years, a handful of studies have focused on the factors influencing the collection and analysis of traces in the criminal justice process (e.g., Bitzer et al., 2016; Resnikoff et al., 2015; Ribaux et al., 2010). Moreover, an increased attention has been dedicated to the study of the impact traces have on the outcome of the case. Yet, results have been scarce and conflicting, leaving us with a limited understanding of the decision making behind the collection and analysis of traces as well as how much these traces really matter in the criminal justice process.

This study aims to contribute to these important questions by examining the process of trace collection and analysis in sexual homicide cases. Specifically, the first step was to determine which crime scene behaviors were more likely to lead to the collection and analysis of traces. The second step was to identify criminal situations in which the collection and analysis of traces led to the solving of the crime. In order to answer these questions, 230 sexual homicide cases that occurred in Canada were examined. Neural network models were computed instead of logistic regression models in order to capture the different combinations of factors that predicted the same outcome (i.e., collection and analysis of traces and crime solving).

Sexual homicide crime scene behaviors to predict the collection and analysis of traces.

In the descriptive part of the study, we found that the percentage of sexual homicide cases for which traces are collected and analyzed (46%) is lower than previous studies of serious crimes. For instance, Mapes et al. (2015) reported that biological traces were analyzed in 57% of serious crimes. In addition, Baskin & Sommers (2010) found higher rates among homicide cases, in which 97% of cases had traces collected and 81% of cases had traces analyzed. Although several reasons may explain the infrequent use of traces (e.g., lack of communication and knowledge; Burrows & Tarling, 2004; Schroeder & White, 2009), our findings highlight two factors that appear determinant in the decision to collect and analyze traces in sexual homicide cases. First, as shown in previous studies (Baskin & Sommers,2010, 2011; Wilson-Kovacs, 2014), the use of forensic science appears to depend on the seriousness of the offence. Thus, our findings show that the variables with the highest predictive power for collection and analysis of traces are related to the greater level of cruelty in sexual homicide, such as evidence of overkill, and the dismemberment or burning of the victims’ body. These crime scene behaviors lead to a higher likelihood that the offender may leave traces behind. Such high-intensity crime scene behaviors consequently influence the finding, the collection, and the analyis of traces. These results suggest that the greater probabilities of trace transfer during such high-intensity criminal activity may influence the finding, the collection and the analysis of traces. Second, the other factor that may influence the decision to collect and analyze traces is the consideration of the physical and situational components of the crime scene. Our findings showed that when the victim’s body was recovered sooner and recovered indoors, traces were more likely to be collected and analyzed. These two variables relate to the consideration of the persistence of traces in particular circumstances, that is, the property of susbtances and materials to persist over time on certain surfaces in sufficient quantity and quality for analytical purposes. In other words, when conditions seem favorable to trace persistence (e.g., the body is recovered indoors and within a few days), more traces could be available for crime scene investigators to collect and analyze. Finally, another possible explanation for the lack of trace evidence is related to the date of the cases in the study’s sample. In older cases, biological traces may not have been routinely collected. In addition, a case-specific context might involve questioning the utility of collecting traces, such as when the suspect is already at the scene.

The collection and analysis of traces is not always associated with crime solving

One of the central questions related to the study of the usefulness of forensic science is whether or not the use of traces leads to the solving of crimes. This question is often the object of myths that is associated with the crime scene investigation (CSI) effect (e.g., Cole, 2010; Cole & Dioso-Villa, 2006, 2007), which gives a prominent and oversimplified role to the collection of traces in the investigative process and the identification of suspects. In the current study, our findings indicate a resolution rate of 75%, which is higher than in previous studies. For instance, Baskin and Sommers (2010) reported an arrest and conviction rate of 50% and 34 % respectively in homicide cases, while the homicide clearance rate for Canada in 2020 is situated around 64% (Statistics Canada, 2021). It is possible that the higher homicide clearance rate observed in our study is due to the sexual nature of these homicides. First, the presence of sexual activities multiplies the victim-offender interactions, which in turn increases the likelihood of leaving traces according to the Locard’s exchange principle (Locard, 1920). Second, individuals involved in sexual homicide present a different criminal career than those involved in non-sexual homicides (Beauregard et al., 2018; Chopin & Beauregard, 2019b). Accordingly, the presence of previous convictions for individuals involved in sexual homicide could result in their inclusion in existing forensic databases (e.g., DNA databases such as CODIS), which would facilitate their identification in case of trace collection and analysis. Both factors would presuppose an impact of forensic science on case resolution. Finally, since sexual homicide is a particularly serious crime, with some of our cases involving great level of violence and cruelty, law enforcement agencies may have deployed greater resources to solve those crime compared to more “typical” homicide.

Our findings also show that the collection and analysis of traces is associated with the crime-solving outcome for certain combinations of homicide crime scene behaviors. Specifically, three combinations of factors predict the crime-solving outcome (nodes 1, 4, 5), while three others are associated with the non-solving crimes (nodes 2, 3, 6). Among the combination of factors predicting the crime-solving, we observe that the collection and analysis of traces is positively associated with crime solving in two combinations (nodes 1, 4). However, our results also show that in some cases, the analysis and collection of traces is not positively associated with crime solving (i.e., node 5) nor is it positively associated with the non-solving of crimes (i.e., node 2). These results reflect the debate surrounding the role and utility of trace collection and analysis for crime solving. While our results show a nuanced picture of the contribution of forensic science to crime solving, they do not suggest that the collection and analysis is counterproductive to the investigation. Rather, they reiterate the complexity of the contribution of material traces to an investigation process, which draws on various other sources of information (e.g. testimony, confession) thus reinforcing the idea that traces are not the silver bullet as often portrayed by the CSI effect. Solving sexual homicide cases is a multifactorial phenomenon (Balemba et al., 2014; Beauregard & Martineau, 2014) partially based on the skills of investigators (James & Beauregard, 2018) or even sometimes luck (Rossmo, 2009). More researcch are thus still needed to better evaluate the overall contribution of forensic science to sexual homicide investigation.

The identification of certain crime scene behaviors in sexual homicides that are associated with the collection and analysis of traces and the resolution of the crime suggests two possible explanations for such relationship. The first is the amount of traces to be exploited. Our results indicate that the presence of sexual penetration, the use of forensic awareness strategies, and the presence of strangulation/asphyxiation during the crime better predict crime solving through trace collection and analysis. These results are congruent with theorical principles (e.g. Locard’s principle) and previous empirical studies suggesting that certain behaviors (i.e., leaving semen at the scene, attempting to destroy the scene, close interactions with the victim) lead to an increase in the amount of traces that can be collected and compared with existing databases (Balemba et al., 2014; Beauregard & Martineau, 2014, 2018; Chopin et al., 2020; Chopin et al., 2019). Second, we observe that when the body is recovered indoors and shortlyafter the murder, it is more likely that traces were collected and analyzed as well as the case being solved. The elapsed time between the commission of the crime and body discovery exposes the crime scene to weather and environmental elements. In turn, the quality of available traces may become degraded, harder to find and appreciate, or even destroyed (Martin et al., 2019). Accordingly, and as highlighted in our study, traces collected promptly in a protected environment (e.g., indoors) seem to be more easily exploited and useful to the investigation.

Conclusion

To the best of our knowledge, the current study is the first to investigate the process of collecting and analyzing traces in sexual homicide cases and its relationship to the resolution of these crimes. First, we determined whether certain crime scene characteristics predicted the collection and analysis of traces. Not surprisingly, the results indicate that trace collection and analysis were more likely to occur in sexual homicide cases with crime scene behaviors exhibiting the highest risk for trace transfer (e.g. close interactions with the victim) as well as the best conditions for trace persistence (e.g. body is found indoors). Second, we examined the situations in which the collection and analysis of traces contributes to crime solving. The results suggest that the collection and analysis of traces does not necessarily predict the resolution of the case. Specifically, the analyses show that the collection and analysis of traces is useful for crime solving when: (1) the offenders' behaviors increase the opportunities for leaving traces at the crime scene, and (2) when the environmental and temporal aspects are favorable to the collection of traces.

There are several limitations to this study. First, the sample consist of only cases from Canada – excluding cases from the provinces of Quebec and Ontario. We cannot exclude that investigative and forensic practices may vary between provinces as law enforcement agencies operate differently across Canada. Therefore, our results may not be representative of the treatment of sexual homicide cases at the provincial level. Similarly, since investigative and forensic practices vary greatly across different countries, de facto generalizations cannot be achieved internationally. Second, the details regarding the type of traces collected and used were not available for analysis. Such data would have contributed additional information to the study by providing explanations as to why certain behaviors that are theoretically riskier for trace transfer (e.g. victim was stabbed/cut, the victim’s body was moved to another location) seem to only be weak predictors of the collection and analysis of traces. Third, although the study focused on the collection and analysis of traces, it was not possible to include the results of the trace analyses (e.g., DNA match). It is possible that some traces have been analyzed in a case, such as a DNA sample, but did not lead to a positive result or to a match in any databases. Finally, a multivariate approach was conducted with a limited sample size. While the use of artificial neural network models is quite appropriate with small sample sizes (Cui et al., 2004; Kim, 2008), other studies have suggested that the ‘factor 10’ rule-of-thumb that was adopted in this study could be insufficient and that a ‘factor 50’ rule of thumb was preferrable (Alwosheel et al., 2018). Despite these analytical limitations, we believe that the results should be understood in terms of trends (i.e., positive or negative) rather than the exact values of the statistical weight of each factor.

The study presents several implications for the investigation of sexual homicide and the crime scene investigation in unusual cases. The results highlight the importance of knowledge on the transfer and persistence mechanisms across the range of traces potentially left by the offender at the crime scene. It calls for an in-depth knowledge about traces for crime scene investigators so that they can infer on which traces may have been exchanged during the criminal event and which ones may still remain to maximize detection (Delémont et al., 2012, 2018). Despite sexual homicide cases posing various challenges to investigators due to their low base rate, the current findings may help identify crime scene behaviors that could lead to a collection and analysis of traces that would subsequently result in the resolution of the case. It could therefore assist investigators who have little experience with this type of crime to make efficient decisions.

Future studies should attempt to replicate our findings with data from other countries to confirm the generalizability of our results or to specify a new model. Moreover, a more in-depth analysis considering the different types of traces collected should be undertaken to provide a better understanding of trace collection and analysis as well as to better inform crime scene investigation practices.

References

Alwosheel, A., van Cranenburgh, S., & Chorus, C. G. (2018). Is your dataset big enough? Sample size requirements when using artificial neural networks for discrete choice analysis. Journal of choice modelling, 28, 167-182.

ANZPAA NIFS. (2012). End-To-End Forensic Identification Process Project.

ANZPAA NIFS. (2016). End-To-End Forensic Identification Process Project—Phase 2.

Baechler, S., Cartier, D., Schucany, P., & Guéniat, O. (2015). Les interventions de la police scientifique suite à des cambriolages: Quelle est la perception des lésés et y a-t-il lieu de s’en soucier? Revue Internationale de Criminologie et de Police Technique et Scientifique, 2.

Balemba, S., Beauregard, E., & Martineau, M. M. (2014). Getting away with murder: a thematic approach to solved sexual homicides using crime scene factors. Police Practice and Research: An International Journal, 15(3), 221-233. https://doi.org/10.1080/15614263.2013.846548

Baskin, D., & Sommers, I. (2010). The influence of forensic evidence on the case outcomes of homicide incidents. Journal of Criminal Justice, 38(6), 1141–1149.

Baskin, D., & Sommers, I. (2011). The Influence of Forensic Evidence on the Case Outcomes of Assault and Robbery Incidents. Criminal Justice Policy Review, 1–25.

Beauregard, E., DeLisi, M., & Hewitt, A. N. (2018). Sexual murderers: sex offender, murderer, or both? Sexual Abuse, 30(8), 932-950. https://doi.org/10.1177/1079063217711446

Beauregard, E., & Martineau, M. M. (2014). No body, no crime? The role of forensic awareness in avoiding police detection in cases of sexual homicide. Journal of Criminal Justice, 42(2), 213-220. https://doi.org/10.1016/j.jcrimjus.2013.06.007

Beauregard, E., & Martineau, M. M. (2016). Does the organized sexual murderer better delay and avoid detection. Journal of interpersonal violence, 31(1), 4-25. https://doi.org/10.1177/0886260514555129

Beauregard, E., & Martineau, M. M. (2017). Introduction. In E. Beauregard & M. M. Martineau (Eds.), The sexual murderer: Offender behavior and implications for practice (pp. 1-10). Routledge.

Beauregard, E., & Martineau, M. M. (2018). Action-Reaction? Forensic evidence analysis and forensic awareness in sexual homicide. In M. DeLisi & M. G. Vaughn (Eds.), The Routledge International Handbook of Biosocial Criminology (pp. 605-622). Routledge.

Bigi, R., Gregori, D., Cortigiani, L., Desideri, A., Chiarotto, F. A., & Toffolo, G. M. (2005). Artificial neural networks and robust Bayesian classifiers for risk stratification following uncomplicated myocardial infarction. International journal of cardiology, 101(3), 481-487. https://doi.org/10.1016/j.ijcard.2004.07.008

Bitzer, S., Albertini, N., Lock, E., Ribaux, O., & Delémont, O. (2015). Utility of the clue—From assessing the investigative contribution of forensic science to supporting the decision to use traces. Science and Justice, 55(6), 509--513. https://doi.org/10.1016/j.scijus.2015.05.005

Bitzer, S., Miranda, M. D., & Bucht, R. E. (2021). Forensic advisors: The missing link. WIREs Forensic Science. https://doi.org/10.1002/wfs2.1444

Bitzer, S., Ribaux, O., Albertini, N., & Delémont, O. (2016). To analyse a trace or not? Evaluating the decision-making process in the criminal investigation. Forensic Science International, 262, 1–10. https://doi.org/10.1016/j.forsciint.2016.02.022

Bradbury, S.-A., & Feist, A. (2005). The use of forensic science in volume crime investigations: A review of the research literature. Home Office Online Report 43/05.

Brodeur, J.-P. (2005). L’enquête criminelle. Criminologie, 38(2), 39–64.

Brown, C., Ross, A., & Attewell, R. G. (2015). Benchmarking Forensic Performance in Australia–Volume Crime. Forensic Science Policy & Management: An International Journal, 5(3–4), 91–98. https://doi.org/10.1080/19409044.2014.981347

Bruenisholz, E., Vandenberg, N., Brown, C., & Wilson-Wilde, L. (2019). Benchmarking forensic volume crime performance in Australia between 2011 and 2015. Forensic Science International: Synergy, 1, 86–94. https://doi.org/10.1016/j.fsisyn.2019.05.001

Burrows, J., & Tarling, R. (2004). Measuring the impact of forensic science in detecting burglary and autocrime offences. Science & Justice, 44(4), 217–222.

Chopin, J., & Beauregard, E. (2019a). Sexual Homicide in France and Canada: An International Comparison. Journal of interpersonal violence, 36(19-20), 10453-10476. https://doi.org/10.1177/0886260519875547

Chopin, J., & Beauregard, E. (2019b). The Sexual Murderer is a Distinct Type of Offender. International Journal of Offender Therapy and Comparative Criminology, 63(9), 1597-1620. https://doi.org/10.1177/0306624X18817445

Chopin, J., & Beauregard, E. (2021). Sexual Homicide and Its Investigation: New Perspectives to Improve Police Practices. In N. Deslauriers-Varin & C. Bennell (Eds.), Criminal Investigations of Sexual Offenses: Techniques and Challenges (pp. 235-251). Springer International Publishing. https://doi.org/10.1007/978-3-030-79968-7_16

Chopin, J., Beauregard, E., & Bitzer, S. (2020). Factors influencing the use of forensic awareness strategies in sexual homicide. Journal of Criminal Justice, 71, 1-9. https://doi.org/10.1016/j.jcrimjus.2020.101709

Chopin, J., Beauregard, E., Bitzer, S., & Reale, K. (2019). Rapists’ behaviors to avoid police detection. Journal of Criminal Justice, 61(2019), 81-89. https://doi.org/10.1016/j.jcrimjus.2019.04.001

Cole, S. A. (2010). Forensic identification evidence: Utility without infallibility. Criminology & Public Policy, 9(2), 375-379.

Cole, S. A., & Dioso-Villa, R. (2006). CSI and its effects: Media, juries, and the burden of proof. New Eng. L. Rev., 41, 435.

Cole, S. A., & Dioso-Villa, R. (2007). The CSI effect: the true effect of crime scene television on the justice system. New Engl Law Rev, 41, 435-455.

Colombet, I., Ruelland, A., Chatellier, G., Gueyffier, F., Degoulet, P., & Jaulent, M.-C. (2000). Models to predict cardiovascular risk: comparison of CART, multilayer perceptron and logistic regression. American Medical Informatics Association Annual Symposium Proceedings Archive, 156-160.

Crispino, F. (2008). Nature and place of crime scene management within forensic sciences. Science & Justice, 48(1), 24–28.

Cui, Y.-J., Davis, S., Cheng, C.-K., & Bai, X. (2004). A study of sample size with neural network. International Conference on Machine Learning and Cybernetics, Shanghai, China

De Forest, P. R. (1999). Recapturing the essence of criminalistics. Science & Justice, 39(3), 196–208. https://doi.org/10.1016/S1355-0306(99)72047-2

Delémont, O., Bitzer, S., Jendly, M., & Ribaux, O. (2018). The Practice of Crime Scene Examination in an Intelligence-Based Perspective. In Q. Rossy, D. Décary-Hétu, O. Delémont, & M. Mulone (Eds.), The Routledge International Handbook of Forensic Intelligence and Criminology. Taylor & Francis.

Delémont, O., Esseiva, P., Ribaux, O., & Margot, P. (2012). La violence laisse des traces: L’homicide dévoilé par la science forensique. In M. Cusson, S. Guay, J. Proulx, & F. Cortoni (Eds.), Traité des violences criminelles. Hurtubise, Montréal.

Delémont, O., Lock, E., & Ribaux, O. (2014). Forensic science and criminal investigation. In Encyclopedia of Criminology and Criminal Justice (pp. 1754–1763). Springer.

Dillard, E., Luchette, F. A., Sears, B. W., Norton, J., Schermer, C. R., Reed II, R. L., Gamelli, R. L., & Esposito, T. J. (2007). Clinician vs mathematical statistical models: which is better at predicting an abnormal chest radiograph finding in injured patients? American Journal of Emergency Medicine, 25(7), 823-830. https://doi.org/10.1016/j.ajem.2006.12.009

Fann, K. T. (1970). Peirce’s Theory of Abduction. Nijhoff The Hague.

Girod, A., Champod, C., & Ribaux, O. (2008). Trace de souliers. Presses polytechniques et universitaires romandes.

Hazard, D. (2014). La pertinence en science forensique Une (en)quête épistémologique et empirique [Phd thesis]. Université de Lausanne.

Hazard, D., & Margot, P. (2014). Forensic Science Culture. In Encyclopedia of Criminology and Criminal Justice (pp. 1782–1795). Springer.

Home Office. (2007). Summary Report of the Scientific Work Improvement (SWIM) Package. Home Office.

James, J., & Beauregard, E. (2020). Murderer vs investigator: Factors influencing the resolution of sexual homicide cases. Police Practice and Research, 21(2), 172–186. https://doi.org/10.1080/15614263.2018.1526683

Julian, R. D., Kelty, S. F., Roux, C., Woodman, P., Robertson, J., Davey, A., Hayes, R., Margot, P., Ross, A., Sibly, H., & White, R. (2012). What is the value of forensic science? An overview of the effectiveness of forensic science in the Australian criminal justice system project. Australian Journal of Forensic Sciences, 43(4), 217–229. https://doi.org/10.1080/00450618.2011.610820

Julian, R., Kelty, S., & Robertson, J. (2012). “Get it right the first time”: Critical Issues at the Crime Scene. Current Issues in Criminal Justice, 24(1), 25–37. https://doi.org/10.1080/10345329.2012.12035942

Kerr, K. J., Beech, A., & Murphy, D. (2013). Sexual homicide: Definition, motivation and comparison with other forms of sexual offending. Aggression and Violent Behavior, 18(1), 1-10.

Kim, Y. S. (2008). Comparison of the decision tree, artificial neural network, and linear regression methods based on the number and types of independent variables and sample size. Expert Systems with Applications, 34(2), 1227-1234.

Liu, Y. Y., Yang, M., Ramsay, M., Li, X. S., & Coid, J. W. (2011). A comparison of logistic regression, classification and regression tree, and neural networks models in predicting violent re-offending. Journal of Quantitative Criminology, 27(4), 547-573. https://doi.org/10.1007/s10940-011-9137-7

Locard, E. (1920). L’enquête criminelle et les méthodes scientifiques [Criminal investigation and scientific methods]. Flammarion.

Ludwig, A., Fraser, J., & Williams, R. (2012). Crime Scene Examiners and Volume Crime Investigations: An Empirical Study of Perception and Practice. Forensic Science Policy & Management: An International Journal, 3(2), 53–61. https://doi.org/10.1080/19409044.2012.728680

Makin, D. A. (2013). Symbolic Evidence Collection or “If All Else Fails, Throw Some Dust Around”. Forensic Science Policy & Management: An International Journal, 3(3), 126–138. https://doi.org/10.1080/19409044.2013.780834

Mapes, A. A., Kloosterman, A. D., & Poot, C. J. de. (2015). DNA in the Criminal Justice System: The DNA Success Story in Perspective. Journal of Forensic Sciences, 60(4), 851–856.

Martin, J.-C., Delémont, O., Esseiva, P., & Jacquat, A. (2010). Investigation de scène de crime: Fixation de l’état des lieux et traitements des traces d’objets. Presses polytechniques et universitaires romandes.

Mucchielli, L. (2012). Homicides in Contemporary France. In M. Liem & W. Pridemore (Eds.), Handbook of European Homicide Research (pp. 301–312). Springer.

Peterson, J. L., Hickman, M. J., Strom, K. J., & Johnson, D. J. (2013). Effect of Forensic Evidence on Criminal Justice Case Processing. Journal of Forensic Sciences, 58(S1), S77–S90.

Price, R. K., Spitznagel, E. L., Downey, T. J., Meyer, D. J., Risk, N. K., & El-Ghazzawy, O. G. (2000). Applying artificial neural network models to clinical decision making. Psychological assessment, 12(1), 40. https://doi.org/10.1037/1040-3590.12.1.40

Quinlan, A. (2011). Tracing the ‘Messy’History of Forensic DNA Analysis in Canada. Studies in Sociology of Science, 2(2), 11-18.

Resnikoff, T., Ribaux, O., Baylon, A., Jendly, M., & Rossy, Q. (2015). The polymorphism of crime scene investigation: An exploratory analysis of the influence of crime and forensic intelligence on decisions made by crime scene examiners. Forensic Science International, 257, 425–434. https://doi.org/10.1016/j.forsciint.2015.10.022

Ressler, R. K., Burgess, A. W., & Douglas, J. (1988). Sexual homicide: patterns and motives. Free Press.

Ribaux, O., Baylon, A., Lock, E., Delémont, O., Roux, C., Zingg, C., & Margot, P. (2010). Intelligence-led crime scene processing. Part II: Intelligence and crime scene examination. Forensic Science International, 199(1–3), 63–71.

Rossmo, K. (2009). Criminal investigative failures. CRC Press.

Schroeder, D. A., & White, M. D. (2009). Exploring the Use DNA Evidence in Homicide Investigations: Implications for Detective Work and Case Clearance. Police Quarterly, 12(3), 319–342. https://doi.org/10.1177/1098611109339894

Schuliar, Y. (2009). La coordination scientifique dans les investigations criminelles. Propositions d’organisation, aspects éthiques ou de la necessité d’un nouveau métier. Université Paris Descartes.

Statistics Canada. (2021). Homicide in Canada, 2020. https://www150.statcan.gc.ca/n1/daily-quotidien/211125/dq211125b-eng.htm?indid=3435-1&indgeo=0

Steadman, H. J., Silver, E., Monahan, J., Appelbaum, P., Robbins, P. C., Mulvey, E. P., Grisso, T., Roth, L. H., & Banks, S. (2000). A classification tree approach to the development of actuarial violence risk assessment tools. Law and human behavior, 24(1), 83-100. https://doi.org/10.1023/A:1005478820425

Stefanska, E. B., Beech, A., & Carter, A. J. (2016). A systematic review of the literature comparing male non-serial sexual killers and sexual aggressors: examining homogeneous and heterogeneous characteristics of these groups. Journal of sexual aggression, 22(3), 323-341. https://doi.org/10.1080/13552600.2015.1126657

Stone, M. H., & Brucato, G. (2019). The New Evil: Understanding the Emergence of Modern Violent Crime. Prometheus Books.

Tu, J. V. (1996). Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. Journal of clinical epidemiology, 49(11), 1225-1231. https://doi.org/10.1016/S0895-4356(96)00002-9

Williams, R. (2007). Policing and Forensic Science. In T. Newburn (Ed.), Handbook of Policing (pp. 760–793). Cullompton: Willan.

Wilson-Kovacs, D. (2014). ‘Backroom Boys’: Occupational Dynamics in Crime Scene Examination. Sociology, 48(4), 763–779. https://doi.org/10.1177/0038038513503741

Wyatt, D. (2014). Practising crime scene investigation: Trace and contamination in routine work. Policing and Society, 24(4), 443–458. https://doi.org/10.1080/10439463.2013.868460

Tables and figures

Table 1. Descriptive analysis of the sample (N=230)

n=

%

Dependent variables

 

 

Collection and analysis of forensic evidence

106

46.09

Crime solved

173

75.22

Violence observed

Evidence of sexual penetration

111

48.26

Victim was beaten

108

46.96

Victim was strangulated/asphyxiated

105

45.65

Victim was overkilled

95

41.30

Victim was stabbed/cut

58

25.22

Victims body was dismembered

14

6.09

Victim's body was burned

13

5.65

Victim sustained gunshot

12

5.22

Forensic awareness strategies

The victim's body was moved from the crime location to another

85

36.96

Crime scene was cleared/modified

68

29.57

Crime parameters

Average Number of days before body recovery (continuous)

581 [SD=190.07, 0-13702]

Body recovery location was an indoor environment

101

43.91

Notes.

1 Corresponds to the mean

2 Corresponds to the range

Table 2. Model 1: Neural network analysis of factors predicting the collection and analysis of forensic evidence (N=230)

Hidden Layer

 

Output Layer

Node 1

Node 2

Node 3

Node 4

Node 5

No collection and analysis of forensic evidence

Collection and analysis of forensic evidence

Input Layer

(Constant)

-0.13

0.45

0.43

-0.37

0.25

Violence observed

 

 

 

 

 

 

 

Evidence of sexual penetration

-0.31

0.73

-0.33

0.02

0.49

 

 

Victim was beaten

0.57

0.61

0.12

-0.36

0.84

 

 

Victim was strangulated/asphyxiated

-1.10

1.14

-1.17

0.11

-0.66

 

 

Victim was overkilled

-0.11

0.88

-1.12

-0.11

0.64

 

 

Victim was stabbed/cut

-1.00

0.33

0.33

-1.44

-0.52

 

 

Victims body was dismembered

-0.89

-0.03

-0.86

-0.44

0.84

 

 

Victim's body was burned

-0.16

1.17

-0.20

-0.60

-0.02

 

 

 

Victim sustained gunshot

-0.27

-0.37

-0.71

0.77

-0.25

 

 

 

Forensic awareness strategies

 

 

 

 

 

 

 

 

The victim's body was moved from the crime location to another

-0.64

0.47

0.28

-0.42

-0.31

 

 

 

Crime scene was cleared/modified

-0.62

0.16

-0.47

0.37

-0.27

 

 

 

Crime parameters

 

 

 

 

 

 

 

 

Number of days before body recovery (continuous)

0.56

-0.37

0.44

-0.20

-0.01

 

 

 

Body recovery location was an indoor environment

-0.24

0.29

0.73

0.45

-0.01

 

 

Hidden Layer

(Constant)

0.26

-0.81

Node 1

0.95

-1.07

Node 2

-1.03

1.51

Node 3

1.36

-0.67

Node 4

-1.01

1.26

Node 5

-0.50

1.06

Classification

 

 

% of correct classification (training sample)

81.30%

 

% of correct classification (testing sample)

82.10%

 

AUC

0.81

 

 

 

 

 

 

 Table 3. Model 1: Best factors classification of factors predicting the collection and analysis of forensic evidence (N=230)

Ordered factors

Importance

Normalized Importance

Evidence of sexual penetration

0.14

100.00%

Number of days before body recovery

0.09

82.40%

Crime scene was cleared/modified

0.08

72.90%

Victim was overkilled

0.08

70.20%

Victims body was dismembered

0.06

64.90%

Victim's body was burned

0.07

52.50%

Victim was strangulated/asphyxiated

0.07

51.00%

Body recovery location was an indoor environment

0.07

50.10%

Victim was beaten

0.05

36.90%

Victim was stabbed/cut

0.04

26.90%

Victim sustained gunshot

0.03

18.40%

The victim's body was moved from the crime location to another

0.03

17.10%

 

Table 4. Model 2: Neural network analysis of factors predicting the crime solving of sexual homicide cases (N=230)

Hidden Layer

 

Output Layer

Node 1

Node 2

Node 3

Node 4

Node 5

Node 6

Unsolved cases

Solved cases

Input Layer

(Constant)

-0.15

-0.53

0.35

0.53

0.67

0.00

 Violence observed

 

 

 

 

 

 

 

 

 Evidence of sexual penetration

0.45

0.09

-0.62

0.26

0.83

-0.40

 

 

 Victim was beaten

0.27

0.23

0.53

-0.05

0.14

0.06

 

 

 Victim was strangulated/asphyxiated

-0.35

-0.45

-0.73

0.36

0.01

0.21

 

 

 Victim was overkilled

-0.12

-0.20

0.15

-0.16

0.38

-0.65

 

 

 Victim was stabbed/cut

-0.36

-0.45

0.11

-0.12

0.33

0.37

 

 

 Victims body was dismembered

0.12

-0.21

-0.51

-0.13

-0.05

-0.24

 

 

 Victim's body was burned

-0.46

-0.47

0.23

-0.57

0.67

0.04

 

 

 

 Victim sustained gunshot

-0.13

-0.26

-0.83

-0.05

0.65

-0.35

 

 

 

 Forensic awareness strategies

 

 

 

 

 

 

 

 

 

 The victim's body was moved from the crime location to another

0.27

-0.34

0.40

0.22

-0.17

-0.40

 

 

 

 Crime scene was cleared/modified

-0.33

0.23

-0.56

0.52

0.88

0.13

 

 

 

 Crime parameters

 

 

 

 

 

 

 

 

 

 Number of days before body recovery

-0.46

-0.14

-0.08

-0.07

0.68

0.45

 

 

 

 Body recovery location was an indoor environment

0.02

0.12

-0.76

0.38

-0.72

0.11

 

 

 

 Collection and analysis of forensic evidence

0.17

-0.06

0.35

0.46

-0.14

-0.31

 

 

Hidden Layer

(Constant)

-0.17

0.52

Node 1

-0.15

0.21

Node 2

-0.23

-0.30

Node 3

0.78

-0.90

Node 4

-0.35

0.11

Node 5

-1.08

0.85

 

Node 6

0.32

-0.46

Classification

% of correct classification (training sample)

87.40%

% of correct classification (testing sample)

83.10%

AUC

0.87

 

Table 5. Model 2: Best factors classification of factors predicting the crime solving of sexual homicide cases (N=230)

Ordered factors

Importance

Normalized Importance

Evidence of sexual penetration

0.14

100.00%

Body recovery location was an indoor environment

0.12

89.10%

Crime scene was cleared/modified

0.11

77.50%

Number of days before body recovery

0.10

74.30%

Victim was gunshot

0.10

72.40%

Victim's body was burned

0.08

59.00%

Collection of biological samples

0.08

57.66%

Victim was strangulated/asphyxiated

0.07

54.20%

Victims body was dismembered

0.06

44.40%

The victim's body was moved from the crime location to another

0.06

40.40%

Victim was beaten

0.05

36.20%

Victim was overkilled

0.04

32.30%

 

Figure 1. Sampling procedures

Figure 2. Model 1: Neural network of best factors predicting the collection and analysis of forensic evidence (N=230)


Figure 3. Model 2: Neural network of best factors predicting the crime solving of sexual homicide cases (N=230)


Appendix

Appendix 1. ROC Curve analysis of Model 1.

Appendix 2. ROC Curve analysis of Model 2.

 

Comments
0
comment

No comments here

Why not start the discussion?