How can I analyse this likert scale? Any help would be greatly appreciated

#1
Hi there,

I am creating a 20 question questionnaire for my dissertation at university. It is a likert scale questionnaire and I will be using a sample (just one sample) of 100 women (young women, aged around 20-28). I will be asking them whether they like a certain personality trait in a male.

The only tests I have found to analyse the likert scale are comparing two groups, but I am not comparing, I am simply using one group of people and seeing whether they like a personality type, by the options they choose. I am so confused. I have no idea how I can analyse it or find any sort of p-value.

Any help would be greatly, greatly appreciated. Thank you :)
 

noetsi

Fortran must die
#2
You can use likert scale [which is usually seen as an ordinal scale] with most statistics. You need to decide what your research question is [or since you probably already have you need to tell us what it is] for anyone here to suggest a method. For example if you wanted to see this variable's impact on some other predictor variable you could run a regression [in which case the likert scale variable would probably be coded into a series of dummies].
 
#3
Thank you for your reply. My question is does a population of young, heterosexual females find the narcisstic personality of the male attractive?
 

noetsi

Fortran must die
#4
It seems to me that the answer to that, assuming you ask this in your questions, would be purely descriptive. Count how many times they say yes they are attractive [as defined by your questions] as compared to no relative to different responses to your questions [or how often the traits you associate with this are viewed as attractive].

Another way, if you asked this question, would to have attractive/not attractive be the dependent variable and run your questions [which again I assume are measuring some type of trait or personality] be the predictor variable. Then see how the odds ratio of attractive changes with answers to your predictors [your IV questions].

In many cases researchers ask individual questions and then collapse them into larger questions [commonly this is done because it assumes they measure some latent trait]. You can collapse your questions many ways, commonly you simply add different questions although there are many other approaches. Factor analysis to do so is one option. You should know that if you add many likert questions together for some combined scale than these are commonly assumed to be interval in nature.
 
#5
Thanks so much for your help. Can you tell me what you mean by the predictor variable and how i'd find out the ratio? I am really confused and stats is all quite new to me! So to analyse this one sample, I could use factor analysis or multiple regression? If you could explain this simply, I would appreciate it! I'm just trying to get the hang of how I can analyse this.
 

noetsi

Fortran must die
#6
In regression, and many methods, it is assumed that some variables are predicting/influencing other variables. These are the variables on the right side of the regression equation. There are many names for these, but I think the most prefered these days are predictor variables (independent variables is what they were called when I first learned statistics and many still use this). They influence what is on the left side of the equation the response variable (many call this the dependent variable).

I think by ratio you mean odds ratio. This is a calculation in logistic regression that shows the odds of being in one state of the response variable for a change in a predictor variable compared to the odds of being in the other state. So if you have an odds ratio of 4 and the state of the response variable you are maximising is attractive, that means for a one unit increase in the predictor variable (controlling for all others) it is four times more likely that a person would consider someone attractive than not attractive.

This is a form of regression called logistic regression (where the response variable has only two levels). You can also use it if the response variable has a few levels not just two but they are ordered (ordinal) - which is common for a likert scale variables. I am not sure that is the best method for someone brand new to statistics, it has taken me a while to learn it and I am still learning it.

A much simpler alternative might be doing something like a chi square test - although the results will not usually be as interesting as regression and you can only test two variables at a time not many predictors as in regression. Factor analysis is entirely different. It is a way of taking large data sets, large sets of variables measured by questions, and reducing them to a smaller set of factors. I was suggesting that you might look for common factors behind the questions you are using to predict the response variable with this. It does not show how one variable influences another variable, it just looks for common relationships behind what you measure.

If you are new to statistics you should get some introductory material. While not supper complex these methods are not simple in my opinion.
 

noetsi

Fortran must die
#7
In regression, and many methods, it is assumed that some variables are predicting/influencing other variables. These are the variables on the right side of the regression equation. There are many names for these, but I think the most prefered these days are predictor variables (independent variables is what they were called when I first learned statistics and many still use this). They influence what is on the left side of the equation the response variable (many call this the dependent variable).

I think by ratio you mean odds ratio. This is a calculation in logistic regression that shows the odds of being in one state of the response variable for a change in a predictor variable compared to the odds of being in the other state. So if you have an odds ratio of 4 and the state of the response variable you are maximising is attractive, that means for a one unit increase in the predictor variable (controlling for all others) it is four times more likely that a person would consider someone attractive than not attractive.

This is a form of regression called logistic regression (where the response variable has only two levels). You can also use it if the response variable has a few levels not just two but they are ordered (ordinal) - which is common for a likert scale variables. I am not sure that is the best method for someone brand new to statistics, it has taken me a while to learn it and I am still learning it.

A much simpler alternative might be doing something like a chi square test - although the results will not usually be as interesting as regression and you can only test two variables at a time not many predictors as in regression. Factor analysis is entirely different. It is a way of taking large data sets, large sets of variables measured by questions, and reducing them to a smaller set of factors. I was suggesting that you might look for common factors behind the questions you are using to predict the response variable with this. It does not show how one variable influences another variable, it just looks for common relationships behind what you measure.

If you are new to statistics you should get some introductory material. While not supper complex these methods are not simple in my opinion.
 

noetsi

Fortran must die
#9
I am glad to help. Statistics is awesome, but it takes a while to learn it. The goal is to eventually be able to ask a question complicated enough to stump someone like Dason :p
 
#10
Thanks again for everything! I have just red through it all again and I've gone confused! I just don't know what test to use to analyse it...I have googled what you said (the tests) and they don't seem to fit with exactly I want (or I can't make sense of them!), all I really want to do is figure out what test I can use to analyse one sample on whether a trait is attractive to them or not and it's taken we weeks to figure out which test to use!