No I do not think so, you need to find a program that handle multinominal analysis. Why would you structure your data that day?

Essentially you are asking about when (if ever) it's ok to treat ordinal data as continuous. There is a lot of discussion online about this, and you can find references in  many of the threads that come up via a Google search. For example, this blog post cites a few relevant papers: http://www.theanalysisfactor.com/can-likert-scale-data-ever-be-continuous/

I think you can use Likert scale data in multiple regression analysis..

Dear Gobinda:

Yes, you can, but only the individual's total score since it can be treated as an interval scale.

Dear Gobinda,

I agree with Chuck. The sum of likert items is a likert scale score and the measurement level of that sum (if there are enough items) is close to interval. Besides, when doing regression check the residuals for the various assumptions (normality, homogenity etc).

greetings, Pierre

You can create binary variables based on Likert scales for right hand side controls & then interpret the coefficients  appropriately.  If you are using the Likert-made variable as the dependent variable, you can use an ordered probit.

Assume there are 5 factors on which you are going to involve these 5 variables under multiple regression. Under each factor you may have various statements which need to be answered by Likert scale responses. In such case, add all the values of all the variables under each factor so that these values can be considered for multiple regression analysis. This is how it works. All the best.

Yes you can  but all the variables should be converted to numerical variables then analysis as usual could be carried out

I think some answers have missed the question... 

Ordinal regression is designed specifically to handle models with ordinal data as the dependent variable.  For an example, see the R package ordinal.  The package requires that the dependent variable be an ordered factor variable, that is, that it be truly ordinal.

https://cran.r-project.org/web/packages/ordinal/ordinal.pdf

Interestingly, I'm not sure how best to handle the ordinal independent variables.

As to whether you should treat ordinal data as numeric data, and use parametric statistics, I would say if you have something like 1 -5 Likert items, there are several reasons why parametric statistics are probably not appropriate.  The data aren't continuous, aren't necessarily equally spaced, and are bound at the ends.

Ordinal regression is good solution.

Some non-parametric analyses are appropriate as well.  For example using multiple Spearman correlations may give you enough information.

And an ordinal regression can be used to estimate the probability that one "likes" or "dislikes" whatever the Likert scale represents, so there is the ability to interpret the results in a meaningful way, too.

In fact your answers have helped me too because i was also wondering which regression model is best fit for a Likert scale data. However I will like to know if there is the need to run a regression analysis if Spearman correlation is already used to run a likert scale data?

You can use  regression analysis to run a Likert scaled data though you should code the data first. The result so obtained will determine the type of regression to be used whether linear or more advanced regression like the logistic or other form of transformation. Good luck

Pls am using excel to analyse a likert scale data of 5 point (Strongly agree to Strongly disagree) from a questionnaire, I code the response as 1-5 in excel, but after analysing the data, I dont know how to go about it again

The the young lady, Chinelo said it depends on the purpose of your research. I see the an application of non-parametric tests, I see the total scores of ith person as a regressed amenable to regression and I see probit, Tobit and logit. But it all depends on the research objectives 

In order to perform any analysis, you should test the assumptions associated with this analysis. Please have a look on the multiple linear regression assumptions that must be met.

https://statistics.laerd.com/spss-tutorials/multiple-regression-using-spss-statistics.php

Generally, there is a big gap here as some researchers recommending the use of non-parametric tests for the ordinal data, while others stated that we can use it, search the google scholar and see the big differences in their opinions and researches. Have a look on the attached editorial article.

Best of Luck

Ahmed Meri

I have this kind of problem, I learnt my statistical class, In Regression Analysis, the dependent and independent variables should be continuous variables, When I read through scholarly articles they have used liker scale as continuous and Multiple regression and Pearson Correlation were used. I wanted to follow them. is it ok? guys,

some scholars pointed that Likert scale should be considered as ordinal, and they are arguing Disagree and Strongly Disagree can be ranged, yes, they can be ranked Disagree is less degree when compare to strongly disagree. but you have to remember that likert Scale contains 5 scales, like Strongly Agree, Agree, Neutral, Disagree and Strongly disagree. you cannot say that Agree is greater degree than disagree, you cannot rank. So, those guys arguing likert

@Abdinasir,

yes, you can use regression analysis on total scores since it is interva. It is the manner in which one uses likert scale that matter and not the scale itself

There is sense in all that were said above. For each respondent and statements measuring the variables a simple average can be taken as data point which can be regressed

Awesome responses! Very insightful and helpful. Thank you all

Glad we could help someone

Imran, your explanation is difficult to follow. Try again ! how does one regress the on the dependent variable? I don't follow at all!

i have encountered a similar problem. my dependent ordinal data does not meet all the assumptions for ordinal regression. so i have opted for multinomial logistic regression (not multiple linear regression as suggested by some comments).

You are correct if the likert scale is the bases of ur analysis, however, when the total score obtained by the ith student is of interest, then multiple regression analysis can be used as an estimator. And nonspherical errors should handled in both cases

Likert scale yields nominal (nonparametric) data. And regression analysis is not applicable to nonparametric variables since they don't have normal distribution.

Are we saying nominal values can't be aggregated by respondent and variable to imitate values having parametric properties, especially random distribution? We don't have to be fixated to the knowledge of the past.

Multiple linear regression is based on the assumptions of OLS and Likert scale are usually nominal and ordinal scale, which violates the assumptions of OLS. Therefore, multiple linear regression analysis is not suitable for Likert scale. It will be more appropriate for researchers to employ other kinds of regression analysis, depending on the data. Some other estimation techniques that can be considered include: Logit, Probit, categorical regression, ordinal regression, among others. It is important to note that these other kinds of regression analysis or estimation techniques equally have their assumptions. I hope it helps

Beneth Eze and Bulent Ozer are wrong. aggregation of of ith score is interval in scale. The main is not to analyze each row of a likert scale. They are given and total score is taken. Please read your books to avoid a mis-guide!1

For the dependent variable, you must employ a logit or probit model. If you have choices where there is a ranking in the dependent variable, then an ordered probit would be best; if no ranking occurs, then a multinomisl probit is used. I really like the old Maddalla (spelling?) book on these models. Wooldridge‘s textbook (grad level) is also good.

For the independent variables, you must create a set of dummy variables for each Likert scale variable you have (omitting one level as your base category).

Hope this helps!

Mary Ellen Benedict

As an econometrician, total score on a test is what we are talking about here and that is different than the general guidelines in these texts.

I thought he said Likert scale not test score. If test score, the researcher must consider that the score is bounded by 0 & the highest score. If Likert, I stand by my earlier statement. And, the two books I mentioned are quite detailed & include examples.

Hi ,

In my case my independent variables as well as dependent variable are in Likert scale. It is a 5 point scale already coded. So ideally all my variables have values 1,2,3,4,5 and they are factors. In this case can I use multi linear regression please ? If not what should be my approach ?

No I do not think so, you need to find a program that handle multinominal analysis. Why would you structure your data that day?

Stata is very good with logits/probits & the website has great “how to” articles & add one if you need them. Wooldrige’s econometric text is helpful too.

PS: if you have only a couple of 1s, I suggest you combine with the 2s to get the maximum likelihood estimation to work. Also, the lowest score when you run the lobit/probit should be 0, so redo the dependent arguable from 0 to 4 (or 0 to 3 if you combine 1s & 2s).

Good luck!

You can use factor analysis through SPSS to generate factor scores thereafter run multiple regression analysis as usual.