About Us
Resources
Documentation
Downloads
- GLIMMPSE Software Downloads
- Other Downloads
Funding
How to Cite
Workshops

Additional Links

Text-only version of this website
Skip navigation
Home page
What's new
Search
Contact Webmaster
Website policies
Disability services

None

Sample Size Shop

Free Sample Size Tools for Multilevel and Longitudinal Data

Home
About Us
Resources
Documentation
Downloads
- GLIMMPSE Software Downloads
- Other Downloads
Funding
How to Cite
Workshops

Calculate Sample Size Now

Read More

GLIMMPSE Does Not Support Internet Explorer Version 9 or Earlier

If you experience problems running GLIMMPSE in your current browser, please try upgrading to Internet Explorer 10 or higher, or try an alternative browser such as Firefox, Safari, or Chrome.

Request More Information

First Name
Last Name
Email*

We do not and never will require you to register to use our open-source GLIMMPSE software. However, to continue to offer this software, we would like to start tracking the locations of users for future grant applications and funding opportunities.
Check to receive future updates
Captcha

Recent News

Short Course Offering: Power and Sample Size for Multilevel and Longitudinal Study Designs

Course Overview This three-day course will give scientists training for calculating power and sample size for multilevel and longitudinal studies. The course is accessible to all, from graduate students to senior […]

GLIMMPSE Version 2.2.0 Released!

GLIMMPSE 2.2.0 has been released. In addition to several bug fixes, this version has a few enhancements, including MANOVA support, improvement of the user interface for specifying group sizes and […]

Biostatistics Research Team Recognized by AMSTAT News for Two Grant Awards

Keith Muller, Ph.D., professor in the Department of Health Outcomes and Policy, and his co-investigator Deborah H. Glueck, Ph.D., associate professor of biostatistics and radiology in the University of Colorado […]

Muller Receives Two Prestigious Grants to Create Statistical Educational Materials

In a single week last September, Keith E. Muller, Ph.D., and his co-investigator Deborah H. Glueck, Ph.D., received two grants from the National Institutes of Health. Muller is a professor […]

JProfiler

Performance profiling performed with Java profiler

GLIMMPSE 2.2.6, 2.2.7, & 2.2.8

Release Notes - SampleSizeShop - 2.2.6, 2.2.7, & 2.2.8

Bug Fixes

Guided Mode, Model > Predictors and Groups: Any change to predictors or their values could cause incorrect adjustments to other parts of the study. Correct adjustments should now be made.
In various circumstances where integer overflow may have caused incorrect results or misleading error messages, GLIMMPSE now detects and reports integer overflow.
An error in the computation of the normal distribution approximation to the noncentral F distribution has been corrected.
Guided Mode, Results > Matrices: With clustering, the wrong value of the Θ matrix was displayed. The correct value is now displayed.
Results > Matrices: With a covariate, the wrong value of the B matrix was displayed. The correct value is now displayed.
Guided Mode, Model > Clustering: It was possible to specify a cluster size of 1. It is no longer possible.
Options > Confidence Intervals: If the error degrees of freedom (the “Total sample size” minus the "Rank of the design matrix”) was less than the rank of the within-participant contrast matrix, incorrect results or misleading error messages appeared. GLIMMPSE does not support this case, the so-called “high dimension, low sample size (HDLSS)” case. GLIMMPSE now detects and reports the HDLSS case.
In certain extreme cases, a bug in the calculation of the Tiku approximation to the noncentral F distribution would result in the message 'Invalid "degrees of freedom" parameter.’ This bug has been fixed.
Cases where corrupted study designs resulted in a report that some covariance is “missing” should now occur less frequently, if at all.
Options > Confidence Intervals: In defining confidence interval parameters, there was no check that the total sample size was greater than the rank of the design matrix. There is now.
In sample size calculations, certain error conditions were reported globally, masking possibly valid calculations for other combinations of study design parameters. This did not happen in power calculations; in power calculations, these error conditions were reported locally, that is, only for the combinations of study design parameters for which they actually occurred. Sample size calculations are now handled the same as power calculations in this respect.

Improvements

Guided Mode, Model > Group Sizes: When other changes to the study resulted in automatic addition of new entries in the table of Relative Group Sizes, they were initialized to the value 1. This gave the user no visual indication that review of those entries was desirable. They are now left blank, so that the user cannot proceed without setting them.
Guided Mode: Buttons to (1) set all blank fields to a common value and (2) set all fields to a common value now exist on the Model > Group Sizes, Means > Means, and Variability > Covariate pages.
Guided Mode: On the Variability > Covariate and Variability > Within-Participant pages, entering correlations outside the open interval (-1, 1) did not result in a warning message. Now it does.
Some study designs that would necessitate unreasonably large sample sizes would fail after several minutes of computation, with the message 'Failed to find valid upper bound on sample size.' Such study designs now fail more quickly, with the message 'The total sample size for this case would be unreasonably large. For performance reasons, we are not computing its exact value.’
Guided Mode, Variability > Within-Participant: In choosing a LEAR correlation matrix, users now have the option of scaling the spacing values so that the minimum distance or time between measurements is 1 unit. For new study designs, this option is the default.
In some cases, when uploading a study design failed, it was possible to proceed with its calculation nevertheless, resulting in spurious behavior. This should no longer be possible.