Question

我正在尝试使用ggplot2来绘制负二项式回归的预测值，一个打开二进制变量，另一个打开它。因此，可以比较两个两个图。

链接here演示了如何在页面底部执行此操作，但我希望能够使用强大的标准错误围绕预测值的绘图创建着色。我没有看到如何从predict（）函数中获取它。这段代码示例是否有任何解决方法可以获得强大的标准误差以遮盖绘制的线条？

我使用this site中的代码生成强大的标准错误：

require(sandwich)
cov.nb1 <- vcovHC(nb1, type = "HC0")
std.err <- sqrt(diag(cov.nb1))
r.est <- cbind(Estimate = coef(nb1), `Robust SE` = std.err, `Pr(>|z|)` = 2 *
    pnorm(abs(coef(nb1)/std.err), lower.tail = FALSE), LL = coef(nb1) - 1.96 *
    std.err, UL = coef(nb1) + 1.96 * std.err)

r.est

我正在使用的模型是：

nb1 <- glm.nb(citecount ~ expbin*novcr + expbin*I(novcr^2) + disease + length +
as.factor(year), data = nov4d.dt)

我正在使用的数据样本是：

nov4d.dt  <-
    structure(list(PMID = c(1279136L, 1279186L, 1279186L, 1279187L, 
    1279187L, 1279190L, 1279257L, 1279317L, 1279332L, 1279523L), 
        min = c(0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L, 0L), max = c(32L, 
        32L, 32L, 32L, 32L, 32L, 32L, 32L, 32L, 32L), mean = c(11L, 
        13L, 13L, 19L, 19L, 16L, 24L, 15L, 8L, 19L), length = c(45L, 
        120L, 120L, 78L, 78L, 136L, 45L, 36L, 171L, 78L), threslength = c(13L, 
        20L, 20L, 7L, 7L, 26L, 4L, 6L, 77L, 14L), novlength = c(5L, 
        6L, 6L, 3L, 3L, 6L, 3L, 3L, 36L, 5L), novind = c("TRUE", 
        "TRUE", "TRUE", "TRUE", "TRUE", "TRUE", "TRUE", "TRUE", "TRUE", 
        "TRUE"), novcr = c(0.111111, 0.05, 0.05, 0.0384615, 0.0384615, 
        0.0441176, 0.0666667, 0.0833333, 0.210526, 0.0641026), novcrt = c(0.288889, 
        0.166667, 0.166667, 0.0897436, 0.0897436, 0.191176, 0.0888889, 
        0.166667, 0.450292, 0.179487), year = c(1991L, 1991L, 1992L, 
        1992L, 1992L, 1992L, 1992L, 1992L, 1991L, 1992L), disease = structure(c(1L, 
        4L, 2L, 4L, 2L, 1L, 4L, 4L, 2L, 4L), .Label = c("alz", "bc", 
        "cl", "lc"), class = "factor"), citecount = c(5L, 8L, 8L, 
        12L, 12L, 0L, 1L, 0L, 92L, 0L), novind2 = c(TRUE, TRUE, TRUE, 
        TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE), rad = c(FALSE, 
        FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE
        ), exp = c(260, 351, 351, 65, 65, 480, 104, 273, 223, 0), 
        novind4 = c(FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, 
        FALSE, TRUE, FALSE), novind5 = c(FALSE, FALSE, FALSE, FALSE, 
        FALSE, FALSE, FALSE, FALSE, TRUE, FALSE), novind6 = c(FALSE, 
        FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE
        ), expbin = c(TRUE, TRUE, TRUE, FALSE, FALSE, TRUE, FALSE, 
        TRUE, TRUE, FALSE), expbin2 = c(TRUE, TRUE, TRUE, FALSE, 
        FALSE, TRUE, FALSE, TRUE, TRUE, FALSE)), .Names = c("PMID", 
    "min", "max", "mean", "length", "threslength", "novlength", "novind", 
    "novcr", "novcrt", "year", "disease", "citecount", "novind2", 
    "rad", "exp", "novind4", "novind5", "novind6", "expbin", "expbin2"
    ), sorted = "PMID", class = c("data.frame"), row.names = c(NA, 
    -10L))

Answer 1

您提供的链接会创建一个模型，创建一个合成数据集，其中一个预测变量沿其整个范围变化，将模型和合成数据集传递给predict（），然后绘制结果预测。您需要做的唯一重要的事情就是将强大的std.err放入数据框中以计算CI。

#look at how model thinks citecount ~ novcr for two values of expbin 
#make synthetic data with a range of range(df$novcr)
#include logical predictor variable expbin
#such that each level of expbin has all the novcr values

newdata2 <- data.frame(novcr = rep(seq(from = min(nov4d.dt$novcr), 
    to = max(nov4d.dt$novcr), length.out = 100), 2), 
    expbin  = rep(0:1, each = 100))

#convert expbin type to logical
newdata2$expbin <- as.logical(newdata2$expbin)

# add in the mean or default values of other predictors
# because I assume predict() needs vals for all parameters in the model
newdata2$length <- mean(nov4d.dt$length,na.rm=T)
newdata2$disease <- factor("alz")
newdata2$year <- factor("1992")

（继续上述步骤，直到合成数据帧具有模型所需的所有变量）

#make predict and add it to synthetic data
newdata2$fit <- predict(nb1, newdata2, type = "response")

# include CIs based on your robust se
newdata2$LL <- newdata2$fit - 1.96 * std.err["novcr"]
newdata2$UL <- newdata2$fit + 1.96 * std.err["novcr"]

#plot
ggplot(newdata2, aes(novcr, fit)) + 
    geom_ribbon(aes(ymin = LL, ymax = UL, fill = expbin), 
    alpha = 0.25) + geom_line(aes(colour = expbin), size = 2)

使用ggplot2绘制具有稳健标准误差的预测值

1 个答案: