Observers' decision moment in deception detection experiments: Its impact on judgment, accuracy, and confidence

Abstract

Research into the nonverbal detection of deception has typically been conducted by asking observers to judge whether a number of videotaped statements are truthful or deceptive. In most cases, the behavioural segments used in this research have been very short. A typical result is that observers tend to judge the statements as truthful (truth bias). In the present experiment, observers watched a series of video clips showing senders answering three questions about an event that they had witnessed. Observers had to indicate whether each sender's statement was truthful or deceptive, their judgmental confidence, and when they had made their decision about the sender's credibility: during his or her first, second, or third answer. Competing predictions were made about the influence of the decision moment on the observers' judgments and accuracy. The results replicated most research findings reported in the US and North‐European literature, including the truth bias phenomenon. However, the proportion of judgments of truthfulness decreased as observers decided later, particularly for the deceptive statements. This yielded an increase in accuracy in judging deceptive accounts. These results are consistent with the idea that initial credibility judgments are made heuristically, either because there is not enough information available or because observers are in the first, automatic stage of current attribution and person perception models. Heuristic decision making may produce a high proportion of judgments of truthfulness. Later judgments would be made in a systematic manner. The truth bias detected in deception research may be caused by researchers having used very brief and uninformative behavioural samples. The moment when observers made their decision had only a marginal negative influence on confidence.

The authors are grateful to the research assistants Gema Martín and Monica Sánchez, to Yago Reis and the actors Manolo Muñoz, Marta Calzada and José Romo, to the technician Javier Tamames, to José L. Vega, the Dean of the Faculty of Psychology of the University of Salamanca, for permitting us to use his office to record the tapes, and to two anonymous reviewers for their helpful comments on an earlier draft of the manuscript. This research was supported by the Junta de Castilla y León, Programa de Apoyo a Proyectos de Investigación, Ref. SA023/03.

La recherche sur la détection non verbale de la duperie a typiquement été conduite en demandant à des observateurs de juger dans quelle mesure un certain nombre d'affirmations filmées sont véridiques ou mensongères. Dans la plupart des cas, les segments comportementaux utilisés ont été très courts. Généralement, les résultats ont montré que les observateurs ont tendance à juger les affirmations comme étant véridiques (biais de vérité). Dans la présente expérimentation, les observateurs ont visionné une série de vidéos montrant des individus répondant à trois questions concernant un événement dont ils ont été témoins. Les observateurs devaient indiquer dans quelle mesure chaque affirmation des individus étaient véridique ou mensongère, leur niveau de confiance en leur jugement et à quel moment ils ont pris leur décision à propos de la crédibilité de l'individu: durant la première, la deuxième ou la troisième réponse. Des prédictions concurrentes ont été avancées concernant l'influence du moment de décision sur les jugements et l'exactitude des observateurs. Les résultats obtenus ont confirmé la plupart des résultats de recherche rapportés dans les écrits états‐uniens et nord européens, incluant le phénomène du biais de vérité. Cependant, la proportion de jugements de véracité diminuait alors que le temps de prise de décision augmentait, ceci étant particulièrement le cas pour les affirmations mensongères. Ceci a mené à une augmentation de l'exactitude dans le jugement des mensonges. Ces résultats sont en accord avec l'idée que les jugements initiaux de crédibilité sont le produit d'heuristiques, soit parce qu'il n'y a pas suffisamment d'informations disponibles ou parce que les observateurs se trouvent dans un premier stage automatique d'attribution du moment et de modèles de perception de la personne. La prise de décision heuristique peut produire une proportion élevée de jugements de véracité. Les derniers jugements peuvent être faits de manière systématique. Le biais de vérité détecté dans la recherche sur la duperie peut être causé par les chercheurs qui ont utilisé de très brefs échantillons comportementaux contenant des informations limitées. Le moment auquel les observateurs prennent leur décision a eu une influence seulement marginalement négative sur la confiance des observateurs.

Normalmente, la investigación sobre la detección no‐verbal del engaño se ha realizado pidiendo a unos observadores que juzguen si una serie de declaraciones en vídeo son verdaderas o falsas. En la mayoría de los casos, los segmentos conductuales empleados en esa investigación han sido muy cortos. Un resultado habitual es que los observadores tienden a juzgar las declaraciones como verdaderas (sesgo de veracidad). En el presente experimento, los observadores visualizaron una serie de fragmentos de vídeo en que aparecían unos emisores respondiendo a tres preguntas sobre un acontecimiento que habían presenciado. Los observadores tuvieron que indicar si la declaración de cada emisor era verdadera o falsa, su confianza en este juicio, y cuándo habían hecho su decisión sobre la credibilidad: durante su primera respuesta, la segunda, o la tercera. Se hicieron predicciones mutuamente excluyentes sobre la influencia del momento de decisión sobre los juicios y la precisión de los observadores. Los resultados replicaron la mayoría de los hallazgos de la investigación estadounidense y noreuropea, incluyendo el fenómeno del sesgo de veracidad. Sin embargo, la proporción de juicios de verdad decreció a medida que los observadores respondían más tarde, especialmente para las declaraciones falsas. Esto generó un aumento en la precisión obtenida al juzgar tales declaraciones. Estos resultados son consistentes con la idea de que los juicios de credibilidad iniciales se hacen heurísticamente, ya sea porque la información accesible es insuficiente o porque los observadores están en el primer paso, automático, de los modelos actuales de atribución y percepción de personas. La toma de decisiones heurística puede producir una alta proporción de juicios de verdad. Los juicios posteriores se harían de forma sistemática. El sesgo de veracidad detectado en la investigación del engaño puede deberse al empleo de muestras conductuales muy breves y poco informativas. El momento en que los observadores hicieron su decisión sólo tuvo una influencia marginal de signo negativo sobre la confianza.

Notes

The authors are grateful to the research assistants Gema Martín and Monica Sánchez, to Yago Reis and the actors Manolo Muñoz, Marta Calzada and José Romo, to the technician Javier Tamames, to José L. Vega, the Dean of the Faculty of Psychology of the University of Salamanca, for permitting us to use his office to record the tapes, and to two anonymous reviewers for their helpful comments on an earlier draft of the manuscript. This research was supported by the Junta de Castilla y León, Programa de Apoyo a Proyectos de Investigación, Ref. SA023/03.

1. It could be argued that if the sender chooses not to communicate, then more time does not result in an increase in information. However, as Watzlawick, Beavin, and Jackson (1967) stated, it is not possible not to communicate, because even silence and stillness are informative. A sender trying not to convey information might, for instance, take a long time before answering questions, speak slowly, pause often, provide few descriptive details, and inhibit his or her body movements. Latency period, speech rate, number and length of pauses, quantity of details, and inhibitions of movements might be actual, perceived, or stereotypical deception cues (e.g., Vrij, 2000). Therefore, they are often very informative.

2. This assumption might seem questionable, since extant research indicates that human observers' accuracy in judging the senders' credibility falls within the 45–60% range. However, as stated in the text, these conclusions are largely based on research using very brief behavioural samples, forcing observers to make their judgments heuristically and preventing them from using enough information to make a more accurate judgment. Even in the studies in which longer behavioural samples have been used, it is not clear whether the observers waited until the end of the senders' performance to make their judgment. The common assumption that observers' judgmental accuracy is poor may be an artifact based on how deception research has been conducted so far.

3. It is apparent that the interviewer was not blind to the veracity of the senders' statements. This could potentially have influenced the way he interacted with each sender. While this is a frequent risk in much deception research, a number of safeguards were taken in this experiment to prevent this from happen. The interviewer was more an experimenter giving instructions to the sender than a real interviewer. In other words, the substantial part of the interview was not a free interaction, but was tightly scripted, and the wording of the questions was always the same. In no case did the interviewer interrupt the interviewee to make questions or comments. In addition, before the actual data collection began, the interviewer rehearsed his performance while interviewing mock participants in the same room where the actual interviews were to be conducted. Several research assistants were present in these rehearsals and provided feedback as to the verbal and nonverbal behaviour of the interviewer, which had to be neutral.

4. In order to examine whether accuracy for later judgments increased, manipulation check analyses were conducted. There were two sets of senders: odd senders and even senders. Odd senders were in videos A1 and B1, whereas even senders were in videos A2 and B2. The order of senders in A1 was the same as in B1, but those senders who lied in A1 told the truth in B1, and vice‐versa (Table 1). The same can be said of A2 and B2. All videotapes contained 12 statements. Accuracy in judging the last 6 statements (which contained 3 truths and 3 lies) was compared with accuracy in judging the first 6 statements (also 3 truths and 3 lies) for both odd and even senders. These comparisons were not significant, t(26) = 1.40, p = .173, for odd senders, and t(26) = 1.68, p = .104 for even senders.

5. The experimental instructions were tested on three samples. The convenience of including all this additional information was assessed. It became apparent from the comments of the participants in the pilot tests (undergraduate students who received the instructions and performed part of the task) that, unless they received this information, they assumed that all senders had watched the same videotape and then tended to compare different statements in order to assess their veracity. Furthermore, it was certain that, except for a few cases, the observers did not make major changes when lying. Also, since the questions had been removed from the tapes, we thought that telling observers what the questions were about would help them to focus upon the experimental task—instead of, for instance, trying to guess the interviewer's question. The pilot test supported our views. Observers in the actual experimental task were also informed that the three questions had been counterbalanced.

6. In order to conduct Pearson correlations the distributions must be normal. The skewness and kurtosis of accuracy, the PJT, confidence, and the moment variable (for truthful statements, deceptive statements, and across all statements) were calculated using the SPSS “frequencies” command and were then divided by their standard error. Most of the resulting values were lower than 1, and all of them were lower than 2. Therefore, the distributions did not depart significantly from normality.

7. An exaggerated number of judgments of truthfulness were only apparent when deceptive accounts were judged. Although the PJT when judging the truthful accounts (.59) was greater than chance, it was far from 1. As for the deceptive statements, the .50 PJT might reflect, as an anonymous reviewer suggested, that participants relied on mere guesswork. Alternatively, since the moment–PJT correlation was significant, it can be assumed that at Moment 1 the PJT in judging deceptive statements was greater than the average .50 rate, whereas at Moment 3 it was lower. Therefore, a truth bias was presumably present when judgments of deceptive statements were made early. Unfortunately, the design of this study made it difficult to calculate the PJT separately at Moments 1, 2, and 3. Ongoing research is addressing this issue.