Written and maintained by CASRAI Editorial Board
Published · Last updated
On this page
The Gaussian distribution in R is implemented through four foundational functions in the base stats package: dnorm for the probability density function, pnorm for the cumulative distribution function, qnorm for the quantile function, and rnorm for pseudo-random number generation. By default, these functions operate on the standard normal distribution with a mean of zero and a standard deviation of one, but they accept custom parameters for arbitrary normal distributions. Researchers use these functions across quantitative disciplines to model empirical measurement errors, evaluate hypothesis test statistics, generate simulated cohorts, and verify distributional assumptions through graphical diagnostic plots and formal omnibus normality tests.
What it is used for in research
In empirical scientific inquiry, the Gaussian (or normal) distribution serves as the primary parametric benchmark for continuous variables. Theoretical foundations rooted in the Central Limit Theorem establish that the sum or average of a large number of independent and identically distributed random variables tends toward a normal distribution, regardless of the underlying population distribution. Consequently, researchers routinely encounter Gaussian models across clinical trials, psychological assessments, biological assays, econometric forecasts, and physical measurement instrumentation.
Within statistical analysis workflows, Gaussian distribution functions in R are essential across several operational stages:
- Linear Model Residual Evaluation: Classical ordinary least squares (OLS) regression, analysis of variance (ANOVA), and linear mixed-effects models rely on the assumption that model residuals are independently and identically distributed following a Gaussian curve with mean zero and constant variance. Researchers evaluate these assumptions prior to drawing inferential conclusions using hypothesis testing with R.
- Probability and Critical Value Determination: Computing p-values, confidence interval bounds, and statistical power metrics requires evaluating cumulative probability distributions and inverse quantile cutoffs. While small samples often necessitate comparing test statistics against Student distributions, as sample sizes grow the t-distribution converges to the normal curve, as detailed in our t-distribution vs normal distribution comparison.
- Monte Carlo Simulation and Power Analysis: Synthetic cohort simulation enables researchers to assess experimental designs, evaluate statistical estimator bias under varying sample constraints, and project empirical power curves before embarking on resource-intensive laboratory data collection.
- Standardized Metric Transformations: Researchers convert raw empirical scores into standardized z-scores to compare disparate measurement instruments or establish standardized clinical reference percentiles across diverse study populations. The mathematical mechanics governing these density curves are further explored in our guide to the probability distribution function of the normal distribution.
- Multidimensional Extensions: In complex biomedical and psychometric studies with correlated endpoints, univariate Gaussian techniques form the direct mathematical foundation for modeling multivariate vectors, as demonstrated in our multivariate normal distribution guide.
How to install or access it
The core functions for modeling the Gaussian distribution—including dnorm, pnorm, qnorm, rnorm, shapiro.test, qqnorm, and qqline—are integral components of R’s standard stats package. Because stats is a pre-loaded base package maintained directly by the R Core Team, no third-party package installation or repository configuration is needed to execute standard normal distribution operations.
To install or update R itself from official distribution channels:
- Official CRAN Download: Obtain the current, verified release of R directly from the Comprehensive R Archive Network (CRAN) at
https://cran.r-project.org/. CRAN provides pre-compiled binary packages for macOS and Windows, as well as source tarballs and repository binaries for major Linux distributions (Debian, Ubuntu, Red Hat Enterprise Linux, Fedora, and SUSE). - Verifying Base Installation: Launch your preferred terminal or command prompt and enter
R --versionto verify that the R runtime environment is active and up to date. Once inside the interactive R console, the basestatspackage is automatically attached to the search path upon session startup. - Accessing Recommended Modeling Packages: While base R provides complete univariate Gaussian functionality, advanced distribution fitting and expanded omnibus normality testing benefit from two widely cited CRAN packages:
MASS(authored by Venables and Ripley, bundled with standard R distributions) for maximum likelihood parameter estimation, andnortest(authored by Gross and Ligges) for large-sample normality testing. These can be installed from CRAN within an active R session via:
# Install optional distribution fitting and normality testing packages
install.packages(c("MASS", "nortest"))
# Load packages into current session
library(MASS)
library(nortest)
Step-by-step instructions
This practical walkthrough demonstrates the end-to-end analytical workflow for working with the Gaussian distribution in R, encompassing pseudo-random data simulation, density evaluation, cumulative probability estimation, quantile calculation, parameter fitting, and normality verification.
Step 1: Set a reproducible seed and simulate Gaussian data with rnorm
The rnorm function generates pseudo-random variates drawn from a normal distribution with specified parameters. The function signature is rnorm(n, mean = 0, sd = 1), where n represents the integer number of observations to generate, mean is the distribution mean (μ), and sd is the standard deviation (σ). To ensure numerical reproducibility across research runs, always initialize R’s random number generator state with set.seed() prior to simulation:
# Step 1: Set reproducible seed and generate synthetic research cohort
set.seed(20261114)
n_obs <- 250
true_mean <- 100
true_sd <- 15
# Generate simulated observations (e.g., standardized cognitive scores)
sim_data <- rnorm(n = n_obs, mean = true_mean, sd = true_sd)
# Inspect summary statistics of generated data
summary(sim_data)
Step 2: Calculate probability density using dnorm
The dnorm function evaluates the Gaussian probability density function (PDF) for a vector of quantiles: dnorm(x, mean = 0, sd = 1, log = FALSE). The resulting output represents the height of the normal curve at value x. For continuous distributions, this value is a probability density rather than a discrete probability. When working with likelihood-based estimation or joint independent probabilities, set log = TRUE to compute log-densities directly, avoiding floating-point underflow:
# Step 2: Evaluate Gaussian density at specific points and along a continuum
# Density at the theoretical mean
density_center <- dnorm(x = true_mean, mean = true_mean, sd = true_sd)
print(density_center) # Height of the curve at mu
# Evaluate log-density to prevent arithmetic underflow in likelihood calculations
log_dens <- dnorm(x = true_mean, mean = true_mean, sd = true_sd, log = TRUE)
print(log_dens)
# Generate a continuous grid to plot the theoretical density curve
x_grid <- seq(from = true_mean - 3.5 * true_sd, to = true_mean + 3.5 * true_sd, length.out = 300)
y_density <- dnorm(x = x_grid, mean = true_mean, sd = true_sd)
Step 3: Compute cumulative probabilities using pnorm
The pnorm function computes the cumulative distribution function (CDF), defined as P(X ≤ q): pnorm(q, mean = 0, sd = 1, lower.tail = TRUE, log.p = FALSE). To evaluate the upper tail probability P(X > q), setting lower.tail = FALSE is mathematically superior to computing 1 - pnorm(q) because it maintains high precision in extreme tails without catastrophic cancelation. Detailed applications of this parameter are described in our guide on cumulative probability calculations with pnorm in R.
# Step 3: Compute cumulative and tail probabilities
# Probability that an observation falls below 85 (one standard deviation below mean)
p_below_85 <- pnorm(q = 85, mean = true_mean, sd = true_sd, lower.tail = TRUE)
print(p_below_85) # Approximately 0.1586553
# Upper tail: Probability that an observation exceeds 130 (two standard deviations above mean)
p_above_130 <- pnorm(q = 130, mean = true_mean, sd = true_sd, lower.tail = FALSE)
print(p_above_130) # Approximately 0.02275013
# Probability of an observation falling within one standard deviation [85, 115]
p_within_1sd <- pnorm(115, true_mean, true_sd) - pnorm(85, true_mean, true_sd)
print(p_within_1sd) # Approximately 0.6826895 (68.27%)
Step 4: Calculate distribution quantiles with qnorm
The qnorm function provides the inverse cumulative distribution function (quantile function): qnorm(p, mean = 0, sd = 1, lower.tail = TRUE, log.p = FALSE). Given a cumulative probability p between 0 and 1, qnorm returns the threshold value x such that P(X ≤ x) = p. This function is essential for establishing two-sided 95% confidence limits and experimental rejection regions:
# Step 4: Determine critical values and percentiles
# 95th percentile cutoff for a standard normal distribution
z_95 <- qnorm(p = 0.95, mean = 0, sd = 1)
print(z_95) # 1.644854
# Critical z-values for a two-tailed alpha = 0.05 test (2.5% and 97.5% cutoffs)
crit_lower <- qnorm(0.025, mean = 0, sd = 1)
crit_upper <- qnorm(0.975, mean = 0, sd = 1)
print(c(crit_lower, crit_upper)) # -1.959964 and +1.959964
# Interquartile range (25th and 75th percentiles) for the simulated cohort
iqr_bounds <- qnorm(c(0.25, 0.75), mean = true_mean, sd = true_sd)
print(iqr_bounds)
Step 5: Fit Gaussian parameters using MASS::fitdistr
When working with empirical datasets, researchers must estimate distribution parameters from observed data. In base R, sample mean and sample standard deviation are calculated via mean() and sd(). Alternatively, the fitdistr function in the MASS package fits univariate distributions by maximum likelihood estimation (MLE):
# Step 5: Fit parameters using closed-form sample moments and MASS::fitdistr
library(MASS)
# Analytical sample estimates
sample_mean <- mean(sim_data)
sample_sd <- sd(sim_data)
# Maximum likelihood estimation via fitdistr
fit_result <- fitdistr(sim_data, densfun = "normal")
print(fit_result)
Step 6: Diagnostic visual inspection with qqnorm and qqline
Visual inspection using normal quantile-quantile (Q-Q) plots is the established scientific standard for evaluating Gaussian assumptions. The function qqnorm() plots the sample quantiles against theoretical standard normal quantiles, while qqline() draws a reference line passing through the first and third quartiles (25th and 75th percentiles):
# Step 6: Construct Q-Q plot and empirical histogram
# Generate Q-Q plot with quartile reference line
qqnorm(sim_data, main = "Normal Q-Q Plot of Simulated Data", pch = 19, col = "darkblue")
qqline(sim_data, col = "firebrick", lwd = 2)
# Overlay theoretical density on sample histogram
hist(sim_data, prob = TRUE, breaks = 20, col = "gray92", border = "gray40",
main = "Histogram and Gaussian Density", xlab = "Value")
curve(dnorm(x, mean = mean(sim_data), sd = sd(sim_data)), add = TRUE, col = "blue", lwd = 2)
Step 7: Conduct formal omnibus normality tests
To supplement visual diagnostics with numerical hypothesis tests, base R provides the Shapiro-Wilk test via shapiro.test(). For sample sizes exceeding the 5,000-observation limitation of the Shapiro-Wilk algorithm, researchers utilize the Anderson-Darling test from the nortest package:
# Step 7: Omnibus normality testing
# Shapiro-Wilk test (valid for sample sizes 3 <= N <= 5000)
sw_test 5000)
library(nortest)
ad_test <- ad.test(sim_data)
print(ad_test)
Output and how to read it
Correct interpretation of R’s statistical outputs is vital to avoid misleading conclusions regarding distribution shape and statistical significance. Below is a structured guide to reading the output objects generated by each core routine:
- Reading dnorm Output: Returns a numeric vector of density values f(x). For example, evaluating
dnorm(0)yields0.3989423, which equals 1/√(2π). Because density represents the derivative of the cumulative distribution function, values of f(x) can exceed 1.0 when the standard deviation is less than 1/√(2π) ≈ 0.3989. It must never be interpreted as an individual probability. - Reading pnorm Output: Returns a bounded probability between 0 and 1 representing the area under the density curve. An output of
0.9750021forpnorm(1.96)indicates that 97.5% of the distribution lies at or below 1.96 standard deviations above the mean. Whenlower.tail = FALSEis used, an output of0.0249979indicates the complementary upper tail probability. - Reading qnorm Output: Returns a numeric value along the measurement scale corresponding to the given cumulative percentile. An output of
qnorm(0.50, mean = 100, sd = 15)yields exactly100, confirming that the mean equals the median in a symmetric Gaussian distribution. - Reading MASS::fitdistr Output: Prints the parameter estimates and estimated standard errors in parentheses below each parameter:
mean sd
100.84120 14.89231
( 0.94187) ( 0.66599)
The first row displays maximum likelihood point estimates (μ̂ and σ̂), while the second row provides their analytical standard errors. The ratio of point estimate to standard error allows computation of Wald confidence intervals. Note that fitdistr computes the maximum likelihood estimate of standard deviation using a divisor of N rather than N – 1, producing a slight downward adjustment compared to base R’s sd() function.
- Reading shapiro.test Output: Prints an object of S3 class
htestcontaining two primary values: the test statisticWand thep-value:
Shapiro-Wilk normality test
data: sim_data
W = 0.99341, p-value = 0.3128
The null hypothesis (H₀) of the Shapiro-Wilk test posits that the data are normally distributed. The test statistic W ranges between 0 and 1, with values close to 1 indicating strong agreement with normality. When the p-value exceeds the selected significance threshold (typically α = 0.05), researchers fail to reject the null hypothesis, concluding that the data exhibit no statistically detectable departure from normality. Conversely, a p-value less than or equal to 0.05 indicates statistically significant non-normality.
- Interpreting Q-Q Plot Morphology: When examining plots generated by
qqnorm()andqqline(), points conforming tightly to the diagonal line indicate empirical normality. Deviations in specific plot regions reveal distinct distributional anomalies:
- S-shaped curve with points below the line at the left and above at the right: Indicates heavy tails (leptokurtosis), where extreme outliers occur more frequently than predicted by a normal distribution.
- Inverted S-shaped curve: Indicates light tails (platykurtosis), where data are more tightly clustered with fewer outliers than a Gaussian model.
- Upward convex curve: Indicates positive (right) skewness, where the right tail is prolonged.
- Downward concave curve: Indicates negative (left) skewness, where the left tail is prolonged.
Troubleshooting
When working with Gaussian distributions in R, researchers frequently encounter specific warnings, numerical errors, or algorithmic boundary issues. Below are verified diagnostic solutions for the most common technical hurdles:
- Warning: NaNs produced with Negative Standard Deviation: In mathematical statistics, standard deviation is strictly non-negative (σ ≥ 0). If you pass a negative value to the
sdargument indnorm,pnorm,qnorm, orrnorm, R returnsNaNand emitsWarning message: In ... : NaNs produced. To resolve this, wrap parameter inputs inabs()or add validation assertions checking thatall(sd >= 0)before executing distribution functions. Note thatsd = 0is supported as a limiting delta-distribution where all mass resides at the mean. - Error: sample size must be between 3 and 5000 in shapiro.test: Base R’s
shapiro.testrelies on Patrick Royston’s AS R94 polynomial approximation algorithm, which was mathematically parameterized specifically for sample sizes between 3 and 5,000 observations. Passing a vector with fewer than 3 or more than 5,000 non-missing elements triggers a fatal error. For large-sample research datasets (N > 5,000), replaceshapiro.testwith the Anderson-Darling test (nortest::ad.test) or the Lilliefors test (nortest::lillie.test), neither of which imposes an upper sample size limit. - Numerical Underflow in Extreme Tails: Evaluating extreme probabilities (e.g., z-scores where |z| > 6) using standard linear arithmetic can cause values to round to 0 or 1 due to IEEE 754 double-precision limitations. For example, computing
1 - pnorm(38)evaluates to 0. To obtain exact numerical results, uselower.tail = FALSEcombined withlog.p = TRUE:pnorm(38, lower.tail = FALSE, log.p = TRUE)returns-725.6881, representing log(P), which preserves numerical stability without precision loss. - Vector Recycling Pitfalls: R functions automatically recycle parameter vectors whose lengths do not match the input vector length. If you pass a vector of quantiles
x <- 1:10and supply two meansmean = c(0, 10), R alternates means across indices without error. However, if the longer vector length is not a clean multiple of the shorter vector length, R emits a warning:longer object length is not a multiple of shorter object length. Always verify vector lengths or usemapply()when evaluating heterogeneous distributions. - Unintended Propagation of Missing Values (NA): The presence of a single
NAin your empirical vector causesshapiro.test(),mean(), andsd()to return errors orNA. Prior to normality evaluation or distribution fitting, sanitize data vectors usingsim_data[!is.na(sim_data)]orna.omit(sim_data).
Free alternatives
While base R is the academic gold standard for statistical computing, quantitative researchers frequently operate across diverse open-source computational environments. The comparison table below highlights verified free and open-source alternatives for Gaussian distribution modeling, comparing programming interfaces, normality testing capabilities, and optimal research use cases:
| Software Environment | License & Access | Core Gaussian Functions | Normality Testing Support | Primary Research Strength |
|---|---|---|---|---|
| R (base stats) | GNU GPL v2/v3 (Free & Open Source) | dnorm, pnorm, qnorm, rnorm | shapiro.test, qqnorm, qqline, ks.test | Native statistical syntax, unmatched bioinformatic and academic package ecosystem |
| Python (SciPy & NumPy) | BSD 3-Clause (Free & Open Source) | scipy.stats.norm (pdf, cdf, ppf, rvs) | scipy.stats.shapiro, normaltest, probplot | Seamless integration with machine learning pipelines, deep learning, and web production APIs |
| GNU Octave | GNU GPL v3 (Free & Open Source) | normpdf, normcdf, norminv, normrnd | kolmogorov_smirnov_test, qqplot | Drop-in open-source script compatibility with proprietary MATLAB numerical workflows |
| Julia (Distributions.jl) | MIT License (Free & Open Source) | Normal(), pdf(), cdf(), quantile(), rand() | HypothesisTests.jl (ExactOneSampleKSTest) | High-performance just-in-time compilation for large-scale Monte Carlo and numerical simulations |
| JASP | GNU GPL v3 (Free & Open Source) | GUI-driven probability distribution calculators | Automated Shapiro-Wilk and Q-Q diagnostics | Accessible graphical user interface offering both frequentist and Bayesian parameter analysis |
| Jamovi | GNU GPL v3 (Free & Open Source) | GUI-driven statistical modules backed by R | Built-in normality assumption checks and Q-Q plots | Spreadsheet-style interface ideal for teaching, clinical research, and exploratory data analysis |
As summarized in the data table below, R provides specialized algorithmic precision (including Cody and Wichura implementations) that ensures consistent numerical behavior across base statistical distributions.
Frequently asked questions
What is the difference between dnorm, pnorm, qnorm, and rnorm in R?
In R, the four functions represent distinct mathematical operations on the Gaussian distribution. The function dnorm calculates the probability density function (PDF), returning the height of the curve at a given point. The function pnorm evaluates the cumulative distribution function (CDF), providing the probability that a random variable is less than or equal to a specified cutoff. The function qnorm is the quantile function (inverse CDF), returning the value corresponding to a designated cumulative percentile. Finally, rnorm generates pseudo-random variates drawn from a normal distribution with the specified mean and standard deviation.
How do I plot a normal distribution curve over a histogram in R?
To overlay a theoretical Gaussian curve on an empirical histogram in base R, first generate the histogram with probability density scaling by setting prob = TRUE in hist(data, prob = TRUE). Then, use the curve() function with add = TRUE to draw the theoretical density: curve(dnorm(x, mean = mean(data), sd = sd(data)), add = TRUE, col = “blue”, lwd = 2). Setting prob = TRUE scales the histogram bars by probability density rather than raw frequency counts, ensuring that the histogram area matches the area under the dnorm curve.
Why does shapiro.test produce an error for large datasets in R?
Base R’s shapiro.test function relies on Royston’s AS R94 algorithm, which restricts sample sizes to between 3 and 5,000 observations. For datasets larger than 5,000 rows, R halts execution with an error indicating that the sample size must be between 3 and 5000. For large research cohorts, analysts should apply the Anderson-Darling test via ad.test() or the Lilliefors test via lillie.test() from the nortest package, neither of which imposes an upper sample size limit.
How do I test if my data follows a normal distribution in R?
A rigorous normality assessment in R combines graphical diagnostics with numerical hypothesis tests. First, create a normal quantile-quantile plot using qqnorm(data) and overlay the quartile reference line with qqline(data). If the plotted points adhere closely to the straight diagonal line, the data conform to Gaussian expectations. Next, perform a formal hypothesis test using shapiro.test(data) for sample sizes up to 5,000, or ad.test(data) from the nortest package for larger samples. A p-value greater than 0.05 indicates failure to reject the null hypothesis of normality.
How does R handle extreme tail probabilities for the Gaussian distribution?
For extreme tail quantiles where standard linear probabilities risk floating-point underflow to zero, R provides two specialized logical parameters in pnorm and qnorm: lower.tail and log.p. Setting lower.tail = FALSE computes the upper tail probability P(X > q) directly using specialized asymptotic expansions rather than evaluating 1 – P(X ≤ q), avoiding catastrophic cancelation. In addition, setting log.p = TRUE returns the natural logarithm of the probability, allowing accurate representation of extreme tail probabilities across high-magnitude z-scores.
What is the difference between sd() and the standard deviation estimate in MASS::fitdistr()?
Base R’s sd() function computes the unbiased sample standard deviation using N – 1 in the denominator, which corrects for sample bias in estimating population variance. In contrast, the MASS::fitdistr() function computes the maximum likelihood estimate (MLE) of the standard deviation, which divides the sum of squared deviations by N rather than N – 1. For small sample sizes, the MLE standard deviation is slightly smaller than the sample standard deviation, though both estimators asymptotically converge as sample size N approaches infinity.
Sources
- R Core Team. (2024). The Normal Distribution. R Documentation (stats package). https://stat.ethz.ch/R-manual/R-patched/library/stats/html/Normal.html (accessed 2026-11-14).
- R Core Team. (2024). Shapiro-Wilk Normality Test. R Documentation (stats package). https://stat.ethz.ch/R-manual/R-patched/library/stats/html/shapiro.test.html (accessed 2026-11-14).
- R Core Team. (2024). Quantile-Quantile Plots. R Documentation (stats package). https://stat.ethz.ch/R-manual/R-patched/library/stats/html/qqnorm.html (accessed 2026-11-14).
- R Core Team. (2024). Kolmogorov-Smirnov Tests. R Documentation (stats package). https://stat.ethz.ch/R-manual/R-patched/library/stats/html/ks.test.html (accessed 2026-11-14).
- Venables, W. N., & Ripley, B. D. (2024). fitdistr: Maximum-Likelihood Fitting of Univariate Distributions. MASS Documentation. https://stat.ethz.ch/R-manual/R-patched/library/MASS/html/fitdistr.html (accessed 2026-11-14).
- Gross, J., & Ligges, U. (2024). nortest: Tests for Normality. Comprehensive R Archive Network (CRAN). https://cran.r-project.org/web/packages/nortest/index.html (accessed 2026-11-14).
- R Core Team. (2024). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. https://www.r-project.org/ (accessed 2026-11-14).
Data: R Gaussian Distribution Functions and Normality Diagnostics
Technical specifications, mathematical definitions, default arguments, and computational constraints for Gaussian distribution modeling and normality assessment functions in R base stats, MASS, and nortest packages.
Period covered: 2024/2026. Coverage: Global.
| Function Name | Operation Type | Mathematical Target | Default Parameters | Core Algorithm | Sample Size Limit (observations) |
|---|---|---|---|---|---|
| dnorm (stats) | Probability Density Function | f(x) density height | mean = 0, sd = 1, log = FALSE | Direct evaluation of Gaussian PDF formula | Any finite numeric vector |
| pnorm (stats) | Cumulative Distribution Function | P(X | mean = 0, sd = 1, lower.tail = TRUE, log.p = FALSE | Cody (1993) rational Chebyshev approximation | Any finite numeric vector |
| qnorm (stats) | Quantile Function | x_p such that P(X | mean = 0, sd = 1, lower.tail = TRUE, log.p = FALSE | Wichura (1988) AS 241 minimax polynomial | Probabilities in [0, 1] |
| rnorm (stats) | Random Deviate Generation | Pseudo-random normal sample | mean = 0, sd = 1 | Mersenne-Twister / Inversion or Box-Muller | n >= 0 integer |
| shapiro.test (stats) | Omnibus Normality Test | Shapiro-Wilk W test statistic | x (numeric vector) | Royston (1982, 1995) AS R94 approximation | 3 to 5000 non-missing |
| qqnorm (stats) | Quantile-Quantile Diagnostic | Empirical vs theoretical quantiles | y (numeric vector), ylim, main | Sample quantiles against qnorm percentiles | n >= 2 non-missing |
| qqline (stats) | Q-Q Reference Line | Line through 1st and 3rd quartiles | probs = c(0.25, 0.75), qtype = 7 | Linear interpolation between quartiles | n >= 2 non-missing |
| ks.test (stats) | Goodness-of-Fit Test | Kolmogorov-Smirnov supremum distance D | alternative = 'two.sided', exact = NULL | Empirical vs theoretical CDF comparison | n >= 1 non-missing |
| fitdistr (MASS) | Maximum Likelihood Estimation | Point estimates of mu and sigma | densfun = 'normal' | Closed-form analytical MLE with standard errors | n >= 2 finite values |
| ad.test (nortest) | Empirical Distribution Function Test | Anderson-Darling A statistic | x (numeric vector) | Stephens (1974) EDF weighted tail statistic | n >= 8 non-missing |
| lillie.test (nortest) | Composite Normality Test | Lilliefors corrected D statistic | x (numeric vector) | Dallal and Wilkinson (1986) approximation | n >= 5 non-missing |
Sources
- R Core Team - The Normal Distribution Documentation (accessed 2026-11-14)
- R Core Team - Shapiro-Wilk Normality Test Documentation (accessed 2026-11-14)
- R Core Team - Quantile-Quantile Plots Documentation (accessed 2026-11-14)
- Venables and Ripley - MASS fitdistr Documentation (accessed 2026-11-14)
- Gross and Ligges - nortest: Tests for Normality Documentation (accessed 2026-11-14)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







