Random number to divide visitors into two groups -- is my code wrong?

[Mike521]

we recently launched a new version of our internal search results page. to test the difference, I'm dividing visitors into two groups at the time they do a search. They get a cookie telling which group they're in, so they remain in it when they do searches later.

 

For some reason I'm getting about 70% in group A, and 30% in group B. I can't figure out what, if anything, is wrong with my code. Can someone review and tell me if I'm crazy?

if ( $_COOKIE["SearchTestGroup"] ) {
// they're already in a group
$group = $_COOKIE["SearchTestGroup"];
echo "you're already in group $group";
if ( $group == "B" ) {
	echo "<br>doing group B search";
	//doGroupBSearch( $theSearch );
} else {
	echo "<br>doing group A search";
	//doGroupASearch( $theSearch );
}	
} else {
// they're not in a group
$group = rand()&1;
// 0 = A, 1 = B
echo "<br><br>your random number is $group.";
echo " A value of 0 puts you in group A. 1 puts you in group B.";
if ( $group ) { // group B
	setcookie( "SearchTestGroup", "B", $cookieExpiration, "/", ".oursite.com" );
	// now perform the search
	echo "<br><br>cookie set as B";
	//doGroupBSearch( $theSearch );
} else {
	setcookie( "SearchTestGroup", "A", $cookieExpiration, "/", ".oursite.com" );
	//now perform the search
	echo "<br><br>cookie set as A";
	//doGroupASearch( $theSearch );
}
}

[soycharliente]

How many test cases have you done? Is your sample size large enough for you to even come to that conclusion? You have to also think that, just because you have a 50% chance, that's a long run 50% distribution.

[discomatt]

Try

 

$group = mt_rand(0,1);

 

Running that (100x10000 time), I never had anything beyond +/- 5% difference.

[soycharliente]

I was also going to suggest using the range parameters. mt_rand() is faster than rand() correct?

[Mike521]

I ran a modified version of the script that simply looped 1 million times and tallied up A vs B (using the method I posted earlier). it was almost exactly 50/50, so I'm confident that it's accurate.

 

Regardless, so far we have 2,296 searches. 1,539 were in group A (67%), 757 were in group B (33%).

 

everything seems correct to me but I can't figure out why there's such a big discrepancy

[Mike521]

I was also going to suggest using the range parameters. mt_rand() is faster than rand() correct?

 

hmm I'm not sure which is faster myself. I'll switch to it, but I don't think the method is the cause of the problem, does everyone agree? If not, try this yourself:

$groupA = 0;
$groupB = 0;
for ( $i = 0; $i < 1000000; $i++ ){
$random = rand()&1;
if ( $random == 1 ) {
	//echo "<br> it's a 1 (group B)";
	$groupB++;
} else {
	if ( $random == 0 ) {
		//echo "<br> it's a 0 (group A)";
		$groupA++;
	}
}
}
echo "<br>group A: $groupA<br>group B: $groupB";

[soycharliente]

What are your expected results?

 

My output:

A: 499939

B: 500061

 

Looks like it's working fine to me.

[Mike521]

yep that's the goal, I want it to be as close to 50/50 as possible

 

but for one reason or another, 67% of our visitors are doing group A's search, rather than 50%. I guess there's a problem outside this PHP code

[soycharliente]

I ran your function with rand/mt_rand and with/without ranges. About 50% every time. Also included how long each took in seconds.

 

mt_rand()&1

A: 499430

B: 500570

2

 

mt_rand(0,1)

A: 500009

B: 499991

1

 

rand()&1

A: 500060

B: 499940

2

 

rand(0,1)

A: 500779

B: 499221

2

[Mike521]

well if nothing else at least I can switch to mt_rand. I'll have to look at javascripts and such to see if there's something redirecting people from page B to page A..

 

thanks everyone for taking a look at the code for me

[discomatt]

From the manual

 

Many random number generators of older libcs have dubious or unknown characteristics and are slow. By default, PHP uses the libc random number generator with the rand()  function. The mt_rand() function is a drop-in replacement for this. It uses a random number generator with known characteristics using the » Mersenne Twister, which will produce random numbers four times faster than what the average libc rand() provides.