Chapter 2. Scientific Method and Experimental Design

General Background Information

Science is conducted in many ways. The two most common are discovery science and experimental science. Discovery science is where new things or ideas are revealed. Experimental science is where predictions are made and then experiments are designed to test those predictions.

Experimental science routinely follows the scientific method. The scientific method is a method to logically answer questions (Figure C-1).

This process begins with observations. These observations lead to the formulation of models about a particular phenomenon. Based on the generalized model, specific hypotheses can be formulated. These hypotheses predict what will happen under specific conditions. As a result of these predictions, the hypotheses can be tested experimentally, producing new observations. These observations may be interpreted and then be used to improve the model. Science moves forward by producing more and more accurate models of natural phenomena.

The formulation of a good null hypothesis (H₀) and a good alternative hypothesis (H_a) will ensure that when the data from the experiment is analyzed one of the hypotheses will be rejected, thereby allowing the other one to be supported. The null hypothesis and alternative hypothesis for your specific experiment should be written in your laboratory notebook prior to beginning the experiment.

Experimental Design

Once you have determined the hypotheses to be tested there are several other issues that you must decide as you begin to design your experiment. Experimental design deals with selection of a variable to be studied and the choice of a sampling program. It does not deal with experimental techniques used to gather data.

In the planning of an experiment the following information should be written in your laboratory notebook prior to beginning the experiment:

1. How many variables and which variables are you going to manipulate, and what is the treatment being studied?
2. Will you have a control for the treatment?
3. How many replicates of this experiment will you do?
4. What is the appropriate scale of the measurements in your experiment?
5. What are the criteria for accepting or rejecting a hypothesis?

The most commonly used experimental design is the two-sample comparison. To do this you select two situations in which all conditions but one are the same. One situation, usually more “normal,” serves as the basis for comparison. The other situation, which is our focus, is the experiment in which you vary the factor of interest. By comparing data obtained from varying the factor of interest, you can make some conclusions about the effect of the variable on the process being studied. If more than one variable is altered, there is no certainty about which variable caused the effect you were examining—at least without using a very complicated experiment and extensive data analysis.

You need a standard to compare your results to—something in order to know if the condition you varied had any effect. The “normal” or unvaried situation serves as the control.

For instance, to know if additional light resulted in an increase in plant growth rate, you first must know the plant’s growth rate under the same conditions without additional light.

You need to control for any factors that might create an effect due to your experimental design. For example, the lamp in a plant growth experiment will give off both light and heat.

Replicates assist in determining whether the difference detected is experimental error or a difference which is related to the variable being manipulated. The number of replicates used is often a balance between practical considerations and how reliable the data need to be. In some experiments, replicates are impractical or maybe even impossible. Where replicates are possible then consideration of time and available material may factor into the determination of how many are done. In general, more replicates provides for increased reliability of the data. For example, if you are testing the safety of a drug, then the data would need to provide answers that were very reliable and there would be a need for a large number of replicates. For the experiments you will conduct, three replicates will be used.

If you are measuring the rate of a response, you need to pick a reasonable time frame. For instance, some reactions occur very fast (seconds or less), while other reactions occur more slowly (minutes or longer). If you are dealing with a fast reaction, then taking the measurement after 2 minutes means you could easily miss some of the most important observations. On the other hand, for a slower reaction, taking lots of measurements at short intervals is a waste of effort. For example, you would not measure the growth of a child by taking daily measurements. If you are uncertain of the reaction rate, then you may do a trial run reaction before your experiment to get some information about this aspect of the experiment.

How big of a difference do you need to see before you consider it a meaningful difference? Generally, any difference should allow you to reject the null hypothesis and accept the alternative. However, if there is a lot of variability in the data then the difference you detected may not be meaningful (or meaningful enough to reject the null hypothesis). How can you decide if a difference is meaningful or not? The most accepted way is to use a statistical analysis of the data, which will be covered in a separate section.