Question

我正在尝试在直方图的顶部添加概率曲线，但是它似乎不起作用。例如

enter image description here

我想在右侧添加一条概率线，以便可以在左侧显示密度，并在右侧显示发生概率。

我完成的代码

x <- Delays_weather0$dif
h<-hist(x, breaks=10, col="red", xlab="Delays", 
   main="Flight Delays")

以及我要添加的概率曲线

my <- pnorm(-18:265, mean = mean(Delays_weather0$dif), sd = sd(Delays_weather0$dif), lower.tail = FALSE)
plot(my, type = "l")

我希望这是可以理解的

Answer 1

我们无权访问Delays_weather0数据集。因此，我将使用dep_delay数据包中提供的flights数据集nycflights13的前100个观察值。

由于默认情况下R中的直方图会绘制频率，所以我将概率乘以观察数（即1000）以使两个图具有可比性。

我首先使用的是lines函数。

library(nycflights13)
dataset <- flights$dep_delay[1:1000]
hist(x = dataset,
     breaks = 10,
     col = "red",
     xlab = "Delays", 
     main = "Flight Delays")
range_dataset <- range(dataset,
                       na.rm = TRUE)
equidistant_points_in_range <- seq(from = range_dataset[1],
                                   to = range_dataset[2],
                                   length.out = length(x = dataset))
upper_cdf_probabilities <- pnorm(q = equidistant_points_in_range,
                                 mean = mean(x = dataset,
                                             na.rm = TRUE),
                                 sd = sd(x = dataset,
                                         na.rm = TRUE),
                                 lower.tail = FALSE)
lines(x = length(x = dataset) * upper_cdf_probabilities,
      col = "blue")

^{由reprex package（v0.2.1）于2019-03-17创建}

使用curve函数的另一种方法。

dataset <- nycflights13::flights$dep_delay[1:1000]

range_dataset <- range(dataset,
                       na.rm = TRUE)

upper_tail_probability <- function(x)
{
  pnorm(q = x,
        mean = mean(x = dataset,
                    na.rm = TRUE),
        sd = sd(x = dataset,
                na.rm = TRUE),
        lower.tail = FALSE)
}

vectorized_upper_tail_probability <- Vectorize(FUN = upper_tail_probability)

hist(x = dataset,
     freq = FALSE,
     col = "red",
     xlab = "Delays", 
     main = "Flight Delays")

curve(expr = vectorized_upper_tail_probability,
      from = range_dataset[1],
      to = range_dataset[2],
      n = 1000,
      add = TRUE,
      col = "blue")

^{由reprex package（v0.2.1）于2019-03-17创建}

Answer 2

在@yarnabrina的可复制示例之后，（1）使用核密度估计器而不是假设正态性，（2）在右侧放置概率轴：

library(nycflights13)
npts <- 1000
dataset <- flights$dep_delay[1:npts]
par(las=1,bty="l",      ## cosmetic
       mar=c(5,4,2,5))  ## expand R margin to make room for second axis
h0 <- hist(x = dataset,
           breaks=100,
           col = "red",
           xlab = "Delay (min)",
           ylab="",
           main="",
           xlim=c(-50,200))   ## cosmetic: leave out a few extreme values
## put axis label at *top* of axis
mtext(side=2,at=550,"Frequency")
## compute kernel density estimate
dd <- density(na.omit(dataset), adjust=3)
dx <- diff(h0$mids)[1]  ## histogram bin width
## scale density to match count vales
lines(dd$x,dd$y*npts*dx,lwd=2,col="blue")
## set up auxiliary axis
dbrks <- seq(0,0.05,by=0.01)
axis(side=4,at=dbrks*npts*dx,label=dbrks)
mtext(side=4,at=550,"Probability")  ## axis label

在密度直方图上添加概率曲线

2 个答案: