cross country analysis of different years?

[Dan_yu_W]

Hello.

I was asked from my employer wether it is possible to make a regression from a cross country survey dataset collected over several years (see example):

Example:

Happiness	Income	Country	Year
8	100.000	Sweden	2001
7	150.000	Sweden	2003
4	90.000	Poland	2001
4	80.000	Poland	2003
5	120.000	GB	2001
6	120.000	GB	2001
5	110.000	GB	2003

The model is: Happiness = Income * X

But I want to take the country of the survey and the year the survey was performed into consideration. Do I have to make some country and year dummies and the inteaction dummy (between income, country and year)? Do SAS have a procedure for this kind of analysis?

Thanks for the help

[SteveDenham · Accepted solution]

I would try PROC GLIMMIX.  It appears that you have an ordinal multinomial response, one continuous variable, one classification variable (country), and one variable that is going to cause problems (year).  I don't think year has the same meaning in each country, and in some sense, it is a repeated measurement on the country.

Something like this (untested) might be a good start:

proc glimmix data=yourdata;

class country year;

model happiness= income country income*country/dist=mult link=clogit;

random year/subject=country residual type=un;

run;

Options for odds ratios could be added.  Example 40.13 in the GLIMMIX documentation may help as well.

Steve Denham