A statistical analysis of Zea mays grain yield response to nitrification-inhibitor amendments under field conditions utilizing parametric and non-parametric techniques
Nitrification is a matter of economic and ecological importance in nitrogen-tensive agricultural practices such as corn (Zea mays) grain production. When ammonium is bacterially oxidized into nitrate it becomes more susceptible to leaching and denitrification as nitrous oxide. The nitrification process not only adds to water and air pollution but also impacts on farm productivity by reducing grains yields and increasing fertilizer demands. Nitrification inhibiting fertilizer amendments can potentially improve grain yields by retarding the growth of nitrifying soil bacteria. Multiple field studies conducted across various locations and timespans have generated mixed results regarding the grain yield difference attributed to nitrification-inhibitor (NI) usage; likely due to diverse soil and climate factors and management practices. A quantitative meta-analysis can determine if grain yield differences caused by NIs (effect size) are statically significant across varying field conditions; and meta-regression can be used to investigate which variables have significant influence over effect size. Calculations of effect size homogeneity, between-study variance and between-group homogeneity all rely on inverse-variance weights derived from the standard deviations of the grain yield means. Many field studies fail to report the variances of recorded grain yield means and thus cannot be included in parametric analyses for NI meta-effects. A database comprised of 266 Logarithmic-response ratio (LRR) effect sizes was derived from 36 separate field studies, yet only 185 (from 26 studies) were utilized for SAS- and R-based meta-analysis and meta-regression. Both methods inferred a statistically-significant positive effect of NIs on grain yield at α=0.05. SAS- and R-based meta-regressions also infer a statistically-significant positive correlation between NI effectiveness and lower yearly minimum temperatures, finer-textured soils, and higher clay, SOM, and SOC percentage. This can be attributed to the effect these factors will have on soil microbe metabolism, rates of environmental degradation of the NI molecules, and the rate of N-cycling in the soil; all of which will entail back to NI performance regarding the degree of retention of N in its positively charged form. Repetition-based “bootstrapping” is a proposed non-parametric alternative to meta-analysis which utilizes repetition-based weights for effect sizes instead of variance-based, and thus can include a larger database for calculations of effect-size homogeneity. Both SAS- and R-based bootstrap analyses generated marginally (but still statistically-significant) positive confidence-intervals from the 266 LRR values. Although the reliability of these non-parametric techniques is questionable due to their inability to separate out within and between study variances, results were directionally similar to those from conventional meta-analyses. In order to improve meta-analyses for field data, such as evaluated here, the variances of any-and-all calculated effect sizes must be reported by authors, especially in field studies assessing agricultural products, as cumulative parametric analyses are expected to become increasingly more important in the field of agronomy and ecology.