What 911 Call Analysis Is and Why It Keeps Showing Up in Your Cases

A detective listens to a 911 call and tells the jury the caller "didn't sound right." That single line of testimony can carry a case. It's also junk science wearing a badge.

911 call analysis traces back to Tracy Harpster, a former Ohio deputy police chief who claims you can detect guilt from a caller's word choice and cadence. His method runs on a proprietary scoring system called the COPS Scale, now 38 indicators across six categories, up from 16 in his original thesis.

ProPublica's investigation documented more than 100 cases across 26 states where the method played a role in arrests, prosecutions, and convictions. The reporters note that's likely a fraction of the real number.

Here's the part that should worry you: most defenders don't know it's in play. A Virginia public defender told reporters he'd "never heard of any of that claptrap" in his jurisdiction. North Carolina's Office of Indigent Defense Services found the pattern widespread enough to issue a statewide warning to every defense attorney in the state.

The "guilt indicators" themselves read like parody. Harpster's materials flag the following as suspicious:

• Saying "Huh?" to the dispatcher, or asking "God, who would do this?"
• An isolated "please," or saying "I'm sorry" or "thank you"
• Saying "blood" instead of "bleeding," or saying "somebody"
• Self-interruption mid-sentence

Real people pay for these inferences. In a Washington case, a mother's first word, "hi," got flagged because "being too polite indicates that the caller may be hiding something." Her baby's death was reclassified as homicide and she took an Alford plea to manslaughter. In Michigan, a caller got tagged as a suspect for saying "I need help."

The science collapsed too. Two 2025 studies, one from the FBI's own Behavioral Analysis Unit, found the COPS Scale can't separate guilty from innocent callers. Sworn officers misclassified calls over 40% of the time, and the FBI authors recommended the scale "should not be used in actual investigations."

Harpster disputes this, writing that he's "unaware of any case in which an innocent person was convicted using this analysis" and claiming it has exonerated people. Co-author Susan Adams says no single indicator proves guilt and the method belongs "in combination with case facts." Both positions conflict with the documented record in Faria and Burton.

The reason this keeps landing in your cases is the workaround. The method rarely survives Daubert or Frye, so prosecutors introduce it through lay testimony: detectives and dispatchers narrating what the caller's tone "indicated." Harpster handed prosecutors ten tips for doing exactly this without seeking expert qualification. Missouri prosecutor Leah Askey admitted in writing to disguising the method in court to avoid admissibility hearings.

You can't challenge what you can't see. Pull the 911 audio and work the transcript yourself, with word-level timestamps so every characterization you make is citable to the exact second. That's the workflow behind 911 call transcript analysis, and the same discipline applies when the state mines custody calls for similar behavioral tells (jail call transcription).

Screen every case with a 911 call. Check discovery for "Harpster," "COPS Scale," "statement analysis," or any detective offering behavioral readouts of the caller. If you find it, you have a fight worth having.

The Science Has Failed: Five Studies, 40% Misclassification, and the FBI's Own Rejection

If the FBI's own Behavioral Analysis Unit says the technique doesn't work, you'd expect prosecutors to stop using it. They haven't. The scientific record against 911 call analysis is now deep enough that any Daubert or Frye challenge should start with it.

Two 2025 studies published in Psychology, Public Policy, and Law delivered the hardest blow. A study by the FBI's Behavioral Analysis Unit found sworn officers misclassified calls over 40% of the time using the COPS Scale, and concluded the scale "should not be used in actual investigations to evaluate callers' potential innocence or guilt." The Villanova replication found the scale "failed to differentiate between guilty and innocent callers," with core assumptions (deceptive callers avoid urgency, are unemotional) not holding up. In total, five independent studies from Villanova, James Madison, and the FBI contradicted Harpster's findings; the Innocence Project counts at least eight replication attempts that failed.

The structural weaknesses matter as much as the replication failures. When you brief the court, lay out the foundation:

• The original thesis was exploratory, built on a sample of just 100 calls (50 guilty, 50 not guilty) from Harpster's hometown of Moraine, Ohio, a city of 6,500 averaging less than one murder per year.
• The scale grew from 16 indicators in the thesis to 38 today, without published validation of the expanded version.
• There is no peer review of the underlying method and no public dataset other researchers can test.
• The FBI's 2020 study warned the method "may exacerbate bias," and the University of Michigan's STPP policy brief tied it to investigative confirmation bias and wrongful convictions.

The flat-affect research gives you a counter-narrative for trial. Dr. Jessica Salerno's studies found that suspicion tracks low emotional display, not guilt: with identical scripts, both civilians and police became significantly more suspicious of low-emotion callers, and the pattern held with real 911 calls regardless of whether the caller was later convicted or exonerated. That explains cases like Huwe Burton, who spent 20+ years in prison after testimony that he "did not seem emotional enough," and Robert Roberson, whose autism diagnosis years later explained the "cold" behavior authorities read as guilt.

For balance, expect Harpster's rebuttal. He denies any innocent person has been convicted using the analysis, claims it has exonerated people, and says he has examined over 2,500 equivocal death cases. Co-author Susan Adams maintains no single indicator determines guilt. Those claims sit uneasily beside the documented convictions and the failed replications, and a court weighing admissibility should see both.

When you're pulling the call audio and transcript together for a motion or hearing, a workspace like Evidence Copilot keeps every quote tied to a timestamp, so your characterization of the call is verifiable against the original.

The Backdoor: How Prosecutors Sneak Call Analysis Past Daubert Through Lay Testimony

If your case has a 911 call and a detective who wants to testify about the caller's "tone," assume 911 call analysis is in play, even though no expert witness is listed. Since the method rarely survives Daubert or Frye scrutiny, prosecutors increasingly route its conclusions through lay witnesses under FRE 701, and the practice is documented, not hypothetical.

The paper trail is unusually candid. According to a University of Cincinnati Law Review analysis, Harpster himself provided prosecutors ten tips for using the analysis without seeking expert qualification, encouraging them to "guide juries through both the audio and transcript of the calls." In Missouri, prosecutor Leah Askey admitted in writing to "skirting legal rules," presenting the method in court without calling it "science" to avoid an admissibility hearing. Some prosecutors have acknowledged the method "will not hold up to scientific scrutiny," so, as one put it, "they get creative."

The pattern to watch for in discovery and on witness lists looks like this:

• A detective or dispatcher, never a qualified expert, is slated to testify about the caller's word choice, cadence, or emotional state.
• Discovery reports describe "behavioral observations" or "statement analysis" of the 911 call without naming Harpster or the COPS Scale.
• The testimony tracks the method's documented indicators: an isolated "please," saying "I'm sorry," asking "God, who would do this?", or sounding insufficiently emotional.
• Training records show attendance at Harpster's $3,500 eight-hour class, often taxpayer-funded.

When you spot it, move to exclude under FRE 701/702. Lay opinion must be rationally based on the witness's perception and helpful to the jury; a detective reciting COPS Scale indicators is offering disguised expert opinion, and the ProPublica investigation gives you the record to argue that. Dr. Jessica Salerno's research cuts the other way too: suspicion correlates with low emotional display, not guilt, in both civilians and officers.

Your best affirmative move is working the call yourself. A timestamped, diarized transcript lets you walk the jury through the full context and counter cherry-picked "indicators," and the same discipline applies to jail calls when prosecutors mine custody audio for similar inferences. Evidence Copilot's grounded review workspace produces word-level timestamps with citations to the exact moment, so every defense characterization of the call is verifiable in court.

Your Four-Step Defense Workflow: Screen, Challenge, and Work the Call Yourself

If a 911 call is in your discovery, assume someone on the other side has run it through the COPS Scale until you've proven otherwise. The ProPublica investigation found defense attorneys in dozens of jurisdictions had never heard of the technique, with one comparing it to "medieval trials by fire." Screening is step one.

Step 1: Screen every case. Check the discovery dump for mentions of Harpster, the COPS Scale, or "statement analysis," and flag any detective or dispatcher offering behavioral observations about the caller. That last pattern matters most: since the method rarely survives admissibility scrutiny as expert testimony, the documented route into court is lay testimony. North Carolina's Office of Indigent Defense Services issued a statewide warning to defense attorneys for exactly this reason, and it's a sensible model for any office-level intake protocol.

Step 2: Build the admissibility challenge from the 2025 studies. Two papers in Psychology, Public Policy, and Law found the scale scientifically invalid, including one from the FBI's own Behavioral Analysis Unit. Sworn officers misclassified calls over 40% of the time, and the FBI authors concluded the scale "should not be used in actual investigations." Pair that with the structural weaknesses: a 100-call exploratory thesis sample, no public data, and 38 indicators that have drifted upward from the original 16.

Step 3: Attack the lay-testimony workaround. When a detective testifies that the caller's tone or word choice "indicated" guilt, move to exclude under FRE 701. The record here is rich: Harpster distributed ten tips for using the analysis without seeking expert qualification, and Missouri prosecutor Leah Askey admitted in writing to "skirting legal rules" so the method wouldn't be called science. The University of Cincinnati Law Review documents this pattern in detail, and it makes potent cross material.

Step 4: Work the call yourself. The strongest affirmative tool is a timestamped, diarized transcript you control. When the state cherry-picks an isolated "please" or a "Huh?" as a guilt indicator, you walk the jury through the full context, second by second, with every characterization citable to the exact moment. This is where grounded 911 call transcript analysis fits the defense workflow: upload the call, get word-level timestamps and diarized speakers, and counter selective quoting with the whole record. Extend the same review to jail call transcription, because prosecutors mine custody calls for behavioral inferences the same way.

Dr. Jessica Salerno's research gives you the closing argument: suspicion tracks low emotional display, not guilt, and the pattern held whether callers were later convicted or exonerated. Don't let the state's theory of the call go unanswered.