如何在R中近似绘制分组分配概率函数曲线？

Alright, let's break down how to approximate and plot that assignment probability curve in R—since you mentioned it's a research scenario where probability shifts with a continuous X, I'll cover flexible approaches you can tweak to match your specific chart.

1. Define Your Probability Function First #

The key is translating the pattern from your chart into a mathematical function. Here are common examples matching research scenarios:

• Logistic curve (super common for propensity scores or binary assignment):

  # Adjust the intercept and slope to match your chart's shape
  prob_function <- function(x) plogis(-1 + 0.6*x)
  

• Linear probability (simple linear trend, but clamp values to 0-1 to avoid invalid probabilities):

  prob_function <- function(x) pmin(pmax(0.1 + 0.1*x, 0), 1)
  

• Piecewise (segmented) function (if your chart has breakpoints where the trend changes):

  prob_function <- function(x) {
    ifelse(x < 2, 0.2,          # Flat below X=2
           ifelse(x <=5, 0.2 + (0.6/3)*(x-2),  # Linear increase between 2-5
                  0.8))         # Flat above X=5
  }
  

2. Generate a Dense X Sequence #

To get a smooth curve, create a dense sequence of X values covering the range shown in your chart:

# Replace min_x and max_x with the actual range from your chart
x_vals <- seq(from = -3, to = 5, length.out = 1000)

3. Calculate Probabilities & Plot #

You can use either base R or ggplot2—here are both options:

Base R Version

# Compute probability values
y_vals <- prob_function(x_vals)

# Plot the curve
plot(x_vals, y_vals, type = "l", lwd = 2, col = "steelblue",
     xlab = "Continuous Variable X", ylab = "Assignment Probability",
     main = "Assignment Probability vs. X")

# Optional: Add a reference line (e.g., 50% probability)
abline(h = 0.5, lty = 2, col = "gray50")

ggplot2 Version (Cleaner for Customization)

library(ggplot2)

# Create a data frame for ggplot
curve_data <- data.frame(X = x_vals, Probability = prob_function(x_vals))

# Build the plot
ggplot(curve_data, aes(x = X, y = Probability)) +
  geom_line(color = "steelblue", linewidth = 1) +
  labs(x = "Continuous Variable X", y = "Assignment Probability",
       title = "Assignment Probability as a Function of X") +
  geom_hline(yintercept = 0.5, linetype = "dashed", color = "gray50") +
  theme_minimal()

4. If You Only Have Points from the Chart #

If you don't know the exact function but can pull a few (X, Probability) pairs from the chart, use interpolation to approximate the curve:

# Example points extracted from your chart
observed_points <- data.frame(
  X = c(-2, 0, 2, 4),
  Prob = c(0.1, 0.3, 0.7, 0.9)
)

# Generate dense X sequence
x_seq <- seq(min(observed_points$X), max(observed_points$X), length.out = 500)

# Linear interpolation to fill in the curve
interpolated_curve <- approx(x = observed_points$X, y = observed_points$Prob, xout = x_seq)

# Plot the interpolated curve + original points
plot(interpolated_curve$x, interpolated_curve$y, type = "l", col = "darkred", lwd = 2,
     xlab = "X", ylab = "Assignment Probability")
points(observed_points$X, observed_points$Prob, pch = 16, col = "black")

Just tweak the function parameters or interpolated points to match the exact shape of your reference chart, and you'll have a solid approximation.

内容的提问来源于stack exchange，提问作者Simon Harmel