问题
I'm trying to use lmerTest to have p-values for my fixed effects. I have 4 different random intercepts, 3 crossed and one nested :
test.reml <- lmerTest::lmer(y ~ s1 + min + cot + min:cot + ge
+ vis + dur + mo + nps + dist + st1 + st2 + di1 + s1:cot
+ s1:min + s1:cot:min + s1:ge + s1:vis + s1:dur + s1:mo
+ s1:nps + s1:dist + s1:st1 + s1:st2 + s1:di1 + (1|Unique_key)
+ (s1-1|object) + (ns1-1|object)
+ (1|region), bdr, REML=1)
The objects are observed two times and the correlation between the two measures is introduced by the random effect on Unique_key, an unique identifier of the object i in a region j. Each object could be observed in any region. S1 is a binary variable that takes the value 1 if the observation is observed the first time period and 0 either. There is one random intercept for the first periode and one random intercept for the second period for each object. ns1 is in fact a binary variable that is the complement of s1 and s1 + ns1 = 1 for each observation.
I can fit the model and get the estimates and p-values with summary() :
summary( test.reml)
Linear mixed model fit by REML ['merModLmerTest']
Formula: y ~ s1 + min + cot + min:cot + ge
+ vis + dur + mo + nps + dist + st1 + st2 + di1 + s1:cot
+ s1:min + s1:cot:min + s1:ge + s1:vis + s1:dur + s1:mo
+ s1:nps + s1:dist + s1:st1 + s1:st2 + s1:di1 + (1|Unique_key)
+ (s1-1|object) + (ns1-1|object)
+ (1|region), bdr, REML=1)
Data: bdr
REML criterion at convergence: 204569.1
Random effects:
Groups Name Variance Std.Dev.
Unique_key (Intercept) 0.2023 0.4497
object s1 0.3528 0.5940
object.1 ns1 0.5954 0.7716
Region (Intercept) 0.7563 0.8697
Residual 0.1795 0.4237
Number of obs: 113396, groups: Unique_key , 58541; object, 1065; Region, 87
Fixed effects:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 6.7341569 0.2382673 28.263 < 2e-16 ***
s1 0.7391924 0.2004413 3.688 0.000233 ***
min -0.0067606 0.0171385 -0.394 0.694205
cot 0.1235093 0.0353693 3.492 0.000499 ***
ge2 -0.1535452 0.0800998 -1.917 0.055525 .
ge3 -0.2131246 0.0986559 -2.160 0.030982 *
ge4 -0.1032694 0.1115603 -0.926 0.354830
ge5 -0.1769347 0.1296558 -1.365 0.172663
ge6 0.0117401 0.1115897 0.105 0.916231
ge7 -0.2692483 0.1022565 -2.633 0.008589 **
vis2 -0.0928661 0.0607950 -1.528 0.126938
vis3 -0.3026112 0.1246595 -2.428 0.015375 *
dur2 0.1479195 0.0786369 1.881 0.060249 .
dur3 0.1406340 0.0809379 1.738 0.082590 .
dur4 0.2742243 0.0884301 3.101 0.001981 **
dur5 0.1946761 0.1065815 1.827 0.068059 .
mo2 -0.1168591 0.1256017 -0.930 0.352386
mo3 -0.0611162 0.1267657 -0.482 0.629824
mo4 -0.2725720 0.1263740 -2.157 0.031248 *
mo5 -0.6107000 0.1379264 -4.428 1.05e-05 ***
mo6 -0.3635142 0.1299799 -2.797 0.005260 **
mo7 -0.0899233 0.1275164 -0.705 0.480846
mo8 -0.2349548 0.1253422 -1.875 0.061140 .
mo9 -0.2624888 0.1263051 -2.078 0.037934 *
mo10 -0.2882749 0.1244404 -2.317 0.020724 *
mo11 -0.1702823 0.1356031 -1.256 0.209497
mo12 0.1989155 0.1322339 1.504 0.132819
nps 0.0278418 0.0010393 26.790 < 2e-16 ***
dist2 0.4065093 0.1118916 3.633 0.000294 ***
dist3 0.0155691 0.0906664 0.172 0.863693
dist4 -0.2910960 0.1595805 -1.824 0.068424 .
dist5 -0.1316553 0.0913394 -1.441 0.149782
dist6 0.0477956 0.0995679 0.480 0.631308
dist7 0.1383000 0.0981247 1.409 0.159011
dist8 -0.3985620 0.0886316 -4.497 7.69e-06 ***
dist9 -0.2036683 0.0799584 -2.547 0.011005 *
st11 -0.0258775 0.0591631 -0.437 0.661919
st21 0.0089230 0.0573352 0.156 0.876356
di11 -0.0910207 0.0838321 -1.086 0.277846
min:cot 0.0066210 0.0006195 10.688 < 2e-16 ***
s1:cot -0.1505670 0.0443186 -3.397 0.000694 ***
s1:min 0.0079478 0.0015051 5.280 1.29e-07 ***
s1:ge2 0.0329272 0.1007943 0.327 0.743948
s1:ge3 0.2150927 0.1241590 1.732 0.083367 .
s1:ge4 0.1786057 0.1404119 1.272 0.203526
s1:ge5 -0.0422380 0.1631757 -0.259 0.795780
s1:ge6 0.1372051 0.1404415 0.977 0.328717
s1:ge7 0.1343314 0.1287059 1.044 0.296755
s1:vis2 0.1354091 0.0765084 1.770 0.076913 .
s1:vis3 0.2449180 0.1568745 1.561 0.118637
s1:dur2 -0.0888179 0.0989573 -0.898 0.369547
s1:dur3 -0.0532473 0.1018481 -0.523 0.601167
s1:dur4 -0.1239068 0.1112907 -1.113 0.265696
s1:dur5 -0.1191069 0.1341435 -0.888 0.374705
s1:mo2 -0.1357615 0.1574365 -0.862 0.388618
s1:mo3 0.0130976 0.1588743 0.082 0.934306
s1:mo4 0.0343900 0.1579532 0.218 0.827669
s1:mo5 0.2257241 0.1732449 1.303 0.192761
s1:mo6 0.0500347 0.1628755 0.307 0.758728
s1:mo7 -0.0451271 0.1596277 -0.283 0.777435
s1:mo8 -0.0200467 0.1572383 -0.127 0.898564
s1:mo9 0.0394005 0.1584268 0.249 0.803620
s1:mo10 0.0641038 0.1562518 0.410 0.681662
s1:mo11 -0.3136235 0.1703456 -1.841 0.065764 .
s1:mo12 -0.7003775 0.1660455 -4.218 2.58e-05 ***
s1:nps -0.0095428 0.0013077 -7.297 4.31e-13 ***
s1:dist2 -0.3867962 0.1407463 -2.748 0.006050 **
s1:dist3 -0.0516400 0.1140519 -0.453 0.650762
s1:dist4 -0.0567491 0.2008542 -0.283 0.777562
s1:dist5 0.0025780 0.1147143 0.022 0.982073
s1:dist6 -0.1456445 0.1252219 -1.163 0.244940
s1:dist7 -0.0452712 0.1234110 -0.367 0.713785
s1:dist8 0.0546400 0.1114865 0.490 0.624117
s1:dist9 0.0540697 0.1000415 0.540 0.588934
s1:st11 0.0784027 0.0744677 1.053 0.292549
s1:st21 -0.0394419 0.0721720 -0.546 0.584788
s1:di11 0.0463040 0.1055326 0.439 0.660882
s1:min:cot -0.0012850 0.0006004 -2.140 0.032344 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
but with anova() I get :
type3.bonmodele <- lmerTest::anova(test.reml, ddf="Satterthwaite")
Analysis of Variance Table
Df Sum Sq Mean Sq F value
s1 1 7.385 7.385 41.1448
min 1 0.081 0.081 0.4536
cot 1 29.384 29.384 163.7026
ge 6 25.198 4.200 23.3968
vis 2 0.464 0.232 1.2929
dur 4 22.763 5.691 31.7042
mo 11 15.581 1.416 7.8914
nps 1 234.535 234.535 1306.6487
dist 8 18.547 2.318 12.9162
st1 1 0.034 0.034 0.1879
st2 1 0.058 0.058 0.3220
di1 1 0.261 0.261 1.4549
min:cot 1 22.537 22.537 125.5611
s1:cot 1 9.146 9.146 50.9555
s1:min 1 18.383 18.383 102.4171
s1:ge 6 5.152 0.859 4.7843
s1:vis 2 1.698 0.849 4.7311
s1:dur 4 2.829 0.707 3.9404
s1:mo 11 8.157 0.742 4.1312
s1:nps 1 10.102 10.102 56.2803
s1:dist 8 2.233 0.279 1.5550
s1:st1 1 0.188 0.188 1.0481
s1:st2 1 0.046 0.046 0.2560
s1:di1 1 0.035 0.035 0.1927
s1:min:cot 1 0.822 0.822 4.5804
When I try to remove the triple interaction the anova() function returns the p-values...I have also tried to split my data frame and to fit the model on half the data and anova() works well to.
There is no warning when I use the functions and I have also tried to change the ddf option and the method but nothing seems to work.
Here is my session info :
R version 3.0.0 (2013-04-03)
Platform: x86_64-w64-mingw32/x64 (64-bit)
locale:
[1] LC_COLLATE=French_Canada.1252 LC_CTYPE=French_Canada.1252 LC_MONETARY=French_Canada.1252 LC_NUMERIC=C
[5] LC_TIME=French_Canada.1252
attached base packages:
[1] parallel splines stats graphics grDevices utils datasets methods base
other attached packages:
[1] ggplot2_0.9.3.1 snow_0.3-13 Snowball_0.0-10 xtable_1.7-1 lmerTest_2.0-0 pbkrtest_0.3-7 MASS_7.3-29
[8] papeR_0.3 gmodels_2.15.4 survival_2.37-4 nlme_3.1-111 car_2.0-19 lme4_1.1-1 Matrix_1.1-0
[15] lattice_0.20-15
loaded via a namespace (and not attached):
[1] bitops_1.0-6 caTools_1.16 cluster_1.14.4 colorspace_1.2-4 dichromat_2.0-0 digest_0.6.3
[7] gdata_2.13.2 gplots_2.12.1 grid_3.0.0 gtable_0.1.2 gtools_3.0.0 Hmisc_3.12-2
[13] KernSmooth_2.23-10 labeling_0.2 minqa_1.2.1 munsell_0.4.2 nnet_7.3-7 numDeriv_2012.9-1
[19] plyr_1.8 proto_0.3-10 RColorBrewer_1.0-5 RCurl_1.95-4.1 reshape2_1.2.2 rJava_0.9-4
[25] ROAuth_0.9.3 rpart_4.1-3 scales_0.2.3 stringr_0.6.2 tools_3.0.0 twitteR_1.1.7
I can't share the data but I can add more infos if needed! I wanted to use the Satterthwaite approximation for the degrees of freedom but if you have any other suggestions to get p-values please share! Thank you very much!
回答1:
If some error occurs in lmerTest then by default the anova from lme4 will be given. So in your case some error occurred, but it is quite hard to say what without testing on the data. Probably it is due to the simple method for the grad function, which is the default. You may try: anova(test.reml, method.grad="Richardson"). Otherwise again like I said quite hard to say without looking at the example...
Alexandra Kuznetsova
来源：https://stackoverflow.com/questions/20209237/anova-does-not-display-p-value-when-used-with-lmertest