Anchor text usage: how much do additional words dilute the topic?

[Miamacs]

This is both a usability and a ranking related question, although in my opinion the two are inching closer.

I'm not talking about keyword stuffing, but one, two, three word anchor text both of internal and inbound links.

The target page receives votes for all words within the anchor text, but I wonder how much the additional words after the first dilute relevancy.

I can track this to some extent but there are too many parameters to get a clear picture. Suppose there's a 3 word anchor text with the first word being "widgets". And suppose that if a link points to a page with just this single word, it'd get a clear 100% vote.

If you add additional words to this anchor text, making it "widgets website" or "widgets comparison, shopping" the 100% goes for the whole phrase, and the percentage for individual words becomes lower, for not only the second or third words - which makes sense - but also for the first.

But I wonder how much lower it really is?
Would it be 33%-33%-33% or 75%-50%-25% or am I completely mistaken with this idea? The sites I oversee show some decline in the weight of the links when they're multiple word phrases, but can't say to what extent.

Some pages on some sites I work on have more than a single topic when it comes to widgets, and given that links pass most of the relevancy, this is a big issue.

Sometimes breaking the pages into two is not a good option for usability, and the visitors would find the three word phrase more descriptive. But if this is to strip the first or first two words of their weight, there's just no point in adding them.

Do you have any experience with this? Good or bad.

[tedster]

There's an active thread in this forum about Phrase based Indexing and retrieval [webmasterworld.com] where we are trying to understand 5 Google patents. They all contain sections about anchor text as well as body text. My experience is that these patents are much closer to describing how Google works with phrases today much better than a percentage-based model does.

From the perpective of these patents, the amount of "dilution" could range from none at all to lots, all depending on the specific words and phrases involved, and how predictive they are of relevance to the specific search term across a very large body of documents.

[Miamacs]

Yes, I'm aware of that thread, and have posted my thoughts there, and elsewhere on the board about how complex ranking became with thematic phrases.

But this time I'm only asking about this single and rarely discussed factor.

So some additions to my example.

Suppose a homepage is relevant for a theme, and several phrases.
Suppose the subpages are relevant for sets of three words, which are all included in the title, desciption, and present in the body text.
Suppose the homepage has inbound links from trusted sites with either word, by themselves and in combinations of three, in the exact order that the subpages have their content set to.

Or simply put, suppose you can do this experiment in a sealed-off laboratory environment ( impossible ).

In this case, would the anchor text to the subpage pass on the same relevancy for "widgets" if:

- this sole word is the entire link
- this is the first word of a three word anchor text

while

- said source and target pages are relevant for each, and the target page has the three words in the same order

The use of "proper" internal navigation is of key importance to pass on phrase based trust, the themes and keyphrases to the inner pages. For the homepage is most likely receiving most of the "natural" inbounds. Internal navigation thus needs to be set up with much thought, as discussed in this thread [webmasterworld.com].

This is a very important part of optimization for sites that compete for some exact phrases ( I know that a few trophy keywords won't match the long-tail or the "fat belly" discussed before, but a LOT of trophy keywords is very healthy for getting new citations. You know, those "natural" links. )