Phenology of Dent Corn and Popcorn. I. Analysis of Repeated Measurements from Phenology Experiments1

Abstract

AbstractRepeated observations are commonly made on individual or multiple plant samples in order to generate data suitable for developing linear regression models with growth stage as the dependent variable. Most studies using regression, including those conducted at the University of Nebraska, Lincoln (UNL), have implicitly assumed that the data adhere to standard regression assumptions even though these assumptions may be tenuous. Misleading conclusions can be made if basic regression assumptions are not reasonably satisfied. Pilot studies may be used as a strategy for completing preliminary analyses of representative data from a larger phenology study to develop appropriate regression models and substantiate basic regression assumptions. The objectives of this research were to (i) validate regression models used in an interactive agrometeorology (AGNET) offered by UNL (ii) correctly specify thermal‐photoperiod regression models of corn phenology using hourly data generated by an automated weather station, (iii) investigate the validity of underlying assumptions of models developed, and (iv) investigate the application of growth curve analysis as a statistical procedure for comparing response profiles of different treatment effects in corn phenology experiments. A 2‐yr phenology study incorporating three planting dates, one dent corn cultivar (B73 ✕ Mo17), and three differently maturing popcorn cultivars (P410, P609, and Iopop 12) was conducted in eastern Nebraska during 1982‐1983. Climate records from an automated weather data network were used to develop daily and hourly heat sums. Corn growth was described using both current and modified staging procedures. Serious violations of standard regression assumptions were identified in the UNL‐AGNET corn phenology model and were commonly associated with one or more of the following conditions: (i) using incorrectly specified models, (ii) heterogeneity of error variance, and (iii) nonindependence of errors. Violations were considered serious since they were suspected of producing biased mean squares and regression estimates, indicating a need for further refinement of model definitions and growth staging procedures. Alternative regression models were developed and tested using a modified staging procedure combined with partitioning and modeling phenological sequences as pre‐ and post‐anthesis development. When applied to these data, growth curve analysis demonstrated a curvilinear response of corn phenology to planting date (daylength) within the range of 14.3 to 15.2 h, and provided a useful alternative to standard regression techniques for statistically analyzing data from corn phenology experiments.