# Likert Scale Analysis

#### theswits

##### New Member
Hello all,
I am looking to apply a statistical test to some Likert data. My survey responses were scaled 1 to 5 (strongly disagree to strongly agree, with no opinion = 3). I asked 47 people a series of 60 questions each, with all questions formatted so that an answer of "5" supported a general hypothesis, while an answer of "1" would say the hypothesis is incorrrect (in other words, 60 different questions regarding a certain hypothesis).

I tried to treat this Likert data as continuous rather than discrete, so that for each person, I calculated a "mean" answer (for instance, person 1 had an "average" response of 3.63, person 2 was 4.12, etc.). Was this an incorrect step? Can each person's mean be used, for a population of 47 means, in a classical normal distribution analysis?

Is there a brief or paper that I can use apply a statistical analysis to my Likert data?

Thanks,
Elroy

#### Noscito

##### New Member
I don't quite understand what you are doing...

Are you trying to test 1 hypothesis with 60 questions? If so, what is your hypothesis?

#### theswits

##### New Member
Likert Analysis

Hi Noscito,
Yes, all 60 questions refer to one hypothesis.
The hypothesis states that cost certainty for a particular type of business contract is possible. I then canvassed about 50 people in various aspects of that business sector and asked them 60 questions each.

Each question was based on a factor or aspect that previous investigators found affected the cost certainty outcome of the business contract, and furthermore, each question was phrased so that a score of 5 strongly agreed that cost certainty is possible (based on the effect of that question's factor or aspect), and a score of 1 strongly disagreed that cost certainty is possible.

So the outcome I am trying to achieve is "does the business sector believe that cost certainty is possible for this particular type of contract?", and I would like to be able to apply a statistical test to the result.

Does that help?

Thanks, Elroy

#### Noscito

##### New Member
I think I would do as you said before, just calculate the means per respondent, and maybe do a one sample t-test.. although that is not really statiscally valid (likert scales are not normaly distributed)

this a good website on likert scale analysis:

http://www.uni.edu/its/us/document/stats/spss2.html

#### theswits

##### New Member
Thanks

Hi Noscito,
Thanks for the tip and the link to the website.

Kind regards, E.