A nice explanation at honestuniverse.com
One of the most important principles of the scientific method is reproducibility. A valid result should be able to be replicated independently, whereas an invalid result (originally achieved due to some error or perhaps just chance) will not be able to be consistently reproduced […] The reason why this is the case has to do with researcher degrees of freedom. In an original experiment, the experimenters have the freedom to make certain choices. Some choices may have been made beforehand, whereas others are made after the study has been started. These choices may not all be made consciously, or they may be made consciously with only unconscious bias, but the fact that the choices are made at all after the experiment has begun affects the reliability of the results.
which basically means that the kind of replication nature genetics demands for an association, is not a replication at all, but some extended or repeated study.