I have a data set consisting of reaction time (RT) measurements. These will be used to calculate the duration of experimental trials in an MRI study. In each block (experimental condition) there are 10 trials. Now, due to various issues, there are some missing RTs. Due to the nature of MRI analysis, I need values for all 10 trials per block. Where more than 5 values are missing I will probably discard it from the analysis completely, but where only 1 or 2 are missing I plan on using the mean RT value for that condition in place of the missing value. However, I want to make sure that this is a principled decision to make, as the RT values in some conditions, by the same participant, can be quite variable.
How can I use the standard deviation or standard error of the mean to ensure that it is 'fair' to use the mean in place of the missing value? For example, see the data below.
Block 1 - Missing values: 2; Mean: 740; SD: 519; SEM: 196. Block 2 - Missing values: 1; Mean: 2245; SD: 292; SEM: 97.
I'm trying to figure out an honest, consistent way of deciding whether the decision to replace the missing value with the mean is sound. Where it is not, I would rather leave the block out of the analysis than skew the data.
Any advice? I hope this makes sense.