Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | James Laurence <jam_lorenz@hotmail.com> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | st: GLLAMM help please: random coef 3-level model with level 2 * level 3 cross level interaction |
Date | Sun, 25 Nov 2012 20:18:27 +0000 |
Dear Statalist, I have a question regarding the use of gllamm for multi-level modelling of ordinal variables. A brief description of my data and aim. I have individuals nested in schools nested in cities i.e. 3-levels. My dependent variable is 4-options from 1 to 4 on levels of satisfaction i.e. it is ordinal. My key independent variables are: individual (level-1) disadvantage; level of school (level-2) disadvantage; and level of city (level-3) disadvantage. My main aim is to test whether the effect of school disadvantage on individual's satisfaction is dependent on/moderated by the level of disadvantage in a city in which the school is "nested" i.e. whether there is an interaction effect between school (level-2) disadvantage AND city (level-3) disadvantage, after controlling for individual-level (level 1) disadvantage. So I want to include a level-2 * level-3 interaction in the model. Crucially, my outcome is ordinal so, as far as I have been able to tell from reading, I have to use gllamm along with the link(ologit) function. So, I believe I need to use a random-coefficient model where the effect of "school disadvantage" (at level 2) is allowed to have different coefficients at (level 3). So, here is the syntax I was thinking of using, where school(lvl2)disad*city(lvl3)disad is the level2 * level3 interaction term. generate cons = 1 eq inter: cons eq slope: school_disad(lvl2) gllamm satis(lvl1) individual_disad(lvl1) school_disad(lvl2) city_disad(lvl3) school(lvl2)disad*city(lvl3)disad, i(school_ID city_ID) adapt link(ologit) nip(10) nrf(1 2) eqs(inter inter slope) Does this look right? In particular: - I've had trouble working out the nrf() values. I used nrf(1 2) as I am specifying one random-effect at level-2 (i.e. a random-intercept for school_lvl2) AND two random-effects at level-2 (i.e. a random-intercept and -coefficient for city_lvl3) - Is the eqs() option specified correctly. I used eqs(inter inter slope) because of random-intercept at level-2 and random-intercept and random-coefficient at level-3 All advice would be hugely appreciated. Thanks so much in advance, Jamie Lorenz * * 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/