Quick info on my 2010 timing tests ...

[Catalin]

On Jan 1st 2010 I have decided to start some 'extreme timing tests' (with a computer and a digital camera in movie mode together with one of my older programs slightly adjusted so that it will also display milliseconds) - I was initially thinking only at the new E510/E410 pair but in the end I started with 8 watches and today I had the first 'intermediary test' at around 10 days (actually 9) - I will not give yet detailed numbers, only a few things that I have noted:

- the tests for 8 watches take a lot more than I initially though - well over one hour if I include both the time to make the movies (around 3 * 8 about 10-20 seconds each, plus some extras when I see that the focus or light was not that good) plus the time to inspect those movies almost frame-by-frame and calculate the most likely interval ... this also places in the right context some of the results I have seen posted on this forum without really understanding the amount of time lost on them (even if the methodology was probably very different on many);

- this also means that I will probably only do in the future tests at around 2 weeks apart - so as to minimize the time spent yet still have two control points on each month;

- the 8 watches currently in the test are:
2 * E510
1 * E410
1 * 8F56 (JP 2008)
1 * 8F33 (US apx2000)
1 * 9923 (JP 1978)
1 * 5E31
1 * 7223 (non-heq from around 1980, but it seems that this is the second 7223 that I own from a total of 5 that is under 3s/month and I was curious on this one since it came only 1.5 months ago back from a full service at my watchmaker - which apparently did a great job);

- the results on 9 days seem to be in line with what I was expecting;

- the more interesting part of the test is that each month another group of watches 'sleeps' on my warm WiFi router (very close to wearing the watch all the time) - in January I decided to keep there the 8F56 and the initial E510 titanium, with the new steel pair E510/E410 going next on February, most likely together with the twin quartz which I might be tempted to adjust after that !

Small update - here is the procedure using:

a) a fast (>1 GHz) PC connected to the internet on a decent connection and synchronized to one of the major internet time servers just before the tests (this probably provides better than 10 milliseconds errors on the time on the computer - I use AboutTime);

b) a decent 'watch program' on the computer that will display the time including some fractions of a second in a very careful way (so as to always have very constant and small delays); unfortunately none of the major operating systems around are not even soft-realtime but I have modified one of my own programs on Windows and I believe the errors are in the same 10-20 ms interval and more important - very constant;

c) a decent (LCD) display with 60Hz refresh rate or better; the newer 120 Hz (some of which are also '3D ready') are even better!

d) a decent camera that can do movies at 25/30 (maybe even 50/60 or 100-200 if you have a 120 Hz monitor) frames/s ; ideally it should also do those movies in 'macro mode'!

e) a program that can display the movie from the above camera 'frame by frame' (VLC, even BSPlayer).

With all the above I believe you can easily measure with a precision clearly better than 100 milliseconds (and even down to the actual time for each frame on the monitor) - of course probably nothing better than 10 ms but even at 50-100 ms the results will be more than 10 times better than what we normally get with a 'human eye' ...

You can also see in a later post below a few frames extracted from one of the movies.

SECOND UPDATE:

The program I use can be now found at:

http://caranfil.org/timing/setup_ear...0_122_beta.exe

You will note that the 'main window' stays normally hidden and can be shown with either a click on it or just 'hoovering' with the mouse over it (there is a setting to configure that) - normally the milliseconds are not shown,but if on activation either SHIFT or CTRL is pressed the 'millisecond mode' is activated. If both SHIFT and CTRL are pressed the 'seconds beep' mode is also activated. See below a post with pictures from the mini-movies I am using for timing tests.

All WUS members that want to use it will be offered (when the final version is ready) a registered copy - you only need to use it and send me a private message here!

ANOTHER UPDATE with links to the pictures of the watches:

8F56 - http://caranfil.org/tz/R0016809_s.jpg
Exceeds - http://caranfil.org/tz/R0017434b.jpg
twinquartz - http://caranfil.org/tz/R0017293c.jpg
5e31 - http://caranfil.org/tz/R0017165_s.jpg
8F33 - http://caranfil.org/tz/R0017373c.jpg
7223 - http://caranfil.org/tz/R0017420d.jpg

[Mechanikus]

Thanks for sharing your plans.
I am interested in and waiting for the results.

Myself keep running my long term tests with with two new players just starting the qualification year:

http://www.mechanikus.hu/w_TheC.htm

http://www.mechanikus.hu/w_Ci4M.htm

Special thanks to ppaulusz to help me possesing Citizen Crystron 4 Mega in excellent condition.

[uktrailmonster]

Am I missing something here? If you just synchronise your watches with an atomic clock and check them again a month later, you'll get results accurate to within a few hundredths of a second per day. What more are you hoping to achieve?

[Catalin]

Am I missing something here? If you just synchronise your watches with an atomic clock and check them again a month later, you'll get results accurate to within a few hundredths of a second per day. What more are you hoping to achieve?

Well, first of all it is nice to actually measure those errors and not just assume things

Second - the more complex method used will probably accelerate the process of measuring the real accuracy with at least one degree of magnitude compared to the 'generic visual' approach - so for instance instead of having to wait for 1-2 years to have some reliable results for the new E510/E410 pair I might get them in two months (including the research on the next point).

Third - as I have also described at the end of the initial post - I will also have a generic idea on how the precision is actually related to the (average) temperature for wildly different calibers, including a thermocompensated one more than 30 years old (which if possible I would also like to adjust somewhere back to the original performance).

[Mechanikus]

Am I missing something here? If you just synchronise your watches with an atomic clock and check them again a month later, you'll get results accurate to within a few hundredths of a second per day. What more are you hoping to achieve?

I am not sure you are missing something just my focus on accuracy is a bit different than most of others I discuss it.


I am interested in the long term 'maintainable' accuracy what I can count on for a long time. That is why I consider the standard deviation of daily (mechanical watch) or monthly (quartz watch) deviations.


That is my primary qualifier of accuracy.


Obviously as a result I can check the yearly deviations against the manufacturers' specs as well.


I adore the hunt for microseconds and read the results with interest, but it is not in my focus so I do not put together computerised camera system connected to atomic clock.


If I had the opportunity I would rather run more HEQ tests in parallel.

If you are interested in the results of that approach please be my guest on my site.

[South Pender]

I am interested in the long term 'maintainable' accuracy what I can count on for a long time. That is why I consider the standard deviation of daily (mechanical watch) or monthly (quartz watch) deviations.

That is my primary qualifier of accuracy.

I must say that I don't get this at all. Could you explain for us--using some empirical data on, say, a quartz watch--how SDs tell us more than do the actual measurements or means of these?

Let me give you an example of how I'd evaluate long-term accuracy("maintainable" accuracy?). I'd (a) start by setting it to precise atomic time, and then (b) test it against atomic time once a month (or perhaps more frequently), prorating the result to annual time (12 months) each time. Suppose, by doing this, I got the following results--each month's actual deviation prorated to 12 months:

Month 1: 3.5 sec./year; Month 2: 3.75 sec./year;
Month 3: 3.25 sec./year; Month 4: 3.63 sec./year;
Month 5: 3.38 sec./year; Month 6: 3.55 sec./year;
Month 7: 3.45 sec./year; Month 8: 3.70 sec./year;
Month 9: 3.30 sec./year; Month 10: 3.65 sec./year;
Month 11: 3.60 sec./year; Month 12: 3.40 sec./year;

and, therefore:

Months 1-12: 3.50 secs. total. (Don't concern yourself with how I got these results; let's just say that I took several measurements at the end of each month and averaged them.)

The standard deviation values are: (a) .153 (as a parameter) and (b) .160 (as a parameter estimate). To some extent, these SD values can be seen as indexing measurement error (which is inevitable regardless of how carefully measurements are made). However, the actual data points, and their mean or total, are far more important than month-to-month variation in the prorated values. For example, the fact that, in 12 months, this particular watch gained 3.5 seconds is far more important to the accuracy enthusiast than the monthly wobble in prorated values, unless, I suppose, they were phenomenally large, something that we'd never expect from a quartz watch. In quartz watches, time deviations are almost always part of linear function of time. Standard deviation analysis would have to assume that some significant random factors were at work too, and such a random component--at least in my own observations--is generally negligible.

[South Pender]

Second - the more complex method used will probably accelerate the process of measuring the real accuracy with at least one degree of magnitude compared to the 'generic visual' approach - so for instance instead of having to wait for 1-2 years to have some reliable results for the new E510/E410 pair I might get them in two months (including the research on the next point).

Suppose you got results in 2 months using your method. Would you have the same confidence in prorating these to annualized values that you would have in observing the deviation from perfect time after 12 actual months?

[Catalin]

Suppose you got results in 2 months using your method. Would you have the same confidence in prorating these to annualized values that you would have in observing the deviation from perfect time after 12 actual months?

Most likely the confidence interval for predicting next 12 months will not be the same but IMHO will not be a lot different either - and actually if we also consider the fact that in this setup I can probably get some better results on keeping 2-3 watches under more constant conditions over two consecutive single months (than if I would try to keep those conditions unchanged for an entire year) I think that the effort is most likely worth it (I really can't see myself wearing just one watch for 1 year or not wearing some of my favorites for an entire year).

I also find very interesting the experiment in itself, and I am also VERY curious how precise the initial two months will be able to predict the results that I'll see at the end of one year ...

Also I would really like to adjust the twin quartz before getting too bored of it

[Mechanikus]

I must say that I don't get this at all. Could you explain for us--using some empirical data on, say, a quartz watch--how SDs tell us more than do the actual measurements or means of these?

Let me give you an example of how I'd evaluate long-term accuracy("maintainable" accuracy?). I'd (a) start by setting it to precise atomic time, and then (b) test it against atomic time once a month (or perhaps more frequently), prorating the result to annual time (12 months) each time. Suppose, by doing this, I got the following results--each month's actual deviation prorated to 12 months:

Month 1: 3.5 sec./year; Month 2: 3.75 sec./year;
Month 3: 3.25 sec./year; Month 4: 3.63 sec./year;
Month 5: 3.38 sec./year; Month 6: 3.55 sec./year;
Month 7: 3.45 sec./year; Month 8: 3.70 sec./year;
Month 9: 3.30 sec./year; Month 10: 3.65 sec./year;
Month 11: 3.60 sec./year; Month 12: 3.40 sec./year;

and, therefore:

Months 1-12: 3.50 secs. total. (Don't concern yourself with how I got these results; let's just say that I took several measurements at the end of each month and averaged them.)

The standard deviation values are: (a) .153 (as a parameter) and (b) .160 (as a parameter estimate). To some extent, these SD values can be seen as indexing measurement error (which is inevitable regardless of how carefully measurements are made). However, the actual data points, and their mean or total, are far more important than month-to-month variation in the prorated values. For example, the fact that, in 12 months, this particular watch gained 3.5 seconds is far more important to the accuracy enthusiast than the monthly wobble in prorated values, unless, I suppose, they were phenomenally large, something that we'd never expect from a quartz watch. In quartz watches, time deviations are almost always part of linear function of time. Standard deviation analysis would have to assume that some significant random factors were at work too, and such a random component--at least in my own observations--is generally negligible.

I do not want to 'sell' it that is simply my view and I do not want you to adopt it.
For details please visit my site:
http://www.mechanikus.hu/
Tests menu item.

Just see the test of 32 tests (one is done by another guy) 919 measurements and the longest run is 7 years.
The pre-qualification to be admitted to the table is one year.
If you click on the type of watch you can see the time series of deviations.

You will see the random effect even in the case of top 10. That is reality.

If you have time take a look to the tuning fork and mechanical tests as well.

[South Pender]

I do not want to 'sell' it that is simply my view and I do not want you to adopt it.
For details please visit my site:
http://www.mechanikus.hu/
Tests menu item.

Just see the test of 32 tests (one is done by another guy) 919 measurements and the longest run is 7 years.
The pre-qualification to be admitted to the table is one year.
If you click on the type of watch you can see the time series of deviations.

You will see the random effect even in the case of top 10. That is reality.

If you have time take a look to the tuning fork and mechanical tests as well.

I'd really like to understand your methodology here. When I look at the series of tests with quartz watches, I see two watches at the top of the list (in terms of their standard deviations)--01420 and Longines VHP. For 01420, I see that its monthly average is a loss of 3.89 sec., which would presumably prorate to -46.68 sec./year. In second place, the Longines VHP shows a monthly average of +.50 sec., which prorates to +6.0 seconds per year.

As for the Monthly Standard Deviation values (.465 for 01420 and .503 for Longines VHP), are these the standard deviations of the monthly values over the period of the test (36 months for 01420 and 84 months for Longines VHP)? That is, do you have 36 monthly deviations from perfect time for the 01420 and take both the mean of these and the standard deviation for your reported values?

The graph and table seem to indicate that by your standards, you'd consider 01420 superior in accuracy to Longines VHP. Is this correct?

[Mechanikus]

I'd really like to understand your methodology here. When I look at the series of tests with quartz watches, I see two watches at the top of the list (in terms of their standard deviations)--01420 and Longines VHP. For 01420, I see that its monthly average is a loss of 3.89 sec., which would presumably prorate to -46.68 sec./year. In second place, the Longines VHP shows a monthly average of +.50 sec., which prorates to +6.0 seconds per year.

As for the Monthly Standard Deviation values (.465 for 01420 and .503 for Longines VHP), are these the standard deviations of the monthly values over the period of the test (36 months for 01420 and 84 months for Longines VHP)? That is, do you have 36 monthly deviations from perfect time for the 01420 and take both the mean of these and the standard deviation for your reported values?


The graph and table seem to indicate that by your standards, you'd consider 01420 superior in accuracy to Longines VHP. Is this correct?

-----------
Methodology is simple:
1./ take the read-outs at certain interval to all tested watches
2./ calculate standard deviation of all series
3./ sort the results in ascending order.

My professor of statistics used to say: "an honest person if tells an average value, tells the standard deviation attached to it as well"

In case of watches the average is not an appropriate measure of performance for two reasons:
a./ a the good, maybe 0 average can be obtained by big differences of unique read-outs, that is what the professor drew attention.
b./ in watch calibres most cases there is an organ to modify the systematic drift that means the average.

Point b means that average can be calibrated, while deviation is the characteristic of the single piece of the given calibre type.

Your question on comparing O1420 and VHP suggest that it is against your expectations.

I have to admit it is against mine as well but my approach is to respect facts even if that is un-expected.

I document what I measured.

That is not a basis to tell the all (or just most) Omega 1420 calibres would be better than all (or just most) L.564.2 (ETA 252.611) calibres. The results I published means only that 'here and now' that piece of O1420 performs better than that piece of ETA 252.611. Not more and not less.

Let me take a step further in the direction you question directs me.

Even in the world of quartz calibres we live in a world ruled by stochastic events. We can experience positive and negative surprises. Several cases 'fortunate' pieces of calibre type of ordinary or even mediocre design can perform very well since they hide themselves under the left 'tail' of Gauss' bell-curve. In my tests Lip (ISA), Casablanca (TMI) and Timex are in top ten to my biggest surprise. On the other hand some example of good, well-designed calibres can show poor performance since they fell under the right 'tail' of the bell-curve. I can show you my O1342 and Seiko King Quartz (4826A).

Moving even further I have to tell you that my working Hypothesis was that more expensive watch is more precise. Test series of mechanical watches convinced me that it is not true among mechanicals. When I arrived to quartz watches that was my last hope to find reasonable positive correlation between price and performance. I have not found it. What is good that the as a tendency it is true that more expensive watches -let us simplify it - are better, but the correlation value is not reaching 0.4.

[Hans Moleman]

I like your approach!

If a certain movement is more stable than another: So be it.
If we can't find an explanation for that: Too bad.
We might find one later.

If we exclude that knowledge because 'it must be a measurement error', we'll never get closer to the truth.

And yes. Stability rules. Without it we would not have a hope in hell.
Standard deviation is a great qualifier for that.

[Catalin]

-----------
Methodology is simple:
1./ take the read-outs at certain interval to all tested watches
2./ calculate standard deviation of all series
3./ sort the results in ascending order.

My professor of statistics used to say: "an honest person if tells an average value, tells the standard deviation attached to it as well"

In case of watches the average is not an appropriate measure of performance for two reasons:
a./ a the good, maybe 0 average can be obtained by big differences of unique read-outs, that is what the professor drew attention.
b./ in watch calibres most cases there is an organ to modify the systematic drift that means the average.

Point b means that average can be calibrated, while deviation is the characteristic of the single piece of the given calibre type.
...

I can see some point to your line of reasoning, but honestly (with the methodology that you are using) I can ONLY see some value of that with mechanical watches.

Your a) assumption above is generally wrong on quartz - the notable exception would be a watch that is COMPLETELY stopped and on which you do random reads - and here is where I say that METHODOLOGY really matters!

The main reason why I say that is since you already 'average' a certain number of things in your measurements, and in quartz watches the influence of at least one of those - temperature - can be easily be bigger than the actual 'averaged error' that you measure AND also bigger than your measurement accuracy!

Case in point - Omega 1420 vs Longines VHP vs. Omega 1510 - last two are TC or TC-like calibers (and I find very funny that on your 1510 graph you actually can see the difference in temperature from summer to winter) yet for some strange reason on the far less accurate 1420 the numbers fail to see the MUCH bigger influence of temperature ... since as I told before - the averaging PLUS the reporting to a MUCH higher error is hiding that !!!

You might try to suggest that 1420 is a more 'predictable' caliber (while VHP and 1510 are more accurate but less predictable) - but that is NOT true - if you send those 3 to an 'impartial' observer (which however takes ONE indication from me on whether to keep the watches at 20 or 30 degrees Celsius) together with your prediction for the end of a 12 months test I can ALWAYS make your prediction VASTLY wrong for the 1420 - by picking one of the two temperature points - which however will most likely have an entire degree of magnitude LESS influence on VHP or 1510 ... so in other words 1420 SEEMS more predictable in certain conditions of averaging temperatures, but it is actually not so predictable once you do very accurate measurements on non-averaged temperatures ...

That being said I CAN see good value in using standard deviation ONCE YOU HAVE A MEASUREMENT ACCURACY OF AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE ERROR THAT YOU ARE MEASURING !!!

[Mechanikus]

I can see some point to your line of reasoning, but honestly (with the methodology that you are using) I can ONLY see some value of that with mechanical watches.

Your a) assumption above is generally wrong on quartz - the notable exception would be a watch that is COMPLETELY stopped and on which you do random reads - and here is where I say that METHODOLOGY really matters!

The main reason why I say that is since you already 'average' a certain number of things in your measurements, and in quartz watches the influence of at least one of those - temperature - can be easily be bigger than the actual 'averaged error' that you measure AND also bigger than your measurement accuracy!

Case in point - Omega 1420 vs Longines VHP vs. Omega 1510 - last two are TC calibers (and I find very funny that on your 1510 graph you actually can see the difference in temperature from summer to winter) yet for some strange reason on the far less accurate 1420 the numbers fail to see the MUCH bigger influence of temperature ... since as I told before - the averaging PLUS the reporting to a MUCH higher error is hiding that !!!

You might try to suggest that 1420 is a more 'predictable' caliber (while VHP and 1510 are more accurate but less predictable) - but that is NOT true - if you send those 3 to an 'impartial' observer (which however takes ONE indication from me on whether to keep the watches at 20 or 30 degrees Celsius) together with your prediction for the end of a 12 months test I can ALWAYS make your prediction VASTLY wrong for the 1420 - by picking one of the two temperature points - which however will most likely have an entire degree of magnitude LESS influence on VHP or 1510 ... so in other words 1420 SEEMS more predictable in certain conditions of averaging temperatures, but it is actually not so predictable once you do very accurate measurements on non-averaged temperatures ...

That being said I CAN see good value in using standard deviation ONCE YOU HAVE A MEASUREMENT ACCURACY OF AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE ERROR THAT YOU ARE MEASURING !!!

-----------------
Saying that:
"That being said I CAN see good value in using standard deviation ONCE YOU HAVE A MEASUREMENT ACCURACY OF AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE ERROR THAT YOU ARE MEASURING !!! "
you are right...

... but please realise the time base I use the signal of atomic clocks transferred via radio is AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE watches I qualify.

What is true my results cannot directly be compared to any other results obtained using some other method. Anyway that is true for any standardised measurement methods. That is why I carry out a lot of measurement to have statistical amount of data to end up with reliable results among the given constraints.

Regarding temperature effect:
Same location same chances - The HEQs except Spring Drive are sitting in the same wooden box in the same house of fairly stable temperature.

Regarding special nature of quartz watches.
I do not see why would the quartz watch be a different than other products of engineering. I see that the different types of errors are present even in case of HEQs no matter how small those errors are:
- systematic error - e.g. the drift
- stochastic error - for what we do not find systematic reason behind

------
Dear All,

Let me tell you something in general.

0. I do not think I am right or what I say is better than of others.
1. I do not want to convince anybody.
2. I do not criticise anyone.
3. What I do is might be rare: I show what I found 'sine ira at studio'

Finally I believe if we meet a side of reality we have not seen before we should not deny its existence rather think about why it is like that.

[Mechanikus]

I like your approach!

If a certain movement is more stable than another: So be it.
If we can't find an explanation for that: Too bad.
We might find one later.

If we exclude that knowledge because 'it must be a measurement error', we'll never get closer to the truth.

And yes. Stability rules. Without it we would not have a hope in hell.
Standard deviation is a great qualifier for that.

I am happy you agree with my approach.

Please tell me more about this sentence since I do not understand what you are pointing at:

"If we exclude that knowledge because 'it must be a measurement error', we'll never get closer to the truth."

[Hans Moleman]

I am happy you agree with my approach.

Please tell me more about this sentence since I do not understand what you are pointing at:

"If we exclude that knowledge because 'it must be a measurement error', we'll never get closer to the truth."

I meant that all measurements should be included even if they look odd.
It is quite tempting to throw those away since they don't fit the current theories. If those theories are wrong however, they will never be exposed as wrong, since all measurements fit so nicely. Better theories don't have any chance in such a way of thinking.
You'll get into a vicious circle of only looking at things you want to see.

Keep on open mind, as you do, and you'll get a lot further.

Actually, I would go one step further:
If I would see measurements that exactly fit the theory, I am sure they have been doctored!

You've got a scientific approach, stick to it!

[Catalin]

-----------------
Saying that:
"That being said I CAN see good value in using standard deviation ONCE YOU HAVE A MEASUREMENT ACCURACY OF AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE ERROR THAT YOU ARE MEASURING !!! "
you are right...

... but please realise the time base I use the signal of atomic clocks transferred via radio is AT LEAST ONE DEGREE OF MAGNITUDE BETTER THAN THE watches I qualify.
...

You misunderstood my point - the point was not that the atomic radio clock was not accurate enough, the point was that 'comparing two watches by naked eye' is not so reliable and only has an accuracy in the range of 0.5-1.0 seconds - that is perfect if you compare errors in the range of two degrees of magnitude better (100s/month like mechanicals), it is still very good (and undetectable in the numbers) when you compare errors in the range of 10s/month (one degree of magnitude) but once you get in the same range of 1 second/month the standard deviation starts to show more the degree of randomness in the measurement than the degree of randomness in the error that you measure !!! (and I wonder if the results don't actually show that the Omega 1420 is simply more accurate to measure with under-1-second accuracy).

That being said your results are by far the most comprehensive long-term averaged constant-condition set available on the net and most likely are rather reliable and scientific - except on this matter when we differ in the INTERPRETATION on what the standard deviation shows on under 1s/month measurements ...

The other thing that puzzles me a little is that we can see the slight variation from summer to winter on the graph of the 1510 and to some extent for the SpringDrive, but not for the 1420, Timex or Lip - where we know that the variation should be even bigger ... this might be showing that the length of the interval (or some part of the setup of the experiment) is averaging the results in a way that selectively hides some of the information ...

That being said your have convinced me that on my own tests I should also look at the standard deviation

[Mechanikus]

You misunderstood my point - the point was not that the atomic radio clock was not accurate enough, the point was that 'comparing two watches by naked eye' is not so reliable and only has an accuracy in the range of 0.5-1.0 seconds - that is perfect if you compare errors in the range of two degrees of magnitude better (100s/month like mechanicals), it is still very good (and undetectable in the numbers) when you compare errors in the range of 10s/month (one degree of magnitude) but once you get in the same range of 1 second/month the standard deviation starts to show more the degree of randomness in the measurement than the degree of randomness in the error that you measure !!! (and I wonder if the results don't actually show that the Omega 1420 is simply more accurate to measure with under-1-second accuracy).

That being said your results are by far the most comprehensive long-term averaged constant-condition set available on the net and most likely are rather reliable and scientific - except on this matter when we differ in the INTERPRETATION on what the standard deviation shows on under 1s/month measurements ...

The other thing that puzzles me a little is that we can see the slight variation from summer to winter on the graph of the 1510 and to some extent for the SpringDrive, but not for the 1420, Timex or Lip - where we know that the variation should be even bigger ... this might be showing that the length of the interval (or some part of the setup of the experiment) is averaging the results in a way that selectively hides some of the information ...

That being said your have convinced me that on my own tests I should also look at the standard deviation

First of all I am very happy that you think it is worth while taking into account standard deviation as analysis method.

To tell the truth you are right that the standard deviation is very sensitive to the error of read-out (reading only full second).
That can be eliminated by long series of data that is why I test the best ones for several years. After 7 years the result of VHP is very reliable but even the 3 years of Omega 1420 is a good basis. On the other hand the 26 month of Omega 1510 was just enough to see how the 'old-champion' can perform after more than 35 years.

[Catalin]

First of all I am very happy that you think it is worth while taking into account standard deviation as analysis method.

To tell the truth you are right that the standard deviation is very sensitive to the error of read-out (reading only full second).
That can be eliminated by long series of data that is why I test the best ones for several years. After 7 years the result of VHP is very reliable but even the 3 years of Omega 1420 is a good basis. On the other hand the 26 month of Omega 1510 was just enough to see how the 'old-champion' can perform after more than 35 years.

I have collected the raw data (the movies) yesterday on the 21st and tomorrow I will try to start for myself a long-term 'database' for that and eventually post some quick numbers - but my feeling is that the info will start to look somehow interesting only after the 3rd interval (on the 31st most likely) - and actually much better than that at the very end of February when I will also have some 'less averaged' data regarding the relation with the temperature ...

[Eeeb]

Ooooooohhhhhhhh DATA!