Categorical variables can absolutely used in a linear regression model. I am not sure how interval data look like, but suggest you directly put those categorical variables in the model without any data transformation.

Of course you can include categorical variables in a linear regression model, just codify first those variables as dummy variables, see Gujarati D. Basic econometrics for further information.

Thanks Dr. Luis

, can i convert categorical data into scale data and code them as 1, 2, 3, 4...and used them directly or i have to convert them as 0, 1 (dummy variable) basis only?

Check this out:

http://socserv.socsci.mcmaster.ca/jfox/Courses/SPIDA/dummy-regression-notes.pdf

You do not convert categorical variables into continous variables to use them in regression models. You use them as categorical (not necessarily being binary!). You must make multiple dummy variables from them, not to put them directly as single variables. But there are many different ways in making dummy variables, each has a different meaning and purpose.

Here is a link that helps you. It is for R but gives you the idea.

http://www.ats.ucla.edu/stat/r/library/contrast_coding.htm

Thanks a lot for all your good suggestions. Can i  use these categorical variable by putting them as  likert scale and then codify as 1, 2, 3....?

HI Vinod:  Your question was answered above, though you may not have noticed. The answer is no, do not code the categorical variables as 1, 2, 3, ... as if on a likert scale. Doing so would give you misleading results. You need to use dummy variables (aka indicator variables). The indicator variables code the input on a 0 or 1 scale for each level of the categorical variable, except for one which is considered the base level.  Most statistical packages will do this automatically if you enter the input is a factor.

As well said above, you shouldn't treat the categorical variables as likert scale. You need to use dummy variables. Here are some additional info

https://www.moresteam.com/whitepapers/download/dummy-variables.pdf

https://stat.ethz.ch/~maathuis/teaching/stat423/handouts/Chapter7.pdf

Categorical variables  are  nominal scale (they don't have an arbitrarily defined zero origin and they don't have a true zero point) therefore they cannot be converted into interval scales.

In order to bring categorical variables into a regression model as independent variables you have to create  k  -  1  vectors of dummy variables whereby  K  is  the  number of categories

As others said you have to create dummy indicator variable. The link that Nedim Yel sent is perfect

https://www.moresteam.com/whitepapers/download/dummy-variables.pdf

Here is a good explanation with a step by step example of exactly WHY you can't directly use categorical variables in regression by using a numberical value for each category

https://www.statisticssolutions.com/dummy-coding-the-how-and-why/

you can use categorical regression option available in SPSS

I have both the dependent variable (dichotomous) and independent variable (multiple subcategories). Can I do a regression analysis with two such dependent and independent variables?

e.g. Dependent Variable: Response to a drug (Yes and No)

Independent Variable: Ethnicity (Caucasian, Asian, Black, and others).

Kindly please suggest what would be the best approach.

N.B: I have created dummy variables for the Independent variable categories

Yes Snehali you can proceed with logistic regression. It is used when your dependent variable is dichotomous and your independent variable is either catagorical or continuous

Back to the original thread, for a multiple linear regression model with a continuous dependent variable, whatever variables you use to create your predicted-y values you should then do a "graphical residual analysis" (which may be researched online) to check the fit for your particular sample. A "cross-validation" may then help you to avoid too closely fitting just to that sample.

A graphical residual analysis might also help you check on heteroscedasticity. But note that the familiar "fan shape" is not necessary for heteroscedasticity. You only need less density in the residuals to have higher variance. (Also see https://www.researchgate.net/publication/354854317_WHEN_WOULD_HETEROSCEDASTICITY_IN_REGRESSION_OCCUR, Pakistan Journal of Statistics.)