Subject: Re: near/medium future digital media economics
From: "Ben Tilly" <btilly@gmail.com>
Date: Mon, 22 May 2006 16:38:07 -0700

 Mon, 22 May 2006 16:38:07 -0700
On 5/21/06, Stephen J. Turnbull <turnbull@sk.tsukuba.ac.jp> wrote:
> >>>>> "Ben" == Ben Tilly <btilly@gmail.com> writes:
>
>     Ben> You don't necessarily know how to readily quantify [the value
>     Ben> of a potential connection] (truthfully neither do I, though
>     Ben> I've put some energy into it), but it *means* something to
>     Ben> you.  You have a referent in your experience for it.
>
> Which is different from yours, and therefore I cannot extract meaning
> from it.  Maybe you just haven't lived in Japan, or watched a Mormon
> missionary at work.  I've done both, and I know that although I can't
> quantify that value very well, I can say for sure that until we agree
> on quantification to some degree, we aren't communicating.

I'm sorry Stephen, but this seems like sophistry to me.

Would you say, "I don't know what you mean by happy" because you don't
know that the happy experiences that I've had match the happy
experiences that you've had?  It is completely and utterly true that
you don't exactly understand what happy means to me, but it is
completely and utterly false that the word "happy" coming from my
keyboard is meaningless to you.  When I say, "I'm happy", a reasonable
person is not going to say, "Communication is not happening here"
because PRECISE communication is not happening here.

Likewise when I say, "People do not place equal value on all possible
connections", it is true that we do not share the experiences that
I've based my claim on.  But unless human nature is utterly different
than I understand it to be, I'm guaranteed that you have experiences
that support this point.  Therefore it means something to you, whether
or not that meaning is precisely mine, or whether that meaning has
much precision at all..

> Note that I don't ask that we agree on which quantification is
> appropriate at this stage.  Just that at any given point we agree on
> which measure is under discussion.

There are a many measures that it would be nice to get at.  "This is
worth at least $X to me" is an important one.  Another useful one is,
"This is sufficiently worth my while that I'll pay attention to it."

However most of the reasonable measures are highly correlated.
Therefore worrying about which one is under discussion is not
important until you try to make exact statements.

>     Ben> The more quantifiable a measurement of value is, the easier
>     Ben> it is to argue that "this isn't what I mean by value!"  I
>     Ben> agree that it is a debatable measure of value.  But [count of
>     Ben> data items returned] is not an unreasonable one, and it has
>     Ben> the strong benefit that I at least have some idea how to
>     Ben> quantify it.
>
> It *is* unreasonable.  Not because it's unreasonable per se, but
> because we know that it's reasonable only as an average of very
> different extremes.

No, that was NOT the count of data items returned!

That was the count of papers about which the researcher said, "This is
interesting, I'll need a copy of this."  The count of data items
returned went up linearly with the number of journals searched, this
is the count of items  of at least a specific threshold of interest .

> On the one hand, if you're assembling a statistical data set, then
> there's decreasing returns.  In the realm of economics, a statistic
> based on 300 observations is much more accurate than one based on 30
> observations, but increasing that to 3000 observations buys very
> little, and going from there to 30000 buys nothing at all.

One would assume that if a researcher found that there were more
interesting papers than could be handled, said researcher would set a
higher threshold for "interesting" and redo the search.  (Well
actually the researcher likely would look for a review paper as an
intermediate step...)

> On the other, if you're searching for a specific thing, you get no
> returns until you find it, then you get a discontinuous jump when you
> do find it.  Increasing returns with a vengeance that would satisfy
> even the Godfather.

True.  The power law that I'm quoting is based on observations of
researchers doing a literature search.  That is not representative of
all searches that people do.

>     >> You're forgetting that I'm a coarse-filtered economist, who
>     >> only knows of three kinds of scaling behavior: decreasing
>     >> returns to scale, constant returns to scale, and increasing
>     >> returns to scale.  Obviously, I will find that O(n^2) and O(n
>     >> log n) are in the identical class.
>
>     Ben> In which case you don't care about the entire subject and we
>     Ben> can skip this conversation. :-P
>
> I do care because *you* distinguish.  For the purposes of
> understanding "Tom's Problem" I think that's a mistake.  YM obviously
> does V, but anyway, I think it needs discussion.

I distinguish because the difference has huge business implications.

Consider competition between two rival networks, one of which has 2/3
of the market and the other of which has 1/3.  According to Metcalfe's
Law, the larger one has 4 time the value of the smaller.  Per
participant it is worth double.  That's an advantage that is so
dominant that the smaller participant might as well shut up shop.  Go
home.  Lights out.

Furthermore with THAT kind of value proposition, nearby niches that
the larger network can be used for will be hard to occupy.  Sure, it
might not be the perfect fit.  But the scale advantage overwhelms
that.

Therefore Metcalfe informs us that the correct business strategy is to
grow, grow, grow.  First to market wins.  And doesn't just win, but
dominates.  Therefore you must overinvest to grow fast enough, early
enough, to take the whole pie.

By contrast n log(n) doesn't produce such drastic network effects.
The larger network is worth more than the smaller, sure.  But if the
smaller one is better, it can still outcompete.  First mover is an
advantage, sure.  But it isn't nearly so overwhelming, so you have
more time to get things right.  There are more opportunities for
leisurely deciding to interconnect.  (Incidental note, interconnection
is worth more to the smaller partner than the larger, therefore one
would expect to see the smaller partner paying for the privilege.
This is what actually is observed in practice.)  And there is more
opportunity for nearby business niches to be viable in the face of a
large competitor.

Sure, in the long run the results look similar.  But the middle game
is very different.

>     Ben> It used to be easy for me to predict who would use Macs.
>     Ben> However there are now a lot of people using them who didn't
>     Ben> not that long ago, and predicting is getting harder... :-)
>
> Aside: In my experience the Mac newbies are people like me, who insist
> on a man's OS for a man's work, but like the occasional eye candy
> too (or need "Official Doc2Text" to deal with those colleagues still
> trapped on the Great Wheel).

In my experience that population is outnumbered by people who are
relatively computer clueless whose computer guru has said, "I'm not
fixing Windows for you, get a Mac."  (As someone who has pushed
Macintosh on others for that reason, my thinking was that I didn't
want to deal with spyware and viruses, and I didn't want to teach
Linux to someone who just Doesn't Care.)

>     >> Two is not competition.
>
>     Ben> It is more competition than one...
>
> You can't quantify things that way.  There are models which are
> plausible in some situations which give competition; there are models
> which give more competition than price-taking behavior would give
> ("patent races" are a standard example); and there are models which
> give monopoly behavior, without explicit collusion.  Almost all of
> these models do converge to price-taking behavior as n -> infinity,
> though.

Oh, I CAN quantify things that way, I just might be WRONG. :-)

Metcalfe's Law basically says that the first to significant
marketshare with a good enough product will, barring extreme stupidity
or unforseen disaster, own that market.  Period.

n log(n) says that overwhelming success is far from locked in at that
point, and it is realistic to believe that a complacent market leader
can be toppled.  (However you need both a good start and a significant
quality advantage to do it.)  That means that competition lasts
longer.

Cheers,
Ben