G*Power is a powerful and versatile statistical software tool widely used by researchers, academics, and students across various disciplines. But *what is GPower* exactly, and why is it so essential in the world of statistical analysis? GPower is primarily designed for performing power analysis, a crucial step in research planning. It helps users calculate the statistical power of a test, determine the sample size needed to achieve a desired level of power, and explore the relationships between effect size, alpha level, power, and sample size. This software is invaluable for ensuring the rigor and validity of research findings by helping to avoid Type II errors (false negatives).
This article delves into the updates and improvements made to GPower over time, showcasing its continuous development and commitment to providing robust statistical tools. By examining the release notes, we can understand how GPower has evolved to address user needs, fix bugs, and enhance its functionalities for more accurate and efficient statistical power analysis. Let’s explore the history of G*Power updates and appreciate the dedication behind this essential statistical software.
G*Power Updates and Bug Fixes: A Chronological Overview
This section provides a detailed look at the updates and bug fixes implemented in various releases of G*Power. These updates reflect the ongoing efforts to refine the software, improve its accuracy, and expand its capabilities to meet the evolving needs of statistical analysis.
March 17, 2020 – Release 3.1.9.7
Windows
- Enhanced User Experience for X-Y Plots: An improvement was made to the “X-Y plot for a range of values” feature. Previously, users could plot graphs after changing input parameters in the main window without pressing the “Calculate” button. However, this led to inconsistencies as the “X-Y plot” would not reflect the new parameters until “Calculate” was explicitly pressed. This release rectifies this by ensuring that the “X-Y plot for a range of values” accurately updates with the latest input parameters from the main dialog, providing a more intuitive and reliable user experience. This ensures that researchers get accurate visualizations based on their most recent parameter adjustments.
February 21, 2020 – Release 3.1.9.6
Mac and Windows
- Bug Fix: Corrected Critical Z-Value Calculation in Generic Z-Tests: A critical bug was identified and fixed in the z tests: Generic z test: Analysis: Criterion: Compute alpha module. The software was incorrectly calculating the critical z-value, which is fundamental for determining statistical significance in z-tests. This correction ensures the accuracy of hypothesis testing when using generic z-tests in G*Power, leading to more reliable research outcomes.
- Bug Fix: Improved Calculation of |sy/sx| in Linear Bivariate Regression (t-tests): Another bug was addressed in t tests: Linear bivariate regression: One group, size of slope. In certain cases, the calculation of |sy/sx|, a component used in effect size calculations for linear regression, was inaccurate. This fix ensures the correct computation of effect sizes in bivariate regression analyses, enhancing the precision of power analysis for regression models within G*Power.
January 14, 2020 – Release 3.1.9.5
Mac
- Bug Fix: Restored “Options” Button Visibility: A user interface bug on the Mac version was resolved. When the “Hide distributions & control” option was selected, the “Options” button, which provides access to additional settings for some tests in the main window, would disappear. This update restores the visibility of the “Options” button even when distributions and controls are hidden, ensuring users can always access all available settings and functionalities of G*Power, regardless of their interface preferences.
February 6, 2019 – Release 3.1.9.4
Mac and Windows
- Bug Fix: Enhanced Handling of Negative Effect Directions in Linear Bivariate Regression (t-tests): Further refinement was implemented in t tests: Linear bivariate regression: One group, size of slope. The software’s handling of negative effect directions, specifically when slope|H1 < slope|H0, was not consistently accurate. This update improves the software’s ability to correctly analyze and account for negative relationships in bivariate regression, providing a more comprehensive and reliable power analysis for researchers investigating negative effects.
- Enhancement: Effect Size Restriction Enforcement in Linear Bivariate Regression: A previously ignored restriction on effect size (|slope| < |sy/sx|) in linear bivariate regression was now implemented. This ensures that effect size specifications adhere to theoretical constraints, preventing users from entering unrealistic or invalid effect sizes, thereby promoting more statistically sound power analyses.
July 7, 2017 – Release 3.1.9.3
Mac
- Bug Fix: Stability Improvement – Crash Prevention: A general bug fix was implemented for the Mac version to address potential crashes. This update focuses on enhancing the software’s stability and robustness, ensuring a smoother and more reliable user experience by preventing unexpected program terminations.
March 28, 2014 – Release 3.1.9.2
Mac
- Bug Fix: Effect Size Display in Chi-Square Goodness-of-Fit Tests: A bug in the χ2 tests: Goodness-of-fit tests: Contingency tables module for Mac was fixed. This bug prevented the computed effect size from being displayed in the effect size drawer. This correction ensures that users can readily access and utilize the effect size information calculated by G*Power for chi-square goodness-of-fit tests, facilitating better interpretation and reporting of power analysis results.
Windows
- Bug Fix: Effect Size Transfer from Drawer to Main Window: A bug specific to Windows was resolved, which occurred under very specific circumstances during the transfer of an effect size value from the effect size drawer to the main window. This fix ensures a seamless and reliable transfer of effect size values, preventing potential data entry errors and streamlining the workflow for Windows users.
March 10, 2014 – Release 3.1.9.1
Mac
- Feature Addition: Inclusion of Calculator in Mac Version: The calculator feature, previously available only in the Windows version of G*Power, was added to the Mac version. This integration provides Mac users with convenient access to a built-in calculator directly within the software, enhancing usability and efficiency for quick calculations needed during power analysis.
February 4, 2014 – Release 3.1.9
Mac and Windows
- Bug Fix: Corrected Effect Size Offset in Sign Test Sensitivity Analysis: A bug in the sign test’s sensitivity analysis was fixed, which caused an offset of -0.5 in the reported effect size. This correction improves the accuracy of sensitivity analyses for sign tests, ensuring more reliable assessments of how power changes with varying effect sizes.
- Behavior Change: Adjusted Binomial Distribution-Based Tests Limits: A significant change was implemented in the behavior of all tests based on the binomial distribution. The upper and lower limits for these tests are now consistently within the range [0,n] instead of the previous [-1,n+1]. While this might lead to alpha values slightly larger than the requested values in some cases, it ensures that the limits correspond to actual decision boundaries. This modification provides a more conceptually sound and practically relevant interpretation of results, especially for tests with small sample sizes (N).
January 31, 2014 – Release 3.1.8
Mac and Windows
- Improvements in Logistic Regression Module:
- Enhanced Numerical Stability: The logistic regression module underwent improvements to enhance its numerical stability, particularly for lognormal distributed covariates. This makes the software more robust when dealing with complex data distributions in logistic regression analyses, improving the reliability of results.
- Additional Input Parameter Validity Checks: More comprehensive validity checks were added for input parameters in both the logistic and Poisson regression modules. This proactive measure helps prevent errors by alerting users to potentially invalid input values, ensuring that analyses are conducted with appropriate and meaningful parameters.
- Improved Power Sensitivity Analysis Handling: Sensitivity analyses were refined to better handle cases where power does not increase monotonically with effect size. An “Actual power” output field was added to indicate such instances. Deviations between the requested and actual power highlight situations where the power-effect size relationship is non-standard, prompting users to examine the power dependency on effect size in the plot window for a more nuanced understanding of their analysis.
April 19, 2013 – Release 3.1.7
Mac and Windows
- Bug Fix: Exact Test of Proportions – Inequality (Two Independent Groups): A bug was fixed in the exact test of proportions for inequality between two independent groups (unconditional). This issue specifically occurred when p1 > p2. The correction ensures the accurate calculation and application of this test, particularly in scenarios where the proportion in the first group is expected to be greater than the proportion in the second group.
February 18, 2013 – Release 3.1.6
Mac and Windows
- Bug Fix: Sensitivity Analysis of Logistic Regression (p2 >= p1): A bug was resolved in the sensitivity analysis of logistic regression, specifically when p2 >= p1. This fix ensures the correct execution and interpretation of sensitivity analyses in logistic regression even when the second probability is greater than or equal to the first, expanding the reliability of this feature across different data scenarios.
Mac
- Bug Fix: Fisher’s Exact Test: A bug in Fisher’s exact test was fixed for the Mac version. This correction enhances the accuracy and reliability of Fisher’s exact test, a crucial non-parametric test for analyzing contingency tables, particularly with small sample sizes.
September 12, 2012 – Release 3.1.5.1
Mac
- Bug Fix: Effect Size Drawers in ANOVA: Fixed Effects: A problem with the effect size drawers in ANOVA: Fixed effects was corrected. The drawers now appear correctly after clicking the “Determine” button, ensuring that users can seamlessly access and utilize the effect size determination functionalities within ANOVA fixed effects analyses.
August 20, 2012 – Release 3.1.5
Mac and Windows
- Bug Fix: Test of Equality of Two Variances (Unequal Sample Sizes): A bug was fixed in the test of equality of two variances. This issue did not occur when sample sizes were identical, implying it was related to scenarios with unequal group sizes. The fix ensures the accurate application of this test even when comparing variances from groups with different sample sizes, broadening the applicability of this variance comparison test.
- Bug Fix: Effect Size Calculation in Repeated Measures ANOVA (Variances): A bug in calculating the effect size from variances in repeated measures ANOVA was resolved. This correction improves the accuracy of effect size estimations in repeated measures ANOVA when using variances as input, leading to more precise power analyses for within-subjects designs.
July 3, 2012 – Release 3.1.4
Mac and Windows
- Feature Addition: Options Dialog for Repeated-Measures ANOVA: An options dialog was added to the repeated-measures ANOVA module. This enhancement provides users with more flexibility in specifying effect sizes for repeated measures ANOVA, allowing for more tailored and nuanced power analyses for complex within-subjects research designs.
- Bug Fix: Sample Size Calculation for Fisher’s Exact Test (A Priori Analyses): A bug in calculating the sample size for Fisher’s exact test was fixed. This problem did not occur in post hoc analyses, indicating it was specific to a priori power analyses (determining sample size before data collection). The fix ensures the correct sample size calculation for Fisher’s exact test in a priori power analyses, crucial for planning studies with adequate statistical power from the outset.
June 22, 2011 – Release 3.1.3
Mac and Windows
- Bug Fix: ANCOVA Module – Degrees of Freedom Adjustment: A bug in the ANCOVA module was fixed. Changing the number of covariates now correctly leads to the appropriate change in the denominator degrees of freedom. This correction ensures the accurate calculation of degrees of freedom in ANCOVA when adjusting the number of covariates, which is essential for correct statistical inference in analysis of covariance.
January 5, 2010 – Release 3.1.2
Mac and Windows
- Parameter Renaming: “Repetitions” to “Number of measurements” in Repeated Measures Procedures: The parameter label “Repetitions” in repeated measures procedures was renamed to “Number of measurements.” The original term was misleading as it incorrectly suggested that the first measurement was not counted. This change clarifies the parameter’s meaning, improving user understanding and reducing potential confusion when specifying repeated measures designs.
- Bug Fix: Sensitivity Analysis of Logistic Regression – Odds Ratio Effect Size: A bug was fixed in the sensitivity analysis of the logistic regression procedure. This bug occurred when “Odds ratio” was selected as the effect size but not when the effect size was specified in terms of “Two probabilities.” The correction ensures that sensitivity analyses function correctly regardless of whether the effect size is defined by odds ratio or probabilities, providing users with consistent and reliable sensitivity analysis results for logistic regression.
Mac
- Feature Addition: “Hide distributions & protocol” Option in Window Menu: The Window menu on the Mac version now includes the option to hide the distributions plot and the protocol section (Hide distributions & protocol menu item). This feature, already available in the Windows version (View menu), allows users to optimize the G*Power interface for smaller screens, improving usability in environments with limited display space.
December 16, 2009 – Release 3.1
Mac
- Expanded Test Procedures: Significant additions were made to the Mac version, expanding the range of statistical tests G*Power could analyze. New procedures were added to analyze the power of tests for:
- Single correlations based on the tetrachoric model
- Comparisons of dependent correlations
- Bivariate linear regression
- Multiple linear regression based on the random predictor model
- Logistic regression
- Poisson regression
These additions significantly broadened the scope of statistical analyses that G*Power could support on the Mac platform, making it a more comprehensive tool for researchers using diverse statistical methods.
December 8, 2009 – Release 3.1.1
Windows
- Feature Re-enabled: Options Dialog for Correlation: Bivariate Normal Model: The options dialog for Correlation: Bivariate normal model was re-enabled in the Windows version. This restores access to potentially useful settings and customizations for correlation analyses under the bivariate normal model, providing Windows users with complete functionality for these analyses.
June 29, 2009 – Release 3.1
Windows
- Expanded Test Procedures (Windows): Similar to the Mac release on December 16, 2009, this Windows release added the same expanded set of test procedures:
- Single correlations based on the tetrachoric model
- Comparisons of dependent correlations
- Bivariate linear regression
- Multiple linear regression based on the random predictor model
- Logistic regression
- Poisson regression
This update brought the Windows version feature parity with the Mac version in terms of the newly added statistical tests, ensuring both platforms offered a similarly comprehensive suite of power analysis tools.
January 24, 2008 – Release 3.0.10
Mac and Windows
- Bug Fix: X-Y Plot for Generic F-Tests – Degrees of Freedom: A bug was fixed in the X-Y plot for a range of values feature for Generic F tests. The degrees of freedom were not being properly set in the graph, leading to erroneous plot values. This correction ensures the accuracy of visualizations generated by X-Y plots for generic F-tests, providing users with reliable graphical representations of power analysis results.
January 22, 2008 – Release 3.0.9
Mac
- GUI Enhancement: Eliminated Brushed Metal Look (Mac): The brushed metal look in the Mac OS X 10.5 interface was removed to improve readability. This user interface adjustment focused on enhancing the visual clarity and usability of G*Power on Mac, making it easier for users to navigate and interact with the software.
Windows
- GUI Fixes (Windows):
- Corrected Button Titles (German to English): Some buttons with German titles were corrected to English titles, addressing a localization issue and ensuring a consistent English language interface for international users.
- File Dialog Button Label Correction: The button in the file dialog was renamed from “open” to “save,” correcting a misleading label and improving the intuitiveness of file saving operations.
- Distribution Plot Improvements: Issues with distribution plots were addressed. Plots were sometimes not appropriately clipped when copied or saved as metafiles, and drawing glitches occurred with very steep curves. These fixes improve the visual quality and accuracy of distribution plots, ensuring that graphs are rendered correctly for both on-screen viewing and export.
- Default Directory for File Dialogs: The file dialog for saving graphs or protocols now defaults to the user’s home directory (myDocuments). This change enhances user convenience by making it easier to locate and save files in a familiar and frequently used directory.
October 10, 2007 – Release 3.0.8
Mac and Windows
- Critical Bug Fix: CDF Routine of Noncentral t-distribution: A serious bug in the CDF (Cumulative Distribution Function) routine of the noncentral t distribution, introduced in the bugfix release 3.0.7, was fixed. This was a critical update, and users who installed version 3.0.7 were strongly advised to update immediately. This bug fix was paramount as it affected the accuracy of all power routines based on the t distribution, which are fundamental to many statistical analyses.
October 8, 2007 – Release 3.0.7
Mac and Windows
- Bug Fix: CDF of Noncentral t-distribution (p close to 1): A bug was fixed in the function calculating the CDF of the noncentral t-distribution. This bug occasionally led to incorrect values when p (probability) was very close to 1. As all power routines based on the t distribution were affected, this fix was crucial for ensuring the accuracy of power calculations across a wider range of probability values.
August 14, 2007 – Release 3.0.6
Mac
- Bug Fix: Central and Noncentral t-distributions for Two-Tailed Tests: A bug in the routine that draws the central and noncentral t distributions for two-tailed tests was fixed for the Mac version. When alpha was very small, this bug could cause G*Power to crash. This fix improves the stability of the software when visualizing t-distributions, particularly for analyses with very stringent alpha levels.
July 3, 2007 – Release 3.0.5
Mac and Windows
- Bug Fix: Power Plot for MANOVA F-Tests: A bug in the Power Plot (accessed via the X-Y-plot for a range of values button) for F tests, MANOVA: Global effects and F Tests, MANOVA: Special effects and interactions was fixed. Sometimes, variables were not correctly set in the plot procedure, leading to erroneous values in graphs and tables. This correction ensures the accuracy of power plots for MANOVA F-tests, providing users with reliable visualizations of power relationships in multivariate analysis of variance.
June 26, 2007 – Release 3.0.4
Mac and Windows
- Bug Fix: Power Plot for Multiple Regression F-Tests: A bug in the Power Plot for F Tests, Multiple Regression: Special (R2 increase) was fixed. The numerator degrees of freedom (df) value was not always correctly determined in the plot procedure, leading to erroneous values in graphs and tables. This fix ensures accurate power plots for multiple regression F-tests, improving the reliability of graphical power analysis for regression models.
- Minor Problems Fixed in t-tests: Some minor problems in t tests were addressed, contributing to the overall refinement and accuracy of t-test functionalities within G*Power.
- Enhancement: Invalid Null Effect Size Check in t-tests: G*Power now checks for invalid null effect sizes in a priori t-test analyses. This proactive check helps prevent user errors by identifying and flagging nonsensical null effect size specifications, promoting more valid and meaningful power analyses.
- Robustness Improvement: t-distribution PDF Routine for Large Degrees of Freedom: The t distribution Probability Density Function (PDF) routine was made more robust for very large degrees of freedom by explicitly using a normal approximation in such cases. This enhancement improves the computational efficiency and accuracy of t-distribution calculations, especially when dealing with large samples.
- GUI Adjustment: Power Plot Default Plots Reduced: When the Power Plot window was first opened via the X-Y plot for a range of values button, the default was to show 4 plots simultaneously. This default was changed to 1 plot. This GUI adjustment likely aimed to improve initial clarity and reduce visual clutter when first accessing power plots.
- Parsing Error Corrections in Calculator (Mac): Parsing errors in the calculator were corrected, particularly concerning text input in normal input fields in the Mac version. This fix improves the reliability and user-friendliness of the calculator feature on Mac.
- Label Correction: ANOVA Effect Size Drawer: A label in the effect size drawer for ANOVA: Fixed effects, omnibus, one-way was corrected. In the From variance input mode, the Variance within group field was erroneously labeled “Error variance.” This label correction enhances clarity and accuracy in the user interface, reducing potential confusion about input parameters for ANOVA effect size calculations.
Windows
- GUI Fix: Main Window Movement with Open Effect Size Drawer: A problem with moving the main window when the effect size drawer was open was fixed. Sometimes the mouse pointer appeared to be “glued” to the window, hindering smooth window movement. This fix improves the user experience by ensuring consistent and predictable window behavior.
March 1, 2007 – Release 3.0.3
Mac and Windows
- Bug Fix: X-Y Plots for ANOVA F-Tests (Main Effects and Interactions): A bug in the X-Y plots for a range of values for F Tests, ANOVA: Fixed effects, special, main effects and interactions was fixed. The df1 (numerator degrees of freedom) value was not always correctly determined in the plot procedure, leading to erroneous plot values. This correction ensures accurate power plots for ANOVA F-tests focusing on main effects and interactions.
- Bug Fix: X-Axis Last Point Inclusion in Plots: A problem in the plot procedure was resolved where the last point on the x-axis was sometimes not included in the plot due to rounding errors. This fix ensures that plots accurately represent the full range of x-axis values, providing a complete and visually accurate depiction of the data.
February 28, 2007 – Release 3.0.2
Mac and Windows
- Bug Fix: Tooltips for Effect Size Conventions: A problem with tooltips for effect size conventions, which were not always shown, was fixed. This correction ensures that users can reliably access helpful tooltips providing context and guidance on effect size conventions, improving user understanding and facilitating informed decisions in power analysis.
Windows
- GUI Enhancements for Low-Resolution Displays (800×600):
- Hideable Distribution/Protocol and Test/Analysis Views: Options were added to temporarily hide the distribution/protocol view and the test/analysis selection view in the main window to save screen space, specifically for users with low-resolution displays (800 x 600 pixels). These views can be toggled using F4 (plot/protocol) and F5 (test/analysis) keys, or via the View menu. These enhancements make G*Power more usable on systems with limited screen real estate.
- Resizable Graph Window Option: The Graph window can now be made resizable by selecting “Resizable Window” in the View menu of the Graph window. This option also initially shrinks the window to fit an 800 x 600 screen. Deselecting the option restores the fixed-size window. This feature provides users with greater control over the Graph window size, improving adaptability to different screen resolutions and user preferences.
Mac
- Cosmetic Fix: “Extreme” Distribution Drawing: A cosmetic problem when drawing “extreme” central and non-central distributions on Mac was fixed. This visual refinement improves the aesthetic quality of distribution plots, particularly for distributions with extreme parameters.
January 15, 2007 – Release 3.0.1
Windows
- Bug Fix: Effect Size Calculation for Independent Means t-test (Equal n): Effect size calculation for t Tests, Difference between two independent means (two groups), in the case of equal sample sizes (n1 = n2), was corrected. The wrong means (those from the unequal sample size case n1 ≠ n2) were being used to calculate the effect size. This fix ensures the accurate calculation of effect sizes for independent means t-tests when group sizes are equal.
- Minor Cosmetic Changes: Minor visual refinements were made to the user interface.
- Updated Documentation: The Readme file and installation instructions were updated to reflect the latest version and features of G*Power.
January 12, 2007 – Release 3.0.0
- First Release of Version 3.0.0: This marks the initial release of G*Power version 3.0.0, a significant milestone representing a major update and expansion of the software’s capabilities.
Conclusion: The Continuous Evolution of G*Power
This detailed overview of G*Power updates highlights the ongoing dedication to refining and enhancing this crucial statistical tool. From critical bug fixes that ensure the accuracy of core statistical calculations to user interface improvements that enhance usability, each update reflects a commitment to providing researchers with a reliable and user-friendly software for power analysis. Understanding *what is GPower* involves not only recognizing its core functionalities but also appreciating its continuous evolution to meet the ever-changing demands of statistical research. By staying updated with the latest releases, users can leverage the most accurate and feature-rich version of GPower for their research endeavors.
