This week, Rand is joined by one of our in-house data geniuses, Ben Hendrickson, to talk more about some of our recently released correlation data to support guidelines for SEO best practices.
While correlation doesn't always equal causation, it's still very interesting to look at the attributes and features high ranking sites tend to have in common. Comparing this data to known and accepted SEO practices can help to reinforce widely held notions or give us some insight into how the algorithms are changing; both important areas of analysis for successful online marketers.
SEOmoz Whiteboard Friday - Correlation, Causation & SEO from Scott Willoughby on Vimeo.
Want to learn even more about what Ben discovered in our data correlation analysis? He'll be presenting his detailed findings at our SEO Training Series Seminar right here in Seattle, August 24th & 25th.
Ben Hendrickson is definitely a name to remember! :) But he seems to have problem explaining things in a simple way. Rand, do you think you could post some of the data in the PRO content with Ben's fully statistical explanation? For those who understand statistics...
I liked the depth Ben added to the video. I was nice to have someone that knows his stuff to comment about the data :D
As Tybi suggested it might be nice with a video where Ben goes though his charts and explain his findings.
Yet better, written report.
Yeah - I think we can definitely get something into PRO very soon :-)
You might want to check out my presentation on this topic here, too.
Good on you Rand and Ben! Thanks for sharing this
"Having bad sites linking with anchor text" brought my attention.
Firstly:
what scale are we talking about? the site I'm looking after is PG3 has got around 700 links out of which quite some are from low pagerank private blogs etc. (though still within the "related field"). As I just started doing SEO on it, there's a lot of higher priority work to be done, but I'm eager to know the impact od those low pagerank links. Also, how bad is it when it's a targeted keyword anchor text or just a domain name?
Secondly:
If for some reasons I am not able to reach the webmaster of the site that is linking to us, how do I guard myself against having that link pointing to the site?
While the data may indicate that these are best practices for rankings, I hope everyone can see that they're not necessarily best practices. After all, rankings aren't (or at least shouldn't be) the ultimate goal.
Let's say the fine folks at Schmeckgut Confectioners want more business for the query "exotic chocolate truffles". They could put up a page with a title of "Exotic Chocolate Truffles - Schmeckgut Confectioners" (keyword - company name), and following the other best practices maybe they'd get to the top of the SERP.
But what if they decided to use the title tag for more than just rankings, and went with something like "Exotic Chocolate Truffles To Die For - Schmeckgut Confectioners"? Even if that pushed them down to 3 or 4 on the SERP, it wouldn't surprise me if they ended up with a higher CTR.
Gladstein -
If achieving top rankings isn't one of the ultimate goals in your business model I'd be happy to see you lay it out here what exactly it is you're telling your clients. Also, your explanation here is that optimizing a title with "Exotic Chocolate Truffles - Schmeckgut Confectioners" is better then "Exotic Chocolate Truffles To Die For" for rankings but not CTR and that you'd rather be ranked #4 compared to number #1 with a title like the second one is questionable reasoning in my opinion.
For one, I would never advise my client to do anything that would hamper his/her top rankings for a 'theory' that this title may have a higher CTR. Secondly, I've personally seen and can vouch for studies that have been made on the CTR on organic placement e.g. #1 gets approximately X amount of hits per month #2 gets approximately Y amounts of hits and so on from analytic data of sites that are ranked 1, 2, 3 and 7(#7 fluctuating) for over 2 years.
I know there's room for argument because the results would vary from industry to industry, keyword to keyword, and other factors e.g. "Exotic Chocolates Truffles To Die For" but approving a method that you believe in or assume and could decrease your visibility (Microsoft Eye-Tracking Study of Informational Searches & Enquisite) e.g. position one has more visibility compared to position 3 or 4 no matter what the title says.
In saying that, I don't feel the first title is better then the second (unless the business brand name is deeply woven into targeted markets) and I don't believe you would achieve higher rankings with the first title then you would from the second if you were trying to rank for "Exotic Chocolate Truffles", in my opinion.
Gladstein If achieving top rankings isn't one of the ultimate goals in your business model I'd be happy to see you lay it out here what exactly it is you're telling your clients.
It's simple: the ultimate goal is the success of my client's business: conversions, sales, ROI, recommendations and reputation.
The example I gave above was off the top of my head, so I'm not making any absolute claims about what the rankings or the CTR would be. I didn't even look to see how competitive the query is. My point is simply that ranking is just the first step in the process, and if getting a top ranking requires a page that's not friendly to people -- that doesn't lead to sales -- then that ranking is useless.
I would agree with this assumption to a certain extent - it's not all about rankings. However, rankings play an extremely large part in the equation.
If, "conversions, sales, ROI, recommendations and reputation" is what you're trying to obtain for your clients, advising them to (If it were possible) to have position #3 or #4 with a title as you mentioned earlier directly conflicts with achieving conversions, sales and ROI. I'd also like to point out that you and I are talking about the title so, if being ranked #1 gets more CTR then being ranked #4 and both sites had the same type of content, besides the title tag of course I think you would serve your clients best if you helped them see it from this light also.
I'm not suggesting that being in a lower position on the SERP = a better ROI. I'm saying that if the goal is a better ROI, there are many factors involved, not all of which are about the number and relative positions of words and characters within a given element.
Every case is different. If I get a client's page to #1, and its conversion rate drops, I've failed. I can tell them, "I've got your page ranking at the top," but that isn't going to help them meet their payroll. If I make changes that improve the conversion and clickthrough rates, even if the page now appears lower on the SERP, I've succeeded.
The point is not "don't go for the top of the SERPs." The point is to do what's best for the client, test your results, and apply that information. If that means not following to the letter instructions that tell you to structure the content of a given tag in a particular way, so be it.
So, if you get fantastic results (beyond just rankings) with a title tag of [keyword phrase 1] - [keyword phrase 2] - [Company], that's great. Do it. I'm not saying that can't or won't work. However, if you find that you get better results with something else, don't beat yourself over the head because the formula is sacrosanct and must never be messed with.
The real audience of a page is people: the potential customers of your client's business. For a lot of people, the result at the top of the SERP is worth checking out, so it's absolutely a worthwhile effort to try to get to the top of the SERP, but that isn't necessarily enough to convince them to buy. If something beyond just keywords can help you whet their appetite while they're still on the SERP, and that change doesn't push you down too far, it might be worth trying.
In short, this post is about the science of SEO. All I'm trying to say is that it can be very beneficial to mix a little art in with that science.
Although I certainly think you're right that best-ranking title doesn't necessarily equal best title, I think the whole point of this kind of analysis is to try to separate out the factors, which means making an "all else being equal" assumption. Once you isolate the data and can start to look at the contributions of individual factors, then you can use that information to piece together the larger puzzle.
In other words, this isn't the end of the story, but that doesn't mean it's not very useful information.
Ever thought about releasing a search engine in competition to the big three!! Perhaps call it Linkscape.... Dr Evil....
seomoz pro community building a search engine perhaps?
happy to do polish localization if this ever happens:]
Nice to see that data like this confirms what is already being practiced for most.
It would be interesting to see your findings on domain age, the number of years registered, and number of years (or days for some) to expiration...curious how strong the correlation in fact is.Â
@RChurt
I guess the domain age would be correlated with rankings, but most probably only because old domains had more time to attract backlinks.
I'd second that. =oP
As an experimental psychologist, I wanted to leave a brilliant comment about correlation and causation, but since it's Friday, all I could come up with is: Ben is tall.
LOL, hilarious!
Hi,
Ben is good. But I think coding is not talking. It is difficult to grasp it without good explanation.
I wish someone would explain it better that what he has in mind from developing point of view.
When you code most part of your life, it really becomes hard to explain what you coded.
This is what happens to him.
Overall it is is a great video.
Thanks for sharing honestly.
Most interesting for me personally was the descending importance of keywords in the title tag, and the importance of the alt tag (which I don't think our company is utilizing nearly well enough and that will be changing on Monday).
One subject that I'd love some research on (and it's probably already out there somewhere) is how big of factor does the age of the link play into this equation.
Ben Ben Ben Ben! He has a future in Whiteboard Friday's for sure. :)
I like Ben too. He's smart, he's skeptical. He's more interested in what's right than marketing. He's real.
OMG, I just had the laugh of my life ;-)
 You have to watch the last 3 seconds. Good bye ;-)
 Very interessting stuff, though.
No kidding the last 5 secs is the best part. Ben, You are a rockstar of data analysis, stats, and delivering a solid explanation and I think Rand was doing a practical joke by messing with your iphone teleprompter : )Â
. . . It does exist
https://www.ipodnn.com/articles/09/04/22/iphone.becomes.teleprompte/
Take a cue from my life hero Bill Oreilly and "Do it live, #$@ it we'll do it live"
Interesting. Another option besides a prompting device would be if people could just convince me that the camera wasn't "turned on" :-)
Rand are you tiny or is Ben a giant? Maybe stand on a box for the next one ;)
Good video still.
Ben is a giant. I'm 6'5" and he makes me feel short.
Good stuff here... would love more hard numbers and graphs here though. Did I just miss them? The Title tag graph was great. I want to see that ALT tag graph!
Definitely understand the disclaimers here - I use this argument all the time with Brand in the TITLE tag. Sure putting the brand first and KW second is not as good purely for rankings as the other way around. That said, if you brand is Nike or something, put that first. From what we've seen, the boost in CTR is worth the tradeoff in pure ranking power. Especially considering most of these household brands already have lots of link juice so they can afford to slightly under-optimize on page elements a bit if it is better for the big picture.
Totally understand correlation vs. causation... any good causation numbers you have here? I think people would be willing to pay / join / link / whatever for that data :) If I had unlimited time I'd take a few months and just rock out tons of totally controlled tests. But alas, I don't have such time. I trust you guys to do a great job of this though...
I think we want to balance between keeping this data behind the PRO wall (for some in-depth stuff) and pushing it publicly so everyone in SEO can benefit. For example, Nick's new Labs Visualization tool leverages the ranking models data to build his radar graphs :-) We think uses like that will probably stay behind PRO, but advice like "H1s may not be that awesome" should be open to the whole community.
Some interesting work. Thanks for sharing.
Some questions:What's the profile of the sites used? Are they your clients? To what degree are they big brand names? If the profile doesn't match a readers', any conclusions might not be relevant.
Eg, the effect of 'brand' in page title tag. This might be different for big brand and non-brand sites. Did you 'test' 'no brand in title tag' for sites with no brand presence.
Glad to see some data backing up keyword prominence idea. In order to get the point across to readers, I moved my main keyword to the end of my title tag instead of the beginning, this resulted in a loss of 20 places, I moved it forward the week after and whalah, back to it's usual position.
I'd love to see more of this data on SEOmoz, especially on the anchor text as all research I have done in the past places anchor text as one of the top factors related to higher rankings.
Either way, a worthwhile watch :)
Was this filmed in an Ames Room?
Hey! I'm just short, OK... :-)
Hey, great video and post. Enough information my head is spinning. I'm new to SEO and have a quick question. When optimizing title tags and keeping it below 66 characters, are the "<"'s counted? For example, is "<questions>" 9 characters or 11?
Great Whiteboard Friday.
I am really looking forward to seeing what correlation anchor text and title tagging the links have in rankings.
I have for a long time used this as a way to assist the page the text link is pointing too to be found for the specific anchor text phrase. From what I see this works very well.
I have done a lot of testing on this and would like to see what you come up with from your own tests.
Keep up the good work. Â
This is very intersting about correlation of different parts of the page with google ranking as far as h1 tags and meta data. I have been watching as many of these google webmaster videos as I can and I heard the same kind of thing that h1s may not matter as much as a the page having a good outline. And while its best practice it may not matter that much.
https://www.youtube.com/watch?v=GIn5qJKU8VM
https://www.youtube.com/watch?v=iR5itZlq8sk
That's some mighty fine Whiteboard Friday, fellas. I guess I didn't know that the closer the keyword to the beginning of the title tag the better. This is why I keep coming back everyday for more SEOmoz. Thanks again!
Slightly more intelligent, post-coffee comment. I know some of this is proprietary, but have you guys done much in the way of multivariate models? I'm curious, because it seems like there are a lot of dependencies in certain SEO factors. For instance, having an effective title leads to better ranking and CTR, which naturally leads to more inbound links, creating a dependency. I think it would be fascinating to combine 10-20 of these factors in a multivariate model and see which ones come out on top.
Brilliant topic. I love the data based approach to affirming/disaffirming best practices for title tags. I'm really excited to see/read more about this research as it evolves.
Maybe a top 5 conclusions list from Ben's data in the near future?Â
Great study, good video. BTW, what exactly is antidotal evidence, is that evidence that fixes a poisonous problem? :-)
I was wondering the same thing...  lol
Hello! I mean, goodbye!
It's the evidence of the poison you just drank, Dr. Jones...mwah-ha-ha-ha-ha."
Very interesting analysis.. Can you please go more in the methodology used and some graphs showing the correlation data.
Thanks
Great Whiteboard Friday guys! Excellent reminder of the complexity of drawing conclusions from correlational data.
Speaking of which, I'm getting ready to release some correlational data I've collected on the factors influencing site traffic. Should prove interesting.
This WBF was worth every single minute of it, particularly the stuff about long urls. I also wonder at what point can a url be tagged by google as 'too long'. I tend to deal with really long urls on the government website I do 'inhouse' for and have never noticed a problem, though I have admit I haven't made any benchmarking or experiments. However, for smaller sites with less trust and authority the story reverses and yes, I have experienced difficulty to outstrip the competition and rank higher, even when critical terms were used along the urls. Anyway, Ben, Rand, thanks for this WBF. Excellent stuff.
.gov has more authority than .com, naturally - therefore, your urls are predisposed to more authority than a .com with the same url.
But to flip that situation. If you're up against a seasoned SEO who's working for the competition. Ranking higher than them may still pose a problem due to the experience of the competition. That's the beauty of SEO, it uses all the factors not just Title and URL for rankings. So, with the right moves an seo expert can out manuver competitors with predefined advantages.
I was a little confused on one thing, did the "long urls hurt" comment refer to volume of characters or number of url variables? I think Ben meant volume of characters and would appreciate any insight as to where the pain begins to show up. 15 characters? 30? 50? Is it linear?
Another great WBF!
My personal experience is not only having long URL in path e.g. domain.com/my-long-%*^-keyword but also having long domain names.
And this might well be a correlation we don't want to confuse with causation. Eg, long URLs are often associated with deep pages with much less link juice.
I would bet that it's a dilution issue regarding the keyword specified. Obviously, the more words used in a URL the less effective the keyword is overall.
It isn't linear, although I haven't made a chart for it yet. Danny also asked for that last week and I told him I'd produce one somewhat soon, so it is on my queue.
The regression we did was with characters in various parts of the url. It is possible they actually try to tokenize the URL and so what matters is the number of "word like things" and not characters. I didn't make features for that, and so can't say if that would have been closer.
Regarding Mark Nunney's point, the correlation with long paths was in a regresison including features like mozrank, external mozrank and toolbar page rank. That means as those are closer to the juice feature used by the engines, we wouldn't see a correlation also with the path length. But his point is still good, as it is plausible there could be some link based algorithm different than the features we used that is fairly well correlated with url length. For instance, we don't have any sort of "juice from the front page" metric, which I can imagine might be used and might be rather correlated with path length.
So ditch the H1's and hit the alt tags...got it.
Hello, goodbye I guess...lol
Just kidding. Thanks for sharing your research.Â