Can invalid D scores be considered missing at random? I'm looking to do multiple regression with D and several other variables as predictors. Besides multiple imputation (or listwise deletion), any ideas for other ways of dealing with cases who do not have valid D scores because of excessive reaction times or errors?
Dear Paul,
As a start on this discussion: If you did not state a limit for the reaction times beforehand as we used to do in the mouse laboratory with our mice ambulating in the open field.; We attributed our beforhand denoted max time to those animals who froze or did not locomote e.g. to 2 min. If you did not beforehand apply a max reaction time you can ascribe now a max reaction time based on the normal curve. You can apply a nomal curve to your data and write in the method section that you excluded those values that exceeded 2,75 SD from the mean for example. You can do what you want if you clearly state what you did in the method section so that your research can be replicated whenever considered needed. You can also compute differences between extreem groups: between those with extremely long reaction times compared to those with extremely short reaction times. Then if there is no difference between the groups you know that.
You can also consider these extreme data as outliers and You can then read how to handle outliers - there is more than 100 different methods - or you can ask Jochen here at RG.
I agree with Beatrice. The best approach, as these are not really missing but rather 'outlier' responses, would be to create a new variable, or new variable category that indicates that these were beyond your time limit. However, if you are unsure of why the values are missing, that raises other questions. As Alvin mentioned, you can use SPSS, or similar, to check for MCAR, or MAR. If your data is MCAR, it is usually fine to use an iteration method to replace missing.
Dear Paul,
which D measure did you use for computing the IAT effects? The improved scoring algorithm (D1, Greenwald et al., 2001) requires to delete RTs > 10.000 ms and to include error latencies in analyses.
If you were considering imputation you'd perhaps be better off trying to impute the trials in the raw data based on stimulus characteristics and responses (if of course it was MAR or MCAR).
As suggested before, I recommend using the improved scoring algorithm to calculate D scores by Greenwald and colleagues (2003). May be you may have less number of missing values in D scores:
Greenwald, A. G, Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the
Implicit Association Test: I. An improved scoring algorithm. Journal of Personality
and Social Psychology, 85, 197-216.
You can find more information on how to handle missing scores on Lane and colleagues (2007):
Lane, K. A., Banaji, M. R., Nosek, B. A., & Greenwald, A. G. (2007). Understanding and
using the Implicit Association Test: IV. What we know (so far) (Pp. 59–102). In B.
Wittenbrink & N. S. Schwarz (Eds.). Implicit measures of attitudes: Procedures and
controversies. New York: Guilford Press.
- Badges
- Science topic
- Similar topics
- Psychology
- Social Psychology