Association for Public Policy Analysis & Management

Over the past few decades, a scholarly consensus has coalesced around the idea that high-quality child care is important to children’s short-run development and long-run schooling and labor market success. Yet many U.S. children—particularly those in low-income families—attend child care programs that are of low- to mediocre-quality. To explain this phenomenon, some analysts point to the presence of information asymmetries, in which providers know more about their quality attributes than do consumers. In this context, market competition will not improve program quality because consumers cannot accurately assess their options. Although this line of reasoning is compelling, it has proven difficult to empirically verify whether parents’ child care preferences are congruent with the definition of “quality” as promulgated by early education scholars and professionals. More fundamentally, there are gaps in our understanding of what parents look for, or value, in a child care provider.

This project exploits novel data and empirical techniques to shed new light on these issues. Specifically, we construct a dataset of consumer reviews of child care programs in 40 large, U.S. cities from the website Yelp.com. Founded in 2004, Yelp is now the predominant website for hosting consumer reviews of virtually all business-types, including those providing child care services. Several attributes make this website a powerful laboratory for studying parental assessments of child care. First, Yelp contains a large number of reviewed child care establishments; indeed, our dataset includes nearly 50,000 unique reviews of about 9,800 businesses. Second, Yelp users can provide detailed open-ended, text-based reviews of child care programs. Such unstructuredness is advantageous because it provides consumers with the autonomy to evaluate any dimension of the program deemed important, and to do so using a variety of emotional tones.

Empirically, we exploit both deductive and inductive approaches to classifying parents’ child care preferences. Regarding the former, we assess to what extent parents’ child care reviews overlap with the quality domains measured in the Early Childhood Environment Rating Scale-Revised (ECERS-R), which is the most widely-used observational measure of child care classroom quality (Harms et al., 1998). Regarding the latter, we utilize Latent Dirichlet Allocation (LDA) methods, which analyze the co-occurrence of words and phrases to produce a set of common topics or themes in parents’ child care reviews. In addition, we subject the reviews to a Linguistic and Word Count (LIWC) analysis, which codes various pieces of text according to its cognitive (e.g., analytical) and emotional (e.g., positive, negative, anger, etc.) components.

Armed with these data, we empirically examine (i) the attributes of child care that parents comment on and (ii) the sophistication and emotional tone deployed in the comments. We examine variations in parents’ child care reviews in two ways: over the distribution of various local demographic characteristics (e.g., median household income and fraction non-white) and to parent reviews of K-12 schools. Finally, we examine to what extent parents’ thematic comments (i.e., whether they comment on prices or quality features) are related to their assigned star-rating.