Split Testing Question

[trigatch4]

I've got two contextual ads running on a page and I've got a ton of PPC traffic incoming. I want to use the PHP redirects I'm using now but send half to one offer and half to another to see what converts better.

Here is the code I'm using in the PHP file:

<?php
$offer = $_GET['offer'];
$type = $_GET['type'];

 if ($offer == "redirect-name1")
    {$jumplink= "http://offerurl.com";}

 if ($offer == "redirect-name2")
    {$jumplink= "http://offerurl.com";}
    

header("Location: ".$jumplink."&sub=".$type."");
?>

So each of these "redirect-name" should split their time equally amongst two different actual offers...

Any help? T.I.A.

 

[trigatch4]

not necessary, but it would also be nice to see the difference between converting if it opens in a NEW WINDOW versus the SAME WINDOW... may be a bit more tricky. I'll take what i can get

 

[mason]

If I were to split test I would do one page at a time. Like test 1 page for a while and then switch to the other. I think it's better b/c there is still a possibility of one page getting all the leads/sales simply b/c the "buying" users simply just got redirected to the same page.

but if you do it that way, you can simply generatea number 1 or 2. if 1, then goto this page. if 2 then goto other page.

that's not really equal disbursement though. to make it complicated, I would use a mysql database. have a column that is for numbers only. everytime a user redirects to a page, it will read this number in the DB, and send them to the correct page. right before you redirect them, all you have to do is make a line of code that changes that number in the database to like "2". The next time it reads the DB, it will send it to the other page and reset it back to "1".

 

[nis]

You can use the rand() function.

rand(0, 2) will give you a random number between 0 and 2.

As far as I can read the code, witch offer it should jump to is already decided when the user hits the php-script.

 

[Someone]

Try this:

<?php
$type = $_GET['type'];

if(rand(0,1)) {
  $jumplink= [URL]http://offerurl1.com[/URL];
} else {
  $jumplink= [URL]http://offerurl2.com[/URL];
}
    

header("Location: ".$jumplink."&sub=".$type."");
?>

This will have a 50/50 chance of going to each page.

 

[mhmhmh]

If I were to split test I would do one page at a time. Like test 1 page for a while and then switch to the other. I think it's better b/c there is still a possibility of one page getting all the leads/sales simply b/c the "buying" users simply just got redirected to the same page.

No disrespect intended, but that is absolutely the wrong way to do any sort of split test if you intend to end up with useful data.

It has been documented and proven time and time again that the aforementioned method is prone to calendar issues, purchase latency, and several other conversion factors.

The most proper way to do a split test, in its' simplest form:

- Establish a baseline page that you know the average conversion rate of
- Come up with 2 new variations to test.
- If doing PHP, and you don't want to involve a database, etc. Use code like the following:

<?php
$type = $_GET['type'];

$a = rand(1,3);

switch ($a) {
  case 1:
      $dest="http://offerurl1.com";
  esac;
  case 2:
      $dest="http://offerurl1.com";
  esac;
  case 3:
      $dest="http://offerurl1.com";
  esac;
}

header("Location: ". $dest);
?>

Please note that I did not check this for syntax, I just typed it here in the forum, and it likely contains errors. This is just to explain the logic.

-mh

 

[mhmhmh]

Also, if you're not PHP savvy, and you want to do something a little more complex, you might want to check out Google's Site Optimizer.

It uses multivariate methods.

Also, for more explanations of how to use these methods and why, I am in love with MarketingExperiments.com.

I believe it's still 100% free.

 

[Propellorhead]

Marketing experts is a great resource tool. I can tell you from experience that multi variant testing is the only true way to test. Split run testing is old skewl bros. I suggest coming up with a few elements that you want to test and then work outwards from here. Everything ought to be tested, ( one at a time wiht everything else remaining constant ) from headlines, to hero shots, to call to actions, to buttons; you name it you can test it. We are working on a multi variant system and when we go into alpha I will let you all have ride on it.

 

[mason]

I only mentioned the DB option because he wanted it to "split their time equally". last time I checked there's a possibility that one page will get more attention than another if using rand. BUT WHAT DO I KNOW.. ?!?