For regression, assumption of continuous data is for the dependent variable; no problem for binary variables like gender. And you can choose whatever coding you like for the two variables, it will change the meaning of the numerical results, eventually making interpretation not straightforward, but not the model by itself.

Hi Eshrat,

You are mixing up two different approaches here. Correlation is not synonymous with regression in what you are trying to accomplish. 

In regression you can certainly regress the outcome on a binary variable such as gender (regardless if it is coded 0,1 or 1,2). You could also use a t-test in which you compare the difference in means between the two gender types. These are indeed similar approaches.

Where things differ here is that in multiple regression you will likely add additional covariates to the model, typically as control variables. T-tests do not allow for covariates.

I hope this helps

Ariel

For control variables, most people do this using dummy variables that are coded 1/0. Thus the coefficient tells you whether the amount added by going from 0 to 1 is significantly different from 0. So a non-significant result means that nothing is being added by moving from 0 to 1.

As for moderation, the fact that gender is nominal makes sense because you are dividing the total sample into two subgroups and then asking whether the coefficients are equivalent between them.

One of the assumptions of multiple linear regression is that your Dependent Variable (DV) is measured in a continuous scale (either interval or ratio), and not the independent variable(s) as you mentioned. When treated as a dichotomous variable (variable having only two possible values), gender can readily be entered as a predictor variable. The problem occurs when you have nominal variables with more than two categories. Then you may have to dummy-code those categorical variables to be able to enter into regression analysis.

On a side note, if gender is on the right hand side (i.e., a DV) of your regression model, then you should not use multiple regression, rather it would call for using logistic regression. If your DV has more than two categories, then you have to use Multinomial logistic. Thus, it is the characteristics of DV that tells you what you can do and what you cannot. On another point, coding gender as 0,1 or 1,2 does not change the interpretation of the regression coefficients but has an effect on the contstant in the regression output (that is altogether another discussion). Hope it helps.

I think Khandoker Mahmudur Rahman is correct we cannot conduct multiple regression analysis but you can go for logistic regression.

Sangita, I did not say logistic regression in this case, I was referring to what happens when DV is categorical. On the other hand, the question talks about gender as an IV. So multiple regression can be done. Just for clarification. Thank you.

Sorry, but in logistic or logit regression the dependent variable is categorical.

I would use the binary logistic regression. Maybe the analysis of covariance (ANCOVA or MANCOVA) would solve the problem, too.

Sangita,

Any dichotomous variable is by definition also categorical, so logistic regression is by far the most common solution for a two-category dependent variable.

David Sir,

              Can we use multiple regression analysis for this question?

Multiple logistic regression is the preferred method to use with a 2-category dependent variable.

If you code gender as "1" and "2 " and other variables are categorical then use Elaboration of Variables. If Gender is coded as "1" and "0" otherwise then you may bring it explicitly into your regression model as a dummy independent variable