projects

The COVID-19 pandemic has affected the public health, business activity, and life of individuals on an unprecedented scale. In March 2020, the Dutch government introduced a number of measures in order to contain and control the virus spread. Their success relied to a great extent on citizens’ collective mindset and willingness to adhere to the new rules and standards. As the government currently explores new public health surveillance technology, such as through so-called “corona apps”, it is important to understand the limits of citizens’ willingness to accept governmental interference in their daily life for the reward of the common good. Solutions that require sharing sensitive health data touch upon not only the issues of political trust and trade-offs between individual freedom and governmental authority, but also upon people’s perceptions of technology (e.g., trust in technology, transparency of data usage, privacy considerations, effectiveness, etc.). This study maps people’s attitudes in all these key areas. This way it gives an insight into people’s willingness to share potentially sensitive data with the government and private companies for the sake of the health and wellbeing of vulnerable groups in society. In particular, it helps to understand which personal and background characteristics, which type of health data, and which conditions affect this willingness. Extending the previous survey, we suggest to identify factors that determine whether people are willing to share personal information with the government for the benefit of collective public health using a Big Data approach. With an eye on policy makers considering ideas for new health surveillance technology, we aim to explore people’s perceptions of data sharing, in relation to their perceptions (e.g., trust) of relevant factors, such as the government, politicians, experts, technology, and the media.

We expect that the decision to share personal information depends on a number of factors. One of these is the perception towards the technology used: will it be able to safeguard users’ privacy? Will it be resilient against hackers? An additional factor may be people’s trust in the government, pertaining to questions as: Will personal information be handled correctly? Will it be used for the specified purposes only? Etc. These perceptions are to a large extent based on what people learn from various sources, such as newspaper articles, public figures, and experts, but also “important others”, e.g., friends, neighbours, and vocal lay persons. The impact of information from most of these sources on individuals and groups within society, in turn, greatly depends on their perceptions of the channels through which this information became available to them – the media. Interestingly, for some “the media” represent merely a range of information channels, whereas for others they form an institution in and of itself that often warrants distrust (MSM; “main-stream media”).

With the help of the BMS’ Behavioural Data Science Incubator (BDSI), we will develop a big-data approach to analyse the influence of these and other factors on the willingness to share personal information. Publicly available responses of people on Internet forums or social media such as Twitter will be mined, categorized, and labeled. Subsequently, each labeled response will be subjected to valence analyses (i.e., to assess how postivepositive or negative these responses are) and sentiment analyses (to what extent these responses reveal emotions such as anger).

The proposed study is a large-scale quantitative sentiment analysis of the public discourse on the Covid-19 vaccination in the Netherlands based on big data, scraped from both social and traditional media. The sentiment analysis will provide insights in the legitimacy of a vaccination in terms of expectations (positive or negative), emotions behind these expectations (anger, fear, sadness, disgust, surplice, anticipation, trust, joy), and the prominence and contents of the different pillars of legitimacy (cognitive, normative, pragmatic, regulative). As Covid-19 has a great influence on society and vaccines seem to be the only solution for ending the pandemic, it is important to study the prominence and strength of negative or positive feelings regarding a potential vaccine at an early stage, whether they are included in the mainstream debate, and how the public discourse is developing in this regard. This study provides crucial information to policymakers about how to inform the general public about vaccinations for Covid-19. By insights in how the public discourse is shaped along the pillars of legitimacy, the most prominent sentiment and underlying emotions, policy makers can proactively shape their communication and information campaigns for gaining public support for a vaccine.

References

Acklin, M. (2020, 12 May). Most Americans would opt for coronavirus vaccine, but some are wary. Retrieved from Civic Science: https://civicscience.com/most-americans-would-optfor-coronavirusvaccine-but-some-are-wary/

Binz, C., Harris-Lovett, S., Kiparsky, D., Sedlak, L. D., & Truffer, B. (2016). The Thorny Road to Technology Legitimation – Institutional Work for Potable Water Reuse. Technological Forecasting & Social Change, 103, 249–263.

Chang, B. (2020, 23 May). More than 40% of Republicans think Bill Gates will use Covid-19 vaccine to implant location-tracking microchip in recipients according to survey. From Business Insider: https://www.businessinsider.nl/republicans-bill-gates-covid-19 vaccine-tracking-microchip-study-2020-5/

Chafale, D., & Pimpalkar, A. (2014). Review on developing corpora for sentiment analysis using Plutchik’s wheel of emotions with fuzzy logic. International Journal of Computer Sciences and Engineering, 2(1), 14-19.

Clarke, C. E. (2011). A case of conflicting norms? Mobilizing and accountability information in newspaper coverage of the autism-vaccine controversy. Public Understanding of Science, 20(5), 609-629.

D’Souza, G., & Dowdy, D. (2020, 10 April). What is herd immunity and how can we achieveit with Covid-19. Retrieved from John Hopkins School of Public Health: https://www.jhsph.edu/covid-19/articles/achieving-herd-immunity-with-covid19.html

Dun, A. G., Surian, D., Leask, J., Dey, A., Mandl, K. D., & Coiera, E. (2017). Mapping information exposure on social media to explain differences in HPV vaccine coverage in the United States. Vaccine, 35, 3033-3040.

Hoffman, B. L., Felter, E. M., Chu, K. H., Shensa, A., Hermann, C., Wolynn, T., Williams, D., & Primack, B. A. (2019). It’s not all about autism: The emerging landscape of anti-vaccination sentiment on Facebook. Vaccine, 37, 2216-2223.

Jansma, S.R., Gosselt, J. F., & De Jong, M. D. T. (2020). Technology legitimation in the public discourse: applying the pillars of legitimacy on GM food. Technology Analysis & Strategic Management, 32(2), 195-207.

Johnson, N. F., Velásquez, N., Restrepo, N. J., Leahy, R., Gabriel, N., El Oud, S., Zheng, M., Manrique, P., Wuchty, S., & Lupu, Y. (2020). The online competition between pro- and anti-vaccination views. Nature, 1-7.

Rutjens, B. T. (2020). Spiritual skepticism? Heterogeneous science skepticism in the Netherlands. Public Understanding of Science, 29(3), 335-352.

Tempelman, O. (2020, 8 May). Dit zijn de zeven zondebokken van de coronacrisis. Retrieved from De Volkskrant: https://www.volkskrant.nl/nieuws-achtergrond/dit-zijn-de-zeven zondebokken-van-de-coronacrisis~bc609ea5/

Yaqub, A., Castle-Clarke, S., Sevdalis, N., & Chataway, J. (2014). Attitudes to vaccination: A critical review. Social Science & Medicine, 112, 1-11.

The race is on to develop smart and sustainable solutions to today’s most pressing challenges. Congestion, air pollution, rising energy costs, and public safety are just a few of the challenges developers of connected and automated vehicles (CAVs) claim to address. CAVs have captured the imagination of a whole host of stakeholders and sparked a new “gold rush” in the automotive industry.

In its wake, this emerging technology has given rise to a yet uncharted regulatory landscape within which governments must fulfill their regulatory role while still enabling further innovation on CAVs. Achieving such a delicate balance is tricky, making this an opportune time for research on the governance of CAVs.

Enter our collaborative project, which will track how the safety discourse on CAVs changes over space and time and will interrogate the role that governments play in the development and deployment of emerging technologies such as CAVs. To that end, our partners at the BDSi have leveraged machine learning to build an original dataset of newspaper articles on this topic.

We are currently exploring and applying diverse approaches to analyzing these data including automated text mining, topic analysis, sentiment analysis, and inferential and Bayesian statistical methods. This pilot is the foundation on which an international funding application will be based. Our team includes (in alphabetical order): Karel Kroeze (UT-BMS), Dasom Lee (KAIST), Le Anh Long (UT-BMS), and Anna Machens (UT-BMS). Please feel free to contact us if you’d like to learn more about our work.

Eye Tracking is a powerful method for understanding human attention, visual processing, problem solving and preferences. It is useful in a variety of applied research domains, such as Human Factors, Communication Science, Marketing and Education. Eye tracking is expected to gain even more significance with the emerge of VR/AR systems in research and training.

BMS Lab is developing the GazeFlow system, which turns a regular webcam into a low-resolution eye tracking device. This budget-friendly system allows us to equip all our students with a working eye tracking system and has been piloted for teaching in two bachelor modules. Based on this and earlier experience, we will create a multi-disciplinary educational framework for eye tracking in BMS education, Etmerald (Eye Tracking Made Easy)

goals of the project

• Developing a sequence of eye tracking learning units, spanning from first year bachelor to master level (CODE).
• Further develop GazeFlow, a budget-friendly eye tracking platform based on webcams (BMSLab).
• Create a Data Science workflow and an easy-to-use R package for eye tracking data (BDSI).
• Facilitate broad adoption of eye tracking within BMS education (PCRS, ETM, CS).

We will produce the necessary tools and create case studies, documentation, teacher guidelines, learning material for eye tracking on first year, second year and master level. The approach will be implemented and evaluated in three existing courses: B-Psy Module 3 (Cognition and Development), module Human factors & Engineering Psychology and M-Psy ARM (Advance Research Methods HFEP).

The goal long-term is to create eye tracking spin-offs in at least three other BMS programs/themes. To keep an eye on the big picture, we include stakeholders from two other programs (CS and ETM) and one other psychology theme (PCRS). We will seek to get in contact with more potential adopters during the project.

benefit to education and/or students

Familiarity with technology is giving our social science graduates a unique edge on the labour market. With our project we want to secure this advantage for the future. Eye tracking is a standard technique in diverse fields of applied and industrial research and experts are highly regarded.

Before it was called Data Science, social scientists were often hired as experts in data analysis (because they know how to operate SPSS). With the new paradigm, the job market is exploding, but the niche for social scientist is even so quickly eroding. Thank to our modernized Statistics curriculum, this niche is secured at least for our students. By interfacing eye tracking with R, we give our students more opportunities to develop professional Data Science skills in a fun way.

The Etmerald framework can be used to facilitate learning in the classroom by creating first-hand experience on theoretical topics, such as visual attention, face recognition and bottom-up processing, as well as applied topics, such as interface design, data visualization or marketing of products.

The GazeFlow system can be used in online education and is therefore robust to lockdowns.

Data visualisation refers to the techniques used to communicate data through visual objects (points, lines or bars) in graphics. Currently, data visualisation is not systematically taught in skills lines in the various BMS programs. The goal of this project is to bring together and further develop teaching materials to enable and stimulate BMS students to find and use novel data, and to visualise these data in engaging ways using R and related programming languages.

BDSi’s contribution to this project comes in the form of an app to aid students in selecting, understanding and crafting visualizations that best explains their data and analyses.

Henk van der Kolk explains the project, and presents an early preview of the app in the below video. A continuously updated development version of the app is available at https://karel-kroeze.shinyapps.io/henk-vis-browser/.

In this project, we professionalise existing BMS educational material and extend it to include online web apps. We also aim to empower BMS teachers to make their own web apps.

Data analytics is taught using the framework of linear models. Because of a lack of textbooks on this topic suitable for behavioural and social science students, we’ve written our own textbook, available free of charge to anyone. The book could benefit from the professionalisation of presentation, layout, and availability in other formats than pdf, making it more accessible to both students and teachers. The approach used in the book could benefit from web applications (shiny apps) that could explain important concepts in an interactive and visual manner. We would like to integrate such apps directly into the textbook online, for instance using HTML rather than pdf. It would both benefit teaching and students. By teaching how to make web apps we can help teachers both to keep updating the material in the future and to make other apps that support their own teaching. It will involve Teaching professionalization, Talent development of students, and Learning facilities. It strengthens hybrid education by making the material accessible online.

Covid-19 antibody tests used to identify whether an individual has been infected with SARS-CoV-2, can be employed to estimate the prevalence of this disease on a more aggregate or population level. If, for instance, such a test is taken among a number of Health Care Workers (HCW) in a well-specified region, then a sufficiently large sample suffices for a reasonably accurate estimation of the infection rate (proportion) among all HCWs in that region. The central ideas are that investigating a sample is quicker, cheaper and uses fewer scarce resources (e.g., the tests themselves) than investigating the entire population of interest. Moreover, the evolution in the rate can be monitored by taking samples distributed over time. The knowledge in the form of qualitatively sound estimates obtained may be relevant for policy decisions, for instance, in the case that a second wave is approaching, we may want to have an informed idea about the speed of the spreading of the virus as a crucial input for a second lockdown decision. Later on, we might want to have an idea about whether thresholds regarding the decline of the speed of spreading, or even the famed herd immunity, have been reached in order to end a lockdown.

The general public, and even medical professionals, may suffer from a series of misinterpretations when faced with test results (cf., e.g., Uffrage et al. [2000], Bar-Hillel [1980]), and debunking these requires insights from the rather counterintuitive field Bayesian statistics. To keep it simple, the proportion of people receiving a positive test result is not without reservations to be extrapolated to estimate the rate of the total population. To obtain reliable estimates regarding the prevalence the reliability (reliabilities) of a test must be taken into account as well.

We have designed a procedure based on Monte Carlo simulation to estimate rates and confidence intervals, and applied the procedure with success, albeit to experimental cases (cf., Joosten & Abhishta [2020]). The inputs for the MC simulation base procedure are quite simple, the test results need to be known, the sample size is required, and we need the reliability data regarding the test itself, and the procedure will provide several useful point estimators and a confidence interval of desired precision for the prevalence rate. Moreover, the confidence intervals can be used fruitfully to determine whether the sample size suffices, or should be increased for real world tests in order to enhance precision of the estimators. A documentation of this procedure and results of small scale tests can be found here.

The Monte Carlo simulation approach is quite robust to alternative assumptions and specifications regarding the reliability of the test. We are near completion of a paper on how to deal with stochasticity in the test reliabilities, as in real world situations numbers on test reliabilities derive from validations of the tests themselves. Hence, the reliability numbers provided by producers, governmental reviewing agencies (e.g., FDA in the USA or RIVM in the Netherlands) or others, are to be viewed as stochastic variables themselves, which have their own distributions, means and variances.

Our procedure and the estimators are still upheld, albeit that we have to add some layers of simulation in order to deal with the additional (but more realistic) assumptions regarding the stochasticity inherently contained in the test reliability numbers.

While deepening our knowledge regarding the topic of our research on behalf of our two papers mentioned, we discovered other items of interest regarding the nature of test reliabilities. Potential problems resulting from these peculiarities can presumably be tackled by means of our procedures or necessary, yet relatively minor adaptions of them. For the near future, we envision to be able to solve the problem of finding reliable estimators and confidence intervals for the situation that reliability numbers correlate systematically with the rate of prevalence of a disease.

Our, thus far limited, experiences show that the computational requirements for our Monte Carlo base procedure increase quite considerably in every step incorporating more realistic scenarios. We have thus far, provided our computer code for free on the internet as an online tool for practitioners to find estimators, confidence intervals and implicitly to determine sufficient bounds for their sample sizes. We noted a keen interest in some communities for the procedure, but our more extensive versions require more sophistication not only on our part but also on the side of the intended users.

To summarize our plans: we want to continue our investigations regarding the use of small-scale sampling for estimating population level prevalence rates of Covid-19 by means of Monte Carlo estimation-based procedures, and to assist practitioners and decision makers by providing our procedures as computational tools (almost) ready to use.

References

Bar-Hillel, M, 1980, The base-rate fallacy in probability judgments, Acta Psychologica 44, 211-233.

Joosten, R & A Abhishta, 2020, A simulation-based procedure to estimate base rates from Covid-19 antibody test results I: Deterministic test reliabilities, Working paper University of Twente, MedRxiv 10.1101/2020.04.28.20075036.

Hoffrage U, S Lindsey, R Hertwig, G Gigerenzer, 2000, Medicine. Communicating statistical information, Science 290, 2261-2262, DOI: 10.1126/science.290.5500.2261.

Creating a recommender system for suggesting research topics and relevant thesis supervisors based on the professional ambition and educational path of the student.

Today, the allocation of thesis supervisors is done on an ad-hoc basis by every study program within BMS. We can broadly categorize thesis projects into two categories based on the process followed to “match” students with thesis supervisors:

• Projects where students first look for an assignment at a company and then approach a university supervisor (e.g. IEM);
• Projects where students are free to either do an assignment at a company or focus on solving an academic research question with the help of a university supervisor (e.g. BIT).

In both of these cases students have little knowledge about the possibility of projects available or the expertise of the university supervisors. When students look for company assignments, they are restricted by the companies they already know or reach out to. Many times this leads to thesis assignments that may not directly confirm with the study program or future job aspirations of the student. We explain in more detail the shortcomings of the current process while describing the potential benefits of our project below. In order to provide an overview of available project opportunities to the students, we propose SmarT rEsEarch Recommender System (STEERS), a recommender system for students to help them choose or to propose a thesis project based on their educational path and future career goals.

This project has the following three goals:

1. To valorize and expand the use of the existing repository of past thesis projects by scraping and linguistically linking its data on the thesis topics, supervisors, companies, and/or study programs (among others).
2. To create a dashboard with insights on network associations between research topics, companies, supervisors, students, and study programs.
3. To create a working prototype of a recommender system that helps in matching appropriate supervisors with students for bachelor and master thesis projects.

STEERS will provide students with adequate information to brainstorm on relevant thesis topics that can help them achieve their career goals. As students will have an overview of past supervision and potentially company involvements for each thesis topic, they will be able to reach out to appropriate university supervisors and/or companies. This will especially help students working on interdisciplinary topics.

Last year, BDSi created YET (is your eye tracker), a capable, DIY eye tracking device, that many students have used for their own projects, since. This year, we are back with: YLab (is your Lab)

YLab is funded as a WSV Innovation in Education project to create a student-friendly platform for programmable sensor arrays, with focus on physiological signals. Physiological signals can be used in many practical research situations to measure activity, workload or stress, but also to study patterns in learning new skills. This fall, YLab was born, including support for electrodermal activity, electrocardiography and 3DoF motion capture. We have also added moment-of-interest recording (MOI), which allows collecting self-report measures.

With the current hardware choices, a setup with EDA and ECG sensors can cost less than a textbook. Beyond the price tag, YLab is designed to promote ownership in terms of command. With the YLab packages a useful application can be written in under 50 lines of Python code, which makes for a fun way to teach first programming courses. As of writing, 70 Psychology students are doing their first steps in Python on YLab, and by the end of the quarter, some capable research instruments will be in capable hands.

YLab comes as a set of four Python packages and is available on GitHub, which all include some demos and examples. It is programmed for CircuitPython for the widely available RP2040 microcontroller.

• Sensory (is your sensor array) unifies the process of connecting to and sampling from sensors
• YUI (is your user interface) makes creating user interfaces easy
• Ydata (is your data) provides a simple and secure way to store sensor data
• Ynot (is your network of things) will allow connecting programmable sensors to form larger arrays

The second phase of the project is ahead of us and will be all about bringing physiological measures to the classroom as a first-hand experience. We will create a set of standardized learning material, making it easy to integrate the matter in your courses. Multiple student projects and a challenge-based teaching course will further explore the possibilities of YLab.

For further questions regarding the project, contact m.schmettow@utwente.nl