Unintended bias in the pursuit of collinearity solutions in fMRI analysis
Presented During:
Friday, June 27, 2025: 11:30 AM - 12:45 PM
Brisbane Convention & Exhibition Centre
Room:
P2 (Plaza Level)
Poster No:
1051
Submission Type:
Abstract Submission
Authors:
Jeanette Mumford1, Michael Demidenko1, Russell Poldrack1
Institutions:
1Stanford University, Stanford, CA
First Author:
Co-Author(s):
Introduction:
In task-based fMRI, collinearity between design matrix regressors can impact power. Optimal task design involves evaluating multiple designs to maximize efficiency and minimize collinearity. Here, we highlight inappropriate strategies to reduce collinearity that introduce biases, impair contrast interpretability, and potentially increase false positives. Using Monetary Incentive Delay (MID) task fMRI data from the Adolescent Brain Cognitive Development (ABCD) study, we show that omitting regressors, using impulse regressors for extended activations, and ignoring response times bias contrast estimates, sometimes inflating error rates in a sample of 500 subjects. We propose a "Saturated" model that includes all stimuli and response times, eliminating bias and providing valid estimates of task-related brain activity.
Methods:
A trial of the MID task includes Cue, Fixation, Probe and Feedback components. There are five Cue types: Win $5 (Large Win), Win $0.20 (Small Win), No Money at Stake (Neutral), Don't Lose $5 (Large Loss), and Don't Lose $0.20 (Small Loss). Based on the subject responses, Feedback is classified as either a Hit (successful probe window response) or Miss (Too Soon/Too Slow).
The ABCD sample used here is from the ABCD-BIDS Community Collection (Feczko et al., 2021). We used the onset files from 500 subjects who were subsampled from the Siemens data Year 2 data (Demidenko et al., 2024) to generate design matrices for the purpose of studying efficiency, variance inflation factors and to simulate data for the purposes of assessing bias and error rates. We compare the original ABCD model-which omits probe and fixation stimuli and uses impulse functions-to our proposed Saturated model, which includes all stimuli with their durations and response times (Mumford et al, 2024) (Fig. 1).
Design matrices were constructed by concatenating each subject's two runs into a single run. Synthetic data were generated by multiplying the Saturated model design matrix by the desired true signal components, and then adding independent Gaussian noise to the true signal. Aside from the Null simulation, the true signal component magnitudes of the synthetic data were set to reflect 80% power in group 1-sample t-tests with N=500. Between-participant variability was introduced by sampling the true signal components from a Gaussian distribution with the intended mean; the standard deviation of the between-participant and within-participant variability were set as equal to 1.
The ABCD sample used here is from the ABCD-BIDS Community Collection (Feczko et al., 2021). We used the onset files from 500 subjects who were subsampled from the Siemens data Year 2 data (Demidenko et al., 2024) to generate design matrices for the purpose of studying efficiency, variance inflation factors and to simulate data for the purposes of assessing bias and error rates. We compare the original ABCD model-which omits probe and fixation stimuli and uses impulse functions-to our proposed Saturated model, which includes all stimuli with their durations and response times (Mumford et al, 2024) (Fig. 1).
Design matrices were constructed by concatenating each subject's two runs into a single run. Synthetic data were generated by multiplying the Saturated model design matrix by the desired true signal components, and then adding independent Gaussian noise to the true signal. Aside from the Null simulation, the true signal component magnitudes of the synthetic data were set to reflect 80% power in group 1-sample t-tests with N=500. Between-participant variability was introduced by sampling the true signal components from a Gaussian distribution with the intended mean; the standard deviation of the between-participant and within-participant variability were set as equal to 1.
Results:
The collinearity in the Saturated model is well-controlled, with variance inflation factors at or near 5, which is on par with the original ABCD model. Also the contrast efficiencies are within 25% of each other between the two designs, indicating a limited change in power when using the Saturated design. Due to the biasing, we cannot compare power between the two modeling approaches directly.
Figure 2 shows the bias is present using the ABCD model in all but the Null setting for at least one contrast, with the strongest bias present when signal is introduced for the fixation stimulus. Although not all of these biases inflate error rates at this sample size of 500, a bias with a Cohen's D of .1 would inflate Type I error rates to 15% using a sample size of 9000 subjects, which is close to the full ABCD sample size.
Figure 2 shows the bias is present using the ABCD model in all but the Null setting for at least one contrast, with the strongest bias present when signal is introduced for the fixation stimulus. Although not all of these biases inflate error rates at this sample size of 500, a bias with a Cohen's D of .1 would inflate Type I error rates to 15% using a sample size of 9000 subjects, which is close to the full ABCD sample size.
Conclusions:
While collinearity should be avoided in fMRI design matrices, omitting stimuli, neglecting response time modeling, or using impulse regressors for longer duration stimuli undermines analysis integrity by introducing bias and inflating error rates. Though illustrated with the ABCD study's MID task, these issues apply to all fMRI task designs.
Emotion, Motivation and Social Neuroscience:
Reward and Punishment
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI) 1
Methods Development 2
Univariate Modeling
Keywords:
Design and Analysis
FUNCTIONAL MRI
Modeling
Statistical Methods
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Feczko, E., Conan, G., Marek, S., Tervo-Clemmens, B., Cordova, M., Doyle, O., Earl, E., Perrone, A., Sturgeon, D., Klein, R., Harman, G., Kilamovich, D., Hermosillo, R., Miranda-Dominguez, O., Adebimpe, A., Bertolero, M., Cieslak, M., Covitz, S., Hendrickson, T., … Fair, D. A. (2021). Adolescent Brain Cognitive Development (ABCD) Community MRI Collection and Utilities (p. 2021.07.09.451638). bioRxiv. https://doi.org/10.1101/2021.07.09.451638
J. A. Mumford, P. G. Bissett, H. M. Jones, S. Shim, J. A. H. Rios, and R. A. Poldrack. The response time paradox in functional magnetic resonance imaging analyses. Nature Human Behaviour, 8(2):349–360, Feb. 2024. ISSN 2397-3374. doi: 10.1038/s41562-023-01760-0. URL https://www.nature.com/articles/ s41562-023-01760-0. Publisher: Nature Publishing Group.