Snap Shot
- The question: If a behavior such as delay, continued contact, or calm demeanor is common among sexual assault victims, what does its presence say about whether a particular report is accurate?
- The key concept: A behavior supports accuracy only if it is more common in accurate reports than in inaccurate ones (a likelihood ratio above 1); a ratio of 1 means it is evidence of nothing. Its frequency among victims, standing alone, does not establish probative value.
- What the research does not supply: Studies usually report only the frequency among self-identified victims; for most behaviors, reliable figures for inaccurate reports are not available. “Consistent with” assault, and “consistent with” a motive to fabricate, have not been shown to carry diagnostic value by themselves.
- The law: Testimony that a behavior is common remains relevant to rebut an argument or assumption that the behavior shows a report is false. Group statistics about how often reports are true or false may not be used to measure a witness’s truthfulness (Brooks, Mullins, Lindsey, Kinney, Grimshaw), and that logic applies equally to statistics suggesting a complainant is lying. Haskie cautions against categorical claims without empirical support.
- Issues for each side: Prosecutors should frame the testimony as rebuttal of an invalid inference and stop there, avoiding “matches true victims” language and false-report statistics; defense counsel should test what “consistent with” means and whether any study compared true and inaccurate reports, while recognizing that the same non-specificity defeats “consistent with fabrication” arguments.
Counterintuitive behavior testimony typically establishes that a behavior, such as delayed disclosure, continued contact with the accused, or a calm demeanor, is common among people who have been sexually assaulted. That proposition is often well supported. Sexual assault is substantially underreported, responses to it vary widely, and the behaviors addressed by the testimony genuinely occur in many cases in which an assault took place. The question this paper addresses is different: what follows, as a matter of logic and probability, from the fact that a behavior is common among victims?
The answer turns on a few concepts from the science of diagnostic inference: sensitivity, specificity, base rates, and likelihood ratios. These concepts explain why a statement that a complainant’s behavior is “consistent with” sexual assault can be accurate and still carry no information about whether an assault occurred. They also explain why the same limitation applies with equal force to arguments that a behavior is “consistent with” fabrication. Courts have reached similar conclusions through the rules on relevance, unfair prejudice, and vouching.
This paper explains these concepts in plain language, works through a numerical example using hypothetical figures, and then examines why most research on counterintuitive behavior cannot supply the comparison these concepts require. It closes with the legal rules that govern probative value and statistical credibility evidence.
Two Different Questions
Every behavioral indicator raises two questions that are easy to confuse. The first is how often the behavior occurs among people who were assaulted. The second is how often it occurs among people whose reports are not accurate, whether because of mistake, misunderstanding, or fabrication. Counterintuitive behavior research usually answers the first question. Only a comparison of the two answers tells a factfinder anything about whether a particular report is accurate.
A simple hypothetical outside the courtroom makes the point. Suppose nearly everyone with a particular illness gets a headache. A headache would still be a poor indicator of that illness, because headaches are also common among people who do not have it. How often a sign appears among people who have a condition says little, by itself, about whether a person with that sign has the condition. The same reasoning applies to behavior after an alleged assault: a behavior can be very common among victims and still be equally common among people whose reports are inaccurate.
Sensitivity, Specificity, and Likelihood Ratios
Diagnostic research uses precise terms for these ideas. Sensitivity is the proportion of people who have a condition who show the sign; in this context, the proportion of people who were assaulted who display a given behavior. Specificity is the proportion of people who do not have the condition who do not show the sign; here, the proportion of people whose reports are inaccurate who do not display the behavior. A sign with high sensitivity and low specificity is common in both groups and distinguishes poorly between them.
The likelihood ratio combines the two. It is the probability of observing the behavior if the report is accurate divided by the probability of observing it if the report is inaccurate. A likelihood ratio of 1 means the behavior is equally likely either way and is not evidence of anything. A ratio greater than 1 means the behavior is more common in accurate reports and therefore supports accuracy to some degree; a ratio less than 1 means it is more common in inaccurate reports. Lyon and Koehler (1996) applied this framework, which they called the relevance ratio, to expert testimony about child sexual abuse, and their analysis applies equally to adult cases. They showed that the frequency of a behavior among abuse victims is not by itself a measure of its probative value; what matters is its frequency among victims compared with its frequency among non-victims. They also showed the converse: testimony that a behavior is common among victims can be relevant for the limited purpose of rebutting an argument that the behavior shows the report is false, because that argument implicitly assumes the behavior is rare among victims.
A Worked Example
The following figures are hypothetical. They are chosen only to make the arithmetic clear and are not estimates of any real proportion of accurate or inaccurate reports, which is not known for any particular setting and which no one should infer from these numbers.
Suppose a group of 1,000 reports in which, hypothetically, 500 are accurate and 500 are inaccurate, a split chosen only because it makes the arithmetic easy. The starting odds that a report from this group is accurate are 500 to 500, or 1 to 1, a probability of 50%. Now suppose a behavior, such as waiting more than a week to disclose, occurs in 40% of accurate reports and also in 40% of inaccurate reports. Among the accurate reports, 200 show the behavior; among the inaccurate reports, 200 do. Of the 400 reports showing the behavior, 200 are accurate, which is 50%, exactly the starting probability. The likelihood ratio is 0.40 divided by 0.40, or 1. Learning that the complainant delayed would tell the factfinder nothing in this hypothetical, even though “40% of victims delay” is a true statement and even though an expert could correctly say that delay is “consistent with” assault.
Change one number. Suppose the behavior occurs in 40% of accurate reports but only 20% of inaccurate reports. Now 200 accurate and 100 inaccurate reports show the behavior, so 200 of 300, about 66.7%, are accurate. The likelihood ratio is 0.40 divided by 0.20, or 2, and the odds move from 1 to 1 to 2 to 1. The behavior is now modest evidence of accuracy. The absence of the behavior also becomes informative in the opposite direction: 300 accurate and 400 inaccurate reports lack the behavior, so the probability of accuracy among reports without delay is 300 of 700, about 42.9%. A behavior that genuinely discriminates between groups does so in both directions.
Three lessons follow. First, the frequency of a behavior among victims, standing alone, does not establish that it is evidence an assault occurred, although it can be relevant to rebut an argument that the behavior shows a report is false; the comparison group is essential. Second, a likelihood ratio adjusts whatever the starting odds happen to be; it cannot replace them, and the starting odds in a particular case depend on all the other evidence, which is the factfinder’s province. Third, even a behavior with a likelihood ratio of 2 shifts the probability only modestly, so a single behavioral indicator will rarely be strong evidence for either side.
Why “Consistent With” Has No Diagnostic Value by Itself
Expert testimony often states that a complainant’s behavior is “consistent with” that of sexual assault victims. In the vocabulary above, that phrase usually means only that the behavior’s sensitivity is above zero: some victims display it. It does not say whether the behavior is more common among victims than among people whose reports are inaccurate, and therefore it does not say whether the behavior is evidence of assault. Because responses to assault vary so widely, nearly any behavior is consistent with having been assaulted. A federal appellate court noted this feature of the testimony when an expert conceded on cross-examination that “everything and anything” can be consistent with abuse (United States v. Roberts, 2023).
The same analysis applies to every other “consistent with” claim. Behavior described as “consistent with” a motive to fabricate, such as a complainant’s anger at the accused, a new relationship, or a disciplinary problem, is equally uninformative unless that behavior is more common among people whose reports are false than among people whose reports are true, and I am not aware of research establishing that for ordinary social circumstances. An argument that the complainant’s calm demeanor shows she was not assaulted assumes that calm demeanor is rare among victims, an assumption contradicted by research going back to the earliest clinical descriptions, which reported that many women seen after a rape presented in a controlled, calm manner (Burgess & Holmstrom, 1974). Non-specificity is a limit on inference, not an argument for either side.
Affirming the Consequent
The logical error underlying much “consistent with” reasoning has a name. Affirming the consequent takes a true conditional statement, “if a person was assaulted, the person may delay reporting,” observes the consequent, “the complainant delayed,” and concludes the antecedent, “the complainant was assaulted.” The conclusion does not follow, because other circumstances also produce delay. The defense version of the error is identical in form: “a person fabricating an allegation might delay; the complainant delayed; therefore the complainant fabricated.” Neither inference is valid.
Educational testimony used properly avoids this error. Its legitimate function is to defeat an invalid inference, that delay shows no assault occurred, by showing that the premise behind it, that genuine victims report promptly, is false. It does not supply the reverse inference.
Comparison Groups and Ground Truth
The worked example requires two numbers, but research on counterintuitive behavior almost always supplies only one. Studies of tonic immobility, delayed disclosure, and continued contact typically recruit people who identify themselves as having been assaulted, through clinics, rape crisis services, college surveys, or treatment programs, and describe how often they report particular behaviors. That design is appropriate for its purpose, which is to understand victims’ experiences, and its findings establish that the behaviors occur among victims, often at substantial rates. It does not produce the second number, the frequency of the behavior among people whose reports are inaccurate.
There are reasons that number is rarely available. Researchers seldom know with certainty which reports are accurate, because “ground truth” in sexual assault cases is usually unavailable. Studies of police case files can classify some reports as false after investigation; the commonly cited estimates of confirmed false reports range from roughly 2% to 10% of cases, with a substantial further proportion that cannot be classified either way (Lisak et al., 2010; Spohn et al., 2014; Ferguson & Malouff, 2016, 5.2%, 95% CI 3.0% to 8.9%), and these figures cannot be applied to any one complainant; in the university study by Lisak and colleagues, for example, 8 of 136 reports (5.9%) were coded false, while 44.9% did not proceed to prosecution or disciplinary action and 13.9% had too little information to classify. That undetermined group is not evidence of falsity; it consists of cases in which the evidence did not permit a determination. In addition, comparing behaviors across police classifications risks circularity. Spohn and colleagues found that a complainant’s recantation was the strongest predictor of a police decision to classify a report as unfounded, so a study that compared recantation rates between “unfounded” and other reports would partly be measuring the investigators’ own decision rules rather than an independent association.
Research on deception detection more generally confirms how weak behavioral cues tend to be. A meta-analysis of 120 independent samples found that most behavioral cues to deception were only weakly related to lying (DePaulo et al., 2003), and a meta-analysis of deception judgments found that people distinguish truths from lies at about 54% accuracy, only modestly better than chance (Bond & DePaulo, 2006). Research on the “emotional victim effect” shows that observers nonetheless rely on demeanor: complainants who display expected distress are judged more credible than those who appear controlled, largely because the controlled demeanor violates observers’ expectations (Ask & Landström, 2010). This body of work supports caution about inferring either truthfulness or deception from behavior, which is the same caution the base-rate analysis requires.
Limits of the Evidence
The likelihood-ratio framework is a way of reasoning about evidence, not a source of numbers for a particular case. For nearly every behavior at issue in counterintuitive behavior testimony, reliable estimates of the behavior’s frequency among people whose reports are inaccurate do not exist, and estimates of its frequency among victims vary with the sample, the definition of the behavior, and the time since the assault. The starting probability that any particular report is accurate is unknown and depends on case-specific evidence. Deception-detection research largely uses laboratory lies and student participants and cannot be applied directly to sexual assault reports. False-report studies use differing definitions and investigative records of varying quality. No research can tell a factfinder whether a particular complainant was assaulted; it can only describe what is known and not known about how informative a behavior is.
Questions to Ask About This Research
When an expert testifies that a behavior is “consistent with” sexual assault, counsel can ask what the base rate is and what a positive result means: in the hypothetical worked example, a behavior found in 40% of accurate and 40% of inaccurate reports has a likelihood ratio of 1, and studies rarely supply the second figure. When a behavior is offered as “consistent with” a motive to fabricate, the question is the same, since no study located shows that such behavior is more common among false reports. When either side cites the estimate that confirmed false reports make up roughly 2% to 10% of cases (Lisak et al., 2010; Spohn et al., 2014; Ferguson & Malouff, 2016), counsel can ask what exactly was counted, because the same studies leave a substantial further proportion that cannot be classified either way, and that undetermined group is not evidence of falsity. When the finding that people distinguish truths from lies at about 54% accuracy (Bond & DePaulo, 2006) is cited, counsel can ask who was studied, since deception research largely uses laboratory lies and student participants. These questions are explained in “Reading Research Claims: A Guide for Attorneys” on this site.
Base Rates, Probative Value, and the Law
Federal and Military Standards
Federal Rule of Evidence 401 defines relevant evidence as evidence having any tendency to make a fact of consequence more or less probable, and Rule 403 permits exclusion when probative value is substantially outweighed by the danger of unfair prejudice, confusing the issues, or misleading the jury. The likelihood ratio is, in effect, a quantitative expression of Rule 401’s “tendency”: a behavior with a likelihood ratio of 1 has no tendency to prove or disprove that an assault occurred. Counterintuitive behavior testimony remains relevant for a different reason, because it rebuts an argument or a juror assumption that a behavior shows the report is false. Its probative value is tied to that rebuttal purpose, and the Rule 403 balance shifts when the testimony is presented or argued as affirmative proof. Under amended Rule 702, the proponent must also show by a preponderance of the evidence that the expert’s opinion reflects a reliable application of the research to the facts, which bears directly on whether a group-level frequency can support the inference being drawn. Military Rules of Evidence 401, 403, and 702 are to the same effect, and United States v. Houser (1993) incorporates the Rule 403 balance as one of its six factors.
The Court of Appeals for the Armed Forces has applied these principles to statistical credibility evidence. In United States v. Brooks (2007) (a child-complainant case), an expert’s testimony about how rarely allegations like the complainant’s prove false was held to be the functional equivalent of vouching, because it amounted to a statement that the complainant was very likely telling the truth. In United States v. Mullins (2010) (also a child-complainant case), the court held that an expert may describe behaviors associated with abuse and whether the complainant exhibited them, but may not testify about credibility or guilt, and found error in testimony implying a very low probability that the allegation was false. In the Eighth Circuit, United States v. Roberts (2023) held that testimony describing complainants’ responses as fawning, flight, and freezing, and as not inconsistent with trauma, came close to the line of improper vouching, though the court found any error harmless.
State courts have reached the same result. In State v. Kinney (2000), the Vermont Supreme Court approved general testimony about victim behavior but held inadmissible testimony about the frequency of false rape reports. In State v. Grimshaw (2020), the Montana Supreme Court reversed a conviction because an expert testified that only 2% to 8% of sexual assault reports are false, reasoning that the clear inference was a 92% to 98% probability of guilt and that a defendant’s denial did not open the door to such evidence. Nothing in that reasoning is confined to one side. The logic of Brooks, Kinney, and Grimshaw, that group-level statistics about how often reports are true or false cannot be used to measure the truthfulness of a particular witness, applies equally to statistics offered to suggest that a complainant is likely lying. Military Rule of Evidence 608(a) likewise limits evidence of a witness’s character for truthfulness to opinion or reputation about that witness, not statistics about a category of witnesses.
Arizona
Arizona Rules of Evidence 401, 403, and 702 track the federal rules, and Rule 702 was amended effective January 1, 2024, to conform to the 2023 federal amendment. In State v. Lindsey (1986), the Arizona Supreme Court held that an expert may not opine on the truthfulness of a particular witness or of witnesses of that type, and may not quantify the percentage of victims who are truthful; admitting such testimony was reversible error (149 Ariz. at 475–77). State v. Salazar-Mercado (2014) permits educational testimony about general behavioral patterns, and State v. Haskie (2017) directs trial courts not to allow experts to speak in broad, categorical terms about what is “common” or “usual” without empirical support. Read together with the base-rate analysis, these decisions support testimony that describes behaviors and their documented frequency among victims while excluding testimony, from either side, that converts group statistics into a probability that the complainant is truthful.
Special Issues in Legal Practice
The principles described above apply with equal force to every party in a legal proceeding. The science operates the same way regardless of who is asking the questions, and the research summarized here does not by itself favor either side. What differs between the parties is not the science but the vantage point — the particular places where these issues tend to surface, and the particular errors each party may need to guard against. The two summaries below identify those recurring issues. They are offered for orientation, not as litigation strategy; in every case the controlling question is what the evidence in that case will, and will not, support, and the expert’s role is to inform the trier of fact rather than to advocate for either side.
Special Issues for Prosecutors
The base-rate analysis supports the core use of counterintuitive behavior testimony. When the defense argues, or jurors may assume, that a behavior such as delay or continued contact shows that no assault occurred, that inference depends on the premise that the behavior is rare among victims, and research showing it is common directly rebuts that premise. Lyon and Koehler’s analysis confirms that testimony serving this rebuttal function is relevant. Prosecutors can frame the testimony accordingly: describing how often a behavior has been found among victims in identified studies, explaining why jurors should not treat it as evidence of fabrication, and stopping there.
The analysis also counsels caution about statements that go further. Testimony or argument that a complainant’s behavior “matches” or is “consistent with” that of true victims, offered as support for her account, implies a likelihood ratio the research does not establish, and statistics about false-report rates were the basis for reversal in Brooks and Grimshaw. Keeping the testimony on the rebuttal side of that line avoids both the scientific overstatement and the appellate risk.
Special Issues for Defense
The base-rate analysis supports a precise challenge to overstatement. An expert who states that a behavior is common among victims has not stated that it is evidence of assault, and questions about whether any study compared the behavior’s frequency in true and inaccurate reports, and what the expert’s “consistent with” language actually means, test whether the testimony has strayed beyond its supported purpose. Defense counsel can raise this point before trial, under Rules 401, 403, and 702, and to request limiting instructions confining the testimony to its rebuttal use.
The analysis also counsels caution about the mirror-image argument. A behavior’s absence is not evidence that no assault occurred unless the behavior is shown to be more common among victims than among others, and arguments that a complainant’s conduct is “consistent with” a motive to fabricate fail the same test as “consistent with” assault. False-report statistics are no more admissible to suggest that a complainant is lying than to suggest that she is truthful, and the undetermined cases in false-report research are not false reports. Applying the non-specificity point evenly is more accurate than applying it to one side only.
This paper is for educational purposes only and is not legal advice. The author is a psychologist, not an attorney; summaries of law are general and must be verified against current authority or with a licensed attorney.
References
Ask, K., & Landström, S. (2010). Why emotions matter: Expectancy violation and affective response mediate the emotional victim effect. Law and Human Behavior, 34(5), 392–401. https://doi.org/10.1007/s10979-009-9208-6
Bond, C. F., Jr., & DePaulo, B. M. (2006). Accuracy of deception judgments. Personality and Social Psychology Review, 10(3), 214–234. https://doi.org/10.1207/s15327957pspr1003_2
Burgess, A. W., & Holmstrom, L. L. (1974). Rape trauma syndrome. American Journal of Psychiatry, 131(9), 981–986. https://doi.org/10.1176/ajp.131.9.981
DePaulo, B. M., Lindsay, J. J., Malone, B. E., Muhlenbruck, L., Charlton, K., & Cooper, H. (2003). Cues to deception. Psychological Bulletin, 129(1), 74–118. https://doi.org/10.1037/0033-2909.129.1.74
Ferguson, C. E., & Malouff, J. M. (2016). Assessing police classifications of sexual assault reports: A meta-analysis of false reporting rates. Archives of Sexual Behavior, 45(5), 1185–1193. https://doi.org/10.1007/s10508-015-0666-2
Lisak, D., Gardinier, L., Nicksa, S. C., & Cote, A. M. (2010). False allegations of sexual assault: An analysis of ten years of reported cases. Violence Against Women, 16(12), 1318–1334. https://doi.org/10.1177/1077801210387747
Lyon, T. D., & Koehler, J. J. (1996). The relevance ratio: Evaluating the probative value of expert testimony in child sexual abuse cases. Cornell Law Review, 82(1), 43–78.
Spohn, C., White, C., & Tellis, K. (2014). Unfounding sexual assault: Examining the decision to unfound and identifying false reports. Law & Society Review, 48(1), 161–192. https://doi.org/10.1111/lasr.12060
Legal Authorities
State v. Grimshaw, 2020 MT 201 (Mont. 2020).
State v. Haskie, 242 Ariz. 582, 399 P.3d 657 (2017).
State v. Kinney, 171 Vt. 239, 762 A.2d 833 (2000).
State v. Lindsey, 149 Ariz. 472, 720 P.2d 73 (1986).
State v. Salazar-Mercado, 234 Ariz. 590, 325 P.3d 996 (2014).
United States v. Brooks, 64 M.J. 325 (C.A.A.F. 2007).
United States v. Houser, 36 M.J. 392 (C.M.A. 1993).
United States v. Mullins, 69 M.J. 113 (C.A.A.F. 2010).
United States v. Roberts, 86 F.4th 1183 (8th Cir. 2023).
Fed. R. Evid. 401.
Fed. R. Evid. 403.
Fed. R. Evid. 702.
Ariz. R. Evid. 401.
Ariz. R. Evid. 403.
Ariz. R. Evid. 702.
Mil. R. Evid. 401.
Mil. R. Evid. 403.
Mil. R. Evid. 608.
Mil. R. Evid. 702.
Related topic papers: Introduction to Counterintuitive Behavior Testimony · Fit and Generalizability: Matching the Research to the Case · Variability, Confounds, and Individual Differences in Responses to Sexual Assault · Vouching, Profile Evidence, and the Human Lie Detector Rule · False, Unfounded, and Unresolved Reports: What the Research Shows
Related topics from Arizona Forensic Services: Memory in the Courtroom · Courts-Martial