ACD

What I think is that people would have a curiosity satisfied by my including that Google Reader RSS subscriptions number. It would as well be interesting, if only marginally useful, to see that number juxtaposed immediately next to the Backward Links rank.

What the hell. I think I’ll do it (and credit you and this back-and-forth with provoking the idea).

Good jousting with you.

ACD

I think the point you are making is that the sample is not random because we are selecting GR RSS subscribers, correct? However, since we are comparing classical music blogs this is only important if GR subscribers are naturally biased toward specific classical music blogs over others. It is “random enough” given that we are comparing things which should not be affected differently by the bias of the sample selection.

In short, I think that other factors (such as correctly dealing with multiple feeds per site) limit the usefulness of the RSS numbers way more than random sample concerns do.

Incidentally, the correlation of the entire data set is about 0.65.

I still disagree with your “universally accepted” phrasing but I feel that at this stage we are arguing way more about the choice of words than the usefulness of the actual statistic.

If you feel that people would benefit from the RSS numbers (something which I am quite interested in finding out) and you have the patience, I think it would definitely enhance the statistics. Actually what I would really like is a table which lets you rank by every possible available ranking, similar to the adage ones, but that kind of mammoth production is for people who actually get paid to write about these things, rather then us poor lucky fellows who do it for pleasure…

Ben

Typo up there. My, 0.7 should have read 0.3.

ACD

ACD

Yes it does. Does that mean it is “universally accepted as a measure of importance?” No, etc.

Not “the feelings of the entire internet.” The feelings of those — universally — who compile such Web statistics.

As to your two points:

1) Even allowing on your insistence on considering total GR subscribers (in which case, the stat of interest becomes the fourth subset level of five total levels), that “huge sample” is still NOT a random sample; ergo, useless for our purpose.

2) Strongly correlated? I suspect not. From the cursory investigation I made on that point, I would guess that if the correlation coefficient could be computed it would fall well below 0.7 except in the cases of the Big Boys (like Alex Ross, who’s the biggest of the Big Boys). And in any case, since the RSS sample is almost certainly not a random one, one can’t even begin to use such a statistical measure.

But you have not even begun to prove that statement.

I don’t have to. That incoming links is (and if you quote this, spare me the sic as in the context of this thread we’re talking about the concept — singular — of incoming links) the universally accepted measure of relative importance is proof enough of the statement for this purpose.

All the above notwithstanding, this back and forth between us has given me what I think might be a useful idea and a useful thing to do.

In the next quarterly S&F Top 50 rankings (to be published 1 July), I’ll of course still rank by Google Backward Links, but parenthetically display the Google Reader RSS subscription number for each entry. Better would be to use Google Reader + Bloglines subscription numbers summed for all three feeds (Atom, RSS 1, and RSS 2), but since I have to do everything by hand (the last programming I did was a utility application written in straight assembly for the Intel 8088 — the original IBM PC — some 25 years ago), that would be far too onerous a job for the compiling of a mere curiosity statistic.

Does that sound like a useful thing to do?

ACD

The acceptance of incoming links as a measure of relative importance goes well beyond that.

Yes it does. Does that mean it is “universally accepted as a measure of importance?” No. Beside your extravagant claim that you are aware of the feelings of the entire internet (“universally accepted”) you are in fact implicitly defining “importance” as “being placed well on google”:

… the success of the concept of PageRank is the reason that incoming links has [sic] become the universally accepted measure of relative importance on the Web.

Which is exactly what i was saying.

I have several issues which I am not going to get into with your “three subsets down” argument, and I am not going to start getting into sampling theory. I will simply state the following two things:

1) The Rest is Noise has a subscription number of 946, which means there are at least this many classical music blog readers using the service. Even this tiny fraction of the GR subscribers is a huge sample, statistically.

2) The GR subscription rankings and the backlink rankings are strongly correlated.

You can claim all you like that the backlinks method is:

“… merely the most appropriate and statistically robust presently available.”

But you have not even begun to prove that statement. It’s outrageous to make that claim when you have essentially no idea about hugely important things such as how Google decides on the number of links to display in the “out of xxx links” line. You are dramatically overselling your chosen method.

I wrote that the success of the concept of PageRank is the reason that incoming links has become the universally accepted measure of relative importance on the Web, and that’s precisely what I meant. Period. Full stop. No qualifying or limiting, “for Google searches.” The acceptance of incoming links as a measure of relative importance goes well beyond that.

In this case the sample population is the entire user base of Google Reader…., etc.

Not quite. It would have to be a sample population of classical blog readers in our case, and that sample population is three subsets down, and NOT random; ergo, useless as a statistical metric for our purpose, as I wrote, even if one accepts that a measure of relative popularity rather than one of relative importance is a fit metric for ranking purposes in our case which I do not; ergo, my choice of Google’s Backward Links (i.e., incoming links) as the metric best fit for the purpose.

I’ve elsewhere several times made the point that my chosen method of ranking for the S&F Top 50 is far from perfect. It’s merely the most appropriate and statistically robust presently available.

ACD

That’s not what I meant. To clarify, I said that the number of quality incoming links is one of most important determining factors for keyword searches. i.e. Google searches. When you enter a search term Google identifies appropriate websites based on the keywords entered and ranks them according to their Pagerank, which is calculated partially by the number of quality incoming links and partially based on other factors which nobody except the folks at Google know for sure. Pagerank is nowhere near to being purely a function of the incoming links, but this is at the core of the algorithm. I’m sure you are already familiar with this. To paraphrase your words, this is why the concept of incoming links has been accepted universally as a measure of relative authority for Google searches. This is far more specific than being some universally accepted measure of importance of a webpage.

I am not arguing with the fact that this is widely accepted as a fairly important rating of authority for searches, but this is a far weaker statement than it being a universally accepted measure of importance.

Anyway, I think we both agree that incoming links is a pretty good metric to use. To be honest, if you had said “widely” instead of “universally” I would probably have gone along with it.

I think we can be pretty confident that the number of Google Reader subscribers is a fairly consistent percentage of the total number of RSS subscriptions when comparing blogs of the same topic (that is to say, since essentially the same group of people are reading our blogs, we expect similar percentages to be using the same RSS feed readers) therefore it is a useful measure of RSS readership if the sample is large. In this case the sample population is the entire user base of Google Reader. Essentially each subscription number result is answering the question: of everybody who uses Google reader, how many are subscribed to blog X? This is a far larger population then just the classical music blog readership — as you mistakenly state the sample consists of — and IS in fact a pretty decent “random” (clearly it’s not entirely random as it requires people to be using GR, and there is some bias between blogs) sample.

There are of course a variety of different reasons why it is not an ideal measure…

Incomplete last graf up there. Graf should have read:

“And as to your, ‘most preferential surveys use only a small sample of an entire population, but are indicative of the population as whole,’ that’s true if, and only if, the small sample is a genuine random sample, a condition which cannot at all be verified to be the case with the small, third-subset-level sample of online RSS classical music blog readership, and is almost certainly NOT the case.”

ACD