McDar Experiment

Yes, there is definitely something going on here. I ran a test as well, and here are my results for 'sleeping bag' (singular):

http://www.sleeping-bag-shop.com
allinanchor:not listed (678 results)
normal: not listed (826 results)
sandboxed: #17

http://www.mcdar.com/camping1/sleeping-bags.htm
allinanchor: not listed (678 results)
normal: #506
sandboxed: not listed (801 results)

It seems like something drastic is happening to the first of your sites (http://www.sleeping-bag-shop.com). From being at #99 (according to your last data), it has now fallen completely out of both the allinanchor SERP and the regular SERP. But I find it a bit strange that it should have been replaced by the experimental site. The two pages are too different, as far as I can see it.

Are there any other ways the page can have provoced the filter Google applies on the regular search?

Regarding the second of your sites (the experimental site), I don't find my results as disturbing as those you mentioned above. I found it at #506, not #25 (I only checked one DC though). I couldn't, however, find it in neither the allinachor SERP nor the sandboxed SERP. How come?

Maybe the filter that differentiates the sandboxed search and the regular search includes some sort of stemming algorithm (as also have been mentioned in this thread earlier)? In such case, all the anchor text connected to the experimental site ('sleeping bags') could be given some value also for 'sleeping bag' in the regular seach...?

I really hope your site comes back for this search term, Caryl, as seeing something like this is always pretty depressive.

 

When were these links added? And is the anchor text an exact match? According to the theory, which may or may not be proven -- fully or partly, the effect of links are more or less immediately factored into the allinanchor results (quickly after the page where the link is located has been spidered) and the sandboxed search. Then gradually, over the next 3 months or so (timing is uncertain) the links should gain more effect in the normal search as well.

This could explain why you are still at #8 in the regular search and at #1 in the sandboxed search (at least if your links have been acquired during the last three months).

Please let me know what happens to your site in the future (monitor both the regular search and the sandboxed search), as I would be very interested in as much data as possible to test this theory.

 

I've been away for a couple of hours and obviously everyone has been busy. But unless I've missed something I thing the answer to that question is fairly simple.

The experimental page is only going to show up on a search for "sleeping bag" singular as a result of stemming. We do not use the singular anywhere in the anchor text.

So one can only conclude that that "-das" search bypasses any stemmed results. It only applies to the exact keyword or anchortext phrase.

This seems so fundamental that I fear I missed something in the discussion. If so please set me straight.

 

This is a very interesting theory. I've had a number of links added in the past few months - a significant number, so I'll try and keep all posted regarding my experience with any significant impact upon rankings in the next few months...

 

No, Bob, I think you're right on target. Very well phrased.

 

Precisely Bob,

If this is the case, then the -dsa... search is NOT an accurate representation of the serps without "sandbox" as it is obstructing other very important factors in the algorithm.

It is not just adding full values to links. It is distorting other aspects of the serps as well.

View the difference in the two searches with Keyword Analysis Tool

 

If that is your point then I agree.

I'm sure the "-das" search is not so simply as to just remove all links from the sandbox. It is obviously interfering with the way Google is presenting their SERPs. If as we suspect it is interfering with the sandbox effect, who knows what else it is interfering with?

Only time will tell how good a prediction tool this search is, but I don't think you can expect it to give absolutely normal results other than removing the sandbox effect.

It is a bastardized search that I'm sure Google never anticipated. I wouldn't be surprised that they adjust for it like they did after Florida. But we shouldn't be looking for perfect results from it. It is an interesting curiosity at best.

 

Based on the SERP results in relation to McDar's site above, it seemed - to my great sorrow - conclusive that the sandboxed search did not include stemming, which the regular Google search does.

Therefore, today, I decided to check one of my own sites for stemmed phrases -- and the results I found makes yesterday's conclusion less certain, at least in my eyes.

Here are my results:

1) Normal phrase
allinanchor: #4 (929 results)
sandboxed: #5 (1.82 mill results)
regular: #21 (1.76 mill results)

2) Stemmed phrase (learning vs. learn)
allinanchor: not listed (149 results)
sandboxed: #19 (1.72 mill results)
regular: #20 (1.98 mill results)

3) Related phrase A (inner page)
allinanchor: #20 (67 results)
sandboxed: #42 (1.77 mill results)
regular: #293 (1.76 mill results)

4) Related phrase B (another inner page)
allinanchor: not listed (4.470 results)
sandboxed: #59 (1.3 mill results)
regular: #276 (1.2 mill results)

Related phrase B is plural of related phrase A, and of these four SERPs, the #1 is most competitive, followed by #4, #3 and #2.

Regarding search #2:
I checked the search for the stemmed phrase on another datacenter as well, and got #11 on both the sandboxed and the regular search -- and not listed in the allinanchor SERP.

Further: there is not a single link pointing to the index page with this as anchor text, and I think the term occurs one time in the text of the page. Google's stemming can, in other words, be pretty powerful.

Regarding search #3 and #4 (the related searches):
I included these because each of these pages has only one link from the main page (and only one of these links contains the search term). So, in some way or another, Google also gives weight to related search terms on other pages on the same site. Maybe ontology does matter?

As you can see, I've found both stemmed and related search terms using the sandboxed search -- something which is in contradiction to yesterday's findings.

Any comments in relation to this? If anyone has data/tests to confirm/contradict this -- or explain why McDar's site wasn't found yesterday using the sandboxed search -- I would very much like to hear about it.

 

NEW PAGE UPDATE:
*Google PR/Backlink update

The "New Page" sleeping-bags.htm
04/07/2004 "New Page" went live

Search for "Sleeping Bags"
_______________
NOTE: Due to size constraints, the entire report can no longer appear in a post.
You can view the entire report here... Complete Report
________________
Last 10 days

Position #15 allinanchor: #5 - 05/15/2004 [12 datacenters #19]Note: Allinanchor #6 on 12 DCs
Position #27 allinanchor: #5 - 05/16/2004 [11 datacenters #31]Note: Allinanchor #6 on 12 DCs
Position #28 allinanchor: #5 - 05/17/2004 [5 datacenters #33]Note: Allinanchor #6 on 6 DCs
Position #28 allinanchor: #4 - 05/18/2004 [2 DCs #25, 10 DCs #27, 4 DCs #29, 6 DCs #34]Note: Allinanchor #5 on 6 DCs
Postion #28 allinanchor: #4 - 05/19/2004 [8 DCs #27, 6 DCs #34]Note: Allinanchor #5 on 6 DCs
Postion #29 allinanchor: #4 - 05/20/2004 [9 DCs #28, 2 DCs #34]Note: Allinanchor #5 on 2 DCs **27 DCs are NOT responding**
Postion #27 allinanchor: #4 - 05/21/2004 [10 DCs #26, 2 DCs #33] Note: Allinanchor #5 on 2 DCs **31 DCs are NOT responding**
Postion #27 allinanchor: #3 - 05/22/2004 [1 DCs #25, 7 DCs #26] **31 DCs are NOT responding**
Postion #26 allinanchor: #4 - 05/24/2004 [1 DCs #24, 10 DCs #25] Note: Allinanchor #3 on 3 DCs & #5 on 7 DCs **29 DCs are NOT responding**
Postion #28 allinanchor: #4 - 05/25/2004 [3 DCs #26, 6 DCs #27] Note: Allinanchor #5 on 4 DCs **29 DCs are NOT responding**

_____________________________________________________
note: these pages are found using ( site:www.compar.com +sleeping Bags )
number of Bob's PR6 and PR5 links found:
56 - 5/18/2004 (All datacenters)
Unchanged from previous date
__________________
note: this page is found using ( site:www.komar.org +sleeping Bags )
Alek's PR 7 Link:
Not reported by Google 4/17/2004
Reported by Google 4/18/2004 (all datacenters)
Unchanged from previous date
__________________
note: this page is found using ( site:www.ski-france-ok.com +sleeping Bags )
Foxy's PR5 and two PR4s
0 - 4/20/2004
3 - 5/04/2004
Unchanged from previous date

__________________
Bob and Foxy added link Sleeping Bags to their sig files on this forum 5/06/2004
Results for - ( site:forums.digitalpoint.com +sleeping bags )
217 - 5/06/2004
241 - 5/07/2004
246 - 5/08/2004
901 - 5/09/2004 ( Looks like the sig links were picked up )
908 - 5/10/2004 ( 18 Datacenters report 1660)
1670 - 5/11/2004
1990 - 5/12/2004
2290 - 5/13/2004
2670 - 5/14/2004 ( 4 Datacenters report 2960)
2950 - 5/15/2004
3100 - 5/16/2004
3350 - 5/17/2004
3500 - 5/18/2004
Unchanged from previous date
3510 - 5/20/2004
3620 - 5/21/2004
3650 - 5/22/2004
3640[/COLOR] - 5/24/2004
3500 - 5/25/2004
__________________

Caryl and Foxy added link Sleeping Bags to their sig files on the SEO Chat forum 5/11/2004
Results for - ( site:forums.seochat.com +sleeping bags )
9 - 5/11/2004
10 - 5/12/2004
13 - 5/13/2004
16 - 5/14/2004 ( 4 Datacenters report 110)
108 - 5/15/2004
108 - 5/16/2004
181 - 5/17/2004
230 - 5/18/2004
232 - 5/19/2004
254 - 5/20/2004
298 - 5/21/2004
318 - 5/22/2004
435 - 5/24/2004
490 - 5/25/2004
__________________

"Sandbox" info - this search theoretically removes "sandbox" link filter
Results for - ( sleeping bags -dsa -dsa -dsa -dsa -dsa -das -das -dsa )
#3 - 05/15/2004
Unchanged from previous date
#2 - 05/21/2004
Unchanged from previous date
#3 - on 17 DCs and #2 on 14 DCs 05/24/2004
#2 - on 16 DCs and #3 on 15 DCs 05/25/2004

__________________

Caryl's Links [links posted 04/07/2004]
2 PR6s & 5 PR5's
__________________
PR/Backlink info

sleeping-bags.htm has PR5
49 Backlinks
(backlinks reported in the 49)
5 of Caryl's links
42 of Bob's Links
1 of Foxy's Links
2 for Digitalpoint forums
Nothing for Alek's PR7 link(the link on Alek's page is 53rd link - may validate the 50 link/page theory)
__________________

 

Both this test and McDar's are interesting, but I'm sure you both know that we are trying to draw conclusions from a way too small data set. We would have to repeat these tests on 50 or 100 different web sites before we could really get a general case.

The problem is we think of Googles algos as some straight forward arithmetic calculation. If this is so then results should be completely consistent and a pattern established from a single test would be meaningful. I think the algos must include a lot of bayesian logic and conditional statements. The result of this would be significantly different results for the same test on different sites, based on issues and items that we would never perceive or consider.

Let's for instance introduce just one simple variable, size. What was the size difference between the site that McDar tested and the one Rational tested? How do we know that that variable alone didn't trip some conditional criteria that resulted in handling the search results entirely differently? And of course size is only one of potentially 50 or 100 variables that Google may be looking at in deciding how to respond to the search.

I hope this post doesn't sound preachy, but I think this tendency to draw conclusions from a single point of data is tempting, but potentially misleading.

 

Rational,

As the experimental page is very controlled, I can tell you that the search term sleeping bag occurs 23 times on the page. the keyword density for the term is 10.02%.

The off-page factor is most likely the culprit. There is NO occurrance of sleeping bag (singular) as the anchor text. 100% of the anchor text is the plural version.

If stemming is effected by using -dsa... then that would be a big factor.

I am certainly not rying to be a stick-in-the-mud here. But, if we are trying to systematically prove/disprove commonly held ideas or theories, I think it would be irresponsible of us NOT to investigate this one as well.

If the use of -dsa... disrupts other aspects of Googles algorithm it is not an accurate indicator of where a site would be IF all of the current links had their full effect.

This is not to say that we cannot achieve #2 in the serps for the experimental page (as the use of -dsa... predicts), it is just that we will not achieve this simply because Google releases the full effect of our current links.

At best, it is an indicator that we might be doing much better, but I have heard some folks state that "the #(*) position is rightfully mine, if Google wasn't penalizing me." It is not that simple.

I have no doubt that Google is doing something to "dampen" the full effect of newly acquired links. AND it would be wonderful IF we had some gauge to view this with. But we would be remiss if we did not report that the use of -dsa.. is only a rough indicator and not entirely accurate.

Caryl

 

I agree with you both, Bob and Caryl, we need more data. And as time goes by, my hope is that we will either gather it ourselves or get it from others viewing this thread. After all, this is what we do, and I think we've made very good progress over a very short span of time.

By studying Google's regular search function, the allinanchor search and the sandboxed search, both isolatedly, and the relationship between them, I think we are learning a lot. This latter part of the discussion, for example, with focus on stemming has enlighten us. Now we know that we need to look out for this when conducting our research, and that this in particular needs to be investigated more closely. Who knows what will come up next?

The sandboxed search in general will probably only give us a rough indication, as Caryl mentioned, but even that is huge progress -- and very helpful. But that's in general... in some cases it is seems like it is going to be very accurate... what if we were able to isolate those cases and compare them with cases where (after 3 months or so) there still was a difference of 100-200 positions between the SERPs?

I think such comparisons could really learn us something. Even if there are a huge number of variables we don't know, I think lack of anticipated/expected upwards movement in the regular SERP despite a good number of backlinks with anchor text -- is because of something major, and when seeing these effects (lack of upwards expected movement) we might also find the cause.

By discussing these issues like this, and having different opinions, we also have a better chance of getting where we all want in the end -- namely understanding the market we are operating in better.

 

One thing I'll be very interested in finding out, by the way, is how the fact that all the links to the experimental page are coming from so few domains will effect the ranking in the long run.

If it doesn't move up, then maybe there is some extra dampening done on links from the same domain. In such case the sandboxed SERP position and the regular SERP position will never converge.

There are still a lot of interesting theories to work with and investigate. We will, in other words, have much fun also in the future

 

I'd be happy to start backing off my links if someone on a different web site with a different IP address can replace them with like quality links.

However again I don't think we should do anything until we think we are both out of the sandbox and have seen some stability.

 

Sorry to seem as tho I'm not here but I've got guests for these three days and will be free again-ish tomorrow so I should be able to comment by the weekend - in the meantime guys keep up the good work - it is obviously in good hands

 

Oh, I didn't mean to suggest that you should do that. I too think it is best to wait this one out as long as possible -- at least until it is reasonably certain that the (potential future) lack of progress is not due to the age of the links.

If there is still no progress after for example three months, then I'm afraid we can't blame the lack of upward movement on the sandbox any more -- and consequently will have to seek other explanations (and thereafter test them, one at a time).

 

Here's an observation I just made regarding links from the same domain, by the way:

Six days ago I sat up a new site and gave it apx. 750 links from another domain of mine. The next day Google spidered the index page, found the new link, and the new site was placed in position (apx.) #180 for both the sandboxed search and the allinanchor search. The next four days the site stayed between 215-220 in both these SERPs.

Tonight Google deep crawled the older site, and found apx. 600 new links. I've checked the stats a couple of times today, and the result is that the new site disappeared from the allinanchor SERP and was moved down to #455 in the sandboxed search.

There are also two other sites affected by this. The older site had already one link to an external site on each page. For this site, the allinanchor and sandboxed SERPs positions remained stable (#5 and #4 respectively). In the regular search it fell 6 positions, the largest movement so far (it generally moves 1 position, the largest before this was 3).

The site giving the links, which had been between 10-11 allinanchor the last 10 days, fell to 21, and the sandboxed search (also 10-11 the last 5 days), fell to 16. In the regular search the site fell 50 positions.

The results for the site giving the links may be skewed, since I at the same time as I added the new links removed the javascript PR channeling navigation. This shouldn't matter though, since Google followed the links in the javascript code ("Home", in the navigation was for example in javascript; I used the alt text of the logo to ensure that the right anchor text was associated with the index page). If this matters, then the internal anchor text of a site might be more important than I had imagined.

I will test this, and make other changes as soon as the results have been stable for a few days.

 

We need to be very careful NOT to ASSUME or attribute reasons for results that we have no evidence to support. This is how many of these "filter" theories get started in the first place.

There can be many reasons why we would NOT abvance further in the serps. If we look at the results of the Keyword Analysis Tool we can see that the size of the site is small, relative to other competitors. Also, it is evedent that manufacturers seem to have an advantage. Another reason could be that we just do not have enough backlinks period.

At this point, I would ONLY ASSUME some wierd dampening effect on current links if we had the competition beat, hands down, with links. This is nowhere near the case!

As far as I know, the sandbox theory was to do with a time delay between the discovery of a new links and that link being credited with it's full strength. The sandbox theory did not also encompass a permanent devaluation of links coming from the same source.

It is my take on this experiment that we are trying to present hard evidence that will challange some of the current thinking regarding linking strategies.

We are focused mainly on anchor text. Another aspect is the type of sites you receive links from (themed or related to your field). This is what we have controlled for.

Remember, evidence is what we are looking for. Speculation can be done with a lot less time and effort.

Yesterday, I was able to present evidence from this experiment on another forum when questioned whether 100% of the same anchor text would be penalized.

I was able to say...

http://forums.seochat.com/t10987/s.html

Caryl

 

NEW PAGE UPDATE:
*Google PR/Backlink update

The "New Page" sleeping-bags.htm
04/07/2004 "New Page" went live

Search for "Sleeping Bags"
_______________
NOTE: Due to size constraints, the entire report can no longer appear in a post.
You can view the entire report here... Complete Report
________________
Last 10 days

Position #27 allinanchor: #5 - 05/16/2004 [11 datacenters #31]Note: Allinanchor #6 on 12 DCs
Position #28 allinanchor: #5 - 05/17/2004 [5 datacenters #33]Note: Allinanchor #6 on 6 DCs
Position #28 allinanchor: #4 - 05/18/2004 [2 DCs #25, 10 DCs #27, 4 DCs #29, 6 DCs #34]Note: Allinanchor #5 on 6 DCs
Postion #28 allinanchor: #4 - 05/19/2004 [8 DCs #27, 6 DCs #34]Note: Allinanchor #5 on 6 DCs
Postion #29 allinanchor: #4 - 05/20/2004 [9 DCs #28, 2 DCs #34]Note: Allinanchor #5 on 2 DCs **27 DCs are NOT responding**
Postion #27 allinanchor: #4 - 05/21/2004 [10 DCs #26, 2 DCs #33] Note: Allinanchor #5 on 2 DCs **31 DCs are NOT responding**
Postion #27 allinanchor: #3 - 05/22/2004 [1 DCs #25, 7 DCs #26] **31 DCs are NOT responding**
Postion #26 allinanchor: #4 - 05/24/2004 [1 DCs #24, 10 DCs #25] Note: Allinanchor #3 on 3 DCs & #5 on 7 DCs **29 DCs are NOT responding**
Postion #28 allinanchor: #4 - 05/25/2004 [3 DCs #26, 6 DCs #27] Note: Allinanchor #5 on 4 DCs **29 DCs are NOT responding**
Postion #28 allinanchor: #4 - 05/26/2004 [7 DCs #26, 8 DCs #27, 5 DCs #29] Note: Allinanchor #5 on 3 DCs **29 DCs are NOT responding**

_____________________________________________________
note: these pages are found using ( site:www.compar.com +sleeping Bags )
number of Bob's PR6 and PR5 links found:
56 - 5/18/2004 (All datacenters)
Unchanged from previous date
__________________
note: this page is found using ( site:www.komar.org +sleeping Bags )
Alek's PR 7 Link:
Not reported by Google 4/17/2004
Reported by Google 4/18/2004 (all datacenters)
Unchanged from previous date
__________________
note: this page is found using ( site:www.ski-france-ok.com +sleeping Bags )
Foxy's PR5 and two PR4s
0 - 4/20/2004
3 - 5/04/2004
Unchanged from previous date

__________________
Bob and Foxy added link Sleeping Bags to their sig files on this forum 5/06/2004
Results for - ( site:forums.digitalpoint.com +sleeping bags )
217 - 5/06/2004
241 - 5/07/2004
246 - 5/08/2004
901 - 5/09/2004 ( Looks like the sig links were picked up )
908 - 5/10/2004 ( 18 Datacenters report 1660)
1670 - 5/11/2004
1990 - 5/12/2004
2290 - 5/13/2004
2670 - 5/14/2004 ( 4 Datacenters report 2960)
2950 - 5/15/2004
3100 - 5/16/2004
3350 - 5/17/2004
3500 - 5/18/2004
Unchanged from previous date
3510 - 5/20/2004
3620 - 5/21/2004
3650 - 5/22/2004
3640 - 5/24/2004
3500 - 5/25/2004
3470 - 5/26/2004
__________________

Caryl and Foxy added link Sleeping Bags to their sig files on the SEO Chat forum 5/11/2004
Results for - ( site:forums.seochat.com +sleeping bags )
9 - 5/11/2004
10 - 5/12/2004
13 - 5/13/2004
16 - 5/14/2004 ( 4 Datacenters report 110)
108 - 5/15/2004
108 - 5/16/2004
181 - 5/17/2004
230 - 5/18/2004
232 - 5/19/2004
254 - 5/20/2004
298 - 5/21/2004
318 - 5/22/2004
435 - 5/24/2004
490 - 5/25/2004
467 - 5/26/2004
__________________

"Sandbox" info - this search theoretically removes "sandbox" link filter
Results for - ( sleeping bags -dsa -dsa -dsa -dsa -dsa -das -das -dsa )
#3 - 05/15/2004
Unchanged from previous date
#2 - 05/21/2004
Unchanged from previous date
#3 - on 17 DCs and #2 on 14 DCs 05/24/2004
#2 - on 16 DCs and #3 on 15 DCs 05/25/2004
Unchanged from previous date

__________________

Caryl's Links [links posted 04/07/2004]
2 PR6s & 5 PR5's
__________________
PR/Backlink info

sleeping-bags.htm has PR5
49 Backlinks
(backlinks reported in the 49)
5 of Caryl's links
42 of Bob's Links
1 of Foxy's Links
2 for Digitalpoint forums
Nothing for Alek's PR7 link(the link on Alek's page is 53rd link - may validate the 50 link/page theory)
__________________

 

Wow! Rational, you need to draw me a flow chart of this one. I thought I was with you until you suddenly started talking about the old page. What old page? What is it's relationship to the new page?

And then you are changing multiple variables all at the same time.

I'm really lost on this entire explanation. Maybe a summary of your speculations would be helpful.