我正在尝试使用rms包的fastbw函数进行反向回归，如下所示（使用mtcars数据集）：
> mod = ols(mpg~am+vs+cyl+drat+wt+gear, mtcars)
> mod
Linear Regression Model
ols(formula = mpg ~ am + vs + cyl + drat + wt + gear, data = mtcars)
Model Likelihood Discrimination
Ratio Test Indexes
Obs 32 LR chi2 58.26 R2 0.838
sigma 2.7008 d.f. 6 R2 adj 0.799
d.f. 25 Pr(> chi2) 0.0000 g 6.383
Residuals
Min 1Q Median 3Q Max
-4.3807 -1.4314 -0.5405 1.5828 5.4703
Coef S.E. t Pr(>|t|)
Intercept 39.9804 8.8745 4.51 0.0001
am 1.5981 1.9927 0.80 0.4301
vs 0.8011 1.9201 0.42 0.6801
cyl -1.3163 0.7033 -1.87 0.0730
drat 0.3488 1.6201 0.22 0.8313
wt -3.0390 0.9510 -3.20 0.0038
gear -1.1450 1.1420 -1.00 0.3256
> modbw = fastbw(mod)
> modbw
Deleted Chi-Sq d.f. P Residual d.f. P AIC R2
drat 0.05 1 0.8296 0.05 1 0.8296 -1.95 0.838
vs 0.17 1 0.6800 0.22 2 0.8974 -3.78 0.837
am 0.58 1 0.4473 0.79 3 0.8509 -5.21 0.833
gear 0.42 1 0.5194 1.21 4 0.8766 -6.79 0.830
Approximate Estimates after Deleting Factors
Coef S.E. Wald Z P
Intercept 39.686 1.8040 21.999 0.00000000
cyl -1.508 0.4362 -3.457 0.00054706
wt -3.191 0.7962 -4.008 0.00006128
Factors in Final Model
[1] cyl wt
以下是该模型的结构：
> str(modbw)
List of 10
$ result : num [1:4, 1:8] 0.0463 0.1701 0.5775 0.4152 1 ...
..- attr(*, "dimnames")=List of 2
.. ..$ : chr [1:4] "drat" "vs" "am" "gear"
.. ..$ : chr [1:8] "Chi-Sq" "d.f." "P" "Residual" ...
$ names.kept : chr [1:2] "cyl" "wt"
$ factors.kept : int [1:2] 3 5
$ factors.deleted: int [1:4] 4 2 1 6
$ parms.kept : int [1:3] 1 4 6
$ parms.deleted : int [1:4] 5 3 2 7
$ coefficients : Named num [1:3] 39.69 -1.51 -3.19
..- attr(*, "names")= chr [1:3] "Intercept" "cyl" "wt"
$ var : num [1:3, 1:3] 3.254 -0.303 -0.358 -0.303 0.19 ...
..- attr(*, "dimnames")=List of 2
.. ..$ : chr [1:3] "Intercept" "cyl" "wt"
.. ..$ : chr [1:3] "Intercept" "cyl" "wt"
$ Coefficients : num [1:4, 1:7] 41.26 43.17 42.39 39.69 1.68 ...
..- attr(*, "dimnames")=List of 2
.. ..$ : NULL
.. ..$ : chr [1:7] "Intercept" "am" "vs" "cyl" ...
$ force : NULL
- attr(*, "class")= chr "fastbw"
以下是摘要的输出结构（summary.lm函数不适用于此模型）：
> summary(modbw)
Length Class Mode
result 32 -none- numeric
names.kept 2 -none- character
factors.kept 2 -none- numeric
factors.deleted 4 -none- numeric
parms.kept 3 -none- numeric
parms.deleted 4 -none- numeric
coefficients 3 -none- numeric
var 9 -none- numeric
Coefficients 28 -none- numeric
force 0 -none- NULL
>
> summary.lm(modbw)
Error in if (p == 0) { : argument is of length zero
但我无法在其中任何一个中找到P值。如何获得fastbw函数最终模型的P值列表？
答案 0 :(得分：2)
p值的计算发生在print.fastbw函数中，由于某种原因，它们不会从函数返回。我打算使用print.fastbw的源代码自己重新计算它们，但我发现重写我自己的print.fastbw函数来返回p值要快得多。
这是重新设计的功能（注意print2不是通用的）：
print2.fastbw <- function (x, digits = 4, estimates = TRUE, ...)
{
res <- x$result
fd <- x$factors.deleted
if (length(fd)) {
cres <- cbind(dimnames(res)[[1]], format(round(res[,
1], 2)), format(res[, 2]), format(round(res[, 3],
4)), format(round(res[, 4], 2)), format(res[, 5]),
format(round(res[, 6], 4)), format(round(res[, 7],
2)), if (ncol(res) > 7)
format(round(res[, 8], 3)))
dimnames(cres) <- list(rep("", nrow(cres)), c("Deleted",
dimnames(res)[[2]]))
cat("\n")
if (length(x$force))
cat("Parameters forced into all models:\n", paste(x$force,
collapse = ", "), "\n\n")
print(cres, quote = FALSE)
if (estimates && length(x$coef)) {
cat("\nApproximate Estimates after Deleting Factors\n\n")
cof <- coef(x)
vv <- if (length(cof) > 1)
diag(x$var)
else x$var
z <- cof/sqrt(vv)
stats <- cbind(cof, sqrt(vv), z, 1 - pchisq(z^2,
1))
dimnames(stats) <- list(names(cof), c("Coef", "S.E.",
"Wald Z", "P"))
return(stats)
}
}
else cat("\nNo Factors Deleted\n")
cat("\nFactors in Final Model\n\n")
nk <- x$names.kept
if (length(nk))
print(nk, quote = FALSE)
else cat("None\n")
}
输出：
> results <- print2.fastbw(modbw)
Deleted Chi-Sq d.f. P Residual d.f. P AIC R2
drat 0.05 1 0.8296 0.05 1 0.8296 -1.95 0.838
vs 0.17 1 0.6800 0.22 2 0.8974 -3.78 0.837
am 0.58 1 0.4473 0.79 3 0.8509 -5.21 0.833
gear 0.42 1 0.5194 1.21 4 0.8766 -6.79 0.830
Approximate Estimates after Deleting Factors
> results
Coef S.E. Wald Z P
Intercept 39.686261 1.8039853 21.999216 0.000000e+00
cyl -1.507795 0.4362091 -3.456588 5.470608e-04
wt -3.190972 0.7961871 -4.007817 6.128261e-05
和p值：
> results[,4]
Intercept cyl wt
0.000000e+00 5.470608e-04 6.128261e-05
答案 1 :(得分：2)
它们是即时计算的。在rms:::print.fastbw内挖掘（类print的对象的fastbw方法），您可以找到：
cof <- coef(x)
vv <- if (length(cof) > 1) diag(x$var) else x$var
z <- cof/sqrt(vv)
stats <- cbind(cof, sqrt(vv), z, 1 - pchisq(z^2, 1))
（如果您想要更准确的小p值，最好将pchisq(z^2,1,lower.tail=FALSE)替换为1-pchisq(z^2,1)）
如果x是fastbw个对象，则stats的最后一列会显示您的p值。