为 r 中的检测概率创建箱线图

Question

我有一个类似于此的数据集：

sex	observed	date	idtag
M	0	10/20/2019	12
M	0	10/20/2019	12
F	0	10/20/2019	21
F	0	10/20/2019	21
M	0	10/21/2019	12
M	1	10/21/2019	14
F	0	10/21/2019	21
M	1	10/21/2019	14
M	1	10/21/2019	14
F	1	10/21/2019	21
M	0	10/23/2019	12
M	0	10/23/2019	12
F	0	10/23/2019	21
F	0	10/23/2019	22
M	0	10/23/2019	14
M	1	10/23/2019	12
F	0	10/23/2019	22
M	1	10/23/2019	14
M	1	10/23/2019	12

我想创建一个按性别划分的检测率箱线图。即，我想比较（每个性别 1s 的 sex/number 的观察总数）。我使用此代码按性别计算检出率：

drrate_sex <- detectiondata %>% 
    group_by(sex) %>% 
   summarise(dr = mean(observed))

这是我通常使用的标准箱线图代码：

boxplot(? ~ sex, data=drdata, main="Detection by sex",
        xlab="Sex", ylab="Detection rate (%)", notch=T, par(mar=c(4,12,4,12)))

我不确定如何将检测率（我在代码中输入 ? 的位置）合并到代码中以在 r 中生成一个箱线图来比较女性和男性的检测率。任何帮助将不胜感激。

Answer 1

对于此类数据，最接近箱线图的可能是比例图，其中置信区间由误差条显示。

获得所需数字的一种方法是运行逻辑回归，并使用其中的系数来获得比例的置信区间：

# Logistic regression model
regression <- glm(observed ~ sex, data = detectiondata, family = binomial)

# Extract coefficients
coefs <- summary(regression)$coef

# Convert coefficients to odds
coefs[2, 1] <- coefs[1, 1] + coefs[2, 1]
odds <- exp(cbind(mean = coefs[,1], 
                  upper = coefs[,1] + 1.96 * coefs[,2], 
                  lower = coefs[,1] - 1.96 * coefs[,2]))

# Convert odds to probabilities
probs <- odds/(1 + odds)

# Create data frame and add column for sex
df <- as.data.frame(probs)
df$sex <- c("Female", "Male")

给出以下数据框：

df
#>                  mean     upper      lower    sex
#> (Intercept) 0.1428571 0.5806110 0.01966987 Female
#> sexM        0.5000000 0.9168654 0.08313460   Male

我们可以这样绘制：

library(ggplot2)

ggplot(df, aes(sex, mean)) +
  geom_errorbar(aes(ymin = lower, ymax = upper, color = sex), size = 3,
                width = 0.3) +
  geom_point(size = 4, shape = 21, fill = "black")

数据

detectiondata <- structure(list(sex = c("M", "M", "F", "F", "M", "M", "F", "M", 
"M", "F", "M", "M", "F", "F", "M", "M", "F", "M", "M"), observed = c(0L, 
0L, 0L, 0L, 0L, 1L, 0L, 1L, 1L, 1L, 0L, 0L, 0L, 0L, 0L, 1L, 0L, 
1L, 1L), date = c("10/20/2019", "10/20/2019", "10/20/2019", "10/20/2019", 
"10/21/2019", "10/21/2019", "10/21/2019", "10/21/2019", "10/21/2019", 
"10/21/2019", "10/23/2019", "10/23/2019", "10/23/2019", "10/23/2019", 
"10/23/2019", "10/23/2019", "10/23/2019", "10/23/2019", "10/23/2019"
)), class = "data.frame", row.names = c(NA, -19L))

^{由 reprex package (v2.0.0)}

于 2021-11-05 创建

为 r 中的检测概率创建箱线图

Create boxplot for detection probabilities in r

r

boxplot

detection