What Makes a Good Science Experiment

Whether you are a student competing in a science fair, a teacher looking for an interdisciplinary classroom project full of teachable moments, or a researcher with Ph.D.s and fellowships, the first step to conducting a good science experiment is to be interested in your subject and curious about the results. Though some will head straight to the exotic, like electromagnetism and seismic geology, when in search of a subject, just take a look at your life. What do you do, and what are you interested in? Is there something you think is true even though everyone else says it isn’t?

People have been observing things and telling other people about their observations ever since there were people. An experiment is a systematic way to test those observations and give your results credibility. By providing the steps you took to achieve the results, or methodology, you provide the basis for other people to challenge or accept your results.

The main purpose of an experiment is to determine whether X causes Y and why X causes Y. X is defined as the independent variable, the treatment given to experimental groups, whereas Y is the dependent variable, the effect of the treatment given to the experimental group. Imagine that we have a garden full of sunflowers. We want to determine whether fertilizer will help them grow faster and larger. X is the application of fertilizer, and Y is the growth size and rate of the sunflowers.

We carefully measure the amount of fertilizer and keep notes on when we applied it. We create a spreadsheet to track the growth rate and size of the sunflowers. After it all, we find that they grew faster and larger than the packet of seeds said they would. The results sound great, so we put our entire experiment on a webpage for other people to see how great fertilizer works on sunflowers. We’ve got several problems here though, and all of them point to important considerations in the experimental process.

What if Betty applies the fertilizer exactly as we did, and her sunflowers are stunted? Can we say why this happened, based off our experiment? A science experiment requires exactness. We need to describe everything about the X we used. In the case of fertilizer, it will be enough to state the brand and basic product information (like N-P-K ratio and trace mineral content), as well as the variety and source of the seeds. That way Betty knows exactly which fertilizer and variety of sunflower to use to attain our results.

What if Archie applies the same fertilizer we did and does not achieve our results? At this point we might start guessing that maybe Archie’s sunflowers did not get enough water or enough sun. Maybe Archie’s flower bed has a lot of clay and rocks, while ours has six inches of rich loam. These are referred to as extraneous variables, and they are defined as possible other causes of the results. In a science experiment we attempt to account for and control as many extraneous variables as possible. In order for Archie to check to see what went wrong in his experiment, he needs to be able to compare how much rainfall, sunlight, and the type of soil our sunflowers had with those his sunflowers had.

And then comes Veronica. She is not sold on our recommendation to use that type of fertilizer on sunflowers. Her big question is how do we know that the fertilizer caused the growth? Maybe our seed packet was collected from a batch of sunflowers that were bred to grow faster and larger. Maybe our area’s rainfall was heavier that season than normal. To address Veronica’s concerns, we should have added a control group.

A control group is the primary way we eliminate extraneous variables in an experiment. After cataloging the soil, water, and sun conditions of our plants, we then establish two groups of sunflowers under the same conditions, using seeds from the same packet. If the fertilized group still grows faster and larger, then we can say with much more certainty that the fertilizer was responsible. The conditions of the control group should match those of the experimental group as closely as possible.

Consider an experiment on lesson plans. We want to find out whether one is better than the other for teaching children the scientific method. Since we can’t split a class in half and have the teacher use the two lessons at the same time, we pick what we consider the most important extraneous variables and find a matching classroom. We’ll want both classes to have similar demographics and socioeconomic statuses, both teachers to have similar teaching styles, experience, and past success, and both schools to be located in similar neighborhoods with similar facilities and funding. Controlling variables simply means that we eliminate them as a possible reason for the results of the experiment.

So far, we have covered three basic elements to a good science experiment:

1) a clear hypothesis that states the independent variable we are testing and the dependent variable(s) we are measuring;

2) a methodology that includes exact information on the process we used to achieve the results;

3) and a control group to eliminate the influence of extraneous variables and establish certainty that the independent variable was responsible for the results.

Assuming we did all this, could we say for sure that the particular fertilizer used in the particular conditions will cause sunflowers to grow larger and faster? Simple answer, not without a lot more information. Perhaps our experimental test bed had a higher nitrogen content than the control. Perhaps the control bed was deficient in trace minerals. A nearby shrub or tree might have grown roots under the control bed after we cultivated it, and this is sapping nutrients from those sunflowers, making the fertilized ones look larger by comparison. Can you think of more extraneous variables that might cause the results?

To run a good science experiment is to recognize all the limitations created by the real world and look for creative ways to control them. Consider this real experiment designed to answer the question, does the presence of green spaces reduce violent tendencies? It’s a hard question to answer, but the researchers happened upon an excellent way to control many variables at once. They used the Robert Taylor Homes project (which has since been demolished), because all the apartments and buildings were the same, demographics and socioeconomic status were the same, and average time lived in the area was the same. Further, apartments were randomly assigned, so that none of the participants chose where they lived. Why do you think this would be important?

What differed was that some buildings were surrounded by parking lots and others by green spaces. By having independent people code the apartments as green or not green, and having another group of independent interviewers ask randomly chosen participants a set of questions, the researchers found that green space does tend to curb violent tendencies, resulting in lower domestic abuse and less physical punishment of children.

Without knowing anything else, can you see any possible limitations? Could the results of this experiment be generalized to the entire population? Why might it be difficult to run this experiment somewhere else? Even a good experiment will raise questions, and the best will attempt to answer the more obvious ones.

Finally, but most importantly, a good science experiment is based on an understanding of the subject gained from reading the work of people with more experience studying it. The authors of the green spaces experiment were basing their work on Attention Restoration Theory. In this case, they were attempting to prove the theory, or at least add to the proof for it. Not all experiments are based on theory, but all good ones are careful to review past experiments on the same topic.

To the list above, we are ready to add a final two criteria to a good science experiment:

4) alternative explanations (another way of saying extraneous variables) should be accounted for, especially if they aren’t easily dismissed;

5) the work of other scientists on the same topic should be consulted before the experiment is designed. It’s always likely that someone has already dealt with a problem that you are going to experience, and knowing about this beforehand will only help you design a better experiment.

If this article leads to more questions than answers for you, then feel free to write to me!