(Dots are individual time measurements. Positive values on the vertical axis mean the bell is ringing early, negative late. The slope of the red line, given in the chart title before *x, is the speed error: how many seconds per day the clock is currently gaining or losing.)
8/8/26 - Clock once again had to be restarted after a stopage. Initial time error: tolling begins 43 seconds before the top of the hour. [View Tracking Data]Back in 2019 my girlfriend was working at the Haydenville Congregational church, and she introduced me to Collin Black, who, with his daughter Penelope, was tending to and winding the church tower clock. I got to go up and see the clock, and I was fascinated by what I saw. (See pictures here.) So old, so beautiful, so totally mechanical. Human powered!
The clock was often pretty far from showing the correct time, with the bells ringing a minute or two away from the top of the hour, and it occurred to me to start monitoring the time error. After a bit of thought, I came up with a way of getting very precise measurements of the clock's time offset:
Using a digital sound recorder, I recorded shortwave time signals leading up to the top of the hour. When the bells were due to ring, I turned the recorder towards my deck, which has a line-of-sight view of the church tower (through a few trees), then back to the shortwave for more time signals.
I would then open the recording in sound editing software (CoolEdit):
Switching the view to waveform mode and zooming in makes it possible to set the selection endpoints very precisely:
Packrat that I am, I started collecting data... Here's the result of the first week of measurements:
The tiny plus signs in the graph are the individual timing measurements, and the red line is a least-squares fit to the data. The clock is running quite fast, gaining about 27 seconds per day, but the speed is surprisingly stable. Two months later, after an adjustment to the clock's speed control and the advent of colder weather, the clock was losing time but was running much closer to the correct speed, falling behind by about 4 seconds per day:
Another week's measurement showed an interesting pattern that was seen repeatedly:
I thought about it, then remembered that Collin described how the timekeeping cable spool should be wound so the cable filled the width of the spool, and then wound a bit more so that there were two turns in a second layer of cable. I realized that this second layer, being farther away from the center of the drum, causes the weights to exert more torque on the drum due to the extra leverage. So if the timekeeping drum is wound with a second layer, the clock will run noticeably faster for a couple of turns of the drum than it does the rest of the time.
In 2019, I got to meet David Grof, the master clock mechanic who had restored the Haydenville clock to working order in 2008 after a period of disrepair. He asked me if I thought the clock ran faster or slower in cold weather. Former physics student that I am, I replied that the pendulum would shrink in cold weather, causing it to swing more rapidly, so I guessed the clock would run faster. David replied that it was just the opposite: The pendulum, being made of wood, shrinks very little, and the clock is actually engaged in a battle of friction vs. torque. In the winter, the oil lubricating the escapement mechanism gets thicker and provides more resistance, and this slows the clock down.
This temperature-sensitive battle between friction and torque is beautifully illustrated in the time-error graph for all the data I collected during 2019:
The discontinuous jumps in the graph are from the clock time being reset. The graceful arc from late September to mid-November — initially gaining time, then slowing down and losing time — was the result of steadily decreasing temperatures. The clock's speed dial wasn't altered and the clock's time wasn't reset during this period, yielding an excellent illustration of clock speed versus temperature: the colder it gets, the slower the clock runs.
Producing the graphs as I was doing it in 2019 was laborious, and they had to be emailed to Collin for them to serve any useful purpose. I didn't keep this up for very long, but I kept making daily recordings of the clock through 2022 and squirreled the recordings away on my computer's hard drive.
Years later, I learned how to make graphs in Google Sheets — graphs that could be seen by anyone simply by following a web link. When the bells started ringing again in 2026 after a period of inactivity, I once again started collecting timing data. This time, though, I put the data in a Google spreadsheet with a graph that was automatically updated every time I added a new measurement. Something like this:
The red line is the least-squares fit to the data points, and the equation for the line in the chart title gives the speed error: the clock is running fast, gaining 7.57 seconds per day.
Having the data online makes it available, via smartphone, to the clock-minder when they're up in the tower winding the clock. (The latest data is displayed in the graph at the top of this page.) Using this live data, it should be possible to avoid having to reset the clock's time and instead make adjustments to the speed of the clock based on how far off, and in what direction, the clock is from the correct time. It should also be possible to keep the clock running accurately to within a few seconds of the correct time. People might actually start setting their home clocks by the time the tower bell rings!
A sad note: I recorded shortwave time signals from the Canadian time station CHU, transmitting from Ottawa and receivable round-the-clock in Western Massachusetts. When the bells started ringing again on June 24, 2026, for some reason CHU wasn't coming in on the shortwave. I thought radio propagation was unusually bad, but I couldn't find it on any frequency at any time of day. I finally Googled CHU and learned that, after 100 years on the air, it had been decommissioned on June 22! (The recording above is now a historic relic, a souvenir of time gone by.) With CHU off the air, I have to use the U.S. time station WWV, located in Fort Collins, Colorado, for time signals. Being so far away, WWV's signal comes in only late at night until early in the morning — which turns out to be a problem because the town may be planning on silencing the bells at night, just when I've got live clock-monitoring in place, d'oh...
Further Reading
About Radio Station CHU
RIP
National Association of Watch and Clock Collectors
- including forums for discussing everything clock-related
http://leapsecond.com
Timekeeping fanatic Tom Van Baak's site
http://leapsecond.com/nawcc2013/tvb-2013-Extreme-Amateur-Timekeeping.ppt.pdf His powerpoint presentation on Extreme Amateur Timekeeping is a fun read.
Get the family involved in relativity!
