Hi everyone,
I need some advice about statistical tests during longitudinal studies. In my current research, I want to describe second language learners reading skill development throughout an academic year. My eye tracking sessions will have a reading task and I will use eye tracking metrics such as saccade length, fixation duration and regressions. I am planning to correlate them with learner grades in my reading course and describe pedagogical development and reading behavior relationship. Due to the participant limitations, I assume that multilevel modeling, logistic regression or linear regression are not suitable and may not meet required assumptions. However, I insist that second language reading behavior is quite similar to first language reading in many ways and eye tracking metrics I mentioned above are strong predictors. I am planning to include 6 periodical eye tracking tasks for each participant throughout an academic year. And 6 different reading exams also. What do you think? Which statistical procedures are best in longitudinal research in education? Thanks in advance.
I've linked below to two books that are positively wonderful for readable approaches to longitudinal data analysis, particularly on the mixed-linear modeling approach. I have drawn heavily on these methods for my own eye-tracking and other-sort perceptual-motor studies.
You can find nice implementations of these methods in SAS and R as well. There is a lot of online help/forums for navigating both software applications.
Best wishes,
Damian
http://www.amazon.com/Applied-Longitudinal-Analysis-Garrett-Fitzmaurice/dp/0470380276
http://www.amazon.com/Applied-Longitudinal-Data-Analysis-Occurrence/dp/0195152964
Thanks Damian, books are great. It seems I need to change my research design as answers to my recent research questions are available only with hierarchical regression and logistic regression. I cannot apply them yet due to sample size limitations yet. My new design include 2 groups (Experimental and control groups) each of which will take 2 different second language instruction. I want to compare them in terms of 3 continuous variables as fixation duration, saccade length and regressions. I will acquire data in 3 time periods and I want to scrutinize the developmental differences. I also include pass/fail as dichotomous variable and grades as continuous variable. Can I use two way repeated measures ANOVA or do you have any suggestions? Thanks in advance.
Hi Emrah,
just a quick reply. First, I would also stick to mixed linear modelling, as Damian suggests. There is lots of help on the internet on that. Furthermore, I don't see why sample size should matter in Linear mixed modelling, but not in ANOVA. But maybe I'm wrong here. Second, you seem to have 4 independent variables: group, point in time of the data collection, pass/fail, grades. Two-way ANOVA are applied to designs with only two independent variables. Third, if your variable 'grade' is something like grades in school (A, B, C, D, or 1, 2, 3,4 ,5, 6) it should be considered as a discrete variable.
check out: "stackoverflow" combined with keywords like "longitudinal data", ... This is a good internet forum for people who kind of know their way around in R.
All the best,
Johannes
Thanks for the great discussion. Depending on the recommendation of Damian and Johannes, I made a quick research on mixed models. It seems I can apply linear mixed model repeated measures to have accurate results. I prepared a sample dataset in SPSS depending on my estimations and theory and tried to figure out if it works or not. After restructuring data, I applied the procedure. Although I am not yet proficient in mixed model interpretation, it seems that this procedure will work and I can focus on it. I also attach my output file and open for any advice and recommendations. Thanks in advance.
- Badges
- Science topic
- Similar topics
- Psychology