|
Ryan,
It's hard to convince me to analyze means at the state level (level 2)
when you have the actual data to work with (level 1). A random coef model
in MIXED will compute the mean predicted value for each state, based on
the actual (level-1) data as follows:
proc mixed data = indat;
class state;
model y = x / solution ddfm = contain outpm=yhatm(keep=state pred );
random intercept x / subject = state type = un solution;
run;
proc sort data=yhatm; by state;
data yhatm; set yhatm; by state; if first.state;
proc print data=yhatm; run;
where y and x are the level-1 (actual data). You can also get the level-1
predicted values with outp=yhat(keep= _____)
Robin High
UNMC
Ryan <Ryan.Andrew.Black@GMAIL.COM>
Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
06/03/2008 04:26 PM
Please respond to
Ryan <Ryan.Andrew.Black@GMAIL.COM>
To
SAS-L@LISTSERV.UGA.EDU
cc
Subject
How to Set Up Regression Model in SAS?
Hi all-
There may be a very easy answer to this, but I just can't seem to
figure it out...
I want to test a regression model where the dependent variable is NOT
at the "individual" level. I want to create a model to predict, for
instance, quality of health (QOH) at the state level, not "individual"
level by socio-economic status (SES) at the state level.
Let's say I have data on 20 states, here's what the dataset would look
like:
Participant State SES_level_at_subj_level
SES_at_state_level QOH_at_subj_level QOH_at_state_level
1 1
24 27.75
56 65.75
2 1
25 27.75
62 65.75
3 1
30 27.75
77 65.75
4 1
32 27.75
68 65.75
. . . . . .
. . . . . .
. . . . . .
200 20
45 49
55 68.75
where
SES_at_state_level = average SES score of all participants who reside
in that state
and
QOH_at_state_level = average quality of health of all participants who
reside in that state
-----------------------------------
---So I think my equation would look like this (although I'm sure it's
wrong, particularly when calculating df):
QOH_at_state_level = b0 + b1(SES_at_state_level)
---I'm worried that the type I error will be too high when I analyze
under OLS---particularly if I have a low number of states (e.g. 10
instead of 20) even though I have 200 participants.
I'd really, really appreciate any help in terms of the SAS syntax or
an answer to the statistical issue.
Thanks,
Ryan
PS I have other predictors, but I figured one predictor would be
enough to explain the issue
| 
