Sources of Uncertainty in Quantum and Classical Measurement Dataset

This dataset consists of participants' natural language responses to a series of questions on the Measurement Survey, a Physics Education Research assessment survey that studies physics students’ thinking about sources of measurement uncertainty across experimental contexts, including classical and quantum mechanics contexts. We consider this dataset to be especially useful for evaluating and comparing different multiclass classification machine learning techniques for two reasons: i) the text responses are well segmented into distinct ideas due to the structure of the survey question, and ii) the coding scheme applied to the data is mutually exclusive, presenting a true multiclass problem for machine learning. The format of the survey is as follows: first, the survey presents an experimental context, then, the survey prompts participants to answer questions in the context of that experimental context. The responses in this dataset come from the first version of the Measurement Survey which consisted of four experimental contexts: Projectile Motion (PM), Brownian Motion (BM), Stern Gerlach (SG), and Single Slit (SS). Each experimental context included a text description, a diagram of the experiment, and a histogram of fictitious experimental results. For each experiment, participants were asked to generate a list of sources of uncertainty in response to the prompt: “What is causing the shape of the distribution? List as many causes as you can think of.” The data are presented in long format, where each row represents an item in the lists of sources of uncertainty generated by participants. The Response Id corresponds to each unique participant who responded to the survey, therefore, rows with the same Response Id represent multiple sources of uncertainty listed by the same participant. Responses were coded by the research team based on an emergent coding scheme with four categories defined as follows: Limitations (L): Reasoning that uncertainty arises from an inability to perfectly measure and model all aspects of a real-world experiment. Physical Principles (P): Reasoning that uncertainty is inherent in the theoretical abstraction of an experiment. Statistics (S): Reasoning that uncertainty is inherent to the nature of experiments and repeating trials. Other (O): Non-empty responses that are vague or do not fall into one of the other three codes. The dataset is multi-institutional and was collected over multiple semesters. For more details about data collection and the demographics of participants, see the journal articles listed under related works or contact the creators.