Precision and accuracy discussion

http://ee.stlcc.info/130/accuracy.htm

That's a pretty good article.

Your friends are correct in the abstract, but you are correct in practical terms. Engineers tend to think in abstract terms, which is great for R & D but terrible for race timing, where stupid practical bulls*%#t trips more people up than anything. Example: Tissot falsely claimed a World 100m dash record for Justin Gatlin in 2006 because a Tissot technician had one setting wrong in their software. The result was truncated instead of rounded, and the world record was later nullified by the IAAF.

https://www.lindyssports.com/content.php?id=175

http://news.bbc.co.uk/sport2/hi/athletics/4989558.stm

Your friends are correct in that - in theory, and for example - if you've got a time base measuring autocrossers' starts and finishes perfectly and consistently, except the time base is precisely x.xxxxx second slow per minute, and all time bases used have the identical flaw, then you do not have a problem (except for absolute course records). But in practice, failures rarely present themselves in such perfect terms.

The fact that you're out in the field timing events and they're in a cubicle somewhere comparing pocket protectors makes the two viewpoints in this discussion very predictable.

The argument was made that "as long as it's repeatable it's fine". Well I for one am not fine with a repeatable 0.25 second error. Just what is the accuracy of their timing device, claiming to be accurate to the thousandth? As Jenna said on the phone, if it's not precise then it's probably not accurate either.

I am working on the club letting me time with an 8001 to check the accuracy.

Thank you Jim.

Sorry for the n00b questions... I'm finding myself more and more confused on the topic, and had some questions for the timing elite:

Stability:

Say you know that your reference clock is only stable within a certain range, how do you properly convert that to a measure of accuracy and/or precision?

With the knowledge that your stability is limited, do you still use the maximum amount of information from your reference for computations?

Accuracy:

Say you know your reference clock is 'so' precise, but 'so' inaccurate, are there any accepted methods for reporting or automatic compensation?

If you can reliably determine a reference clock's inacccuracy by long-term comparisons with another reference (eg. ntp), is it ok to correct your reference time values by a correction factor? If so, do you apply the correction to time of day values, or to net times extracted from time of day values?

When your accuracy is dependant on the duration of measurements, how do you report to officials something like : "any time over 1 minute, will only be accurate to the tenth of a second"?

Any info or discussion greatly appreciated

The way to determine the accuracy of your clock is to compare it to other clocks which compare to known, accurate clocks, through testing. I think you get that.

NO, it is not ok to "correct" a time from a clock either TOD or Net time. How do you know, at any given instance, if your clock was less accurate or "drifted" by a certain amount even if it drifted over a long period. That inference can't be made.

Remember, also environmental and power issues can impact the drift of a clock at any given point in time, so you can never know, over a minute, what a drift was, even if you know, it was X over 24 hours.

The easy answer is:
1. Buy HIGH quality clocks, with known accuracy ranges. Currently, the major manufacturers are producing timing equipment accurate to 1/125,000 PER HOUR!! Times to the 1/1000 of a second are possible, with the right input devices.

If you have a clock you want to use, and know it may be drifting, get it repaired. This is a good reason to always run multiple clocks and compare them. If one starts to drift, you will know, thought possibly not which clock. You could then compare each clock individually against a third, or fourth clock and figure out where your issue is.

Thanks for that, good info. Only one minor qualm (devil's advocate):

quote:

issues can impact the drift of a clock at any given point in time, so you can never know, over a minute, what a drift was, even if you know, it was X over 24 hours

Doesn't that also apply if you are assuming your drift is zero (normal case)? So how is correcting your clock by a factor of 1.0 much different to correcting it by eg 1.00001345?

If you are assuming error and constantly measuring it, isn't that in fact better than assuming no error and periodically checking the assumption?

Seriously though, thanks for the info - the equipment I'm using probably just needs a good service to shut me up.

Of course we're checking for deviation, though I wouldn't call that error! We check for it, but can't do anything about it, during the race as you wouldn't know which clock is the one that's off. If you did know, why would you be using that clock? If I had a clock that I believed was deviating, I would 1. Pray, 2, hope I didn't need to use it, 3. wish it was the same deviation for everyone.
Seriously, it depends on the event and the measurement precision. At a MB XC, we use clocks capable of timing to at least 1/1000 of a second, but only time to the 1/10 as the input isn't that accurate. At a DH mtb event when we use photocells, we publish times to the 1/100 and time out to the 1/1000 and see deviation at most to a few thousands over 4-5 hours.

Timing Auto-x using 3 timers. One is TAG 520 or 540. Two is Race America & three is a TAG 520 or 540 piggy backed onto one to run the scoreboard (TAG cponfigured Race America Scorboard that won't work with AXware). Usually no issues. No drift. My time diff between clocks is caused by height of car bumper since Tag eys are on 8" tirpods and Race America eyes are on the ground. Using FIS EET math if I need to go to timer 2.( has not happened yet.)