Skip to main content
v2026.11,986 entries · CC-BY 4.0

How to Detect and Handle Outliers in GraphPad Prism

Learn how to detect and clean outliers in GraphPad Prism using the ROUT method and Grubbs test, with step-by-step column analysis and output interpretation.

Written and maintained by CASRAI Editorial Board

Published

On this page

To identify and handle an outlier in GraphPad Prism, enter your dataset into a Column or XY data table, select Analyze, and choose Identify Outliers from the Column Analyses menu, or enable automatic outlier elimination within the Nonlinear Regression dialog. Prism identifies outliers using either the ROUT method (Robust regression and Outlier removal, based on False Discovery Rate coefficient Q) for single or multiple outliers, or the two-sided Grubbs’ Extreme Studentized Deviate (ESD) test (significance level α) for a single outlier in a Gaussian distribution. Once analyzed, Prism generates non-destructive output sheets containing cleaned data with outliers excluded, an outliers-only table, and an analysis summary without altering your original raw data.

What GraphPad Prism is used for in research

GraphPad Prism is a biostatistical and scientific graphing application tailored for pharmacology, physiology, cell biology, and preclinical investigations. Unlike general spreadsheets or code-based pipelines, Prism couples laboratory data entry directly with automated nonlinear curve fitting and parametric tests. Biomedical researchers routinely use Prism statistics to analyze replicate bioassays, construct dose-response models, evaluate enzyme kinetics, and perform comparative hypothesis testing.

In laboratory assays, anomalous values frequently appear due to distinct technical and biological causes:

  • Technical execution errors: Pipetting volume inaccuracies, microplate bubbles, partial fluidic clogs in flow cytometers, or optical detector saturation.
  • Biological variation: True physiological diversity, drug-resistant cellular subsets, or variable metabolic phenotypes.
  • Clerical mistakes: Transcription mistakes or mislabeled specimen tubes during assay transfers.

Informal visual inspection—deciding to delete data points by eye—introduces confirmation bias and inflates false positive findings. Prism addresses this vulnerability by applying formalized statistical algorithms with transparent decision criteria. Crucially, Prism preserves full data provenance: raw data tables remain untouched, while identified outliers can be flagged with blue strike-through text, excluding them from downstream procedures such as an unpaired t test while maintaining complete audit trails.

How to install or access GraphPad Prism

GraphPad Prism is commercial software developed by GraphPad Software (Dotmatics) and must be installed through official distribution channels. Prism operates as a local desktop program on Windows and macOS, executing computations locally without transferring experimental data to external servers.

Operating system hardware specifications include:

  • Windows: 64-bit editions of Windows 11 and Windows 10, requiring 2 GB RAM and 100 MB disk space.
  • macOS: macOS 12 (Monterey), macOS 13 (Ventura), macOS 14 (Sonoma), and macOS 15 (Sequoia), running natively on Apple Silicon (M-series) and Intel processors.

To install Prism from the official distributor:

  1. Navigate to the official portal (graphpad.com) or your university licensing repository.
  2. Download the official platform installer: the .msi installer for Windows or the .dmg disk image for macOS.
  3. Execute the installer and follow setup wizard prompts to install the binaries.
  4. Launch Prism and activate your license using an institutional serial number, named-user Prism Cloud credentials, or a machine activation key. An unrestricted 30-day trial is available directly from the official portal.

Researchers needing single-column outlier verification without desktop software can access GraphPad QuickCalcs (specifically the Outlier Calculator at graphpad.com/quickcalcs/Grubbs1.cfm). This free browser utility runs Grubbs’ ESD test for continuous datasets up to 2,000 values. Investigators can also utilize the GraphPad P value calculator to examine probability boundaries.

How to identify outliers in GraphPad Prism step by step

Prism provides two analytical routes for outlier detection: column analyses for discrete groups and nonlinear regression for curve fitting.

Step-by-step outlier identification in column data tables

To detect outliers across grouped numerical columns, follow these numbered steps:

  1. Format data table: Create a new project, select Column from table types, and click Create. Enter numerical values into column stacks. Each column must contain at least three values, representing Prism’s mathematical floor for estimating variance.
  2. Open analysis selection: Click the Analyze button in the main toolbar, or select Insert → New Analysis.
  3. Select outlier module: Under Column analyses, select Identify outliers. Confirm target columns in the right pane, then click OK.
  4. Configure detection parameters: In the Parameters dialog, choose your algorithm:
    • ROUT method: Select ROUT (Robust regression and OUTlier removal). Enter coefficient Q (default 1.0%), specifying the maximum desired False Discovery Rate. Choose Q = 0.1% for stricter detection, or Q = 5% for broader screening.
    • Grubbs’ method: Select Grubbs’ method (ESD). Specify two-sided significance level alpha (typically α = 0.05 or α = 0.01). Select this method only when theoretical considerations ensure at most one outlier exists per group.
    • Iterative Grubbs’ method: Official GraphPad guidance discourages iterative Grubbs testing due to outlier masking in small cohorts.
  5. Generate results: Click OK. Prism computes residuals and outputs three sheets: Cleaned data, Outliers only, and Summary.

Step-by-step outlier detection in nonlinear regression curve fitting

In dose-response assays and binding kinetics, aberrant points distort curve parameters. Prism embeds the ROUT algorithm directly into nonlinear regression:

  1. Format XY table: Enter the independent variable in column X and response replicates in Y subcolumns.
  2. Launch regression: Click Analyze and select Nonlinear regression (curve fit) under XY analyses.
  3. Select curve model: In the Model tab, pick your mathematical function (such as [Four-Parameter Logistic ELISA curve fitting](/guides/four-parameter-logistic-elisa-standard-curve-fitting) or sigmoidal response).
  4. Enable outlier elimination: In the Model tab, check Automatic outlier elimination (ROUT).
  5. Define FDR coefficient: In the Method tab, verify ROUT coefficient Q (default 1%). Prism fits a robust initial curve where outliers exert minimal pull, calculates residuals, flags outliers using FDR threshold Q, excludes them, and completes standard least-squares regression on the remaining data.
  6. Inspect curve output: Click OK. Prism renders the fitted curve across valid points, highlights eliminated outliers with red markers, and reports the outlier count in the results table header.

Step-by-step exclusion of outliers in the raw data table

Prism does not overwrite values in your original data table. To remove an identified outlier from downstream statistical comparisons:

  1. Review the Outliers only sheet to identify the row index and column position of the flagged observation.
  2. Navigate to your primary raw data table in the project navigator.
  3. Click on the cell containing the outlier value.
  4. Right-click and select Exclude Values, or press Ctrl+E (Windows) or Cmd+E (macOS).
  5. The value displays with blue strike-through formatting, remaining visible for laboratory record-keeping but omitted from linked graphs and hypothesis tests.

How to read the output

Prism organizes column outlier results across three dedicated output tabs within the Results section:

1. Cleaned data (outliers removed)

The Cleaned data sheet reproduces your original table layout with all flagged outlier cells left blank. This sheet serves as direct input for subsequent analyses. Clicking Analyze from this sheet enables running an ANOVA or t test on the outlier-depleted dataset without manually copying values or deleting experimental records.

2. Outliers only

The Outliers only sheet displays a grid containing only the values flagged as statistical outliers, positioned in their respective row and column coordinates. All non-outlying cells are left empty, enabling researchers to check whether outliers cluster in specific assay plates or cohorts, signaling systematic artifacts.

3. Summary sheet

The Summary sheet reports quantitative parameters for each analyzed dataset:

  • Number of values (N): Total count of numerical observations present prior to analysis.
  • Number of outliers: Total count of data points meeting the outlier threshold.
  • Threshold setting: The active ROUT coefficient Q percentage or Grubbs’ alpha value.
  • Critical values: For Grubbs’ test, the summary displays the calculated Grubbs Z statistic and critical Z cutoff.

Mathematical interpretation of detection statistics

Grubbs’ ESD Test: The Grubbs statistic Z measures the distance of the most extreme value from the sample mean, scaled by the sample standard deviation:

Z = |Y_i – Y_mean| / s

where Y_i is the suspected observation, Y_mean is the sample mean, and s is the sample standard deviation across all N values. Prism compares Z against critical threshold Z_crit derived from the Student’s t distribution with N – 2 degrees of freedom at two-sided significance level α / (2N). In Prism, specified α (e.g., 0.05) represents the probability of falsely identifying an outlier when all data are sampled from a Gaussian distribution, applying to the entire dataset as a whole.

ROUT Method: Developed by Harvey Motulsky and Ronald Brown (2006), ROUT combines robust nonlinear regression with the False Discovery Rate (FDR). When testing a column of numbers, Prism fits horizontal model Y = M, where M is a robust mean calculated by minimizing Huber-like loss functions. Residuals are normalized by a robust scatter estimate (RSDR, the 68.27th percentile of absolute residuals) and evaluated using an adaptive Benjamini-Hochberg FDR procedure.

Parameter Q represents the maximum allowable False Discovery Rate. Setting Q = 1% means that no more than 1% of identified outliers are expected to be false discoveries (chance tail observations from a Gaussian distribution), ensuring that at least 99% represent genuine anomalies from an external distribution.

Troubleshooting GraphPad Prism

When applying outlier detection in laboratory workflows, consider these technical and statistical factors:

Outlier masking in small sample datasets

A frequent error is attempting to detect multiple outliers by repeatedly running Grubbs’ test sequentially (Iterative Grubbs). Grubbs’ test assumes at most one outlier departs from a Gaussian distribution. When two outliers sit on the same side of the mean in a small sample, their combined magnitude inflates the sample standard deviation s. Because s appears in the denominator of the Grubbs ratio, this inflation depresses calculated Z below critical cutoff Z_crit, causing both outliers to be missed. This failure mode is termed masking.

In Monte Carlo simulations conducted by Motulsky and Brown (2006) on small datasets (n = 10) containing two added outliers, Grubbs’ test failed completely: it missed both outliers in 98.8% of simulations, identified only one outlier in 1.2%, and never identified both. Conversely, the ROUT method (Q = 1%) successfully identified both outliers in 92.8% of simulations and missed both in only 6.0%. Consequently, whenever experimental datasets may contain two or more outliers, select the ROUT method rather than iterative Grubbs testing.

False positives from lognormal and skewed distributions

Both Grubbs’ test and the ROUT method assume uncontaminated data follow a Gaussian distribution. In biological research, many experimental variables follow lognormal distributions characterized by heavy right tails, including serum antibody titers, cytokine concentrations, viral copy numbers, and qPCR fold changes.

Applying Grubbs’ test or ROUT directly to raw lognormal values will cause valid high-concentration replicates to be flagged as outliers simply because they depart from a symmetrical bell curve. Discarding these points artificially distorts the dataset mean. To troubleshoot this issue, log-transform data (using Prism’s Transform analysis with Y = Log(Y)) prior to running outlier identification. If log-transformed values conform to a Gaussian distribution, evaluate outliers on the logarithmic scale.

Unequal weighting conflicts in robust nonlinear regression

In binding assays and kinetics, experimental variance often scales with signal magnitude, leading researchers to apply weighting schemes such as 1/Y² or 1/Y. However, robust regression already applies dynamic residual weighting, creating a mathematical conflict with user-specified weights.

When you select both unequal weighting and automatic outlier elimination in Prism’s Nonlinear Regression dialog, Prism resolves the conflict in two stages: it first fits the model using robust regression with equal weights to identify outliers; after eliminating confirmed outliers, it performs standard weighted least-squares regression on the remaining data using your specified weighting model.

Scientific evaluation vs. mechanical data deletion

Outlier algorithms identify mathematical deviation, not experimental culpability. Before excluding an outlier, conduct a laboratory review:

  • Examine laboratory records for documented spills, air bubbles, or pipetting alerts.
  • Review raw instrument data files for detector saturation flags.
  • Assess whether the extreme value reflects authentic biological variation, such as a drug-resistant cellular sub-clone.
  • Never remove an outlier purely to alter a p-value or force statistical significance. For foundational statistical rules, see Casrai guidance on how to find an outlier in statistics.

Free alternatives to GraphPad Prism

While GraphPad Prism is widely adopted in biomedical laboratories, several free, open-source computational environments provide equivalent outlier detection capabilities. The data table below compares statistical outlier detection methods across scientific software and reference standards.

Software Platform Supported Detection Methods Primary Package / Function Outlier Handling Mode Workflow Interface
R (CRAN Ecosystem) Grubbs ESD, Dixon Q, Rosner Many-Outlier ESD, ROUT equivalent, Boxplot IQR outliers::grubbs.test(), EnvStats::rosnerTest(), boxplot.stats() Programmatic vector indexing and logical subsetting Script-based (RStudio / Posit)
Python (SciPy / Statsmodels) Z-score studentization, Chauvenet criterion, Huber robust regression, RANSAC scipy.stats.zscore(), statsmodels.stats.outliers_influence NumPy boolean masking and Pandas DataFrame query filtering Script-based / Jupyter Notebooks
Jamovi Exploration boxplots (Tukey IQR), studentized residuals in linear models Exploration module / Descriptives / GAMLj module Data filter conditions (dynamic row subsetting) Interactive GUI (R engine backend)
JASP Boxplot IQR, Mahalanobis distance, Cook’s distance, robust regression Descriptives / Regression diagnostics / Robust statistics modules Dynamic column filtering and spreadsheet row exclusion Interactive GUI (Univ. Amsterdam)
GraphPad QuickCalcs (Online) Two-sided Grubbs’ ESD test (single outlier) Web browser calculation form (up to 2,000 values) Calculates Z ratio and p-value; manual exclusion by user Zero-install browser interface

R Programming Environment

The open-source R statistical environment provides specialized packages for outlier screening. To perform Grubbs’ test in R, install and load the outliers package:

library(outliers)
# Run two-sided Grubbs test on sample data vector
data_vector <- c(12.1, 12.4, 12.8, 12.2, 12.5, 28.4)
grubbs.test(data_vector, type = 10, two.sided = TRUE)

When multiple outliers threaten masking, R provides Rosner’s generalized extreme studentized deviate test via the EnvStats package (rosnerTest(data_vector, k = 3)), where users declare an upper bound k of potential outliers. In addition, the robust regression principles underlying Prism’s ROUT method can be mirrored using M-estimators in the MASS::rlm() package.

Python Scientific Computing Stack

In Python, the SciPy and Statsmodels ecosystem provides automated outlier identification within data analysis pipelines. Robust regression that minimizes outlier pull can be implemented using Huber regression in scikit-learn:

import numpy as np
from sklearn.linear_model import HuberRegressor

# Example robust model fit minimizing outlier leverage
X = np.array([1, 2, 3, 4, 5, 6, 7, 8]).reshape(-1, 1)
y = np.array([2.1, 3.9, 6.2, 8.1, 10.0, 12.2, 14.1, 35.0]) # 35.0 is an outlier
huber = HuberRegressor().fit(X, y)
residuals = np.abs(y - huber.predict(X))
threshold = 3.0 * np.median(residuals)
outliers = np.where(residuals > threshold)[0]

Jamovi and JASP Graphical Environments

For investigators seeking an interactive visual interface without writing code, Jamovi and JASP provide free, open-source alternatives. Both execute on Windows, macOS, and Linux desktop systems. Jamovi generates boxplots using Tukey’s interquartile range rule, which flags points exceeding 1.5 times the IQR from the first or third quartile. In JASP, linear model regression diagnostics automatically compute Cook’s distance and studentized residuals to identify influential data points.

Frequently asked questions

How do I identify outliers in GraphPad Prism?

In a Column table, click Analyze on the toolbar and select Identify Outliers from the Column Analyses list. Choose between the ROUT method (setting coefficient Q) or Grubbs’ ESD test (setting alpha), and click OK. For curve fitting, check Automatic Outlier Elimination directly on the Model tab of the Nonlinear Regression dialog.

What outlier methods does GraphPad use?

Prism primarily uses the ROUT method and Grubbs’ ESD test. The ROUT method combines robust nonlinear regression with False Discovery Rate testing, identifying single or multiple outliers across columns and curves. Grubbs’ test identifies a single extreme value from an assumed Gaussian distribution.

What is the ROUT method in GraphPad?

The ROUT method (Robust regression and Outlier removal) is an algorithm developed by Harvey Motulsky and Ronald Brown. It fits data using robust regression where outliers exert minimal leverage, calculates residuals, and evaluates them against False Discovery Rate threshold Q (default 1%) before final least-squares fitting.

How do I remove or exclude an outlier in Prism?

To exclude an outlier without deleting the raw record, select the cell in your raw data table, right-click, and choose Exclude Values (Ctrl+E on Windows, Cmd+E on macOS). The value appears with blue strike-through text and is automatically excluded from linked graphs, summaries, and statistical tests.

When should you remove an outlier?

Remove an outlier only when verified experimental, technical, or clerical error exists, such as documented pipetting failure, microplate bubbles, or instrument error. Outliers should never be removed arbitrarily simply to improve p-values or force statistical significance.

Sources

Data: Statistical Outlier Detection Methods in Laboratory and Clinical Research Software

Comparative benchmark of statistical outlier detection techniques across laboratory data analysis workflows, compiled from GraphPad Prism 11 documentation, NIST Handbook 1.3.5.17, and Motulsky and Brown (2006) simulation parameters.

Statistical Outlier Detection Methods in Laboratory and Clinical Research Software
Method NameSoftware ImplementationAssumed Data DistributionOutlier Detection CapacityDetection Threshold ParameterMinimum Sample Size (observations)Masking ResistancePrimary Standard or Citation
ROUT Method (Q=1%)GraphPad Prism 11Gaussian with potential outliersMultiple (up to 30% of sample)False Discovery Rate Q = 1%3High (92.8% recovery for n=10, 2 outliers)Motulsky & Brown (2006)
ROUT Method (Q=5%)GraphPad Prism 11Gaussian with potential outliersMultiple (up to 30% of sample)False Discovery Rate Q = 5%3High (robust regression residual thresholding)Motulsky & Brown (2006)
Grubbs Test (ESD, alpha=0.05)GraphPad Prism & QuickCalcsStrictly GaussianSingle outlier onlyTwo-sided alpha = 0.053None (fails in multiple outlier scenarios)NIST Handbook 1.3.5.17 / Grubbs (1969)
Grubbs Test (ESD, alpha=0.01)GraphPad Prism & QuickCalcsStrictly GaussianSingle outlier onlyTwo-sided alpha = 0.013None (fails in multiple outlier scenarios)NIST Handbook 1.3.5.17 / Grubbs (1969)
Iterative Grubbs TestPrism (Discouraged) / R outliersStrictly GaussianMultiple (sequential)Sequential alpha = 0.053Very Low (98.8% miss rate for n=10, 2 outliers)GraphPad FAQ 1600
Tukey IQR Boxplot RuleR stats / Prism BoxplotsNonparametric / Distribution-freeMultiple (upper and lower tails)1.5x Interquartile Range (IQR)4Moderate (robust median and quartiles)Tukey (1977) EDA
Rosner Generalized ESDR EnvStats / NIST GuidelinesApproximately GaussianMultiple (user specifies k max)Per-test alpha = 0.0510High (immune to masking up to k outliers)Rosner (1983) Technometrics

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · full access

Ask about How to Detect and Handle Outliers in GraphPad Prism

Full access: large monthly usage, up to 10 source passages behind each answer, and MCP and API access for Claude Code, Codex and other CLIs. Cancel any time.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →