st: Regression/Dependent Variable Question

[Sherouk Fetaih <sf773@york.ac.uk>]

Dear Statalist,

I’m having a problem that is more of a statistics/econometrics issue
than a Stata issue. I'm using Stata/IC 13 on my Mac.

I've been consulting various Stata/econometrics textbooks for my
questions, but my dataset is unique, which is why I'm hoping you can
help me. I conducted a qualitative survey that assessed attitudinal,
behavioural and information asymmetry elements, which I am trying to
correlate to voting intentions and the salience of a particular energy
policy to these voting intentions (e.g. since this energy policy will
cease to exist if citizens vote for a particular party). Many of my
attitudinal/behavioural questions used a Likert scale. The information
asymmetry questions tested a respondent's knowledge on a particular
issue relevant to this policy. I also had a few ranking questions,
asking respondents to rank the top 3 policy issues out of 10 issues,
from instance. I used the encode command to transform the string data
to numerical data and recoded the numerical values accordingly based
on my own categories. For instance, for the knowledge-based questions,
I used 1 for a correct answer, 0 for I don't know and -1 for an
incorrect answer. Now, I'm trying to regress the data, based on my
hypotheses - but since I have a combination of ordinal, nominal and
cardinal values, this has proven difficult. Even the
socio-demographics variables are mostly ranges (e.g. age, income,
etc.)

Since my dependent variable, voting intentions is split into a few
questions (depending on the answer given by respondents) I'm not sure
how to go about it. The reason it's split, is that I first asked if
respondents know which party they will be voting for in the next
election.... If yes, I asked how they will cast their ballot. If no,
respondents were sent to the second voting question asking that given
the fact that a particular party will dismantle this energy policy,
could this influence their vote in this upcoming election. The 'yes'
respondents from the first voting question were given a similar
(second) question but I had to use less hypothetical language. So,
although respondents were asked different subsequent questions
depending on the first voting question, they were exactly the same,
just the wording was slightly different depending on their first
answer.

So, my questions are the following:
- Is there a way to lump a few dependent variables together (e.g. all
my voting questions)
- Which model would you suggest to use for correlation purposes?
Ordered or Multinomial Logit or probit models? The problem I'm finding
with this question is that there is variation between variables under
the same hypothesis that I'm testing for... so what if some variables
are ranked, while others are not. Also, I've tried to use these models
but they didn't converge.
- How would you take into account survey weights if I'm using these
models? It wouldn't allow me to do survey weights when I was testing
some of mlogit and ologit models.

Thank you very much for your help in advance. Apologies for the
lengthiness - I just thought I'd present the whole situation. I'm also
new to Stata, but I'm eager to be really proficient using it. Thanks
again.

Best wishes, Sherouk
-- 
Sherouk Fetaih
MSc. Environmental Economics

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/