RAPID: Analyzing forced habit change from COVID-19 using large-scale data

在COVID-19大流行期间，人们被迫在学校、工作和家庭中尝试新的常规-“被迫探索”。也就是说，习惯用来指导我们如何工作，进行日常生活，锻炼，吃饭，上学，与朋友和邻居互动，习惯的形成是为了让同样的日常生活毫不费力，节省时间和精力，当日常选择工作良好时。然而，虽然习惯带来的心理节约是一种好处，但当人们习惯性地选择时，他们可能不会探索其他可能更好的选择-这就是习惯的隐藏成本。强迫探索实际上是有益的，如果它向人们展示了更好的学习，工作和社会交往方式。这种探索就像去你最喜欢的餐馆，发现他们已经用完了你最喜欢的菜-现在你必须尝试一些新的东西，如果没有中断，你就不会探索。这一波被迫的探索提出了重要的问题：什么样的新习惯会形成并持续下去--什么将是“新常态”？例如，考虑在房子外面戴口罩。这正是一种“肌肉记忆”的行为，通常habitizes-它可以触发时，走出你的车，或进入一个商店，并迅速成为自动和毫不费力。是否有很多人戴着口罩也可能是一个触发因素，促使人们养成习惯（无论是哪个方向）。同样的问题也会出现在所有人身上：人们会回到电影院（还是呆在家里看流媒体）？餐馆会完全重新开放还是送货上门？知识型企业会转向更远程的“远程工作”吗？学校会找到更好的家庭学习和校内活动的混合体吗？该项目将分析两种不同的大数据，以测试这种强迫性探索是否真的会导致新的习惯。在社会科学中，习惯通常是用一个简单的方程来数学建模的，即过去做的活动越多，未来做的活动就越多。这被称为“简化形式”方法，因为它将生物学上复杂的机制简化为简单得多的机制。这是一个很好的起点，但不能回答诸如“如果过去的行为被打乱了怎么办？”这项研究项目使用了一种基于动物学习和人类认知神经科学的新方法来研究习惯。出发点是，习惯已经形成，以节省体力和精神上的努力。这里提出的“神经自动驾驶仪”框架预测，个人发展的行为习惯，经过反复的决定，已被证明是可靠的回报。这种习惯性的行为消耗较少的体力和精神资源。与此同时，当人们习惯于运动、饮食或工作时，他们会忽视他们实际尝试过的新商品和活动。虽然神经自动驾驶方法已经在许多动物和人类习惯化的实验室研究中进行了测试，但从未使用大量关于人们在日常生活中实际行为的数据对其进行过系统的探索。对这个模型的理想测试是在一个选择集被人为截断的现场设置中，这样人们就会求助于新的选择;而这正是正在进行的封锁期间发生的事情。该项目将使用来自微博聊天数据和Fitbit健身和睡眠跟踪的数据。这些大型数据集包含细粒度的行为度量。使用这些数据，我们将开发和测试一个统计神经自动驾驶仪模型，以恢复模型的主要参数值。这些参数是衡量每个人习惯形成的速度以及打破习惯并探索可能更好的东西的阈值的数字。估计的参数值将用于预测在大流行期间养成的习惯将持续下去，以及哪些行为将恢复到大流行前的模式。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

During the COVID-19 pandemic, people were forced to try new routine at school, work, and home — “forced exploration”. That is, habits used to guide how we work, conduct our daily routine, exercise, eat, go to school, and interact with friends and neighbors, Habits are formed to make the same routines effortless, to save time and mental effort, when routine choices work well. However, while the mental savings from habits are a benefit, when people are choosing habitually, they may not be exploring other options which could be even better— that is the hidden cost of habit. Forced exploration can actually be beneficial if it shows people better ways of school, work, and social interaction. This kind of exploration is like going to your favorite restaurant and finding out they have run out of your favorite dish — now you have to try something new, which you would not have explored without the disruption. This wave of forced exploration raises important questions: What new habits are formed that will persist— what will be the “new normal”? Consider, for example, wearing a face-mask outside of the house. This is exactly the kind of “muscle memory” behavior that usually habitizes— it can be triggered while stepping out of your car, or entering a store, and quickly becomes automatic and effortless. Whether a lot of other people are wearing masks or not can also be a trigger that prompts habit (in either direction).The same question arises across the board: Will people go back to movie theaters (or stay home with streaming)? Will restaurants fully reopen or will home delivery take over? Will knowledge firms switch to more remote “tele-work”? Will schools find better mixtures of home learning and in-school activity? This project will analyze two different kinds of big data to test whether or not this kind of forced exploration really does result in new habits.In social sciences, habits are usually modelled mathematically using a simple equation in which the more an activity has been done in the past, the more it is done in the future. This is called a "reduced form" approach because it reduces a biologically complicated mechanism to something much simpler. It is a good starting point but cannot answer questions such as "What if past behavior is disrupted?” This research project uses a new approach to habits based on animal learning and human cognitive neuroscience. The starting point is that habits have developed to save effort ⎯— both physical and mental. The “neural autopilot” framework proposed here predicts that individuals develop habits for actions which, after repeated decisions, have proven to be reliably rewarding. Such habitual behavior drains fewer physical and mental resources. At the same time, when people are habitized⎯- about exercise, eating, or work — they ignore new goods and activities they would prefer if they actually tried them. While the neural autopilot approach has been tested in many lab studies of animal and human habituation, it has never been systematically explored using a large amount of data about how people actually behave in everyday life. An ideal test of this model is in a field setting where choice sets are artificially truncated, so people resort to new choices; and that is exactly what happened during the ongoing lockdowns. This project will use data from Weibo chat data and Fitbit fitness and sleep tracking. These large sets of data contain fine-grained measurements of behavior. Using this data, we will develop and test a statistical neural autopilot model, to recover values for the model’s main parameters. The parameters are numbers that measure, for each person, how fast habits are formed and the threshold to break out of a habit and explore something that might be better. The estimated parameter values will be used to make predictions about which habits acquired during the pandemic will persist, and which behavior will revert to pre-pandemic patterns.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.