Sampath Jayasinghe, Senior Research Analyst, Decision Innovation Solutions
Decision Innovation Solutions (DIS) has proven experience in analyzing data related to livestock and poultry experimental feed trials. Initially, to explain this process some introductory definitions in livestock feed experiments are necessary for clarity. Next, an overview of the key principles and concepts of an experimental design will be provided. In conclusion; a brief overview of our work experiences in feed trial analysis will be highlighted. Overall this synopsis emphasizes the value of rigorous data analysis on feed and feed additive research and development.
Definitions
In a livestock feed trial, we are dealing with experiments with various test subjects, for example sows in a hog barn. Whenever an experiment is performed, first there is a need to identify whether a treatment was imposed or not. If a treatment is not imposed, it is an observational study. Therefore, there is an applied experimental condition which is called a treatment in every case. Next, there are explanatory variables that are called “factors”. For example, in a livestock feeding trial involving sows, if you are trying to test a new feed additive named “BetaX” on a few indicator measurements of sow performance, the BetaX additive is the factor. If varying amounts of the BetaX are used, they are called “levels”.
Key Principles and Concepts
There are three basic principles of any experimental design. 1) Randomization, 2) Replication, and 3) Control. Randomization is a way of assigning treatments to the experiment making sure that each possible allocation to treatments has the exact same probability. Replication is the number of repetitions of the basic experiment using enough subjects to reduce chance of variation. The number of replications is an important parameter in the calculation of experimental error as well as the final confidence of the applied treatment effects. Control is a way to minimize all extraneous sources of variation in the experimental design. This will decrease the experimental error and increase the efficiency of the design. The control can be either no treatments or a known treatment.
The concept of statistical significance is important in how we are drawing conclusions using the data generated in the experiment at the end. A statistical significance event means an observed effect is so large, that it would not occur by chance. Doing so we can check to see if the observation between treatment and control is due to the effectiveness of the treatment.
This will complicate things when it comes to time series data. The issue here is, can the observed difference between treatment and control be due to time trend and not purely due to the effectiveness of the treatment? Delignating effects of time trend and treatment is very important in deriving a conclusion at the end. There are ways in statistical theory to solve these complicated data analysis problems.
DIS Application of Experimental Design Analysis
Decision Innovation Solutions has worked most recently with clients such as Nutriquest, Phileo Lesaffre Animal Care, and Feed Energy in feed trial data analysis. DIS has contracted as their statisticians to analyze their feed trials data and provide rigorous and in-depth analysis for the final data collection. Tests are then performed on many different hypotheses to see if given treatments to subjects have shown statistically significant differences compared to the control subjects. DIS’s analyses have been able to determine that there are significant observed effects so statistically overwhelming, that when compared to the control, these occurrences happening by chance or time trend can be ruled out. This provides evidence of the real effectiveness of a feed or feed additive on real test subjects.
For an additional example specific to sow feed trial data, DIS statisticians have looked at whether new feed additive treatments on sows have had statistically significant effects on key sow growth performance measures, including but not limited to:
- Conception rate
- Farrowing rate
- Non-productive days
- Sow mortality
- Total born
- Total born live
- Pre-weaning mortality
- Weaned piglets per sow
- Piglets per sow per year
We used parametric as well as non-parametric tests to examine whether there are significant differences under each performance measure. We also applied a novel statistical method to isolate the time trend effect and treatment effect as these data have time series properties.
With respect to clients that provided data on turkeys, we analyzed some key performance metrics between control and treatment flocks for a given number of closeouts. Some the key performance variables analyzed in these studies include:
- Livability
- Net pound per net number sold
- Average live weight gross
- Actual condemn
- Feed efficiency
For these and all clients, statistical tests are performed for sample means taking into account any issues that arise in regard to sample variances.
DIS statisticians also have experience in experimental data analysis related to bacteria such as E. coli and clostridia in ilium, ceca and litter in turkey barns. By doing rigorous data analysis, DIS has helped our clients determine if the treatment has statistical significant impact on reducing the bacteria, compared to the controls.
Combined, these “third party” analyses provide our clients with the highest quality information to make the best decisions and find the best and most innovative solutions. This can give the competitive edge necessary to remain ahead, and provide the best inputs to producers and by default the highest quality products to the consumers of animal products.