# Multiple Regression Question

##### New Member
Hello,
I have a question about multiple regression. My project involves non-political group involvement predicting political participation. My hypothesis is that being involved in various social groups (youth organizations, PTA, unions, those who volunteer for various causes, etc.) is an indicator of participating politically.

I have broken my independent variables (non-political group involvement), all originally dichotomous, down into four indexes:
-Formal group involvement (consisting of 18 variables)
-organized group interactions (consisting of 4 variables)
-informal social interactions (consisting of 5 variables)
-giving and volunteering (consisting of 8 variables)

So, I want to use those four indexes as my independent variables to predict political participation. In my very, very limited understanding of statistics, multiple regression can only have one dependent variable (political participation in my case). So that would have to be one dichotomous variable, for example, voting as either one voted or didn't vote. It couldn't be more than one variable correct? Or can I/should I do several regressions (using the same independent variables) with several different dependent variables, one dependent variable at a time? I would like to do an index of political participation where I combine several different variables but I am not sure if that is appropriate.

Sorry for the long posting, but I figure the more info the better and this is my final paper for my degree so I am very nervous about it's completion.
Thank you for any help.
Regards,

#### JohnM

##### TS Contributor
You can certainly do regression with multiple dependent variables - this is referred to as canonical correlation - but from a methodology standpoint, it would be easier to interpret the results if you did several separate multiple regressions, each with one dependent variable (or just a single multiple regression if you used a single index of political participation as the dependent variable).

If you go with one dichotomous dependent variable, this would be logistic regression (or discriminant analysis), and you would need a large sample size to detect significant relationships.

On the topic of indexes - are these indices something you've made up, or are they substantiated by prior published research? Be careful here - you'll need to justify how these indices are comprised...

- i.e., as the index "increases," you'll need to show that this indicates an increase in the measured "activity."
- you'll also need to show that the index is "sufficient" and/or "complete" - in other words does it adequately measure what you think it measures - prior published research would help here

##### New Member
Multiple Regression Question Continued

Ok, so if I understand you correctly, I out of the 30-something independent vairables I have listed, instead of combining them into various indexes, it would be better for me to run a few multiple regressions with one dependent variable at a time (so say that would be three or four different regression models depending on how many independent variables I choose to use?).

So, for simplicity sake, I am then entering into SPSS the various independent variables with the one dependent variable which will make my results easier to interpret. Like I stated earlier I am by no means a statistics wiz and am therefore trying to make the statistics portion of my paper as uncomplicated as it can be so I am not making mistakes with advanced analyses that are out of my league and that I do not understand. Thank you, your response was very helpful.

#### JohnM

##### TS Contributor

Sorry to add to the confusion....

I don't think you should necessarily abandon the idea of your indices (combinations of independent variables) - they could actually make it easier to analyze/interpret the results.

Where it can get overly complicated is if you have multiple dependent variables - multivariate methods can get a bit messy...

I would settle in on a single dependent variable (perhaps an index of political participation, as you mentioned), and run a multiple regression with your 4 indices as independent variables.

I would avoid measuring political participation as a simple dichotomy (voted / didn't vote) because logistic regression often needs large sample sizes to detect significant relationships. Develop a quantitative index for it, just like you're doing for the independent variables.

Just make sure you justify how you "built" your indices - talk it over with your advisor...I'm sure he/she can point you toward references...

Study Design:

Dependent Variable: Political Participation (consisting of xx variables)
Independent Variables:
-Formal group involvement (consisting of 18 variables)
-organized group interactions (consisting of 4 variables)
-informal social interactions (consisting of 5 variables)
-giving and volunteering (consisting of 8 variables)

Good luck,
John

##### New Member
That helped very much and didn't confuse me (any more than I already am), it clarified what I originally wanted to do. Like I said I have forgotten much of my stats and I wasn't sure if using an index of political participation would function well as my dependent variable in multiple regression.

Therefore if I combined say five dichotomous variables into the index (low, medium and high levels for example) of political participation, with a sample of 3,000, my results using the various organizational idicies as the independent variables would yield me a more robust multiple regression result than with five separate logistic regressions one on each variable included in the political participation index?

I wanted to do logistic regression since it is not as intimidating as multiple regression (as silly as that sounds). I just figured the results would be more difficult to interpret with multiple regression. Most texts delve into the function and structure of the equations and not so much the practical aspect and analysis of the SPSS product. Thank you again...this site and your advice are a wonderful resource.
Regards,