Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: complex multilevel analysis


From   Christian Deindl <deindl@soziologie.uzh.ch>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: complex multilevel analysis
Date   Wed, 17 Dec 2008 18:57:22 +0100

I am familiar with books and papers on dyadic data like
Snijders & Kenny 1999 and Kenny, Kashy, & Cook 2006, but their approach is somewhat different and doesn't really suit my kind of data.
I have no knowledge of the methods applied by Peter Hoff.

All my dyadic data analysis are working fine.

I just recently started with triadic data and I've never seen a paper on it. So I try to follow the methods in dealing with cross classified date as described in Hox 2002 and mix it with the "Social Relation Model" from Snijders & Kenny 1999.

I am pretty sure that I'm close to a working solution, but something is still not quite right.

christian



Stas Kolenikov schrieb:
I am personally pretty sure it is not quite right. But unfortunately I
cannot claim I am a huge expert on dyadic and triadic data. I know
there's been a couple of books out about analysis of dyadic data, but
I've been most convinced by the work of Peter Hoff from U Washington
who has thoroughly derived all those weird likelihoods that take into
account all the necessary symmetries in the data. How familiar are you
with those methods?

On Wed, Dec 17, 2008 at 8:05 AM, Christian Deindl
<deindl@soziologie.uzh.ch> wrote:
hi,

I have a question regarding the analysis of triads using multilevel-models.

I'm conducting an international comparison of financial transfers from
parents to their children.
each respondent can have up to four children, and each child is an
observation, with children nested within respondents nested within
households nested within countries.

So I have four levels.
Since transfers are not only affected by the characteristics of children
but also by parents I 'm trying to analyse triads.

the datastructure is as follows:

Triad    Dyad(Parent)    Dyad(Child)    respondent
1        mother1        child 1        1
2        father1        child 1        1
3        mother1        child 2        1
4        father1        child 2        1
1        mother2        child 1        2
2        father2        child 1        2
3        mother2        child 2        2
4        father2        child 2        2
.        .        .
.        .        .


two problems in regard to mulitlevel-analysis arise with this structure.
1) each child is doubled for each parent and for each respondent
2) parents are doubled with children, but unique for each respondent

as far as I know this is a case of cross-classification.

to deal with the nonindependence of children and parents I build a dummy
variable for children (1 for the first appearance, zero for any further
appearance) and for parents.
this dummy-variables are included in the modell as random slopes for the
respondents (see syntax below).

since I couldn't find a clear expamle in the literature I'm not quite
sure if I'm correct.

Can anyone give me some advice?

best regards,

christian



*Syntax:

gen     dumkind=1
replace    dumkind=0 if dyadKINDER==dyadKINDER[_n-1] & persid==persid[_n-1]

gen     dumeltern=1
replace    dumeltern=0 if dyad_e==dyad_e[_n-2] & persid==persid[_n-2]


xtmelogit   transfer_k  /*
*/ || land:  || hhid: || persid:  dumkind dumeltern,  or


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





--
__________________
Christian Deindl
Universität Zürich
Soziologisches Institut
Andreasstr. 15
CH - 8050 Zürich
Tel: 0041/(0)44/635 23 46

http://www.suz.unizh.ch/ages
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index