How could the change spread through the population?
Robertsonian translocation happens often enough that about one in a
thousand of the population owns a fused chromosome, giving their
owners 45 rather than 46 chromosomes in total
(Ref 1). A chromosome count of 45 is not stable for a population because they
still have both normal and fused chromosomes. This means only 50% of
their children would get the fused chromosome, in other words they
would not breed true. In addition they would have a higher risk of
abnormal gamete production as has been previously discussed.
Neither of these problems would be true for individuals possessing two fused chromosomes and thus a chromosome count of 44. That number would both breed true and would not have the same problems producing viable gametes as the 45 chromosome individuals. So how could a population of 44 chromosome individuals happen?
One possible scenario is where one group of related individuals, who are carrying a mutation, have a vastly disproportionate number of children. This is thought to explain the prevalence of unusual Y chromosomes in the Mongol population, they are probably descended from Genghis Khan (Ref 2).
Another scenario is a population bottleneck in which a small population become separated from the main population. In this scenario the genetic makeup of a small number of founders makes up the gene pool for the entire future population. Any rare mutation in the founders will be present in much higher proportions than in the outside populations, so it has a higher chance of becoming fixed in the population.
Imagine a scenario in which there are two large families with one parent in each family having a fused 14/21 chromosome (and 45 chromosomes). In each family 50% of the children would be carriers of the fused chromosomes because the fused chromosome would be passed on 50% of the time.
If these families became separated from the main population by, for example, being isolated on an island after a plane crash or as survivors of a nuclear war then they would become the founders of any future population. Instead of the fused chromosome being present in 1:1000 of the population it would be present in much higher proportions.
The graph below demonstrates how even a small advantage gained from the fused chromosome would spread populations over time. This graph is produced by a perl program (Ref 3) which simulates the spread of fused chromosomes through a population of 1000 individuals over time. To demonstrate how little reproductive advantage is needed in order for a mutation to dominate a population, the program assigned the fused chromosomes just a 1% chance of an additional offspring at each random mating.
This graph shows the cumulative shape of the population change in the 75% of 100 simulations where the fused chromosome individuals completely dominated the population (the data from 25% of runs where the unfused chromosomes won is not included). On average the fused chromosomes won by generation 2607. This demonstrates that even with only a very slight advantage it is possible for novel mutations to dominate a small population.
Note that there is also a more detailed explanation of how the graph was generated and what it shows.