Psychology Thesis Frequently Asked Questions (FAQ) Part 2
“ANOVA? But I am not manipulating anything!”
They are often not sure if they can use means-comparison tests in non-experimental design and the predictors are categorical such as gender or race, because they are accustomed to the tests being for experimental designs, in past studies as well as their lab reports, and therefore assume the tests are only for manipulative IVs. However, statistical tests are just algorithms to analyse data, and they do not care if you manipulate the variables. All they tell you is that if the difference in the means among the levels within the samples is due to chance whenthe difference is absent in the population. Just because you do not manipulate the variables, if you have only categorical predictors, you do not need to dummy code them because regression is not your only choice in non-experimental design. Still, you cannot infer causality regardless of the tests, and this is the only thing you need to concern about.
If you use any tests or inventories, you may ask:
“How to score the items? Do I really need to use all of them? The author did not give any instruction.”
They often have this doubt when they only have the items and basic instruction, especially when it only says “score on a 7-point Likert scale”, or the author only sent them the items without scoring. They do not know how to convert the responses into numbers or calculate the total score. I told them what matter were the items and what each of them measure, i.e. subscales, and it is up to them to convert into numbers in the way that is convenient for them. In an example of a 7-point Likert scale, the author may not show how to score each response, but you can score with the range of 0-7, 1-8, or even from -3 to 3 for each item.In addition, if the author did not specify, or you have a good reason to do so, you can even choose the type of scoring, such as 3-point Likert scale or even true-or-false, so far the items are not significantly changed due to copyright. For thetotal score, they can simply sum up all the items, or take the average of each item. Personally I would recommend the average rating because the number is easier to interpret. Nonetheless, you should get the same inferential result regardless the scoring you choose.
One thing you should remember is you MUST know the meaning of score because sometimes combined score does not make sense. For instance, it is not reasonable to combine all the items in the scale that measure gender identity (i.e. masculinity and femininity), and the subscales should be separated. However, if they measure the same construct despite different subscales, you can sum up the scales to indicate the level of overall construct, e.g. exhaustion and depersonalization subscales in burnout inventory, and the combined score shows the level of overall burnout.Also, one thing to note is you do not need to use all of the subscales, and you can choose some of them depending on your aim, such as you can choose only to use exhaustion subscale in burnout inventory.
Also, maybe they are influenced by the “psychological” tests from newspapers or magazines, they think the scale should have specific cut-off points. In psychological research, the scales often do not have aspecific cut-off to categorize high or low level. When you operationalize the scale, just simply state what the higher score means, e.g. in Rosenberg Self-esteem scale, the higher the score, the higher the self-esteem. We cannot, and it is not always necessary, to categorize them into high or low self-esteem. Nevertheless, in some situations, a continuous scale need to be classified, e.g. when another IV is categorical in quasi-experiment, so the data is easier to be analysed. In this case, you can categorize the data with various ways, and one of them is using the pre-existing cut-off in some articles. However, I do not recommend it (and do not like it personally) because the samples can be different from the past research using or validating the scale. The other way is to categorize based on descriptive statistics, and the medium split is the common method. Personally I like to categorize into three categories with one standard deviation from the mean as the cut-offs. Regardless of the ways to categorize, you can use any if you can justify it, and the justification can be as simple as easy computation.
Everything is done now, and you need to discuss your result, you may ask:
“How to write my discussion? And what are implications?”
For a simple answer, you can write almost anything in your discussion. Did the result match your hypotheses and support the theory? In your opinion, why did you get such results? How is it related to past literature and your aim? What are the limitations and how to improve it in the future study? As long as they are logical and rational, you can discuss everything, but be careful not to write mainly the limitations. Perhaps because of culture, many Asian students tend to write their discussion in the way the research is worthless. While doing research, you can be humble but not incompetent, and you can discuss the strength of your study relating to past studies, e.g. yours is one of the few studies done in experimental design to infer causality, or you considered a variable that only few research has thought of.
For implication, basically you need to answer the question “So what”, that is, you got such result, what does it mean? Why people should care? How does it contribute, to existing literature or the society? The two main implications are theoretical and practical implications. In theoretical implication, you talk about how your study contribute to the literature and current knowledge, e.g. Does it support a theory? Or does it refine or extend the theory? You can relate the result to your literature review, discussing how your study fills the gap of knowledge, and how it further supports or rejects past studies. In practical implication, you discuss how people can use your research, e.g. How your studies help others? Does it potentially make people’s lives better? Different researchers and lecturers prefer one kind of implication than the other, so I often suggest thesis students to know the preference and expectation of their supervisors, or even better, to write both kinds of implications.
Featured header image credit: http://exclusive-executive-resumes.com/
Jordan Oh (Veng Thang) is a 3rd year psychology student in HELP. He studied and has the experience in Education (Teaching Chinese as Second Language), and now is a member of Peer Mentors and PAL (Peer Assisted Learning) tutor in quantitative research and cognitive psychology. His interest is in soft science like statistics and psychology, especially about how people acquire knowledge and anxiety issue in academic setting, that’s why he loves the course. Also, he is gay.