Flip Flop the displayed page

Jon12345

I want to test different content on a landing page to see which my visitors like the most. Version A is to be shown 50% of the time, Version B to be shown 50% of the time.

So, visitor #1 gets shown Version A, visitor #2 is shown Version B, visitor #3 shown Version A and so on.

My idea is that I would have some code that would say include(content-a.php) or include(content-b.php) depending on which is due to be shown next.

Any suggestions on how I might achieve this?

Thanks,

Jon

brianadkins

test to see if the last digit of the current time is even or odd and based your decision on that.

Jon12345

I would prefer a system where it does things sequentially. i.e. it looks at a status flag and changes according to that. This way, if I don't have too many visitors, I can be sure that each page would be shown 50% of the time. This would not be the case with small numbers of visitors using a current time.

Weedpacket

Or just use the value returned by rand(0,1). I don't expect visitor #3 is going to know about visitor #2, so wouldn't have any reason to expect Version A instead of Version B.

Jon12345

If I used rand(0,1) and I had 100 visitors, it is possible that Version A is shown 70 times and Version B only 30. This would make it innaccurate for my testing.

Bobulous

Funny enough I've just written a small test harness for mt_rand.

On an example run of 100 random numbers of either 0 or 1, zero came up 57 times, and one came up 43 times. There seems to be a difference of fifteen or less between the two numbers almost every time.

However, being random it is possible that you will end up with zero showing up once and the number one showing up 99 times. But that's very unlikely to happen often.

Weedpacket

Jon12345 wrote:
it is possible that Version A is shown 70 times and Version B only 30. This would make it innaccurate for my testing.

Then you should yell at whoever made the compiler, since with any decent random number generator (and even many poorly-written ones) what you describe should only happen 0.0023% of the time (i.e., once out of every four million page views or so).

Bobulous

A probability of 0.00002317 would mean that the situation was expected to occur once in every 43,158 trials rather than one in four million. Or have I messed up somewhere?

madwormer2

Sheesh, some people, he plucks two numbers out of the air, and you correct him because his second number didn't match the first.

He was speaking in "general terms", just so you get an idea of how stuff would happen.

And in solution to you're question, easy, on the page that displays it, in order to get a number 1 or 0, just write the no. to a file. If it reads 1, display one, and change it to a 0. If it reads 0, change it to a 1. That way it's guarunteed 50-50.

Don't like files? Database. Don't like database? Use little stones.

Bobulous

What can I say, madwormer2, I'm a fussy b******. And I'm not sure I'm right about the numbers anyway.

As for the question at hand, the database option would work, because MySQL will deal with the file-locking. But if you use a file, I reckon you'd have to use flock() to avoid multiple writes to the file ruining the sequence.

. . . actually, flock() doesn't sound too reliable. Maybe stick with the database option if you really want a totally reliable zero-one-zero-one-zero-one sequence.

Weedpacket

Bobulous wrote:
Or have I messed up somewhere?

Yeah, I said "page views" - counting one hundred page views per trial (otherwise you wouldn't get 70 of one and 30 of the other) 🙂

madwormer2 wrote:
Use little stones.

The secret's to bang the rocks together, guys!

But hey, the results would be worthless anyway if there aren't enough hits in the first place to get a decent sample....

Bobulous

Ah, see, that's why I was never any good at maths.

Weedpacket

Bobulous wrote:
Maybe stick with the database option if you really want a totally reliable zero-one-zero-one-zero-one sequence.

I suspect that artificially imposing a strict pattern may bias the results (with some viewers getting disproportionately more of one than the other).

If the poster really want to ensure that there is an exact 50/50 split (as if a couple one way or the other is going to have a bigger effect on the results than variations in the sample), then you should pregenerate a random sequence of bits that has exactly the right numbers of 0s and 1s that you're after (put them all in an array, and then shuffle, say), save it to a file, and then read from that. If you save them as '0' and '1' characters you can use fseek() and filesize() to find the last character in the file, fread() it, and then ftruncate().

You still have all the hassles of file-locking to contend with, and the same approach could be done with a database, but it would be less likely to contaminate your results with unwanted patterns that a zero-one-zero-one pattern might produce.

More hassle than it seems worth bothering about, in my opinion, though. If you're going to do anything statistical, it helps to know a bit about the odds. A sense of proportion helps 🙂

Pregenerated tables of random numbers ... that takes me back. In my day we didn't have random number generators we could just use. Even effective pseudorandom number generators were beyond our means. We had to get our random numbers generated in factories, where they'd be printed into books and published, and we'd read our numbers from out of those books (I think I might still have one around here somewhere).

Kids these days don't know how easy they have it....

Bobulous

They used to print books of random numbers? I had no idea. I suddenly have a new gratitude for mt_rand and such similar functions. (I did test its conformance to the uniform distribution, and the mean was bang on, but the variance was higher than it should have been. I'm not sure what the consequence of that is.)

As for a zero-one-zero-one pattern producing disproportionate outcomes, would that be because of file-locking issues, or something I haven't thought of?

phpshift

Create a file called "visitnum.dat" that just has the number 0 in it, then put this code in "index.php":

<?
$vn=fopen("visitnum.dat","r");
$num=fread($vn,8);
fclose($vn);
$num+=1;
if($num==2)
$num=0;
$vn=fopen("visitnum.dat","w");
fwrite($vn,$num);
fclose($vn);

if($num==0)
header("Location: home1.html");
else
header("Location: home2.html");
?>

Now, put one version of teh page in "file1.html" and the other version in "file2.html" or just change those values in the code to the URLs you are already using.