Steps for Experimental Design
1. Figure out what you want to explore and formulate a research question based on a previous theory.
2. Formulate a hypothesis
3. Define your variables
4. Decide which type of experiment is appropriate
5. Carry out statistical analysis and discussion
When you are designing an experiment in psychology you will have an experimental hypothesis and a null hypothesis. Both of these will predict how the variables relate to another.
The null hypothesis predicts that there will be no relationship between the variables.
The experimental hypothesis predicts that where will be a relationship between the variables; however, there are two kinds of experimental hypotheses. A non-directional hypothesis will only say that there will be a relationship between the variables. For example, caffeine will have an effect on motor skills. A directional hypothesis will state what kind of relationship will exist between the variables. For example, caffeine will have a positive effect on motor skills OR caffeine will have a negative effect on motor skills. If we can accept the experimental hypothesis, we can reject the null hypothesis and vice versa.
In an experiment you must have an independent variable and a dependent variable. The independent variable is the manipulated variable or in the case of categorical variable, one with limited levels. For example nationality (Levels: English and Chinese). The dependent variable is what the independent variable will have a causal effect on.
There are also confounding variables. However, unlike the independent and dependent variable, we want to avoid confounding variables in our experiments as much as possible. Confounding variables are variables that could influence the dependent variable. They have a systematic effect on the conditions. To reduce the effect of confounding variables you need to ensure the the independent variable is the only difference between experimental conditions. This includes random assignment, random sampling, controlling for age, gender and skills, etc.
Deciding on the Right Experiment
In psychological experiments your two options are within participants design (or related samples) and between participants design (or unrelated samples).
Within participants designs compare the performance of the person or participant across all the conditions. This is done through repeated measures or a related design. This type of design is usually preferable. The advantages are that you need fewer participants, you have much better control of confounding variables because you are comparing the person against themselves. Of course, there are also disadvantages of a within participant design. The first one is called the carry-over effect, which means that once you have learnt a skill it’s hard to unlearn it or forget it. This means that if a person makes progress in a different condition it could just be down to practice. The second disadvantage is the order effect, which is when practice, fatigue or just plain boredom affects the performance of the participant. Luckily, most order effects can be avoided by counter balancing. This is when you randomly assign participants to the conditions, so that one person’s first task might be the second person’s fourth task. Controlling for confounding variables helps increase the experiments internal validity (the extend to which we can relate changes in the IV to the DV). External validity relates to how well the findings can be generalised to the population at large. External validity relies on random assignment, random sampling and ensuring that factors such as setting do not differ from experimental settings to the real world.
Between participants designs compare the performance of participants each in different conditions. One participant is only exposed to one of the conditions. Between participant design is the best alternative for within participant design. It should be used when you are trying to make comparisons in performance across different groups like gender, age groups, culture, etc. The main advantage is not having to deal with order effects or the carry-over effect; however, you do need more participants that are far more similar to each other to avoid confounding variables. This must be done before the experiment and is known as matching participants.
Sometimes there is also a third type of experiment called a one-sample design. This is when a sample mean is compared against a known population.
ZHENG, Y. (2013). Referencing and citation – Harvard style, from PSY104 Methods and Reasoning for Psychologists. University of Sheffield, Richard Roberts Building on 18th February. Available from: Blackboard.