It’s been a long time since Moz last published an in-house ranking factor study, and also a long time since I last published one prior to joining Moz. In my case, this is partly due to my long-standing skepticism and caution around how studies like these are typically very loudly misinterpreted or misrepresented. There’s also the complexity and difficulty of quantifying on-page factors within Google’s increasingly nuanced and sophisticated interpretation of relevance (although, yes, we’re working on it!).
Nonetheless, I think there’s value in a narrower study (or studies), for a few reasons. Firstly, it can be useful to set a comparison point that we might revisit — perhaps if we notice a change in Google’s algorithm, or if we think a given industry or set of keywords might be untypical. Secondly, we might still wish to compare narrower sets of metrics — such as link vs. domain level linking factors, follow vs. dofollow links, or branded search volume vs. Domain Authority — and this, too, requires a baseline. Lastly, there’s some merit in reaffirming what we would expect to be true.
How to interpret a correlation study
It’s a cliché to say that correlation does not imply causation, but one that few seem to remember in this context. I’ve written before at length about interpreting correlations, but if you don’t want to go back and read all that, I think the main thing to check before you go any further is whether you can simultaneously accept all of the following to be true:
Links are a fundamental part of how Google works
Links are correlated with rankings
Building links may not always improve rankings
Sometimes links are a symptom, rather than a cause, of SEO performance
I’m not asking you to agree with all those statements, just to be open to this kind of interplay when you consider studies like this one and how they affect your worldview.
As it happens, though, whatever you or I may think, most SEOs do still hold that links directly improve rankings, which seems reasonable. But surprisingly, a narrow majority will not say this without qualification: this recent study from Aira shows the commonly-cited caveats of a lack of technical issues, and of some verticals not really benefiting.
What counts as a good correlation?
When looking at large datasets and very complex systems, any one metric having a non-zero correlation is worth paying attention to, but obviously some context is needed, and comparison between metrics can be useful for this. For the sake of this study, it’s probably more useful to compare correlation values between metrics than to get hung up on specific absolute values.
With all that said, then, let’s get into the data.
This study is based on the first 20 organic results for every MozCast keyword (10,000 keywords), on both desktop and mobile, from a suburban location in the USA.
Spearman’s rank correlation is used, as we’re comparing ranked variables (organic ranking) with logarithmic(ish) metrics like DA, and variables with extreme high-end values (like link counts). Using Spearman’s rank allows us to ask whether the order in which results appear is the one we’d expect based on a given metric, rather than getting bogged down in issues around different SERPs having vastly different distributions of link-count or DA.
Page, Subdomain, and Domain-level external links
In this chart, we look at how the number of links to a page’s domain predicts its ranking, compared to the number of links to a subdomain, compared to the page itself. Keen students of SEO theory will be unsurprised to see page-level links being by far the most potent predictor.
I’m sure this data will feel vindicating to SEOs and digital PRs who swear by building links directly to product or category pages, and they may have a point. However, there are a couple of things to keep in mind:
Often, homepages are the most linked-to page on a site. We shouldn’t be surprised to see homepages rank well in the SERPs where they’re relevant, and that is some of what this data describes.
You can achieve, from a PageRank perspective, a similar effect to direct page-level link building through the use of internal links. (Depending where your built links are pointing, of course.)
Links vs. Authoritative Links
This is perhaps another chart that more reaffirms what we’d hope than blows anyone’s mind, but yes, Moz’s DA and PA metrics — which look at the overall authority as well as quantity of links to a domain or page — do outperform raw followed link count.
That said, I may find this unsurprising, but plenty of brands and agencies out there still do KPI link building campaigns based on link count, so perhaps this chart will be of particular interest in their case!
Branded Search Volume vs. Domain-level
This comparison is an old favorite of mine, and illustrates some of the reasons why link-level factors are valued by Google in the first place: they were, originally, a proxy for popularity.
Those of you paying attention may actually be surprised that DA outperforms Branded Search Volume here. That does tend to be the case as you get deeper into search results. If we look at the top 10 only, you see lower correlations in general (due to the smaller dataset), but the ordering is a little different:
This is a similar finding to studies I’ve published before, and makes sense when you consider the competitive and data rich environment on the first page for competitive terms.
Does this mean branded search volume is a ranking factor?
Not necessarily! And this is the type of conclusion I was seeking to warn you about earlier. Brand very likely is an important part of what Google is trying to measure with links, as ultimately they want to give us results that we trust and want to click on. Presumably, Google’s engineers are not narrow-minded enough to think that links are the only way they could measure brand, given the wealth of data at their disposal, but whether branded search volume specifically is used is anyone’s guess. What we can see is that it very likely correlates with things that are used — just as DA is not directly used by Google, but correlates very well with things that are.
Similar to click-based metrics, there’s a semantic debate to be had here around whether something that Google is optimizing towards in its algorithm — but possibly not directly using as an input — constitutes a ranking factor.
Certainly you should not take away that your best bet is to directly manipulate branded search volume by generating a load of artificial searches. That said, naturally causing people to search for your brand, especially in conjunction with relevant product terms, can only be a good thing. Whatever Google is measuring (whether it be links, search volume, clicks, or anything else) is likely to be improved by the same activities you’d use to naturally raise branded search traffic. Which is, of course, probably why it correlates so well.
No major shocks: “links correlated with rankings, SEO study finds!”
But, there are some important reminders here:
Page-level performance is important, however you go about achieving it
Raw link count isn’t a great metric
Demand for your brand is at least as good a predictor of rankings as domain strength on the first page
Like I said above, though, please do remember in any incendiary tweets you’re now penning that the relationships behind these correlations can be more complex than meets the eye!