How accurate is the 66i for reporting elevation?

As a rule of thumb, vertical error is about 1.5X larger than horizontal error. Horizontal or vertical error depends on the quality of the fix. The unit reports approximate (horizontal) error as a distance (call it x). This value is only good for relative measurement (for instance, for comparing the quality of the fix "now" with the quality of the fix "before I entered the narrow defile"). The actual value of the error distance means nothing. The meaning of this number is roughly "the difference between the reported position and the actual position is x with probability p". No GPS vendor discloses the value of p which they use. But it's something on the order of 95%.

My personal take is that barometric altimeter is the better choice provided you manually calibrate it at least once a day. As long as weather-related barometric pressure is not changing rapidly, this setup should be good enough (for me - maybe not for a paraglider). Especially for short activities, I would leave auto-calibration off. No point in dragging GPS altitude into it when we have an accurate barometric calibration. If the weather is highly variable, consider continuous automatic calibration. But now you've dragged in the somewhat inaccurate GPS altitude.

Thanks for that.  How about this....

My old paragliding GPS receiver (the Garmin 76CSx) with WAAS/EGNOS corrections enabled should manage horizontal accuracy of  ±3m under an open sky. When I took it for a trip round town it really did that well away from buildings. I was impressed. www8.garmin.com/.../waas.html

If vertical errors are roughly 1.5x the horizontal, that makes for  ± 4.5 m vertical error. That's very good.

Tomorrow looks like a reasonably stable day and the weather forecast shows sea level pressure going from 1030mb at 0900 to 1035mb at 1800. A barometric altimeter set at 0900 would be under-reading by 50m at 1800. Not so good.

I'd expect a dual-band receiver to do at least as well as the 76CSx but maybe more reliably. Is the barometric altimeter such a good idea?

Thoughts?  I'm struggling to find anything definitive on this. When flying my paraglider I will be under a completely open sky if I'm worrying about infringing airspace. So any GNSS receiver will be benefitting from near ideal receiving conditions.

I doubt that you're going to find anything definitive. The problem is that there are lots of variables no matter which approach you take. 

GPS is subject to the qualify of the fix. Even with open sky, it's going to vary some based on the constellation overhead at any given time. 1.5x for vertical error is a rule of thumb, not reality. Some sources put the rule of thumb at 1.7x. Then you have ionospheric disturbances, which are partially compensated with WAAS/EGNOS (depending on your location). Local variations of the ellipsoid vs. reality. And so on.

As you point out, changes in atmospheric pressure introduce errors in pressure-derived altitudes. 

This is why a lot of users engage auto-correction when using pressure-derived elevation. And why some users manually calibrate frequently. And why some users opt in favor of GPS-derived altitude. All depends on your use case, conditions, and preferences.

That said, aircraft altimeters have always been based on pressure, not GPS. Must be a reason.

In Googling about, ran across this article regarding your specific sport. Pretty old, but still has an interesting perspective. Probably makes more sense to you than it does to the rest of us.

xcmag.com/.../

I know that article well. It's very good but old. I'm keen to know what the up-to-date info is. I'll keep digging. Thanks for Googling on my behalf, appreciated :-)

I will answer that at two ends of the spectrum, elapsed time.

Barometric:  Measuring a 50 foot hill by taking a reading at the top and another at the bottom 2 minutes later and subtracting it.  I did this where surveyors had done the same and left sticks in the ground with there elevations.  My height was very consistent with theirs. 

GPS:  Taking readings during a one week, or greater, period when no calibrations of barometric are available.

Basis:  About a dozen years ago I performed a study to determine the precisions of the methods by gathering a statistically significant number of data points, elevations, for each.  For the pure barometric, I took 20 readings at 5 second intervals to eliminate weather impacts.  For GPS and weather affected barometric, I took 3 per day, at least 3 hours apart,  for a week without barometric calibration.  The standard deviation of the weather impacted barometric elevations was 12 times greater than that of the GPS elevations while that of the GPS was 6 times greater than that of the pure barometric elevations.

Note that I assumed no weather impact on the barometric readings gathered in 100 seconds.  Also, the lessor standard deviation indicates greater precision, or lessor variation.

I do not disagree with twopert's post.