Experimental Design
The experimental design aims at identifying the workers’ and firms’ determinants of AI-adoption. The goal is to estimate what specific characteristics of AI applications fosters or hampers willingness to implement. The key features of the experimental design follows Maestas et al. (2023).
Each worker (firm) respondent participates in 10 (5) stated-preference experiments. In each experiment, survey respondents are asked to select between two AI applications (A and B), each defined by a partially varying set of attributes.
The hypothetical AI applications A and B differ in the following characteristics. The attribute levels are randomly chosen for each AI application.
Characteristics [levels]:
Area of Application: [Internal company process, Customer product/service]
Functionality of the AI: [Identifies patterns, sorts data, Predicts future developments, Creates new content such as texts]
Origin of the AI: [Developed internally, Sourced externally]
Role of the Human: [Active decision-maker/supervisor, Passive observer]
Initiator in the Company: [Primarily employer, Primarily employee]
Main Advantage: [Increased efficiency, Improved quality, Enhanced safety, Greater convenience, Cost reduction, Increased flexibility]
Total Costs: [Low, Medium, High]
Cost Predictability: [Highly predictable, Somewhat predictable, Difficult to predict]
In every AI application choice, respondents see two applications next to each other where two randomly selected job attributes vary randomly. Those three attributes are marked in red and all the other job attributes are the same as in their current job. The respondents were asked to select which one of the AI applications they would prefer.
Reference: Maestas, Nicole, Kathleen J. Mullen, David Powell, Till von Wachter, and Jeffrey B. Wenger. 2023. "The Value of Working Conditions in the United States and the Implications for the Structure of Wages." American Economic Review, 113 (7): 2007-47.