For a long time, empirical economic research was conducted using a limited set of variables deemed sufficient to model economic behavior. Economists focused primarily on data reflecting observed choices and the objective conditions influencing those choices (such as prices and income). The prevailing view was that “preferences” could not be directly measured but only “revealed,” under certain assumptions, through choices and circumstances. This revealed-preference approach, grounded in the work of Samuelson, Arrow, Houthakker, and others, dominated the field, reinforced by debates in the 1950s among scholars such as Tobin, Okun, Juster, and Katona. While some researchers combined choice data with other measurable constructs (e.g., health or biological indicators), skepticism persisted toward variables capturing intentions, tastes, beliefs, attitudes, or subjective expectations.
In recent years, this paradigm has shifted. Experimental economists initially attempted to elicit preferences and attitudes in laboratory settings, often isolating subjects from their real-world contexts and collecting little background data. More recently, however, survey-based research has expanded to include subjective expectations, attitudes, and intentions alongside choice data. Following Manski (2004), the systematic collection of subjective expectations about future and uncertain outcomes has become widespread. Field experiments—especially in developing countries—have extended this effort, producing innovative tools to measure drivers of behavior such as tastes, attitudes, expectations, and social norms. These measures are now frequently combined with traditional economic data.
Significant methodological advances have emerged. For example, elicitation of subjective expectations has progressed from simple point forecasts to full probability distributions, with multiple elicitation techniques now in use. Despite this progress, major design challenges remain, and developing reliable measurement tools continues to require extensive experimentation and validation. Because many such tools originate in other disciplines, interdisciplinary collaboration has become increasingly important.
A central question arises from these developments: what should guide the choice of variables to measure? The answer lies in the needs of the models under study. Measurement gaps should neither impose restrictive assumptions on models of behavior nor constrain policy design. For instance, policies addressing cognitive and socio-emotional development delays among disadvantaged children require understanding parental behavior. Recent research shows that parents’ beliefs about child development, rather than assumed knowledge, are crucial for effective policy design. Similar lessons apply in other policy domains.
Parallel innovations in data sources and methods further enrich this landscape:
- Administrative data: Increasingly available and reliable, administrative datasets provide strong measures of certain variables. They can also be matched with survey data to validate new tools and explore behavioral properties. For example, Caplin et al. (2022) combined Danish administrative records with subjective expectations data to show that uncertainty measures from subjective responses are much smaller than those inferred from models based solely on observed outcomes.
- Online surveys: Widespread internet access has improved the representativeness of online survey samples. The New York Fed’s Survey of Consumer Expectations exemplifies the growing importance of this method.
- Latent factor models: Widely used in psychometrics, these models are now increasingly applied in economics to uncover underlying behavioral constructs. Though discussed as early as Goldberg (1971), their use became prominent only recently (see Cunha et al. 2010). They hold promise for designing and validating new measurement tools.
- Natural Language Processing (NLP): Advances in machine learning now allow researchers to transform qualitative text or open-ended survey responses into quantitative constructs, broadening the scope of measurable variables.
Together, these developments reflect a broadening of economics’ empirical toolkit. The challenge ahead is to refine and validate these new tools, integrate them with traditional measures, and align measurement strategies with the needs of economic theory and policy design.
