Location: Vegetable ResearchTitle: Using weighted trait indices to select the best performing broccoli hybrids in multi-site and multi-year trials Author
|Bjorkman, T - Cornell University - New York|
Submitted to: HortScience
Publication Type: Abstract Only
Publication Acceptance Date: 5/1/2015
Publication Date: 9/1/2015
Citation: Stansell, Z.J., Farnham, M.W., Pelc, S., Couillard, D.M., Bjorkman, T. 2015. Using weighted trait indices to select the best performing broccoli hybrids in multi-site and multi-year trials. HortScience. 50:S171.
Interpretive Summary: N/A
Technical Abstract: Understanding and implementing evaluation data from vegetable trials conducted across multiple years and environments by multiple raters presents numerous challenges. In order to select new broccoli hybrids suitable for eastern production, the SCRI East Coast Broccoli Project has conducted over 32 phase I and 80 phase II trials which have included over 120 unique broccoli hybrids grown at different locations along the eastern seaboard (SC, NC, NY, ME, and VA). These broccoli hybrids were evaluated for 10 traits including bead uniformity, head color, head firmness, head smoothness, bead size, color, and overall quality. Initially, hybrid selections and recommendations were based predominately on the “overall quality” trait. Theoretically, this trait rating takes into account many different attributes; however, it is also likely the most susceptible to human perceptual bias. In the spring of 2014, two East Coast Broccoli phase I trials conducted at the U.S. Vegetable Laboratory in Charleston, SC, were concomitantly evaluated by three different raters in an effort to elucidate and account for potential rater bias. Four evaluation instruments or indices (e.g., different weighted linear combinations of traits) were proposed including an instrument which accounted for variation in overall quality relative to the more specific traits measured. Intra-class correlation (ICC), a statistic used to quantify both rater agreement and rater consistency, was used to compare methods. The evaluation instrument which accounted for human perceptual bias was shown by these experiments to significantly increase both ICC agreement and consistency between raters (p<<0.001). This improved evaluation instrument allows for both greater selectivity and precision in the analysis of multi-site and multi-rater data sets, and also for use with comparative methods such as stability or principal component analysis. Moreover, this work should be readily applied to trials of other horticultural crops.