Thrilled to share this new paper: in a pilot randomized trial, text messages with causal information led to meaningful increases in physical activity, while non-causal texts did not. Surprisingly, people also wanted more messages than the every other day text schedule doi.org/10.2196/80090 #medsky
MIKE
Chappell Roan
CMAT
Lady Gaga
Valerie June
Causal messages share what an action can do (like reduce risk of diabetes) while non-causal ones only suggest the action. In qualitative feedback, participants did notice the connection to their goals.
Thrilled to share this new paper: in a pilot randomized trial, text messages with causal information led to meaningful increases in physical activity, while non-causal texts did not. Surprisingly, people also wanted more messages than the every other day text schedule doi.org/10.2196/80090 #medsky
New paper! How many days of dietary data do we really need to collect? We find that during pregnancy it is more than many studies collect, and depends on the target being estimated (macronutrients needing most days, HEI and timing least). #nutsky doi.org/10.1016/j.aj...
New work with Dave Lagnado out now! Using qualitative + quantitative data we find people reason differently with causal information and see how they bring in prior beliefs. Call to action on methods: if we want to make claims about real world, we need to use realistic stimuli doi.org/10.3389/fcog...
I’m so excited to be speaking at this conference and looking forward to the contributed talks! What is (and should be) a variable is my absolute favorite under discussed topic in causality. Everything follows from this choice, but how should we make it?
To Be or Not to Be Included in a Causal Model
February 28, 2026 @ 9:00 am - March 1, 2026 @ 5:00 pm EST
Venue and Date: Center for Philosophy of Science, February 28 & March 1, 2026
Keynote Speakers:
Samantha Kleinberg (Stevens Institute of Technology)
Lily Hu (Yale University)
SO important, especially as people try to argue science somehow exists separately from values. 🤔
This is a great article on why bias against applied ML matters: www.technologyreview.com/2020/08/18/1...
The Hidden Bias Pushing Women Out of Computing. Excited to see our paper finally being promoted! Regardless of gender, if you do applied work we find there is a real career penalty for it. But women are disproportionately affected because of diffs in research focus www.stevens.edu/news/the-hid...
Not your fault! Most people just do not need this data and forget that it can have downsides. It’s always a good time for a data diet: www.statnews.com/2024/01/08/f...
I’ve written two books on causality, yet was thrilled to find a study on associations involving an understudied health condition where I think an exposure may be causal. Why? Because there was zero evidence before and without even associations it’s hard to argue it’s worth studying (and funding!).
I love showing screenshots when I teach health informatics along with how much companies spend on EHRs. Students are always shocked!
🚨big new paper🚨Using surveys of faculty and data on >10K researchers we find 1) faculty have more negative views of applied researchers vs theoretical, 2) data shows there is a career penalty for doing applied work, and 3) women are more represented in applied research. doi.org/10.1109/ACCE... 1/5
There are so many ways to improve conferences but yikes at the idea of focusing on innovation and concepts rather than whether things actually work. Our recent paper shows this may worsen gender disparities in CS ieeexplore.ieee.org/document/109...
What can we do?
1) Raise awareness in hiring and evaluation
2) Rethink use of CSRankings, which excludes many applied venues
3) Understand how perceptions of a candidate’s research type may unfairly shape perceptions of them as a researcher (like judging them less technically capable). 5/5
Women are more likely to do applied research: Using data from papers and grants we found women are more represented in applied CS relative to their share of CS faculty. Biases against applied research may disproportionately affect women. 4/5
There’s a real penalty for applied research: it is underrepresented at top conferences, among recipients of major awards, and in hiring at top-ranked CS departments. 3/5
Applied research is seen as lower status: Theoretical researchers were rated more likely to publish, get tenure/grants, and win awards. They were also seen as more brilliant/creative/skilled, despite the work itself being valued. Like teaching and service someone should do it, but maybe not them 2/5
🚨big new paper🚨Using surveys of faculty and data on >10K researchers we find 1) faculty have more negative views of applied researchers vs theoretical, 2) data shows there is a career penalty for doing applied work, and 3) women are more represented in applied research. doi.org/10.1109/ACCE... 1/5
Now published! authors.elsevier.com/a/1k%7EYs5SM... Code available here: github.com/health-ai-la... really excited about this one!
Wow, so sorry to hear this. I hope you’ll find a way to share the results and continue your work!
I guess “data diet” is becoming our brand www.statnews.com/2024/01/08/f...
Nice overview of our work showing responses to meals can be predicted even without microbiome and other invasive data! www.stevens.edu/news/stevens... 🩺 🖥️
🤣 we need to expand to other time series data types to see the effect of sampling frequency and nonstationary at short timescales. But we are glad to provide support for common practice with CGM!
If you use CGM or activity data our paper accepted to #CHIL2025 has big news: we benchmarked lots of imputation methods using real missing data mechanisms (not just random deletion) and it turns out … linear interpolation is the winner no matter what the mechanism or % missing data.
Two exciting new papers coming to CogSci 2025: Go Big or Go Hoax (the bigger the more believable for both conspiracies and facts), Causal and Counterfactual Reasoning about Gradual and Abrupt Events (timing matters for token causality and causal attribution doesn’t always track counterfactuals)
New paper just accepted to JBI: we introduce new methods for causal inference in health data with different variables for different patients. Such a common problem and yet there haven’t been any solutions!
Apparently people didn’t learn much from how IBM Watson unfolded (which we discussed in class last week)!
