289-28 Beyond ANOVA... Exploratory Analysis for a Germplasm Evaluation Under Shade.

Poster Number 711

See more from this Division: C05 Turfgrass Science
See more from this Session: Poster Session: Breeding, Genetics, Selection, and Weed Control
Tuesday, November 4, 2014
Long Beach Convention Center, Exhibit Hall ABC
Share |

Jeffrey C Dunne, 101 Derieux Place, North Carolina State University, Raleigh, NC, Susana R. Milla-Lewis, Crop Science, North Carolina State University, Raleigh, NC and Consuelo Arellano, Statistics, North Carolina State University, Raleigh, NC
Analysis of variance (ANOVA) has been the traditional and often only approach to modeling complex, multi-factor experimental designs in applied field research. In becoming more acquainted with the data through a variety of exploratory analysis methods, the researcher can gain more insight into individual factors or treatments while still answering the intended research objectives. A data set collected from a multi-year bermudagrass shade study was used and analyzed with exploratory modeling techniques. Data was collected for two years (2011-2012) on a split-split plot design nested within three shade treatments (0%, 63%, and 80%). Twelve bermudagrass entries under each level of shade were split for two levels of fertility. The plots were evaluated for turfgrass quality, turfgrass cover, and normalized difference vegetation index (NDIV) at four rating dates across the two years of the study. Three analyses were conducted, along with the ANOVA for the appropriate experimental design, including a cluster analysis, logistic regression, and survival analysis. Cluster analysis provided a blind evaluation of the treatment relationships to the response variables collected. For instance, the CLUSTER procedure in SAS was used to reveal percentages of entries performing at the upper or lower boundaries of turfgrass cover, turfgrass quality and NDVI.  Logistic regression, which is best applied to scaled, discrete choice data like turfgrass quality (1-9; according to NTEP standards), was used to calculate probabilities of treatments ascending to a higher order or rating in comparison to others. Similarly, data manipulation of turfgrass cover and NDVI to create binary dummy variables (1 and 0) allowed for treatment comparisons to estimate the likelihood of an entry being more or less acceptable than the other.  In addition to these modeling techniques, survival analysis using the LIFETEST, LIFEREG and/or PHREG procedures was used to estimate the time to reach below acceptable turfgrass coverage, turfgrass quality and NDVI, due to the addition of a particular treatment or factor. These analyses can be conducted prior to or in verification of the traditional ANOVA results, providing a more in depth analysis and further discussion of treatment responses for applied field research.
See more from this Division: C05 Turfgrass Science
See more from this Session: Poster Session: Breeding, Genetics, Selection, and Weed Control