Likert scale

#1
Hello all,
I seriously need help on my research analysis. I am conducting a qualitative research with structured interviews i.e predetermined questions ( almost like questionnaires) with 35 respondents. I have about 13 questions and some of the questions are answered using a likert scale of strongly agree agree not important, strongly disagree , disagree and assigned to numeric value of +2 +1 0 -1 -2....
what is the best possible way to analyse the likert scale bit of the interview questions. I have about 6 questions using the likert scale and the others are open ended discussion question which i can analysed comfortably.

Thanks.
 
#2
As a minor aside most qualitative researchers would not describe your design as "qualitative" which always involves the analysis (coding) of text not likert scale responses. I don't take such orthodoxy all that serious, but a lot of quantitative and qualitative professors (and reviewers) will.

To answer your question you have to know what your research question is, what your assumptions about the data are and how many individuals you interviewed. You have to decide, for example, if the likert scale data is ordinal or interval in nature which in turn will determine (along with what you are trying to answer) what method you use.
 
#3
Thanks for quick response. Individuals interviewed is 35 in numbers and the research questions is based on perception of customer satisfaction on 2 main attractions in a destination. some of the interview questions are not rated on likert scale which i can easily analysed. but the likert part of it is where the problem lies.
 
#4
To follow on noetsi's comment, I would say you have a "mixed methods" design--very trendy :)

You have a reasonably small sample and a small number of questions, so you might wish to "work around" the statistics, at least at first. Have you graphed your responses to look for visual trends? It will help you to get a feel for your data.

After looking at the graphs, I would start to look at some cross-tabs. What you are looking for now are relationships between responses (e.g. Those who agree or strongly agree with Likert Q7, tend to say X in the open reponse Q9, etc.)

Are you analyzing the open ended responses by hand, or with software? I use QDA Miner from Provalis software, and it handles this sort of mixed method design beautifully. Your sample is small enough to do by hand, but good mixed methods software would be very helpful.

John
 
#5
Yeah it is trendy (although cynically I suggest mixed methods will get you in trouble with both quantitative and qualitative researchers - those who take this type of issue seriously that is). As SmoothJohn I think is suggesting you really have too few cases (35) to have adequate power for quantitative approaches (qualitative rules are different, there you have had to reach theoretical saturation which depends on what you are finding and the amount of comments from each respondent. Likert type data does not normally meet this criteria).

I don't know if your questions can be addressed by t-tests (assuming you feel your likert scale data is interval like which is required to do this method and that your power is adequate). If it is you might consider that. Or you might do simply descriptives, show how people answered specific question possibly controlling for factors such as gender, organizational position etc. Crosstabs, as noted above, are one way to address that.
 
#6
As SmoothJohn I think is suggesting you really have too few cases (35) to have adequate power for quantitative approaches (qualitative rules are different, there you have had to reach theoretical saturation which depends on what you are finding and the amount of comments from each respondent. Likert type data does not normally meet this criteria).
Actually, I was tapdancing around it. Thank you for explicitly questioning the sample size.
 

SiBorg

New Member
#7
35 may well be enough - if the differences in response are great enough you can get stat. sig results with that many responses so it's always worth a try (and doesn't take that long).
 
#8
I used Gpower to look at a one sample t test (although I dont know which type of t test would be used here). For a one tailed test, alpha of .05 and a power of .8 (the minimum you should use) your effect size has to be .44 or greater to use 35 cases. If you think that is a reasonable number you can use it.