Assessments are critical in education, but creating them can be difficult. To address this challenge in a grounded way, we partnered with 13 teachers in a seven-month codesign process. We developed a conceptual model that characterizes the iterative dual process where teachers develop assessments while simultaneously refining requirements. To enact this model in practice, we built Ripplet,1 a web-based tool with multilevel reusable interactions to support assessment authoring. The extended codesign revealed that Ripplet enabled teachers to create formative assessments they would not have otherwise made, shifted their practices from generation to curation, and helped them reflect more on assessment quality. In a user study with 15 additional teachers, compared to their current practices, teachers felt the results were more worth their effort and that assessment quality improved.

Recovering a continuous colormap from a single 2D scalar field visualization can be quite challenging, especially in the absence of a corresponding color legend. In this paper, we propose a novel colormap recovery approach that extracts the colormap from a color-encoded 2D scalar field visualization by simultaneously predicting the colormap and underlying data using a decoupling-and-reconstruction strategy. Our approach first separates the input visualization into colormap and data using a decoupling module, then reconstructs the visualization with a differentiable color-mapping module. To guide this process, we design a reconstruction loss between the input and reconstructed visualizations, which serves both as a constraint to ensure strong correlation between colormap and data during training, and as a self-supervised optimizer for fine-tuning the predicted colormap of unseen visualizations during inferencing. To ensure smoothness and correct color ordering in the extracted colormap, we introduce a compact colormap representation using cubic B-spline curves and an associated color order loss. We evaluate our method quantitatively and qualitatively on a synthetic dataset and a collection of real-world visualizations from the VIS30K dataset [9]. Additionally, we demonstrate its utility in two prototype applications-colormap adjustment and colormap transfer-and explore its generalization to visualizations with color legends and ones encoded using discrete color palettes.

Nishant Balepur, Vishakh Padmakumar, Fumeng Yang, Shi Feng, Rachel Rudinger, Jordan Lee Boyd-Graber. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Nishant Balepur, Matthew Shu, Yoo Yeon Sung, Seraphina Goldfarb-Tarrant, Shi Feng, Fumeng Yang, Rachel Rudinger, Jordan Lee Boyd-Graber. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Visualization items-factual questions about visualizations that ask viewers to accomplish visualization tasks-are regularly used in the field of information visualization as educational and evaluative materials. For example, researchers of visualization literacy require large, diverse banks of items to conduct studies where the same skill is measured repeatedly on the same participants. Yet, generating a large number of high-quality, diverse items requires significant time and expertise. To address the critical need for a large number of diverse visualization items in education and research, this paper investigates the potential for large language models (LLMS) to automate the generation of multiple-choice visualization items. Through an iterative design process, we develop the VILA (Visualization Items Generated by Large LAnguage Models) pipeline, for efficiently generating visualization items that measure people's ability to accomplish visualization tasks. We use the VILA pipeline to generate 1,404 candidate items across 12 chart types and 13 visualization tasks. In collaboration with 11 visualization experts, we develop an evaluation rulebook which we then use to rate the quality of all candidate items. The result is the VILA bank of ~1, 100 items. From this evaluation, we also identify and classify current limitations of the VILA pipeline, and discuss the role of human oversight in ensuring quality. In addition, we demonstrate an application of our work by creating a visualization literacy test, VILA-VLAT, which measures people's ability to complete a diverse set of tasks on various types of visualizations; comparing it to the existing VLAT, VILA-VLAT shows moderate to high convergent validity (R = 0.70). Lastly, we discuss the application areas of the VILA pipeline and the VILA bank and provide practical recommendations for their use. All supplemental materials are available at https://osf.io/ysrhq/.

A year ago, we submitted an IEEE VIS paper entitled "Swaying the Public? Impacts of Election Forecast Visualizations on Emotion, Trust, and Intention in the 2022 U.S. Midterms" [50], which was later bestowed with the honor of a best paper award. Yet, studying such a complex phenomenon required us to explore many more design paths than we could count, and certainly more than we could document in a single paper. This paper, then, is the unwritten prequel-the backstory. It chronicles our journey from a simple idea-to study visualizations for election forecasts-through obstacles such as developing meaningfully different, easy-to-understand forecast visualizations, crafting professional-looking forecasts, and grappling with how to study perceptions of the forecasts before, during, and after the 2022 U.S. midterm elections. This journey yielded a rich set of original knowledge. We formalized a design space for two-party election forecasts, navigating through dimensions like data transformations, visual channels, and types of animated narratives. Through qualitative evaluation of ten representative prototypes with 13 participants, we then identified six core insights into the interpretation of uncertainty visualizations in a U.S. election context. These insights informed our revisions to remove ambiguity in our visual encodings and to prepare a professional-looking forecasting website. As part of this story, we also distilled challenges faced and design lessons learned to inform both designers and practitioners. Ultimately, we hope our methodical approach could inspire others in the community to tackle the hard problems inherent to designing and evaluating visualizations for the general public.

Trust in high-profile election forecasts influences the public’s confidence in democratic processes and electoral integrity. Yet, maintaining trust after unexpected outcomes like the 2016 U.S. presidential election is a significant challenge. Our work confronts this challenge through three experiments that gauge trust in election forecasts. We generate simulated U.S. presidential election forecasts, vary win probabilities and outcomes, and present them to participants in a professional-looking website interface. In this website interface, we explore (1) four different uncertainty displays, (2) a technique for subjective probability correction, and (3) visual calibration that depicts an outcome with its forecast distribution. Our quantitative results suggest that text summaries and quantile dotplots engender the highest trust over time, with observable partisan differences. The probability correction and calibration show small-to-null effects on average. Complemented by our qualitative results, we provide design recommendations for conveying U.S. presidential election forecasts and discuss long-term trust in uncertainty communication. We provide preregistration, code, data, model files, and videos at https://doi.org/10.17605/OSF.IO/923E7.

Visualization literacy is an essential skill for accurately interpreting data to inform critical decisions. Consequently, it is vital to understand the evolution of this ability and devise targeted interventions to enhance it, requiring concise and repeatable assessments of visualization literacy for individuals. However, current assessments, such as the Visualization Literacy Assessment Test (VLAT), are time-consuming due to their fixed, lengthy format. To address this limitation, we develop two streamlined computerized adaptive tests (CATs) for visualization literacy, A-VLAT and A-CALVI, which measure the same set of skills as their original versions in half the number of questions. Specifically, we (1) employ item response theory (IRT) and non-psychometric constraints to construct adaptive versions of the assessments, (2) finalize the configurations of adaptation through simulation, (3) refine the composition of test items of A-CALVI via a qualitative study, and (4) demonstrate the test-retest reliability (ICC: 0.98 and 0.98) and convergent validity (correlation: 0.81 and 0.66) of both CATs via four online studies. We discuss practical recommendations for using our CATs and opportunities for further customization to leverage the full potential of adaptive assessments. All supplemental materials are available at https://osf.io/a6258/.

We conduct two in-lab experiments (N = 93) to evaluate the effectiveness of Gantt charts, extended Gantt charts, and stringline charts for visualizing fixed-order event sequence data. We first formulate five types of event sequences and define three types of sequence elements: point events, interval events, and the temporal gaps between them. Our two experiments focus on event sequences with a pre-defined, fixed order and measure task error rates and completion time. The first experiment shows single sequences and assesses the three charts' performance in comparing event duration or gap. The second experiment shows multiple sequences and evaluates how well the charts reveal temporal patterns. The results suggest that when visualizing single fixed-order event sequences, 1) Gantt and extended Gantt charts lead to comparable error rates in the duration-comparing task; 2) Gantt charts exhibit either shorter or equal completion time than extended Gantt charts; 3) both Gantt and extended Gantt charts demonstrate shorter completion times than stringline charts; 4) however, stringline charts outperform the other two charts with fewer errors in the comparing task when event type counts are high. Additionally, when visualizing multiple point-based fixed-order event sequences, stringline charts require less time than Gantt charts for people to find temporal patterns. Based on these findings, we discuss design opportunities for visualizing fixed-order event sequences and discuss future avenues for optimizing these charts.

We conducted a longitudinal study during the 2022 U.S. midterm elections, investigating the real-world impacts of uncertainty visualizations. Using our forecast model of the governor elections in 33 states, we created a website and deployed four uncertainty visualizations for the election forecasts: single quantile dotplot (1-Dotplot), dual quantile dotplots (2-Dotplot), dual histogram intervals (2-Interval), and Plinko quantile dotplot (Plinko), an animated design with a physical and probabilistic analogy. Our online experiment ran from Oct. 18, 2022, to Nov. 23, 2022, involving 1,327 participants from 15 states. We use Bayesian multilevel modeling and post-stratification to produce demographically-representative estimates of people's emotions, trust in forecasts, and political participation intention. We find that election forecast visualizations can heighten emotions, increase trust, and slightly affect people's intentions to participate in elections. 2-Interval shows the strongest effects across all measures; 1-Dotplot increases trust the most after elections. Both visualizations create emotional and trust gaps between different partisan identities, especially when a Republican candidate is predicted to win. Our qualitative analysis uncovers the complex political and social contexts of election forecast visualizations, showcasing that visualizations may provoke polarization. This intriguing interplay between visualization types, partisanship, and trust exemplifies the fundamental challenge of disentangling visualization from its context, underscoring a need for deeper investigation into the real-world impacts of visualizations. Our preprint and supplements are available at https://doi.org/osf.io/ajq8f.

We propose a new approach to uncertainty communication: we keep the uncertainty representation fixed, but adjust the distribution displayed to compensate for biases in people’s subjective probability in decision-making. To do so, we adopt a linear-in-probit model of subjective probability and derive two corrections to a Normal distribution based on the model’s intercept and slope: one correcting all right-tailed probabilities, and the other preserving the mode and one focal probability. We then conduct two experiments on U.S. demographically-representative samples. We show participants hypothetical U.S. Senate election forecasts as text or a histogram and elicit their subjective probabilities using a betting task. The first experiment estimates the linear-in-probit intercepts and slopes, and confirms the biases in participants’ subjective probabilities. The second, preregistered follow-up shows participants the bias-corrected forecast distributions. We find the corrections substantially improve participants’ decision quality by reducing the integrated absolute error of their subjective probabilities compared to the true probabilities. These corrections can be generalized to any univariate probability or confidence distribution, giving them broad applicability. Our preprint, code, data, and preregistration are available at https://doi.org/10.17605/osf.io/kcwxm

How deep neural networks can aid visualization perception research is a wide-open question. This paper provides insights from three perspectives—prediction, generalization, and interpretation—via training and analyzing deep convolutional neural networks on human correlation judgments in scatterplots across three studies. The first study assesses the accuracy of twenty-nine neural network architectures in predicting human judgments, finding that a subset of the architectures (e.g., VGG-19) has comparable accuracy to the best-performing regression analyses in prior research. The second study shows that the resulting models from the first study display better generalizability than prior models on two other judgment datasets for different scatterplot designs. The third study interprets visual features learned by a convolutional neural network model, providing insights about how the model makes predictions, and identifies potential features that could be investigated in human correlation perception studies. Together, this paper suggests that deep neural networks can serve as a tool for visualization perception researchers in devising potential empirical study designs and hypothesizing about perpetual judgments. The preprint, data, code, and training logs are available at https://doi.org/10.17605/osf.io/exa8m.

Immersive visualization in virtual reality (VR) allows us to exploit visual cues for perception in 3D space, yet few existing studies have measured the effects of visual cues. Across a desktop monitor and a head-mounted display (HMD), we assessed scatterplot designs which vary their use of visual cues-motion, shading, perspective (graphical projection), and dimensionality-on two sets of data. We conducted a user study with a summary task in which 32 participants estimated the classification accuracy of an artificial neural network from the scatterplots. With Bayesian multilevel modeling, we capture the intricate visual effects and find that no cue alone explains all the variance in estimation error. Visual motion cues generally reduce participants' estimation error; besides this motion, using other cues may increase participants' estimation error. Using an HMD, adding visual motion cues, providing a third data dimension, or showing a more complicated dataset leads to longer response times. We speculate that most visual cues may not strongly affect perception in immersive analytics unless they change people's mental model about data. In summary, by studying participants as they interpret the output from a complicated machine learning model, we advance our understanding of how to use the visual cues in immersive analytics.

Data can be visually represented using visual channels like position, length or luminance. An existing ranking of these visual channels is based on how accurately participants could report the ratio between two depicted values. There is an assumption that this ranking should hold for different tasks and for different numbers of marks. However, there is surprisingly little existing work that tests this assumption, especially given that visually computing ratios is relatively unimportant in real-world visualizations, compared to seeing, remembering, and comparing trends and motifs, across displays that almost universally depict more than two values. To simulate the information extracted from a glance at a visualization, we instead asked participants to immediately reproduce a set of values from memory after they were shown the visualization. These values could be shown in a bar graph (position (bar)), line graph (position (line)), heat map (luminance), bubble chart (area), misaligned bar graph (length), or 'wind map' (angle). With a Bayesian multilevel modeling approach, we show how the rank positions of visual channels shift across different numbers of marks (2, 4 or 8) and for bias, precision, and error measures. The ranking did not hold, even for reproductions of only 2 marks, and the new probabilistic ranking was highly inconsistent for reproductions of different numbers of marks. Other factors besides channel choice had an order of magnitude more influence on performance, such as the number of values in the series (e.g., more marks led to larger errors), or the value of each mark (e.g., small values were systematically overestimated). Every visual channel was worse for displays with 8 marks than 4, consistent with established limits on visual memory. These results point to the need for a body of empirical studies that move beyond two-value ratio judgments as a baseline for reliably ranking the quality of a visual channel, including testing new tasks (detection of trends or motifs), timescales (immediate computation, or later comparison), and the number of values (from a handful, to thousands).

Data visualizations convert numbers into visual marks so that our visual system can extract data from an image instead of raw numbers. Clearly, the visual system does not compute these values as a computer would, as an arithmetic mean or a correlation. Instead, it extracts these patterns using perceptual proxies; heuristic shortcuts of the visual marks, such as a center of mass or a shape envelope. Understanding which proxies people use would lead to more effective visualizations. We present the results of a series of crowdsourced experiments that measure how powerfully a set of candidate proxies can explain human performance when comparing the mean and range of pairs of data series presented as bar charts. We generated datasets where the correct answer-the series with the larger arithmetic mean or range-was pitted against an "adversarial" series that should be seen as larger if the viewer uses a particular candidate proxy. We used both Bayesian logistic regression models and a robust Bayesian mixed-effects linear model to measure how strongly each adversarial proxy could drive viewers to answer incorrectly and whether different individuals may use different proxies. Finally, we attempt to construct adversarial datasets from scratch, using an iterative crowdsourcing procedure to perform black-box optimization.

We present exploratory research of virtual reality techniques and mnemonic devices to assist in retrieving knowledge from scholarly articles. We used abstracts of scientific publications to represent knowledge in scholarly articles; participants were asked to read, remember, and retrieve knowledge from a set of abstracts. We conducted an experiment to compare participants' recall and recognition performance in three different conditions: a control condition without a pre-specified strategy to test baseline individual memory ability, a condition using an image-based variant of a mnemonic called a "memory palace," and a condition using a virtual reality-based variant of a memory palace. Our analyses show that using a virtual reality-based memory palace variant greatly increased the amount of knowledge retrieved and retained over the baseline, and it shows a moderate improvement over the other image-based memory palace variant. Anecdotal feedback from participants suggested that personalizing a memory palace variant would be appreciated. Our results support the value of virtual reality for some high-level cognitive tasks and help improve future applications of virtual reality and visualization.

We investigated the effects of example-based explanations for a machine learning classifier on end users' appropriate trust. We explored the effects of spatial layout and visual representation in an in-person user study with 33 participants. We measured participants' appropriate trust in the classifier, quantified the effects of different spatial layouts and visual representations, and observed changes in users' trust over time. The results show that each explanation improved users' trust in the classifier, and the combination of explanation, human, and classification algorithm yielded much better decisions than the human and classification algorithm separately. Yet these visual explanations lead to different levels of trust and may cause inappropriate trust if an explanation is difficult to understand. Visual representation and performance feedback strongly affect users' trust, and spatial layout shows a moderate effect. Our results do not support that individual differences (e.g., propensity to trust) affect users' trust in the classifier. This work advances the state-of-the-art in trust-able machine learning and informs the design and appropriate use of automated systems.

Recent visualization research efforts have incorporated experimental techniques and perceptual models from the vision science community. Perceptual laws such as Weber's law, for example, have been used to model the perception of correlation in scatterplots. While this thread of research has progressively refined the modeling of the perception of correlation in scatterplots, it remains unclear as to why such perception can be modeled using relatively simple functions, e.g., linear and log-linear. In this paper, we investigate a longstanding hypothesis that people use visual features in a chart as a proxy for statistical measures like correlation. For a given scatterplot, we extract 49 candidate visual features and evaluate which best align with existing models and participant judgments. The results support the hypothesis that people attend to a small number of visual features when discriminating correlation in scatterplots. We discuss how this result may account for prior conflicting findings, and how visual features provide a baseline for future model-based approaches in visualization evaluation and design.

Insider Threat Detection is one of the greatest challenges for organizational cybersecurity [2]. In this paper, we designed and evaluated visually compressed cyber event sequence (ViC2ES) to assist analysts with building mental models about user activity for Insider Threat Detection. Our visualizations, which show user activity on a daily level, are purpose-built to be embedded in our in-house active learning tool called "CHISSL." [3], [4] We explored different visual compression techniques with binning or run length encoding, resulting in four unique designs built upon the same icon array presentation. We evaluated these four designs for both low-level and high-level tasks in two experiments: in Experiment I, participants performed perceptual tasks such as selecting the most and least similar activities for each of the designs; in Experiment II, participants used one of the designs in CHISSL for eleven reasoning tasks. The results suggest that participants preferred the high level of aggregation, but made the fewest errors with the low level of aggregation; they were able to interact with CHISSL and accomplish the tasks using both designs. We believe our aggregated designs are effective regarding both task performance and screen space; the high and low levels of aggregation designs are valid for user activity modeling.

Despite years of research yielding systems and guidelines to aid visualization design, practitioners still face the challenge of identifying the best visualization for a given dataset and task. One promising approach to circumvent this problem is to leverage perceptual laws to quantitatively evaluate the effectiveness of a visualization design. Following previously established methodologies, we conduct a large scale (n=1687) crowdsourced experiment to investigate whether the perception of correlation in nine commonly used visualizations can be modeled using Weber's law. The results of this experiment contribute to our understanding of information visualization by establishing that: (1) for all tested visualizations, the precision of correlation judgment could be modeled by Weber's law, (2) correlation judgment precision showed striking variation between negatively and positively correlated data, and (3) Weber models provide a concise means to quantify, compare, and rank the perceptual precision afforded by a visualization.