On average, if you cross two late-flowering daylilies and grow a number of their seedlings the average flowering time of the group will be earlier than their parents. If you cross two early-flowering daylilies and grow a number of their seedlings the average flowering time of the group will be later than their parents. Early for this is defined as flowering before the average date and late means flowering after the average date of the population of all daylilies.
Geneticists call it "regression toward the mean".
The figure above represents an imaginary example of the flowering time (day of the year, January 1 = 1 to December 31 = 365) of the population of all daylilies. The population has an average. We select two cultivars with late flowering (their first day of flowering is greater than the average).
In the figure above we have chosen two parents which both first bloom on day 270 of the year. We cross them with each other and grow a number of their seedlings. In a world where the day of first flowering was 100% genetic we would find that the seedlings would first flower on day 270. But nearly all characteristics in the real world are determined by both genes and the environment and interactions between them. So the parents do not have perfect genes for flowering on day 270 and so their seedlings will not all flower on day 270. In fact some will flower before day 270 and some will flower after day 270.
In the last figure we can see that although we chose as parents daylilies that first flowered on day 270 their offspring on average flower before day 270. However the average day of first flowering of those seedlings is greater (later) than the average day of first flowering of the original population. And there will be a greater proportion or percentage of the seedlings that flower after day 270 than there was in the original population of daylilies.
By selecting the parents for later flowering we have changed the average date of flowering in the seedling population (the offspring) and made it later. We can continue selecting (by choosing later flower seedlings to use as parents) and we will continue to change the average date of first flowering. If we simply choose as parents seedlings that flower on day 270 then with each passing generation of selection our seedlings will come closer and closer to having an average first day of flowering of 270. But if with each passing generation we choose as parents seedlings with even later first days of flowering (first generation 270, next 275 then 280, etc) we will move the average date later and later.
If we continue selecting for enough generations we can produce daylilies that all flower later than those in the original population and if we graphed the first flowering dates of the original and selected populations there would be little or no overlap in first flower dates.