Hi Pamela,

You can use Spearman's (rho) correlation coefficient. It is used when variables x and y are both measured on ordinal scale or transformed to ordinal scale. Offcourse you can transform the norminal to ordinal.

Auwal

Hi Pamela,

@Auwal Abdullah says we can transform the norminal to ordinal but we can't.

Nominal to ordinal transform? How can we tranform nominal scale to ordinal scale?

If your variable sum of some likert type variable you can use biserial or point-biserial correlation coefficient.

@Abdullahi Faruk,

Thanks for your contribution. I will read about the biserial correlation coefficient.I am hearing about it for the fist time. Or does it have another name?

Auwal

Biserial and point biserial correlation are their names and i dont know if they have any other name.  You can check this link for something about that.

http://www.statisticshowto.com/point-biserial-correlation/

Thanks a lot for your responses. :)

@Abdullah Using the point biserial correlation means that I would consider the Likert scale scores, which are ordinal data, as interval data. Is that alright?

Moreover, i will use a non-probability sample (a purposive one), is it ok to use the point-biserial correlation for it?

The n would not be lesser than 30.

If you use rpbis you assumed that your data is interval.

If you want to calculate a dichotomous variable and sum of likert scale (e.g. gender and sum of a likert scale data (50 point)) data, you can use rpbis.

But you want to calculate a dichotomous variable and a likert scale item (like 5 point likert item) you can use polyserial correlation coefficient. 

Kline (2016) says that:

"The polyserial correlation is the generalization of rbis that does basically the

same thing for a naturally continuous variable and a theoretically continuous

but-polytomized variable (i.e., categorized into three or more levels). Likert-type response scales for survey or questionnaire items, such as agree, undecided, or disagree, are examples of a polytomized response continuum about the degree of agreement. " page:43

Kline, R. B. (2016). Principle and practice of structural equation modelling (4. Edition). New York, NY: The Guilford Press.

@Abdullah,

Many thanks. I will check the link.

Auwal

Thank you very much! I have one *maybe* last question, if the average, not sum, of the likert scale score per individual would be used, is point biserial still applicable? Or should I use the polyserial correlation?

Of course, you can use point biserial correlation. Because if you calculate sum or mean (average) of score you assumed that your data is interval at least. In this case your variables are a dichotomous and an interval data. So you can can calculate point biserial for this purpose.

I see. That's not a good question haha. I realized just now. Thanks a lot! :)

Your welcome Pamela, Good luck.