my dear fellows. I mainly do experimental research, so I run experiments in which participants play a game repeatedly (for 20 periods for example). The dataset I get after the experiments would be that each individual's decision is measured 20 times consecutively/repeatedly.
Consider a MWE of my data as follows (in R):
Subject_ID <- c(1, 2, 3 ,1, 2, 3) Period <- c(1, 1, 1, 2, 2, 2) DV <- sample.int(20, 6, replace = TRUE) IV1 <- c(5, 2, 7, 5, 2, 7) df <- bind_cols(Subject_ID = Subject_ID, Period = Period, DV = DV, IV1 = IV1) df # A tibble: 6 x 4 Subject_ID Period DV IV1 <dbl> <dbl> <int> <int> 1 1 1 14 5 2 2 1 4 2 3 3 1 17 7 4 1 2 15 5 5 2 2 5 2 6 3 2 15 7
In this MWE, DV is the dependent variable, IV1 is the first independent variable. This is a dataset for one treatment, and usually, I would combine the dataset for all treatments together.
For data analysis, I usually do rank-sum test for treatment effects first, then I will do some regressions. If the outcome of the game in each period is only provided at the end of the experiment, the data would usually pass the serial correlation test, and thus I can use the pooled OLS regression like
DV ~ Constant + IV1 + error(heteroskedasticity-robust standard errors can be used if heteroskedasticity is detected).
When the dependent variable is a binary variable, I would use the pooled logit regression or pooled probit regression.
However, recently, I read some articles and posts on generalized estimation equations, conditional logistic regressions, (generalized) linear mixed effects models, and I am a bit overwhelmed. I really don't know which is the right one to use for the kind of data that I collected from experiments.
Is there anyone who can help me? Or is there any book/article that you would recommend to me which clearly deals with the dataset that I mentioned at the beginning?
Thank you so much!