Resources & FAQs
Towards Understanding the P-Value: Has it Lost its Significance?
Understanding Effect Size and Power Analysis
Reclaim Your Time: Upgrade Your Research with Smart Data Management
What does the 2023 NIH Data Management & Sharing Policy mean for you?
BERD Community Forum
One limitation of standardized effect sizes is its interpretation. In the research community, the magnitude of effect size is set too low, for example, 0.2 is a small effect size for Cohen’s d. This enables researchers to claim that they found practical significance when there is no actual difference of real-world importance. In fact, Cohen referred to an effect size of 0.2 as “difficult to detect”. Secondly, large effect size can also mask the actual benefits which could have been tiny and of no practical benefit. Hence, it is imperative to interpret standardized effect size in range of contexts and not based on conventional values.
Main effect is the effect of one of the independent variables (predictor) on the dependent variable (outcome) while other variables’ effects are ignored.
BMI = sex + smoking status
In this example, if we only wish to know the effect of the smoking status on BMI, then you will take note of the estimate of the smoking status while the estimate of sex will be ignored. However, in reality, variables depend on each other and hence, we need to consider the effect of interactions between two or more variables. Including an interaction in the model is a conscious decision based on a prior knowledge of the subject.
When the effect of one independent variable on the outcome depends on another independent variable, then we call it an interaction effect. When both main and interaction effects are found, then the effect of interaction has higher preeminence in reporting since the main effect of the independent variable under consideration may not reflect the actual magnitude or sometimes direction of the effect.
Let’s say, available literature suggests that sex and smoking status interact with each other. So, we decide to add this interaction in our model.
BMI = sex + smoking status + sex* smoking status
In this case, if we find a considerable interaction effect of sex and smoking status, then we will report this effect. Otherwise, we will proceed to interpret the main effects.
*P-values only tell statistical significance and not practical/clinical significance. Hence, interpret p-values within context.
Graduate student projects are considered on a case-by-case basis, and the proposed work must serve to advance the research and/or funding portfolio of Illinois investigators.
Statistical consulting services geared toward students are also available across campus. For example, the Scholarly Commons at the Main Library offers statistical support for both undergraduate and graduate students. Certain colleges also offer no-cost statistical consulting to their students.
Depending on the complexity of the project, the time to provide the requested deliverable(s) ranges from three weeks up to four months. While all efforts will be made to accommodate deadlines, we strongly urge investigators to allow sufficient time to enable core members to work effectively with your team.
For example, additional time may be needed to add new content and incorporate feedback into the statistical analysis plan (SAP), or to revise a deliverable based on changes in the data set and discussions with the project team. For those projects which require a SAP, the project PI’s approval of the SAP is required to initiate the analyses. Moreover, if the direction of the project changes and additional analyses are required, the project timeline will be reassessed.
Please contact email@example.com.
Please acknowledge Illinois Biostatistics, Epidemiology, and Research Design (BERD) Core support by including the following in your submission(s): "We thank the Interdisciplinary Health Sciences Institute (IHSI) at the University of Illinois at Urbana-Champaign Illinois Biostatistics, Epidemiology, and Research Design (BERD) Core for supporting the work on this project.”
It is our policy to include the biostatistician as co-author for abstracts, manuscripts, posters, presentations, or other materials reporting statistical methods and analyses carried out by, or with significant contributions from, BERD biostatisticians. Please refer to Writing BERD Services into Grants.
As standard practice dictates, core biostatisticians are expected to have the opportunity to review and approve, prior to submission, all final versions of proposals and other products to which they contributed.
We ask that you complete the first and last name fields in a new Project Initiation Form and check “Yes” when asked whether the request is regarding a resubmission or renewal. We will then work to set you up with the biostatistician with whom you previously worked, if available.
The BERD team will send you an evaluation form upon completion of your project. You may also contact Gillian Snyder, IHSI Assistant Director for Research, firstname.lastname@example.org
It may be necessary to replace the biostatistician working on your project due to workload concerns, conflict resolution purposes, or changes in the project requiring a different level or area of expertise. Replacement decisions are made only with consent of all relevant parties.