Progress 10/01/04 to 09/30/05
Outputs
Randomization of treatments to experimental units is often infeasible in on-farm research. Instead, treatments may be assigned in sequence to rows of plants, with one treatment applied to the first set or rows, the second applied to the second set, etc. This is known as "pseudo-replication." The same phenomenon occurs in demonstration project. In observational studies, the objective may be to identify the effect of an intervention in time or space. This also results in pseudo-replication. Whereas there is no fool-proof way to judge statistical significance of studies with pseudo-replication, it is sometimes evident that a change in the sequence could not have occurred by chance alone. Techniques based on spatial distritutions and subjective assessment of variation are investigated to obtain approximations and bounds to significance probabilities and standard errors.
Impacts
Data from non-randomized trials can provide information about treatment effects or the impact of intervention, but without the level of objectivity expected in randomized trials. Such information can be useful for planning future randomized studies.
Publications
- No publications reported this period
|
Progress 09/15/01 to 09/30/04
Outputs
The project began with an internet survey of on-farm research in the Caribbean Basin, which produced very few results. It was then deemed impractical to conduct the research as originally planned. Focus then turned to demonstration and unreplicated research conducted in Florida. The principal study concerned effects of extended day length on grass varieties in north florida. This project had two large plots. One plor had artificial light installed to simulate extended day lenth. The other plot was nearly adjacent and had no artificial light. Within each plot, numerous entries of pasture grass planted in replicated within the large plots. The design provides no conventional method for assessing the effect of extended light due to non-replication of the treatment. Spatial variation models were used to assess trends in variation, and effects of extended light were assessed relative to spatial variation.
Impacts
Data from non-randomized trials can provide information about treatment effects or the impact of intervention, but without the level of objectivity expected in randomized trials. Such information can be useful for planning future randomized studies.
Publications
- No publications reported this period
|
Progress 10/01/02 to 10/01/03
Outputs
Due to limitations in conducting on-farm research, randomization of treatments to heterogeneous experimental units often is not possible. Sometimes only one treatment (or farming system) can be employed in a given farm, or pehaps two or more can be employed but not replicated within a farm. Replication is achieved by repeating the experiment in several farms, and combining results over farms. Each farm may have its own set of covariates, such as soil type or slope of land. A central problem is finding an appropriate error term for tests of hypothesis and confidence intervals on treatment differences. Mixed model methods can be used to assess error, at least approximately. Similar methods are used to analyze multicenter clinical trials. In some cases, only one farm is available, and treatments cannot be replicated in the farm. This case is especially troublesome. Methods of spatial data or time series analysis can be used to model variation, with the objective of
assessing the statistical signifiance of an abrupt change at the point of treatment change. This technique is recommended only in the most difficult cases.
Impacts
Using methods of multi-center clinical trials, spatial data and time series analysis, data from on-farm research may be statistically analyzed. These methods must be used only by competent statististicians.
Publications
- Eilitta, M., L.E. Sollenberger, R.C. Littell, and L.W. Harrington. "On-farm experimentation in teh Los Tuxtlas region of Veracruz, Mexicl. I. Mucuna biomass and maize grain yield." Expl. Agr. (2003), vol.39, 5-17.
- Eilitta, M., L.E. Sollenberger, R.C. Littell, and L.W. Harrington. "On-farm experimentation in teh Los Tuxtlas region of Veracruz, Mexicl. II. Mucuna variety evaluation and subsequent maize grain yield." Expl. Agr. (2003), vol.39, 19-27.
|
Progress 10/01/01 to 10/01/02
Outputs
Internet-based resources were used to identify on-farm researchers. Some agricultural colleges have web-pages that show previous and current on-farm research projects, both foreign and domestic. A survey form was constructed to list various statistical methods that are commonly used in on-farm research. The form was sent to 66 researchers who had at some time engaged in on-farm research. Many responded, but of those, most were not currently conducting on-farm research. Only eleven reported current investigations. Most of the identified statistical methods were being used. Seven computer programs were reported as having been used. Most prevalent statistical problems were data collection on location, no replications, different varieties being compared in different sites, lack of statistical methods for "whole system" analysis, and loss of unknown yields.
Impacts
Results of the survey indicate areas of statistical problems for future research. They also indicate problems that are not addressed in traditional statistical training programs.
Publications
- No publications reported this period
|
|