Application of Mathematics in supporting the Agricultural Experiment Design Model

[Kikis]

Application of Mathematics in supporting the Agricultural Experiment Design Model

 

The experiment is a series of activities in which each stage in the series is really defined; done to find answers about the problems examined through a hypothesis testing. The experiment used to test a hypothesis or provisional estimates of a natural occurrence or non-natural in the physical and non physical environment resulting from the hypothesis or answer these allegations. The hypothesis to be used can be positive or negative in the sense of receiving the results of the experiment as a positive value and reject the results of the experiment as a negative value.

By the time we observe a set of objects, for example a set of cultivation in paddy fields, we see different things that look cons, giving rise to these allegations or hypotheses, including:

1. Why is the appearance of each object seen varies, there is a high, low, good, bad growth, not attacked by pests and pest attack.

2. What are the causes of differences or variations in these?

3. What actions can we do for there uniformity in the object?

4. How the application or treatment provided?

To answer these questions have to do a series of studies or a series of experiments. The experiment became a tool for researchers in seeking answers to the allegations or non-natural and natural conditions that happen. The answer is a statement of the results of scientific experiments during the experiments conducted to follow the concepts and principles of scientific research.

 

Definition and Types of experiment

 

Pattern or ordinances implementing measures (treatment and non treatment) in an experiment on certain environmental conditions that later became the basis of the arrangement and method of statistical analysis of outcome data, called the Experimental Design (Experimental Design). In the field of agriculture are known several types of experimental design, among others:

1. Completely Randomized Design (completly randomized design)

2. Randomized block design (randomized Block Design)

3. Latin Square Design (Latin Square Design)

4. Factorial Design (factorial design)

5. Split Plot Design (Split Splot Design)

In this course will only be discussed Rancangang completely randomized, randomized block design, and factorial design. Usefulness of rancagan-bill vary, depending on environmental conditions.

 

 

The main principle in the design of experiments

 

In the experimental design is known that some key principles, which are an important part in using a model of experimental design. These principles should be owned in every model of experimental design used. As these principles are:

1. Replication (Deuteronomy)

If a treatment more than one pattern of study, then we say, a repetition occurs in the experiment. The purpose of the repetition is: in order to assess the diversity of experimental material. The more closely replicates the outcome, but rather the higher cost and difficult to control.

2. Randomization (Randomization)

Randomization performed for all variables or treatment used to have an equal opportunity to receive placement and observation, Randomization aims to ensure proper assessment of experimental error and avoid bias, and some things that need to provisions in determining the opportunities associated with the lapse of estimators and null hypothesis testing, making it possible to do statistical tests in which the changes are changes observed yyang have free distribution.

3. Local Control (Local Control)

What is meant by local control is actually the setting of research units in groups or blocks of a diverse group.

The objective is to enhance local control efficiency of a trial because of trial error becomes smaller so as to show the differences between the results.

Some terms:

1. Trial and error Trial Unit

Experimental error is to describe the results between the two sets of experiments with the same treatment that should give the same result, while the unit is an attempt to give the same result, is a unit test is a unit where a single experiment performed on a replication / replication of the experiment.

Experimental error can be minimized by means of:

a. Enhance the uniformity of experimental material

b. Careful in supervising the experiment

c. Using a more efficient experimental design

2. Treatment (Treatment)

Treatment means a set of special conditions imposed on the experiment on a limited set of experiments and design were selected.

3. Factor and Level

Is a factor that affects things or experimental measurable parameters that can be either deliberate treatment and can also not a treatment. While the level is the level of treatment in units smaller than a factor in each treatment.

4. Influence (effects) and interaction

Understanding the effect intended to investigate differences in outcomes of treatment when an environmental change. Medium interaction is intended to investigate the effect of differences in the results of the 2 factors on a treatment if one does not change or both.

[Bignose]

Is there a question here or something you want to discuss? Or are you just posting just to post? Because this looks like a random cut and paste job, which makes me suspect that someone is gearing up to do some spamming or similar.

[Kikis]

For Mr. Bignoise, i am really sorry for your discomfort, i just want to invite everyone for sharing and discuss about my post, and that's my post from my teaching material. And i am really sorry again and again if you really mad with my post.

Is there a question here or something you want to discuss? Or are you just posting just to post? Because this looks like a random cut and paste job, which makes me suspect that someone is gearing up to do some spamming or similar.

[Bignose]

For Mr. Bignoise, i am really sorry for your discomfort, i just want to invite everyone for sharing and discuss about my post, and that's my post from my teaching material. And i am really sorry again and again if you really mad with my post.

 

Its ok. It is just that this is a discussion forum, and unless there is something to discuss, posting something pretty much misses the point.

 

If you want to post something just to have it on the web, there are plenty of sites where you can start your own blog or website.

 

But if you want to post something to have a discussion about, then this is a good place.

[ajkoer]

I am a firm believer in Analysis of Covariance which is basically regression anlysis with continuous and dummy variables. Statistical significant of regression coefficients are hypothesis test on the factors. But why do I favor this approach? Because differences in variables, other than the major one you are testing, can be better detected (meaning a sharper analysis can be performed). Parsimony in variables, is however, important due to confounding and the actual accuracy for which regression coefficients can be estimated. You may wish to build a simulation model using a hypothetical distribution of reputed effects for the variable of interest and ACTUAL DISTRIBUTIONS FOR ALL OTHER VARIABLES IN THE REGRESSION MODEL. This will test and give guidance on the issues of variables construction, number of variables including interaction terms, and assuming a positive effect for the variable of interest, how good is your model design (statistically speaking since you know apriori the strenght of variable of interest hypothetically) in actually uncovering the effect.

 

Talk to some experts in the field to discuss why yield results vary (as a potential variable of interest). For example, sunlight may vary from plot to plot, so create a dummy variable to measure the differences in access to light(or water or wind or nitrogen delivery). Whatever factors you are using MUST be determined in advance (by a prior experiment, for example) and supported by expert opinion.

 

As a final word, which you may of aware of, Analysis of Variance models (ANOVA) can all be transformed into equivalent regression model forms. This is important because if some of the underlying assumptions are invalid (for example, normally distributed error terms), use the regression literature to suggest fixes (for normality issues, Box-Cox Analysis of Transformation) and even Robust Regression models (like Least-Absolute Deviations, Median Regression, etc). Also, from the perspective of Bayesian Regression Analysis, the results go from yes/no in ANOVA to a whole distribution of possible results.

 

Don't be surprised however if this suggestion is resisted by your professor as he may be poorly schooled in regression analysis (and how to fix assumption violations), spreadsheet based skills needed in constructing random variables for your simulation model, and anything about Robust Regression Analysis. By the way, if the simple ANOVA model has good attributes (or weak), your simulation exercise we confirm (or warn you).