Each of the three chandeliers in the Poughkeepsie Train Station sports 36 bulbs in two rings. When the station opened in 1918 they installed those newfangled incandescent bulbs that were all the rage at the time. The color of the bulbs in this Wikipedia picture, dated October 2007, suggests that tungsten ruled for at least nine decades:
Since then, they installed chunky compact fluorescent bulbs that probably provide the same amount of light, minus the pinpoint highlights from tungsten filaments in clear bulbs. This view from below the central chandelier shows the layout and some detail of the carving & decorative sockets:
In addition to being decorative, those chandeliers also give useful data on the reliability of compact fluorescent bulbs. With the contrast stretched the other way to make the bulbs easier on the eye, count the number of deaders in …
Chandelier 1:
Chandelier 2:
Chandelier 3:
I took each picture from a vantage point showing all the deaders; the bulbs hidden behind the central dingus work.
Let us assume all 108 bulbs were installed at the same time and, given the number of deaders, haven’t been touched since then (although they’re not covered in fuzz, which suggests that they’ve been dusted within living memory). I was there in mid-afternoon, so the bulbs probably burn 24 hours/day and aren’t subject to early failure from frequent starts.
So, in no more than five years, 108 CFL bulbs have a 4.6% failure rate, which works out to 0.9%/year, more or less, ignoring any infant mortality. If they’ve been up there for the last 2.5 years, then it’s 1.8%/year. Replacing deaders since installation, of course, makes it worse than that.
Over the course of a decade, a compounded 0.9% failure rate will kill 9.4% of the bulbs. After 20 years, 20% will be dead. A 1.8% annual failure rate kills 20% and 43%, respectively.
Now, I’ll grant you that tungsten bulbs burn far more energy over that time, but replacing a percent or two of those complex and somewhat eco-hostile CFL bulbs every year cuts away a big chunk of the rainbows-and-pink-unicorns delight involved in Saving The Planet.
The Smell of Molten Projects in the Morning 
Ed –
You mentioned startup failures. This design indicates to me one of the problems with CFL retrofits.
Immediately after the bulb plasma breaks down, the electron avalanche causes the bulb to have negative resistance, (dV/dI) causing a spicy DC transient. Most of the CFLs I have worked with have snubbers which soften the blow on the way out, but they do not suppress it entirely. I used to have a cheap amp that would pick up these transients and “click” whenever a CFL on the same circuit was turned on.
I can only imagine what 108 CFLs in parallel, switched from the same switch, would do to everything else on the same line.
The power factor for those things must be a sight to behold, too…
Although any switch that could handle, oh, 10.8 kW of incandescent lighting ought to survive the CFL surge!
Do CFL’s fail that way (i.e. x% per year), or do you get a vast increase in failures in year x?
Admittedly, I’ve noticed no discernible pattern in failures in my house…some last for a month, and some for years. I tried swapping sockets with one bulb that has lasted through three bulb changes in the other socket on the same switch (all from the same box of bulbs), and that bulb has continued to survive with no problems.
Without knowing anything else, I’d expect the usual bathtub failure curve: plenty of infant mortality tapering off to nearly nothing for at least a few years, then a steep rise as the remainder fail. But I have absolutely no data to support that expectation!
The same thing happens here: some fail almost immediately, others last basically forever. On the whole, however, there are more failing much sooner than they should.
I’ve definitely pulled far more than a few percent per year of deaders around here…
But you do know something else: the advertised life of the CFL.
The ones I have on hand claim a life of 12,000 hours which is more than 10 times the expected life of an incandescent but much less than five years. They don’t last that long for me (5/2011 to 3/2012 for the last one for my porch which is on all the time.) so use one year as a ballpark figure.
I suspect that it is quite a production to drag out the ladders and safety equipment to replace those lights. I suspect that doing that once a year instead of once a month or so is considered a big win.
Which is 500 days at 100% duty cycle, minus damage from vibration & power glitches. Looks like a year or two, at most.
Of course, the box will claim that to be 8 years at 4 hours/day, which I think we all agree is bogus.
They can’t just stand on a chair like I do, that’s for sure. I’m sure there’s a scissors lift platform involved!
I’ve found a few ways to get CFLs to fail more often than normal. First, the older ones tend to last a long time. We have a couple of Osrams left over from the ’90s that are still going. The newer curlieque bulbs will pack it in much sooner in the same fixture. (The really early Panasonics still used electro-mechanical starters, so I’m not counting them.)
Heat is an issue. I have a couple of dual-bulb fixtures on the same switch as single-bulb (er, the twins use 25W, while the singles are 13) and the twins to tend to crap out faster.
Vibration is murder on a CFL. I tried one for illuminating workpieces on a mill-drill, and it lasted an hour. When I was reroofing the house a few years ago, I had to replace a fair number of ceiling bulbs during the project. Those in hanging fixtures or floor/table lamps installed at the same time, no problem. In general, the closer the bulb was to my banging around, the more likely it was to fail. (We’re in a well-insulated triple wide manufactured home. The problematic areas were where we were about a foot between the ceiling and the roof.)
Back when our power quality was fairly bad (lots of glitches and outages) we lost bulbs quickly. Mostly, the appliance and other electronics did OK (barring the late bread maker), but CFLs weren’t happy. Fortunately, we got another substation online and the power was improved. Ah, life in the country!
Which suggests they predated the cheapnification that allows current models to sell for next to nothing… and fail next to instantly. Verily, they don’t make ’em like they used to!
I have several can lights in the Basement Laboratory Office Wing ceiling and specifically got CFL bulbs rated for “burn base up” duty (more specifically, ones that didn’t prohibit BBU installation). Most died over the course of a few years. That’s tough service, but … c’mon, that’s how ceiling lights work.
Maybe the new LED bulbs will be better; I’m hoping so, anyway.
The cheapification of LED’s is going to be a make-or-break. Newer drivers are pretty good (well, from my biased viewpoint) but the price is so breathtaking that there’s huge pressure to go for cheaper drivers, diodes, heatsinks, and lousy thermal bonding between the diodes and the heatsinks. There was an enormous move towards LED area and street lighting in China that’s stalled almost completely because they were buying the least expensive bulbs which were promptly dying of heat failure However, they should be vibration-proof.
Here’s a decent comparison of some parameters:
http://www.thelightauthority.com/compare.html
During some HAST testing of incandescent bulbs in the 90’s found that life varied inversely to the fifth power with increasing voltage. Line drop gave long life to incandescent bulbs. Haven’t seen any HAST results for CFL or LED bulbs.