Mixing of heritable characteristics in humans over multiple generations

One question that sometimes nibbles on the minds of those of us who study human forms and behaviors is "How durable, from a generational standpoint, are innate abilities and behaviors?"

The research on the topic has been intriguing although statistically mixed.

For example:

A researcher named Bäumler proposed that a person's surname (last name) might be linked to their physical strength even though surnames solidified between 6 and 26 generations ago. That is, surname usage became common and were handed down generation-to-generation between 1450 and 1850.

Bäumler focused on surnames that were descriptors of professions and self-professed aptitude for sports and later actual track-and-field performance metrics. He divided various sports into "heavy" sports like Rugby and Weight lifting and "light" sports like "Distance Running" and "Darts".

Regression-to-the-mean effects SHOULD have totally obliterated the differences between "Smith", "Wright", "Miller", "Brewer" (professions that selected for physical strength) and "Tailor", "Clark/Clerk" and "Scribner" (sedentary professions) over the +6 generations.

Results 

The last column is the difference between the percentage of participation-rate in "Heavy" sports and "Light" sports. 20532 data points.

The results are not a total slam-dunk but it strongly suggests that there is some factor in play that is resisting the regression-to-the-mean effect.

It could be that large, muscular men don't pick tiny, ectomorphic women and that ectomorphic men are less inclined to select endomorphic women.

It could be that men are more likely to select mates who are daughters and sisters of fellow professionals. 

Larger implications

It suggests that the great melting-pot model of assimilation will not be as robust as commonly assumed. Stupid people will beget and raise stupid children. Violent, low-impulse-control parents will raise children who are inclined to violence.