I am analyzing a 17 items Likert Scale in SPSS. I have numerically coded the items and wish to compare the medians of two populations (Generation Z (group 1) and Generation Y (group 2) as it is a Likert Scale (the scale is 0-6), I am considering it to be of ordinal nature as the distance between the two responses are not equal. I wish to use Mann-Whitney U test. How can I compare the medians of these two populations in SPSS?
- Should I calculate the median for each respondent in the group 1 and group 2 by using the compute variable option in SPSS (Function (Median (Item1,Item2,Item3)) ?
- As this compute variable option will create a new target variable (let say "median"), should I use this new computed variable "median" in the Mann Whitney U Test by putting Variable "Generation" (Gen Z and Gen Y groups) as the independent variable?
- I also do not know how to check whether the likert data is normally distributed or not? The reviews that I reading suggest that we cannot check the normality of the Likert Scale? Suppose, I treat it as continuous data (as there is no escape category e.g neutral/do not know/ in between the scale 0-6)... then how can I check the normality of a likert scale data assuming it to be continuous?
Kind regards
Amit
An important question, Amit.
There are several issues to consider.
First, understand that a Likert scale is not a single 1-5, or 1-7, or 0-6 response to a question. It is the sum, or mean, of several such items - usually six or more. Traditionally, a Likert item was a 5-point agree-disagree response. We construct the scale by calculating the mean of those items that theory tells us are manifestations of a construct.
Think about how you answered the questions when you first read them. You could put a mark on the scale quickly on some, but you others made you think, "maybe 4, maybe 5: can't I check somewhere in between?" The logic of a Likert scale is that a overstatements on some items will be compensated by understatements on other items so that, when they're averaged, they give you a score that is closer to the "true" value.
For example, the following four questions from the World Values Survey (http://www.worldvaluessurvey.org) are highly correlated, and can be used to infer the extent that a person is Altruistic and concerned for the community.
- Looking after the environment is important to this person; to care for nature and save life resources (V78)
- Help Others: Mean of two variables: It is important to this person to do something for the good of society (V74) and It is important to help people living nearby; to care for their needs (V74B)
- Tradition is important to this person; to follow the customs handed down by one's religion or family (V79)
- It is important to this person to always behave properly; to avoid doing anything people would say is wrong (V77)
Another three questions are not correlated with the above four, but are highly correlated with each other. We could call their average "hedonism"
- It is important to this person to be rich; to have a lot of money and expensive things (V71)
- Adventure and taking risks are important to this person; to have an exciting life (V76)
- Being very successful is important to this person; to have people recognize one's achievements (V75)
To get a measure of one person's score for "altruism" we simply calculate the mean of those first four items from that one person. And we calculate her score for "hedonism" with the mean of her answers to the last three items.
If you wanted to compare a single Likert item between two groups then yes, you'd use a Mann Whitney U Test. But you should not ever do this anyway. You would not find the median of several items for the group or for any individual respondent.
I'm guessing that your 17 Likert items are designed to capture 3 or 4 different theoretical constructs. So, your strategy would be to construct a composite measure in SPSS:
- Construct_1 = mean(Var01, Var02, Var03)
- Construct_2 = mean(VAr04, Var05, Var06)
- and so on.
Then you can use a regular independent-samples t-test to see if the mean for each construct is the same for both groups.
Use a non-parametric test if you like, but it almost certainly will give you the same interpretation as the parametric test.
If your data set or the data dictionary does not give you any clear guidance on what constructs are supposed to be captured with your 17 Likert scales, then you can run Exploratory Factor Analysis, with rotation, on the data to discover what variables do share the same information. You can then use these as your constructs.
Normality
Don't worry about it.
T-tests and non-parametric tests are extremely robust against violations of any normality assumption.
Other analyses - say using your Likert scale as a predictor variable in a regression model - do not in themselves assume that your data are Normally Distributed. They assume that the independent variables are IID (independently and identically distributed). Such models do assume that the errors in any regression model are normally distributed. But errors are nothing at all the same as the predictors: (1) errors are a part of the dependent variable, and (2) a dependent variable can be skewed and still have normal errors. Even with violations of the IID assumption, you still get unbiased estimates. The standard errors of the estimates may be unreliable though.
Hello Amit. Hume F. Winzar has already given you a very complete response. I'll add one small point. Contrary to what many people believe, the (Wilcoxon-)Mann-Whitney test does not test a null hypothesis stating that the population medians are equal. See this Stata Journal article by Ronán Michael Conroy for further info and a nice example.
https://www.stata-journal.com/article.html?article=st0253
HTH.
Hi,
I will recommend these following materials:
1. Principles of Biostatistics (Authors- Marcello Pagano and Kimberlee Gauvreau)
2. Medical Statistics at a Glance (Authors- Aviave Petrie and Caroline Sabin)
Best,
Oloyede
Dear Hume F. Winzar , Bruce Weaver and Oyekola Oluyimika Oloyede ,
Thank you so much.
I am working on my thesis by seeking the invaluable advise from you.
Thank you for guiding me. Your selfless efforts to share the knowledge about the research area help so many of us who are new to this research world.
Warm regards,
Amit Basra
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS