help with Proc Mixed

[sky1]

I was trying to do repeated measure ANOVA using Proc mixed to analyze a
longitudinal data. In the data, there are about 40 subjects who had been
followed at 4 different time points. A few subjects had missing data at some
time points.

My code is as following:

proc mixed;

class id time;

model dependent = a b time / ddfm=kr solution;

repeated time / subject=id type=un;

run;

* a is time-variant continuous covaraite, b is time-invariant continuous covariate;

SAS output:

effect              time               estimate             standarderror              DF             tvalue               P

Intercept                         3.3442                1.5310                   74.2          2.18          0.0321

a                                     -0.00269             0.02882                  73.5           -0.09       0.9258

time               0                 -1.0780            0.7036                     76.2          -1.53       0.1296

time               1                 -1.1664            0.5936                    75.2           -1.97       0.0531

time              2                   -0.7012           0.3935                     70.6          -1.78       0.0791

time              3                           0                 .                                .                 .                .

b                                      0.000704         0.000153                   38.3           4.59        <.0001

Type 3 test fixed effects

Effect           NuMDF          DenDF             F-value            P

a                        1                    73.5                  0.01               0.9258

time                  3                    31.8                  7.87               0.0005

b                       1                     38.3                 21.08              <.0001

from the upper panel ("solution"), the time effect is not significant. Coefficients of "time" have p-values > 0.05. But in the lower panel (type 3 test of fixed effects), time effect is significant. P of time is 0.0005.

I don't understand why they are not consistent. which one should I use?

Thank you.

[SteveDenham]

It is important to know what is being tested.  In the solution panel, each level's estimate is tested against zero, whereas in the Type 3 tests, the 4 effects are tested against one another (a 3 df test).  In other words, are any of the time points 0, 1, and 2 different from time point 3.  You get these by adding the estimate for each time point to the intercept.  The joint F test is the one that probably addresses your question--are any of the responses different by time?

Steve Denham

[lvm]

No conflict at all, as indicated by Steve.Each of the first three time parameters is being tested versus 0 (which means, in this, case, with an over-parameterized model: Is the expected value for each time different from the last time?). The type 3 test consists of three contrasts simultaneously. Perhaps the mean for one time is different from a different time, or the mean of the first two times is different from the mean of the second two times, and so on and so forth.