Chapter 4 surveyed several measures and methods of diversity. An experimental study demonstrated the behaviour of several of these measures. The search process in genetic programming must find suitable structures on which to represent content. Thus, measures of diversity which provide an accurate depiction of the structure and content in a population, such as edit distance diversity, are likely to capture important dynamics. However, a main driving force behind the algorithm is selection and recombination. Therefore, diversity measures which capture the property of fitness distributions will indicate the population's effect on selection pressure.
Many methods have been used to control diversity. Chapter 5 demonstrated the use of lineage selection and its good and bad effects on performance. The concept of genetic lineages can be an accurate indication of diversity loss in a canonical genetic programming system. Some methods encourage high genetic diversity or fitness-based diversity, while others are adaptive. However, as seen in Chapter 6, changing population diversity will also change other dynamics in the algorithm. The lack of a clear correlation between diversity and fitness in Chapter 4 emphasises the danger in assuming that high diversity (of a some type) leads to better fitness.