I am analyzing a data set of 100 subjects. I am looking at reaction time in different bins of 50 ms. However, each participants contains very few number of trials. I tried to group all trials from each subjects and treated all trials as if coming from a single participant. Is it correct? any other suggestions?
Hi Indrajeet Indrajeet
I'll repeat the question to make sure I have understood it correctly. You have a number of 50ms bins (consecutive?). For each bin you have trials (can be more than one?) for some of the participants out of 100. If this is the case, here are my suggestions for the analysis.
1. There is no harm in averaging the data across participants for each bin and reporting that average. But average by itself is not adequate, and it should be accompanied by standard deviation, confidence interval, etc.
2. In essence you have repeated measures imbalanced data and the most effective tool for this kind of data is a linear mixed model (LMM). You can construct a linear mixed model in one of the following two ways.
a. Bins are consecutive: A fixed slope effect for bin number (1, 2, 3) entered as a continuous variable. Or a fixed natural spline effect for bin number entered as a continuous variable. A random intercept effect for participants.
b. Bins are not consecutive: A fixed intercept effect for bin number entered as a categorical variable. A random intercept effect for participants.
In either case if the residuals of the model are not normal, you may use a Gamma distribution with a log link as the outcome is reaction time which is always positive. Here a few useful resources for LMMs.
Article Linear models and linear mixed effects models in R with ling...
https://ourcodingclub.github.io/2017/03/15/mixed-models.html
Article To transform or not to transform: Using Generalized Linear M...
Why are you bining the RTs rather than just using the times (perhaps transforming them). And if you expect differences among individuals, why average for bins not taking this perhaps valuable variable into account? @Usman and Jimmy Y. Zhong says there is no harm doing this, but I think they mean still keeping different values for different individuals. There are lots of RT models, depends what you are using them for so be good if you could provide more information.
ps. Also, depending on what the RTs are of, you should expect things like the mean and median to differ, so not sure about @Jimmy's advice on this. RTs usually fit a log-normal more closely than a normal. Maybe he can clarify what he means.
Thanks Usman Rashid Jimmy Y. Zhong Daniel Wright . I will try to follow your valuable expert suggestion
- Badges
- Science topic
- Similar topics
- Biological Science
- Neuroscience