? about using categorical variables in correlations

#1
Can I use categorical and continuous variables as control variables in a partial correlation which is looking for the relationship between categorical and continuous variables?
Can I use categorical variables as control variables in a partial correlation which is looking for the relationship between categorical and continuous variables?
Can I use categorical variables as control variables in a partial correlation?

I can use Spearman to see the relationship between categorical and continuous variables, right?

:)
Many thanks.
 

Karabiner

TS Contributor
#2
For Spearman, variables have to be measured on an ordinal or an interval scale.

If you want to predict an interval scaled variable, using categorical and interval scaled predictors at the same time, then multiple linear regression or ANCOVA can be used.

With kind regards

Karabiner
 

Karabiner

TS Contributor
#4
This doesn‘t make sense to me. Where is the “nonparametrical“ part?
He could just have the 2 variables transformed into ranks and used
the partial correlation procedure on the ranked variables. Whether it
is justified to treat a binary variable like an ordinal scale variable, I do
not know.

You wrote about categorical variables, did you mean dichotomous or
binary variables throughout?

With kind regards

Karabiner


By the way, the formulation of the result is annoying. “There was
no relationship“ is silly. Obviously, there was a relationship, the
coefficient is different from zero. It‘s just that it is not statistically
significant (which is not surprising given the low power of the
test , due to very small sample size).
 
#5
This doesn‘t make sense to me. Where is the “nonparametrical“ part?
He could just have the 2 variables transformed into ranks and used
the partial correlation procedure on the ranked variables. Whether it
is justified to treat a binary variable like an ordinal scale variable, I do
not know.

You wrote about categorical variables, did you mean dichotomous or
binary variables throughout?

With kind regards

Karabiner


By the way, the formulation of the result is annoying. “There was
no relationship“ is silly. Obviously, there was a relationship, the
coefficient is different from zero. It‘s just that it is not statistically
significant (which is not surprising given the low power of the
test , due to very small sample size).
I thought they all meant the same thing but if I had to pick, I'd pick binary. My categorical variable is coded as : 0/1 or No/Yes.
And I also thought nonparametric meant categorical (in a sense at least). I guess it actually refers to normality..

I have one more Q for you though. I'm now trying a multiple regression to find the relationship between 1 continuous IV and 1 continuous DV while controlling for 1 categorical variable and 4 continuous variables. Even if the model is not significant I can use the p and r values of the relationship I'm looking for? The model will have 6 predictors. The resulting p and r values for each predictor is gathered controlling for the rest of the variables--can I say that?
And I got SPSS to report to me Zero-Order, Partial, and Part correlations. Which of these would I need to report? I guess I'd need to look deeper into differences of those types..

I changed route of this question. I alredy asked this question in a more worded way elsewhere, so I'll link these two if I get a response :)
http://www.talkstats.com/threads/easy-regression-qs.73147/

Thanks for the help.
 

noetsi

Fortran must die
#8
One of my best stats classes used a psychology test. For a number of years I worked with factor analysis models (one of my big projects at work was analyzing satisfaction surveys which use Likert scale data). I spent a lot of time reading about correlations which of course are central to factor analysis. There is significant disagreement if you can use pearson's r or even spearman with likert data. I think the general view is that polychoric correlations are better although when I ran them and compared them to spearman's I did not think the differences were that great.

A practical issue with using polychoric correlations is that (at least last time I did this) the key software like SAS or SPSS did not do them. SAS has a macro that will run them (it is what I used). SPSS used R I think to do so, so R probably will calculate them.