Case Study 25.2 — Lisa Roberts: When the Tower Lies About the Spot

Sourcing and tone. This case study draws on the documented public record of an Oregon exoneration, as reported by the National Registry of Exonerations and contemporaneous coverage. It is used to teach a single, decisive methodological point from §25.5: that historical cell-site/location data places a phone in a coverage area, not at a spot — and that overstating the difference has sent innocent people to prison. We treat a real death soberly and confine ourselves to documented facts. Lisa Roberts's guilty plea was vacated and the charge dismissed; she is an exoneree (Chapter 34). Specific distance figures are given as reported in the public record; the load-bearing lesson does not depend on their exact values.

Background

In 2002, Jerri Williams was found dead in Kelley Point Park in Portland, Oregon. Lisa Marie Roberts, who had had a turbulent relationship with the victim, became the suspect. Facing the prospect of trial and a potentially much longer sentence, Roberts entered a guilty plea (reported as an Alford-type plea, in which a defendant pleads guilty while maintaining innocence to take a certain sentence over the risk of a worse one) and was sentenced to a long prison term. She served roughly nine and a half years before her conviction was undone — and the thing that undid it is the subject of this chapter.

The case is studied in digital-forensics and wrongful-conviction courses for one reason: the evidence that loomed largest in pushing Roberts toward that plea was cell-phone location data, and that data was understood — by attorneys, and very likely by Roberts herself — to mean something it does not mean.

The forensic evidence and the overstatement

The cell records showed that on the morning of the murder, Roberts's phone connected to a cell tower in the vicinity of Kelley Point Park — by the reported figures, a tower within a few miles (about 3.4 miles) of where the body was found. To a layperson, and to the lawyers in the case, this sounded like the phone — and therefore Roberts — had been at the park. It was, in the atmosphere of the case, powerful placement evidence: her phone "pinged" near the crime scene at the relevant time.

Here is the §25.5 problem, in life rather than in a textbook. Roberts was reportedly more than eight miles away when that call was made, driving — a location supported by a witness account. Her phone had simply been routed through the tower near the park, exactly as Figure 25.2 of this chapter predicts: a phone connects to a tower that gives it usable signal, which is usually but not always the nearest one, and a tower's coverage area can span a large region. The call did not happen at the tower or at the park. The network routed it through that tower while the phone was miles away. The record was real; the interpretation laid on top of it — "the phone was at the park" — was false.

This is the chapter's central caution made tragically concrete. As commentary on the case put it, attorneys, prosecutors, judges, and juries routinely overestimate the precision of cell-phone location records — records that can place a phone within an area of several square miles in a city and, in less congested terrain, far larger. The tower told the truth about which antenna the phone used. It said nothing reliable about the spot. And the gap between those two facts — which this whole chapter exists to teach — was the gap Roberts fell through.

A teaching note on the mechanism. Trace the failure step by step against §25.5, because it is the cleanest real-world illustration the chapter has. (1) The phone connected to a tower near the park. (2) That fact is consistent with the phone being near the park — but it is equally consistent with the phone being anywhere else in the tower's coverage area, including eight miles away on a road. (3) The network's choice of tower depends on signal, congestion, and geometry, not strictly on proximity, so a phone miles away can route through a given tower. (4) The honest reading was therefore "the phone was somewhere in this tower's coverage area," which does not place Roberts at the scene. (5) The reading actually operating in the case was "the phone was at the park" — a pinpoint the data cannot support. Every error here is an error this chapter named in advance.

What the evidence did — and didn't — establish

Strip the case to its forensic logic and the lesson is stark.

• What the cell record actually established: that Roberts's phone connected to a particular tower at a particular time — i.e., that the phone was somewhere in that tower's coverage area. That is the entire honest content of the record.

• What it was taken to establish: that Roberts was at Kelley Point Park. This is the §25.5 overstatement — converting a coverage area into a spot, and converting the phone's location into the person's. Both leaps are unsupported, and both are exactly the leaps the chapter warns against (area vs. point; device vs. person).

• What independent evidence later showed: post-conviction DNA evidence (Chapters 7–9) reportedly pointed to another person at the crime scene — a different suspect entirely. The strong, quantified method (DNA) contradicted the conclusion the weak, overstated method (cell-site pinpointing) had seemed to support. The validity spectrum (Chapters 1, 6) is visible in a single case: the method at the top of the spectrum corrected the error made by relying on a method whose pinpoint claim sits near the bottom.

In 2014, a federal judge vacated Roberts's guilty plea, finding that her trial counsel had been constitutionally ineffective for failing to investigate the cell-tower evidence — that is, for failing to discover and explain that the data did not place her at the park. The prosecution then dismissed the charge. The legal hook was ineffective assistance; the underlying scientific failure was the one this chapter is about.

The lesson

Three lessons, all central to this chapter and to the book:

1. Cell-site data is area, not point — and the difference is not academic. The single most important sentence in §25.5 is that historical cell-site data places a phone in a coverage area, not at a spot. Roberts is what happens when that sentence is ignored: a phone routed through a tower near a park was treated as a person standing in the park, and a likely-innocent woman pleaded guilty in the shadow of that "evidence." The case is the ⚠️ Junk-Science Alert of §25.5 in flesh and blood — a method genuinely useful for placing a phone in a broad area, catastrophically misused to pinpoint.

2. The error is in the interpretation, not the records — so attack the interpretation. The cell records were not forged; they were over-read. As with the dueling-experts lesson elsewhere in this book, the failure here lived in the inferences laid on the data, not the data themselves. A competent defense (Chapter 30) does not deny the records; it forces the witness to concede the size of the coverage area, the fact that phones don't always use the nearest tower, and the consequent impossibility of pinpointing — precisely the questions §25.5 lists. Roberts's first lawyer asked none of them, and that failure became the legal ground for relief.

3. Exclusion over proof, and convergence as the corrective. The case is a near-perfect illustration of the book's first theme. The exclusionary power of DNA — pointing to another person — is what ultimately corrected an error built on the affirmative, overstated cell-site claim. Had the case rested on convergence of independent, honestly weighted evidence (the standard the cold case follows and Chapter 39 demands), the thin cell-site inference could never have carried the weight it did. Roberts shows the cost of letting a single overstated thread stand in for proof — and shows the surest forensic power, exclusion, finally doing its work.

A closing tie to the cold case. The Mill Creek file uses cell-site data the honest way: not to pinpoint Keller at the cabin, but to contradict his alibi — to show his phone in the scene's coverage area when he claimed to be an hour away (§25.5 and the Case File). Roberts is the warning that runs alongside that use. The same data type that legitimately breaks an alibi by placing a phone in an area must never be inflated into placing a person at a spot. The cold case stays on the right side of that line. Lisa Roberts is the record of what happens when a case does not.

Discussion questions

1. The cell record showed Roberts's phone connecting to a tower near the park while she was reportedly eight miles away. Using §25.5 and Figure 25.2, explain mechanically how a phone miles from a tower can route through it — and why this makes "the phone connected to this tower" unable to support "the person was at this spot."

2. Distinguish what the cell record actually established from what it was taken to establish. Identify which two of the chapter's three core distinctions (integrity/truth, device/person, area/point) were violated.

3. Roberts's conviction was vacated for ineffective assistance of counsel — her lawyer failed to investigate the cell-tower evidence. Using §25.5's list of cross-examination questions, write three questions the lawyer should have asked, and state what each would have exposed.

4. Post-conviction DNA pointed to another suspect. Using the validity spectrum (Chapters 1, 6), explain why it is fitting that the method at the top of the spectrum corrected an error produced by a method whose pinpoint claim sits near the bottom. What does this say about which evidence should anchor a case?

5. Contrast the honest use of cell-site data in the cold case (breaking Keller's alibi by placing his phone in the scene's area) with the dishonest use in Roberts (pinpointing a person at a spot). Why is the first defensible and the second not, when both rest on the same kind of record?

6. The CSI effect, both ways (Theme 4). Jurors and even attorneys over-trusted the precision of cell-site evidence in this case. Connect this to §1.2: how does the cultural belief that "the phone places you at the scene" function exactly like the CSI effect's credulity mode — over-trusting weak evidence — rather than its skepticism mode?