In statistics—a big part of my day job—an often misunderstood idea is “regression to the mean.” This is sometimes misinterpreted as some sort of mystical process where the universe balances things out, but it really is an expression of a simple idea in probability theory. If you just observed the least likely outcome, then the next observation is expected to be a more likely outcome. This reflects the simple fact that you can’t be better than the best, worse than the worst, or higher than the highest possible score.

We have run into some regression to the mean in our watchmaking. Our last Illinois 16s “Bunn Special” pocket watch hit the elusive moment of adjusting perfection: 0 rates in all 6 positions. Because you can’t be better than perfect, probability theory implies that our next watch will be worse.

And worse it was. Here’s another 16s Bunn Special from the stash of watches awaiting repair. I had high hopes for this movement. It had no faults aside from a broken mainspring, and it was in better condition overall than the other one. But the process of adjusting went a bit wayward in a manner that could be educational for understanding practical aspects of watch adjusting.

After getting a thorough cleaning, a new alloy mainspring, demagnetizing, and a couple days to run, it was time to adjust it to positions.

The key step in dynamic poising is to find the heavy spot on the balance wheel. The movement, when running at low amplitude, was fastest in this diagonal position. We then visualize a vertical line through the wheel: the heavy spot is directly below the staff, and the light spot is directly above it.

Here’s the screw that’s exactly at the light spot. Perhaps this screw was overly undercut at one point.

Because the watch was running fast dial-up, I wanted to add a timing washer under that screw. But then the trouble began: that screw, like most of the screws on the balance wheel, was stuck. Really stuck. In fact, only a couple screws on the whole wheel could be removed. Uh oh.

The Perils of Seized Screws

You occasionally encounter balance screws that are seized: they’re stuck on there and can’t be unthreaded. Screws get stuck for a few reasons.

One is the usual crusty gunk and grime that collects in old watches. A good cleaning should remove it, but oils can nevertheless migrate into the threads and cause them to stick.

A much worse reason is that someone has used an adhesive, such as a thread-locker compound, shellac, or super glue, to fix the screw into the hole. If someone strips a screw, they might glue it in out of desperation. You wouldn’t know this until you try to remove the screw.

Trying to force a stuck balance screw is a recipe for disaster. They are made from relatively soft metals (brass or gold), so forcing it can cause a few bad outcomes. The screw head can snap off, leaving the thread end stuck in the hole. You might cause galling: part of the thread can stick to the threads and shear off, leaving you with both a damaged screw and damaged threads.

What To Do?

You can try a few remedies for stuck screws:

• Run the balance through another cleaning cycle.
• Soak the balance in denatured alcohol after removing the roller. (Denatured alcohol will dissolve the shellac.)
• Heat the balance to cause the metals to expand and the crusty gunk to melt.
• Freeze the balance to cause the metals to contract at different relative rates. (I haven’t tried this myself.)

After additional cleaning and heat treatments, I could get a couple screws to unthread.

Adjusting a pocket watch to high accuracy is a lost cause when you can’t remove the screws. Nevertheless, our fall-back aim in most cases is to see if we can get a good 3-position adjustment. Most pocket watches spend nearly all of their time either dial up, dial down, or pendant up, so a 3-position watch will serve its owner well.

For this watch, I broke out a few tricks, spent more time than I should have, and grumbled and groused more than was really unnecessary. My aim was tight rates for the 3 main positions.

As for tricks, in some cases the screw next to a light or heavy spot could be removed. Although not ideal, modifying an adjacent screw can improve poise. In other cases, I could apply a screw slot file without removing a screw. And in one situation, I could modify two screws that were equidistant from a heavy spot. These sorts of maneuvers are useful for situations like this one, but you can’t attain fine, tight rates when nearly all the screws are stuck.

Here are some charts that show how the watch turned out after 9 rounds of adjusting. This plot shows the dispersion of the 6 rates (the open dots) as well as the mean of the 6 rates (solid dot) and the standard deviation (the line). You can see that the final couple rounds were not actually the ones with the lowest variance (t7 was the tightest). This is because we were focusing on 3 of the 6 positions.

Another way is to view the rate profiles for each round.

Probably the best plot is this one: it shows how the rates changed for each position. Round 7 had the lowest standard deviation, but the final round yielded the closest rates for our 3 main positions.

Wrapping Up

I’m happy with the final 3-positions rates (DU: 1, DD: 3, PU: -1) but am a bit grumpy about what could have been for this watch, which was otherwise in fine condition. Nevertheless, this watch will keep superior time in the main pocket watch positions and looks quite sharp. There’s just no fighting regression to the mean.