Hi, I'm one of the co-authors on the study, and also a co-founder of Quantified Citizen which developed the platform used in this study. I just left a reply fur

Hi, I'm one of the co-authors on the study, and also a co-founder of Quantified Citizen which developed the platform used in this study. I just left a reply further up this thread detailing some of the logic behind our approach, and you're asking some great questions here. A simple way of thinking about this is that you're almost always making a trade off between quality and quantity when you collect data, with enough quantity you can hope to overcome some of the quality concerns. For instance our study also included a finger tapping test (analysis will be in an upcoming paper). This would have traditionally be administered in a lab by a technician by placing electrodes on a persons thumb and forefinger and asking them to tap them together. It is a validated practice for measuring Parkinson's symptoms or neuromuscular integrity. We can easily replicate something like this on a mobile app. Obviously doing it this way might result in noisier or less reliable data because there isn't a technician standing next to them helping them to get it right. Over time we have improved this by adding better instructions, addressing training effects and providing better instructions. We also collect data like screen resolution and phone model with each response so we can account for people who switch devices between tests or device-specific issues. And having datasets measured in the thousands means that a lot of these confounders come out in the wash, which isn't the case with study populations measured in the tens. We're picking up the pace on our technical development. Version 2 of the app will be out in the next week or two; we planned to get it out before this paper was published, but they surprised us with a very fast turn-around on the proofs. One of our goals is to make this type of research more agile, in a similar way that the software industry moved from waterfall project management to more rapid iterations. If the cost of designing and deploying a study is low enough, and data can get back fast enough, researchers can do more rapid pilot studies and iterations, and more quickly make adjustments to add new scales and assessments. On the subject of direct HealthKit measurements that's also a trend that we're excited about. The new version of our app will read from HealthKit and Google Fit, and we will be adding more integrations to other wearables and passively collected data with a privacy-by-design approach. We'll also be moving towards a design where users can consolidate different feeds of data and assessments that they are interested in collecting for intrinsic reasons, and then choose to make these available to different research projects.