May 2020 Core Update: Help me with my analysis

Hey guys,

As some of you know I run a pbn network here on bhw, so I have access to quite a few sites and it's given me some insights into the update.

However, a good chunk of my clients are local businesses who haven't been affected too much by the update. I have some ideas as to what's behind it, but I need more data.

If anyone wants to help out I would appreciate if you could pm me your site. I need sites that had strong drops or strong rises since the may 5 update.

I can't do this with random sites, because I need to be able to ask you questions to get more data that I can't get from random sites.

The more the better. If I can get about 20-30 sites that had a big drop, and 20-30 sites that gained positively then that would be extremely valuable.

And remember, my main business is running the wolf network, so people naturally share their sites with me all the time. I'm an agency and I make my money from selling links and advising my clients. Data and contacts are far more valuable to me than anyone's niche.

Anyone who does help me out with this can also get some advice from me on their site or about any aspect of SEO in exchange for the help.

And remember, it has to be sites with significant increases or drops. I'm looking for the outliers. Generally a drop of 50% or more, or an increase of 20-25% or more is ideal.

I'm going to pick a series of data points for the sites, based on on-page data, backlink data and the information you give me, ie, is it an aged domain, fresh domain, is your content native, non-native, from an expert, general writer etc. Then I'll do a correlation analysis on each data point to look for positive and negative correlations, and also calculate if it's statistically significant.

The results will be shared in a post here.


Thanks, guys.

Are you gonna share the results publicly or you gonna use them for your business?
Can i trust you with the niche?
How can i manage that you can really help in return?
I had 2K/day ---> 600/day drop, I might help, more details if you can (about what you are going to analyse)

hideath said:

Are you gonna share the results publicly or you gonna use them for your business?
Can i trust you with the niche?
How can i manage that you can really help in return?
I had 2K/day ---> 600/day drop, I might help, more details if you can (about what you are going to analyse)

Click to expand...

Read the OP complete and you'll see where he said the result will be shared in a post on the forum.

I don't have a site that meets the criteria described, would have submitted.

I don't have sites that meet the criteria. Sorry. But I would love to follow this thread so I can get the insight. Thanks.

hideath said:

Are you gonna share the results publicly or you gonna use them for your business?
Can i trust you with the niche?
How can i manage that you can really help in return?
I had 2K/day ---> 600/day drop, I might help, more details if you can (about what you are going to analyse)

Click to expand...

I'll be sharing the results in full here on bhw as I always do with my guides.

Yes, you can absolutely trust me. I'm a professional and my business and reputation are built on helping businesses. It would make no sense for me to give up my 80 hour a week business I'm passionate about to jump into some other niche. I love my work and I make good money. I don't need niches. I need things to write about to spread my name and get more business. My name is my business.

The offering help for you is just giving you access to me to ask me some questions and get my advice on your site. I am not saying I can fix anyone's site. I'm just offering my knowledge/advice in return for people's help, but the biggest value is the analyse. I'm going to a proper statistical analysis to get some hard data. It may turn up nothing. Data is data. All I can do is work with the data I get and see what kind of insight we can get. Our power here is as a community. With single sites it's impossible to determine what went wrong, but looking at dozens, I am confident we'll see some interesting correlations.

I am going to look at single data points and compare them with the percent drop or increase in traffic and then do a correlation analysis.
I'm also going to plot histograms of the data points for different levels of increase and decrease to get some further visual insight into patterns.

Some of the data points I have in mind are :-

Aged domain
Expired domain
Fresh domain

Native writer
Non-native writer
Expert writer - writer understood the niche well and wrote with technical language (Hence why I do need to ask the site owner questions)
non-expert writer - general writer, not particularly expert in the area

has brand searches
no brand searches

Then ones with affiliate links/ads breakdowns
Say 5 levels, ad heavy down to ad light, and I'll classify based on number of pages with ads, and typical ads/affiliate links per page

Then commercial intent data points
Heavy, medium, light - Where heavy is 70%+ of the site is high commercial intent pages, 40-70% is medium, < 40% is light. That could be broken down further if there's interesting data.

I'll need to think up some more, but these are the type of data points I'm interested in. I'm not interested in things related to anchors, content length etc as these definitely don't have anything to do with the latest update. The latest update is going to be related to the machine learning in the algorithm and google trying to match user intent with pages better, so these will drive my data points.

Details sent

I don’t have any that meet the criteria otherwise I’d send you the info.

I had a bad hit on this update which in my guess was because of some cheap PBN links I bought for a page on the website. Interestingly, only the page which had these PBN links got kicked out of the search engine but rest of the other pages on the same website which had legit links and diversified links are still ranking strong.
This keyword is for local business.
If need any other kind of info on this, let me know. I am interested in studying more details about this G update too.

Looking forward to your detailed analysis hopefully not a long a wait

I think it's a good initiative all the way, and senior members must jump in at the opportunity if the OP delivers as promised.

tousif1988 said:

I had a bad hit on this update which in my guess was because of some cheap PBN links I bought for a page on the website. Interestingly, only the page which had these PBN links got kicked out of the search engine but rest of the other pages on the same website which had legit links and diversified links are still ranking strong.
This keyword is for local business.
If need any other kind of info on this, let me know. I am interested in studying more details about this G update too.

Click to expand...

It's interesting you replied, because after I created this thread and headed to bed, I was thinking that very thing.

What I'd like to see is, is it targeting just specific pages, AND, what are those pages, are they from high CI keywords, pages with content that perhaps doesn't match searcher intent so well or pages with too many ads.

Send me a pm.

Cshark said:

I think it's a good initiative all the way, and senior members must jump in at the opportunity if the OP delivers as promised.

Click to expand...

I definitely will. I've already grabbed a copy of IBM SPSS for this.

But.

I *NEED* data.

This will NOT work if I only get a handful of people working with me on it.

I cannot do this kind of analysis on random sites. It's extremely limited. If I had 100+ sites to look at data we would have a very powerful analysis.

The kind of data I'll get from ahrefs etc is not that valuable for these types of updates. It's ok for certain types of analysis, but generally you need specific data from the site owner.

What I'm going to do is create a form for people to fill out so I have a systematic way of collecting the information I need.

There's a reason the kind of analysis you get on websites where they look at a bunch of "winners and losers" is near useless. They have access to very limited data and they categorize things so broadly.

They look at such old useless metrics and do very basic correlation analysis. There are so many powerful statistical analysis methods available, not to mention machine learning algorithms.

Looking at things like content length, keyword density, number of h tags is all complete and utter nonsense and will never yield any viable results. Google is so far beyond this. They are very heavily using machine learning to match pages with keywords, not keywords on your page.

They are using machine learning to understand what type of pages people want to see for certain keywords/topics. For example if someone searches for X, then Y, then Z straight after, THEN clicks on something, they understand that the results for X and Y weren't useful. Over time they learn through machine learning what kind of pages should be served for users searching for certain keywords/topics.

So we want to be looking at more general data points and rather than looking for correlations with a spearman correlation analysis, we should be doing a difference between groups test with different numbers of independent variables and analysis like independent samples t-test, one-way ANOVA, two-way ANOVA, three-way ANOVA and factorial ANOVA. Things like "expert writer vs non-expert writer" for a topic. Native vs non-native content. Looking for differences between the groups. This is very different than just doing correlation analysis, and it works better for updates where I have cases of strong positive increases in traffic vs strong negative increases. We don't care about rankings. Just traffic, which is why I can't do this without your help.

So for example here's 1 type of analysis we could do where we would use an independent samples t-test.

This is where we have 1 continuous dependent variable. Traffic.

And 1 independent nominal variable(nominal is a category type, like native vs non-native rather than a continuous like traffic, distance, speed etc)

The independent samples t-test works if your dependent variable is continuous, and your independent variable(IV and DV from now on) have only 2 categories. Ie, native vs non-native. If we have 3, expensive native, cheap native, non-native, we would need to use one-way ANOVA.

This type of analysis will tell us if there's a statistically significant difference between the means of the 2 groups.

I have *never* seen this type of analysis done in SEO before. Everyone just does very basic, naive correlation analysis, and they don't do it right. You can't just randomly pick spearman or pearson correlation analysis. You choose them based on your data. pearson is for 2 continuous variables. spearman is for ordinal. For your typical seo rank analysis, ie, does moz DA correlate with rank, you would use spearman, because google rank and DA are ordinal.

Anyway..

We could get some amazing insights here, folks, but I need people to step up. So far I've only had 2 people pm me.

If you guys want to see some hardcore, bad-ass statistical analysis based on this update, then step up. You'll never have access to something like this. A) Most people don't have the math background to do this and don't even understand statistics at the most basic level, let alone enough to setup a study and choose the right data points and test types, then interpret it properly.

This only works with traffic. Rankings are ambiguous. Traffic is ABSOLUTE and it's a cumulative effect of your rankings, therefore it's an absolute snapshot in time of your site based on how google sees it. A single keyword ranking means very little as you're only analysing 1 keyword on 1 page which doesn't give you much information at all.


So you guys have a chance at getting access to something unique and rare, and this study is only the first. With your data I can do tons of interesting statistical analysis and we can all gain more insight.

Even doing correlation analysis with traffic data is useful. We can also combine your data with things like the google natural language API which lets you classify content. The possibilities are endless if I have enough data. Specifically site + traffic history.

This is interesting indeed.

I'll send over a couple of my sites via Skype.

Why not just scrape 10.000 websites and analyze their rankings prior and after the update?

laur.laurix said:

Why not just scrape 10.000 websites and analyze their rankings prior and after the update?

Click to expand...

Because as OP mentioned in his first post

splishsplash said:

I can't do this with random sites, because I need to be able to ask you questions to get more data that I can't get from random sites.

Click to expand...

laur.laurix said:

Why not just scrape 10.000 websites and analyze their rankings prior and after the update?

Click to expand...

Read up my post above where I talked about the types of statistical analysis. This can't be done with rankings.

Think about it.. You don't have a ranking for a website. You have a ranking for 1 keyword, on 1 page. That's not useful.

This is also why typical correlation studies are not useful. Looking at how content length and h2 headers correlate with rankings for 1 keyword is just so incredibly limited. You rank for a keyword based on your overall page, your overall site, your overall links, your overall internal links. Not just a single factor on that page. Try to rank a site with 1 page, for 1 keyword. That's just not how seo works, so these correlation studies are just for show.

They were more useful in the past when seo was simpler, but it's much more "big picture" now with higher level concepts being more important like "having a good writer", and "matching user intent" rather than "number of h2 tags" or "content length".

So when analysing these types of things single rankings aren't useful. What we want to be able to compare are traffic levels. However. You can't normally do an analysis with traffic levels for a normal analysis because they can't be compared. The traffic for 1 site can't be compared to the traffic for another site. But with an update, what you can do is compare increases and decreases in traffic to see if there's a difference in various groups.

ahrefs traffic values are so so inaccurate it would skew the data in a study that's already very difficult and sensitive to outliers and just a mass of data making it hard to pin-point things.

Following.

Hope you can find some strong correlation.

I am a big fan of your posts. Looking forward to see the results from your research/analysis.

K114a said:

Following.

Hope you can find some strong correlation.

Click to expand...

Roger Marquez said:

I am a big fan of your posts. Looking forward to see the results from your research/analysis.

Click to expand...

Cheers, guys.

Update(important):

I think it's better if I expand the analysis to not just include extreme cases. Since I've been stuyding more types of statistical analysis I've realized I don't need just the extreme cases, because I'm not doing just a correlation analysis. I'm doing a "compare groups" analysis. This is the difference

Correlation example :-

Is there a correlation between maximum aerobic capacity and age. Ie, as age increases does aerobic capacity decrease.

This is what most of the seo analysis do. "As moz DA increases, does average rank increase" - And as already discussed, these are just awful for seo as you cannot pin point 1 single factor and get any sort of reasonable information. This type of analysis is completely wrong for SEO.

I'm using this here https://statistics.laerd.com/ which is an amazing bit of kit that lets you put in all the details about your data and it tells you the type of test that's appropriate and it has guides for those tests. Very powerful stuff. I'm good at math and I've studied it at a fairly high level, including lots of stats, but with this tool it lets you operate like a higher level mathematician without the experience.

Group Differences example :-

Is there a difference in salary between male and female doctors.

Is there a difference in the average page views per visitor based on the three types of site design?

Is there a difference in productivity amongst packers at a factory based on the use of background music?


So this is more appropriate for seo.

Especially for an update. We can look at

Is there a difference in the percent change of traffic between sites with native content vs non-native.

Is there a difference in percent change of traffic between sites based on 3 types of domains, fresh, aged and expired.

Is there a difference in percent change of traffic between sites based on 3 ranks of ads/affiliate links usage, light, medium, high

Is there a difference in percent change of traffic between sites based on high commercial intent targeting, low (< 30% of pages for example, medium 30-70%, high 70%+)

So with that in mind anyone who's had a site with at least a 5% increase or decrease in traffic from may 5 is still going to be useful.

I can also swap the IV and DV's around, so we can ask "Is there a difference in X between groups who lost traffic and those who gained", or "is there a difference in X between groups who lost 5-10% vs those who lost 10-25% vs those who lost 25%+"


I've got 5 people so far who are in. I'll probably be able to get about 20-25 from my clients, but I need MORE. 100+ would be the best.

There needs to be enough in each group, ie, if we want to test aged vs fresh vs expired domains I need at least 15 in each group for a valid study.

I have pinged you on Skype and sent my website details

Power Booster said:

I have pinged you on Skype and sent my website details

Click to expand...

I'll reply to people who've pm'd soon. Just catching up with some wolf work then I'm going to create a form for people to submit the details I need.