Validity and effectiveness of interrogation techniques: A meta-analytic review

ABSTRACT

The gathering of information through the use of interrogation techniques in the context of human intelligence (HUMINT) has a long and elusive history within applied settings of law enforcement and the military and civilian intelligence/counterterrorism community. However, psychological research has yet to catch up to systematically address pressing matters regarding the validity and effectiveness of common interrogation methods and a conceptual framework for relevant psychological factors. A promising, comprehensive contribution is the Taxonomy of Interrogation Methods (ToIM), which aims to integrate multiple approaches within the field of interrogation. In this paper, we utilized the ToIM model as a foundation for a meta-analytic review on the validity and effectiveness of interrogation techniques. We systematically integrated the existing evidence from 60 studies in order to determine which techniques from six domains of the ToIM produce valuable information. The results indicate that Rapport and Relationship Building, Presentation of Evidence and Cognitive Facilitation (an additional domain beyond the ToIM) are valid approaches to optimize both the amount of information gathered as well as its accuracy. The evidence is insufficient to conclude the effectiveness of techniques from the other four domains. Overall, the results are in line with the general notion in the field that a positive relationship with a suspect/source is the key to gather valuable information.

What is the public significance of this article?—This meta-study aggregates the results from existing studies on the efficacy of different interrogation techniques in order to elicit valuable information. The results indicate that building a positive relationship with a source, the tactical use of information, and offering cognitive support and aids for a source’s statement are valid approaches to optimize both the amount of information gathered and its accuracy.

Introduction

Human intelligence (HUMINT) is the process of gathering information from human sources in the context of military intelligence, espionage, counterintelligence, or counterterrorism and is one of the oldest intelligence disciplines. Yet, with few exceptions, increased public research in psychological aspects of HUMINT has only been conducted within the past two decades. This can be attributed to a growing need for effective methods in HUMINT since 9/11, primarily in the operations type of interrogations (Brandon, 2011), accompanied by increased publication of scientific literature (Brandon et al., 2019). Interrogation can be defined as “all manners of directed information gathering, including interviewing, investigation, or HUMINT collection” for both cooperative and uncooperative sources (Kelly et al., 2013, p. 166). Although its practical relevance ranges from military and intelligence contexts to the domain of law enforcement, it is pertinent to delineate HUMINT interrogations from criminal interrogations (Evans et al., 2010). HUMINT interrogations possess a present and prospective intent, for example, in obtaining intelligence about individuals and networks, as well as their capabilities, intentions, or plans. They are therefore not necessarily tied to single events such as investigating a crime or obtaining a confession (Hartwig et al., 2014).

Naturally, interrogation settings are complex psychological situations. Rather than conducting the questioning of sources as random processes, efforts have been made to influence the outcome of these settings by the use of interrogation techniques in order to gather information. In our study, we aim to synthesize research on the effectiveness of these efforts. We lay our emphasis on HUMINT interrogations effectiveness, but since the empirically based knowledge is limited, we conceptualize interrogation in a broader sense as all manners of directed information gathering (Kelly et al., 2013). Thus, we will deviate from the usual objective of interrogations in practice: we do not limit ourselves to interrogating persons in custody but examine the usability of research findings in HUMINT activities including debriefing, interviewing, and contact handling, as common types of source operations.

Prior research on interrogation techniques

Within the scientific literature, several concepts and research approaches can be found regarding the topic of HUMINT interrogations. In addition to individual specific interrogation techniques, comprehensive interrogation methods have been established, which are constituted by a partly standardized application of certain interrogation techniques or conceptual principles. Three of the most prominent HUMINT methods in the psychological literature are the Scharff Technique (Granhag et al., 2015), the Cognitive Interview (CI; Fisher & Geiselman, 1992) and the Strategic Use of Evidence (SUE; Granhag & Hartwig, 2015) Technique.

The Scharff Technique is based on stories about Hanns-Joachim Scharff, a German Air Force interrogator during World War II. Researchers have found that Scharff´s approach was characterized by four key techniques. These included (1) a friendly, rapport-based approach, (2) information gathering especially through free narrative without pressing for information, (3) working with a confirmation/disconfirmation-format instead of asking direct questions, and (4) creating the impression that one already has much or all information (Granhag et al., 2015). Additionally, Scharff relied on multicultural competence: he had a good understanding of British culture (and to a lesser extent American) and spoke polished English (Oleszkiewicz, 2016).

The CI is a method for interviewing victims, witnesses, and suspects alike. The method is based on concepts of social dynamics, takes the thought processes of the participants into account and aims to improve communication between them. At the beginning of the interview, techniques are used to establish rapport. During the conversation, different techniques are used to facilitate the recall of memories. The focus of the communication is to encourage and facilitate detailed statements. Sources´ information is first reproduced according to their own recollections. Memory gaps are clearly pointed out, making guesses and responses based on suggestion less likely. In the further course of the CI, sources are encouraged to narrate freely again, this time with the classification of memories into exemplary statements, putting themselves into situational framework conditions while the eyes are closed, telling backwards what they have experienced or taking the perspective of other persons. This is beneficial since the interrogator can use different techniques, e.g. due to time constraints, to evoke additional memories without introducing own information, to generate false memories or to fall back on suggestive questions.

Taking a different tack, the SUE technique is rooted in research on deception detection and aims at the elicitation of one or several cues to deception that may guide an interviewer on whether the suspect is being truthful. It is thought that a mixture of withholding, matching, and presenting evidence in various phases of the interrogation (SUE tactics) affects the suspect’s perception of the evidence and moderates the choice of counter-interrogation strategy, which is mirrored in the suspect’s verbal responses (Granhag & Hartwig, 2015). The idea is to withhold actual evidence for a longer period of time to evaluate if the free recall of witnesses has within-statement inconsistencies. The witness is then repeatedly confronted with these inconsistencies or the actual evidence in order to make the witness believe that the interviewers are well-informed. After that, the interviewers shift to a new topic – a topic for which the interviewer lacks critical information. It is thought that the witness now reveals more accurate information.

There have been recent efforts to evaluate which interrogation techniques are beneficial on a grander scale. A meta-analysis on the CI (Memon et al., 2010) revealed large effects in regard to the number of reported correct details in statements, but also a small increase of errors, however no effects in regard to confabulation. In another meta-analysis the Scharff Technique led to a substantial increase in new information (Luke, 2021). This also resulted in the sources largely overestimating the interviewers` knowledge and underestimating the amount of new information they provided. Another prominent example is a report by the High-Value Detainee Interrogation Group (HIG), which initiated and summarized a substantial body of research after 2016 (Brandon et al., 2019). The report concluded that it is possible to conduct non-coercive interrogations in order to gather information and provided concrete recommendations for practitioners in developing and maintaining rapport, active listening, providing assistance in recalling memories, using open-ended questions, being aware of cross-cultural differences and strategically introducing evidence or new information.

In addition to this solid empirical foundation there seems to be a consensus in the field that rapport (i.e., building a positive working relationship between operator and source) is widely regarded as a key foundation to most interrogation settings (Sivasubramaniam & Goodman-Delahunty, 2021). This is reasonable for a number of reasons. First, HUMINT is based on frequent collaboration between sources and interviewers: This collaboration can only be successful if both parties trust each other to some degree which can be characterized as a “working alliance” (Nunan et al., 2022, p. 3). Second, a friendly approach that preserves the sources´ autonomy (regarding sharing information) impedes the emergence of guilt within the source: Hence the source might be willing to share more information (Alison & Alison, 2017). Third, the social norm of reciprocity applies to the interrogation setting as well: When a small favor is given (without asking for something in return), the initial favor creates a felt obligation to return the favor, i.e., sharing one´s knowledge (High-Value Detainee Interrogation Group, 2016). Fourth, a friendly, motivational approach makes it harder for sources to withdraw from the social situation, for instance by refusing to look at interviewers or remaining silent (Alison et al., 2014). Note that accusation-oriented, coercive interrogation methods are powerful tools to make sources “confess,” however these confessions often fail to be trustworthy (Meissner et al., 2014). While interrogations are psychologically highly complex settings, simply acting friendly seems to be a potent tool to elicit information.

The taxonomy of interrogation methods

In order to evaluate the effectiveness of interrogation methods a fitting overarching framework is crucial. The Taxonomy of Interrogation Methods (ToIM) introduced by Kelly et al. (2013) provides a comprehensive framework for research and practice on the phenomenon of interrogation. It includes a classification of interrogation methods at three levels derived by theory and practitioners´ expertise. The top level is formed by dichotomous and broad categories, e.g., information-gathering versus accusatorial approaches (Macrolevel Categories). Below this broad level are six categories of interrogation methods (Mesolevel Domains), namely (a) Rapport and Relationship Building, (b) Context Manipulation, (c) Emotion Provocation, (d) Confrontation and Competition, (e) Collaboration, and (f) Presentation of Evidence, which in turn are constituted by 71 individual techniques (Microlevel Techniques). Rapport and Relationship Building includes techniques that reflect a working relationship between operators and sources based on mutual respect and understanding of each other´s goals. Context Manipulation comprises changing the circumstances or physical space in which the interrogation takes place or the choice of time of day when the interview takes place. Emotion Provocation techniques include, for example, rationalizations, appeals to self-interest, the source´s conscience, pride, or fear, and feelings for others with the goal of evoking emotion. Confrontation and Competition includes techniques for asserting authority and control over the source through direct blaming, implicit threats for noncompliance, and disallowing denials. Using Collaboration techniques aims for a cooperative relationship between interrogator and source, ergo collaboration, e.g., by offering rewards in exchange for information. Techniques of Presentation of Evidence involve the provision of documentation that establishes the sources´ guilt or complicity, including bluffs and misinformation or audio files and visual aids.

Kelly et al. (2013) see the intended practical applicability of the ToIM in law enforcement, counterterrorism, military contexts, as well as in other HUMINT operations, primarily regarding the interrogation of individuals accused of some wrongdoing. However, they concede that the ToIM might be applicable in the context of interviewing witnesses and victims as well. For this reason, we believe the ToIM is a useful approach to the spectrum of military source operations, favoring a controlled and systematic approach by the HUMINT operator.

The ToIM itself is not without limitations. While Kelly et al. (2013) also propose a model for psychological processes that underlie (effective) interrogations (ergo interactions between the mesolevel domains during interrogation), bear in mind that the assumptions regarding these causal intertwining of mesolevel domains are not sufficiently validated at this point. On a similar note, it is unclear if the ToIM captures the full range of interrogation techniques. Although the authors used thorough inductive methodology to identify and categorize many techniques, it appears they were unable to capture some highly relevant approaches in the field (which will become evident later on). This might be rooted in the decision of the authors to focus on the interrogation of “guilty” individuals, not mere innocent witnesses (see above). Nevertheless, the ToIM is suitable for the study of interrogation at a meta-analytic level, as it provides a uniform framework with which researchers and practitioners can describe and explain a variety of interrogation techniques (Kelly et al., 2016, p. 297) – regardless of their presumed effectiveness.

Research question

For our meta-analysis, we chose the ToIM as an underlying framework to determine the effectiveness of the mesolevel domains. But what does effectiveness mean in this context? Although there might be varying goals within interrogation settings and a number of parameters to be considered (e.g., the reliability, the role or long-term cooperation of a source), the main objective is always gathering of information. We focused on two dimensions of interest regarding the effectiveness of interrogation techniques. First, there is the quantity of information that can be gathered. The quantity of information is a relatively simple measure, e.g., the number of details a source is reporting or how many chunks of a narrative are revealed. Second, the quality of information needs to be considered, since more information does not necessarily mean that the information is reliable or useful. Quality of information can be operationalized, e.g., the accuracy rate (proportion of correct details in relation to all reported information) or the relative importance of the reported information. This directly contributes to the subsequent evaluation of the information obtained in the form of a better assessment of the reliability of the source and credibility of the information, as is common in military intelligence reports.

Every interrogation technique is designed to be beneficial in the interrogation context in some way. Consequently, for our meta-analysis, we test for each mesolevel domain separately if they have a meaningful effect in terms of improved quantity and quality of information. Although, our approach is mostly exploratory in nature with respect to the rather broad mesolevels, based on our theoretical and empirical considerations above, we expect rapport to be a meaningful predictor of higher quantity and quality of information.

Method

Inclusion criteria and systematic literature search

The systematic literature search was conducted in late 2021 and ended in November 2022. The databases PsycINFO, PsycArticles, PSYNDEX, Psychology and Behavioral Sciences Collection, and Web of Science were searched by using pertinent interrogation terms and their combinations for publications in German or English.

In addition to searching the databases, reference lists of well-known work on interrogation (Bull, 2014; Gordon & Fleisher, 2010; Gudjonsson, 2018) and websites of the International Military Testing Association (IMTA), the National Consortium for the Study of Terrorism and Responses to Terrorism (START) and the RAND Corporation were inspected in order to identify additional relevant publications.

To be included in the meta-analysis, studies had to incorporate an interrogation technique or a set of techniques that could be categorized within a ToIM mesolevel domain and measure quantity and/or quality of information as a criterion of outcome – preferably in an experimental design. Many of the included studies incorporated variants of the Scharff technique, the CI or the SUE framework (see above). Naturally, almost all studies only included mock interrogation scenarios with no aversive consequences for participants. In some cases, we also accepted field studies (n = 3) or studies with a correlational design (n = 2) if the ToIM-domains were operationalized in an adequate manner, since their inclusion added to the methodological variety of our study pool. During the systematic literature research, we found a substantial body of research that did not fit to any ToIM domain but appeared as relevant to our meta-analysis. Interestingly, these studies had a commonality: They all incorporated some form of cognitive support, special witness instruction or helpful answering format, e.g., a timeline technique where participants witnessed a conversation and were then asked to visually sort bits of conversations or actions on a timeline – comparable to a movie script or SMS-conversation. Hence, we decided to take an inductive approach: after we identified these similarities during the literature review and were unable to integrate them under the ToIM-umbrella, we decided that their inclusion is reasonable and formed a sufficiently homogeneous category. Although we developed no detailed rationale on how to combine these studies with existing theoretical frameworks, we chose to include these studies under the subcategory Cognitive Facilitation. This category encompasses all the techniques that required some kind of cooperation from the source where the source was willing to share information and the interviewer actually helped the source with generating and improving the statement – primarily by cognitive aids.

Note that some approaches from the field of interrogation describe the psychological processes that might underlie the effectiveness of cognitive techniques: The core principle of the CI for example is “guided retrieval” (Fisher & Geiselman, 1992, p. 15): Eyewitnesses receive support in order to improve the quality of their statement, e.g., by encouraging in-depth reporting or by creating a relaxed atmosphere to enhance their cognitive capacities. Bear in mind that relaxation appears reasonable in many interviewing situations since acute stress has shown to have a detrimental on memory and cognitive ability in general (Shields et al., 2016).

Studies had to provide sufficient information for an effect size and an accompanying sample size. In most cases, we had to calculate effect sizes ourselves and since most studies were experimental with at least two groups (the specific interrogation technique vs. standard questioning), we chose the point-biserial correlation as the common effect size. Therefore, the meta-analytic overall effect size is equivalent to a common correlation coefficient.

The final study sample for the meta-analysis (n = 60) comprised up to k = 22 effect sizes for a single ToIM domain. All studies were coded with a coding manual by both authors and disagreements could be solved by discussion. Overall, the agreement was 98% (11 initial disagreements for 604 coding decisions) for a subset of 50 studies. Figure 1 shows a PRISMA flow diagram documenting the literature search results. The dataset and the included studies can be inspected online (see Data Availability Statement).

Figure 1.

PRISMA flow chart for the systematic literature search.

Method of meta-analysis

The meta-analytic model used throughout this meta-analysis is the fixed-effect model (FE model), since the heterogeneity variance that plays an important role in the alternative random-effects model (RE model) cannot be estimated with sufficient precision to avoid biased results if less than 30 studies are used to aggregate correlations in meta-analyses (Schulze, 2004). Additionally, it is difficult to argue that the available studies are a random sample from a well-defined universe of studies on that specific topic. Hence, the fixed-effect model was used because it is appropriate for the intended inference and does not suffer from statistical result distortions under the given circumstances. As a result, the inference needs to be restricted to the set of studies included in the meta-analysis. For the aggregation of effect sizes, the minimum variance unbiased estimator as proposed by Olkin and Pratt (1958) was used. All computations were conducted with the package metafor in R using the inverse sampling variance as weights. We deem overall meta-analytic results reliable in terms of interpretability if (a) the number of effect sizes is k ≥ 10 since this is a common number that is deemed necessary for many important procedures in meta-analysis (Higgins & Green, 2017) and (b) the publication bias analyses (see supplementary online materials) do not suggest severe distortions.

Results

The results for the meta-analyses or rather the overall effect sizes of the ToIM mesolevel domains regarding quantity of information are depicted in Table 1. While the number of effect sizes is negligible and too small to interpret for Emotion Provocation, Confrontation and Competition and Collaboration, the overall effects for Rapport and Relationship Building and Presentation of Evidence are small and positive and the overall effect for Context Manipulation is positive and moderate in magnitude. As mentioned before, we included some studies that could not be subsumed under the ToIM-umbrella (Cognitive Facilitation). The overall effect size for these studies was surprisingly large with $\bar{r}$ = .5498 (CI [.5192; .5804]; k = 31; N = 2032). Furthermore, we calculated an overall effect size for all studies within the ToIM (effect sizes from different ToIM domains but the same sample were averaged) which resulted in a small effect of $\bar{r}$ = .2255 (CI [.1962; .2547]; k = 33; N = 3957). There was considerable heterogeneity for almost all effect sizes as can be seen in the I²-values, however the number of effect sizes was too low to “switch” to the RE model or conduct moderator analyses. Taken together, the results show that there is solid evidence that interrogation techniques from the ToIM extract more information in comparison to standard questioning – particularly for Rapport and Relationship Building, Presentation of Evidence and to a lesser degree Context Manipulation. Due to the sufficient numbers of effect sizes, the positive effects for Rapport and Relationship Building, Presentation of Evidence and Cognitive Facilitation on quantity of information are depicted in detail in three forest plots as well (see Figures 2, 3 and 4): Each effect size is listed with its accompanying confidence interval and weight to generate an overall effect (FE model).

Table 1.

Meta-analytic overall effect sizes for the ToIM-interrogation mesolevel domains´ effectiveness regarding quantity of information.

Mesolevel domain	k	N	$\bar{r}$	95% CI	I2	95% CI (I2)
Relationship and Rapport Building	22	2624	.1978	[.1619; .2338]	83.67	[70.80; 91.54]
Context Manipulation	6	918	.3253	[.2676; .3829]	71.44	[28.98; 94.03]
Emotion Provocation	1	123	.2851	−	−	−
Confrontation and Competition	4	459	.0701	[−.0207; .1610]	76.79	[.00; 99.02]
Collaboration	2	127	.2158	[.0449; .3867]	.00	[.00; 99.24]
Presentation of Evidence	12	1185	.2341	[.1808; .2874]	75.48	[51.78; 93.66]
Overall effect ToIM mesolevel	33	3957	.2255	[.1962; .2547]	75.02	[52.76; 81.88]
Overall effect beyond ToIM mesolevel	31	2032	.5498	[.5192; .5804]	84.72	[74.40; 90.74]

Overall effect ToIM mesolevel: overall effect size for all studies that could be subsumed under the ToIM model. Overall effect beyond ToIM mesolevel: overall effect size for every other study that could not be subsumed under the ToIM model (Cognitive Facilitation).

k = number of independent effect sizes; N = aggregate sample size; $\bar{r}$ = overall effect size for a FE model; 95% CI = 95% confidence interval for $\mathit{\rho }$ or I²; I² = proportion of variance in observed effect sizes that is due to heterogeneity.

Figure 2.

Forest plot for all included studies on relationship and rapport building regarding quantity of information.

Figure 3.

Forest plot for all included studies on presentation of evidence regarding quantity of information.

Figure 4.

Forest plot for all included studies on cognitive facilitation regarding quantity of information.

In a similar fashion, the overall effect sizes of the ToIM mesolevel domains regarding the quality of information are depicted in Table 2. Note that the number of effect sizes is even lower in comparison to quantity of information and the results should be interpreted with great caution. Nevertheless, we found a small effect size for Rapport and Relationship Building on the quality of information with $\bar{r}$ = .1728 (CI [.1044; .2412]; k = 6; N = 720). Again, we included some studies that could not be subsumed under the ToIM-umbrella. The overall effect size for these studies was considerably smaller compared to the effect on quantity of information with $\bar{r}$ = .1464 (CI [.0774; .2155]; k = 17; N = 828). Parallel to the quantity of information, we calculated an overall effect size for all studies within the ToIM (effect sizes from different ToIM domains but the same sample were averaged) which resulted in a small effect of $\bar{r}$ = .1246 (CI [.0711; .1781]; k = 10; N = 1301). Taken together, the results show that there is some evidence that interrogation techniques from the ToIM-framework and beyond (Cognitive Facilitation) extract more precise information in comparison to standard questioning – particularly for Rapport and Relationship Building.

Table 2.

Meta-analytic overall effect sizes for the ToIM-interrogation mesolevel domains´ effectiveness regarding quality of information.

Mesolevel domain	k	N	$\bar{r}$	95% CI	I2	95% CI (I2)
Relationship and Rapport Building	6	720	.1728	[.1044; .2412]	48.29	[.00; 93.00]
Context Manipulation	3	463	.3099	[.2310; .3887]	96.37	[86.53; 99.90]
Emotion Provocation	0	−	−	−	−	−
Confrontation and Competition	3	400	−.0443	[−.1408; .0522]	90.79	[56.85; 99.80]
Collaboration	1	68	−.0200	−	−	−
Presentation of Evidence	3	178	.3338	[.2002; .4674]	21.82	[.00; 96.68]
Overall effect ToIM mesolevel	10	1301	.1246	[.0711; .1781]	59.67	[13.00; 90.68]
Overall effect beyond ToIM mesolevel	17	828	.1464	[.0774; .2155]	.00	[.00; .00]

Overall effect ToIM mesolevel: overall effect size for all studies that could be subsumed under the ToIM model. Overall effect beyond ToIM mesolevel: overall effect size for every other study that could not be subsumed under the ToIM model (Cognitive Facilitation).

k = number of independent effect sizes; N = aggregate sample size; $\bar{r}$ = overall effect size for a FE model; 95% CI = 95% confidence interval for $\mathit{\rho }$ or I²; I² = proportion of variance in observed effect sizes that is due to heterogeneity.

Observe that the Confrontation and Competition-effect – the technique domain that can be described as relatively coercive and aversive – did not reach statistical significance for neither quantity nor quality of information. Additional analyses concerning a possible publication bias can be inspected in the supplementary online materials (see Data Availability Statement).

Discussion

Taken together, for a quantity of information overall effects for Rapport and Relationship Building and Presentation of Evidence are small and positive and a (less reliable) overall effect for Context Manipulation is positive and moderate in magnitude. The effects are very similar for quality of information. Hence, our hypothesis regarding the effect of rapport on quantity and quality of information is corroborated. The results on the other three mesolevels are too fragile to be interpreted due to the very low number of effect sizes. The general notion within the field that a friendly, respectful interrogation style in combination with a slightly manipulative use of evidence presentation is clearly supported by our analyses. Note that our results are comparable in direction and magnitude to earlier meta-analytic results on the efficacy of the (rapport-oriented) CI and the Scharff Technique (see above). Even the non-significant results for Confrontation and Competition indirectly support that an approach similar to coercion is not beneficial for the interrogation process. This is in line with recent work on the efficacy of the infamous enhanced interrogation techniques (EITs) that illustrates how claims that coercion “works” are methodologically flawed and rarely based on empirical evidence (Duke & Van Puyvelde, 2017).

Our “new” mesolevel domain Cognitive Facilitation beyond the ToIM surprisingly comprised a solid body of research. Apparently, there is a variety of ways to support sources to improve and increase the usefulness of their statements – at least in terms of quantity of information. While the existing domains of the ToIM are set up to influence the social dynamics between interviewer and source or the perception of one´s own circumstances, techniques of Cognitive Facilitation aim to foster the sources´ ability to report accurately. Bear in mind that the use of these techniques is probably unreasonable with uncooperative sources, since as long as someone is unwilling to share information, they might simply ignore cognitive aids. Hence, the inclusion of Cognitive Facilitation acknowledges the psychological complexity of the interrogation setting. We believe that this area should receive more attention not only in law enforcement research, but also in military HUMINT activities. In future research, cognitive facilitation should therefore be integrated as an additional area in the conceptual framework of source operations. The inclusion of a well-defined additional mesolevel domain to the ToIM should be considered.

Limitations

We need to acknowledge a few limitations of our meta-analysis. First, some challenges emerge when using the ToIM as a conceptual framework for the meta-analysis: The categorization of the studies into the mesolevel domains was an ambiguous and complicated process and needed intense conceptual discussion between the authors. We had to exclude several studies since they could not be included under the ToIM, although many of these studies contain valuable knowledge for the field of interrogation. Note that some methods (e.g., the Scharff Technique or the Cognitive Interview) encompass elements from several distinct mesolevel domains and not just one. Therefore, in 18 cases, we decided to use one and the same effect size for more than one mesolevel domain since the methods or the study design included elements from several mesolevel domains. Unfortunately, it was not possible to untangle the effects from these combined techniques for each mesolevel domain respectively. As a consequence, the manifest effect sizes cannot be attributed to only one explaining factor – here the mesolevel domain – and interpreting these effect sizes becomes vague to some degree. However, the alternative would be to use no interrogation model at all. In order to better understand actual psychological processes, we believe that our model-based approach outweighs the disadvantages. On a similar note, it may have been appropriate to consult with a third expert to resolve coding disagreements. Furthermore, the included studies´ respective operationalization of interrogation techniques did not portray the full range of microlevel techniques that are described in the ToIM for each mesolevel domain, e.g., identifying contradictions within a source´s story for Presentation of Evidence did not occur in any of the included studies. Hence, our study sample has no representative sample for its respective domain. Note that the ToIM primarily applies to sources that are suspects or captured persons – although some studies we included were designed to treat participants as suspects, in most studies the participants had a simple witness role. Previous discussions on the topic of interrogation are about the effectiveness of torture or at least coercion. While coercion has undoubtedly been an element of interrogation throughout history (Hartwig et al., 2014), Kelly et al. exclude such coercive techniques in their model.

Second, our criteria of effectiveness – quantity and quality of information – are difficult to operationalize. Only n = 24 studies (primarily beyond ToIM) measured the quality of information in any way, however most of these studies focused on accuracy, that is the proportion of correct information in relation to all information gained from the source. In other words, these studies examine what makes participants´ statements more precise – not necessarily “more true.” Evaluating the quality of information in a real interrogation setting comes with additional challenges – especially with the truthfulness of statements made by a source. Note that more is not necessarily better: Some interrogation techniques might be prone to produce a lot of information, however it might be virtually useless, because the source might tend to lie if confronted with that specific technique. Although it is conceptually plausible, we would refrain from the interpretation that our results definitively show that some interrogation techniques make sources tell the truth.

Third, there are limitations due to the use of meta-analysis: The number of effect sizes (and also the overall sample sizes of the primary studies) for most ToIM-domains was low (especially quality of information). Consequently, many analyses that might have enhanced the understanding of interrogation effectiveness were not possible (use of RE model, moderator analyses, subgroup analyses, consideration of other criteria of effectiveness besides quantity/quality of information, …). Furthermore, the inference that can be made in a meta-analysis is limited by the study design and study quality of the primary research included. Most of the included studies were experimental studies which is beneficial in terms of internal validity, ergo causal interpretability. However, most studies included student samples and the study design was necessarily restrained due to ethical considerations, but therefore hardly comparable to actual interrogation settings. Most study samples knew that they would not face severe consequences depending on their behavior in the study and ecological validity is probably limited. We recommend that – in addition to experimental studies – more field studies on actual interrogations should be conducted (see Alison et al., 2014).

Practical implications

Despite the aforementioned limitations of this work, we try to formulate some concrete recommendations for practitioners: First, HUMINT operators should always try to start on a positive note and form a friendly relationship with the source by treating the other person respectfully. Second, they should present evidence to evoke the impression that the HUMINT operators are well aware of all relevant circumstances to appear in a benevolent-superior way. The presentation of manufactured evidence might be reasonable in very specific situations. Third, HUMINT operators should assist willing sources by providing them with cognitive aids, such as a timeline format, a detailed description of the information needed by the source, or a description of events from another person’s perspective. Fourth, HUMINT operators should not use coercive techniques or suggestive interviewing since these techniques did not have beneficial effects in our meta-analysis and have shown to be related to the risk of questionable intelligence (Duke & Van Puyvelde, 2017). In particular, the possibility of false memories (Loftus & Pickrell, 1995) and false testimonies (Leo, 2009) have extensively been discussed as a potential problematic outcome of improper interviewing. Note that the characteristics of the source (adolescence, intellectual disability, mental illness, and certain personality traits) and interview situation (excessive interrogation time, presentations of false evidence) are related to the emergence of false testimonies (Kassin et al., 2010). Additionally, it impedes the option to implement techniques from psychological credibility assessment to assess the sources´ information (Volbert & Steller, 2014).

Concluding remarks

The field of interrogation research is in good shape, and it is growing. We applaud the many authors for their excellent work and their approach to test their interrogation techniques by using randomized controlled trials. However, effectiveness of interviewing techniques depends on the objective one wants to achieve and differs per circumstance and per type of interviewee. Only some studies consider this, e.g., by assigning participants the role of a witness or possible perpetrator. Also, the sources` personality, cultural background, motivation and training certainly plays a role in which techniques might work, and which might not. Future research should therefore test the possible effects of varying circumstances to determine which settings require which interrogation technique or combination of interrogation techniques. This may include the level of training of HUMINT operators and, by extension, consideration of framework activities in which information gathering is not conducted by HUMINT specialists but by conventional or special forces. A future outcome of this research should be a portfolio of techniques derived directly from theory, practical knowledge and empirical evidence that can be used across the military HUMINT spectrum of activities.

On a different note, the actual effective implementation of the mesolevel domains needs clarification: How can rapport be operationalized within a practical HUMINT setting without appearing manipulative to the source? An increase of field studies might solve some of these issues, however not as an alternative but an addition: Experimental interrogation studies are elaborate and expensive as can be observed by the rather small sample sizes. However, they are without alternative to gain insight in the causal structure of these complex social situations, which by their very nature take place during military operations in conflict and war zones. Hence, the field simply needs variation and replication – especially for the ToIM domains Context Manipulation, Emotion Provocation, Confrontation and Competition and Collaboration.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The dataset, the analysis script, all supplementary material and all references of studies included in the meta-analysis are available in the online material (https://osf.io/f3e7q).

Open scholarship

This article has earned the Center for Open Science badge for Open Data. The data are openly accessible at (https://osf.io/f3e7q).