# ??multi-level/hierarchical or cluster analysis??

#### barbi

##### New Member
Hi! I hope there's somebody who can help me with my thesis analysis (hierarchical logistic regression). i know the basic concepts and am actually confused if it is truly hierarchical/multi-level or just cluster analysis.

Here is the situation:
I have for level 1, individual respondents with variables i want to correlate with this level (socio-demographics, etc.)

In level 2, would be the province level (where individuals are grouped into geographical location). I don't have variables for this level but i would like to take into account the effect of the place where they live~ in this case, the province.

Is this a legit multi-level/hierarchical or cluster analysis? What software can I use and how do i go about my data? How would my equation look like if there isn't any 2nd level variables?

I hope somebody can help me. Thanks!

#### JohnM

##### TS Contributor
Barbi,

Welcome to the forum.

-what is the objective of your thesis, i.e., what research question(s) do you want to answer; what research problem(s) do you want to solve?
-what are the independent variables, and how will they be measured?
-what are the dependent variables, and how will they be measured?

I know this probably sounds tedious, but having written a thesis for my MS, it is absolutely necessary to develop answers to these questions I've listed before starting a discussion of appropriate analysis methods.

Very often, grad students want to do the opposite - they want to run a particular analysis they find interesting or perceive to be sophisticated, and they're not even sure exactly what they're trying to achieve, or they are having difficulty articulating it...

Also - do you have a thesis advisor? Have you discussed this with him/her?

John

#### barbi

##### New Member
The title of my thesis is "Correlates of Modern Contraceptive Use Among Married Filipino Women". I'm using secondary data from a baseline survey. The variables are: age of the woman, number of children, proximity to health facility, religion, woman's influence in household decision, past contraceptive experience, exposure to media. I don't know if i've mentioned everything here already.

Yes, i do have a stat adviser but, she's also trying to figure out how to do multi-level analysis, because she has not applied this kind of analysis yet. What I actually plan to do is divide my data into two parts, 60% of it, roughly 3000+ will be for modelling and the 40% will be for validation of the model.

I know how to do logistic regression because that is taught in grad school. My thesis adviser, wants me to do a new kind of analysis so we will both learn a new method.

My objective is to determine the relationship of each independent variable to use of modern contraception. I hope you'll be able to help me on this

Thanks!

#### JohnM

##### TS Contributor
Unfortunately I'm not familiar with hierarchical logistic regression, nor do I have much experience with cluster analysis....

I'm fine with everything in your post except the third paragraph - just my opinion, but theses and dissertations should not be used to "try out new methods," unless the primary objective of the study is to explicitly assess and evaluate the new method.

They should be used to extend the knowledge base of a field of study....it sounds like you have a very interesting topic and obviously you've given it a great deal of thought - but I wouldn't attempt to apply a "new" method in the hopes that it will answer your research questions.

#### barbi

##### New Member
JohnM said:
Unfortunately I'm not familiar with hierarchical logistic regression, nor do I have much experience with cluster analysis....

I'm fine with everything in your post except the third paragraph - just my opinion, but theses and dissertations should not be used to "try out new methods," unless the primary objective of the study is to explicitly assess and evaluate the new method.

They should be used to extend the knowledge base of a field of study....it sounds like you have a very interesting topic and obviously you've given it a great deal of thought - but I wouldn't attempt to apply a "new" method in the hopes that it will answer your research questions.