In psycholinguistics research, why often people use per subject and per item analyses of variance (ANOVAs) instead of Mixed Effect Linear Regression? Which one is better to analize nested reaction time data?
Hello Tekabe Legesse Feleke. I think there is pretty good consensus nowadays that the mixed model approach is better. The 2007 article by Hoffman & Rovine has some relevant discussion and an example you might find helpful.
https://digitalcommons.unl.edu/cgi/viewcontent.cgi?article=1420&context=psychfacpub
HTH.
For reaction times you should use survival analysis: Article Analyzing Response Times and Other Types of Time-to-Event Da...
That looks like an interesting paper, Sven Panis. I look forward to reading it when I have time. As you know, psychologists who treat RT as a primary outcome often analyze error percentages in the various conditions too. And as you know, they often use ANOVA for that purpose too. But nowadays, they could use a mixed effects logit model (or a logit model with GEE) instead. ;-)
As explained in the paper, the discrete-time hazard analysis of event times [h(t) = P(T=t|T>=t)] can be extended with a micro-level speed-accuracy tradeoff analysis by plotting discrete-time conditional accuracy functions [ca(t) = P(correct|T=t)] in case of discrimination data. Hazard models and conditional accuracy models can both be implemented as generalized linear mixed effects regression models.
I assume when you say ANOVA you mean random effects ANOVA as used in Clark's (1973) fixed effect fallacy paper in JVLVB. If so, both that and the mixed/multilevel models are accounting for variation by people and items as random, so for some designs they will be operating similarly. As Bruce Weaver says m/m models are more frequently used as it is easier to extend them. As far as time as a special variable, the early use of this is covered well by Luce (1986) book, but as Sven Panis shows more recent models exist, and lots of those (e.g., the sample methods like Ratcliff's), but they will depend on your research question.
In terms of analyzing nested reaction time data, both per subject and per item ANOVA and mixed effects linear regression can be used. But mixed effects linear regression may be a more flexible and powerful tool for modeling these complex relationships. It allows you to model the relationships between the predictor variables and the outcome while taking into account the nested structure of the data.
However, it is important to note that the choice between these methods ultimately depends on the specific research question and the nature of the data. In some cases, a simpler method such as per subject or per item ANOVA may be sufficient, while in others a more complex approach like mixed effects linear regression may be necessary.
The choice between the ANOVA and the MELR depends on the objective of your research, the study's design, and the type of data being used. Each has its advantages and disadvantages. MELR is probably the better option when the research issue entails analyzing the impact of several independent factors and their interactions on a dependent variable, or when the data have a hierarchical structure.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts