4 out of 6 HGN

The 2015 NHTSA manual claims that if an officer observes 4 clues he will be able to classify the subject's BAC as greater than .08 with 88% accuracy.  I found this statement troubling because in two recent cases it was relied upon by prosecutors and judges to find 4 clues sufficient to support a correct "arrest decision."  So I dove into this issue deeper than the sentence or two reported in the recent NHTSA Manuals.

The NHTSA funded and published HGN Robustness Study of 2007 was the first study to examine this test performed by expert practitioners, under laboratory conditions, and where the BAC was known.  This stated that HGN clues are appropriate where the BAC is as follows: 
< .03 = 0 - 2 Clues
.03 - .059 = 2 - 4 Clues
> .06 = 4 - 6 Clues

In fact, the data from this study shows that 67% of people under the legal limit would be incorrectly arrested by this criteria.  While this issue should be clear after being published in a study funded by NHTSA in 2007, NHTSA still stated in the 2013 and 2015 manuals that the presence of 4 clues is 88% conclusive that the subject has a BAC greater than .08.  

According to NHTSA's own study, however, NHTSA should only state the following:
4 Clues indicates a BAC only greater than a .03
6 Clues indicates a BAC only greater than a .06

Unfortunately, the ramification of such a publication would be that the HGN test would not support an arrest decision at the legal limit of .08.

It is true that if you have 100 people and 88 are clearly intoxicated and an officer has at least 4 clues on those 88 subjects alongside an additional 12 people who are objectively sober, then NHTSA can claim an 88% accuracy of the test but it is still bad science based on the data set.  All of NHTSA's "validation studies" have traditionally been loaded with intoxicated people.  In the San Diego study, the study NHTSA instructs officers to rely on the most, 72% of the people were over the legal limit.  This means that even if the officers decided to arrest every single person, the officers would make more correct arrest decisions at a rate of 72% and the tests would appear to be effective.  None of the close to 300 people tested in that study had an alcohol level of less than .03, 99.3% had a BAC greater than .04, and 72% were over .08.

Chuck Hayes of the IACP who is listed on the Acknowledgements page of the last two NHTSA Manuals presented a webinar on the process and changes that went into the 2013 manual for the Law Enforcement Agencies of the State of Montana.  In that webinar, around the 28-minute mark, he spoke regarding the San Diego Study from 1998.  He said the following:

"And then it was really kind of the very first of the research studies that successfully evaluated the accuracy of the SFST battery identifying BAC levels below .10 and so they identified that the validity of the SFST’s for both .08 and also for .10 but one of the things that that study did, if you read it and read over it well, you will see that it also was helpful in identifying those blood alcohol levels maybe at a .04, .05 in that area.
And so for many years, we’ve always said that once you see two clues of nystagmus, so that lack of smooth pursuit and then you see distinct and sustained nystagmus at maximum deviation, well, we’ve said and believed for a long time, because of the research, that that is a good indicator of blood alcohol level at around a .04 - .05.  Well the San Diego study validated that for us because that’s what they actually saw in those low blood alcohol levels, so 4 clues now indicates that it could be a blood alcohol level at a .04 - .05 in that area.  So the San Diego study helped with that as well and helped clarify that for us."

The San Diego study doesn't report the statistics the way Mr. Hayes does, but the Robustness Study supports his position.  In the San Diego study, the data was not published, but I assume Mr. Hayes was able to review it.  Despite evidence to the contrary, the study puts in its abstract:
2 Clues indicates a BAC > .04
4 Clues indicates a BAC > .08

If the data had been provided, I believe it would not support the statements in the abstract.  Take for instance even the title of the San Diego study:
"Validation of the SFST Battery at BAC's Below .10"
The abstract itself says the average BAC of the subjects tested was .117-.122.  It is impossible to validate SFST's below a .10 if the majority of your subjects are over a .10.

Despite what the manual states, the available information does not indicate that a subject with 4 or even 6 clues on the HGN has a BAC in excess of .08.