So I have done a research on the difference in the perception of basic emotions between American and Japanese culture. I had 14 pictures of random people portraying a certain basic emotion. Each of this picture has a correct answer (the actual emotion being portrayed). The second group of questions were always related to the picture and it was about intensity of the emotion portrayed. So if the participants said that the person on the picture is portraying for example Happiness, the following question would be to determine the intensity of emotion being portrayed from 1 (Very little) to 5 (Very). So the first dependent variable is whether the person was able to determine the correct emotion, and if not if it had any similarities with others of the same nationality, and the second dependent variable is to determine whether there is any difference in perceiving the intensity of those same emotions between Americans and Japanese. And of course, the independent variable is nationality (American & Japanese).
What would be the process in SPSS? Can anyone help?
Also, If anyone can answer if there is any difference in doing the process in SPSS if my sample size is not balanced (31 American participants & 23 Japanese participants) - I know the sample size is not enough, however, since the coronavirus situation and me being forced to collect all my data through online means, it was very difficult to gather participants so I was excused.
MANOVA requires two or more continuous (i.e. interval or ratio) dependent variables, and so, is not appropriate here. You will need to conduct two separate tests: one for accuracy and one for the continuous DV. As shown here, there are several statistical tests one can use with a categorical dependent variable. Chi-square or Fischer's exact test would work in your case.
https://stats.idre.ucla.edu/other/mult-pkg/whatstat/
Regarding the samples, no there is no difference in the procedure. The output in SPSS will give you the results of the statistical test with (and without) assuming equal group variances. You can use Levene's test to assess equality of variance.
Blaine Tomkins Thank you so much for you answer! I really appreciate it a lot! However, unfortunately I have never done Fischer's Exact Test at my university so I am unfamiliar with it. I tired to do it in SPSS and something keeps confusing me... I will try to explain it here. So in my questionnaire I had 14 pictures. Each picture is a different individual portraying a specific emotion (Happiness, Disgust, Sadness... and so on). Under each picture there were offered 10 multiple choice answers/emotions from which one was correct (the same option of answers under all 14 pictures). Then after each picture there was a question about the intensity of emotion being portrayed on the photo they were just identifying (Likert scale from 1 [Very Little] - 5 [Very]). Some of the emotions were portrayed in more than one photo. Then 31 American and 22 Japanese participants filled my questionnaire.
Now. When I attempted to do Fischer's Exact Test (following steps from Youtube video tutorial) I got confused as for me each question is separate - meaning I get 14 different tables (individually for each question/picture). Is there a way to combine them into fewer tables or am I supposed to report each table individually and then the same for intensities (additional 14 tables)? Cause that seems like a lot of work and I don't know if I am doing it right (most probably not).
Thank you in advance!
Hello Zoran Sukurma. As Blaine Tomkins suggested, you will have to estimate two separate models, one for the dichotomous outcome, and another for the ordinal outcome with 5 levels. As you have explained, you have 14 measures for each participant, 1 per picture. Your models must take into account the correlated nature of those repeated measures. The two main ways of doing that are generalized estimating equations (GEE) and multilevel modeling, with pictures at level 1 clustered within participants, who are at level 2. The GENLIN command allows you to implement GEE, if that's the route you choose. For a multilevel modeling approach, use GENLINMIXED. The book by Heck, Thomas & Tabata (2012) shows both approaches, I think. (My copy is in the office on campus and inaccessible right now, as I am WFH!).
https://www.routledge.com/Multilevel-Modeling-of-Categorical-Outcomes-Using-IBM-SPSS/Heck-Thomas-Tabata/p/book/9781848729568
HTH.
Bruce Weaver Hi, thank you for your response! I have never done any of those at my university so I am unfamiliar on how to use it. I am [in a way] trying to replicate the following study:
http://www.davidmatsumoto.com/content/1989%20American-Japanese%20Cultural%20Difference%20in%20Intensity%20Ratings%20of%20Facial%20Ex.pdf
which I simplified by removing several variables such as the importance of which nationality is the person posing on the picture, Caucasian male, Caucasian female, Japanese male, and Japanese female, and so on. Dr. Matsumoto and P. Ekman analyzed all the data by using ANOVA's and I am unsure how can I do the same as they did. As what you are suggesting is way more advanced than I was taught. If you have time to look at the link I posted above from the pages 149 - 154 and the way they did it and tell me if there is any way I can get to to what they did?
Thank you in advance!
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS