I’d like to add one more piece to the good explanations here, by explaining genetic drift.
Suppose you have a population of six individuals, three of which have gene A, and three of which have gene B. There isn’t any advantage associated with either gene- the difference between the two genes is purely neutral as far as selection goes.
Now, after those six individuals mate to have offspring, it might just be the case, purely by chance, that you end up with four A’s and two B’s in the next generation. (You could just as well have two A’s and four B’s, depending on the roll of the dice.)
When that generation reproduces, there are more A’s having kids, so there’s more of a chance that the A’s will have a few extra children than there were in the last generation. I realize that this is a little oversimplified (in reality the genes A and B would be distributed throughout the population, and some individuals would have both genes, etc.,) but if you use a very famous equation from evolutionary biology called the Hardy-Weinberg equilibrium, you find that the chances of going from 4A, 2B to 5A, 1B are a little higher than the chances of going back to 3A, 3B because there are more A parents, so any statistical fluctuations in the number of offspring they have will tend to be amplified. Think of it this way: if there’s only room for 1 million people in america, and 90% of them are blonde and 10% are redheads, then it’s possible for the blondes to squeeze out the redheads just by having an average of 2.1 children per family rather than just 2. If the redheads have 2.1 children per family, they will only increase their percentage in the population ever so slightly.
In the next generation, there’s an even better chance that you will go from 5A 1B to 6A 0B. That means that A has become “fixed”; everyone has it now, so all the offspring will have A for all generations to come, until a new mutations appears. Meanwhile, B is lost.
Now, bear in mind that this would appear to make it even more likely that a beneficial gene could be lost, but that’s not quite so. The equations take into account the amount of selective advantage that a gene confers, so if a gene is beneficial, it has a better chance of snowballing from a single mutation to being fixed in the population. Also, as has been mentioned before, small populations tend to evolve faster. If you have a small population of organisms that are cut off from the rest, then that single B is a larger percentage of the total population, because there’s less A to compete with. That, too, makes it more likely to become fixed. This is presumably the reason behind punctuated equilibrium, and is generally referred to as allopatric speciation: a large population has a lot of genetic “inertia” that soaks up new mutations, so it remains unchanges for a long time. If a small population is isolated for some reason (for example, by the severing of a land bridge, or the introduction of transposons (genetic parasites, essentially) into the population which prevent some animals from interbreeding with others) then it will evolve very fast, as good mutations can be better fixed.
-Ben