Portal de Periódicos da CAPES

Confessing their Crime: Factors Influencing the Offender’s Decision to Confess to the Police

2009; Taylor & Francis; Volume: 28; Issue: 1 Linguagem: Inglês

10.1080/07418820903218966

1745-9109

Nadine Deslauriers‐Varin, Patrick Lussier, Michel St‐Yves,

Policing Practices and Perceptions

Abstract Confessions are crucial to successful police investigations but scholars have significantly overlooked factors that contribute to an offender's decision to confess a crime. This study aims to examine a large array of factors that play a role in the offender's decision to confess a crime to the police and potential interaction effect among them. A total of 221 adult males incarcerated in a federal Canadian penitentiary were recruited. Correctional files, police reports, and offenders' self‐reported data were collected and analyzed. Controlling for sociodemographic, criminological, and contextual factors, a series of logistic regression analyses were conducted. Findings highlighted the predominant role of police evidence over and above other factors analyzed. Furthermore, sociodemographic and criminological factors played a more important role in the offender's decision to confess when police evidence was weak. Findings are discussed in light of the current scientific literature on the determinants of offenders' decision to confess their crime. Keywords: police interrogationconfessionpolice evidenceadult offenderspolice investigation Acknowledgments The authors wish to thank the Centre International de Criminologie Comparée, the School of Criminology of the University of Montreal, and the Social Sciences and Humanities Research Council of Canada for their financial support. The authors also wish to thank Dr. Tewksbury and the anonymous reviewers for their comments on an earlier version of this manuscript. A preliminary version of this study was presented in 2008 at the Third International Investigative Interviewing Conference in Nicolet, Quebec, Canada. Notes 1. In the UK, the law permits no ruses or persuasive techniques to be used during the interrogation. At the beginning of the interview, it is also forbidden to discuss anything other than the index crime. In the USA, as in Canada, courts allow the use of such persuasive techniques. In both the USA and the UK, however, the offender's confession has to be obtained voluntarily, meaning that the self‐incriminating statement has not been obtained from the suspect by fear of prejudice or hope of advantage, exercised or held out by a person in authority, or by oppression (Gudjonsson, Citation2003). In spite of this condition of admissibility, it is important to mention that the UK legal system does not recognize the right to remain silent, as it is the case in the USA and Canada. Hence, the caution made to suspects in the UK informs them that they do not have to say anything but it may, however, harm their defense if they do not mention when questioned something they later rely on in court (Home Office, Citation1995). This means that the judge and jury can draw adverse inferences of guilt against a suspect who decides to remain silent or fail to provide relevant facts during police questioning (Gudjonsson, Citation2003), putting pressure on the suspects to self‐incriminate themselves and collaborate with the police. 2. On one hand, the confession rate is higher among cases that go to court than among those that do not go to court (Phillips & Brown, Citation1998). On the other hand, studies have also shown that confession rates vary according to crime type. For example, one study showed that the confession rate decreased as the seriousness of the crime increased: less serious offenses (e.g., prostitution, shoplifting) revealed a confession rate of 72%, whereas moderately serious offenses (e.g., burglary, arson) revealed a confession rate of 49%, and very serious offenses (e.g., homicide, rape) revealed a confession rate of 46% (Phillips & Brown, Citation1998). 3. A French translation (Blais, Lachance, & Richer, Citation1991) of the short version (20 items, true or false) of the Marlowe‐Crowne questionnaire (Crowne & Marlowe, Citation1960) was used (Strahan & Gerbasi, Citation1972). The questionnaire measures the level of social desirability or the tendency to act (consciously or not) according to the expectations of others. The score on the questionnaire was used to control for the possibility that offenders might be biased in responding to the questionnaire by trying to protect their self‐esteem and/or their self‐image. The questionnaire was administered to the first 126 research participants (54.8%). The scale was not administered to the full sample due to logistic concerns (i.e., not to interfere with the offenders/correctional activities and schedule). The statistical analyses presented here were conducted with this subsample. The mean score on the scale was 12.2 (s = 3.8; range 3–20). Among the correlations between scores on the social desirability scale and the independent variables, two were statistically significant (16.7%): (1) age (r = .29, p < .01); and (2) marital status (r = .23, p < .05). Therefore, older offenders, in a relationship, seemed to be more inclined to respond in a more socially desirable way to the survey. The correlation between score on the social desirability scale was unrelated to parental status, ethnicity, level of education, feeling guilty, crime seriousness, length of prison sentence, perception of police evidence, and use of legal advice. Importantly, no statistically significant association was found between the decision to confess and social desirability. 4. The sociodemographic variables in the group of deleted cases were compared to those cases included in the study to make sure the deleted questionnaires were not reflecting special characteristic of the offenders (e.g., education, reading and writing disabilities, age, etc.). None of the comparisons yielded significant differences at p < .05. 5. In Canada, as in the USA, prior to commencing interrogation police must inform custodial suspects of their Constitutional rights to silence and appointed counsel, and that anything they say can be used against them (St‐Yves & Deslauriers‐Varin, in press). These rights have to be actively, voluntarily, and knowingly waived by the accused in order for their statements to be admissible at trial (Kassin et al., Citation2007). The caution is made to any person in custody, arrested or not, and interrogated for a crime he or she is likely to be implicated in. Contrary to the practice in the USA and the UK, Canadian laws do not recognize the moral necessity for the attorney to be present in the interrogation room; it only recognizes the right to inform the suspect of his or her right to remain silent and to contact an attorney without delay, usually by telephone. However, where the interviewee is an adolescent (aged 12–18 years), the Youth Criminal Justice Act provides for the presence of counsel during interrogation. With adults, police retain a discretionary power to allow the attorney to be present during the interrogation. In such a case, the attorney would become a witness and could be summoned as such if the statement is contested and a voir dire becomes necessary. Police officers will generally insist that the suspect contacts an attorney and will make sure that this strictly confidential conversation is entirely satisfactory to the suspect. 6. The ROC (Receiver Operating Characteristic) curve allows examining simultaneously the sensitivity (i.e., the ability to identify true positives, or those predicted as having confessed and who did confess), and the specificity (i.e., ability to identify true negatives, or those predicted as non‐confessors and who did not confess). ROC curves are typically used by risk assessors to examine the predictive accuracy of risk assessment tools and the ability of the instrument to distinguish recidivists from non‐recidivists. The same logic can be applied here by using ROC curves to assess the ability of the predictive models in distinguishing confessors from non‐confessors. The ROC curve can be interpreted by looking at the area under the ROC‐curve (AUC), and the relative improvement of the prediction model over chance (i.e., 50%, or .50). The AUC varies between 0 and 1.0, with values closer to one indicating perfect prediction of both true positives and true negatives. For example, an AUC value of .70 for a logistic regression model indicates that, 70% of the time, a randomly selected confessor would have a higher predicted probability of occurrence of the event (i.e., making a confession) than a randomly selected non‐confessor. 7. Lipsey's & Wilson's (Citation2001) algorithms were used to compute Cohen's d effect sizes. According to Cohen (Citation1988), in behavioral research, a d coefficient of about .20 represents a small effect (equivalent to a correlation of about .10), .50 represents a medium effect (equivalent to a correlation of about .25), whereas .80 represents a large effect size (equivalent to a correlation of about .40). 8. Recall that outliers deleted from the analyses may artificially increase predictive accuracy of the model. The deletion, however, was necessary as outliers may create important biases in the parameter estimates (Tabachnick & Fidell, Citation2007). In this case, outliers had a minimal but significant impact on the overall classification accuracy of the predictive model. With the outliers included, the overall classification accuracy was 76.6% while the correct classification of confessors and non‐confessors were, respectively, 46.7% and 90.6%. Careful examination of the deleted cases (n = 7) showed that the majority of those offenders were Caucasian (n = 6), were multi‐recidivist (n = 6), had confessed their crime (n = 6), and had received the minimal federal sentence of two years (n = 4), many of which were for a drug related offenses (n = 3). For unknown reasons, the predicted model did not work well for this very small subgroup of offenders. A different dynamic might be at work that was not captured by the model tested. The subgroup, however, was too small to produce firm conclusions. 9. In order to test for a more parsimonious model, only the statistically significant predictors in Model III (marital status, index of crime seriousness, prior conviction, use of legal advice, guilt) were retained into a final model (model not shown). This allowed us to reduce the number of predictors and increase the statistical power of the analysis. The logistic regression model was statistically significant [χ 2 (6) = 43.5, p < .001], with a pseudo‐explained variance of 39.4% (Cox & Snell). This parsimonious model correctly classified 93.7% of the non‐confessors and 70.8% of the confessors. Confessors were statistically more likely to have committed a more serious crime (OR = 1.58; 95% CI = 1.14–2.19) and have feelings of guilt (OR = 5.82; 95% CI = 2.19–15.50), but were less likely to be multi‐recidivists (OR = .15; 95% CI = .03–.77) and to have used legal advice (OR = .02; 95% CI = .00–.18). For this model, marital status and being a recidivist were no longer predicting confession at p < .05. Hence, crime seriousness, feeling guilty, being a multi‐recidivist, and using legal advice were the most significant factors in explaining the decision to confess or not when the offender perceived police evidence as weak or somewhat weak. 10. A parsimonious model (not shown) was tested with only the two significant predictors (multi‐recidivist and use of legal advice). The results were unchanged from those observed in Model III, which show much stability in the findings. Hence, when police evidence was perceived as either strong or somewhat strong, being a multi‐recidivist and using legal advice were the most significant factors associated with the offender's decision to confess or not.