Hi, I'm running a mixed model for longitudinal data with a two by two categorical interaction (all other variables being continuous). grceintra is coded like 0 for low ec and 1 for high ec. time is coded as 1 for time1, 2 for time 2 3 for time3 and 4 for time4.
here is the mixed command and the results :
Code:
 xtmixed rmssd i.grceintra##i.time alc caf cig bmi ||id:alc caf cig bmi , residuals(un, t(time))
Code:
Mixed-effects ML regression                     Number of obs     =        268
Group variable: id                              Number of groups  =         68

                                                Obs per group:
                                                              min =          3
                                                              avg =        3.9
                                                              max =          4

                                                Wald chi2(11)     =      45.69
Log likelihood =  -56.68586                     Prob > chi2       =     0.0000

--------------------------------------------------------------------------------
         rmssd |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------+----------------------------------------------------------------
   1.grceintra |  -.1516471   .1237396    -1.23   0.220    -.3941723    .0908781
               |
          time |
            2  |  -.0828655   .0517872    -1.60   0.110    -.1843665    .0186355
            3  |   -.235263   .0556458    -4.23   0.000    -.3443267   -.1261993
            4  |  -.1622465     .04143    -3.92   0.000    -.2434478   -.0810453
               |
grceintra#time |
          1 2  |   .1037683   .0721843     1.44   0.151    -.0377103    .2452469
          1 3  |   .0890882   .0782335     1.14   0.255    -.0642466     .242423
          1 4  |   .1410346   .0577477     2.44   0.015     .0278511    .2542181
               |
           alc |  -.1268449   .0641427    -1.98   0.048    -.2525624   -.0011275
           caf |  -.0009671   .0437007    -0.02   0.982    -.0866189    .0846846
           cig |   .0116784   .0402781     0.29   0.772    -.0672652    .0906221
           bmi |   .0007147    .019349     0.04   0.971    -.0372087     .038638
         _cons |   4.209916   .4915439     8.56   0.000     3.246508    5.173325
--------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
id: Independent              |
                     sd(alc) |   1.81e-09          .             .           .
                     sd(caf) |   1.99e-09          .             .           .
                     sd(cig) |   1.65e-10          .             .           .
                     sd(bmi) |    .022004          .             .           .
-----------------------------+------------------------------------------------
Residual: Unstructured       |
                      sd(e1) |   .1679547          .             .           .
                      sd(e2) |   .2756804          .             .           .
                      sd(e3) |   .3655816          .             .           .
                      sd(e4) |    .129396          .             .           .
                 corr(e1,e2) |   .1695978          .             .           .
                 corr(e1,e3) |   .5009059          .             .           .
                 corr(e1,e4) |  -.2689602          .             .           .
                 corr(e2,e3) |   .6809035          .             .           .
                 corr(e2,e4) |  -.0108121          .             .           .
                 corr(e3,e4) |   .4826205          .             .           .
------------------------------------------------------------------------------
LR test vs. linear model: chi2(13) = 330.39               Prob > chi2 = 0.0000
when I use the contrast command to test main and interaction effect, the result is the following :
Code:
 contrast time##grcetot

Contrasts of marginal linear predictions

Margins      : asbalanced

------------------------------------------------
             |         df        chi2     P>chi2
-------------+----------------------------------
rmssd        |
        time |          3       34.92     0.0000
             |
     grcetot |          1        3.15     0.0760
             |
time#grcetot |          3        6.26     0.0998
so, it's a little bit disturbing as :
1. mixed results show that high EC have lower rmssd (the DV) than low EC (coef = -.15) but the contrats command tells us that there is no main effect of the IV (grcetot chi2= .3.15, p=.076).
2. the interaction terme show that high EC group exhibit a significant gain of .14 between time1 and time4 than low EC group. but again, the overall interaction term is not significant (chi2 = 6.26, p = .09).

In social science we are not used to compute follow-up analysis after regressions because all coef in the mixed table are sufficient. But i'am a little bit obsessive with stat !!! (sorry).
I don't know what to conclude with such discrepancy. any help is welcome...
best
carole