Likert-type questionnaire analysis

#1
I am trying to analyze a questionnaire that has the following types of questions:

"I am satisfied that I can turn to my family for help when something is troubling me." Possible answers range from 0 to 2: 0=Hardly ever; 1=Some of the time; 2=Almost always.

The questionnaires were answered by both males and females both prior to treatment in a drug rehab center and at discharge from the center. Family counseling was a major part of the treatment.

Aside from percentages, what type of statistical analysis can I do to test the hypothesis, for example, that there is no difference between the responses of males compared to females? The mode for nearly all questions answered by both men and women is 2. What tests can I do for pre- and post-treatment paired responses?
 

trinker

ggplot2orBust
#2
The time to ask this question was before you made your survey. Your data is not Likert (see WIKI LINK). Now you have data that's more categorical than numeric and have to make decisions based on that. Look at the wiki link I provided as this gives some good initial direction.

Hopefully Spunky, a TS contributor with a background in psychometrics, will see this post and give you more direction.
 

noetsi

Fortran must die
#4
The time to ask this question was before you made your survey. Your data is not Likert (see WIKI LINK). Now you have data that's more categorical than numeric and have to make decisions based on that. Look at the wiki link I provided as this gives some good initial direction.

Hopefully Spunky, a TS contributor with a background in psychometrics, will see this post and give you more direction.
I am not sure why this would not be likert scale data. It is certainly ordered and has 3 levels. Likert scale data does not (as ordinal data) have to have the same degree of difference between each level. It is not, however, interval like likert data because you have so few levels and you can not reasonably comment on the distance between each level.

A chi square test would tell you if males and females varied on this statistic at either the pre-test or post test stage. If you want to see if they influenced the result (and you had interval data for your dependent variable which you don't) something like a two way ANOVA within subject design (that is two way repeated measures) might work. You might see if any of the non-parametric equivilents of ANOVA can deal with that type of analysis.

Trinker will likely tell you I am on drugs tommrow, wait to see what trinker or Dason says of my suggestions before you consider them :p
 

trinker

ggplot2orBust
#5
Agreed Noetsi (though I think a true Likert is > 5 pts but am going from memory). I guess the point I was trying to make was what you captured. You can't treat this data as interval.
 

Karabiner

TS Contributor
#6
test the hypothesis, for example, that there is no difference between the responses of males compared to females?
Two Chi square tests, one with the pretest data and one with the posttest data.
What tests can I do for pre- and post-treatment paired responses?
That's a sign test.

If you want to test whether changes between pre and post are different
for men vs. women, then things are a little bit more complicated. One
possibility might be to categorize the pre-post combinations (e.g. increase /
unchanged / decrease) and perform a Chi square test with gender.

Kind regards

K.