[R-sig-ME] Tukey after lme does not match

Dennis Murphy djmuser at gmail.com
Fri Jan 6 05:36:41 CET 2012


Hi:

In addition to Dr. Robinson's comments, I would add the following:

(i) Your model specification indicates that you want random slopes for
trial by subject with correlated intercepts. Is that what you
intended?
(ii) I'm wondering why you're not treating angle as a continuous
variable and looking for potential trends in the response as a
function of angle. If you have a discernable trend, its form would be
more useful than a collection of multiple comparisons. Did you plot
the response by subject and angle (either a conditioning plot by
subject or a 'spaghetti plot' of individual profiles of (angle, RMSy)
pairs)?

My 2c,
Dennis

On Thu, Jan 5, 2012 at 4:22 PM, Alen Hajnal <Alen.Hajnal at usm.edu> wrote:
> Dear R users:
> I have a simple lme model based on the following data:
> sub     trial   angle   RMSy
> 1       1       30      3.745084
> 1       2       0       7.520667
> 1       3       90      11.17038
> 1       4       15      7.581526
> 1       5       60      11.17822
> 1       6       75      8.440891
> 1       7       45      13.19024
> 1       8       15      9.822035
> 1       9       60      6.002665
> 1       10      75      4.393961
> 1       11      0       7.436676
>
> When I run the model the results show that 0vs45, 0vs60, 0vs75, and 0vs90 are all significant main effects:
>
>> m.base01<-lme(RMSy ~ angle+trial, data=data,  random=~ trial|sub,method='ML')
>> summary(m.base01)
> Fixed effects: RMSy ~ angle + trial
>                        Value   Std.Error       DF     t-value     p-value
> (Intercept)  5.236020   0.6312637       267 8.294505  0.0000
> angle15     -0.687669   0.4140277       267 -1.660925  0.0979
> angle30     -0.571092   0.4129365       267 -1.383001  0.1678
> angle45     -0.984597   0.4139330       267 -2.378638  0.0181
> angle60     -0.874718   0.4135615       267 -2.115085  0.0353
> angle75     -1.389835   0.4113411       267 -3.378788  0.0008
> angle90     -1.620493   0.4133209       267 -3.920666  0.0001
> trial       -0.009372   0.0299782       267 -0.312620  0.7548
>
> However, when I try to run a Tukey posthoc test, I get different results (notice that the estimates are the same as the main effect values above): ONLY 0vs90 is significant at p<.05 :
>
>> summary(glht(m.base01, linfct=mcp(angle = "Tukey")))
>
>         Simultaneous Tests for General Linear Hypotheses
>
> Multiple Comparisons of Means: Tukey Contrasts
>
>
> Fit: lme.formula(fixed = RMSy ~ angle + trial, data = data, random = ~trial |
>    sub, method = "ML")
>
> Linear Hypotheses:
>             Estimate Std. Error z value Pr(>|z|)
> 15 - 0 == 0   -0.6877     0.4082  -1.684  0.62652
> 30 - 0 == 0   -0.5711     0.4072  -1.403  0.80077
> 45 - 0 == 0   -0.9846     0.4081  -2.412  0.19329
> 60 - 0 == 0   -0.8747     0.4078  -2.145  0.32592
> 75 - 0 == 0   -1.3898     0.4056  -3.427  0.01070 *
> 90 - 0 == 0   -1.6205     0.4075  -3.976  0.00138 **
> 30 - 15 == 0   0.1166     0.4079   0.286  0.99996
> 45 - 15 == 0  -0.2969     0.4078  -0.728  0.99090
> 60 - 15 == 0  -0.1870     0.4078  -0.459  0.99931
> 75 - 15 == 0  -0.7022     0.4050  -1.734  0.59342
> 90 - 15 == 0  -0.9328     0.4073  -2.290  0.24842
> 45 - 30 == 0  -0.4135     0.4082  -1.013  0.95124
> 60 - 30 == 0  -0.3036     0.4074  -0.745  0.98971
> 75 - 30 == 0  -0.8187     0.4052  -2.021  0.40123
> 90 - 30 == 0  -1.0494     0.4076  -2.574  0.13377
> 60 - 45 == 0   0.1099     0.4077   0.269  0.99997
> 75 - 45 == 0  -0.4052     0.4059  -0.998  0.95449
> 90 - 45 == 0  -0.6359     0.4078  -1.559  0.70835
> 75 - 60 == 0  -0.5151     0.4050  -1.272  0.86496
> 90 - 60 == 0  -0.7458     0.4074  -1.831  0.52730
> 90 - 75 == 0  -0.2307     0.4049  -0.570  0.99763
>
> Why does the Tukey not match the lme results?
> Any help would be much appreciated!
> Thanks, Alen
>
>
>
>
> ----------
> Alen Hajnal, PhD.
> Department of Psychology
> The University of Southern Mississippi
> 118 College Drive #5025
> Hattiesburg, MS 39406
> USA
> Tel. +1 (601) 266-4617
> alen.hajnal @ usm.edu
> http://ocean.otr.usm.edu/~w785427/lab.html
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models




More information about the R-sig-mixed-models mailing list