Hi,
I am relatively new to statistical testing and with universities closure I'm struggling with which test to use.
I am using a within-subjects factorial design. Where all participants view a series of 100 random images of objects e.g. a cat, slice of cheese etc (where 25% are pointed right to left and are 'agentic', 25% are pointed left to right and are 'agentic', 25% are pointed right to left and are 'non-agentic', 25% are pointed left to right and are 'non-agentic') then they are shown 100 images, 50% were in the previous trial (with 25% agentic, 25% non-agentic, 25% pointed left right, 25% right left) 50% are new images (the same directions and agency proportions apply). Participants must hit one key if they recognise the image and a different key if they do not recognise the image.
I want to test if the participants have faster reaction times to agentic images pointing left to right than other images. I also want to analyse the false alarm/hit rates.
I know for reaction time speed I should conduct an ANOVA but I am confused over which type as different textbooks give different indications.
I am available at all time and I do have a video of and microphone connection.
Thank you in advance for any help!
I don't disagree with Daniel Wright and Jochen Wilhelm. (They're both smart guys, and it doesn't usually pay to disagree with them.) But I do wonder if they are advising you to run before you can walk. Are you a student? If so, at what level? This sounds to me like it could be about an undergraduate final year project (or an hono(u)rs thesis as it might be called in North America). If it is, you must have a supervisor. What does he or she suggest? And what software do you have access to? Thanks for clarifying.
The response times are likely skewed. Models like the log-normal are often used. If you want to look closely at the relationship between response times and accuracy, I recommend looking at diffusion models. If you are more just interested in accuracy, signal detection theory models, framed as a logistic/probit multilevel regression could work (see Article Functions for traditional and multilevel approaches to signa...
). You could predict what people say from the design (direction, agentic, previous trial). Neither of these are trivial procedures, so I recommend connecting with someone at Cardiff. I should add both of these are commonly used in cognitive psychology so there should be people there with expertise.
I think Daniel is right. Reaction times are often assumed exponential distributed (more often than log-normal), so you might consider a generalized linear model of the gamma-family of distributions (gamma GLM).
To analyze the correct/wrong hit rates, I would suggest using a (beta-)binomial GLM.
I would also try to use the hits (correct/wrong) as another predictor as in a three-factorial design, so that you might be able to adress interactions of correct/wrong hits, pointing direction and agenticity.
I don't disagree with Daniel Wright and Jochen Wilhelm. (They're both smart guys, and it doesn't usually pay to disagree with them.) But I do wonder if they are advising you to run before you can walk. Are you a student? If so, at what level? This sounds to me like it could be about an undergraduate final year project (or an hono(u)rs thesis as it might be called in North America). If it is, you must have a supervisor. What does he or she suggest? And what software do you have access to? Thanks for clarifying.
Thank you Daniel Wright and Jochen Wilhelm for your help!
And thank you Bruce Weaver you're right I am a third year student doing my dissertation. I am using SPSS for this.
My tutor recommends I do a 2-way repeated measures ANOVA for my reaction time data, which has analysed the effects of direction and agency on rts and their interaction but I am confused as to how to incorporate the new and old levels of participant exposure to the images.
He has also told me to use d prime for my hits/misses data and to draw conclusions about direction and agency's effect on response tendency and discrimination performance.
I have never used d'prime before and I am struggling to find instructions for how to analyse the data it produces
If you outcome (response times) is count, you can use Poisson regression which is a GLM.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA