Re: st: Nested design problem

[Steven Raemaekers <s.raemaekers@sig.eu>]

Hello David,

Thanks for your response! You can view the connections between people as a directed graph, with arbitrary arrows between people when they have a connection. 
There is not a value for each possible pair of people, a lot of connections are missing and are then considered irrelevant, but I'm only interested in the subjects which have connections to other people so this should not be a problem.

Maybe it is indeed an analysis problem because I have already collected the data and I already know my research question. Then the question becomes more how to choose a statistical model that does most justice to reality. 

Regards,

Steven

On 16 nov. 2012, at 13:13, David Hoaglin wrote:

> Hi, Steven.
> 
> The problem you describe is interesting.  I don't recall seeing
> anything like it in the statistical literature.  I view it as an
> analysis problem, rather than a design problem (you have already
> collected the data).  Some of the work in sociology on networks may be
> relevant.
> 
> I think it would be more appropriate to consider closeness as the
> dependent variable.
> 
> Do you have a value of closeness for each of the possible pairs of
> people?  If not, what is the structure of the subset of people for
> whom you do have closeness?
> 
> David Hoaglin
> 
> On Thu, Nov 15, 2012 at 4:44 PM, Steven Raemaekers <s.raemaekers@sig.eu> wrote:
>> Hello,
>> 
>> I do not know where else to post this problem so I hope somebody can help me here. I have a statistical design issue which is as follows.
>> As an example, let's say I have a table with information on people in the following format:
>> 
>> Id                      IQ              Introvertness           …
>> person1         120             80                              …
>> person2         110             70                              …
>> person3         130             40                              …
>> 
>> This table contains a number of properties of certain persons, each person is a unique entry in this table. I also have a table which contains relationships between those persons:
>> 
>> Id1             Id2             Closeness
>> person1 person2 80
>> person1 person3 70
>> 
>> "Closeness" is a property of the relationship between person1 and person2. The numbers do not make sense but are just for illustration.
>> 
>> Now I want to test whether people that are in relationships "closer" to each other are more intelligent. I also possibly want to take into account other properties of people.
>> In regression terms: I want to regress the dependent variable IQ on the independent variable closeness and introvertness. How can I do this?
>> 
>> The problem is, persons can appear multiple time in the list so applying normal linear regression or correlation on this table produces incorrect results.
>> Additionally, there are two sides to the relationship, so there is a connection between the IQ of person1, the closeness of their relationship and the IQ of person2.
>> I believe the thing I need is called a nested hierarchical model/mixed model, but I have no idea how to design this.
>> 
>> My questions are therefore the following:
>> 1) What statistical test do I need to test my hypothesis on this data?
>> 2) What assumptions does this test make and how can I check these assumptions?
>> 3) How should I format my data?
>> 4) What are the commands in Stata to execute this test?
>> 
>> Thanks very much!
>> 
>> Regards,
>> 
>> Steven Raemaekers
>> PhD student
>> Software Improvement Group/TU Delft
> 
> *
> *   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/



*
*   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/