Applying quantile regression to analysis of AFIS cotton fiber distribution

Varying fiber length distributions of cotton, Gossypium hirsutum L., impacts its spinning performance. Advanced Fiber Information System (AFIS) facilitates the analysis of the length distribution of individual fibers in cotton. Quantile regression is a variant of standard regression with which conditional quantile values can be calculated by minimizing weighted sums of absolute deviations across the entire distribution. Quantile regression was used to analyze AFIS fiber length distribution among five genotypes of upland cotton grown at the Texas Agricultural Experiment Station Research Farm near College Station, TX during 2001 and 2002. The shape of the distribution of 'CAMD-E', a short-staple variety, was actually similar to 'Acala 1517-99', a long staple variety with good spinning quality, even though CAMD-E had consistently lower fiber lengths. 'FM 832', and `TAM 94L-25' had similar mean fiber lengths to Acala 1517-99, but their distribution shape was less skewed. 'TTU 202' had high cross entropy values, but little difference was detected in distribution shape by quantile regression. Year had a significant impact on distribution of fiber lengths, affecting distribution scale and location, which may be due to lower fiber fineness and maturity in 2001. Quantile regression was found to be an effective method for analyzing AFIS fiber length distributions, although further testing with a larger set of genotypes and environments with spinning data is needed.