我正尝试在R中开发一个模型，该模型说明使用从软件包MASS中使用stepAIC函数选择的预测变量子集的函数所收获的（眼扭曲）成眼的数量。为此，请参见下面的数据框结构：
> str(dat)
'data.frame': 92036 obs. of 47 variables:
$ month : Factor w/ 7 levels "4","5","6","7",..: 1 1 1 1 1 1 1 1 1 1 ...
$ day : int 1 2 13 13 13 15 16 16 16 16 ...
$ year : Factor w/ 29 levels "1989","1990",..: 28 28 28 28 28 28 28 28 28 28 ...
$ area : Factor w/ 9 levels "1","2","3","4",..: 1 1 1 1 1 1 1 1 1 1 ...
$ loc : int 42 13 42 42 31 42 31 42 42 12 ...
$ dow : int 5 6 3 3 3 5 6 6 6 6 ...
$ type : Factor w/ 3 levels "1","2","3": 1 3 2 3 1 3 1 1 1 1 ...
$ ang : int 3 8 3 6 3 6 2 2 3 3 ...
$ grid : int 903 801 903 803 903 804 903 903 903 801 ...
$ zone : int 18 10 18 16 18 16 14 12 14 14 ...
$ seek : int 334 334 334 334 334 334 334 334 334 334 ...
$ tis : num 10 7 6 7.5 9 7 8 7.5 8 6.5 ...
$ tif : num 15.5 12 13.5 14 13 15.5 11.5 12 12 10.5 ...
$ tii : num 15.5 12.5 14 14.5 13.5 15.5 11.5 12 12 10.5 ...
$ F1 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F2 : int 5 32 12 24 0 11 8 2 5 1 ...
$ F3 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F4 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F5 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F6 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F7 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F8 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F9 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F10 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F11 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F12 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F13 : int 0 1 1 0 0 0 2 0 1 0 ...
$ F14 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F15 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F16 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F17 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F18 : int 0 0 0 0 0 0 0 1 0 0 ...
$ F19 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F20 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F21 : int 0 0 0 0 0 0 0 0 0 0 ...
$ F22 : int 0 0 0 0 0 0 0 0 0 0 ...
$ method : Factor w/ 2 levels "1","2": 2 1 1 2 1 2 1 1 1 1 ...
$ WDWE : Factor w/ 2 levels "1","2": 1 2 1 1 1 1 2 2 2 2 ...
$ anghrs : num 16.5 40 22.5 39 12 51 7 9 12 12 ...
$ X : int NA NA NA NA NA NA NA NA NA NA ...
$ basin : Factor w/ 3 levels "east central",..: 2 2 2 2 2 2 2 2 2 2 ...
$ totwal : int 5 33 13 24 0 11 10 2 6 1 ...
$ time : num 12.75 9.5 9.75 10.75 11 ...
$ TOD : Factor w/ 2 levels "AM","PM": 2 1 1 1 1 1 1 1 1 1 ...
$ triplen: num 5.5 5 7.5 6.5 4 8.5 3.5 4.5 4 4 ...
$ effort : num 16.5 40 22.5 39 12 51 7 9 12 12 ...
$ WAE_POS: num 1 1 1 1 0 1 1 1 1 1 ...
我正在考虑的模型的主要影响如下：
f_main_effects <- as.formula(paste("totwal ~ year + month + TOD + ang + triplen + method + WDWE + basin + type + offset(log(effort))"))
有关模型和摘要，请参见下文：
> res_WAE_poiss <- glm(formula = f_main_effects, data = dat, family=poisson(link="log"))
Call:
glm(formula = f_main_effects, family = poisson(link = "log"),
data = dat)
Deviance Residuals:
Min 1Q Median 3Q Max
-13.053 -2.354 -0.815 1.235 36.543
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.7710820 0.0099014 -77.876 < 2e-16 ***
year1990 -0.3666092 0.0075276 -48.702 < 2e-16 ***
year1991 -0.5909582 0.0079259 -74.561 < 2e-16 ***
year1992 -0.2591255 0.0073414 -35.296 < 2e-16 ***
year1993 -0.0669383 0.0063155 -10.599 < 2e-16 ***
year1994 -0.3658504 0.0070433 -51.943 < 2e-16 ***
year1995 -0.2206199 0.0066573 -33.139 < 2e-16 ***
year1996 -0.1035683 0.0063993 -16.184 < 2e-16 ***
year1997 -0.4919732 0.0072490 -67.867 < 2e-16 ***
year1998 0.1029357 0.0065669 15.675 < 2e-16 ***
year1999 -0.4383308 0.0071890 -60.972 < 2e-16 ***
year2000 -0.4835465 0.0074930 -64.533 < 2e-16 ***
year2001 -0.2362845 0.0077348 -30.548 < 2e-16 ***
year2002 -0.4938362 0.0076850 -64.260 < 2e-16 ***
year2003 -0.3045370 0.0083361 -36.532 < 2e-16 ***
year2004 -0.1870754 0.0067665 -27.647 < 2e-16 ***
year2005 0.0084350 0.0074392 1.134 0.257
year2006 0.1609717 0.0063515 25.344 < 2e-16 ***
year2007 0.0352623 0.0068963 5.113 3.17e-07 ***
year2008 -0.1079411 0.0077131 -13.995 < 2e-16 ***
year2009 -0.2403672 0.0082776 -29.038 < 2e-16 ***
year2016 0.2797565 0.0061268 45.661 < 2e-16 ***
year2017 0.7801561 0.0057861 134.834 < 2e-16 ***
month5 0.2492277 0.0071682 34.769 < 2e-16 ***
month6 0.5002337 0.0069455 72.022 < 2e-16 ***
month7 0.5930889 0.0070208 84.476 < 2e-16 ***
month8 0.4875154 0.0074904 65.086 < 2e-16 ***
month9 0.3895295 0.0086433 45.067 < 2e-16 ***
month10 0.1483472 0.0108386 13.687 < 2e-16 ***
TODPM -0.1060913 0.0024875 -42.650 < 2e-16 ***
ang -0.0328293 0.0008702 -37.726 < 2e-16 ***
triplen -0.0924310 0.0006020 -153.527 < 2e-16 ***
method2 0.2659051 0.0027271 97.506 < 2e-16 ***
WDWE2 -0.0662235 0.0020664 -32.048 < 2e-16 ***
basinwest 0.3995765 0.0036221 110.316 < 2e-16 ***
basinwest central 0.1031291 0.0038891 26.518 < 2e-16 ***
type2 0.2337072 0.0032363 72.214 < 2e-16 ***
type3 0.5876307 0.0038143 154.060 < 2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for poisson family taken to be 1)
Null deviance: 865844 on 79916 degrees of freedom
Residual deviance: 602809 on 79879 degrees of freedom
(12119 observations deleted due to missingness)
AIC: 863541
Number of Fisher Scoring iterations: 5
然后，我尝试使用stepAIC考虑二阶交互并限制其始终包含年份和精力来选择一组预测变量。
res_WAE_pois_step <- stepAIC(glm(totwal ~ year + offset(log(effort)), data = dat, family=poisson(link="log")), scope=list(upper = totwal ~ (year + month + TOD + WDWE + method + basin + ang + triplen)^2 + offset(log(effort))), direction = "forward", trace=5)
这会导致以下错误：
Start: AIC=1119511
totwal ~ year + offset(log(effort))
Error in x[good, , drop = FALSE] : (subscript) logical subscript too long
我不确定在给定情况下如何解释此错误。值得一提的是，同一脚本在共享相同结构但观测值较少的数据集上可以正常运行。有谁知道是什么原因导致此错误，或者对如何解决该问题有建议？请让我知道更多信息是否有用，但是我不能合法共享数据。
感谢您的时间和帮助。
更新： 问题已经解决。 WDWE值在2010-2015年是不适用的，那时我不应该这样做。我能够用完整的数据更新数据框，脚本再次运行良好。就是说，有人可以帮助我理解1）为什么glm函数会默默地删除那些年而不是抛出错误？和2）如何根据新信息理解从stepAIC收到的错误？这些错误是否也是这些年来默默放弃的结果？