The Online Photographer

By Ctein

My last column, published on June 20th, explained that there's lots of unit-to-unit variability in photographic gear. Recently I had reason to investigate variability in Epson 3880 printers.

Before I get into this, let me emphasize that these are great printers and Epson is a great company. I've been entirely satisfied with the quality of their goods, and their customer support and warranty on their professional printers is almost inhumanly superb. I will continue to buy Epson printers in the future, along with an extended warranty.

Further, my decades of experience as a reviewer do not give me any reason to think that my results are unique to Epson. I expect I would find similar variability if I looked at any maker's printers. In short, this is not Epson bashing, nor an excuse for such. Diving in...

Last fall my Epson 3880 printer developed a head clog. The heaviest of heavy-duty cleaning cycles did not clear it, so I called Epson support: "No problem, you'll have the replacement printer tomorrow."

The next day I did, along with the prepaid pickup label to return the old printer. Out with the old, in with the new. A quick test showed the tone and color balance were nearly identical to the old printer. I didn't even need to make new profiles. Hurray!

When I looked more closely at the new prints, they looked "grainier." The old printer produced invisibly-fine droplet patterns; I could just see the individual ink droplets in the new prints. Not so "hurray."

After doing my own version of "is it plugged in," I called back Epson support. Got this great technician who knew even more about Mac systems than I did. She suggested a couple of more plugs to check, so to speak. We fiddled around for half an hour, then she shipped me another printer. It wasn't a lot better.

I put the matter on hold for a few months. At the end the last year, I called support again; the fellow I got this time clearly didn't know much about Macs, but he spent a good 90 minutes with me on the phone as we tried to figure things out, even waiting while I made scans of sections of prints and e-mailed to him. He was pretty convinced I was seeing normal product variation, but I wasn't. So, Epson sent me a third printer.

It was a little better than the second, but nowhere near as good as my original printer. Still, I agreed that there was hardly much point in sending me further printers without having some good reason for thinking they would be better.

Me, I'm thinking thusly: I bought this really great printer in late 2010. I got three replacement printers in late 2011 and they all looked worse. Something had slipped in quality control at Epson. A slight timing jitter in the placement of the ink droplets would explain the degraded pattern I was seeing. There are ways that manufacturing can go awry that will cause that.

I put my suspicions forward to Epson. They accepted a packet of sample prints from me, which were shown to the production managers. They still thought it was normal sample variation.

I persevered. I have a private contact inside Epson, someone positioned to make things happen if there really was a problem and someone who knows me personally and professionally, so he knows exactly how careful I am about testing and making these kinds of claims. He investigated my reports and my data. His conclusion was that I had lucked into a "cherry" unit the first time. A random roll of the dice.

I still thought (well, hoped) that Epson might be wrong. I decide to test their conclusion. If I looked at prints from a whole bunch of 3880 printers manufactured at different times and found a systematic variation in "graininess" with time, that would prove that something had changed in the manufacturing process. I knew how to collect that data—crowdsource it to you, my faithful readership.

So now you know what my request four months back for people to make Epson 3880 prints was all about. People came through. I collected 40 respondents, with printer purchase dates scattered over more than a two-year period that nicely bracketed my units. Yes, I know that purchase date is not the same as manufacture date, but it correlates roughly. I had people provide me with their printers' serial numbers in case I needed more accurate dates down the line.

I provided a test file, told folks exactly how I wanted it printed out, and had them send me their prints. I ranked the prints by graininess from first/best to last/worst. I plotted those rankings against the purchase dates, confident a pattern would emerge. I didn't know if I'd see a gradual degradation in quality with time, or a relatively sharp transition when something changed in the manufacturing line, but I expected there'd be something.

Fig. 1

Figures 1 and 2: scatter plots of print quality (by ranking number) vs. date of printer purchase. It's the same data in both figures, the axes are swapped between the two of them because sometimes correlation patterns stand out better one way then the other. In this case, there is no pattern; it's about as random as you can get.

Fig. 2

Well, a pattern sure jumped out at me. The print quality was totally uncorrelated with time. This is about as close to a random scatter plot as I'd ever expect to see from real-world data. Epson was right. I was wrong.

Dammit.

In both plots I've labeled my current printer and my original printer. Yup, the original printer was considerably better than average while the current one is worse. Ah well.

Mind you, my new one is still a very good printer. Nobody who's ever gotten prints from it has said a negative word about the print quality (including the 750 of you who recently took advantage of the $19.95 print offer). The printer gets high marks. It's just that my old printer got an A+. He whined.

Fig. 3

Figures 3 and 4: high-resolution scans of prints from the best and worst printers in the lot. This is about a 40X magnification; you'd have to stand back 60 feet from your display for the ink droplets to appear as they would in a real print viewed closely.

Fig. 4

Still, the difference in quality from best to worst is considerable. Figures 3 and 4 show very high-resolution scans of prints from the best- and worst-ranked printers. The difference you're seeing there is greater than the difference between printing at 1440 dpi high-speed and 2880 dpi low-speed. In fact, it's comparable to the difference you'll see between entire generations of printers. In other words, the very best of these 3880 printers is right up there with the 4900 printer quality. The very worst is down at the printer quality of the previous generation of printer.

So, what can you do about this? Nothing. What can I do about this? Nothing. But, ain't it interesting!

Getting back to the point of Part I, it's a good lesson in why you should take reviews that peer too minutely into equipment quality with a certain very small grain of salt.

Or perhaps a rather large chunk of halite.

It's one of those random variation things, you know.

Ctein

Custom Printer Ctein's regular weekly column usually appears on Wednesday but was delayed this week because of Editor Presence Variability (EPV).

Send this post to a friend

Please help support TOP by patronizing our sponsors B&H Photo and Amazon

Note: Links in this post may be to our affiliates; sales through affiliate links may benefit this site. More...
Original contents copyright 2012 by Michael C. Johnston and/or the bylined author. All Rights Reserved.

Featured Comment by Sareesh S: "This is an excellent article in the best traditions of information-sharing on the web. Thank you for taking the effort and sharing!"

Mike adds: That's what I thought too. Taking Part I and Part II together, this is among Ctein's best articles for TOP. First an informed and judicious discussion of the general principles, then a specific example with original site-sourced research, clearly presented and well illustrated. Kudos to columnist Ctein from the Humbl. Ed.

Featured Comment by Paul Maxim: "I've scanned through all of the other comments and didn't notice anyone bringing this up, but if I missed it, I apologize.

"As a statistician, it occurs to me that something is missing here. While I don't doubt your testing (subjective or not), there's nothing in your published data that shows the 'within printer' level of variability. That is, if you take some of the individual printers and repeat the test multiple times (with independent setups, of course), how much within printer variability do you see? And if you then compare that level of variability with the 'between printer' variability that you've measured, is there a detectable (significant) difference? In other words, is the 'between printer' variability detectably larger than the 'within printer' variability?

"The result of that test will determine what conclusions, if any, that you can draw from the tests.

"I suspect that Epson already knows the answer to that question. And that's why they've said that you're looking at 'normal' printer to printer variability. At least I hope that they know. Otherwise, there truly is a hole in their quality system."

Ctein replies: Ah, good point! For a given printer, printing on a given paper, there's very little variation in quality from print to print or session to session. I can't say for sure there's none, but it's so small that it's smaller than any of the other sources of data "noise" I'm dealing with.