Fuhai Hong, Tanjim Hossain, John A. List, Migiwa Tanaka. 2013. Testing the theory of multitasking: Evidence from a natural field experiment in Chinese factories. Working Paper 19660. National Bureau of Economic Research
Abstract: A well-recognized problem in the multitasking literature is that workers might substantially reduce their effort on tasks that produce unobservable outputs as they seek the salient rewards to observable outputs. Since the theory related to multitasking is decades ahead of the empirical evidence, the economic costs of standard incentive schemes under multitasking contexts remain largely unknown. This study provides empirical insights quantifying such effects using a field experiment in Chinese factories. Using more than 2200 data points across 126 workers, we find sharp evidence that workers do trade off the incented output (quantity) at the expense of the non-incented one (quality) as a result of a piece rate bonus scheme. Consistent with our theoretical model, treatment effects are much stronger for workers whose base salary structure is a flat wage compared to those under a piece rate base salary. While the incentives result in a large increase in quantity and a sharp decrease in quality for workers under a flat base salary, they result only in a small increase in quantity without affecting quality for workers under a piece rate base salary.
摘要：在有关“多任务”的著述中，一个被熟知的问题就是工人可能由于寻求在产出可被观察到的任务上的奖励而实际上削减在产出不好被观察到的任务上的努力。由于与“多任务”相关的实证研究仍旧是几十年前的，多任务情景下标准化激励计划的经济成本仍然是个极大的谜团。本研究给出了实证的洞见，通过在中国工厂中进行实地实验量化这样的影响。通过使用来自126名工人的超过2200个数据，我们找到了强有力的证据——工人们的确会由于计件工资奖金分成计划牺牲不能被激励的产出（质量）而换取能够被激励的产出（数量）。与我们的理论模型一致的是，平基薪的工人比计件工资的工人的反应更为强烈。激励导致了平基薪工人工作成果在数量上大幅提升并在质量上大幅下降，而它们对计件工资工人的影响仅仅是在不影响质量时，数量的小幅增长。
数据来源：More than 2200 data points across 126 workers
研究方法：实验研究，计量回归
Conclusion: Principal-agent models have become the workhorse framework for modeling asymmetric information settings. In the field, when the agent cannot be certain that a dimension of output is closely monitored, the classic multitasking theory applies. This study provides empirical insights into the multitasking problem By making use of a unique naturally-occurring setting: incentive contracts for workers in China on the floor of various factories. Through our interactions with managers at the plants, we were able to implement a natural field experiment in several plants to explore basic questions within the classic principal-agent setting.
Our main results paint an intriguing picture. First, the first order predictions of the theory are found in our data: as we incent workers on the margin, they move their effort to the incented activity to the detriment of the non-incented one. But, there is an important caveat to this result: we only find this result amongst workers who were previously working under a fixed wage scheme. For those workers previously under a piece rate scheme, their output moves By a small magnitude when we introduce a performance incentive. Our results suggest that, in structural estimation of principal agent models, simplification of the model into a single task dimension when the task is better described By multiple dimensions may yield biased estimates of structural parameters.
Second, we report an interesting Hawthorne-type effect. We find that a simple reminder letter to workers leads to a robust and economically significant increase in worker productivity. In this way, the data are consonant with a Hawthorne effect in Chinese manufacturing plants. Notably, while the actual Hawthorne data from the original experiments do not stand up to closer scrutiny, data from the Chinese factories do. Finally, an overarching lesson learned from this exercise is that one can gain enough control in a field environment to test important theories of multitasking incentive schemes.
结论：委托代理模型已经成为了非对称信息环境建模的研究框架。在这个领域，当代理人不能确定产出的某个维度是被密切监控的，经典多任务理论就能够适用。本研究利用独特的自然发生环境——许多中国工厂中工人的激励合同——提供了关于多任务问题的实证性洞见。通过我们与车间经理的互动，我们能够在多个车间进行实地实验，在经典委托代理环境下去探索基本问题。
我们的主要成果形成了一幅有趣的图画。首先，该理论的首要预言在我们的数据中被发现：我们在边际上激励员工，他们就会将牺牲不被激励的活动而把努力转移到被激励的活动上。但这里有一点值得注意：这个结论仅仅适用于之前在固定薪酬制度下工作的员工。对那些在计件工资制度下工作的工人，当我们引入绩效激励时，他们的产出移动了一个很小的量级。我们的结论表明：在委托代理模型的结构化估测中，当任务应该使用多维度描述时，把模型简化为单一任务维度也许会导致结构化参数的估测误差。
第二，我们发现了一个类霍桑效应。我们发现给工人的一封简单的提醒信可以导致工人生产率的大爆发与经济上的显著增长。如此，数据与中国制造车间中的霍桑效应异曲同工。值得注意的是，当来自原实验的真正的霍桑数据不能详查时，我们可以查看来自中国工厂的数据。最终，我们在这个实践中学到的最重要的是：一个人能够在实地实验中获得足够的控制，以检验多任务激励计划的重要理论。
（By 潘一坤) |
