Year 4 Semester 2

The second semester of the CDP was a research and statistics fiesta for me as I took two graduate level research modules and worked on my thesis.

• PL5222 Multivariate Statistics In Psychology
• PL5225 Structural Equation Modeling

Overall, I greatly enjoyed this semester as the modules were all highly practical. The focus of these modules was to build sufficient understanding and confidence such that we could apply advanced statistical and research techniques to answer research questions. I would recommend taking these modules the first chance you get, as the knowledge gained here can really sharpen the quality of your thesis if you can show that you recognize the advantages and limitations of certain research methods and design your thesis around them. The classes for both these modules were small but that is a good thing, as it is much easier to discuss among your peers.

Now for more specifics about the two modules:

PL5222 Multivariate Statistics

It is quite hard to exactly pin down what exactly is “multivariate”, the most common answer is that “multivariate” refers to statistical techniques that simultaneously handle multiple dependent variables. But look into the module and you will find various topics that do not fit into this definition. A running joke is that anything that is not part of GLM or any existing framework is simply classified into “multivariate”. Here are the list of topics covered during my semester:

• Linear Discriminant Analysis
• Logistic Regression
• Exploratory Factor Analysis
• Multilevel Modeling (4 Lectures)
• Meta-Analysis
• Categorical Data Analysis
• Cluster Analysis

Most of us came in with no knowledge whatsoever about these analysis methods but the module was still manageable. Having a solid understanding of linear regression is enough to grasp most of the topics here. Multilevel Modeling would be the big boss of this module, taking up four lectures. This is for good reason, as it is one of the more complex statistical technique in this list. But it is one of the most common analysis technique used in applied research settings so the pain would be worth it. The module had 4 (12.5% each) assignments, with around 2 weeks spaced between assignments. The final exam was worth 50% and focused more on conceptual understanding (advantages, limitations, interpretation, discussion) than applying formulas. Overall, I would say that amongst the graduate statistics module, this was the hardest. The greatest difficulty came from Multilevel Modeling. But I would encourage most students to take this module for its wide applicability to many applied scenarios such as IO, social service, educational, clinical research.

PL5225 Structural Equation Modeling (SEM)

Compared to Multivariate Statistics, I would say that SEM is harder at the beginning but becomes easier towards the end. The main reason being that there is a very obvious flow from the first to the last week, as the concepts from previous weeks are continuously applied in the later lectures. In Multivariate Statistics, the topics are more heterogeneous and concepts from one week may become irrelevant for the other 11 weeks. In short, SEM is like reading a story from start to end but Multivariate Statistics is like reading multiple distinct journal articles. The list of topics covered in SEM:

• Confirmatory Factor Analysis
• Path Analysis
• Exploratory Factor Analysis
• Structural Equation Modeling
• Latent Growth Modeling
• Multiple-Group Analysis
• Measurement Invariance
• Meta-Analytic Structural Equation Modeling
• Mediation and Moderation Analysis in SEM
• Missing Data Analysis

This module’s strength is the coherent narrative from start to finish. Content from the first week is still relevant up till the last week and personal progress and growth feels much more real compared to learning a different topic each week. Personally, I find the main advantages of the SEM framework is the amount of flexibility offered. There are SEM-equivalent methods for almost all other analysis methods (at least in Psychology), ANOVA/ANCOVA/MANOVA/MANCOVA, the Regression family, repeated-measures, etc…can be performed under the SEM framework. There are even advantages to performing such analysis in the SEM framework as SEM was originally built on the maximum likelihood estimators, which makes the extension to handle missing data easy and intuitive. I would also say that familiarity with SEM is almost necessary if you wish to do more research in Psychology as a career, this is because SEM focuses on the relationships between underlying latent and unobservable constructs, which is what most Psychological constructs are. E.g. We can’t actually see self-esteem, we infer it from behavior.

If you can, definitely take all three graduate-level research modules, even if you are an undergraduate.

Thesis

At this point, I was busy figuring out how to use Linux and utilize the school’s high performance computing system. My thesis was a little different from most others as I did not collect data. But my more normal peers were also in the process of designing and implementing their studies this semester. Those who were quicker were already piloting their data but most of us were still coding and designing our experiments/simulations. Literature review was still ongoing this semester, instead of focusing on crafting our research questions, we were focusing on how previous literature have studied whatever topic we were researching. Personally, this was the semester when I started feeling nervous and unsure of myself. Was I really going to produce something that can build upon what many years of research has already done? And to do it with at least the same quality as researchers who have been in the field for decades? But I look around and see that my peers were in the same boat, and many other graduates have also survived this process. So I just shrugged my shoulders and went back to figuring out how Linux works.

This was the semester where I racked my brain the hardest for the thesis. This phase is less mentally demanding (but more physically demanding), I was busy hauling my portable hard disk around and transferring data, juggling between different simulations and troubleshooting the system. It was also the most stressful phase as screwing up in the design of your experiment/simulation/data collection will severely compromise the thesis, and there is little time to redo this phase if the mistake is found later.

Read part 3 here!