The Court’s analysis focused heavily on the methodological and scientific deficiencies of the three reports Dr. Citek used to support his opinions. None of the studies was designed or executed to validate HGN as a standalone scientific instrument for measuring BAC or impairment—the precise purpose for which the State sought to use the expert testimony.

The Colorado Report (Burns and Anderson, 1995)
This study examined officers’ arrest and release decisions using the full three-test SFST battery on 305 motorists stopped for suspicion of DUI across seven Colorado law enforcement agencies. Only 234 cases with confirmed BAC measurements were ultimately analyzed.

Key shortcomings identified by the Court include:

• The study never isolated HGN performance. It evaluated the combined battery for probable-cause arrest decisions, not HGN as an independent predictor of BAC or impairment. Dr. Citek himself conceded this point.
• The subject population was heavily skewed toward severe intoxication. The mean BAC of arrested drivers was 0.152%, with 78.6% of confirmed specimens above the legal limit (then 0.05% in Colorado). Borderline cases near relevant thresholds were largely absent.
• Officers’ overall decision accuracy was only 86% (93% for arrests, but just 64% for releases). Fully 36% of released drivers should have been arrested according to the study’s own BAC criterion.
• No test-retest reliability data was generated. The report expressly acknowledged that it “did not support an examination of test-retest reliability.”
• Demographic imbalances were significant: 82% male subjects and only 26 drivers over age 50. Gender disparities appeared in error rates.
• The study did not address whether drivers at lower BACs were actually impaired—an issue the report itself flagged as separate and unexamined.
• Trained observers were present for only 41% of subjects, limiting independent verification.

The Florida Report (Burns and Dioquino, 1997)
This Pinellas County study analyzed 313 usable records (from an initial 379) of drivers asked to perform the SFST after roadside detention for suspected impairment. BAC data was available for 256 drivers.

The Court highlighted these deficiencies:

• Again, HGN was never isolated from the full SFST battery or from other non-standardized roadside tests that officers administered on 53 occasions.
• The population was heavily intoxicated: approximately 80% of drivers with measured BACs were above the legal limit, and the mean BAC of the 206 arrested drivers was 0.150% (median 0.147%). Thirty-seven drivers registered BACs between 0.200% and 0.284%—levels at which impairment is obvious without specialized testing.
• Observers monitored only 64% of stops. Portable breath testing devices (PBTs) were used for some released drivers, and the evidentiary instrument was never clearly identified.
• Accuracy figures (approximately 95% correct arrests and 82% correct releases among the analyzed cases) were obtained on a non-random sample of drivers already showing evidence of impairment. The study therefore measured officers’ ability to confirm obvious intoxication rather than the diagnostic power of HGN near the legal threshold.
• Like the Colorado study, it provided no test-retest reliability analysis and did not control for confounding variables such as medical conditions or individual tolerance.

The San Diego Report (Stuster and Burns, 1998)
This is the study most frequently cited to support HGN use at the 0.08% threshold. Officers administered the SFST battery to 297 motorists and used PBTs to obtain BACs for all subjects, including those released. The mean BAC across the sample was 0.122%, with roughly 74% of subjects over 0.08%.

The Court’s critique was especially pointed:

• The report evaluated the full SFST battery (plus observations of appearance and speech) for arrest/release accuracy at 0.08% and 0.04% thresholds. It expressly recommended that “all components of the SFST battery should be administered when possible or practical.” It was never designed to validate HGN as a standalone test.
• Although HGN showed the highest individual correlation with measured BAC (r = 0.65) and the combined battery reached r = 0.69, the Court rejected the notion that correlation equals scientific validation.

The Court emphasized the oft quoted scientific principle that correlation does not equal causation. 

“Nor does the San Diego Report’s correlation analysis impart validity to the study. Under Daubert, correlation is not validation. A validation study must establish that HGN reliably predicts impairment or BAC—or that BAC is above or below 0.08%—not merely that HGN results are associated with BAC. A statistical association between HGN and BAC does not establish that HGN is a scientifically valid method for estimating BAC or determining impairment.” [emphasis added]

• Officers carried PBTs throughout the study—directly contrary to the integrity safeguards emphasized in the Colorado and Florida reports. This access undermined the independence of the arrest decisions being measured.
• Accuracy for estimating BAC dropped sharply at lower levels (only 52% correct for drivers below 0.04%).
• No test-retest reliability was measured, and inter-rater reliability was described as “impossible to measure” under the study design. The authors fell back on a 1981 study with only a 66% reliability rate.
• The sample remained skewed toward higher BACs, rendering claims of reliable discrimination at the 0.08% threshold statistically weak. With approximately three-quarters of subjects already over the limit, high “correct arrest” rates (90%) could be achieved even with limited diagnostic precision.
• None of the three core reports underwent traditional scientific peer review (although a 2006 summary article by one San Diego author was later published).

Across all three studies, the Court found a fundamental mismatch: the reports examined whether trained officers could make probable-cause arrest decisions using a multi-test battery plus observations. They did not—and could not—establish that HGN alone is a reliable, valid scientific method for proving BAC level or impairment beyond a reasonable doubt at trial. Confounding variables (other tests, odor of alcohol, appearance, admissions, high baseline intoxication rates) made isolation of HGN’s predictive power impossible. Demographic imbalances, lack of random sampling, absence of controlled laboratory protocols, and failure to account for alternative causes of nystagmus further eroded reliability.

The Court acknowledged the practical challenges officers face and the importance of public safety, but emphasized that Daubert and due process demand the same scientific rigor applied in other contexts. A properly designed, controlled study isolating HGN across a full spectrum of BAC levels with test-retest protocols has simply never been presented.