For Students : >>> open thesis projects <<<

I am fascinated by the diversity of human individuality and am dedicated to the study of how people differ in their behavior, their perception of the world, and their decisions. I my work, I investigate:

  1. how technology can be used to study human behavior, experience, and preferences objectively

  2. how AI/machine learning can be used to statistically recognize individual differences

  3. how stable user traits and momentary states can be considered in the design of intelligent systems and services to support people in their everyday lives (personalization).

  4. the consequences and implications that arise from the widespread use of digital behavioral data in algorithmic decision making, for individuals and our societies as a whole.

In my work, I combine methods from the social sciences (e.g., questionnaires, experience sampling) with more technical approaches from computer- and data science (e.g., mobile sensing, web scraping, APIs, machine learning). This allows me to objectively quantify and study everyday behavior, situational characteristics, and psychological processes (e.g., cognition, affect) with a data-driven approach.

My research interests in social media language: #behavior, #mobilesensing, #machinelearning, #personality, #privacy, #personalization

Using Technology to Study Human Behavior - Mobile Sensing

In the last century, it has been almost impossible to collect ecologically valid data to study people's daily behaviors and environmental characteristics in an objective and scalable way. However, the pervasiveness and the technical capabilities of modern consumer electronics (e.g., smartphones, cars) in terms of sensors, computational abilities, and connectivity now make it possible to do so. Digital behavioral data will help to overcome some well-known limitations of questionnaire-based research in the social sciences (e.g., response styles, social desirability). It will help to study a wider range of populations unobtrusively, in the wild (more than 90% of people in developed and more than 80% in developing countries own a smartphone).

I am the founder of the PhoneStudy mobile sensing research project. While I led the project from 2013-2020, Ramona Schödel now continues to spearhead the mission of the project: To create a method for the objective quantification of human behavior and its environments. In the project, we use a mobile sensing Android application to automatically collect longitudinal data about daily behaviors, environments, and situations (e.g., communication, app-use, location and physical activity, media use, speech parameters, language use).

We used these data to investigate how the usage of different types of smartphone applications is related to personality traits, demographics, and cognitive abilities (Stachl et al., 2017). Results of this study suggest that extraverted people tend to make more but shorter calls, conscientious people play games less often on their phones, and more agreeable people use public transportation apps more frequently. In a series of follow-up studies in this line of research, we helped to map out relationships between personality traits and sociability behaviors in the wild (Harari et al., 2019; Journal of Personality and Social Psychology) and related social behavior and cognition to autistic traits (Schuwerk et al., 2019; Journal of Autism and Developmental Disorders). Moreover, we investigated the behavioral factors of sensation seeking as a personality trait (Schoedel et al., 2018), and explored day- and nighttime activity with mobile sensing data (Schoedel et al., 2020; European Journal of Personality).


For my work on mobile sensing in personality psychology, I received the Award for Digital Assessment from the German Society for Personality Psychology and Psychological Assessment.


Using AI/Machine Learning to Model Individual Differences

Behavioral Pattern Plot: Shows the ranked importance of different behaviors for the prediction of personality traits.
Stachl, C., Au, Q., Schoedel, R., Samuel. D. Gosling, Gabriella. M. Harari., Buschek, D., Völkel, S., Schuwerk, T., . . . Bühner, M. (in press). Predicting Personality from Patterns of Behavior Collected with Smartphones. Proceedings of the National Academy of Sciences of the United States of America (PNAS). Preprint: https://doi.org/10.31234/osf.io/ks4vd
Fig. and code available at https://osf.io/kqjhr/, under a CC-BY4.0 license

In a recent publication, me and colleagues demonstrated the breadth of behavioral and situational data that can be collected with smartphones and can reveal people's self-reported personality trait levels (Stachl et al., 2020b; Proceedings of the Natural Academy of Sciences of the United States of America).

In this project, we used machine learning models to predict self-reported, personality trait scores from classes of behavior (communication and social behavior, music consumption, app usage, mobility, overall phone usage, and day-nighttime activity). Additionally, we used interpretable machine learning techniques to identify which behaviors were most predictive for specific personality traits. For example, we found that conscientiousness levels were best predicted by a mix of overall smartphone use (e.g., periods of non-use), day nighttime activity (e.g., mean time of last activity) and app-use. Levels of extraversion were best predicted by communication and social behavior (e.g., average number of outgoing calls per day).



>>> EXPLORE THE RESULTS <<<




Improving Design through Personalization

In addition to my research on mobile sensing in academic settings, I have been able to closely work with engineers, data scientists, and computer scientists in the industry. In this part of my research agenda, I initially focused on the general optimization of digital systems with regard to human factors (e.g., Stachl & Bühner, 2012). In consequent research projects, I focused on the design and conceptualization of user interfaces and AI systems based on user characteristics (e.g., personality traits, demographics; Scheider et al., 2017, Schoedel et al., 2018a, Stachl et. al 2015), and user-states (Völkel et al., 2018).

From 2017 to 2020, I was leading my own team of excellent PhD students (Ramona Schödel, Sarah Theres Völkel, Quay Au) in a industry-funded research project (“User-Centered Machine Learning”) together with the Audi car company. The vision of this project was to rethink adaptive user interfaces by considering user-characteristics in the design process (e.g., demographics, personality traits, mental states) and to improve the interaction between the user and the system through automated personalization, based on these characteristics. Most recently, we finished work on a personality trait theory for artificial, conversational agents (Völkel et al., 2020)

Below, Sarah is presenting our research approach at CHI 2020:

Machine Learning Methods & AI Implications

Class-based greedy forward search plot. Shows which classes of behavior, if combined are most predictive of personality overall and across multiple resampling iterations.
Stachl, C., Au, Q., Schoedel, R., Samuel. D. Gosling, Gabriella. M. Harari., Buschek, D., Völkel, S., Schuwerk, T., . . . Bühner, M. (in press). Predicting Personality from Patterns of Behavior Collected with Smartphones. Proceedings of the National Academy of Sciences of the United States of America (PNAS). Preprint: https://doi.org/10.31234/osf.io/ks4vd (SI Appendix)
Fig. and code available at https://osf.io/kqjhr/, under a CC-BY4.0 license

In my work, I often collaborate with data scientists and statisticians. This is very rewarding, because it helps me to use latest modeling techniques most rigorously. Sometimes I get the chance to help them in the development of new methods. During the last six years, I helped to publish on the development and usage of machine learning methods (Au et al., 2018, Probst et al., 2017, Stachl et al., 2020) and in the conceptualization of a R-package for easy aggregation of grouped sensing data (Au et al. , 2019).



The critical investigation of machine learning models and the consequences of their application is a meta-research theme that bridges all my projects and that I am increasingly focusing on as a main line of research. For example, which private traits and states can be inferred from relatively rough estimates of behavior, what are the consequences of this being possible (e.g., smartphone use, Stachl et al., 2020b, PNAS)? In another recent publication, me and colleagues have critically discussed the use of machine learning methods in personality resarch and assessment, and highlighted key issues and opportunities in doing so (Stachl et al., 2020a, EJP).

Open Science

Blohowiak, B. B., Cohoon, J., de-Wit, L., Eich, E., Farach, F. J., Hasselman, F., … Riss, C. (2019, November 14). Badges to Acknowledge Open Practices. Retrieved from osf.io/tvyxz

In my research and scholarly activities, I am dedicated to follow open science principles. Specifically, I seek to maximize the reproducibility, transparency, and accountability of my work by pre-registering confirmatory work and by using reproducible methods (e.g., version control). Additionally, I make code, data, and materials openly available, whenever ethically possible. I publish pre-prints of my work on preprint servers (e.g., PsyArXiv) for other researchers to criticize and to make changes during the review process transparent. Finally, I collaborate with others to create software and processes to standardize research methods and to quantify the impact of researcher-degrees of freedom on results (e.g., multiverse-analyses, Schoedel et al., 2020).

I signed the Commitment to Research Transparency.

My profile at the Open Science Framework (OSF)