I am not sure how I would analyse this data:
The experiment is looking at the accuracy of emotion recognition in children with ADHD. The children will be asked to identify what emotion a face is portraying (out of 6 expressions; happiness, sadness, fear, anger, disgust and neutral). They will firstly be shown the photographs and asked to identify the emotion, then they will be instructed to focus on the eyes and asked to identify the emotions and then asked to focus on the mouth. (three conditions- free gaze, eye gaze and mouth gaze). The experiment involves 2 groups of participants, ADHD children and matched controls. The DV is accuracy of emotion recognition.
How would I analyse this, would this be a repeated measures ANOVA to compare accuracy of emotion recognition by participant group (ADHD or controls) gaze condition (free, eye, mouth) and emotion type? I think this will determine whether there was a significant difference in the accuracy of identifying emotions in the difference gaze conditions and types of emotion portrayed. And if significant, follow up Bonferroni tests will be conducted to show which emotions were recognised the least accurately? t tests? planned comparisons? Im not too sure what design this is or what type of ANOVA should be conducted. Would I do seperate ANOVAs for each emotion? I am a bit confused!
Best Answer
The form of your data is going to be a table giving each emotion three times (once for each gaze instruction) as columns two rows, control and ADHD, and the entries being the number of kids who got that identification correct (assuming the total number of children allocated to each group is identical). Do I have that right? If so, then you're looking at a contingency table. Here's a nice introduction.
But now we need to know exactly what hypotheses you want. The simplest one is just 'The children with ADHD correctly identify emotions less well than the children without.' Then you would just run your contingency table saying that the null hypothesis is that the two rows in each column have, on average, the same number of counts.
However, you probably want to pull apart more than that. Perhaps you also want to know whether different gaze instructions help more. You can just run the table again, but this time taking the three gaze types as the rows and the two groups of children for each emotion as additional columns.
Also, you can do much better than Bonferroni for corrections. Bonferroni makes no assumptions whatsoever about the hypotheses, so it's the most conservative possible use. Look into stepdown methods, or for hypotheses that are probably independent such as gaze instructions being helpful vs ADHD kids doing better or worse (are they independent? that's a theoretical question for the domain specialist) you can at least use the Šidák correction, which is far more powerful.