What 911 Call Analysis Is and Why It's in Trouble

A former Ohio police lieutenant built a forensic technique in his 2006 master's thesis, and it's now reaching jurors in more than 100 documented cases across 26 states. The problem isn't just that 911 call analysis lacks empirical support. It's that the method was designed to slip past the rules meant to keep junk science out of courtrooms.

Tracy Harpster's technique claims to identify whether a 911 caller is guilty based on word choice, phrasing, and cadence. His COPS Scale grew from 16 guilt indicators in the original thesis to 38 in a 2017 book, yet the underlying research has never been released and the data behind those indicators has never been made public. Harpster himself has consulted on at least 2,500 homicide investigations but has never testified as an expert in a criminal trial.

That structural gap is what makes the technique dangerous. According to a 2026 Cornell Law Review article by Christopher Lau of the Wisconsin Innocence Project, 911 call analysis is "particularly pernicious because it operates in the shadows." Prosecutors are coached to present the analysis through officers or dispatchers as lay opinion under FRE 701, bypassing the expert-qualification requirements of FRE 702 entirely. The methods stay undisclosed, the founders stay off the stand, and the technique largely escapes Daubert or Frye scrutiny.

Courts have pushed back in isolated cases. Harpster was rejected as an expert in Alabama in 2009. A Nevada judge rejected a detective trained in the method. In Michigan, a conviction was overturned because 911 call analysis and other statement-analysis techniques should never have been admitted. The pattern is clear: when courts actually examine the technique under proper standards, it fails.

The institutional response is building. A Governing.com investigation documents prosecutors, judges, and defense attorneys pushing for bans, conviction reviews, and sanctions against prosecutors who used the technique while disguising it. NACDL President Nellie King called it another addition to "the junk pile of fake scientific theories." An Ohio judge told reporters he "never anticipated that prosecutors, officers of the court, would engage in systematic organized frauds."

For defense teams, the practical reality is this: a junk-science method is still reaching juries through disguised channels. The audio and transcripts don't lie, but the interpretation layered on top of them often does. Working the actual recordings with word-level timestamps and speaker diarization lets you test claimed "guilt indicators" against what the caller actually said, frame by frame. Evidence Copilot's 911 call transcript analysis and jail call transcription support that review with citations to the exact source moment, so every challenge you raise is grounded in the record rather than an analyst's untested framework.

How the Technique Evades Daubert and Frye (and How to Attack It)

Here's the uncomfortable truth about 911 call analysis: it has been used in more than 100 documented cases across 26 states, yet its creator has never once testified as an expert in a criminal trial. Tracy Harpster, the former Moraine, Ohio police lieutenant who built the technique in his 2006 master's thesis, has instead spread it through publications, nationwide training, and consulting on over 2,500 homicide investigations, all while staying off the witness stand where his methods would face scrutiny.

That's not an accident. It's the design.

The technique survives admissibility challenges because it's structured to avoid them. As Christopher Lau's Cornell Law Review article puts it, the "underlying methods are never disclosed and the founders of the technique are never proffered as experts." Harpster and co-author Susan Adams have never released how they obtained the calls behind their COPS Scale, how the research was conducted, or made any data publicly available. You can't cross-examine a dataset no one will show you.

The result, per Lau: the technique "largely escapes Daubert or Frye evidentiary challenge." Collin Derrig's analysis in the University of Cincinnati Law Review reaches the same conclusion, describing how the method fails both Daubert and Frye but gets "sneaked" past expert-testimony standards anyway.

The workaround is lay opinion under FRE 701. Harpster has written ten tips for prosecutors on using call analysis without ever seeking expert qualification under FRE 702, including guidance on walking juries through the audio and transcripts to "create lasting impressions of this critical evidence." The officer who took a weekend seminar then testifies about caller guilt as though it were common observation.

Attack the workaround early and explicitly. Object the moment an officer starts interpreting word choice or phrasing, and argue it's expert opinion in disguise: conclusions drawn from a proprietary methodology the witness cannot explain, validate, or attribute to general experience. Lay opinion under FRE 701 requires rationally derived from perception, and a trained scoring rubric is the opposite of that.

Build the Daubert/Frye record on the gaps. The undisclosed error rate, absent peer review, and opaque data fail the standard factors: testability, known error rate, controlling standards, and general acceptance. Courts have already agreed at least three times:

• An Alabama trial court rejected Harpster as an expert witness in 2009.
• A Nevada judge rejected a detective trained in the technique.
• In Michigan, the conviction of 16-year-old Riley Spitler was overturned because 911 call analysis and similar statement-analysis techniques should never have been admitted.

When you prepare the challenge, work from the actual recording rather than the analyst's transcript. Testing claimed "guilt indicators" against the audio itself, with word-level timestamps and speaker diarization, is where a grounded review workspace like Evidence Copilot earns its keep: every claimed indicator gets checked against a cited, timestamped moment of the call, and every AI output stays flagged for human review before it touches a brief. Pair that transcript work with the same rigor you'd apply to jail call transcription review, because the prosecution's audio evidence rarely survives contact with an accurate record.

The momentum is with the defense on this one. Prosecutors, judges, and NACDL leadership are pushing bans, conviction reviews, and sanctions for prosecutors who used the technique knowing it was inadmissible. Cite the Alabama, Nevada, and Michigan rejections, force the lay-testimony workaround into the open, and make the prosecution defend a method whose own creators won't defend it under oath.

Hearsay Attacks on 911-Call Testimony

When an officer takes the stand and walks the jury through a 911 call, that testimony is often inadmissible hearsay, and in U.S. v. Nelson (6th Cir. 2013) it cost the government a conviction. The Sixth Circuit ruled that police testimony about an anonymous 911 call was hearsay, and reversed because the error was not harmless: it was "more probable than not" that the hearsay materially affected the verdict. That case is your template for 911 call analysis objections at trial.

The most common fight is over the "background" framing. Prosecutors will offer the call as mere context, then have the officer repeat the caller's accusations in detail. Nelson shows the distinction that matters: a minimal statement that "police received a 911 call" avoids the prejudice problem, but once the officer relays what the caller said, it's substantive hearsay dressed up as scene-setting. Object at the first sentence, not after the details land.

Expect prosecutors to invoke the standard exceptions:

• Excited utterance and present sense impression, arguing the caller's distress or contemporaneous narration guarantees reliability.
• Medical diagnosis, where the call contains statements made to obtain emergency treatment.
• Forfeiture by wrongdoing, claiming the defendant caused the caller's unavailability.
• Impeachment, using the call to contradict the defendant's statements.

Test each exception against the actual audio, not the officer's summary. A dispatcher's calming questions or a caller's detached, narrative tone can undercut an excited-utterance claim, which is why working from a timestamped transcript of the recording itself, rather than the state's characterization, matters. That's the review workflow behind 911 call transcript analysis: pin every claimed "guilt indicator" or accusation to the exact second it was spoken, with citations a court can check.

Don't count on curative instructions. In Nelson, jury instructions alone did not cure the hearsay prejudice, and appellate analysis confirmed that a less detailed statement would have avoided the problem entirely. The lesson is preventive: a motion in limine, a standing hearsay objection, and a clear record that the officer's retelling goes beyond permissible background.

The same rigor applies to other recorded statements in your case. When investigators lean on recorded inmate conversations, jail call transcription with speaker diarization and word-level timestamps lets you verify what was actually said before the state paraphrases it for the jury. Grounded review, with every answer cited to the source recording and marked for human review, keeps the challenge where it belongs: on the evidence itself.

Preserve the Audio Before It Disappears

In 911 call analysis cases, the audio you need to rebut the state's expert may already be gone by the time you open the file. Retention windows at some California agencies run as short as six months, and prosecutors rarely volunteer that fact.

According to practitioner reporting, the California Highway Patrol, Orange County, and Riverside each purge 911 recordings at the six-month mark, while Beverly Hills keeps them for a year, Long Beach for three, and LAPD for two years or longer. Your jurisdiction's schedule may differ, and agencies aren't obligated to tell you. Treat every case with a 911 call as if the audio has an expiration date, because it does.

The practical move is a records request on day one, and it should cover more than the recording itself:

• The 911 audio file, in its native format, not a compressed copy
• The call log showing time, duration, and caller number
• The CAD (computer-aided dispatch) log with dispatcher actions and timestamps
• Any transfer or secondary calls to other agencies

Why the logs matter so much: they survive after the audio is deleted. In one case described by trial attorneys, the recording was long gone 18 months after arrest, but the call and CAD logs preserved the caller's phone number. That number identified an eyewitness who exonerated the client. The logs turned a dead end into the defense's best fact.

Expect friction. Roughly half the time, agencies redact identifying information from calls and logs before release. When a public-records request comes back scrubbed, be ready to subpoena. A subpoena gets you the unredacted caller data that matters for witness identification, and it puts the agency on notice not to purge anything else while the case is pending.

Once you have the audio, work from the recording itself. Word-level timestamps and speaker diarization let you test every claimed "guilt indicator" against what the caller actually said and when, which is the grounded way to cross-examine an analyst whose methods were never validated. Evidence Copilot's 911 call transcript analysis supports that review with citations to the exact second of audio, and the same workflow extends to jail call transcription when the state leans on recorded calls elsewhere in the case. Every AI output is marked for human review; the calls you preserved are the ones you control.

Working the Recording Itself: Rebutting the Analyst with Grounded Review

The analyst's report lists "guilt indicators" like a checklist. Your job is to take each claimed indicator and put it against the actual recording, word by word, second by second.

Start with a timestamped, diarized transcript of the call. When the analyst says the caller used a "passive defense" at a particular moment, you need the exact spoken words and the exact timestamp to test that claim. Word-level timestamps turn a subjective impression into a checkable fact. Speaker diarization matters just as much: a dispatcher's leading question and the caller's answer land on different speakers, and conflating them manufactures indicators that aren't in the caller's mouth at all.

The COPS Scale has grown from 16 factors in a 2006 thesis to 38 indicators in a 2017 book, all built on samples of 100 to 200 calls whose underlying data has never been released, according to law-review analysis. That opacity is your cross-exam opening.

Build your cross-exam questions from quoted source moments, not from memory of the audio:

• Tie each question to a quoted line with a timestamp: "You testified the caller showed 'resistance to the dispatcher' at 1:14. The transcript reads '[quote].' That's what was said?"
• Separate speaker turns. Ask whether the claimed indicator sits in the caller's words or the dispatcher's prompt.
• Test timing claims. If the analyst says the caller hesitated before requesting help, pull the word-level timing and show the actual gap.
• Build a parallel timeline: analyst's claimed sequence versus the recording's actual sequence, side by side.
• Flag every indicator that requires inferring mental state from phrasing alone. Those are the ones most vulnerable on cross.

This is where 911 call transcript analysis earns its keep: diarized, word-timestamped transcripts where every answer carries a citation into the exact second of audio, so your cross-exam prep quotes the source rather than paraphrasing it. The same grounded approach applies when the state's theory leans on jail calls; jail call transcription gives you the same word-level footing. Every AI output stays marked for human review, and you make the credibility calls.

Finally, use the 2026 policy momentum. Prosecutors, judges, and defense leaders are now pushing for bans and conviction reviews, and NACDL leadership has publicly grouped 911 call analysis with discredited forensic methods, per reporting on the growing reform push. If you're handling a post-conviction matter where call analysis shaped the investigation, that institutional shift gives you documented support for the challenge.