To reuse an old post:

[I]t has to do with relative growth/shrinkage and the base of the positional-numbering system you're using. If you have a random starting value (X) multiplied by a second random factor (Y), most of the time the result will start with a one.

You're basically throwing darts at logarithmic graph paper! The area covered by squares which "start with 1" is larger than the area covered by square which "start with 9".