The scientific method is a mystery to many people. They may hear about scientific discoveries in the news but they don’t understand how an experiment is conducted. So I wrote this for those who are curious about the scientific method.
A scientific experiment requires four basic components; a researcher controlled independent variable, a dependent variable, a random sample and random assignment. First I will define each then I will put them together.
Say you want to find out if fertilizer X is better than fertilizer Y for growing tomatoes. What would you do? Well first you would get a bunch of tomatoes of the same variety and both types of fertilizers. You would plant a third of the tomatoes with fertilizer X and a third with fertilizer Y and a third with no fertilizer. Every plant would have equal chance of getting fertilizer X, Y or no fertilizer. You would plant them all the same day. They all would be planted in a place where all plants get equal sunshine. Every plant would receive equal water and fertilizer.
The independent variable (IV) is what you think will have an effect. In this case it is the fertilizer. Take note that one group has no fertilizer. That is the control group, it tells us what would happen without intervention. That gives us a baseline to compare the effects of the fertilizers. Without a control group you don’t know what effect any fertilizer has on tomatoes. Another critical feature of the IV is that the researcher controls it. Since the researcher can control the IV you can make an attribution of causation. That is you can say IF I do X then A occurs, if I do Y then B occurs, if I do nothing then C occurs. If you are slightly confused don’t worry, we will come back to this after explaining the rest of the experiment.
The dependent variable is what you expect the independent variable to cause. In our experiment we could use a number of dependent variables. Maybe you think a better fertilizer leads to larger plants. Maybe you think it leads to more tomatoes. Maybe you think it leads to more nutritious tomatoes. An experiment requires a minimum of one dependent variable but can more than one. Since this is set up as a basic experiment that anybody could do in their backyard it makes sense to select plant size and amount of tomatoes as the dependent variable. Very few have access or the knowledge to test the nutrition of tomatoes. So now that we have selected the dependent variable we have to decide how to measure those variables. For plant size we could measure height or we could measure mass. In order to measure mass we could have to pull the plant out of the ground to weigh it and few people would want to kill their plants for that. Whereas measuring height is simple and easy which makes it a great measure for this experiment. When it comes to measuring the amount of tomatoes we could opt to count the number of tomatoes produced or measure the mass of tomatoes produced. Because it is easy enough to weigh the tomatoes I would chose to measure the mass of tomatoes produced. That allows me to tell if one plant is producing larger tomatoes even if it is producing the same number as another plant. Now one key to the dependent variable is that you must specifically define how it is measured and then apply it consistently. The definition should be clear enough that another person could duplicate your experiment on their own. For example when measuring the mass of tomatoes produced do you ensure that the stem is completely removed or can you leave some of the stem attached? I would remove the stem entirely so I am only measuring the tomatoes but I have to ensure that I include such specifics in my definition of my dependent variable.
Next is random sampling, that requires that every member of a population has the same chance of being part of the study. That means when you go to buy all the tomato plants every one of them should have an equal chance of being picked. We don’t want to go and just pick the healthiest looking ones or the largest. That would skew the results. There are number of ways to accomplish this. You could decide before going to the store that you would buy every 5th tomato plant you see. This type of sampling is done in survey research when dealing with a crowd like a event, rally or demonstration. Or you could decide that you will roll a 6 sided dice for each plant and take any that you roll a 6 for. The key is that all have equal chance. That allows you to generalize your results to the population as a whole because you pulled out a random group from the population so theoretically it should represent the whole group. Whereas if you picked only the healthiest plants then you could only generalize your results to the healthiest population of plants. Whatever group you sample for the study is what group you can generalize your results to.
The final requirement is random assignment. Every member of the sample should be randomly assigned to the different conditions of the IV. In our experiment all tomato plants should have an equal chance of getting fertilizer X as getting fertilizer Y as getting no fertilizer. This is done because there will always be individual differences in a population. There will be genetic and environmental factors in the past that influence the individuals whether those individuals are plants or humans or whatever they are. By randomly assigning individuals to groups then those differences are distributed between the groups. In theory the differences should be equally distributed. In our experiment say that some of the plants we bought were less healthy than others. Well if we randomly assign each plant to the three groups then it is likely that each group would end up with about the same number of unhealthy plants. That equalizes the effects of those unhealthy plants on the final results since they are distributed between all the groups.
Now it is time to put this all together. We have an independent variable (IV) which is the type of fertilizer (X or Y or None). We have two dependent variables (DV); the height of the plants and the mass of the tomatoes. We believe that the type of fertilizer (IV) will have an effect on the tomato plants which is measured as our DVs. In addition we randomly sampled the tomato plants from the store by ensuring each plant had an equal chance of being included in the study. Finally we randomly assigned all of the plants to one of the conditions of the IV. Otherwise every plant should be as equal as possible in treatment. They are planted in a place where all can get equal sunshine. They all receive the same amount of water. The groups that get fertilizer get equal amounts of fertilizer each.
This set up will allow us to determine what effects the fertilizers (or no fertilizer) causes to occur with the tomato plants. So how do we know that? Well we held all conditions we could control equal (water, sun, etc). The conditions we could not hold equal like the individual variations in genetics or past environment we distributed randomly among the conditions of the IV so as to equalize the impact on each group. The only change between groups was the IV which we controlled. So we can say with confidence that changing the IV is what caused the difference in the plants because it was the only feature being changed. Furthermore since we randomly sampled plants from the population we can generalize the results we see to that population. By randomly picking individuals from the population we are able to assume that the sample we tested would be equivalent to any other random sample pulled from the same population because the individual differences should be equally represented in our sample as they are in the population as a whole.
Well there you have it, we have set up an experimental design. This is the same basic design of just about any experiment. It could be applied to tomatoes or psychology or biology or just about anything. At a later point I will go over correlational designs which are used heavily in medicine, economics and sociology.
DISCLAIMER: The only piece that I did not and will not go over in this post is the statistical testing to determine if there really is an effect of fertilizer. Just because one group has slight taller plants or slightly more tomatoes does not mean that we can definitively say the IV caused that effect. That is because there are individual differences in plants and that must be accounted for when determining if there was an effect. We have to statistically test before making that declaration. Unfortunately explaining even a simple T test or F test is quite a lengthy process and beyond the scope here. Though if the differences are large then it is very very likely that the IV caused the differences because it is very likely the statistics would turn out that way. But I am warning against saying the IV caused a difference if the differences are minimal.