A-B advert testing - my results

Discussion in 'Reporting & Stats' started by cornetto, Aug 25, 2006.

  1. #1
    Ok, so I implemented A/B testing (see the Adsense Blog or look here http://forums.digitalpoint.com/showthread.php?t=121080 for info). The general idea is that if someone visits a page of yours, they get to see one of two different page layouts, and you monitor their clicks using a channel. From this you can tell which ad placement / colour scheme gets the best CTR.

    I set it up so that white background pages with black text and blue links were tested against black background pages with white text and very light blue links. After a few days the results were heavily in favour of white background, so I stopped the test.

    Then I tired same colours (white background, black writing, blue links) but tried either a central leaderboard, or two central squares. After a few days, the results show that there isn't much in it, with the leaderboard being slightly ahead. I investigated this, and it seems that my second square doesn't always fill with ads, but the leaderboard always does, so maybe that is why.

    Next test is to put either one or two leaderboards on the page. I'll tell you how I get on.

    Anyone else care to post their test results please?

    I should say that this is for a video game website, and I let each type of ad have 500 page impressions, to make sure that I am getting realistic results.
     
    cornetto, Aug 25, 2006 IP
  2. Submerged

    Submerged Active Member

    Messages:
    132
    Likes Received:
    0
    Best Answers:
    0
    Trophy Points:
    51
    #2
    Just a note - I've heard that a lot of the experts that use A-B tests wait for 30 results (in this case, actual clicks) before judging the results, as opposed to a certain number of impressions. Depending on your CTR, this may or may not matter, but just a note.

    This is a great resource though - I'll definetly be using this on my site :)
    -Alex
     
    Submerged, Aug 25, 2006 IP
  3. stlgatekeeper

    stlgatekeeper Peon

    Messages:
    131
    Likes Received:
    1
    Best Answers:
    0
    Trophy Points:
    0
    #3
    Testing is not statistically reliable without at least 100 results.
    1000 is better but 100 is acceptable.
     
    stlgatekeeper, Aug 25, 2006 IP
  4. Submerged

    Submerged Active Member

    Messages:
    132
    Likes Received:
    0
    Best Answers:
    0
    Trophy Points:
    51
    #4
    I was citing Joel Comm's "Google Adsense Secrets" with the 30 - so 100 might be better. Using statistics, "reliable" usually refers to 95% probability, which using the easy equation (chance = 1/sr(x), sr=square root) means 400 results would equal a 95% chance of the test being accurate.

    At least, I think those equations would be apt in this situation. Either way, unless one is drastically higher than the other, more is always better :)
     
    Submerged, Aug 25, 2006 IP
  5. stlgatekeeper

    stlgatekeeper Peon

    Messages:
    131
    Likes Received:
    1
    Best Answers:
    0
    Trophy Points:
    0
    #5
    Joel already made his money by selling out...
    Now he's simply trying to get others to buy templates for MFA sites.

    He's no fool.... but alot of others are. :)
     
    stlgatekeeper, Aug 25, 2006 IP
  6. Submerged

    Submerged Active Member

    Messages:
    132
    Likes Received:
    0
    Best Answers:
    0
    Trophy Points:
    51
    #6
    Haha... well, I use many different sources for my Adsense info, Joel's book being one of them. Also Peachput Press, and O'Reily. Of course, this forum helps fill in the gaps :)
     
    Submerged, Aug 25, 2006 IP
  7. JC007

    JC007 Active Member

    Messages:
    1,066
    Likes Received:
    11
    Best Answers:
    0
    Trophy Points:
    78
    #7
    Well, I will be waiting to see your results and please keep me posted.

    Thanks!
     
    JC007, Aug 25, 2006 IP
  8. cornetto

    cornetto Peon

    Messages:
    180
    Likes Received:
    4
    Best Answers:
    0
    Trophy Points:
    0
    #8
    Oh, yeah, I leave the page for 500 page impressions, which in my case is easily 30 clicks, before I judge results.

    I have now discovered that 2 square blocks of adverts in the middle of the page does better than one leaderboard in the middle, by a reasonable percentage. I think the problem with 2 sets of ads not filling up only happens on certain days, and other days it improves. This gives the user twice as many ads to look at, so more chance of a click.

    Next I'll try 2 squares versus 2 leaderboards.
     
    cornetto, Aug 29, 2006 IP