iiinova — where your digital self keeps evolving

Abhraneel Sarma

@abhsarma.bsky.social

VIS/HCI Postdoc at TU Graz. Previouly at Northwestern CS, UMich and IITG Author (and maintainer) of the multiverse R pkg, tea drinker, book hoarder, afternoon napper abhsarma.github.io

Posts Replies Media

Abhraneel Sarma @abhsarma.bsky.social · 09/11/2025

I’ve long admired the work done by Alex and the VDL, and I’m really excited to be a part of it!!!

050

Abhraneel Sarma @abhsarma.bsky.social · 03/08/2025

Also thank you to Darren and Fanny for being on my committee! You’ve always given great feedback and been very supportive of my work :)

010

Abhraneel Sarma @abhsarma.bsky.social · 03/08/2025

Thank you @jessicahullman.bsky.social and @mjskay.com for being such wonderful mentors!!! Looking back, one thing I really appreciated was all the times that you’ve challenged me to take on the more difficult problem, and that’s helped me improve my work immensely!

130

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

Takeaway #2: designers should visualise multiple distributions as p-boxes if they want viewers to adopt a risk-averse decision-making strategy regardless of how the forecasts are distributed.

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

Takeaway #1: designers should (probably) visualise multiple distributions using ensembles if they want viewers to base their decisions on the more frequent predictions (aka a probabilistic interpretation of the 2nd order of uncertainty)

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

If the forecasts are clustered and represented as ensembles, then the decision-making strategy that participants adopt is largely based on the cluster of forecasts.

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

We found that p-boxes lead to relatively more risk-averse decision-making where the lower bound of the set of forecasts is weighted more heavily. On the other hand, decisions made with ensembles are only going to be similarly risk-averse if the forecasts are not clustered.

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

We explore how the visual representations used can impact the decision-making strategies that a viewer adopts. We vary how the 2nd order uncertainty is visualised using either ensembles or p-boxes (which only shows the bounds of the distributions.

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

Decision-makers have access to multiple forecasts, often from different models which make different assumptions. Simply taking the average of all of these predictions might not always lead to the best decisions. We consider a range of possible strategies that may be used for decision-making.

100

Abhraneel Sarma @abhsarma.bsky.social · 29/04/2025

🚨Paper alert! 🚨 I'll be presenting my paper visualising multiple forecast distributions at #chi2025! (If you are in-person, the talk is on Wed 11:10am in G414+415) If you are not interested in probability distributions, I will also be talking about 🦙 alpacas 🦙 dl.acm.org/doi/10.1145/...

dl.acm.org

More Forecasts, More (Decision) Problems: How Uncertainty Representations for Multiple Forecasts Impact Decision Making | Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems

You will be notified whenever a record that you have chosen has been cited.

1121

Reposted by Abhraneel Sarma

Matthew Kay @mjskay.com · 26/11/2024

This is a great point, and was one reason @abhsarma.bsky.social (with me + @jessicahullman.bsky.social) built a multiverse analysis tool that tries to help experts understand and evaluate the validity of subsets of a multiverse, rather than just shrugging: "enh, results vary" doi.org/10.1145/3613...

Figure 2: The Milliways interface consists of four panels to support the multiverse analysis tasks and a principled evaluation of the multiverse analysis. The outcome panel Al presents the results of the multiverse-for each universe in the multiverse, we show an outcome or result. The specification panel A2 reveals the options for each parameter which the universe is composed of. The code panel A3 presents the code that was used to construct the multiverse (here, the code was from the multiverse R library). The data panel A4 shows the dataset used in the analysis. Other elements include the add button to add additional outcome variables A5, exclude button to remove options of a parameter AG, link button to link (aggregate) the outcomes across two or more options A7, slider for hierarchical sorting based on average effect AS and the zoom toggle A9

Figure 5: Hierarchical sorting based on the average marginal effect allows users to see the variation in the outcome due to the options of a parameter. It also allows users to inspect the outcomes conditional on the options of parameters

Figure 6: On aggregation, we represent uncertainty in the result using probability boxes

Figure 7: A walkthrough of the steps that Alice performs during her principled evaluation of a multiverse analysis using Milliways which is described in our case study

58430

Abhraneel Sarma @abhsarma.bsky.social · 11/06/2024

Can we make using SPSS (or other proprietary software) for analysis in papers be grounds for desk rejection? (I am only slightly joking)

050

Abhraneel Sarma @abhsarma.bsky.social · 12/12/2023

Once, we did write the results section in RMarkdown (uploaded it to Github for collabs). When things looked good, compiled it to latex and then uploaded to overleaf

020