Some Preliminary Accuracy Results with the Bulova Precisionist Claremont Model

[minidriver]

I have been wearing my Bulova  Precisionist Clarement  now for a little over 4 weeks and decided to do a careful timing of its accuracy over that time in this "wearing" condition.  This accuracy-timing project is planned to go for 8 weeks, but it seemed like a  good idea to show what we've found so far.  I teamed up with South Pender on this project.  I did all the timing (once a week) and then sent the results to South Pender, who did some calculations and set up the graphs.

Here are some of the details. 

Timing Method: The Stopwatch Method.  For all the details on this method of assessing watch accuracy, see:

http://forums.watchuseek.com/f9/methods-determining-accuracy-watch-382752.html

We used 40 timing trials at each of the four time points in the project (using the NIST reference clock).  That is, we took 40 separate timings and then computed their average and standard deviation.  Therefore, the data points in the graphs below are these 40-trial averages.   This number of trials means that the error bandwidth is about .039 sec. at each time point.  That is, in the second graph below (drift from baseline in seconds), the numbers should be understood as XX ± .039.  As an example, the value .868 seconds given for Week 4 should be understood as having a margin of error of .039 seconds, so that the true, exact value could lie anywhere between .829 and .907 seconds.

Conditions:   Ambient (air) temperature 74 - 78 degrees F.  This, combined with the approximately 91 deg. F temperature of the surface of the skin means that, while being worn, the watch was experiencing temperatures lying between something like 84 and 88 deg. F.  The watch was worn for approximately 12 hours a day, so that for about half the time, the watch was experiencing lower temperatures--probably somewhere in the 73- 76 range.  However, this is a "while worn" project, and the 12 hours per day of wearing coincides with the suggested wearing time by Citizen for its Chronomaster and Seiko for its Grand Seiko quartz models for these watches to perform up to their specifications.

Results:  First the drift in seconds per year (or spy):
 


The way to understand the numbers and the graph is to see each number as our best estimate of the spy performance of the watch after 1 week, after 2 weeks, after 3 weeks, and after 4 weeks.  Clearly, the latest estimate is the best because a longer time period has elapsed, making that estimate the most stable.  This graph illustrates the flaws in estimating drift over shorter intervals, like one week or two weeks.  By four weeks into this project, we see that our best estimate of its accuracy is a little over 11 spy--very good for a non-thermocompensated watch.  (To be more precise, our obtained spy value of 11.32 spy has a margin of error that extends from 10.81 spy and 11.82 spy.)  But keep in mind that this is a preliminary result--after only  4 weeks--and we'll have a more definitive result after another 4 weeks. 

The next two graphs give the results from a slightly different perspective.  The first of these gives the drift from baseline results in seconds, instead of spy.  These are the results that came directly out of the timing data.  We converted these results to the spy metric in the graph above.  So, just as an example, we see that after Week 4, the watch had gained .868 seconds over that period.  That value prorates to a figure of 11.32 spy, as seen in the first graph above, where the results are given in the seconds per year form.



The last graph gives the week-to-week fluctuation in rate from one week to the next.  We see that, over the first week, the rate gain was .169 seconds, whereas over the second week, it jumped up to .74 seconds, decreasing to a loss of .255 seconds between Week 2 and Week 3, and a gain of .214 seconds between Week 3 and 4.  This fluctuation indicates a certain degree of instability in the rate, these being rather large fluctuations.  This week-to-week rate fluctuation would be far less in a thermocompensated movement.



Still, the results look pretty good at this point.  We'll continue this project to Week 8, at which time we'll have much more stable results because of the longer time period over which the calculations will have been performed.  At that point, we'll probably put the watch into storage and do a timing experiment on its performance when off the wrist.

We should note a couple of things.  First, the results above are those that were obtained when the watch was actually worn, and, therefore, are not based on static temperature tests that have been used to guesstimate the wearing condition.  Second, the results are over a reasonable (although still pretty short) period of time.  It is evident from the graphs that had we tried to estimate performance on the basis of a single week timing period, our results would have been very different.  For example, if we'd taken the results for Week 1, we would have estimated 8.83 spy, but had we instead used the results for Week 2, our estimate would have been 23.73 spy.  The value calculated after the end of Week 4, therefore, is our best estimate, at this 4-week point, in the project: 11.32 spy.

These results have probably raised a lot of questions.   South Pender and I will be happy to answer these.  My apologies for this overlong post.  It just seemed necessary to provide as much detail as possible to help everyone understand the graphs and the meaning of the results.
 
 

[brett kenny]

interesting to note that there was quite a deviation in the 2nd week when the given conditions being worn for 12hrs would provide as high a real life stability in temperature, short of a climate controlled environment, as possible?
 
still, good to see the prelim results so close to rated spec!
 
thanks for the effort fellas, look forward to more definitive results

[minidriver]

And hence, why we would want to continue these observations for another 4 weeks. It will be interesting to see if another instance of the rather severe deviation shown in Week #2 would manifest again at some point for the remainder of the timing tests.


Still, pretty solid performance by a non-thermocompensated Citizen movement that retails for $350 and change.

[Time]

Thank you and South Pender for taking the time to do this and sharing the results.

[minidriver]

Thank you and South Pender for taking the time to do this and sharing the results.

You are welcome, Time


We also plan to perform similar long term timing tests on my Citizen Chronomaster, Seiko Ananta Spring Drive GMT and others....

[ADB]

Frank,
Two simple questions:

• Did you reset the time to the atomic clock signal during these four weeks or not?
• If not, what is the accumulated deviation from atomic time over four weeks?

Thanks,
Cheers,

[minidriver]

Frank,
Two simple questions:

• Did you reset the time to the atomic clock signal during these four weeks or not?
• If not, what is the accumulated deviation from atomic time over four weeks?

Thanks,
Cheers,

Andrew:

See Southpender's answer folllowing....



Hacking of the watch was indeed performed at the end of week #3 to adjust for non-DST time.

[SouthPender]

Frank,
Two simple questions:
       

• Did you reset the time to the atomic clock signal during these four weeks or not?
• If not, what is the accumulated deviation from atomic time over four weeks?[/l][/l]

Perhaps I can answer that.  At Time 0, a baseline values was established that was the deviation from the atomic clock.  In other words, when the experiment began, the watch was sync'd as closely as possible to atomic time, but the discrepancy from perfect time was noted and formed the baseline measurement for the tests.  After that no further adjustment was made to the watch.  At each weekly time point, the deviation from the atomic clock was recorded (via the average of 40 readings) and compared to the baseline deviation.  Thus, the accumulated deviation from atomic time over the four weeks is .868 seconds.
 
Edit: That's  right, minidriver, we did have to resync because of the time change back to standard time.  I'd forgotten about that.  This new baseline was incorporated into the results.  That is, the measurements after that time were referenced to this new baseline.  But that changes nothing.  The elapsed deviation from atomic time over the 4 weeks was .868 seconds.
[/list]

[nhoJ]

Exceptional real life performance.  How does this compare to a non thermocompensated Seiko movement?

Send it to me for the winter and I'm sure your delta for the year will tighten up  a bit.

[minidriver]

Exceptional real life performance.  How does this compare to a non thermocompensated Seiko movement?

Send it to me for the winter and I'm sure your delta for the year will tighten up  a bit.

John, I no longer have my 8F35 Prospex diver, but seems to me the performance of the Precisionist movement is more stable (Well to a point, minus the "Spikes") than that of the 196Khz 8F seiko quartzes. I kept my 8F35 for around 3-4 months, worn sporadically (Mostly on weekends) and in all that time the watch was no more than +2 seconds over atomic time.


Still, thermocompensation is a key aspect on the overall accuracy and stability of the movement over a long period of time. So far, looks like the 3-pronged design of the Bulova Precisionist may indeed be a breakthrough in the world of non-TC affordable high accuracy watches.

[SouthPender]

Exceptional real life performance.  How does this compare to a non thermocompensated Seiko movement?

My sense of it, from reading a number of accuracy reports involving the various Seiko 8F movements, is that, if this accuracy level for the Precisionist holds up over a longer time period, it will be superior to that obtained with the high-frequency 8F movements.  Their specs are ±20 spy, and, from most reports I've read, most don't meet these specs.  Still, there are exceptions that do better than those specs.

[Time]

I have an 8F56 that gains 1 second EVERY week at room temperature. My other 8F's right next to it gain a second every couple of months...

[ADB]

I have an 8F56 that gains 1 second EVERY week at room temperature. My other 8F's right next to it gain a second every couple of months...

Time, that's exactly one of the points I was going to make: two watches is too small a sample to say anything about a movement in general.
If we extrapolate from your first 8F56 watch, the 8F56 movement is a good, but unexceptional +/- 1 minute per year movement.
On the other hand, if we extrapolate from your second watch, the 8F56 is a Chronomaster-class +/- 6s per year movement.
And statistically speaking, averaging the results of your two watches yields a questionable number with too much uncertainty.
There is only one way to verify the accuracy of any given type of quartz movement and that's by statistical sampling at the factory where one can collect a sufficiently large sample i.e. we as consumers have to rely on figures provided by the manufacturers.
Cheers,

[Catalin]

...
There is only one way to verify the accuracy of any given type of quartz movement and that's by statistical sampling at the factory where one can collect a sufficiently large sample i.e. we as consumers have to rely on figures provided by the manufacturers.
Cheers,

Maybe, but there is one extra thing - the accuracy of quartz movements changes (more or less) with AGE - usually that is hard to see on ordinary 15-20 s/month quartz, but is more obvious in HEQ. So the accuracy 'out of factory' is still relevant, but does not tell you the entire story!

[Catalin]

...
We should note a couple of things.  First, the results above are those that were obtained when the watch was actually worn, and, therefore, are not based on static temperature tests that have been used to guesstimate the wearing condition.  Second, the results are over a reasonable (although still pretty short) period of time.  It is evident from the graphs that had we tried to estimate performance on the basis of a single week timing period, our results would have been very different.  For example, if we'd taken the results for Week 1, we would have estimated 8.83 spy, but had we instead used the results for Week 2, our estimate would have been 23.73 spy.  The value calculated after the end of Week 4, therefore, is our best estimate, at this 4-week point, in the project: 11.32 spy.

These results have probably raised a lot of questions.   South Pender and I will be happy to answer these.  My apologies for this overlong post.  It just seemed necessary to provide as much detail as possible to help everyone understand the graphs and the meaning of the results.
...

I would be most interested if you have also read my comment from:

http://forums.watchuseek.com/f9/methods-determining-accuracy-watch-382752-post2985575.html#post2985575

and what results have you got from the 'self-test' described there at point 3 - further talk on the actual results might be of little use without that initial estimate!

Also it might be very important to estimate major changes in your 'wearing pattern' (or in the ambient temperature) in those weeks.

[SouthPender]

Time, that's exactly one of the points I was going to make: two watches is too small a sample to say anything about a movement in general.
If we extrapolate from your first 8F56 watch, the 8F56 movement is a good, but unexceptional +/- 1 minute per year movement.
On the other hand, if we extrapolate from your second watch, the 8F56 is a Chronomaster-class +/- 6s per year movement.
And statistically speaking, averaging the results of your two watches yields a questionable number with too much uncertainty.
There is only one way to verify the accuracy of any given type of quartz movement and that's by statistical sampling at the factory where one can collect a sufficiently large sample i.e. we as consumers have to rely on figures provided by the manufacturers.
Cheers,

Well, ADB, no one was claiming that the results could be used to estimate the population (of all Precisionists) accuracy of this model.  Of course, a sample is required from which to make inferences about the population.  However, we have to work with what we have.  There will never ever be manufacturer data on the wearing accuracy of their watches.  Static tests, perhaps, as are made with the special-issue 9F Seikos, but these do not reflect wearing accuracy.  The purpose of this thread is to provide information about the performance of one particular specimen under the conditions stated.  We think it is informative.  If one Bulova Precisionist model can perform at this level, we now have some idea of the accuracy of Bulova's pre-issue claims.  Further, although this is but one result, and there will be others  to come.  As these individual timing results accumulate, we will be able to start calculating averages that will bring us closer to our goal of knowing just how accurate this movement is, over a number of watches.  What has been demonstrated here (albeit in preliminary form) is how accurate one specimen is.  It shows results close to stated specifications.  This tells us something.  If, instead, our results had worked out to +65 spy, for example, we could immediately dismiss Bulova's accuracy claims.
 
You need to know that this kind of empirical testing of individual watches is what has given us our knowledge about the performance levels of different movements.  Manufacturer specifications often don't tell us the whole story.  We've seen this with a number of Seiko 8F watches.  We've also seen that the Rolex Oysterquartz models never got close to what we consider HEQ accuracy.  Rolex didn't tell us this.  Through tests like this, we've learned a lot about the various ETA thermoliine movements--again information we would never get from the makers.  Incremental results over time of this kind do give us a good idea of the "accuracy of any given type of quartz movement."  If after 5 or 6 independent tests (i.e., different specimens) of Bulova Precisionist accuracy we have all of the specimens hovering close to specs, or, better, within specs, we have a pretty good idea of what we'll get if we purchase one.

[Time]

Maybe, but there is one extra thing - the accuracy of quartz movements changes (more or less) with AGE - usually that is hard to see on ordinary 15-20 s/month quartz, but is more obvious in HEQ. So the accuracy 'out of factory' is still relevant, but does not tell you the entire story!

thats funny.
the 8F56 thats performing terribly is a 2000 model, battery changed in 2008.

[SouthPender]

I would be most interested if you have also read my comment from:

http://forums.watchuseek.com/f9/methods-determining-accuracy-watch-382752-post2985575.html#post2985575

and what results have you got from the 'self-test' described there at point 3 - further talk on the actual results might be of little use without that initial estimate!

Also it might be very important to estimate major changes in your 'wearing pattern' (or in the ambient temperature) in those weeks.

I read that comment.  I can't speak for minidriver re the self-test, but when I have done timing of my watches, I have found that my self-tests (usually with about 5 trials), which I do to "warm up" before the actual tests and are done on the reference clock readings at both time points (rather than reference clock at Time 1 and the watch at Time 2), come very very close to the 5 second interval (I use 5-second intervals; minidriver used 10-second intervals), with a standard deviation of around 30 ms.  This is a regular result.  Minidriver's standard deviations over his 40 trials at each time point were closer to 110 ms., with some a little less and some a little more.  This is natural as minidriver is using this method for the first time.  Nonetheless, even with the 110 ms. error--arising from a combination of clock and perceptual errors--the mean readings from the 40 trials have an error standard deviation of only 16-18 ms.  It is these figures that we used to arrive at the 39 ms. margin of error.
 
As for the wearing pattern, it is pretty constant at 12 hours worn a day.  In my opinion, ambient outdoor temperature changes of a few degrees F (say 6-8 deg. F) will have very little effect on the results.  This is because the indoor temperatures will not change as much, perhaps hardly at all, and the watch is being exposed to these indoor temperatures for 12 hours a day.  Further, there is a moderating effect comiing from a relatively-constant skin temperature, so that even large changes in outdoor temperatures will be moderated by this constant skin temperature.  Thus, an 8-degree F change in outdoor temperature, for example, might result in only a 2-3 deg. F temperature change experienced by the watch.

[Catalin]

I read that comment.  I can't speak for minidriver re the self-test, but when I have done timing of my watches, I have found that my self-tests (usually with about 5 trials), which I do to "warm up" before the actual tests and are done on the reference clock readings at both time points (rather than reference clock at Time 1 and the watch at Time 2), come very very close to the 5 second interval (I use 5-second intervals; minidriver used 10-second intervals), with a standard deviation of around 30 ms.  This is a regular result.  Minidriver's standard deviations over his 40 trials at each time point were closer to 110 ms., with some a little less and some a little more.  This is natural as minidriver is using this method for the first time.  Nonetheless, even with the 110 ms. error--arising from a combination of clock and perceptual errors--the mean readings from the 40 trials have an error standard deviation of only 16-18 ms.  It is these figures that we used to arrive at the 39 ms. margin of error.
 
As for the wearing pattern, it is pretty constant at 12 hours worn a day.  In my opinion, ambient outdoor temperature changes of a few degrees F (say 6-8 deg. F) will have very little effect on the results.  This is because the indoor temperatures will not change as much, perhaps hardly at all, and the watch is being exposed to these indoor temperatures for 12 hours a day.  Further, there is a moderating effect comiing from a relatively-constant skin temperature, so that even large changes in outdoor temperatures will be moderated by this constant skin temperature.  Thus, an 8-degree F change in outdoor temperature, for example, might result in only a 2-3 deg. F temperature change experienced by the watch.

Well, as I already said in the above link - my own 'self-tests' (with 20 readings) were showing about 100 ms in standard deviation (started around 150-160ms but after some extensive experiments that was improved to about 80ms at best, and the actual experimental average error was never, ever under 10ms and went as high at 55ms - so for me the chrono method is absolutely certainly unusable for 1-week intervals with HEQ) - but I was really, really curious about results from other people!

I also have seen that the 'room temperature' can be vastly more 'variable' than people might believe - once you have a window open in that room for a few hours (from the actual relevant interval) 

[Catalin]

thats funny.
the 8F56 thats performing terribly is a 2000 model, battery changed in 2008.

If you look inside

http://forums.watchuseek.com/f9/quick-info-my-2010-timing-tests-347912.html

you will see that I have two 'old' 8F33 models and two new 8F (one 8F56 and one 8F35) - and the results are much, much better for the 'new' 8F models 

[SouthPender]

Well, as I already said in the above link - my own 'self-tests' (with 20 readings) were showing about 100 ms in standard deviation (started around 150-160ms but after some extensive experiments that was improved to about 80ms at best, and the actual experimental average error was never, ever under 10ms and went as high at 55ms - so for me the chrono method is absolutely certainly unusable for 1-week intervals with HEQ) - but I was really, really curious about results from other people!

I also have seen that the 'room temperature' can be vastly more 'variable' than people might believe - once you have a window open in that room for a few hours (from the actual relevant interval) 

I perhaps should have noted in my above post that my steady value of 30 ms. is with my satellite clock.  With the NIST clock my error standard deviations were around 45 ms. because of the far greater clock error.
 
But, you need to consider what happens to single observations (like one Stopwatch-Method reading) when they are averaged, particularly over a large number of trials.  The error standard deviation is reduced by a factor of 1/√n, where n is the number of timing trials at each time point.  Therefore, if the error standard deviation is 100 ms. for each observation, it is 100/√40 = 15.8 ms. for the average value that is used as your data point for that time point (each week's result).  Since minidriver's single-observation error standard deviations were on the order of 110 ms., his averages (over 40 trials) manifest error standard deviations of about 17.5 ms., and it was this figure that lead to our 95% confidence-interval margins of error of about 39 ms.  This is good precision and far more than adequate for all timing tests.  Compare this error standard deviation for a large-n average with the error you are getting with a single Video-Method reading.
 
With regard to your reference to a one-week timing period, I would just say that no timing results based on that short an interval have much value as stable estimates of a watch's seconds-per-year accuracy,  regardless of how they were gathered.  This is evident from the above graphs, and would be equally apparent from any one-week timing with the Video Method as well.   

[minidriver]

Indeed, it takes quite a bit of practice and good observation skills to develop consistency on each and every timing measurement session. While not 100% perfect, Southpender and I feel that we have developed enough consistency to provide fairly accurate results.

[jason_recliner]

All these graphs, and techincal terms... and 21 replies to read through... it's doing my head in... 
 
In one sentence, how accurate is the damn watch likely to be?!?!?!

[ADB]

Exactly as accurate as spec'd by Bulova!
"With Precisionist, Bulova has added a third prong to a standard two prong quartz crystal
creating a torsional resonator. This innovation results in eight times the vibration frequency of a
traditional quartz watch resulting in the most accurate watch with a continuously sweeping
second hand. The aesthetic movement of the second hand becomes the symbol to the
consumer of Bulova Precisionist and signals a watch with accuracy so exceptional, you can see
it. The technology is accurate to 10 seconds per year as compared to most other quartz
watches, which are accurate to fifteen seconds per month."

[minidriver]

All these graphs, and techincal terms... and 21 replies to read through... it's doing my head in... 
 
In one sentence, how accurate is the damn watch likely to be?!?!?!

So far, 11.32 seconds per year (Official accuracy spec is -/+ 10 secs per year)


Too many pints?