21st August 2013 at 11:32 am #1434Jessica AndersonMember
This may be a silly question but please help me anyway!
I have a large amount of non-numerical data (from likert scales: strongly agree, agree, neutral, disagree, strongly disagree) that I need to analyse. Do I need to change it to numerical form or am I able to perform accurate analyses on the non-numerical data?
I’m confused because Allen and Bennet specified to use numerical data only, while Andy Field seems to use non-numerical in his youtube clips. Any help would be much appreciated!!
Thanks so very much 🙂21st August 2013 at 4:03 pm #1438Dave CollingridgeParticipant
Not a silly question at all if you are not quite sure. Yes, you probably need to convert the names into numerical values. This can be done easily using programs like SPSS, Excel, and R. You might code this way: SA (5), A(4), N(3), D(2), and SD(1). The numerical assignment is somewhat arbitrary. You could reverse the numbering if you like.
Most Likert-type scales like the one you described are actually ordinal in nature because the distance between SA, A, N, D, and SD are not equivalent, but this does not stop people from analyzing individual questions with t-tests and ANOVA. In a theoretical sense, analyzing individual questions with parametric stats like t-test and ANOVA is incorrect. People should be using non-parametric statistics like the Wilcoxon/MWU and Kruskal Wallis to analyze ordinal data. There is one exception to this rule. If you aggregate questions (i.e., sum the responses for several questions and find the mean response for each participant), it is perfectly fine to carry out t-test and ANOVA on the aggregate values.
If you decide to aggregate questions you probably should make sure that those questions address similar issues. If it is not obvious whether two questions address the same issues, you can checking it statistically with principal components analysis.
-DC16th September 2013 at 11:08 am #1437Dr.RamanMember
It is interesting question,
Yes, that can be converted into numerical form by giving scores. If you used Likert scales (otherwise called as summation scales). After converting, it can be put into either statistical tests on just graphical presentation such as bar charts using central tendency. For a similar scales we used chi square tests.
Similarly, the responses may be classified into two category such as “Agree” or “Disagree” and cross tabulate with the independent variables such as age, gender, caste etc.16th September 2013 at 12:49 pm #1436Jessica AndersonMember
Thank you very much 🙂10th December 2013 at 5:00 pm #1435Azam YaacobParticipant
1 – You are absolutely correct that your data is of non-numerical nature. Assigning number is arbitrary. Such assigning exercise is in fact a mere labeling exercise.
2 – the Likert data type since mere ordinal certainly indicate that summation mathematical exercise is not possible.
3 – Your options :
a) stick to non-parametric testing
b) employ Rasch Model analysis
Try this paper from Julie Pallant on Rasch Model analysis
- You must be logged in to reply to this topic.