不从nlme包summary.lmList
我使用Pixel数据的标准误差数（NaN），让模型lmList功能：不从<code>nlme</code>包summary.lmList
dat <- lmList(pixel ~ day+I(day^2)|Dog/Side, data=Pixel[Pixel$Dog != 9,], level=2)
我很好奇，为什么，为什么我得到NaN为Dog==10当我尝试使用summary打印拟合对象？
summary(dat)
Call:
Model: pixel ~ day + I(day^2) | Dog/Side
Level: 2
Data: Pixel[Pixel$Dog != 9, ]
Coefficients:
(Intercept)
Estimate Std. Error t value Pr(>|t|)
1/R 1045.349 6.436476 162.41015 0
2/R 1042.166 6.436476 161.91569 0
3/R 1046.265 7.853767 133.21825 0
4/R 1045.602 7.853767 133.13382 0
5/R 1110.309 27.576874 40.26231 0
6/R 1093.556 27.576874 39.65482 0
7/R 1156.478 30.223890 38.26369 0
8/R 1030.754 30.223890 34.10393 0
10/R 1056.600 NaN NaN NaN
1/L 1046.538 6.436476 162.59486 0
2/L 1050.367 6.436476 163.18985 0
3/L 1047.438 7.853767 133.36754 0
4/L 1050.915 7.853767 133.81027 0
5/L 1068.412 27.576874 38.74306 0
6/L 1089.184 27.576874 39.49630 0
7/L 1139.851 30.223890 37.71356 0
8/L 1086.129 30.223890 35.93611 0
10/L 1041.100 NaN NaN NaN
day
Estimate Std. Error t value Pr(>|t|)
1/R 0.21534820 2.600975 0.08279519 9.343899e-01
2/R 3.82436362 2.600975 1.47035789 1.485802e-01
3/R 8.59752235 1.698113 5.06298479 7.828854e-06
4/R 12.18801561 1.698113 7.17738612 6.287493e-09
5/R 4.91365979 6.709441 0.73235013 4.678382e-01
6/R -0.01159794 6.709441 -0.00172860 9.986286e-01
7/R 0.27908291 7.755457 0.03598536 9.714568e-01
8/R 14.20961055 7.755457 1.83220800 7.369405e-02
10/R 16.10000000 NaN NaN NaN
1/L 2.22308391 2.600975 0.85471187 3.973407e-01
2/L 3.31617525 2.600975 1.27497407 2.090100e-01
3/L 6.03985508 1.698113 3.55680313 9.127977e-04
4/L 12.48222079 1.698113 7.35064026 3.512296e-09
5/L 14.13427835 6.709441 2.10662542 4.088737e-02
6/L 7.22757732 6.709441 1.07722501 2.872506e-01
7/L -0.77719849 7.755457 -0.10021311 9.206304e-01
8/L 3.97248744 7.755457 0.51221835 6.110599e-01
10/L 30.60000000 NaN NaN NaN
I(day^2)
Estimate Std. Error t value Pr(>|t|)
1/R -0.0507392 0.1819114 -0.2789227 7.816110e-01
2/R -0.2228509 0.1819114 -1.2250523 2.270733e-01
3/R -0.3556849 0.0755204 -4.7097854 2.498505e-05
4/R -0.4708779 0.0755204 -6.2351082 1.522147e-07
5/R -0.3510125 0.3639863 -0.9643565 3.401377e-01
6/R -0.0880891 0.3639863 -0.2420122 8.098952e-01
7/R -0.1462626 0.4245106 -0.3445440 7.320786e-01
8/R -0.7429334 0.4245106 -1.7500941 8.707333e-02
10/R -1.6250000 NaN NaN NaN
1/L -0.1649267 0.1819114 -0.9066324 3.695397e-01
2/L -0.2135152 0.1819114 -1.1737319 2.468167e-01
3/L -0.2764050 0.0755204 -3.6600044 6.720231e-04
4/L -0.5425352 0.0755204 -7.1839551 6.150012e-09
5/L -0.8313144 0.3639863 -2.2839170 2.725859e-02
6/L -0.5060199 0.3639863 -1.3902170 1.714560e-01
7/L -0.1847048 0.4245106 -0.4351005 6.656163e-01
8/L -0.1878769 0.4245106 -0.4425729 6.602428e-01
10/L -1.9500000 NaN NaN NaN
Residual standard error: 8.820516 on 44 degrees of freedom
对于Dog==10模型去精确地通过每一个数据点，这会导致NaN为Std. Error。
如果有机会，你可以展示你如何“手工”检查它，这将非常感激。我觉得我遇到了同样的问题，但我想确保它:) –
@YohanObadia在这种情况下，公式如下所示：pixel = a + b1 * day + b2 * day^2，其中a是截距的估计系数，b1是天的估计系数，b2是第^天的估计系数（它们的值在系数表中可见）。因此，如果您尝试在第4天预测“Dog == 10”（10/R）右侧的像素值，则它是：pixel = 1056.6 + 16.1 * 4 - 1.625 * 4^2，您会得到的预测像素值是1095. –
@YohanObadia在加载'library（nlme）'后打印'Pixel'，你会在第4天看到'Dog == 10'右侧的'pixel'的值，完全一样。这意味着预测是完美的。 –
可能错误非常低，无法打印。 – 2016-02-27 21:43:15
你有多少个数据点对于每个狗/侧 – user20650
@ user20650对于'Dog/Side'，有'像素数'的不同数量。对于'狗== 10'有3个措施。 –