So I am moving on with my thesis and have a question.
I want to see the effect of precarity on political participation. However, I have devided political participation in 4 groups (formal, activist, online and voting). Formal has 5 items which can be asnwered with yes or no, activist 6, and online and voting just 1.
I do not really know if I can insert this all into one model, probably not. I think my only option is to use multinomial for formal and activist, and binary logistic regression for online and voting.
Am I right or wrong?
You could dichotomize formal and activist such that formals are comprised of those who answered "yes" to at least 3 items and activists to at least four items. Then binary logistic regression could be used for all four groupings.
Does every person in your sample have one designation (i.e., Person A is Formal, Person B is Online, Person C is Activist, etc), or does every person have four different scores (i.e., Person A has a 4 in formal, a 1 in activist, and 0's in online and voting, Person B has a 0 in formal, a 2 in activist, a 0 in online and a 1 in voting)? That will make a big difference in how you analyze your data.
Hey Julia, thanks again for answering one of my questions.
Every person can have different scores on each variable. For example Person A can score low on activism but high on formal political participation for example.
So the way its structured now is:
-formal = 0-5
-activist= 0-6
-Voting= 0 or 1
-online= 0 or 1
And each person can score differently on each variable. So each person has 4 different scores (unless for example they participate in voting and online).
So that way I can see if certain precarity groups maybe score higher on activist and online political participation (based on literature) than on formal for example.
I appreciate all the help
That makes sense. In this structure, I would do a bit of exploratory analysis on the formal and activist measures to see if there are break-points (descriptive distributions work for this process - see Tukey's book for ideas here (it is old, but useful - Tukey, J. (1977), Exploratory Data Analysis.Reading, Mass. : Addison-Wesley Pub. Co.). If there is a solid logic for dichotomizing those two variables, John Cochran's approach would be the way to go. In my opinion :-)
I would have to order this book tho, I am not sure they will have this in our library. And the deadline is 11 december.
Maybe i can find info about this online.
Is there a way to make the 4 binary items into a categorical scale ?
So that you have
1. formal
2.activist
3. new
For example
And then use this for multinomial
Yes, if you can identify each person as belonging to one category, you can use multinomial logistic regression. I would worry, though, about how you are identifying people if they have the same response for more than one category. For example, if you have a person who responded 1 for both voting and online, how do you identify which group they are in?
That was what i was questioning, how do I identify this.
I will just look at them seperatly.
Sounds like you are trying to create a classification variable. Why not employ the four scales into a cluster analysis and see what groups form. You can fit observations into that latent class for which they best fit then create a nominal scale variable based on these groupings. From there a simple multinomial logistic regression approach seem appropriate.
Sounds really complicated to me. I'll see if I can find more info, since im not very familiar with cluster analysis
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Probability