The Journal of Credibility Assessment and Witness Psychology

2003, Vol. 4, No. 1, pp. 1-36

Published by the Department of Psychology of Boise State University

 

 

WHEN DID YOU CONCLUDE SHE WAS LYING? THE IMPACT OF THE MOMENT THE DECISION ABOUT THE SENDER’S VERACITY IS MADE AND THE SENDER’S FACIAL APPEARANCE ON POLICE OFFICERS’ CREDIBILITY JUDGMENTS.

Jaume Masip, Eugenio Garrido, and Carmen Herrero

Department of Social Psychology and Anthropology,

University of Salamanca, Spain

Correspondence concerning this article should be sent to Jaume Masip (jmasip@usal.es) or Eugenio Garrido (garrido@usal.es), Department of Social Psychology and Anthropology, University of Salamanca, Facultad de Psicología, Avda. de la Merced, 109-131, 37005 Salamanca (Spain).

The research reported here was supported by the Junta de Castilla y León, Programa de Apoyo a Proyectos de Investigación, Ref. 30/98.

Copyright 2003 by the Department of Psychology of Boise State University and the Authors. Permission for non-profit electronic dissemination of this article is granted. Reproduction in hardcopy/print format for educational purposes or by non-profit organizations such as libraries and schools is permitted. For all other uses of this article, prior advance written permission is required. Send inquiries by hardcopy to: Charles R. Honts, Ph. D., Editor, The Journal of Credibility Assessment and Witness Psychology, Department of Psychology, Boise State University, 1910 University Drive, Boise, Idaho 83725, USA.


 ABSTRACT: Two experiments were conducted to explore how the moment observers make their decision about the senders’ veracity affects their judgment and detection accuracy. In Experiment 1 police officers and undergraduates judged the credibility of video-recorded statements. Contrary to our expectation, officers did not judge the statements earlier than the students. An initial lie bias became evident. In Experiment 2 a still face, which could be of the same witness as in Experiment 1, or of two other witnesses, was shown to officers as they listened to truthful or deceptive accounts taken from Experiment 1. There was no effect of the sender’s facial appearance on the lie bias found in the first experiment, which emerged here as well. Accuracy for detecting deceptive accounts decreased across time in both studies, while accuracy for truthful accounts increased only in Experiment 2. How visual and verbal information contributed to these effects is discussed. 

WHEN DID YOU CONCLUDE SHE WAS LYING? THE IMPACT OF THE MOMENT THE DECISION ABOUT THE SENDER’S VERACITY IS MADE AND THE SENDER’S FACIAL APPEARANCE ON POLICE OFFICERS’ CREDIBILITY JUDGMENTS

Introduction

DePaulo and Rosenthal (1979) identified three main areas of inquiry in the field of nonverbal detection of deception: (a) people’s ability to lie successfully and to accurately detect deception, (b) channel or modality effects on accuracy, i.e., what kind of information (visual, vocal, verbal, transmitted by the face, transmitted by the body, etc.) is most useful for untrained observers to detect deception, and (c) the study of the behavioral indicators of deception (real deception cues, perceived deception cues, and behaviors believed by people to be useful to detect deception). Within the first area pointed out by DePaulo and Rosenthal (1979), attention has been paid to sender and/or receiver variables that may affect their ability to deceive or detect deception (variables such as gender, age, experience, personality traits, etc.), as well as to certain situational variables such as motivation to lie successfully, familiarity between sender and receiver, time to create a deceptive story, perceived consequences of being detected, etc. (see reviews by DePaulo, DePaulo, Tang, & Swaim, 1989; DePaulo, Stone, & Lassiter, 1985; Ekman, 1992; Ekman & O’Sullivan, 1989; Ford, 1996; Kalbfleisch, 1992; Köhnken, 1989; Kraut, 1980; Masip & Garrido, 2000, 2001a; Miller & Burgoon, 1982; Miller & Stiff, 1992, 1993; Vrij, 1998, 2000; Zuckerman, DePaulo, & Rosenthal, 1981; Zuckerman & Driver, 1985). In general, meta-analyses on the results obtained from this approach show that detection accuracy (i.e., accuracy at detecting both truths and lies, Miller & Stiff, 1993) by untrained detectors usually falls between 45 % and 60 % correct classifications, where 50 % is the chance level. In addition, it has been found that police officers are no more accurate than lay people in their credibility judgments (e.g., DePaulo & Pfeiffer, 1986; Ekman & O’Sullivan, 1991; Garrido, Masip, Herrero, & Tabernero, 1997; Garrido, Masip, & Herrero, 2003; Henderson & Hess, 1982; Köhnken, 1987; Kraut & Poe, 1980; Sanderson, 1978, cited by Bull, 1989; Vrij, 1992; Vrij & Graham, 1997; see reviews by Bull, 1989, Garrido & Masip, 1999, and Vrij, 2000). Instead, there is some evidence that police officers may even be less precise than non-officers, due to a lie bias they may display when making their judgments (Garrido et al., 1997; Sanderson, 1978, cited by Bull, 1989).

In view of that poor accuracy level among observers trying to discern whether someone is lying or telling the truth, Miller and Stiff (1993) suggested that that issue should be considered from an alternative perspective: instead of investigating detection accuracy, researchers should identify explanations for observers’ errors in their credibility judgments. In line with that suggestion, in the two experiments reported here we try to identify some factors that may have an effect upon observers’ accuracy at judging credibility; specifically we are trying to discern what processes underlie the poor performance attained by police officers in Garrido et al.’s (1997, 2003; see also Masip, 2002) study. In that experiment, officers’ detection accuracy did not differ significantly from chance level, while students’ accuracy was significantly above chance.  The poor accuracy among officers was due to their tendency to judge all statements as false. Officers’ accuracy at judging deceptive statements was as high as that of undergraduates, while their accuracy at judging honest statements was poorer. In fact, the tendency of police officers to judge statements as deceptive was the same regardless of the real quality (truthful or deceptive) of those statements, while students were somewhat more sensitive to the real truth value of the stories.

In an attempt to have a closer look at that lie bias among officers, Garrido and Masip (2001) explored whether it was due to a reduced capacity among them to perceive a general expressive pattern as defined by Becerra, Sánchez, and Carrera (1989). These authors suggested that the accurate detection of deceit is based on observers’ perception of a general expressive pattern in the sender’s behavior, a pattern that changes as the statement quality (value of truth) varies. If so, it could be the case that officers did not perceive that pattern. There may be several reasons for that. For instance, police officers may have a stronger Generalized Communicative Suspicion (GCS) than non-officers. Levine and McCornack (1991) differentiated between GCS and situationally-aroused suspicion or “state” suspicion. The former would be a “predisposition toward believing that the messages produced by others are deceptive” (Levine & McCornack, 1991, p. 328), and is described as a relatively enduring and cross situational cognitive construct. On the other hand, situationally-aroused or state suspicion is prompted by certain contextual cues. It was defined by Levine and McCornack (1991) as “a belief that communication within a specific setting and at a particular time may be deceptive” (p. 328). Unlike GCS, state suspicion is transitory and is based upon certain situational variables.

Detecting deceit is an important task for police officers. During their daily work, they are often involved in social interactions where mistrust and lack of confidence are normal, and where they must question the interviewee’s assertions. That is to say, situations where a state suspicion is aroused. Yet, this suspicion, given its frequency in police work, could become chronic, arousing among officers a belief that the interviewee is probably not being truthful. This process would end up generating a kind of suspicion that would no longer be a response to contextual cues nor would it be transitory anymore. Rather, that suspicion would be a GCS. Research has shown that high GCS ratings are associated with a tendency to make judgments of deceptiveness (Levine & McCornack, 1991). With regard to our police officers, it could be the case that their generalized suspicion prevented them from scrutinizing the witness’s behavioral displays, thus not being able to perceive his or her general expressive pattern. If this were actually the case, then perhaps officers made only a biased “guess” based on their initial suspicion. Alternatively, police officers’ generalized suspicion may have given rise to a confirmation bias, making them attentive to only those behaviors supporting their view that the sender was lying. In either case, police officers would be unable to perceive the general pattern described by Becerra et al. (1989). However, Garrido and Masip’s (2001) results showed that not only officers, but also non-officers, were unable to perceive any general expressive pattern in the sender’s behavior. Thus, that factor cannot account for the differences between the police and lay people.

In this paper we describe some further explorations of the processes underlying officers’ lie bias in Garrido et al.’s (1997, 2003) study. Experiment one looks at whether police officers and students came to their conclusion about the sender’s veracity at different times, and whether this can account for the judgmental differences between these groups which were detected by Garrido and his colleagues. Also, an interesting question is how the moment observers come to a conclusion about whether the sender is lying or telling the truth affects detection accuracy, that is, is there any point in time where accuracy is higher? Experiment 2 is a follow-up study to answer some questions raised by the results obtained in Experiment 1. Thus, the contribution of the sender’s facial appearance and that of her dynamic nonverbal behavior to the profile found in Experiment 1 is explored.

EXPERIMENT 1

The availability of information (both useful and misleading behavioral cues) depends upon the moment observers make their decision about the truthfulness of the senders’ account. If receivers decide at the very beginning of a sender’s performance, the amount of available verbal and nonverbal information from that sender will be very limited. Conversely, if observers decide after the sender’s performance has concluded, they will be able to take into account all the verbal and nonverbal behavior displayed by that sender throughout his or her performance. Thus, if information gathered by observers paying attention to the senders’ behavior is used as a basis for making veracity judgments, accuracy will probably be influenced by the moment observers conclude that the sender is lying or telling the truth, at the beginning, middle, or end of his or her performance.

A first interesting question is whether police officers and non-officers (undergraduate psychology students) tend to decide at different moments in time. Such a difference might  account for the differences between those groups found by Garrido et al. (1997, 2003). Our prediction concerning the moment variable is based on the contributions of Levine and McCornack when conceptualizing their GCS, as well as on the work of Stiff, Kim and Ramesh (1992). Thus, it could be the case, for instance, that officers have a strong generalized communication suspicion as mentioned above, so that they enter the situation with the a priori belief that the sender is lying, while lay observers are more attentive to the sender’s behavioral displays. In that case, officers would tend to decide quickly at the beginning of the statement, because they “would be certain of it” and would see no need to pay attention to the witness’s behavior, while students would tend to decide later in the sender’s performance, after paying close attention to that witness’s behavior and after having processed the information so gathered. Also, Stiff et al.’s (1992) paper permits drawing an alternative process, which would lead to the same prediction. Those authors justify the development and existence of a cognitive heuristic which would lead relational partners –among which mutual confidence and trust are the norm, as well as necessary to maintain the relationship– to judge the other member’s performance as truthful, without even processing the information conveyed by that other member which could potentially be relevant to judge his or her credibility. Probably the same rationale could be used to account for officers’ judgmental tendencies, but in the opposite: instead of those cooperative interactions characterized by relational intimacy and trust that relational partners are involved in, police officers often get into interactions where distrust and suspicion are usual. This could create among officers a belief that the interviewee is not being truthful, in the same way that a belief that the other person is being truthful is aroused among relational partners. And, in the same way that a potential lie detector involved in a close relationship bases his or her credibility judgments on the a priori belief that his or her partner is honest, thus making heuristic judgments of truthfulness without even processing the incoming information, police officers could do something similar to conclude that the witness is being deceptive. On the other hand, lay observers judging the credibility of strangers’ statements would use a rather different strategy. They would be less biased than officers concerning the sender’s honesty, they would be less confident than officers in their skills to assess other people’s credibility (Garrido et al., 2003), and, therefore, they would be more willing than officers to attend to and to take into account the behaviors displayed by the witness during his or her statement.

Both processes, the one based on a GCS among officers and the one derived from Stiff et al.’s (1992) findings concerning lie detection amongst relational partners, suggest that police officers will display a lie bias such as that found by Garrido et al. (1997, 2003) while lay observers will be able to take into account information drawn from the sender’s behavior to make their judgment, which will make them more accurate at assessing credibility. Thus, our first hypothesis predicts that police officers will make their decision about the sender’s veracity earlier than non-officers, because, unlike these, officers will tend not to pay attention to the incoming information which would help them make an accurate judgment.

An important moderating factor on the effects of the moment observers decide on judgmental accuracy may be the value of truth of the statement to be judged. For instance, it could be the case that, for truthful accounts, deciding at the end is beneficial, given the greater amount of accurate information available at that later point. However, the prediction is different when it comes to assessing deceptive statements. Liars may monitor their behavior in order to give a plausible false account (information management) as well as an honest impression (image management and behavior management) (see Buller & Burgoon, 1996, 1998; DePaulo, 1991, 1992; DePaulo & Kirkendol, 1989; Greene, O’Hair, Cody, & Yen, 1985; Masip & Garrido, 1999, 2000, 2001a; Vrij, 1998, 2000; Zuckerman et al., 1981). Therefore, the later observers come to a conclusion about the deceiver’s truthfulness, the more misleading cues that deceiver will have had a chance to display. Not all of his or her cues will be misleading, since some behaviors are hardly controllable (e.g., Ekman, 1992; Ekman & Friesen, 1969, 1974), but in any case, later in the sender’s performance, the amount of misleading information will be greater in false accounts than in honest ones, while the amount of truthful information will be relatively smaller. However, earlier in the account these differences will be less pronounced. Therefore, an interaction between the moment observers decide and the value of truth of the statements could be expected. Thus, our second hypothesis predicts that, as observers decide later in time, there will be a relative increase in accuracy at detecting truthful accounts and a relative decrease in accuracy at detecting deceptive accounts.

It is important to stress that this study was designed to test hypothesis one. Since only one sender was used, either supportive or non-confirmatory evidence for hypothesis two must be taken only as preliminary and suggestive evidence until replications with a large number of senders be conducted.

 

Method

Participants: The sender was a female undergraduate student of psychology at a Spanish University. Observers were 121 police officers studying to become police inspectors at the Police Academy of Ávila (Spain), and 147 undergraduate students of psychology at a Spanish University[1].

Procedure: In order to increase the ecological validity of this study, we addressed some of the concerns expressed by various authors in this area (e.g., Köhnken, 1987, 1989; Miller & Stiff, 1993) by (a) motivating our senders to be convincing, (b) making the content of the statements relevant to police interrogation settings: the topic was the reporting of criminal actions (factual descriptions), (c) by giving senders a few minutes to prepare before giving their statements, (d) by having observers make a dichotomous decision (“true” or “false”) instead of rating the degree of truthfulness or deceptiveness, and (e) by showing observers only two statements of some length (no less than two minutes). Normally, in laboratory research on nonverbal detection of deception a large number of small behavioral samples are shown to observers. However, in the real world officers rarely have to judge the credibility of dozens of statements that are only a few seconds long. We addressed this issue by showing observers only two statements of some length, although this prevented us from using a large sample of senders.

In order to motivate our sender, we offered all psychology students at our University who were taking a social psychology module a substantial academic reward if they participated as witnesses in a lie detection study and were the most convincing of all senders. Four undergraduate females volunteered. Each of them was shown two film sequences depicting criminal actions (S1 and S2). After watching each of these sequences, senders were instructed to work out a deceptive version (D) and a truthful one (T) of the sequence. They were left ten minutes to create each version, and were video recorded as they made their statements -a free narrative account no less than two minutes long. Thus, each sender produced four statements: a deceptive account of the first sequence (S1D), a truthful account of that same sequence (S1T), a deceptive account of the second sequence (S2D) and a truthful account of that second sequence (S2T). A pilot study was conducted with a few undergraduates in order to choose the most convincing liar for the main study[2]. All four candidates received the advertised reward for their participation.

The four performances of the sender who was chosen were edited and shown to 121 police officers and 146 psychology students. Each participant watched two statements: one based on S1 and the other based on S2. These statements could be both truthful (31 police officers and 38 undergraduates were allocated to this condition), both deceptive (29 officers and 40 undergraduates), truthful the first to be shown and deceptive the second to be shown (31 officers and 36 students) or deceptive the first and truthful the second (30 officers and 32 students). All police officers allocated to the same experimental condition were in the same class in the police academy at the moment the experimental session was carried out; allocation of officers to their classes is based on an alphabetical criterion. Allocation of undergraduate students to the experimental conditions was made randomly. Since the number of officers per classroom was not the same across all classrooms, and some students failed to attend their sessions and/or came to a session different from the one they had been assigned to, there were some small variations in size across the experimental groups.

After watching each of the two performances of the sender, observers were given a few minutes to complete a questionnaire. One of the items asked them whether they thought the sender had lied or told the truth. Another item asked observers whether they had reached their conclusion early, as they started to see the sender’s performance (Moment 1), at the middle part of that performance (Moment 2), or at the final moment (Moment 3).

Results

Hypothesis Testing

Data were analyzed separately for S1 and S2. Two stepwise backward hierarchical loglinear analyses were performed using SPSS 9.0. The variables introduced were value of truth of the statement (truthful / deceptive), observers’ occupation (police officer / undergraduate), the hit / miss variable, and the moment observers made their decision (Moment 1 or 2 / Moment 3)[3]. Both a significant association among occupation and moment –as predicted in hypothesis 1– and a value of truth X moment X hit / miss –in the way predicted in Hypothesis 2– would be expected to emerge for both statements. In addition, concerning the first three variables, we expected to find results similar to those reported by Garrido et al. (2003; see also Masip, 2002), which were based on these data.

Concerning S1, k-way effect tests showed the fourth-order interaction was of no relevance, likelihood-ratio chi-square: c2 (1) = 0.04, p = .846, but there were substantial third-, second-, and first-order effects, respectively: c2 (4) = 18.05, p = .001; c2 (6) = 69.54, p = .000; c2 (4) = 9.68, p = .046. Something similar was found for S2, respectively: c2 (1) = 1.59, p = .220; c2 (4) = 22.80, p = .000; c2 (6) = 22.76, p = .001; and c2 (4) = 28.83, p = .000. The best model for S1 comprised three interactions: Occupation X Value of Truth X Hit/Miss (police officers made more errors when judging truthful statements than when judging the deceptive; this effect was presented and discussed by Garrido et al., 2003), Value of Truth X Moment X Hit/Miss, and Occupation X Moment (these interactions are discussed briefly). This model had an excellent goodness of fit: its likelihood-ratio chi-square was c2 (3) = 0.24, p = .971, and the greatest standardized residual had an absolute value of 0.29. The best model for S2 was somewhat simpler, comprising only the two third-order interactions also included in the S1 model, that is, Occupation X Value of Truth X Hit/Miss and Value of Truth X Moment X Hit/Miss. This model had a likelihood-ratio chi-square of c2 (4) = 4.93, p = .294, and the greatest standardized residual had an absolute value of 0.85.

In order to examine the specific contribution of each effect to the fit of the model, attention was paid to partial association tests and parameter estimates. In Table 1 this information is

 

TABLE 1. Partial Association Tests And Parameter Estimates (In Absolute Values) Of The Effects That Had A Relevant Contribution To The Fit Of The Model, Either In S1 Or In S2.

Effects

Partial association

Parameter estimates*

S1

S2

S1

S2

c2 (1)

p

c2 (1)

p

|l|

|z|

|l|

|z|

Third-order effects

 

 

 

 

 

 

 

 

Occupation X Value of Truth X Hit/Miss

8.13

.004

7.04

.008

.20

2.66

.17

2.44

Value of Truth X Moment X Hit/Miss

11.40

.001

12.65

.000

.24

3.12

.24

3.58

Second-order effects

 

 

 

 

 

 

 

 

Occupation X Moment

3.24

.072

0.45

.832

.15

1.92

.02

0.25

Value of Truth X Moment

1.39

.239

12.18

.001

.03

0.42

.19

2.82

Value of Truth X Hit/Miss

49.63

.000

3.87

.049

.47

6.20

.10

1.47

Moment X Hit/Miss

12.25

.001

0.99

.321

.21

2.80

.09

1.32

First-order effects

 

 

 

 

 

 

 

 

Moment

4.09

.043

9.88

.002

.04

.55

.13

1.96

Hit/Miss

3.16

.076

16.11

.000

.16

2.18

.19

2.79

* In absolute values. To examine the direction of effects see Appendix 1.

summarized for all those effects which either approached significance or were significant, whether in S1 or in S2. The direction of effects can be observed in Appendix 1, where the presentation model suggested by Tabachnick and Fidell (1996) was used.

Concerning the occupation, value of truth, and hit / miss variables, Garrido et al.’s (2003) results were here replicated with some minor nuances. In particular, the most relevant effect (the three-way interaction) was found again and, furthermore, it did not interact with the moment observers decided their judgment. Thus, the introduction of that variable in the analyses did not substantially alter the former results. Since they were already discussed elsewhere (Garrido et al., 2003) and are not the main focus of the present report, they will not be discussed here again.

Instead, our focus in the present paper centers on those effects involving the moment variable. A certain tendency was found in S1 to make the decisions at moments 1 and 2 (56.18 % of judgments) instead of making them at Moment 3 (43.82 %). Something similar, although the trend was clearer, happened in S2 (percentages were, respectively, 59.19 % and 40.82 %). This effect could be due to having added the number of decisions made at Moment 1 to those made at Moment 2, since in both statements the latter had a frequency that was quite similar to that of Moment 3 decisions.

The first hypothesis predicted that police officers would hurry and make their judgment earlier than the undergraduates. Therefore, an association between being an officer and deciding at moments 1 and 2, and between being a student and deciding at Moment 3 would be expected. However, the Occupation X Moment association was not significant in S2. In S1 it did not reach statistical significance either, as indicated by the two measures used to explore the individual effects (although it was close to significance: c2 (1) = 3.24, p = .072; z = -1.92), but the program retained the effect while searching for the best model during the stepwise procedure (the associated change likelihood-ratio chi-square was c2 (1) = 5.48, p = .019). In any case, the effect was the opposite to what was expected: police officers did not tend to make their decisions earlier than the undergraduate students, but later (see Appendix 1): While in S1 48.76 % of police officers decided at moments 1 and 2 in comparison with the remaining 51.24 %, who decided at Moment 3, 62.33 % of undergraduates decided during the early moments and only 37.67 % of them did so at Moment 3. In S2 the associations failed to reach significance, but they pointed in the same direction. In summary, our first prediction did not receive empirical support. If there was any occupational group which acted hastily in making their credibility judgments it was not the officers, but rather the undergraduate students.

The second hypothesis predicted an interaction between the value of truth of the statement, the decision-making moment, and the correctness of the credibility judgment (hit or miss), in the sense that deceptive statements would be more accurately detected earlier in the statement than later on, while truthful statements would be judged with higher accuracy at the final moment rather than at the beginning of the statement. To begin with, it should be mentioned that some second-order effects were substantial. In S2, the interaction Moment X Value of Truth indicates that, when judging the false account (S2D), the decision was made basically at the beginning (69.85 % of cases); this did not happen when judging S2T (48.09 %). In S1 it was the Moment X Hit/Miss interaction that was relevant: the decisions made at the beginning of the statement were accurate more often than those made at the final moment. But both of these effects were qualified by the higher-order value of Truth X Moment X Hit/Miss interaction, which lent support to our second hypothesis. When judging the deceptive statements an association was found both in S1 and S2 between making the decision early (moments 1 and 2) and guessing right, as well as between deciding at the final moment and judging wrongly. An opposite tendency became apparent when judging the truthful statements (see Appendix 1). As stated before, this effect was one of the components of the final model in both analyses: the one concerning S1 and the one concerning S2.

Although this interaction was significant it would be interesting to analyze whether, in an absolute sense, there was a significant decrease in accuracy when judging deceptive statements at Moment 3 in comparison with the early moments, as well as whether the increase in accuracy for the truthful statements was significant too. In order to examine these effects, individual chi-square analyses were performed to examine the associations among the hit/miss and the moment variables separately for truthful and deceptive accounts. The results of these analyses are summarized in Table 2. It is apparent that the predicted decrease for the deceptive statements was significant. However, the increase in accuracy for truthful statements across time was not found, although a marginally significant trend in the predicted direction was apparent in S2T.

TABLE 2.  Moment X Value Of Truth X Hit/Miss Contingency Tables, And Chi-Square Analyses For Truthful And Deceptive Statements.

Statements

Moment

c2 (1)

p

Moments 1 and 2

Moment 3

Sequence 1 (S1)

 

 

 

 

   Deceptive (S1D)

 

 

 

 

      Hit

62 ( 1.8)

24 (-2.1)

22.53

.000

      Miss

13 (-2.5)

32 ( 2.9)

 

 

   Truthful (S1T)