#1
  1. No Profile Picture
    Utility Bot
    SEO Chat Good Citizen (1000 - 1499 posts)

    Join Date
    Oct 2003
    Posts
    1,108
    Rep Power
    17

    Google Ranking Positions and %CTR: Estimating Search Engine Visitors


    Have you ever wondered how many visitors you'll attract if you manage to bump your site up a few ranks on Google for particular keywords -- and how that will affect your click-through rate? Now that Google Webmaster Tools offers more accurate data, you can get that answer, and see if the increase will be worth the work you'd need to put in.


    Read the full article here: Google Ranking Positions and %CTR: Estimating Search Engine Visitors

    For more discussion go here: Blog Article Discussion

    Comments on this post

    • pandit je agrees
  2. #2
  3. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    I like what you have done here and I like the logic you have followed but unfortunately I believe it is flawed. When plotting a curve to points of data, for each point on the X axis you need to consider spread.

    For position 2 on Google your CTR ranges from c. 4% to c. 78%. with no particular clustering. So it is just as likely if I achieve positon 2 that I will get a 4% CTR as it is that I will get 78%. In certain instances the data suggests lower positions are better.

    That's not to say that your theory is not accurate and it could be that there is just not enough data to draw anything conclusive.

    Finally I would be interested, and might even try, to see the difference between industries - is there a similar pattern in health related searches compared to IT related searches - otherwise you might be mislead using the behaviour of different users in incorrect places.

    But thanks for a very interesting idea!
  4. #3
  5. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    This actually inspired me to dig deeper - there are a few things here which mean I don't see this model stacking up. I tried it out on 80 pieces of data and here is my graph:

    graph-google-pos-ctr.jpg

    Quick explanation for the following if you don't do stats: R-square is a statistical output that says "according to the line I made fit this graph, what percentage of times will using the line as a predictor be successful". If R-square is .5 that means there is a 50/50 chance that the line that the data makes will make an accurate forecast.

    Now when I used a Power Curve I got an R-square of .54. When I used a regression line R-square was .52. That means my data could be a Power Curve or a straightline regression. There is so much noise that neither is a good indicator or predictor of the data.

    I also saw your R-square is .52 when fitting a Power Curve. That means as a predictor it is only true half of the time - same as my data.

    I would not want to advise clients to invest based on a model that has a 50% chance of being right.

    I actually think what both your data and mine show is the opposite of what you want to prove. I believe the data say:
    - there is little difference between being Google #1 and Google #5
    - there is little difference between being Google #5 and Google #8
    - Google #11 and #12 may be more valuable than #9 and #10 - I've read alot of speculation about this but it needs more data
  6. #4
  7. SEO Consultant
    SEO Chat Genius (4000 - 4499 posts)

    Join Date
    Jul 2004
    Location
    Minneapolis, MN, USA
    Posts
    4,244
    Rep Power
    1105
    I would say that the graph shown shows a significant trend. Sure there is variation, which there is in any statistical set of data.
    If you looked at the mean CTR value for any position with a decent sized data set you would see that the R2 parameter would be a lot closer.

    There are many situations when looking at regression analysis you would want the R2 parameter to be 0.98 or better, otherwise it is not a good fit, but in this instance I think that the mean CTR for each position would show a strong trend that looks very similar to the regression curve. I agree that there is a good deal of noise, but if you did straight linear regression on that data set it would be nowhere near as good a fit.
  8. #5
  9. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    Originally Posted by tstolber
    If you looked at the mean CTR value for any position with a decent sized data set you would see that the R2 parameter would be a lot closer.
    I would 'assume' the same but I have this nagging feeling it might not be true.

    Originally Posted by tstolber
    but if you did straight linear regression on that data set it would be nowhere near as good a fit.
    I did (see earlier post) and the linear regression showed no better fit than the Power Curve.

    All I'm confident about at the moment is that if you drew a line between the top left and the bottom right of the chart all data points would lie to the left of that line within that triangle. But that doesn't mean that a power curve or a regression line are the correct way to create a model for predicting what CTR you would get at any position in the SERPs.

    BTW - I'm rambling about this more here:
    Google SERP positions and CTR

    Mainly about the probability of CTR at any one point rather than just using an average.
  10. #6
  11. No Profile Picture
    Contributing User
    SEO Chat Good Citizen (1000 - 1499 posts)

    Join Date
    Apr 2009
    Posts
    1,317
    Rep Power
    581
    Just found this that breaks things out to types of quiries (NOT result page types). Looks like our suspicion that some queries behave differently than others might be correct (little out of date but not too far)

    Click Through Rates in Google SERPs for Different Types of Queries - YouMoz | SEOmoz
  12. #7
  13. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    Great piece thanks but still not answering one piece.

    "On average" position 1 is better but how confident can a person be that this will be true for their keyword. For this you need to create something that looks like a bar graph but at each position there needs to be a binomial distribution graph.

    So even when we put brand aside you can make a statement, "If you are in #1 the average CTR is 50% but the probability of this being true (or within 5% either direction) for any keyword in particular is 40%"

    It's that final piece of the puzzle I was thinking about but as CTR for #1 differs by industry, by the number of keywords in the phrase, and probably by other factors I'm ready to conclude that you just cannot make the statement "#1 will give you a CTR of x%"
  14. #8
  15. No Profile Picture
    Contributing User
    SEO Chat Good Citizen (1000 - 1499 posts)

    Join Date
    Apr 2009
    Posts
    1,317
    Rep Power
    581
    Yeah. Nobody seems to have any really serious stats. But then I think only Google could really even start to get them.

    CTR varies by a lot more than just position so I think average CTR for type of term or type of results page is the best you are likely to see.

    Makes me sad because I love playing with statistical analysis.
  16. #9
  17. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    Reminds me of "Airplane" the movie:

    "The doctors say she's only got a 50/50 chance. but there's only a 10% chance of that!"
  18. #10
  19. Contributing User
    SEO Chat Adventurer (500 - 999 posts)

    Join Date
    Apr 2011
    Posts
    931
    Rep Power
    1278
    Reminds me of "Airplane" the movie:

    "The doctors say she's only got a 50/50 chance. but there's only a 10% chance of that!"
  20. #11
  21. SEO Consultant
    SEO Chat Genius (4000 - 4499 posts)

    Join Date
    Jul 2004
    Location
    Minneapolis, MN, USA
    Posts
    4,244
    Rep Power
    1105
    This intrigues me a lot, and with lack of available data - lets pull together as the SEO community and get the data.

    Here is a CTR v SERP Position data table

    Here is the link - https://docs.google.com/spreadsheet/...FE&usp=sharing

    Anyone with the link can edit.

    I have populated it with results from one site.

    Go into Google Webmaster Tools and find your CTR and position.

    For example - average position 5.6 with a CTR of 11 would go in position 6 - I am rounding up the average position. Position 5.4 with a CTR of 11 would go in position 6.

    Just add the CTR for your position on a new line Keyword 50 for example.

    As we populate this table would should get sufficient sampling to determine what the average position is and also the range. We can then do a statistical analysis on the data and get some good information out of it.

    I have commented in cells on relevant data - such as branded keywords etc, so feel free to add a comment in the box.

    Thanks in advance to all those who participate, and any thoughts or questions?
  22. #12
  23. SEO Consultant
    SEO Chat Genius (4000 - 4499 posts)

    Join Date
    Jul 2004
    Location
    Minneapolis, MN, USA
    Posts
    4,244
    Rep Power
    1105
    I started a fresh thread for this in test and experiments - CTR v Keyword Position Test

IMN logo majestic logo threadwatch logo seochat tools logo