| Popis: |
Latent constructs are fundamental for theory-building and practice in various disciplines of the social sciences. Passive datasets and machine learning approaches provide an opportunity to predict latent constructs, thereby offering a strong potential to build scalable, granular, and theory-grounded models for various phenomena of interest in social sciences. However, there is little guidance for researchers to develop prediction models for latent constructs, especially given that latent constructs are different from observable variables in several keys ways. To address this gap, and drawing from both theoretical and applied literature across the social sciences, we first develop a taxonomy to classify latent construct models based on their ontology and measurement. Second, we synthesize the relevant literature on using passive data to predict latent constructs. Finally, by applying the developed taxonomy to various applications in the literature, we offer guidance for predicting latent constructs from passive data. By bringing together loosely tied research in the context of scaling up latent construct models, this paper makes two significant contributions: 1) identify key significant and specific gaps in this research area, 2) provide theoretically- and empirically-grounded guidance for social scientists interested in using passive data for developing scalable latent variable models. |
