MAY 21

Flow Metrics vs Moz Metrics vs Page Rank

Filed under: SEO
May 21st, 2012

It looks like this is your first time here?

Please subscribe to my RSS feed! Have a good read...

MajesticSEO have launched a new set of link metrics, based on iterative flow of data through links. I know I am not fooling anyone in the know that I am impartial – as I am Majestic SEO’s marketing director, but I’m also a MozPro user as well.

So I asked one of our team to go to a random Wikipedia article, take every third word, put the word into Google and select the third result 50 times. I then asked him to record the following data for every URL in his list, so we could all start comparing the new metrics with Moz metrics and with Page Rank. Here’s the resulting list with the following:

Correlation Tables with Page Rank summary:

Domain Authority	.787
Moztrust	.119
MozRank	.014
Citation flow	.814
Trust flow	.746

The strongest correlation is Citation flow in our study, even though Citation flow does not try to emulate Page Rank. Domain Authority comes in a close second, but at a URL level there really is no competition. Trust flow is not aiming to correlate with Page Rank but it is interesting to look for yourself to see sites with a high Domain Authority or Citation flow but with low Trust flows. Trust flow is something new – and very enlightening.

(Download the full URL list in this W ord document)

In actual fact – Citation flow is in several regards a stronger metric than Page Rank because:

It updates daily – not once in a blue moon
It is “pure” in that it is not affected by manual penalties in Google
It can be calculated at the URL, subdomain and root levels – whereas Page Rank is only per page.
Links are not created “equal”, because it lets page strength flow over multiple iterations

(Download the full URL list in this W ord document)

page rank comparison < Here it is in Excel with Page Authority added.

Be Sociable, Share!

13 users commented

Hey Dixon – thanks for putting this together! Excited for you guys that citation flow has such a high correlation with Toolbar PageRank.

Something looks a bit weird – in all of our correlation studies previously, mozRank had quite a high correlation (in the 0.8+ range) with Toolbar PageRank (once you control for the additional granularity of two decimal points, which TBPR doesn’t show). Domain Authority has never matched up well to TBPR, so this is a very odd result (since DA is not at all similar to PageRank in how it’s built and is sitewide rather than page specific).

I’ll ask some Moz folks to see if they can reproduce. It would be strange indeed to find that mozRank is no longer correlated with PageRank well but some other algorithms are!

said Rand Fishkin on May 21st, 2012 at 5:31 pm

Hi Rand – Very happy to redo this study if it’s wrong. I am having a fight with the internet, here in Posnan – so the Moz toolbar doesn’t seem to be getting through so can’t check if we used the right numbers… Until I can check – I’ll add a big note at the top warning people that I may need to check this research…

said admin on May 21st, 2012 at 6:27 pm

Hi Dixon – we ran a few tests on a larger dataset (~4200 URLs).

First off – big picture is that you are correct! Since the April PR update, the mozRank scores (which are from data crawled Feb/March) are a bit old and thus the correlations are lower. Meanwhile PA/DA are looking higher (which is weird – we didn’t expect those would have a good correlation, but who knows!?).

From our team:

~4200 PageRank values from a random sample of URLs. I ran Spearman correlations from Index 51 against toolbar PageRank and got:

mozRank: 0.45
external mozRank: 0.45
mozTrust: 0.66
PA: 0.68
DA: 0.37

A few things about your methodology in the XLS files (thanks for providing those BTW!):

- mozTrust correlation was actually negative (not positive; bad for us, good for you )
- You used URLs for which Mozscape has no data, but no URLs where Majestic or Google had no data, which we kinda think might be an unfair bias
- 50 URLs is likely way too small a sample set; ideally, we’d want to do something in the 10s of thousands
- You used Pearson’s correlation coefficient, which as several mathy folks pointed out in our previous studies in 2008/2009, is non-ideal. Spearman’s correlation would be appropriate here.
- Accounting/controls for rounding and granularity weren’t quite clear in the methodology. We like simple rounding (which is close), but we’d both have to know how Google does rounding with PR to really get accurate.

We’re actually going to buy the Majestic API and do a larger scale project in ~6 weeks (when our data science folks have some bandwidth). I bet these new numbers from Majestic are going to look pretty good – seems like you guys have put together something awesome

said Rand Fishkin on May 21st, 2012 at 11:01 pm

Hi Rand,

Thanks for checking. I was worried my guy had collected te wrong data.

Although my numbers vary from yours, seems like they get to similar conclusions right now. You are right that I am using a straight correlation and not Spearmans. Two reasons. One is time (takes seconds on Excel) but the other is that Spearmans in SEO is strong when ranking things – which is something we SEOs love to do – but I am not trying to order these numbers, so did not feel that Spearmans was any better and in some ways worse in this instance.

I’ll defend the URLs for which there was no data on a couple of counts too. The first being that I gave Michael a methodology for finding URLs which I felt would be out of my “control” to influence, so was not trying to consciously find blank urls. Second, Excel ignores lines where there are not two numbers to compare directly – so the lines where there is no data simply reduces sample size of the data set, not the correlation itself. That said – I also agree that having urls for which you have no data is kinda mean. Sorry about that.

Tom also criticized me for only using 50 urls. I’ll take that criticism – but I think 50 random(ish) urls is enough to test data if you then expect others to repeat rinse and wash the same test. If you want to test the ph value of a lake, you usually don’t need more buckets of water to get an accurate reading, even though there could be unknown acid poured in from the other side of the lake. It is only the degree of certainty about the correlation that increases or decreases with volume – not the correlation itself – but I am also hoping that it spurs people into doing similar analyses for themselves. I worked with a larger data set but the source list was not mine to give out. I also managed to break a little tool that I had built which pulled in all the necessary stats from the APIs. I expect Google didn’t want to play, so went manual in the data collection.

On the methodology – I simply used the absolute integer numbers for PageRank (Toolbar). I am a simple man.

Yes – thanks for the nice comment at the end. We are obviously pleased as punch with flow metrics and there were some very sleepless nights over several months from Alex in particular. I know crowing about them is not really “cricket” as we say here – but if I have to crow about them… it’s my job and you guys are the only players in town with metrics robust enough to compare against. I’m sure you guys will be coming right back at us, soon enough.

Congrats on the funding by the way! WAY cool and now we know you’ll keep us on our toes.

Best,

Dixon.

said admin on May 22nd, 2012 at 12:08 am

Hi Dixon Jones,
Thanks for this great post !
The co relation between PR, Flow Metrics and Moz Metrics is really important.
Flow Metrics is the graphical representation with the help of charts.It helps in creating the graph of domain link profile.The Moz rank is also important which represents the link popularity score.Both these metrics are related to get a high PR.

Thanks again
Regards

said Arpita on June 15th, 2012 at 6:58 am

This are some great news. I expected them to present something new
considering the rate at which they developed lately. I’m
looking forward to test this update…

Thanks!
Regards!

said Mobila Comanda on October 29th, 2012 at 7:58 pm

Albeit on a very superficial level, if you graph descending page Rank with the matching Citation and Trust flow there is no correlation. In fact the declining line of the Page Rank contrasts the “U” shape of the other two.

Bearing in mind that the Citation flow and the Trust flow have the purpose of measuring different things, I am a wee bit concerned at their pretty high trending correlation which is very evident in graphical format

Cheers,

Mike FD

said Mike FD on March 17th, 2013 at 10:31 am

Hi Mike,
This test was carried out a year ago. Since then, Google has updated its Page Rank once I think (and Majestic will have updated Flow Metrics > 100 times) so things may well be different. It now seems that sites in the west correlate better with Trust Flow and Sites in Russia still correlate better with Citation Flow, but that is anecdotal only,so perhaps it is time to revisit this.

As you say, though, they do measure different things. Flow Metrics are not trying to emulate Page Rank. It would seem, now, that high Citation Flow and low Trust Flow is a red flag.

said admin on March 18th, 2013 at 7:58 am

As a web-site owner. Do you think like owning a web site or perhaps internet site is profitable?
Also how would you get individuals to stop by your web sites?