Keywords

Investigative interviews, confession, disclosure, police interrogation, Game Theory, dynamic interaction

Introduction

Interest in investigative interviews has increased over the past decades and research in this area has been important in the development of ethical, legal, and effective interrogation practices (Meissner et al., 2017). For example, the High-Value Detainee Interrogation Group (HIG) research program, begun in 2010, has made it possible to identify approaches that are effective in developing cooperation and rapport, eliciting information, challenging inconsistencies through the strategic presentation of evidence or information, and assessing credibility using cognitive cues and strategic questioning based on evidence-based scientific studies. Researchers have also helped develop meaningful reforms in police practices, such as the decision in Canada and some states in the US to legislate mandatory recording of all police interviews and the creation in the UK of the PEACE model of investigative interviewing, now used internationally (Bang et al., 2018; Snook et al., 2020). However, work is still needed to better understand different aspects of investigative interviews, including suspect disclosure. The increased sophistication of the latest methodologies and the development of more efficient tools for data collection offer promising possibilities for the study of investigative interviews according to evidence-based practices (e.g., Clarke et al., 2011; Cleary & Bull, 2021; Kelly et al., 2019; Leahy-Harland & Bull, 2017).

Deslauriers-Varin and Bergeron (2021) reviewed the scientific literature on police interrogation and identified three major waves of research. In the first wave, which was more theoretical in nature, researchers focused largely on explaining why someone confessed or admitted to having committed a crime. Various explanatory models were proposed and the importance of considering emotional, criminogenic, crime-related, and situational factors was discussed. Individual characteristics (e.g., age, relationship status), criminal background, and the context of the interview (e.g., presence of guilt, length of the interview and the conditions under which it was conducted as well as the attitude and behaviors of the police interrogator) were all found to play a role in the offender’s decision to confess during police interrogation (Beauregard et al., 2010).

The second wave is captured in studies published in the 1990s and 2000s (Deslauriers-Varin & Bergeron, 2021). Researchers focused on empirically validating the explanations and motivations put forward in the theoretical models and found significant associations between individual, crime-related, and situational/contextual factors and confession. Ethnicity was also found to be significantly associated with confession as Caucasians were more likely to confess than individuals of other ethnicities (Cleary & Bull, 2021; St-Yves, 2002; Viljoen et al., 2005). Regarding marital status, a study by St-Yves (2002) on sex offenders found that those who are single have a higher confession rate (38%) than individuals in a relationship (24%), perhaps because single individuals are less worried that a confession will lead to rejection (Deslauriers-Varin et al., 2011; St-Yves, 2002). Increased rates of confession have also been found with younger suspects (Beauregard et al., 2010; Viljoen et al., 2005).

The third wave of research, during the last decade, focused largely on further analysis of the influence of situational and contextual factors, particularly the effect of interrogation techniques and the relational context in which interrogations take place (Deslauriers-Varin, 2022). Much of this research examines the role of evidence strength (e.g., Brimbal & Luke, 2019; Deslauriers-Varin et al., 2011, 2020; Moston & Engelberg, 2011), the way in which evidence is presented (e.g., strategic use of evidence; Clemens et al., 2020; Granhag et al., 2013), and the role of police investigators and investigative techniques (e.g., Leahy-Harland & Bull, 2017; Zeng et al., 2021).

Policymakers, practitioners, scholars, and the public have become interested in interrogation methods (Kelly et al., 2016; see for example the “Mendez Principles of Effective Interviewing” in 2022). While recent studies have focused on factors (and the interactions between them) that may encourage or discourage confession, few have looked at the relative importance of these factors. In addition to the identification of static factors that may be associated with disclosure, developing best practices requires empirical information about the effectiveness of various strategies.

Contextual factors related to the disclosure process

There is evidence that confession is a dynamic process that is influenced by contextual elements and researchers have highlighted the importance of changes in the decision to confess during an interrogation (Deslauriers-Varin et al., 2011; Walsh & Bull, 2012). Such changes may be related to contextual variables, such as the attitudes and skill level of interrogators, as well as the strategies used (Cleary & Bull, 2021). Empirical studies have shown that pressure from police interrogators influences suspect disclosure. A suspect who feels respected and understood may have more confidence, which might make it easier to confess (Collins & Carthy, 2019; Kebbell et al., 2010; Kelly & Valencia, 2021; Wachi et al., 2014; Westera & Kebbell, 2014), while suspects who feel they are being pressured by the interviewer, for instance by the use of confrontation, argumentation, or humiliation during the interview process, are more likely to deny that they committed a crime (Holmberg & Christianson, 2002). This suggests that it is more effective for the interviewer to elicit and/or provide evidence of inconsistencies in a suspect’s story in a non-judgmental manner (May et al., 2017); direct accusations appear to lead to denial and decrease the amount of investigative-related information obtained (Adams-Quackenbush, 2019). The quality of available evidence also has an impact on confession (e.g., Gudjonsson & Petursson, 1991; Moston et al., 1992; Phillips & Brown, 1998) as does the way the evidence is presented (e.g., Granhag et al., 2013; Hartwig et al., 2014). Strategies used in an interview influence confession rate, showing that interviewers’ behaviors influence the course of an interview.

Research on contextual factors and interviewer strategies has helped increase our knowledge about investigative interviews. Although researchers have often focused on the correlations between confession and the factors that influence it as if they were stable over time (e.g., Walsh & Bull, 2012; Tekin et al., 2015), there was a clear shift in recent research toward a more dynamic analysis of investigative interviews (e.g., Kelly et al., 2016; Yang et al., 2015). Research shows that interviews are not linear as different elements are prone to change (e.g., Kelly et al., 2016) and that the various phases of an interview must be taken into account (Snook et al., 2021).

A dynamic approach to investigative interviews

Research examining the way evidence related to the case is presented can be seen as the first attempts to account for a dynamic element in interviews (Granhag et al., 2013; Hartwig et al., 2014; Vrij & Granhag, 2012) and has led to the development of approaches such as strategic use of evidence (SUE) and Shift-Of-Strategy (SoS). Luke and Granhag (2020) have shown that information gathering is more effective when the interviewer responds immediately to any discrepancies between the suspect’s statement and the evidence rather than continuing with general questions. Responding immediately to discrepancies is also associated with a change in interview strategies by both interviewer and suspect (Luke & Granhag, 2020). The confrontation of suspect with contradictory information is associated with more skillful interviewers (Walsh & Bull, 2015). SUE and SoS advocate withholding evidence from suspects to influence them and change their counterstrategies (i.e., their actions and how they cope with stress). Researchers have found that how suspects adapt their statements (Hartwig et al., 2014; Luke et al., 2014; Sorochinski et al., 2014) and their strategies (Håkansson, 2019) is related to their perception of the interviewer’s knowledge of available evidence. The Scharff technique is an intelligence gathering technique consisting in developing a friendly relationship with the suspect while giving the impression of knowing all the evidence and not pressing for information (Oleszkiewicz et al., 2017a). Scharff-trained handlers are perceived as less eager to gather information, but collects comparatively more new information (Oleszkiewicz et al., 2017b). The effectiveness of this technique strongly suggests that the perception of evidence and the development of humane relationship with the suspect increases disclosure.

Researchers increasingly take into consideration the importance of time on behaviors during investigative interviews. Studies of both the effect of the timing of evidence presentation and comparison of behaviors at different times during the interview have made it possible to identify changes in behaviors during an interview, creating a more dynamic analysis of the interview process. Pearse and Gudjonsson (1999) divided 18 police interviews in the United Kingdom into 5-minute segments, which allowed them to identify 39 different strategies and show that strategies can change over the course of an interview. Bull and Soukara (2010) adopted the idea of 5-minute segments and found that non-coercive strategies have the most positive influence on disclosure. These two studies analyzed only interviews that led to a confession. Kelly, Miller, and Redlich (2016) and Leahy-Harland and Bull (2017) used the same idea of 5-minute segments to compare interviews that led to a confession with those where the result was no confession, making it possible to account for the effect of time on confession and to observe the differences between a cooperative and a non-cooperative suspect over a given trajectory. For example, a strategy of rapport/empathy, which involves reassuring the suspect and using open-ended questions, was associated with an increased probability of partial or full confession while describing victims’ trauma and using negative questions were associated with a decreased probability (Leahy-Harland & Bull, 2017).

Verhoeven (2018) also used a dynamic methodology to study the relationship between suspects’ statements at the beginning and end of the interview, looking particularly at the influence of interview strategies on suspect disclosure, and concluded that the suspect is more likely to disclose information when the interviewer uses strategies that stress information gathering, while denial is more likely if the interviewer uses a confrontational strategy.

Yang and colleagues (2015) considered the dynamics of investigative interviews based on a decision-making model that highlights the temporal nature of the interview process. They argue that looking at the trajectory of an interview makes it possible to develop a model that can account for situations in which suspects make a series of decisions during the course of an interview, decisions that may change depending on what have happened previously. Their findings show that participants were more likely to change their disclosure decisions to avoid the consequences that are immediate (e.g., being detained for the night) when they perceived the long-term consequences to be more negative (e.g., perception of a harsher sentence).

There has been a shift in the literature in how the concepts of dynamic analysis are treated. Until very recently, the literature on the subject was largely atheoretical and, while the approaches being used are closely related to those in decision-making theories, researchers do not usually refer to any specific theory. We argue here that the concepts developed in Game Theory could be very helpful in understanding suspect disclosures and developing an effective methodology that makes it possible to investigate interviews as a process rather than a static event.

The Game Theory perspective

Game Theory is a mathematical model used to study decision-making (Bicchieri, 2004). It relies on rational choice theory, which sees individuals as rational actors whose actions are based on cost-benefit calculations about both possible gain and the means needed to obtain it (Cornish & Clarke, 1986). Game Theory emphasizes not only decision-making but the importance of interactions, seeing a game as an interaction between several individuals in which the gains of each player are affected by the decisions of the other players (Kelly, 2003). In analyzing both the logic behind rational decisions and in interactions between players, Game Theory deals with both communication and strategic interaction. What happens during a game is not a simple communication between two individuals but an exchange in which each decision (observable through behaviors) alters the objective environment – the concrete, existential situation of the other players – whether or not they are aware of this change (Goffman, 1969).

Applying Game Theory to investigative interviews involves integrating three factors that are seldom considered in most previous research in the area. The first is the players in the interaction. Most researchers have looked at how interviewer strategies influence suspect behaviors. Game Theory proposes that the decisions of both players, discerned through observable behaviors, should be considered.

The second factor is having the ability to describe the interaction, and this necessitate a change in the choice of the main variable used to analyze the disclosure process. Research on investigative interviews has tended to focus on a dichotomous analysis of suspect disclosure (confession vs. denial) (e.g., Tekin et al., 2015; Madon et al., 2013). We suggest that, rather than taking the suspect’s confession as the main object of study, the focus should shift toward investigation-relevant information (IRI), defined as information obtained during the interview that may be of relevance to the ongoing investigation (Baker-Eck & Bull, 2022; Baker-Eck et al., 2021; Bull & Rachlew, 2020; Oxburgh & Ost, 2011). The process of disclosure in the interview can then be considered in relation to each IRI provided by the suspect, making it possible to create both a continuous timeline and a more precise picture of the disclosure process. This ensures to see the interview as process in which the different actions of the players interact with one another, as proposed by Game Theory.

The third factor is time, which is present in Game Theory’s recognition of dynamic interaction, central to this perspective. A time-oriented approach makes it possible to study the probability that an IRI will be provided by the suspect in relation to the different strategies used during the interview. Observing behavior on a continuous timeline is necessary to understand the dynamics – and evolution – of decisions during an interview.

Aim of the study

The principal objective of this study is to observe the impact of behaviors of both the interviewer and the suspect on the production of IRI. The second objective is to look at whether the effect of such behaviors lasted over the course of the interview. This is done through the lens of different concepts of the Game Theory discussed above. Interviews are considered to be a dynamic social interaction that involves both suspect and interviewer and their behavior is observed through a continuous timeline. The sub-objective of the study is to showcase the usefulness of applying Game Theory to the study of investigative interviews by considering time and interaction between players as an integrative part of the analysis.

Methodology appropriate to the analysis of the continuous timelines of investigative interviews is presented and the disclosure process is analyzed in terms of all IRI revealed by the suspect rather than being understood as a dichotomous variable (admission or denial), making it possible to observe a complex and dynamic process and the way it changes over time. The influence of time on suspect disclosure is discussed in relation to the position of different behaviors on a timeline. Multilevel logistic regression is used to examine the behaviors and strategies of both interviewer and suspect that are predictive of suspect disclosure.

The study is based on a sample of investigative interviews with online sexual offenders in Canada. Online sex offenses leave digital traces. Such traces are often not only abundant but also appears in a similar amount through the different interviews, presenting a heterogeneity that is helpful in identifying trends among a sample. The large amount of detail in the interview makes it easier to identify the IRI provided, facilitating a sequential analysis of information presentation.

The interviews in the sample should be understood in terms of the Canadian context for police interviews. Such interviews usually involve only two participants – the interviewer and the suspect. In Canada, the suspect is allowed to contact a lawyer before the start of the interview, but the lawyer is not allowed inside the interview room. The setting thus limits the interaction to two players, enabling efficient application of Game Theory concepts.

Method

Sample

The sample consisted of 130 videotaped investigative interviews by the Internet Child Exploitation (ICE) special unit of Québec’s provincial police department, the Sûreté du Québec (SQ)1. Everyone in the sample had been charged with one or more offenses related to online child sexual exploitation – including child luring or accessing and possessing, distribution of, or production of child pornographic material – and had been convicted. All suspects were male and 95% were Caucasian. Average age was 37.68 years (SD=14.06). The mean length of the interviews was 198.23 minutes (SD= 71.15; range = 46.5-493.5; median = 193.69). In total, 25,770 minutes of interview time were analyzed.

Data collection and grouping of variables

A data collection tool using Observer XT^© was developed specifically for this study. Based on the literature on investigative interviews and adapted to the setting of Canadian investigative interviews, the collection tool made it possible to timestamp preidentified behaviors that took place during an interview, creating a continuous timeline of interview events. A coding grid and a guide were developed based on the literature and the first phase of observation. A research assistant worked closely with the first author to code interviewer’ and suspect’ behaviors. The grid included 77 behaviors (31 for the suspect and 26 for the interviewer)2. The taxonomy of investigative interviews developed by Kelly, Miller, Redlich, and Kleinman (2013) was adapted for the purpose of the study and the Canadian police investigative interview setting3 and used to classify the different behavior into domains. Six classification categories were identified for investigative interview strategies: (a) Rapport and relationship building, (b) Context manipulation, (c) Emotion provocation, (d) Confrontation and competition, (e) Collaboration, and (f) Presentation of evidence. No taxonomy of suspect behavior was available, so the categories of interviewer behaviors proposed by Kelly and colleagues (2013) were adapted to allow comparison of suspect and interviewer behavior. Table I shows how the original taxonomy was adapted to the requirements of the present study to capture observed behaviors for both interviewer and suspect rather than focusing only on interviewer strategies. Frequency of behavior for each category in the sample4 is also presented.

Table I - Adaptation of Kelly, Miller, Redlich, and Kleinman Taxonomy (2013)

Kelly’s (2013) taxonomy raises the importance of context manipulation which includes the physical setting around the suspect. All the interviews in the sample occur in the same type of room¹. There was therefore no possibility to control or measure the effect of this variable in the current research project. The emphasis of the observation is therefore put on the social interaction between the interviewer and the suspect. The second domain in Kelly’s taxonomy – context manipulation – was not applicable and was therefore dropped. The five other categories proposed by Kelly and colleagues are explained below and the adaptations needed to make it possible to take suspect behaviors into account are described. Each domain contains several behaviors.

Rapport and Relationship Building

In Kelly et al.’s taxonomy, Rapport and relationship building involves a working relationship between the interviewer and the suspect based on mutual respect and an understanding of each other’s objectives. Rapport allows the interviewer to develop a working alliance between the interviewer and educing information (Abbe & Brandon, 2013). The behaviors like building emotional bond, identifying common grounds with the suspect, and disclosing personal information to the suspect all contribute to build this collaborative relationship. Behaviors tented of kindness and respect will fall in this category. For example, considering and meeting basic needs (e.g., food, water), and maintaining the well-being of the suspect is essential for developing this rapport. For the purposes of the study the technique of flattering the suspect by using compliments and praise, classified as “emotion provocation” in the original taxonomy, while likely to result in a positive emotion, was considered to have more of an effect on developing a relationship and therefore has been put in the present group. A total of 7 different interviewer behaviors falls into this category.

The interviewer and the suspect are expected to have different attitude because of the nature of their respective roles. For instance, it is very unlikely that a suspect goes through the interview with the objective of developing a warm relationship with the interviewer. To recognize that the roles are very different, and to observe the different behaviors in the analysis, the suspects’ behaviors classified as part of the rapport-based domain are very different from those of the interviewer. While the interviewer attitude of relationship building is observed, the opposite attitude is taken into consideration for the suspect, which is, behaviors that are totally non-collaborative, including active denial of the accusations or remaining silent. A total of 4 behaviors are included in this category.

Emotion provocation and Emotion response

Kelly and colleagues (2013) grouped interviewer strategies according to their underlying principle or aim. Their category of Emotion provocation and response is based on the aim of affecting the suspect’s feelings in order to trigger a response and all strategies that provoke emotions, positive or negative, are included, such as behaviors that involve identifying and exaggerating the suspect’s fears to produce negative emotions such as stress or offering rationalizations that allow the suspect to see the crime as less grave and therefore decrease the level of stress. In the present study, whether the emotion provoked is positive or negative is considered to have different consequences for suspect disclosure. Some negative emotions, such as guilt and anxiety, have been associated with a higher level of confession (Deslauriers-Varin et al.,2011; Gudjonsson, 2003; Sigurdsson et Gudjonsson, 1994; St-Yves, 2002); the effect of positive emotions has not been investigated. Given this, in the present study the category of Emotion provocation includes only those behaviors that lead to negative emotions. For example, if the interviewer suggests that an accomplice could be interviewed, this could lead to an increase in the stress felt by the suspect and is therefore included in this category. Following this logic, two behaviors that are in the Confrontation category for Kelly and colleagues (2013) were judged to be likely to trigger negative emotions such as anxiety and have therefore been moved to the Emotional provocation and response category. The first one is Taking a judgmental attitude toward the suspect since the suspect may see as insulting. The second is Identifying and exaggerating fears by emphasizing the consequences of arrest because it is a behavior that have been highly documented in the literature (see for example Gudjonsson, 2003; St-Yves & Meissner, 2014) as leading to considerable levels of anxiety and distress. Finally, strategies such as reducing fear by minimizing the gravity of the crime or by offering a moral justification for it were also moved to the Collaboration category (instead of the Confrontation category) because the objective of this technique is to reduce anxiety rather than triggering negative emotion. A total of nine interviewer behaviors are included in this category.

To make it possible to consider suspect behavior, the category dealing with emotions was adapted to include Emotion response. This adaptation follows the basic idea of the original taxonomy, which was to account for the largest possible spectrum of raw emotions. In contrast to the interviewer category, all suspect behaviors that demonstrate an emotional response, whether positive or negative, are included in this category (e.g., crying). As some Emotion provocation strategies used by the interviewer are aimed at increasing the suspect’s guilt or anxiety (Kassin et al., 2007), suspect behaviors that are intended at reducing anxiety are also included; for instance, offering a moral justification for an act (e.g., “It was not a crime since the child wanted to have sexual interaction”) or avoiding taking responsibility by blaming someone else can be used by the suspect to reduce anxiety and therefore fall into this category. A total of 6 suspect behaviors are included.

Confrontation

For Kelly and colleagues (2013), the Confrontation category includes behaviors in which the interviewer asserts authority and control over the suspect as a way to achieve compliance. These behaviors include expressing impatience, frustration, or anger, or suggesting negative outcomes for the suspect. They also include behaviors aimed at gaining compliance, such as confronting the suspect with discrepancies in his information. A total of 18 different interviewer behaviors are included in this category.

For the suspect, behaviors in this category include arrogance and attempts at domination, for example, trying to control the interview by making inadequate requests for the situation or by confronting the interviewer. Interrupting the interviewer or adopting an unfriendly stance that includes the use of sarcasm are also examples of the 11 behaviors included in this category.

Collaboration

The inverse of the Confrontation category, Collaboration is aimed at minimizing or eliminating authority and control over the other player (Kelly et al., 2013). For the interviewer, it includes positive behaviors that are not part of the Rapport and relationship building category, such as congratulating the suspect for efforts and collaborations. Strategies aimed at triggering positive emotions by reducing fear through minimizing the gravity of the crime or offering moral justifications for it are considered to be in this category, which includes a total of 5 different interviewer behaviors.

For the suspect, collaboration is associated with cooperation, participation, or the appearance of being eager to please. It can also include expressing remorse or self-criticism for having committed the act. A total of 6 suspect behaviors are included in this category.

Information related to the case

The last category in Kelly et al.’s taxonomy is Presentation of evidence, in which the interviewer provides documentation of the suspect’s guilt or complicity. Presentation of evidence is a central subject in research on investigative interviews and is closely correlated with an increase in the rate of confession (e.g., Brimbal & Luke, 2019; Deslauriers-Varin et al., 2011; 2020; Moston & Engelberg, 2011). The title of this category was adapted for the present study to make it possible to include suspect behaviors. Interviewer behaviors in this category include all information related to the case presented by the interviewer. An information is judged related to the case when it concerns the person who committed the actions, the details related to the action, the location where actions were posed, and the temporal information of the actions (see Oxburgh et al., 2010). For example, such as showing a picture found on the suspect’s computer or mentioning a victim’s name would be elements related to the case. The category includes three different types of interviewer behaviors.

For the suspect, this category includes providing IRI by answering a question from the interviewer concerning the actions and admitting guilt or providing additional information. Similar than for the interviewer, all information related to “who did what where and when” are relevant to the investigation. This category includes 3 different types of suspect behaviors.

Analysis

An investigative interview is a process in which the act of a player at a point in time is, at least in part, contingent upon past acts. Data in the present study were behaviors captured as discrete states on a continuous timeline. The simplest mathematical model of a process involving discrete and continuous times is a Markov model (Brent & Sykes, 1979), a sequence of events in which each event, or state, depends on one or more previous states (Manderscheid et al., 1982). The probabilistic nature of the process can be represented by an initial distribution of states and a transition matrix that specifies the probability of moving from one state to another. An important aspect of Markov models is their recognition of indeterminacy: individual behaviors may vary widely while the overall pattern is maintained, creating a statistical approach that is more reflective of human behavior than models that use estimates of parameters based on means (Hewes, 1975; Leik & Meeker, 1975).

Figure I - Proposed visual presentation of the Markov database concept

We wanted to determine whether a given behavior at time 1 influences subject disclosure at time 2. Each behavior was paired with the one that followed it and then paired with the one after that; that behavior was then paired with the one following, and so on. Figure 1 provides a visual presentation of the Markov data concept developed for this study. Because the objective was to observe the impact of behavior on disclosure, only dyads with disclosure of IRI at time 2 were kept for analysis. Of the 2,561,968 pairs in the data, 89,806 included a suspect presentation of IRI at time 2 and were therefore kept for analysis. Equation 1 presents a logistic regression model predicting the probability that the suspect will provide an IRI (P_info) at a given time (t), with the behavior preceding it at time 1 (beha) as a covariate. To account for dependencies among observations (i.e., each behavior at time B can have more than one pairing), a random effect was included at the interview level (u_e). The coefficient β1 tests the gain of information hypothesis and estimates whether the presence of a behavior at time 1 increases the probability that information will be provided. A positive coefficient indicates that if behavior at time 1 occurs, the probability that the suspect will provide information increases. A negative coefficient indicates that if behavior at time 1 occurs, the probability that the suspect will provide information decreases.

1. Log (P_infoT2/1-P_info) = β0 + β1 + behaT1 * time (t2-t1) + u_e

Generalized linear mixed models

Generalized linear models (GLM) are an extension of linear regression models and are aligned with the data using the maximum likelihood method, making it possible to provide not only estimates of the regression coefficients but also estimated asymptotic standard errors for the coefficients (Fox, 2016), which permits use of non-normal dependent variables. Generalized linear mixed models (GLMM) add a multilevel random effect (Demidenko, 2004) and can combine repeated measurements and process them simultaneously, thus significantly increasing the statistical power of the analysis. Classical statistical analyses assume that observations are independent and identically distributed but this assumption applied to clustered data can lead to false results. The mixed effects model treats clustered data adequately and considers two sources of variation: within clusters, (i.e., variation within an individual - intrasubject variance), and between clusters (i.e., variation between individuals - inter-subject variance). When the number of clusters is small and the number of observations per cluster is large, it is considered best to treat the cluster-specific coefficients as fixed (Demidenko, 2004). There are three assumptions made in using GLMM: 1) the random effect comes from a normal distribution; 2) the chosen link function is appropriate – in this case the logit for binomial data was used; 3) estimation of variance shows an appropriately dispersed variability.

GLMM analyses makes it possible to predict whether behavior at time 2 is influenced by behavior at time 1 as well as the force of this influence. It also makes it possible to insert a polynomial condition – allowing to observe curve in the data instead of a line – to capture the interaction between the variable and the time at which it occurs. One of the advantages of using GLMM is that the structure of the linear predictor has the familiar structure of a linear model (Fox, 2016). This type of statistical analysis makes it possible to assume that observations are independent by adding a random effect for each interview. The software R version 4.1.2 was used to run the analysis with the lme4 package.

The sample was composed of interviews of different lengths, creating the possibility that longer interviews might play a more decisive role in the model. To make it possible to observe the effect of time, interviews must be comparable. The time variable was therefore normalized (scaled) using the mean value and standard deviation to avoid this potential problem.

Results

Logistic regression analysis was used to meet two of the study objectives. The first objective was to identify behaviors that were more likely to be followed by disclosure of IRI by the suspect, i.e., behaviors that influence disclosure. The second was to look at whether the effect of such behaviors lasted over the course of the interview. Analysis was performed on every dyad of behaviors that included a disclosure from the suspect at time 2.

One approach to investigate the interaction between variables is to look at the odd ratio of each one. The odd ratio predicting disclosure derived from the logistic regression analysis is provided in Appendix B. However, interpreting the effect of the interaction term may be challenging (Li & Barry, 2012) especially for categorical variables. The logistic regression compares one category to the others. In our case, the effect of the interviewer behavior of collaboration on suspect disclosure is compared to the other behaviors of the interviewer but also to the behaviors of the suspect, which complicate the interpretation of the model. One solution would have been to run post hoc tests with a model for each behavior as a reference category to be compared to the others in order to create a matrix of comparison. However, each model takes around 12 hours to be generated because of the random effect and the polynomial effect of time which resulted in the elimination of this strategy. Also, the coefficients of the table are difficult to interpret because of the polynomial interaction effect. For transparency purposes, the odd ratios are presented in Appendix B and the simplified model without interaction effect is compared to the actual model discussed in this paper.

Another approach, that is an equally valid technique for exploring the nature of an interaction in a logistic regression model than the odd ratio, is to calculate predicted probabilities (Li & Barry, 2012). Because the interpretation of the behaviors on one another with the interaction of time can be confusing, researchers agree to say that it is typically useful to interpret results using a graph (e.g., Aguinis & Gottfredson, 2010; Aiken et al., 1991). The predicted value of the model, the probability of behaviors to be followed by a IRI from the suspect, is therefore presented in this article in order to facilitate the interpretation of the variables’ coefficients.

Figure II to Figure V show the predicted probabilities of disclosure for each behavior. It is important to keep in mind that there were 24 303 behaviors from the interviewer and the suspect altogether and those have been compared to the probability of being followed by a disclosure which represents 994 of the behavior (3.9%). Based on the predicted probability, the following results section presents the immediate effect of the behaviors on disclosure and then, the impact of time is explored.

Behaviors preceding disclosure of an IRI by the suspect

Behaviors of the interviewer related to disclosure by the suspect at time 2 are presented in Figure II. Information related to the case presented by the interviewer (e.g., presentation of evidence) was one of the best predictors that an IRI would be disclosed by the suspect. Inversely, emotion provocation was unlikely to be followed by an IRI.

Figure II - Probability that an interviewer’s behavior will be followed immediately by IRI from the suspect

Disclosure of IRI is not always linked directly to an interviewer behavior but can be related to subject behavior, as shown in Figure III. The variable with the highest predictive value was having already disclosed an IRI, with the second-best predictor the suspect’s emotional response. Silence, active denial, or confronting the interviewer were less likely to be followed by subject disclosure.

Figure III - Probability of the suspect’s behavior being followed immediately by disclosure

Time effect

Figures 2 and 3 show the probability that a behavior will be immediately followed by IRI. However, the effect on disclosure is not always immediate, and even when the effect is strong and immediate (as it is with provision of information related to the case), its impact may decrease almost immediately. Behaviors other than providing information seem to have an affect later in the interview (see figures 4 and 5).

With interviewer strategies (shown in figure IV), a collaborative strategy (e.g., providing justification for the crime) seems to have a positive influence on the probability of disclosure after approximately 60 minutes, as illustrated by the orange line. Strategies oriented toward developing rapport and relationship (e.g., finding common grounds; blue line) seem to have an optimal impact after 40 minutes. The effect of strategies oriented toward provoking emotion appears after almost 70 minutes (green line). Confrontational strategies (e.g., asking questions that presume guilt) have a stable and low impact on disclosure.

Figure IV - Interviewer’s behaviors prediction of an IRI

For the suspect (see figure V), strategies such as confrontation or active denial were unlikely to be followed immediately by disclosure, but these negative effects diminished over time (purple and red lines) and the probability of disclosure increased around 60 minutes after a confrontational behavior by the suspect. Collaborative behavior from the suspect has a positive but modest impact on disclosure overall and its effect diminishes slowly (dark purple line).

Figure V - Suspect behaviors predictive of IRI

Discussion

The notions of time, interaction, and players were applied to analyze investigative interviews. Because some effects have been observed and more precise information about players’ decision in the interview has been obtained, the findings suggest that Game Theory helps understand the disclosure process. Investigative interviewing is a dynamic process and suspect disclosure is influenced by the behaviors of both players. Different strategies have different effects, with some increasing the probability of IRI, while others decrease it. Results also show that the passage of time seems to have an impact on the relation of behavior to disclosure. Dividing the interviews in our data base into segments linked to seconds rather than several minutes made it possible to analyze behavior during investigative interviews as a continuous process, rather than dichotomous situation of disclosure/nondisclosure, while using Markov chains made it possible to relate the passage of time to behaviors. These two important findings are discussed below in relation to the literature on suspect disclosure.

The impact of strategies on suspect disclosure

The interviewer’s presentation of evidence relating to the case increased the probability of IRI. This supports previous findings that the subject’s perception of how much evidence is available affects the decision to conceal or reveal information (Bull, 2014; Luke et al., 2014; Luke et al., 2016; Tekin et al., 2015). Confrontational behaviors, such as asking negative questions, decrease the probability of disclosure, a finding that also supports previous research (Kelly et al., 2016; Leahy-Harland & Bull, 2017; Verhoeven, 2018).

Two suspect behaviors – providing IRI and showing an emotional response, such as crying – were often followed immediately by disclosure of IRI. Collaboration and demonstrating vulnerability by showing emotion are also associated with disclosure. Future research might investigate the suspects’ behavior during interviews to see if there are identifiable profiles of suspect based on their behavior that would be associated with disclosure. Results also show the advantage of increasing our empirical knowledge about suspect behavior and strategies during investigative interviews to help increase the effectiveness of interviewer strategies.

The impact of time on behaviors

The effect of the passage of time during an interview has seldom been considered in the literature on investigative interviews. For example, previous research has shown that presentation of evidence to the suspect is associated with confession (Granhag et al., 2013; Hartwig et al., 2014; Vrij & Granhag, 2012) and the present study not only supports this association but demonstrates that the effect diminishes over time: if the suspect does not provide an IRI immediately after the presentation of evidence, the chances of disclosure related to this strategy decrease substantially. Inversely, the influence of some behaviors is apparent only in the long-term. Interviewer strategies involving collaboration, provoking emotions, and rapport building have a large impact on the disclosure of IRI but their effect is apparent after a certain period of time. Previous research on confession has found that developing a warm and humane relationship in investigative interviews leads to an increase in disclosure (Leahy-Harland & Bull, 2017; Snook et al., 2015; St-Yves et Kebbell, 2018). Our results support this finding and also suggest that the interviewer must be patient to see the effect of this strategy. The effect of a given strategy or behavior is not always immediately apparent, demonstrating the need to study disclosure in investigative interviews as a dynamic, continuous, step-by-step process rather than a dichotomous event.

A delayed effect of strategy and behavior on disclosure can also be observed for the suspect. Emotional response and provision of IRI were found to be linked to immediate disclosure of IRI, while the longer the period of time after an IRI, the lower the probability that other IRI will be provided. This suggests that suspects tend to provide information in bursts rather than scattered throughout the interview. Active denial and confrontation did not show any immediate relation to disclosure, but this effect diminished over time, suggesting that such behaviors do not indicate that the suspect will not provide information at a later stage in the interview.

These findings raise concerns about the length of investigative interviews. The length of interviews tends to vary according to the characteristics of the crime involved (e.g., type of crime, number of charges, number of victims, amount of evidence available). Studies in the United States suggest interviews are approximately 1.5 hours long (see Kassin et al., 2007; Kelly et al., 2016). Similar results were found in Canadian settings (King & Snook, 2009). The present study suggests that some strategies have an impact on disclosure only after a considerable amount of time. For example, the effect on disclosure of interviewer strategy intended to elicit emotion by identifying and exaggerating the subject’s fears was apparent only after more than an hour, suggesting that this strategy should not be used in shorter interviews. Presentation of evidence, however, appears to be useful in interviews of any length as its impact on disclosure was immediate.

The gradual presentation of evidence

Research on the strategic use of evidence (SUE) suggests that evidence should be introduced gradually to increase the suspect’s perception that the interviewer has access to other evidence that has not been disclosed (Tekin et al., 2015; Tekin et al., 2016). However, we found that even when the evidence was presented gradually, the IRIs revealed by the suspect occurred within a short time: there was a high probability that one IRI would be followed fairly quickly by another, rather than being disclosed gradually throughout the interview. Further research is needed to explain this finding. A possible hypothesis is that the suspect tends to give a burst of information each time rather than one information at the time. Future research should investigate how this burst of information relates to the evidence presentation process and to other possibly implicated variables.

Limits of the study

The interviews were an analyzed as a series of behaviors on a timeline and were broken into dyads, making it possible to analyze each behavior in relation to the one that followed it. Consequently, the same interview repeats itself many times through the dyads that were analyzed. This means that the data were multileveled (i.e., nested). To ensure the independence of the data, a random effect was included in the formula. The analysis is therefore able to take into account when the result is related to the same interview and adjust the standard error accordingly.

Also related to the methodology of using dyads of behaviors to account for the impact of time, another problem arose: the cumulative effect of past behaviors was not calculated and might have an even more important effect on disclosure. For example, the fact that a specific category of behaviors repeated many times before the disclosure might influence the decision of the players. Or the interaction of specific categories of behavior might create a reaction that could not have been measure here. This is an important element that was not taken into consideration in this exploratory study but one that should be considered in future research.

From the start of the project, the interpretation and rating of interrogation techniques were discussed on a regular basis between the research assistant and the authors, a process which can be viewed as an alternative to an inter-observer reliability analysis. In case of discrepancies, the coding was then revised accordingly in the coding scheme. For methodological reasons, one sole research assistant was trained and assigned to do the remaining coding of the interviews used in the current study. Inter-rater correlations therefore could not be calculated. The results of the current study, although based on a particularly large sample of interviews, might still be the reflect of one’s perception of the behaviors and interview process and need to be interpreted as such.

As indicated by the demographic information, the sample was highly homogeneous: suspects were largely white males who had committed an offense related to child sexual exploitation. The results may therefore not be generalizable to other types of offenses. The homogeneity of the sample should, however, allow useful comparisons with future studies using samples of different offenders and different crimes. The type of crime might also have impacted the interviews on many aspects which are important to keep in mind when comparing samples. Whether sex offenders differ from other offenders is often discussed in the literature. For example, sex offenders are likely to have a higher level of education (Babchishin et al., 2011; Faust et al., 2015) and a higher rate of employment (Clevenger et al., 2016; Faust et al.,2015). Cybersexual offenders are less likely to have a criminal history than those who commit other types of sexual crimes (Aebi et al., 2014; Faust et al., 2015; Smid et al., 2015) and also show fewer antisocial traits than other offender groups (Babchishin et al., 2015; Magaletta et al., 2014). Some studies suggest that an individual who has committed a serious criminal act will try to avoid the consequences of the act by denying the accusations (Gudjonsson, 2003; Moston et al., 1992; Phillips & Brown, 1998; St-Yves, 2004) while other researchers have found the confession rate of sex offenders to be higher than that of other kinds of offenders (Mitchell, 1983; Gudjonsson & Sigurdsson, 2000), particularly when the offense has been committed more frequently (Lippert et al., 2010). The level of confession in the sample is high and very often, when the suspects confess and gives details, it increases the amount of IRI provided during an interview, making such interviews of particular interest for this study.

Finally, the Canadian context as well as the interview protocol of the ICE unit of the SQ may also have played a role in the findings of the study. For example, the length of the interviews in our database tended to be longer than those observed in previous studies, which may have had an influence on strategies used as well as suspect reactions. The fact that the sample only includes dyadic situations limits the transferability of the study to other countries given the number of countries permitting a lawyer inside the interview room. The dyads analysis contributed to develop the understanding and the theory around this main police-suspect interaction, but also proposed a methodology that can eventually be tested in other type of settings.

Conclusion

This study used the perspective provided by Game Theory to analyze investigative interviews as a dynamic social interaction involving both suspect and interviewer. Game Theory makes it possible to account for the dynamic process of disclosure, taking into account the impact of player behaviors as well as the effect of the passage of time on behaviors. Markov chains were used to structure the database and a multilevel analysis was then used to investigate the results, making it possible to weigh both the influence of each category of strategies on disclosure and the effect of time on the behaviors studied. The findings indicate that this theoretical frame is useful in understanding investigative interviews and disclosure and suggest the ways both interviewer and subject behaviors are related to the probability of disclosure, as well as the effect of the passage of time on these behaviors. These findings have important implications for police investigators. First, they suggest which strategies are most effective to increase the probability of disclosure. Second, they provide important indications of how strategies should be used in relation to the timeframe available for a specific interview. The new theoretical and methodological approach to analyzing investigative interviews presented here provides a starting point for developing guidelines for practitioners that can help increase the level of disclosure during investigative interviews.

References

Abbe, A., & Brandon, S. E. (2013). The role of rapport in investigative interviewing: A review. Journal of investigative psychology and offender profiling, 10(3), 237-249. DOI: 10.1002/jip.1386

Adams-Quackenbush, N. M. (2019). The influence of guilt presumptive language on investigative interview outcomes. Doctoral Thesis, Maastricht University.

Aebi, M., Plattner, B., Ernest, M., Kaszynski, K. & Bessler, C. (2014). Criminal History and Future Offending of Juveniles Convicted of the Possession of Child Pornography. Sexual Abuse: A Journal of Research and Treatment, 26 (4), 375–390. DOI: 10.1177/1079063213492344

Aguinis, H and Gottfredson, R K (2010). Best-practice recommendations forestimating interaction effects using moderated multiple regression. Journal oforganizational behavior, 31 (6), 776–786. DOI : 10.1002/job.686.

Aiken, L S, West, S G and Reno, R R (1991). Multiple regression:Testing and interpreting interactions. United States:Sage Publications.

Babchishin, K. M., Hanson, R., K., & Hermann, C. A. (2011). The characteristics of online sex offenders: A meta-analysis. Sexual Abuse, 23(1), 92-123. DOI: 10.1177/1079063210370708

Babchishin, K. M., Hanson, R. K., & VanZuylen, H. (2015). Online child pornography offenders are different: A meta-analysis of the characteristics of online and offline sex offenders against children. Archives of sexual behavior, 44(1), 45-66. DOI: 10.1007/s10508-014-0270-x

Baker-Eck, B and Bull, R (2022).Effects of empathyand question types on suspects’ provision of information in investigative interviews. International Journal of Police Science and Management, 1–11. DOI : : 10.1177/14613557221106073.

Baker-Eck, B, Bull, R and Walsh, D (2021). Investigative empathy: Five types of cognitive empathy in a field study of investigative interviews with suspects of sexual offences. Investigative Interviewing: Research and Practice, 11, 28–38.

Bang, B. L., Stanton, D., Hemmens, C., & Stohr, M. K. (2018). Police recording of custodial interrogations: A state-by-state legal inquiry. International Journal of Police Science & Management, 20(1), 3-18. DOI:10.1177/1461355717750172

Beauregard, E., Deslauriers-Varin, N., & St-Yves, M. (2010). Interactions between factors related to the decision of sex offenders to confess during police interrogation: A classification-tree approach. Annals of Sex Research, 22 (3), 343-367. DOI: 10.1177/1079063210370707

Bicchieri, C. (2004). Rationality and Game Theory. In Mele, A., R. & Rawling, P. (Eds.) The Oxford Handbook of Rationality, Oxford University Press, pp.182-205.

Brent Jr, E. E., & Sykes, R. E. (1979). A mathematical model of symbolic interaction between police and suspects. Behavioral Science, 24(6), 388-402. DOI: 10.1002/bs.3830240605

Brimbal, L., & Luke, T. J. (2019). Deconstructing the evidence: The effects of strength and reliability of evidence on suspect behavior and counter-interrogation tactics. DOI: 10.31234/osf.io/vrs7z

Bull, R (2014). When in interviews to disclose information to suspects and to challenge them?. In: R Bull, ed. Investigative Interviewing. Springer.

Bull, R and Rachlew, A (2020). Investigative interviewing: From England to Norway and beyond. In: S Barela, M Fallon, and J Ohlin,eds. Interrogation and torture: Research on efficacy, and its integration with morality and legality. United-States: Oxford University Press, 171–196.

Bull, R., & Soukara, S. (2010). Four studies of what really happens in police interviews. In Lassiter, G. D. & Meissner ,C. A. (Eds.), Police interrogations and false confessions: Current research, practice, and policy recommendations (pp. 81–95). American Psychological Association. https://doi.org/10.1037/12085-005

Clarke, C., Milne, R., & Bull, R. (2011). Interviewing suspects of crime: The impact of PEACE training, supervision and the presence of a legal advisor. Journal of investigative psychology and offender profiling, 8(2), 149-162. DOI: 10.1002/jip.144

Cleary, H and Bull, R (2021). Contextual factors predict self-reported confession decisionmaking: A field study of suspects’ actual police interrogation experiences. Law and Human Behavior, 45 (4), 310–323. DOI: 10.1037/lhb0000459.

Clemens, F., Knieps, M., & Tekin, S. (2020). Untapped potential? A survey study with German police officers into suspect interviewing practices and the strategic use of evidence. Journal of Forensic Psychology Research and Practice, 20(1), 53-79. DOI: 10.1080/24732850.2019.1684123

Clevenger, S. L., Navarro, J. N. & Jasinski, J. L. (2016). A Matter of Low Self-Control? Exploring Differences Between Child Pornography Possessors and Child Pornography Producers/Distributers Using Self-Control Theory. Sexual Abuse: A Journal of Research and Treatment, 28 (6), 555–571. DOI: 10.1177/1079063214557173

Collins, K., & Carthy, N. (2019). No rapport, no comment: The relationship between rapport and communication during investigative interviews with suspects. Journal of Investigative Psychology and Offender Profiling, 16(1), 18-31. DOI: 10.1002/jip.1517

Cornish, D. & Clarke, R. (1986). The reasoning criminal: rational choice perspectives on offending. Springer-Verlag.

Demidenko, E. (2004). Mixed models: theory and applications (1st ed., Wiley series in probability and statistics). Wiley-Interscience.

Deslauriers-Varin, N., Beauregard, E., & Wong, J. (2011). Changing Their Mind About Confessing to Police: The Role of Contextual Factors in Crime Confession. Police Quarterly, 14 (1), 5–24. DOI: 10.1177/1098611110392721

Deslauriers-Varin, N., & Bergeron, A. (2021). Suspect Interviews and Confessions in Sex Crime Cases. In Bennel, C. & Deslauriers-Varin, N. (Ed.). Criminal Investigations of Sexual Offenses, Spinger, 99-113.

Deslauriers-Varin, N. (2020). Facteurs explicatifs de la confession en contexte d’interrogatoire policier: une analyse d’arbres décisionnels. Criminologie, 53(2), 219-254. DOI: 10.7202/1074194ar

Deslauriers-Varin, N (2022). Confession during Police Interrogation: A Decision Tree Analysis. Journal of Police and Criminal Psychology, 37, 526–539. DOI: 10.1007/s11896-022-09507-9.

Faust, E., Bickart, W., Renaud, C., & Camp, S. (2015), Child Pornography Possessors and Child Contact Sex Offenders: A Multilevel Comparison of Demographic Characteristics and Rates of Recidivism. Sexual Abuse: A Journal of Research and Treatment, 27 (5) 460–478. DOI: 10.1177/1079063214521469

Fox, W. P. (2016). Applied Game Theory to Improve Strategic and Tactical Military Decisions. Journal of Defense Management, 6, 147.

Goffman, E. (1969). Strategic interaction. University of Pennsylvania Press.

Granhag, P. A., Mac Giolla, E., Strömwall, L. A., & Rangmar, J. (2013). Counter-interrogation strategies among small cells of suspects. Psychiatry, Psychology and Law, 20(5), 705-712. DOI: 10.1080/13218719.2012.729021

Gudjonsson, G. H. (2003). The psychology of interrogations and confessions: A handbook. Chichester: John Wiley.

Gudjonsson, G. H., & Petursson, H. (1991). Custodial interrogation: Why do suspects confess and how does it relate to their crime, attitude and personality?. Personality and Individual Differences, 12(3), 295-306. DOI: 10.1016/0191-8869(91)90116-S

Gudjonsson, G.H. & Sigurdsson, J.F. (2000). Differences and similarities between violent offenders and sex offenders. Child Abuse and Neglect, 24, 363-372. DOI: 10.1016/S0145-2134(99)00150-7

Håkansson, C. (2019). Do Suspects Use the Counter-Interrogation Strategies They Say They Use?. [Master Thesis]. University Of Gothenburg Institution of Psychology.

Hartwig, M., Granhag, P. A., & Luke, T. (2014). Strategic use of evidence during investigative interviews: The state of the science. Credibility assessment, 1-36. DOI: 10.1016/B978-0-12-394433-7.00001-4

Hewes, D. E. (1975). Statistical tests for Markov chains. Markov Chain Tutorial, International Communication Association, Chicago.

Holmberg, U., & Christianson, S. (2002). Murderers’ and Sexual Offenders’ Experiences of Police Interviews and Their Inclination to Admit or Deny Crimes. Behavioral Sciences and the Law, 20, 31–45. DOI: 10.1002/bsl.470

Kassin, S. M., Leo, R. A., Meissner, C. A., Richman, K. D., Colwell, L. H., Leach, A. M., & La Fon, D. (2007). Police interviewing and interrogation: A self-report survey of police practices and beliefs. Law and human behavior, 31(4), 381-400.

Kebbell, M., Alison, L., Hurren, E., & Mazerolle, P. (2010). How do sex offenders think the police should interview to elicit confessions from sex offenders? Psychology, crime & law, 16(7), 567-584. DOI: 10.1080/10683160902971055

Kelly, A. (2003). Decision making using Game Theory. Cambridge University Press.

Kelly, C. E., Miller, J. C., Redlich, A. D., & Kleinman, S. M. (2013). A taxonomy of interrogation methods. Psychology, public policy, and law, 19(2), 165.

Kelly, C.E., Miller, J.C., & Redlich, A.D. (2015). The dynamic nature of interrogation. Law and Human Behaviour, 40(3), 295–309. DOI: 10.1037/lhb0000172

Kelly, C E, Russano, M B, Miller, J C and Redlich, A D (2019). On the road (to admission):Engaging suspects with minimization. Public Policy, and Law, 25 (3), 166–180. DOI: 10.1037/law0000199.

Kelly, C. E., & Valencia, E. J. (2021). You ask and do not receive, because you ask wrongly. International Journal of Police Science & Management, 23(1), 42-54. DOI: 10.1177/1461355720955077

King, L., & Snook, B. (2009). Peering inside a Canadian interrogation room: An examination of the Reid model of interrogation, influence tactics, and coercive strategies. Criminal Justice and Behavior, 36(7), 674-694. DOI: 10.1177/0093854809335142

Leahy-Harland, S., & Bull, R. (2017). Police Strategies and Suspect Responses in Real-Life Serious Crime Interviews. Journal of Police and Criminal Psychology, 32, 138-151.

Leik, R. K., & Meeker, B. F. (1975). Mathematical sociology. Prentice-Hall.

Li, H and Barry, J (2012). Interpreting interactions in logistic regression. Cornell Statistical Consulting Unit Report.

Lippert, T., Cross, T. P., Jones, L. & Walsh, W. (2010). Suspect Confession of Child Sexual Abuse to Investigators. Child Maltreatment, 15 (2), 161-170. DOI: 10.1177/1077559509360251

Luke, T. J., & Granhag, P. A. (2020). The Shift-of-Strategy (SoS) approach: Using evidence strategically to influence suspects’ counter-interrogation strategies. Psychology Crime & Law, 1-26. DOI: 10.1080/1068316X.2022.2030738

Luke, T. J., Dawson, E., Hartwig, M., & Granhag, P. A. (2014). How awareness of possible evidence induces forthcoming counter‐interrogation strategies. Applied Cognitive Psychology, 28(6), 876-882. DOI: 10.1002/acp.3019

Luke, T. J., Hartwig, M., Shamash, B., & Granhag, P. A. (2016). Countermeasures against the à strategic use of evidence technique: Effects on suspects' strategies. Journal of Investigative Psychology and Offender Profiling, 13(2), 131-147. DOI: 10.1002/jip.1448

Madon, S.,Yang, Y., Smalarz, L., Guyll, M., & Scherr, K. C. (2013). How Factors Present During the Immediate Interrogation Situation Produce Short-Sighted Confession Decisions. Law and Human Behavior, 37 (1), 60–74. DOI: 10.1037/lhb0000011

Magaletta, P. R., Faust, W. Bickart, W., & McLearen, A. M. (2014). Exploring Clinical and Personality Characteristics of Adult Male Internet-Only Child Pornography Offenders. International Journal of Offender Therapy and Comparative Criminology, 58 (2), 137–153. DOI: 10.1177/0306624X12465271

Manderscheid, R. W., Rae, D. S., McCarrick, A. K., & Silbergeld, S. (1982). A stochastic model of relational control in dyadic interaction. American Sociological Review, 62-75. DOI: 10.2307/2095042

May, L., Granhag, P. A., & Tekin, S. (2017). Interviewing suspects in denial: On how different evidence disclosure modes affect the elicitation of new critical information. Frontiers in psychology, 8, 1154. DOI: 10.3389/fpsyg.2017.01154

Meissner, C. A., Surmon-Böhr, F., Oleszkiewicz, S., & Alison, L. J. (2017). Developing an evidence-based perspective on interrogation: A review of the US government’s high-value detainee interrogation group research program. Psychology, Public Policy, and Law, 23(4), 438. DOI: 10.1037/law0000136

Mitchell, B. (1983). Confessions and police interrogation of suspects. Criminal Law Review, September, 596-604.

Moston, S., & Engelberg, T. (2011). The effects of evidence on the outcome of interviews with criminal suspects. Police Practice and Research, 12(6), 518-526. DOI: 10.1080/15614263.2011.563963

Moston, S. J., Stephenson, G. M., & Williamson, T. M. (1992). The effects of case characteristics on suspect behavior during police questioning. British Journal of Criminology, 32, 23–40. DOI:10.1093/oxfordjournals.bjc.a048178

Oleszkiewicz, S, Granhag, P A and Kleinman, S M (2017a).Eliciting information from human sources:Training handlers in the Scharff technique. Legal and Criminological Psychology, 22 (2), 400–419. DOI: 10.1111/lcrp.12108.

Oleszkiewicz, S, Granhag, P A and Kleinman, S M. (2017b). Gathering human intelligence via repeated interviewing: Further empirical tests of the Scharff technique. Psychology, Crime and Law, 23 (7), 666–681. DOI : 10.1080/1068316X.2017.1296150.

Oxburgh, G., & Ost, J. (2011). The use and efficacy of empathy in police interviews with suspects of sexual offences. Journal of Investigative Psychology and Offender Profiling, 8(2), 178-188. DOI:10.1002/jip.143

Oxburgh, G., Ost, J., & Cherryman, J. (2010). Police interviews with suspected child sex offenders: does use of empathy and question type influence the amount of investigation relevant information obtained? Psychology, Crime & Law, 18(3), 259–273. DOI:10.1080/1068316x.2010.481624

Pearse J., & Gudjonsson, G. (1999). Measuring influential police interviewing tactics: a factor analytical approach. Legal and Criminology Psychology, 4, 221–238. DOI:10.1348/135532599167860

Phillips, C., & Brown, D. (1998). Entry into the criminal justice system: A survey of police arrests and their outcomes. London: Home Office.

Sigurdsson, J. F., & Gudjonsson, G. H. (1994). Alcohol and drug intoxication during police interrogation and the reasons why suspects confess to the police. Addiction,89, 985–997. DOI:10.1111/j.1360-0443.1994.tb03358.x

Smid, W., Schepers, K., Kamphuis, J. H., van Linden, S., & Bartling, S. (2015). Prioritizing Child Pornography Notifications: Predicting Direct Victimization. Sexual Abuse, 27 (4), 398 – 413. DOI: 10.1177/1079063213514453

Snook, B., Barron, T., Fallon, L., Kassin, S. M., Kleinman, S., Leo, R. A., ... & Trainum, J. L. (2020). Urgent issues and prospects in reforming interrogation practices in the United States and Canada. Legal and criminological psychology, 26(1), 1-24. DOI: 10.1111/lcrp.12178

Snook, B., Brooks, D., & Bull, R. (2015). A lesson on interrogations from detainees: predicting self-reported confessions and cooperation. Criminal Justice and Behavior, 42(12), 1243-1260. DOI: 10.1177/0093854815604179

Sorochinski, M., Hartwig, M., Osborne, J., Wilkins, E., Marsh, J., Kazakov, D., & Granhag, P. A. (2014). Interviewing to detect deception: When to disclose the evidence? Journal of Police and Criminal Psychology, 29(2), 87-94.

St-Yves, M. (2002). Interrogatoire de police et crime sexuel: Profil du suspect collaborateur. Revue Internationales de Criminologie & de Police Technique & Scientifique, 1, 81-96.

St-Yves, M. (2004). Les facteurs associés à la confession: la recherche empirique. In St-Yves, M. & Landry, J. (Eds.). Psychologie des entrevues d’enquête, de la recherche à la pratique (pp. 53-71). Cowansville, QC: Yvon Blais.

St-Yves, M and Kebbell, M (2018). Investigative interviewing in suspected sexual homicides. Routledge international handbook of sexual homicide studies, 562–573.

St-Yves, M., & Meissner, C. A. (2014). Interviewing suspects. In St-Yves, M. (Eds). Investigative interviewing: The essentials, 145-190. Carswell.

Tekin S., Granhag, P. A., Strömwall L., Giolla E. M., Vrij A., & Hartwig M. (2015). Interviewing Strategically to Elicit Admissions from Guilty Suspects. Law and Human Behavior, 39 (3), 244-252.

Tekin, S., Granhag, P. A., Strömwall, L. A. & Vrijd, A. (2016). How to make perpetrators in denial disclose more information about their crimes. Psychology, crime & law, 22 (6), 561-580. DOI: 10.1080/1068316X.2016.1168425

Verhoeven, W. J. (2018). The complex relationship between interrogation techniques, suspects changing their statement and legal assistance. Evidence from a Dutch sample of police interviews. Policing and Society, 28(3), 308-327. DOI: 10.1080/10439463.2016.1157594

Viljoen, J. L., Klaver, J., & Roesch, R. (2005). Legal Decisions of Preadolescent and Adolescent Defendants: Predictors of Confessions, Pleas, Communication with Attorneys, and Appeals. Law and Human Behavior, 29 (3), 253-277.

Vrij, A., & Granhag, P. A. (2012). Eliciting cues to deception and truth: What matters are the questions asked. Journal of Applied Research in Memory and Cognition, 1(2), 110-117. DOI: 10.1016/j.jarmac.2012.02.004

Wachi, T., Watanabe, K., Yokota, K., Otsuka, Y., Kuraishi, H., & Lamb, M. (2014). Police interviewing styles and confessions in Japan. Psychology, Crime & Law, 20 (7), 673-694. DOI: 10.1080/1068316X.2013.854791

Walsh, D., & Bull, R. (2015). Interviewing suspects: examining the association between skills, questioning, evidence disclosure, and interview outcomes. Psychology, Crime & Law, 21 (7), 661-680. DOI: 10.1080/1068316X.2015.1028544

Walsh, D., & Bull, R. (2012). How Do Interviewers Attempt to Overcome Suspects' Denials? Psychiatry, Psychology and Law, 19 (2), 151-168. DOI: 10.1080/13218719.2010.543756

Westera, N. J., & Kebbell, M. R. (2014). Investigative interviewing in suspected sex offences. In Investigative interviewing (pp. 1-18). Springer.

Yang, Y., Madon, S., & Guyll, M. (2015). Short-Sighted Confession Decisions: The Role of Uncertain and Delayed Consequences. Law and Human Behavior, 39 (1), 44–52. DOI: 10.1037/lhb0000100

Zeng, F, Huang, C Y and Bull, R (2021). Police interview of suspects in China: Developments and analyses. International Journal of Police Science and Management, 23 (1), 29–41. DOI: 10.1177/1461355720947548.

Appendix A - Behaviors by category

Table II - Interviewer behaviors by category

Table III - Suspect behaviors by category

Appendix B – Supplemental logistic regression

Table IV- Summary of logistic regression analysis for variables predicting disclosure with and without time interaction effect