It is not generally possible. First have a look at the distribution in both scales. If skewness and kurtosis are both high, it is more risky to use Pearson corr. Study the crosstab (by comparing expected and observed percentages in each box) and try to understand where the correlation comes from. In case of a linear relationship (overrepresentation in the diagonal boxes) Pearson corr is a possible option. Compare Pearson corr with ETA. In case of a linear relationship they are similar, in case of a non-linear one they differ. Consider if you should recode your Likert values, to make the scale more an intervall scale. Do this only if you have a symmetric distribution with a peak in the middle and decreasing frequencies towards both ends. Generally compare different correlation measures. Have a look into the prerequisites for each of them.

Many fields routine use Liker-scored items to create interval-level scales.

Because there have been so many questions about this topic, I have compiled a set of resources on it:

https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items

If the following criteria are met it is quite common to apply correlation testing.

1. `is/are the agreement questions part of of set of validated scales? Should be yes.

2. Code the option 'That does not apply to my case' as missing as this option should not contribute to a mean/any calculation. Use a 5 or 7 point validated scale

3. The scale should be normally distributed, especially if to be used as a dependent variable (in for example regression)

4. The scale should have an acceptable/reliable cronbach's alpha (>= 0.7)

5. The variable to be correlated against this should obey the same criteria.

Note: in a simple bivariate correlation the variables can be interdependent

Thank you Charles, David and Robert for your answers! They will help me a lot!

Your assumptions for Pearson r are independence of observations, interval or ratio scaling, bivariate normality and homscedasticity of residuals. The Pearson correlation shouldn't be used for ordinal data. The Spearman r or Kendall's tau are better choices. Spearman's r will yield the same outcome if there are no tied scores.  

Thank you, Fergan! 

The standard procedures for creating scales from Likert-scored items typically consist of two choices. One is to simply add the items and assess the reliability of the resulting scale with Cronbach's alpha. The other is to do factor analysis to determine the underlying scales for a set of variables.

Either of these procedures will allow you to create an interval scale, and both are described in the RG "question" that I referenced above.

You are better off using a Spearman's Rho than a Pearsons Correlation when using likert scale variables. Some disciplines believe ordinal variables have the same properties as interval (especially likerts) but I don't think the math works for ordinal (discrete) variables. The Spearman's Rho gives correlation findings that one may read like Pearson's. Good luck.

Dear, Yes you can do correlation as Likert scale is called Summeted scale too.

Thanks.

Thank you very much Bhavesh and Cynthia!!

Pearson correlation is used with variables measured on continious level {Interval or Ratio}. In case of likert scale, you need to compute the total score for the scale, after that do correlation with other continous variables.

Good luck

Thank you, Muayyad!!

You can't use any correlation coefficient if you include "That's not apply to may case" in the rating. The 1-5 Likert-style (not Likert) rating can be used with Pearson correlation and generally will not differ too much from other options. Charles Berg points out issues with skew and kurtosis, but these are generally mild with rating scales because it isn't easy to get very extreme values. As long as the ratings are not clustered at either end and are vaguely unimodal and symmetrical it should be OK.

Thank you!! I am becoming  hopefull with research! ☺

I am not sure if the statement "That's not apply to may case" include the likert scale. Normally likert scale is contain 5 optional and then you have to give score for the each optional.

I disagree heartily with the idea of using Pearson's Correlation for ordinal level data, like the likert scale. I know some people use it that way without a second thought, but to my mind, the math doesn't work. So.... I recommend using the Spearman's Rho. If you are using SPSS, it is very easy to do and easy to interpret, and it is a statistic designed specifically to look at "correlations" (from rank orders) between two ordinal variables, or ordinal and scale variables. It has worked quite well for me in my research, when I have used likert scales. Best of luck to you.

Dear Cynthia and Thoriq, thanks a lot!!

@ Cynthia: for me you are to rigid. You cannot use Pearson Corr for Lickert scales not without caution. This means comparing with other measures of association and analyzing the crosstabs. Then you can take a decision either to use or to drop Pearson Corr.

No problem in using it, regarding that the number of respondents is sufficiently large

Thanks, Miguel!!

Obrigada, Miguel!! 

Help please! I want to see whether there is a correlation between two of my questions. Question 1 was concerning risk tolerance, and used a likert scale (strongly agree, agree, neutral, disagree, strongly disagree). Question 2 was concerning the number of horses owned. I want to know whether there is a correlation between the number of horses owned and the answer given for question 1 (e.g. do people have a higher risk tolerance if they own more horses?). 

Hi there,

I have a validated Likert scale and I want to compare this with Likert items (always, often, sometimes and never). Should I use kruskal Wallis followed by Mann Whitney, when I do this I find that the Kruskal Wallis is significant but when I do post hoc Mann Whitney) I find that only 'often and sometimes' and 'often and never' are actually significant.

Or should I just use Spearman's correlation?

There is no problem in using the likert scale but i can recommend you to use the 5 points scaled as 1=strongly disagree, 2=agree; 3=not sure; 4=disagree; and 5=strongly disagree. and in using this make sure that you provide the mean ranges of the scales and their interpretations. these are important in ranking statements basing on the mean obtained and standard deviation and provide a proper understanding of your data.

When it comes to performing the correlation coefficient make sure that you do the proper entry and coding of your data.

Adding something to my previous answer, because you have 6 Likert scales, this would not be good for detering the mean ranges. i recommend you to use 5 points or 7 points for better analysis.

I suggest using a Spearman's Rho.

As longer as ordinal variables are concerned, it is better to use spearman correlation

Yeah it's feasible doing a correlation. I agree with Charles

Dear all, I needed to use it in my PhD, which I have finished some time ago. Anyway, these question is always coming around again. Then, thanks a lot for all the answers and comments!

Wishing you all a warm Sunday, Regina

You can use both Spearman and Person and compare the results. I had the same case and compared both of them. The results do not differ that much.