Why Accuracy is overrated (though Swiss watches might not be): A timing experiment
Like Tree106Likes

Thread: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

Page 1 of 8 1 2 3 4 5 6 7 8 LastLast
Results 1 to 10 of 80
  1. #1
    Member Purple Hayz's Avatar
    Join Date
    Jan 2015
    Location
    Durham
    Posts
    1,654

    Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    Finally able to share the results of my timing project and get feedback from my fellow WISers on some unresolved questions. Why did I do this? Simple. I’m new to this “hobby” and wanted a way to evaluate and compare watches based on their performance. Fair warning, this is a ridiculously long post, so you may want grab a p̶o̶t̶ ̶o̶r̶ ̶t̶w̶o̶ ̶o̶f̶ coffee if you decide to take the plunge. Happy reading.

    PART I: MOTIVATION AND AIMS
    Despite being the “watch guy” in my social circles, I still marvel at how much I’ve learned in the short time I’ve spent on this site. I couldn’t be more grateful to all of you who make WUS such a fine place for watch-lovers to learn, laugh, and share. With that said, I do on occasion privately bemoan the tendency for so many of the reviews, recommendations, and just good old rants/raves about watches (or brands) to focus primarily on aesthetics. In fairness, other factors are routinely referenced—as one would expect from an enthusiast site like this one—but even then, I’ve found that most rarely venture beyond A) other subjective criteria like brand pedigree or reputation, OR B) static characteristics like materials and fit/finish. The former are endlessly debatable—while the latter are of limited scope and practical utility. Neither scratches the itch for an overly empirical, performance-obsessed enthusiast like myself.

    Not to say styling and brand cachet don’t have a place of course, particularly in shaping individual purchase decisions. But even if we relegate the c̶o̶m̶m̶o̶n̶ nefarious practice of “buying for looks” to the plebes (though I have done this as well), how much WISer (ha!) are we who "̶b̶u̶y̶ ̶t̶h̶e̶ ̶n̶a̶m̶e̶"̶ focus instead on the history, heritage, and haute horlogerie of the house printed on the dial? Both are subjective reasons to buy a watch, IMO. The second simply requires a bit more reading. More importantly, both offer less than ideal criteria for evaluating and comparing watches, much less making purchase recommendations to those less “in the know.” Aesthetics, like beauty itself, are in the eye of the beholder. Arguing the attractiveness of the dial, case, or bracelet of watch A vs. watch B is like arguing one’s favorite color (pointless, since purple is clearly best) or favorite Scotch (equally pointless, because Macallan C/S). All opinions are valid; they differ only in prevalence and conviction.

    A watch’s design parameters, features, materials, and finishing provide a more objective, even quantifiable, basis for comparison, and I could almost, almost be content collecting and critiquing solely on these criteria. Seriously, the knowledge on display here at WUS is one of the biggest reasons I treasure this site. Hearing “real watch guys” (which I am not) argue the particulars of watch design and execution is a real privilege, and to those who live and breathe this stuff, I offer my sincerest praise. Be that as it may, I see build quality and performance as different animals. A toilet made of gold with a brushed platinum handle and diamond coated fittings is certainly more “lovingly crafted” than the average toilet, but it’s not clear that such lavish appointments make for a less leaky or better flushing commode, only a rarer and more expensive one. To be sure, a watch’s “specs” tell us some things about its performance--sapphire trumps glass, screw down crowns tend to seal better, etc.—but this speaks more to durability and longevity, IMO. It’s certainly reasonable to assume that build quality and performance are correlated, but I hear this assumption far more than I see it tested, much less proven.

    As a hard numbers guy, I’ve never gotten comfortable with the idea of a watch’s value being so heavily centered on craftsmanship, pedigree, and (especially) styling, rather than performance. I shop cars based on how they drive, speakers based on how they sound, and whiskies based on how they taste. Why should a watch’s function be any less relevant? Sure, styling and status symbols are important—particularly to those who see watches like earrings and bracelets—but even as our high tech gadgets (network signals and battery life notwithstanding) continue to supplant them, watches remain, first and foremost, timepieces. Excellence in timekeeping is the coin of the realm—everything else is just marketing hype, tastes, and gold toilets, IMO.

    And contrary to the oft voiced opinion, technical inferiority does not alter this equation. Yes, mechanical watches are based on “obsolete,” centuries-old technology. Yes, there are limits to how well said technology can perform. But isn’t part of the joy of our o̶b̶s̶e̶s̶s̶i̶o̶n̶ WIScraft found in testing those limits, pushing the envelope with better engineering, better materials, tighter tolerances, and so on? The Quartz revolution may have reshaped the battlefield, but it didn’t change the mission. No one expects a S.E.T. Tube Amp to match the sonic accuracy of its solid state contemporaries, or a manual transmission to match the speed and efficiency of its twin-clutch, self-shifting sibling. But we still expect our amplifiers to make sweet music, our cars to provide spirited drives, and our watches to keep good time. And the “better” the watch—be it l̶i̶v̶i̶n̶g̶ mechanical or s̶o̶u̶l̶l̶e̶s̶s̶ quartz, the more capably it should do so.

    My specific aims in taking on this project, then, are as follows:
    1) Systematically test how well various automatic watches keep time
    2) Identify key indicators of timekeeping performance
    3) Make limited, but empirically-derived, inferences about the accuracy and precision of various makes and models of mechanical watches
    4) Write up the results in laymen’s terms, share them with the WUS community, and invite feedback
    Last edited by Purple Hayz; July 2nd, 2016 at 04:16. Reason: typos

  2. #2
    Member Purple Hayz's Avatar
    Join Date
    Jan 2015
    Location
    Durham
    Posts
    1,654

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    PART II: BACKGROUND AND PRETESTING
    Like many here (I suspect), I held a detached fascination with mechanical watches well into my young adult years. Though I had little direct experience with them, I knew they were valued for their craftsmanship, durability, and analogue goodness, traits that an LP collecting, stick shift driving, “throwback” millennial like myself could appreciate. As a child of the quartz dominated 80s/90s, however, my only firsthand experience with mechanicals was a hand-me-down Elgin 218 pocketwatch (circa 1899), which oozed vintage cool but seemed “typewriter useless” in an era in which Rolex-besting accuracy could be found inside of a happy meal.

    Fast forward a couple decades, and I’m now two years into a modest but fast growing collection of autowinders. While it’s safe to say that I’ve officially “caught the bug” that you e̶v̶i̶l̶,̶ ̶m̶a̶r̶r̶i̶a̶g̶e̶-̶d̶e̶s̶t̶r̶o̶y̶i̶n̶g̶ wonderful people keep spreading all over the internet, these new additions are still competing for f̶a̶m̶i̶l̶y̶ ̶t̶i̶m̶e̶ wrist time with a couple dozen solar/quartz watches I’ve collected over the years, some of which are actually nice pieces in spite of their a̶p̶p̶a̶r̶e̶n̶t̶l̶y̶ ̶s̶o̶u̶l̶l̶e̶s̶s̶,̶ ̶p̶r̶o̶s̶a̶i̶c̶ electronic movements. My autos’ lack of wrist-time combined with some low priced, non-hacking movements exposed how little I knew about the condition and capabilities of these mechanical marvels. Sure, one or two might be running a bit “fast,” and some seem to last longer (before stopping) on the nightstand or in my (recently unplugged, utterly useless) watch winder. But until very recently, I couldn’t even begin to quantify how well these watches were (or weren’t) keeping time, much less account for any differences in performance that might exist among them.

    Now granted, I’m only a couple G’s into this little hobby (so far!). But like most of you, I’d at least like to know what I’m getting for my money, particularly as I mull dropping real money on a grail or two in the coming months. Stock movement specs are a fine starting point, but with so much diversity in the way movements are finished and packaged by various watchmakers, differences in quality control and testing, and the winding road through retail and gray market channels, published specs (when they can be found) seem a poor substitute for hands-on testing. So after spending several hours plowing through various threads, articles, and videos, I heeded the advice of my fellow WISers and forked over a couple bills for a Timegrapher.

    Eager to get some insight on how my autos were running, I grabbed the one I was wearing at that moment--an Armani Meccanico--slapped it on the mic stand, and switched the timing machine on. The plan was simple. Jot down the timing numbers, open the back of the watch if needs be, and regulate away. “How hard could it be,” I remember thinking. I’d just seen a video of a guy timing his Nixon-era Omega SomethingMaster, and it was in good enough shape (~+4 seconds, small beat error, nice “train tracks” in the beat noise readout) that he didn’t even need to regulate it. Surely my shiny new Meccanico could do the same. I certainly had no a priori reason to expect otherwise. The watch looked well built, had a fancy Italian guy’s name on it, and cost more than triple what I paid for my trusty Seiko SNK. Even the movement looked good to my untrained eye--with decorative engravings, perlage finishing on the bridge, and a brushed, black metal rotor that would surely meet Dr. Odets exacting standards (ok slight embellishment but you get the point). More importantly, I’d maybe worn the thing a half-dozen times—the leather strap was still stiff as a board! Lacking any concrete frame of reference beyond that Youtube video, but not feeling quite as d̶e̶l̶u̶s̶i̶o̶n̶a̶l̶ confident as those WISers whose mechs are never off by more than “one or two” seconds per day , I figured my watch might be running in the +/-10 second range. As the Timegrapher printed the daily rate, I was delighted to see how spot on this prediction was …

    +8 s/d

    Now THAT’S what I’m talking about, I remember thinking. In fact, 8 seconds per day seemed properly good. Doesn’t the lofty COSC allow deviations as large as +6 or something? And here a lowly mall watch is within spitting distance? Suck it, Swiss snobs! Not only is your storied pomp readily bested by the lowly quartz, but even your so-called watchmaking virtuosity affords little real world performance advantage. If a lowly fashion watch with a $30 k̶n̶o̶c̶k̶-̶o̶f̶f̶ ̶ Chinese movement can hang with the industry’s finest, why waste the money on the latter? Seems a hefty price for that “Swiss Made” badge on the dial, no? Just as my smugness was about to crescendo, the Timegrapher updated the Meccanico’s daily rate reading:

    +37 seconds

    WTF? 37 seconds? As in more than half a minute fast? It was only 8 seconds fast just a moment ago! How can a new watch be running so fast? How can any watch’s rate fluctuate so wildly? Surely this must be a glitch. Then another refresh flashed across the screen…

    +9 seconds

    Phew! I just knew that previous reading was a fluke. No way the rate should just bounce up and down by half a minute, I told myself. Let’s wait for one more reading just to be sure though…

    -17 seconds

    Oh come on! Now it’s supposedly LOSING time??? On what planet does that even make sense? It’s like the Timegrapher’s just spitting out numbers at random. [another refresh]

    +5 seconds

    Forget it. Clearly the timing machine is defective and needs to be returned. Not only does it seem to run on the same principle as a roulette wheel, but the “beat noise” readout is bonkers. I mean, just look at all that SNOW on the trace! What is this, a Poltergeist remake?! Never saw any of that white noise on the Omega dude’s YouTube video! Perhaps the Timegrapher was dropped or damaged in transit or something. Just look at this mess...

    Name:  meccanico trace.jpg
Views: 1925
Size:  124.1 KB

    To confirm the hunch, I grabbed another of my recent purchases—a gray market Tissot Visodate—tags still on it. This was my first Swiss autowinder—without one or seven of which no m̶e̶a̶n̶i̶n̶g̶f̶u̶l̶ ̶l̶i̶f̶e̶ watch collection is apparently complete. The Tissot has an ETA 2836 (just a 2824 with day, date)—a “workhorse” according to the WISers. Tried and true but nothing special—the Toyota Camry of watch movements. But it’s Swiss, so it has to cost double what the Meccanico does, even if it performs only marginally (possibly no) better. So I hooked it up to the mic and waited for the snowstorm to resume.

    And waited.

    And waited some more.

    But the snowstorm never came. In fact, I was starting to wonder if the Timegrapher had died. All it was showing was a flat line streaming across the screen like Dracula’s EKG. Occasionally, the line would zig-zag up or down by a measly pixel, but otherwise the display looked completely dead. Even the rate numbers had gone silent. 0………0………0. Then it hit me. The Timegrapher was reading exactly what the Tissot was doing. And what the Tissot was doing was keeping time. Perfect time, in fact. As in zero deviation from the (timing machine’s) reference clock. So I just let the timer run, staring in disbelief as the minutes rolled by, waiting for something to spoil the silence. But the fireworks put on by the Meccanico never returned. Just a steady stream of 0s and (almost always positive) 1s, punctuated by the occasional 2 or (rarely) even a 3. I don’t think the Tissot deviated more than +3 over ten minutes of continuous measurement. It just kept doing this:

    Name:  tissot trace.jpg
Views: 1915
Size:  89.6 KB

    So it turns out the Timegrapher wasn’t wrong at all. I was. I don’t recall which thought predominated at the time, but I definitely remember asking myself:
    1) Was the Meccanico really this bad, or was the Visodate just freakishly good?
    2) Is Armani’s Meccanico line generally poor performing? Are all “fashion” autowinders generally lousy timekeepers? Are Chinese automatics generally poor timekeepers?
    3) Is my Tissot a ringer, or all Swiss watches this good?
    4) If the Meccanico performs about 1/10th as well as the Tissot for half the price, is it really a good deal?

    It took a moment for me to realize that only question 4 could be answered with any certainty. There was no way to draw conclusions even about these two particular watches, much less whole segments of the mechanical watch market, from just a handful of data points gathered in such a haphazard manner. To do so would invite a host of internal and external validity questions. Was the Timegrpaher accurate? Was I using it correctly? Did the Tissot simply have a great day? Were these results replicable? Would other watches of the same make, model, origin, or some combination thereof perform similarly under the same conditions? These questions could only be addressed with data--more cases, more observations, more readings. Until I had such data, any conclusions about Watch A vs. Watch B (or watches from Country X vs. Country Y) would have to wait.

    Albeit not for long.
    Last edited by Purple Hayz; July 2nd, 2016 at 04:22.

  3. #3
    Member Purple Hayz's Avatar
    Join Date
    Jan 2015
    Location
    Durham
    Posts
    1,654

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    PART III: METHODOLOGY
    After more background reading, I learned that several factors influence how well mechanical watches keep time. The ones we care about are exogenous, i.e., established before we take possession of a watch, including how/where it was designed, crafted, and tested, as well as where it is sold and for how much. These factors are usually beyond our control—we decide only how we process and react to them. So for any prospective watch purchase, a typical WISer might study the specifications, read some reviews, bounce ideas off his/her fellow WISers, survey their options, and then decide whether/where to make the purchase (or….just buy every third deal posted on F71’s “Heads up” thread!). The idea is that as long as we’ve done our homework, we should know just what we’re getting well in advance.

    There are problems with this reasoning, however. First, informed choices require good information. Anecdotes, emotional appeals, fanboyism, hype, “hate”, etc. confound efforts to reach sound and timely conclusions by any means other than serendipity. The openness of forums like WUS, while admirable, can be a double-edged sword, particularly if misinformation (or irrelevant information) is in ample supply. Buyers need more than just a steady diet of photos, styling comments, and history lessons. They need more than blanket generalizations based on brand snobbery or endless, unwinnable debates about “innovation” or “authenticity.” Yet hard numbers often seem hard to come by.

    Second, design and build quality are not the only factors that influence a watch’s performance. Things like power reserve, position/orientation, temperature, and service condition all play significant roles. Many watches perform differently as their mainsprings unwind. Most run at different speeds depending on what position they’re in—a phenomenon called positional error. The biggest difference is usually between a watch lying flat (horizontal) vs. sitting on its side (vertical), since the latter generate more stress & friction on parts of the escapement. There are also non-trivial differences between the vertical positions themselves (e.g. crown up vs. crown down vs. crown right), albeit for reasons I won’t get into here. Efforts to minimize positional error date back centuries, and have led some of the most innovative (if not always effective) horological developments (e.g. the tourbillon, the gyro, etc.). It has also fueled all sorts of “nightstand regulation” tips (and myths) for folks looking to correct any daytime loss or gain (e.g. leaving a “slow watch” dial up at night, under the expectation that it will “pick up” time while resting in this position).

    This study was designed to minimize, if not eliminate, both sets of concerns raised above. Unless stated otherwise, I make no subjective assessments in reporting or interpreting the results I present. This an empirical study, not an expert “review,” a puff piece, or a hatchet job. The numbers are the story. My task is to report them, decipher them, and infer from them when appropriate. Any discussion “beyond the numbers” is up to you. I may share my personal opinion about a watch, or a quick snippet about how I came to own it, but any critical discussion of styling, brand identity, materials and craftsmanship, etc. I will leave to those who know far more about such things. In addition, I have made every effort to control for the endogenous factors discussed above. For those of you interested in the particulars, here are the details:

    Data Collection
    This analysis utilizes 12 recently purchased mechanical watches in new or almost new condition. Here is a table describing the watches sampled:
    Name:  table 1.JPG
Views: 1899
Size:  100.5 KB

    As shown in Table 1, eight of the 12 watches were purchased in the past three months, and all but two were purchased in the past year. All watches were purchased new by me, mostly from gray market vendors, and have led fairly pampered lives. None has been exposed to extreme temperatures or magnetic fields, exposed to water, dropped, damaged, or serviced. Only the KC has had the caseback opened (merely to satisfy my curiosity). Few have spent meaningful time on my wrist or in a winder. The average number of days worn is 13, but just 6 not counting the KC (my first autowinder). One watch has been worn for about a month, and another seven have been worn 3 days or less.

    Though similar in age and wear, the watches span a range of manufacturers and price points. Movements range from the lowly tongji (CSM) to the venerable ETA 2892-A2, which I understand is commonly used (or modified for use) in a number of fairly expensive and well-regarded watches (Omega, IWC, Breitling, etc.). The average sticker price for the watches sampled is just over $500, though I’ve typically paid less than half that amount (thank you, F71!) Though I’ve tried to include a variety of watches, the sample is not as diverse as it appears. Of the four Chinese-sourced watches, only the Armani and the KC were purchased to be worn. The two “Winners” (yeah, really XXX) were bought solely for tinkering and experimentation. Both have no-name tongji movements, one of which is self-winding. Also, though I tested five Japanese watches, only four remain in my collection--the first SARB was returned and replaced with the second. The two Orients have different movements, but both are in the “46” family (I believe). Last but not least, all the Swiss watches have ETA movements, and two are from the Swatch group.

    All watches were tested using an ACE TIMER Model 1000 with the sampling frequency set to eight seconds. The baseline timing data were collected using the following procedure:

    1) Fully wind the watch
    2) Place watch dial up on mic stand and input lift angle if known (otherwise leave at 52 degrees)
    3) After ~30 seconds to acclimate to position, record amplitude, beat error, and daily rate
    4) Over the following ~32 seconds, record 4 additional daily rate measures
    5) Record assessments of beat noise/jitter and linearity from the trace
    6) Repeat steps 3 through 5 for the remaining five positions: 12 up, 3 up, 6 up, 9 up, and dial down

    Once completed, I was left with measurements of each watch’s beat error, amplitude, and trace characteristics in each of the six positions. In addition, the daily rate was measured five times per position, or 30 in total. After the baseline measurements were taken, each watch was rested dial up for ~24 hours, then the entire process was repeated. In all, this yielded a final sample of 1,296 data points: 12 watches x 9 measurements x 6 positions x 2 time periods.

    Variables Analyzed

    After collecting the data, I constructed a series of summary variables for analysis. These variables are grouped into four major categories: Accuracy, Variability, Movement Health, and Stability (over time).

    Bias (Accuracy): In my field (as well as the broader scientific community), the term accuracy, when used, describes a measure’s unbiasedness or “closeness to the truth.” In statistical parlance, an unbiased (or accurate) measure is one that captures the “true” quantity of the phenomenon under study. So if the actual temperature is 35° C, an unbiased thermometer is one that reads 35. degrees. Likewise, an unbiased watch is one that keeps pace with an accurate reference clock (the Timegrapher in this case). Deviations from the reference clock are measured in seconds per day, i.e., the daily rate. Positive values of the daily rate indicate the watch is gaining time (i.e. running “fast”), while negative values indicate losing time (running slow).

    As discussed above, each watch receives 30 measurements of its daily rate. The five measurements per position are treated as random trials—there is no reason to prefer one over another. Repeated measurements are necessary to distinguish systematic problems (like movement defects) from random noise (sampling variability). The idea is that a watch keeping perfect time may nonetheless register small, random fluctuations (of perhaps +/- a second or so) when measured, while watches that consistently produce large and/or one-sided errors are probably not keeping good time. Even after averaging across the rate measurements for each of the different positions, this still leaves SIX daily rates from which to choose, and since most watches experience positional variation, the measurements from one position (e.g., dial up = +6 seconds, on average) will be poor substitutes for the measurements from another (e.g., 9 up = -4 seconds, on average). In short, the six position-specific averages must also be combined to get an overall summary of the watch’s accuracy.

    I provide three such measures of the daily rate in my analysis. Overall Rate, Desk Jockey Rate, and Blue Collar Rate. All three combine the six positional averages, varying only in how much weight they assign to each position. A logical starting point is the arithmetic mean of the six averages—essentially the COSC rate but with the “12 up” position also included (COSC uses only 5 positions). This Overall Rate provides a straightforward metric for scoring a watch’s accuracy. Like any unweighted average, it assigns equal weight to all observations, and can thus expose the limits (or design trade-offs) of watches tuned to keep good time in just one or two positions. But while the overall rate is a useful synthetic benchmark, it is also impractical. In the real world, no watch spends four hours a day dial up, another four face down, four more dial down, etc. In fact, some positions are rarely used at all, especially the “12 up” (crown right) “money shot” used in 99% of watch advertisements. For this reason, I do not consider the Overall Rate to be a particularly realistic measure of how fast/slow a watch will run.

    [WISers, why does COSC use an unweighted mean as a benchmark, given these limitations?]

    As such, I provide two weighted averages that approximate more realistic usage patterns. The desk jockey rate mirrors the wear pattern of a person that spends most of his/her waking hours seated at a desk or conference table, while the blue collar rate assigns proportionally more weight to various vertical positions, consistent with the wear pattern of those who spend more of their workday servicing equipment, climbing ladders, or tending bar, etc. Including all three measures of accuracy allows us to easily compare how well a watch will perform in different contexts.

    [WISers, are there any hard numbers on typical usage patterns, i.e. the distribution of hours per day watches spend in various positions? I was unable to find any authoritative sources on this…]

    Variability (precision):
    Though it is tempting to think that a watch’s accuracy is the only performance metric that matters, I remain deeply skeptical of this claim, as I hope to convince you to be as well. In many ways, it was the puzzling performance of the Meccanico (see Part II above) that planted this seed. Recall that when placed dial up on the timegrapher, the Meccanico’s daily rate bounced around quite a bit, with readings ranging from small (+5 seconds) to large (+37), and even switching signs. Thus, rather than providing a single, clear picture of its accuracy, the Meccanico produced a range of values from which we must choose. Averaging across the different rates certainly smooths over the highs and lows, but it also conceals important information. New Mexico and New Jersey have similar average temperatures, though their ranges are quite different. To illustrate the limitations of relying solely on the daily rate, let’s review the pretest results shown earlier, in which our two contestants gave the following performance:

    Meccanico: 8, 37, 9, -17, and 5
    Visodate: 0, 0, 1, 1, and 1

    At first, it looks like a blow-out. The Visodate is just 0.6 seconds fast on average, a truly outstanding performance. But just how far off is the Meccanico? Averaging across the five readings, we get a daily rate of +8.4, which is not only surprisingly good, it’s nearly identical to the first reading (8 s/d) I raved about earlier. How can this be? The answer is a combination of variability and “luck” (random chance). The Meccanico’s ever-fickle rate dips deep into negative territory on the 4th reading (-17). This negative value partially offsets the other four, positively signed readings, resulting in an average rate that looks rather good. Even more frightening comparisons are possible. Imagine I had only recorded the first four scores for each watch, and then dropped the most extreme score to minimize the influence of outliers. In this scenario, the daily rates calculate to:

    Meccanico: 8, 3̶7̶, 9, -17: Average = 0 seconds
    Visodate: 0, 0, 1, 1̶: Average = 0.33 seconds

    So Meccanico…FTW? Yes, but not because it’s the better watch. Rather, it “wins” because this particular combination of timekeeping errors happens to perfectly cancel out. The fundamental problem lies in treating performance and accuracy as one in the same—just as most folks do when they test their watches against the clocks on their phones or computers at night. Accuracy, in truth, is only part of the story. I would even argue that looking solely at a watch’s accuracy provides an inadequate, and potentially deceptive, picture of its performance. Accuracy tells us where the daily rate is centered, and how close that center (i.e. average rate) falls from a reference clock. But any average can be shifted up or down with tuning/regulation. What we still don’t know is the range or “spread” of the scores around that average. This is a serious omission, for even a watch that fluctuates (differs from the reference clock) by an outrageous +/- 1000 seconds at various times throughout the day, can still keep perfectly “accurate” time as long as the positive and negative fluctuations balance out in the long run. As these examples show, what’s needed is a dedicated measure of variability—something to quantify how reliably a watch keeps time throughout the day. Indeed, focusing on variability may provide a very different perspective on performance:

    Meccanico: Average = 0, Std. Deviation = 12.03 seconds
    Visodate: Average = 0.33, Std. Deviation = .47 seconds

    The Meccanico may be slightly more accurate, but it is also a staggering 25 times more erratic, with average rate deviations of nearly 12 s/d, compared to less than half a second for the Visodate. At the very least, I think a strong case can be made for including standard statistical measures of variability in the performance assessments of mechanical watches. For as any Brexit pollster should have known, the margin of error is often just as, if not more important, than the point estimate itself. To that end, I include four common measures of variability: The overall Root Mean Square, the Within Position RMS, the Between Position RMS, and the Standard Deviation of the three accuracy measures detailed above. I’ll forgo the detailed explanation of the methodology (though fellow stats gurus will recognize this as simple one-way ANOVA), and instead focus on what each measure tells us in practical terms.

    The root mean square (RMS) approximates the average or typical fluctuation in the daily rate across all 30 rate measurements (5 measurements in each of six positions). A watch that runs at the same pace over time (whether fast, slow, or spot on) never fluctuates, and will therefore have an RMS of zero. But most watches cannot keep perfectly consistent time, and the more the rate fluctuates from one measurement to the next, the higher the RMS will be. Digging a bit deeper, we see that the RMS is actually composed of two different types of fluctuations (i.e. variance components)—those that take place while the watch is in a stationary position (i.e., the “within position” RMSW), and those that take place between the different positions themselves (the “between position” RMSB). The RMSW tells us how reliable a movement is in a fixed state—with friction and gravity held constant. The RMSB tells us how robust the movement is to changes in gravity and friction (as its orientation changes). The overall RMS is simply the weighted average of these two components. Finally, I include the standard deviation of the three accuracy rates explained above (Overall, Desk Jockey, and Blue Collar). This provides a direct measure of how consistently the watch will perform for different types of wearers.

    [WISers, I’m confident that the better the watch, the lower the overall RMS. I also think that a good watch should not fluctuate much once it is resting in a particular position. In short, I suspect good watches can minimize the RMSW. But what are the optimal ratios of the two variance components? On one hand, if positional variation cannot be “engineered away,” (which make sense since the effects of gravity and friction will change depending on position), than only the RMSW is “treatable,” albeit at an expense. If this is true, then wouldn’t good watches maximize the ratio of RMSB/RMSW? On the other hand, what if someone discovers a mythical three-axis supermega tourbillon that vastly reduces positional variation, perhaps so well that all that remains is sampling variability (i.e. small but random within-position rate fluctuations aka noise). In that case, wouldn’t the best watches be those that minimize RMSB/RMSW?! I’m at a loss here…]

    Finally, I want to make some quick clarifications about terminology. While terms like “accuracy” and “precision” are widely used in discussions of watch quality, they are often vaguely defined, or even used interchangeably. I distinguish these two terms, since they are distinct scientific concepts. As stated above, accuracy (in a statistical sense) is the degree to which a measurement of X comes close to the true value (quantity) of X, be it temperature in degrees, height in inches, or time in seconds per day. In practice, statisticians and behavioral scientists tend to use the term bias (deviation from the truth) rather than accuracy, but the difference is strictly rhetorical: biased measures are inaccurate, and accurate measures are unbiased. Both terms measure the same thing, just in opposite directions. The same holds true for variability, which can be thought of as the opposite of precision. In fact, statisticians commonly calculate precision (and precision weights) as the mathematical inverse of the variance (a common measure of variability). Thus, we can think of variability as a synonym for imprecision.

    Movement Health. The Beat Amplitude and Beat error are just unweighted means of the six positional values of each respective measure. As a refresher, amplitude refers to the distance (in degrees) that the balance wheel rotates during each beat, while the error (in milliseconds) refers to the asymmetry of the clockwise and counterclockwise swings (ticks and tocks). The amplitude is determined by the amount of potential energy stored in the mainspring (which itself reflects the spring’s health, quality, and level of wind). Though I’ve been unable to find any amplitude figures in my watches’ documentation, folks seem to be of the opinion that a healthy, fully wound wristwatch should be in the ~260 to ~310 range when positioned dial up, and perhaps ~30 degrees lower in the higher friction, vertical positions. The optimal beat error is zero milliseconds, which would indicate “ticks” and “tocks” of identical duration. Though I’ve found no “hard rules” for this either, I’ve read that beat errors can be roughly classified as Good (<.6), Acceptable (<1.0), and Poor (>1.0).

    In addition to these standard measures, I also include a semi-subjective measure of the movement’s beat noise printout, which I label “Trace Quality.” This measure, which ranges from 0-100, combines information from measures of trace “jitter” (noise) and linearity, essentially my attempt to quantify how “clean” (free of noise) and “straight” (linear) the beat trace appears. Since trace quality is based on my own assessments and has no defined unit of measure, readers should interpret it more cautiously than the “hard” numbers like amplitude, beat error, and rate.

    [WISers, with that said, there was a fair bit of variation between the beat graphs of various watches, and I would be curious to know if there’s any equipment (or algorithm) out there that has attempted to rate/score the beat trace]

    Stability over time (baseline vs. +24 hours). Last but not least, I include measures of the performance degradation each watch experiences after ~24 hours of rest in the dial up position. By the +24 hour mark, most of the watches should have about 35-50% power remaining (i.e. less than half wound). However, the exact figure depends on the watch’s power reserve, as well as how fully wound the movement was for the baseline measurements. On this a few points bear mentioning. While the Chinese and Swiss watches could all be wound manually, the SARB was the only Japanese model for which this was possible. The other Japanese watches are autowind only, and the first time I attempted the +24 hour follow up on the Bambino and the SNK, both had stopped (suggesting they were not fully wound at baseline). I wound up wearing both watches, one on each arm (solely in the name of science), for two days before attempting to measure them again. I believe this effort was a success, as both watches were still ticking away the morning after the second trial (~34 hours from their baseline winding), but I may measure the Esteem again sometime down the road, as its low amplitude figures give me pause.

    In all, I include three measures of change between baseline and follow-up. The rate shift is defined as the absolute value of the change in daily rate between the two periods. While I toyed with the idea of using directional shifts, it’s not clear that a watch that loses time is any better or worse than one that gains time, since the net effect will still depend on whether the watch was running fast or slow to begin with. For simplicity, the rate shift is calculated using the Overall Rate from the two periods. I also include measures of the (absolute) shift in overall variability (RMS), which give us sense of how much the movement “loosens up” (becomes less precise/robust) as the mainspring loses power. Finally, I include a measure of the beat error shift, though to be honest, it’s not clear to me how a watch can get more (or less) out of beat (i.e. ticks become longer than tocks or vice versa) simply by losing power. Perhaps you folks can shed some light on this?
    Last edited by Purple Hayz; July 2nd, 2016 at 16:44. Reason: emphasis on precision paragraph

  4. Remove Advertisements
    WatchUSeek.com
    Advertisements
     

  5. #4
    Member Purple Hayz's Avatar
    Join Date
    Jan 2015
    Location
    Durham
    Posts
    1,654

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    PART IV: RESULTS

    Name:  table 2.JPG
Views: 1880
Size:  187.9 KB

    Here are the results of the analysis in all their glory. The numbers should be straightforward at this point (if not, refer back to the methodology section. If still not, ask me), though I’ve included some conditional formatting (color shading) to help highlight the big picture. In the simplest terms, green indicates good performance, red is poor, and yellow is in between. For each measure (column), I’ve also bolded the top and bottom performing watches. Note, all color shading is applied separately for the baseline vs. 24 hour follow-up measures (when applicable), hence the two top and two bottom performers in each column (4 total).

    Looking first at the Accuracy measures, we see that most of the watches keep fairly good time, though there are exceptions. The Orient Esteem and KC both run fast (the former gaining more than a minute a day). All other watches are within 15 s/d, and 8 in 12 are within 10 s/d, at least when fully wound. The direction the watches tend to err depends on their origin. All the Swiss and Japanese watches run fast at baseline, while half of the Chinese watches run slow.

    Which group, as a whole, keeps the most accurate time? Each has at least one watch that performs well, but for one group (Swiss), every watch does so. Of the three Swiss watches sampled, none has a daily rate of more than 6 seconds overall. In fact, looking across all 18 daily rate calculations (3 watches x 3 rate averages x 2 time periods), we see that none is higher than 7, and the mean for the group is only 3 seconds a day. The Swiss autowinders may be more expensive (on average) than their Eastern peers, but they are also the most accurate (group of) timekeepers.

    The question of most accurate individual watch is less straightforward. Readers who look solely at the overall rate (which gives equal weight to all six positions) will find several strong performers. Fully wound, the replacement SARB keeps the best time (+2), and gains only one second more at the 24-hour follow-up. The Hamilton, Vicky, and Seiko SNK also have excellent baseline rates. But the trophy goes to those with the best average accuracy at baseline and +24 hours, the Tissot Visodate and a $12, self-winding “Winner” (no pun intended) that I bought strictly to dissect and eventually destroy. You read that correctly. Using only a single, synthetic benchmark, I find that a throwaway (literally) Chinese mechanical runs neck and neck with the most accurate Swiss autowinder I own. That it does so for the cost of a (cheap) pizza is downright stupefying. If I can say nothing else about this bargain basement mechanical, it’s that it punches well above its weight. I’d hardly expected a $12 mechanical to even run, much less keep pace with watches costing 20 to 30 times more.

    Unfortunately, the story changes the moment we move beyond this single synthetic benchmark, for the “Winner” will likely feel less so when worn by an actual human (see part III for details). For the overall rate, the Tissot and the Winner both run 3 seconds fast (fully wound), but for the realistically weighted desk jockey rate, the Tissot improves to a nearly perfect +1, while the Winner drifts to +10. That the Winner’s accuracy changes depending on its usage pattern is not entirely surprising—all watches experience positional variation. The sheer magnitude of the shift bodes poorly, however, for only a dizzyingly imprecise watch could exhibit such wild discrepancies.

    Indeed, shifting our attention to the variance estimates, the performance gradient comes into much sharper relief. The Swiss again earn top marks, but now by an even wider margin, sweeping every single root mean square (within, between, and total) statistic calculated. The RMSW figures are, frankly, astonishing: just one second across the board. Hammy, Vicky, Tissot? Doesn’t matter. Dial up, crown up, six up? Doesn’t matter. Once set in position, any position, none of the Swiss watches’ daily rates fluctuate by more than a second. Even when changing positions, the accuracy rates barely move—3 to 4 second deviations for the Tissot, 7 to 8 for the Hamilton and VSA. All told, the total variation (RMS) for the Swiss trio ranges from just 1 to 4 seconds per day across 30 measurement repetitions.

    Unlike the accuracy rates, there can be no Cinderella stories here. Any watch can “nail” a daily rate if regulated properly—some even do so with luck (i.e. chance). This is possible because negative scores (losses) can offset or “cancel out” positive ones (gains). But variance (errors) is variance—all deviations are cumulative, no matter the direction. An imprecise watch can keep good time if tuned to minimize bias (inaccuracy), but I don’t think one can “tune” a watch to have a small variance. Precision, I suspect, is a question of engineering excellence (and cost, to the extent the two are correlated), though I welcome feedback from the watch experts on this.

    A comparison of the Seiko SNK (with the cheap but popular 7S26C) and the Hamilton Khaki (with the less cheap but equally popular ETA 2824-2) readily demonstrates this point. At +24 hours, a desk jockey would be pleased with either watch: the Seiko is off only 3 seconds, the Hamilton just 2. But they couldn’t be more different in how they achieve these marks. The Seiko’s rate fluctuates wildly, varying by an average of 31 seconds from one position to the next, and 14 seconds overall (RMS). Over 30 measurements, I clocked the Seiko running as fast as +12 and as slow as -35. The Hamilton deviates by just 3 seconds, on average. In fact, over the same 30 repetitions, it never ran faster than 5 seconds, or slower than -4. Thus, getting accurate time from the Hamilton (or its siblings) requires neither effort nor luck. Because its variance is so low, it will keep roughly the same time whether hanging from a tree, resting on a couch, or being carried in a pocket. You just set it and forget it. Positional variation is all but ignorable. As shown in the table, the Hamilton’s gain (+2) is identical for all three usage patterns. The desk jockey can loan the watch to his millworker cousin, or put it on a machine that rotates it evenly across all six positions. The Hamilton will keep the same time regardless. The SNK, meanwhile, doubles its timekeeping error when switched to the blue collar rate (from -3 to -6), and triples it (-3 to -9) when switched to the synthetic benchmark.

    While the Swiss have the highest precision as a group, it is by no means a monopoly. The replacement SARB approaches the lower end of the Swiss range on several variance measures, and is the only non-Swiss watch with an intra-position variance of just one second (rounding to the nearest whole number). What’s more, the SARB is easily the most consistent watch over time. No other watch sampled has a smaller rate shift between baseline and the 24 hour mark, and the variance and beat error deltas are also outstanding. Numbers aside, I really like the SARB. It is a more than capable timekeeper, and easily THE most beautiful watch I own. After all, I bought it twice.

    Name:  sarb.jpg
Views: 1851
Size:  163.1 KB

    My sole knock against it is the relatively ho-hum timekeeping performance of the first one I purchased. Long story short, after getting such terrific baseline numbers from the humble SNK, I expected the SARB, with its premium (6R15) movement, to handily outperform, or at least match, the Swiss in accuracy and precision. When it failed to do either, and was even upstaged by its supposedly lesser sibling, I filed the RMA request even before waiting for the +24 hour follow-up. The replacement (shipped directly from Japan) performed better on almost every measure, but still lags the Swiss in precision.

    The other Japanese standout is the SARB’s homelier, scrappier, kid brother. The SNK was my first “real” watch (i.e. automatic, made by an actual watchmaker, “in house” movement, pedigree, etc.), and it’s gotten more wrist time than everything but the Kenneth Cole. It’s a modestly sized, modestly priced ($60), basic field watch with a dull matte finish and a cheap blue nylon strap (replaced with a leather NATO). Of the 5 Japanese watches I tested, I pegged it to finish last, while still turning in solid numbers. Then I ran the baseline numbers. Not only did the SNK finish first, it so thoroughly upstaged the 6x more expensive SARB that I immediately returned the latter in disgust! The lowly SNK also put up one heck of a fight against the costlier Swiss competition. Fully wound, its accuracy is neck and neck with the Hamilton. More impressive still are its low variance figures, which are only slightly off the marks set by the Hamilton and the Victorinox. Though the SNK degrades severly as the mainspring unwinds—going from best Japanese performer to worst by the 24 hour mark—it still delivers near-Swiss timekeeping and precision when fully wound, which seems better than any $60 watch has a right to be.

    Except for one.

    Which brings me at last to the Kenneth Cole, my first autowinder and the last one I’ll discuss before wrapping up. Though the KC will get a puff piece of its own someday, for now I will say only that I’ve read enough on this site to know how little regard a watch like this typically receives. I understand the reasons and don’t necessarily disagree with (all of) them. But no matter how numerous the KC’s shortcomings, precision is not among them. In fact, when fully wound, this lowly fashion watch lays waste to all comers, including the Hamilton and the Vicky. It even manages to “tie” the (seemingly immortal) Visodate’s variance figures, at least when rounding to the nearest whole number (a mathematically correct, but conservative choice here). Like the SNK, the KC degrades significantly as it unwinds, yet so good are its baseline numbers, that even when down on power, its variance stats still beat nearly every Japanese watch. In addition, no watch, not even the superb Visodate, delivers more consistent timekeeping across the three daily rate measures. This would be fine, if not exceptional, performance from most any timepiece, much less a two year old “mall watch” with apparent health issues (highest beat error by a large margin). [WISers, should I be concerned about this?]

    I close this insanely long post with an honest aside/confession. In case it hasn’t been apparent from the preceding paragraphs, I never really “got” the whole Swiss mechanical thing. Until very recently, my respect for Swiss watches would have generously been described as tepid. Just another shiny, wearable, status symbol, needlessly embellished by storied purveyors of overworked, overpriced, functionally obsolescent tools cum fashion statements. I only bought the Tissot to prove what I had “known” all along—that no one engineered and innovated better than the Japanese, and no one r̶i̶p̶p̶e̶d̶ ̶o̶f̶f̶ copied the innovative engineers better than the Chinese. This wasn’t so much a knock against the Swiss as it was a belief (sans evidence, mind you) that they had simply been upstaged and undercut by their newer, tech savvy rivals. I would have been content with Chinese “homages” and fashion mechs (oohhh look, now there’s sub-$500 Tourbillons!) Seiko and Orient would be my BRANDS. Honest, functional, durable, always accessible, and on rare occasion, beautiful. I could smugly diss the Swiss pedigree, marketing hype, and cringe-inducing brand ambassadors. I would laugh all the way to the bank, I thought, ever confident that I’d gotten Swiss-level performance for pennies on the dollar, you know, because I’m that guy. And when at last I decided I had really arrived, I would indulge myself with the only vanity purchase befitting my lofty (anti-)snobbery, a Grand Seiko!

    Then I saw what these bloody Swiss baubles could do. I watched them, one after the other, lay down these almost impossibly straight, perfectly linear traces for minutes on end. I measured them, again and again, staring in disbelief at the utter absence so much as a single double digit number. They felt, and I hate this word, premium, and not in the wankerish “Oh just pick up a Rolex and you’ll FEEL the quality” kinda way, but rather the superbly designed, meticulously executed, delivers on the hype kind of way. It was sobering, annoying, and seductive all at once. I still don’t know how deep this rabbit hole goes, but I’m sure glad I decided to take a look.

    Yours in WISdom,
    PH

    ADDENDUM:

    Folks, just to be clear, the take home here it is that precision, not accuracy, distinguishes the performers from the pretenders. Put simply, I think a watch that keeps consistent (but inaccurate) time is far preferred to one that keeps accurate (but inconsistent) time. I discovered this truth empirically (because I’m new to this), but I don’t think we even need data to prove it. If a watch is imprecise (high variability), its accuracy is constantly fluctuating. How could a watch that runs +18 one moment, dead accurate (0) the next, and then -12 a few seconds later realistically be regulated? Only with luck (chance) could a watch like this keep good time (average out to 0). A precise watch keeps the same time, all the time. Making it accurate requires nothing more than the slide of a lever or the turn of a screw. Which would you rather have, a thermometer that consistently reads +5 degrees hotter than the real temp, or one that bounces around randomly (sometimes accurate, sometimes too low, sometimes too high) from reading to the next?

    Also, please limit attempts to generalize from my results. As I’ve gone to lengths to point out, my sample (i.e. my collection) is not even remotely representative of the broader populations of Swiss, Chinese, or Japanese movements, and I never presented it as such (see the data description for details).

    This is a descriptive exercise, not a forecast. I am 100% confident that these 12 watches perform exactly as I’ve shown here. I know this because I measured them systematically and repeatedly under controlled conditions. Beyond that I can only speculate, as I would need even more data (and formal statistical tests) to reach definitive conclusions about the general populations from which these watches were sampled. With that said, I think the data do offer some promising clues:

    -I started with a head-to-head matchup of two watches with $400-$650 stickers. The Swiss one curb stomped the Chinese one and then rode off on his bicycle. I was not expecting this.

    -So I decided to look closer, this time with a sample size of 12 instead of two. The results were similar. As a group, the Swiss watches handily outperformed the others. That’s what the data show.

    -The next step will be to kick the sample size up to 50 or so (and include a broader selection of movements and price points), at which point I will probably need to slow down or get a divorce. I can’t tell you (yet) what those results will show, but I can tell you how I’d be inclined to wager
    Last edited by Purple Hayz; July 8th, 2016 at 06:44.

  6. #5
    Member
    Join Date
    Nov 2013
    Posts
    4,604

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    I don't understand why you're talking about "swiss" or any origin here with watches. As for accuracy, how do the watches do compared to their stated accuracy?

    more importantly watches are going to have variance, but if you took 100 watches with a higher standard rating vs. 100 lower rated, you'd expect that the be true. Otherwise you have a brand or group who "under rates" perhaps.

    Either way I hope this was fun, but i'm not sure what value this will have to a consumer. A sample of one is just that. Statistically meaningless.
    MaDTempo, Sevenmack and Morethan1 like this.

  7. #6
    Member MaDTempo's Avatar
    Join Date
    Oct 2012
    Location
    Denver, CO
    Posts
    423

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    Admittedly, I had to give up on reading the entire post as my Dewars White Label and Drambuie (a cheap and rusty nail - I know you said coffee) took hold. However, as a researcher I am grateful for the discussion of accuracy, precision, and variance in your post. When I first joined this forum/hobby as a WIS-in-training, I delved into the concept of the "most accurate" watch only to be quickly relieved of my delusion that a mechanical will ever acheive what I look for in terms of accuracy, precision or variance (I am waiting for the right price on a radio controlled Citizen as I live just miles from an atomic clock tower in Ft. collins but my UHF Bulova Snorkel comes next week). But, I have not yet lost my appreciation for the artistry of measuring time using components that, individually, are inert trinkets of metal, silicone and minerals to measure the time - however imprecisley.

    Hence, your detailed post reveals what, to me, is most enjoyable about watch collecting...that the measurement of time can be accomplished with so many variables at play yet be reasonably accurate while allowing exploration of new materials, techniques, presentations and that this hobby plays to the romantic, the experimentalist, the connoisseur, the historian, the futurist or the philosopher. Like the group,of blind men touching different parts of the elephant, we appreciate different aspects but only when we take in the whole do we get a true idea of a thing.

    thanks for posting!
    It's not a matter of mercy
    It's not a matter of laws
    Plenty of people will kill you
    For some fanatical cause
    - Rush

    C-3PO: "Is there anything I might do, sir?"
    Luke: "Not unless you can alter time, speed up the harvest or teleport me off this rock."

  8. #7
    Member
    Join Date
    Dec 2014
    Location
    Australia
    Posts
    998

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    That is one serious post!!

    But I still like how accurate my omega sm300 and my Rolex OP are!
    Gives me warm feelings inside.
    But I don't bother with watch timers etc, just check its accuracy with my iPhone over a month or so.
    Sorry for the basic low tech reply ;).


    Sent from my iPhone using Tapatalk
    alessandro132 and trf2271 like this.

  9. #8
    Member oak1971's Avatar
    Join Date
    Aug 2013
    Location
    SE Upper Mid West
    Posts
    2,336

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    Nice post. I like using my time grapher just to check up on some of my older watches.

    I could list my watches here, but I won't.

  10. #9
    Member yankeexpress's Avatar
    Join Date
    Apr 2013
    Location
    USA
    Posts
    32,934

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    Such a small sample with so many common movements not included is near meaningless.

    Where is Soprod A-10, any Sellita, Seiko 4 and 8 series, Miyota 9015 and 8215?......add so many others also missing.
    Last edited by yankeexpress; July 2nd, 2016 at 02:13.
    Badbebe likes this.

  11. #10
    Member
    Join Date
    Mar 2016
    Posts
    955

    Re: Why Accuracy is overrated (though Swiss watches might not be): A timing experiment

    Quote Originally Posted by yankeexpress View Post
    Such a small sample with so many common movements not included is near meaningless.

    Where is Soprod A-10, any Sellita, Seiko 4 and 8 series, Miyota 9015 and 8215?......add so many others also missing.
    You wanna donate a few?

    Nice posts OP, kudos for your effort. I'm not that surprised that the 7s26 is similar to the 6r15 in terms of accuracy. The only real difference is the hacking/winding and the mainspring material to extend reserve. Also not surprised that the ETA 2824s outperform Seiko, although I'm a little surprised by how close they are. If the cost in accuracy for a lower beat rate and a chunkier movement is this low, that's a pretty fair trade. Perhaps report back in 10 years to talk about longevity? :)

    I am also surprised by how weak the Orient and Seagull movements are. Although for Seagull, I hear that their ebauche movements are terrible relative to the movements they use for their own watches.

    In terms of your variability statistics - I think if you put standard errors around your estimates, it would be harder to distinguish between the movements.
    Last edited by KtWUS; July 2nd, 2016 at 04:00.
    Purple Hayz and MaDTempo like this.

Page 1 of 8 1 2 3 4 5 6 7 8 LastLast

Thread Information

Users Browsing this Thread

There are currently 1 users browsing this thread. (0 members and 1 guests)

    Similar Threads

    1. Are Sinn watches overrated?
      By dethinker in forum German watches
      Replies: 40
      Last Post: March 17th, 2019, 21:07
    2. G-Shock accuracy experiment: results after one month
      By Lester Burnham in forum Casio G-Shock
      Replies: 27
      Last Post: March 9th, 2019, 17:48
    3. Do You Find Hublot Watches Overrated?
      By Mixalix in forum Public Forum
      Replies: 86
      Last Post: June 7th, 2014, 12:07
    4. Interesting Sumo Timing Experiment
      By kevinkar in forum Seiko
      Replies: 8
      Last Post: February 13th, 2009, 22:35

    Posting Permissions

    • You may not post new threads
    • You may not post replies
    • You may not post attachments
    • You may not edit your posts
    •