pageok
pageok
pageok
[Paul Ohm (guest-blogging), April 13, 2007 at 12:11pm] Trackbacks
The Analog Hole is Worth Twenty-Four Cents (2 of 2):

Recently, Chris Sprigman published an article called the 99 Cent Question in the Journal on Telecommunications and High Technology Law. He asked a simple question: Why are almost all major-label songs sold online for exactly 99 cents? More to the point, why don't music labels price discriminate? They can probably sell higher quality songs at a higher price or intentionally degraded versions for less. They can try charging more for DRM-free music, or they can offer a menu of DRM restrictions; the more restrictions you'll accept, the less you'll pay.

Maybe law professors sometimes can influence decision-making, or at least maybe Professor Sprigman can. With last week's announcements from EMI and Apple, price discrimination is coming to online music in the form of $1.29, DRM-free, higher quality downloads. The songs won't be available for sale for a few more weeks, but many predict that they will sell well.

Unfortunately for economists, Apple decided to alter two variables at the same time -- DRM and audio quality -- so this won't be a pure price discrimination experiment. Even if people show up in droves to pay 30 cents more for these tracks, will they be doing it for the freedom from DRM, the quality, or both?

This brings me back to our paper. We tested whether consumers would be willing to pay different prices for different sound quality. But instead of asking whether higher quality was worth more, as Apple and EMI are asking, we asked whether consumers would pay less for lower quality music. If yes, how much less? We also used this survey to answer some fundamental questions about the analog hole and signal degradation. Before I get to our results, a few words about signal degradation:

Debates about the analog hole inevitably turn to signal degradation. Each trip through an analog-to-digital converter or digital-to-analog converter degrades the signal, as information is lost in the process. For example, ambient noise and distortion can be introduced and stereo information and bass can be lost.

If it is typical for analog hole copies to be significantly degraded, perhaps they are a poor substitute for traditional fair use and First Amendment rights, and maybe the content industry should not worry so much about the analog hole being a "DRM loophole."

In reality, it is hard to say what is "typical" signal degradation. Some analog hole copies are horribly degraded (aim your digital tape recorder out the window in the direction of your neighbor's blaring boombox) and others are near perfect (spend thousands of dollars on professional-grade equipment). We chose to use the "line-out to line-in" and "speaker to microphone" tests I described yesterday, with the understanding that they were just two types of a wide variety.

We generated an online survey, asking respondents to listen to pairs of recordings of songs, and to compare their subjective quality. For each song, there were three different recordings, the digital original, the line-out/line-in copy and the speaker-to-microphone copy. These were played to the respondent in pairs, and each respondent heard and was asked to compare all nine possible pairwise combinations. (In other words, if the three recordings were A, B, and C, each respondent heard AA, AB, AC, BA, BB, BC, CA, CB, CC in some random order during the survey.)

Seventy participants completed the survey. The results suggest first that our listeners could tell the difference between the analog hole copies and the original digital files, but perhaps not on the scale we had anticipated. In side-by-side preference tests, the digital original was picked approximately 51 percent more often than the line-out/line-in copies and 42 percent more often than the speaker-to-microphone copies.

But we didn't stop there. We also randomly assigned hypothetical prices to each song in each pair, and asked our respondents to select based on both quality and price. For example, if A was played alongside B, we would ask "If A cost $0.55 and B cost $0.25, which do you prefer?" In this way, we set up a stated preferences econometric model. This statistical technique allowed us to isolate specific consumer preferences when many factors—utility, cost, and quality—interacted. Read the paper if you want more detail.

The bottom line is the model allowed us to calculate the population's willingness to pay for quality. How much less, if any amount, were listeners willing to pay for signal-degraded, analog hole copies than for their digital originals?

The answer is tantalizingly specific: our respondents found the difference in quality in the analog hole copies to be worth 24 cents (23.9828 cents to more significant digits). In other words, if ordinary digital tracks cost 99 cents, these respondents would be willing to pay 75 cents for a lower-quality copy.

What does this all mean? If it wanted to, the music industry could probably price discriminate in the way we've described. If it offered lower-quality music downloads for less money, it would probably find a market. Although lower-quality tracks are no cheaper to produce than the standard-quality tracks sold today, lower-quality files are usually smaller, resulting in less bandwidth to distribute, leading to possible cost savings. Also, lower-quality tracks may be good enough for an iPod but not for a home audio system, which could possibly spur multiple purchases of the same song by the same consumer.

More likely, the music industry will follow the lead of the EMI/Apple deal, and attempt to price discriminate for higher prices, if at all. It is unclear whether our result is generalizable to that situation.

My stay here is almost up. This weekend, I'll try to post once or twice with some reflections about this week's stimulating discussions.

Related Posts (on one page):

  1. The Analog Hole is Worth Twenty-Four Cents (2 of 2):
  2. The Price of Music and the Analog Hole (1 of 2):
PersonFromPorlock:
Hmmm. Didn't Dogbert do this last week?
4.13.2007 2:20pm
Lior:
I would be hesitant to accept any statistics from this researcher: he claims that his research is accurate to one part in 100,000 (in that all digits of 23.9828 are significant), yet there is no way his methodology allows for such accuracy.

[Paul Ohm: I meant the significant digits as a bit of a (obviously unfunny) joke. I was poking fun at the idea that you could state this result with that level of precision.]
4.13.2007 2:25pm
justzisguyD:
I think the problem that I see with this idea is... that isn't the mindset of the companies or the buyer, for the most part. Music is TIME dependent. People who are hot today, won't be tomorrow, and THAT drives demand. The companies want to be able to charge a premium for what is newest or most in demand... less for the back catalog stuff that is more of a niche, but is still sellable.

The actual quality is less important to most people, they want the best price. That is the upside of the iTunes one size pricing... back catalog is reasonable, and your favorite new band is no more expensive than Britney's next famous song you hate. Many people listen to Mp3 format on crappy little earbuds... they care that they have 3000 songs, not that they sound like a low power fm station. Sure there IS a group that is likely to pay extra for the better stuff, just like there are those that pay $300+ for a pair of shure earbuds. I think the vastness of the market isn't concerned with that.

What some WILL like is the lack of DRM, and a smaller group, the better sound. Mostly people want PRICE. They are willing to pay more for the newest. Give them a really good price and they will buy many, and that is probably the bottom line... IMHO
4.13.2007 3:21pm
Fub:
Paul Ohm (guest-blogging), April 13, 2007 at 12:11pm wrote:
In side-by-side preference tests, the digital original was picked approximately 51 percent more often than the line-out/line-in copies and 42 percent more often than the speaker-to-microphone copies.
Statistical significance aside, that appears an intuitively anomalous result, but possibly explainable. Intuitively, speaker-to-microphone recording should obviously be more likely degraded from the original than line-out/line-in recording.

But the nature of the speaker-to-microphone degradation, assuming sufficiently high quality speakers and microphones in an otherwise quiet environment with no major resonances, would be primarily of three or four kinds: spectrum reduction, loss or garbling of phase information, intermodulation distortion, and (slight) echo.

Of these distortions, by far the most readily noticed by untrained ears is echo.

So here is a fairly readily testable speculative hypothesis to explain the anomaly that subjects preferred original over direct analog copy more often than over the speaker-to-microphone copy: human perception (at least anecdotally) is known to interpret slight and short echo as "room ambiance", an indicator of a "live performance". Slight echo for "ambiance" or "sweetener" is often added to commercially produced recordings.

So it is possible that in your experiment a very slight room echo would be interpreted as indicating a live performance, or a more "real" sound than the electrically direct analog reproduction. That may explain why subjects preferred the original over the speaker-to-microphone less often than over the direct line-in/line-out reproduction.

It could be tested many ways.

An aside: The A-B and A-B-C tests I mentioned in yesterday's comments were of a different sort entirely. The A and B were original 16-bit digital reproduction, and an MP3 (or other scheme) compressed reproduction. The upshot, also statistically insignificant due to small sample size, was that almost all subjects could not reliably distinguish between the two for MP3 bit rates of about 128 Kbps or greater.

Since that's the commonly accepted bit rate for minimal "CD quality" MP3 compression, the result is not surprising. But it does indicate how easily human perception of sound can be fooled.

What almost all these perceptual tests, and many others, tend to show is that untrained human audio perception is not difficult to fool, and that untrained ears will often perceive as more "real" or "high quality" a sound that is actually highly degraded from the original sound.

The prevalence of that general tendency to misperceive is readily apparent in the popularity of extreme dynamic range compression by some radio music broadcasters, the highly processed nature of many popular music recordings, and the popularity of "boomy" bass in small speaker systems, among many things.

For some very advanced technical expertise in these areas of human sound quality perception in your economics experiments, I'd suggest contacting Bob Orban. He is the absolute Dean of sound processing in broadcasting.
4.13.2007 4:12pm
Hucbald (mail) (www):
Ha! Another chink in the DRM debacle's armor. Soon, DRM will be double indemnity dead, and thank God.

Back when this whole DRM "thing" was being proposed, I just knew it would be a bust. At the time, I had an Alesis Masterlink ML9600 hard disk mastering recorder and CD burner, and a pro audio CD player. Both of those units had balanced three-pin connectors - not the little unbalanced line I/O connectors consumer electronics units sport - as well as state of the art D/A A/D converters. So basically, I had the technology to defeat DRM before it was even implemented, and the sound degredation evident with the CD that had gone through the D/A A/D hole was virtually impossible for even me to detect (And I'm a professional musician).

I even wrote to one of the DRM outfits and explained this at the time. Though the gear in question was about $1,500.00 back then, the same stuff can be picked up today for about $750.00 total on eBay.

If you can't figure out how to create high quality DRM-free CD copies today, you just havent put enough thought - or money - into it.

If you are wondering why a musician who stands to gain from royalties would support defeating DRM schemes, it's the purist libertarian consumer talking here: If I buy a CD or MP3, I OWN IT, and Ishould be able to put it into whichever player I want, and make as many copies of it I want.
4.13.2007 4:14pm
M. Simon (mail) (www):
At a 30 cent difference it looks like the record companies have captured the difference on the upside.

What is the difference in cost between a 10 megabit file and a 50 megabit file?

Assume a T1 line (effective transfer rate of around 1.2 megabits a second or better) costs $600 a month and is loaded to 33% capacity (on a monthly basis).

There are 30 X 24 X 60 minutes in a month (roughly). 43,200 minutes. Divide by 3 (the actual useage time). And you have around 15,000 minutes a month. 4 cents a minute effective rate. A 10 Mbit song takes 10/(1.2*60) minutes to load about .6 cents. A 50 Mbit song runs about 3 cents. Meaning that there is significat profit in the 30 cents of added value, even if you count the added capital equipment required to keep the pipeline full at peak times.

In any case - like telephone prices, song prices will get in line with actual costs over time. $1.29 is the ceiling. The floor is probably around 10 to 20 cents. Which means you should be getting an album for $1.29.
4.13.2007 4:14pm
Trey (mail) (www):
As one of the nerds who would pay more for better sound quality, I must agree that there are not too many of us. Why the record biz types never did anything to support and increase those who want a high quality listening experience is just another example of how short sighted and miserly they are.

Imagine the ad campaign, "Closer to what the artist wants you to hear" or other such adspeak, but they could easily grow the market for higher quality sound. Instead, they let sacd and dvd-audio basically die. There would have been no need for drm with those formats as they are too bandwidth heavy.

I can hear and enjoy the difference between lossy and non-lossy compression, and I am certainly not a golden eared audiophile. But creating more people like me, though simple, would require vision and some concern for the listener. Bottom line: not gonna happen.

Trey
4.13.2007 4:53pm
Alec Rawls (mail) (www):
Just a note on terminology: when economists talk of price discrimination, they are talking about selling the same product to different people at different prices, to take advantage of the fact that some people are willing to pay more for a product than others. An example would be means tested pricing (often accomplished in a rough way by tricks like offering student discounts). The subject of the post here is different pricing for different products (different quality levels). To avoid confusion, a better term might be "quality pricing."
4.13.2007 5:54pm
rothmatisseko (mail) (www):
I wonder about the quality of your listening equipment.

Nowhere in this post, nor in the linked article, do you list what the listeners listened on. I'd have expected it at ppg 10-11 of the article, given the other technical information. This seems the most important part of your analysis, and might explain why your results say people prefer the "open air" copy rather than the LI/LO copy (which, as one poster notes, is very very strange).

Cheap earbuds, earphones, headphones, boomboxes, car stereos, and different sorts of home systems will all sound vastly different playing the same song. Maybe the hole is "worth" $.24 if you've got a tin ear, but I wouldn't pay a dime for something like your "open air" copy.
4.13.2007 7:15pm
crane (mail):

Just a note on terminology: when economists talk of price discrimination, they are talking about selling the same product to different people at different prices, to take advantage of the fact that some people are willing to pay more for a product than others. An example would be means tested pricing (often accomplished in a rough way by tricks like offering student discounts). The subject of the post here is different pricing for different products (different quality levels). To avoid confusion, a better term might be "quality pricing."


Price discrimination doesn't necessarily involve selling the exact same product for different prices. In a situation where all customers buy from the same source, and there's no way to differentiate between them without letting them know that you're planning to price-discriminate, companies will often sell different versions at different prices.

For example, a company makes a printer that sells for $500 to the business market but would still make a profit selling at $200 to the student market. How do they keep business users from buying the student version? Alter the student printer so that it prints much more slowly. This is a minor inconvenience to the student, but a major one to the business that needs to print hundreds of pages a day.

Airlines also price-discriminate based on different versions of their product. Tickets cost less when you buy them well in advance and are flexible about when you go. Vacationers, then, pay less than business travelers, who often have little choice about when they have to be at their destinations, and thus less flexibility.

Offering songs at different quality levels, or with different levels of usage restriction, sounds like a perfect example of price discrimination to me.
4.13.2007 7:30pm
k parker (mail):
bass can be lost

That seems counterintuitive to me--the farther below the sampling frequency, the less loss from aliasing etc. Got a link that explains why there would be more loss at lower frequencies?
4.14.2007 12:16am
Fub:
k parker wrote at 4.13.2007 11:16pm:
bass can be lost


That seems counterintuitive to me--the farther below the sampling frequency, the less loss from aliasing etc. Got a link that explains why there would be more loss at lower frequencies?
I think you're referring to this para in Prof. Ohm's original post:
Debates about the analog hole inevitably turn to signal degradation. Each trip through an analog-to-digital converter or digital-to-analog converter degrades the signal, as information is lost in the process. For example, ambient noise and distortion can be introduced and stereo information and bass can be lost.
On that bassis here is a ground to figure on:

A round trip through the analog hole involves passing an analog signal decoded from the original digital signal through at best an effective equivalent linear RLC circuit, say a unit gain amp to feed the a/d encoder.

RLC circuits typical in such an application would have a passband, outside of which frequency response is diminished at 3 dB or 6 dBi per octave. That's the nature of the analog beast. If the particular RLC equivalent circuit through which the analog signal passes has a more limited passband than the original digital signal had encoded, then some spectrum loss would occur in the reencoded digital signal.

Since perfect DC amplifiers (with a lower passband of pure DC or 0 Hz) are unlikely to be used for a number of reasons, including the purpose of avoiding spurious DC offset in the reencoded signal, then some bass loss is likely to occur for some digitally encoded material.

In the case of using electromechanical transducers for the hole, the situation would be even worse, unless one uses transducers with near perfect DC response and sufficient environmental noise rejection or isolation. Such a setup would be impractically expensive at best.
4.14.2007 2:16pm
Eric Andrew Horwitz:
Why are you doing the line out to line in method? The best way to get around the analog hole happens also to be the easiest way. Basically, just use Windows to internally send the output stream to the input stream.

Click on Volume Control (the speaker icon in the lower right portion of screen), go to Options, then Properties, then click Recording, and then click OK. You should see a slider called "Wave" . . . select that. Now just open up some software like Cool Edit that records from the audio input stream. Click record and then start playing some audio file in Windows Media Player or the like; it'll capture the output and it will sound exactly the same.

While it doesn't require you to have the speakers on, the time it takes to record is equal to the time it takes to play the audio file. I suppose it's possible to speed up the playback and then slow down the recorded music by the same ratio, but I've neverd tried that. Also, I've only tried this for Windows XP.
4.15.2007 3:44am