Data analysis mess! Huge data set, 2 factors with 2 levels each, but the levels are the same

Question

To explain the hypothess a little bit: The factors would be emotional words and neutral words; the levels of both factors are left and right visual hemifields. The data is the participants response time in milliseconds. I can explain our hypothesis: We predict a decrease in reaction time for the emotional words as compared to the neutral words; I guess we are expecting a decrease in response time when the stimuli is presented in the right visual hemifield... thats why each factor has the same levels (left/right visual hemifield)

I guess my main problem is that I don't understand how to identify the rows as the response variable... how can I make this dataset easier to run a statistical test on? I originally thought I would have to do ANOVA, but I realize that since each participant was assessed multiple times (with different words displayed on different sides of screen during lexical decision study) my data isn't independent so ANOVA would not be the appropriate test? I'm just not sure, I've tried entering it so many times in both excel and SPSS, I've even tried creating regression models in R... I'm not entirely sure they make sense though. I just need ANY KIND OF help or suggestions on what's the most sensible thing to do with this messy, long dataset.

The data: https://docs.google.com/spreadsheets/d/1czrXmJuxvwalDwNDLHIvt64QqNb8cSTahEEK8DIiQPM/edit?usp=sharing

There's multiple "sheets", because I tried too many times entering it in different ways for it to make sense. The second sheet is what I've mainly been working with.. It's the same dataset, I just tried to clean it up.. please help!!

Answer 1

The most straightforward way to organize data like these is to have 1 row per observation. Each row would then have data in columns for: the response time, the neutral/emotional category, the right/left category, and an ID for the individual tested. That data format then fits quite nicely into standard statistical software programs. You just let the software know which column contains the response values and which columns contain the corresponding predictor-variable values, based on the syntax expected by the software.

You are correct that you should take into account the multiple tests performed on the same individual. One way to do this is to treat the individuals as random effects, in which the variability among the individuals in response times is modeled. The best way to proceed then depends on how you expect the individuals to differ in terms of response times.

For example, if you think that individuals will differ in terms of baseline response times (some are just faster than others) but that right/left and neutral/emotional differences around those baseline times will be the same, then you could include a random-effect term for the intercept in your model. If you think that individuals might have a wide distribution with respect to right/left or neutral/emotional differences in response times, you could also include random-effect terms for the coefficients of those predictor variables, too. This page shows how to specify these types of random effects for one popular package in R.

This sounds like the type of project that is being done at an academic institution. If so, it would be wise to consult with local statistics experts, who might better be able to identify particular issues and approaches related to your study and its data than we can remotely on this site.