«A Household-Centered Approach to Saving: Financial Types and Their Relationship to Saving Behavior Lisa A. Keister Richard Benton James Moody Duke ...»
A Household-Centered Approach to Saving:
Financial Types and Their Relationship to Saving Behavior
Lisa A. Keister
Department of Sociology
*Direct correspondence to Lisa A. Keister at 268 Sociology-Psychology Building, Box 90088,
Durham, NC 27708. Lkeister@soc.duke.edu. Keister acknowledges a grant from the National
Science Foundation (SES-1322738), and Moody acknowledges a grant the National Institutes of Health (HD075712-01) that supported this research. We are grateful for comments from Paige Borelli, David Diehl, Achim Edelmann, Hang Young Lee, and Laura Tesch.
A Household-Centered Approach to Saving:
Financial Types and Their Relationship to Saving Behavior Abstract Household saving is an important indicator of both short-term and long-term financial wellbeing, but Americans do not save much of their incomes. Most explanations for saving isolate and study the role of household-level variables (e.g., education, family structure) and, to a lesser extent, aggregate variables (e.g., region, structural position, business cycles). We move beyond this standard approach and, instead, identify household financial types based on the allocation of household resources across spending categories. Using the Consumer Expenditure Survey, we explore differences across financial types in short- and long-term saving behavior. We find that households cluster into distinct financial types that are meaningfully related in social space and that these household types save in distinct ways that have important implications for understanding inequality and stratification.
Household saving and the resulting assets that accumulate from saving are important components of both short- and long-term financial well-being, yet American households save very little of their incomes. Accumulated savings can be used to pay for current needs and desires or retained to provide a buffer against unanticipated income interruptions, medical emergencies, accidents, natural disasters, and other crises. They can generate interest and dividend income that can be consumed or reinvested—as in the purchase of a home, other real estate, or financial assets—to create additional resources (Kennickell and Starr-McCluer 1997;
Wolff 2010). Accumulated savings can also improve educational attainment, occupational opportunities, political power, and social influence (Domhoff 2013; Freeland 2012). Perhaps most significantly, accumulated savings can be passed to futuregenerations either as inter vivos transfers or inheritance to extend these benefits indefinitely (Avery and Rendall 2002; Gokhale and Villarreal 2006; Laitner 2001). Most Americans recognize the importance of saving, and awareness of asset inequality has also grown (Taylor et al. 2011); but saving rates in the United States remain surprisingly low. Between 2007 and 2012, Americans saved only 4% of their disposable incomes annually, with rates dropping to as low as 2% during the 2007-09 recession (OECD Library 2013). Household saving rates in the U.S. are particularly low compared with rates in other developed countries, such as Switzerland and Germany, where households save as much as 14% of their disposable incomes (OECD Library 2013).
Despite the importance of saving, we know relatively little about the factors that explain it. A growing but still somewhat new literature on inequalities in the related concept of wealth or net worth—total assets less total debts—suggests that some households save considerably more than others at all income and wealth levels. Consistent with this, the association between income and savings is moderate, with income accounting for only 25% to 36% of the variance in saved assets (Wolff 2010). To explain why people save, prior research has typically isolated one or two traits and studied how these variables correlate with saving. For instance, there is evidence that saving is strongly associated with race and ethnicity (Avery and Rendall 2002; Campbell and Kaufman 2006; Oliver and Shapiro 2006; Shapiro 2004), education (Spilerman 2000), marriage and divorce (Zagorsky 2005), gender (Chang 2010), and religion (Keister and Sherkat 2013).
Researchers have also paid some attention to the role of aggregate factors, such as local barriers to accessing financial organizations; regional variation in market conditions; and to a lesser extent, national-level business cycles and cohort processes in predicting saving (Kopczuk and Saez 2004; Shapiro 2004; Spilerman 2000; Wolff, Owens, and Burak 2011). This research provides powerful evidence of numerous bivariate relationships, but the focus on particular variables as independent, disconnected influences strips the meaning of each variable from the context of other behaviors within which it is embedded. Instead of asking how households save, the focus is on average differences (holding all else constant) across households that vary on a single trait.
In this paper, we take a different approach: we examine how households—as whole packages rather than simple markers of particular variables—cluster together based on spending allocations that collectively shape saving. In so doing, we build on a growing body of research in the social sciences that recognizes social variation as case-centered, rather than variable-centered, and that takes aspects of structural position, avocation, and preference as complete packages (Abbott 1998; Garip 2012). Once we identify household clusters, we ask how cluster membership relates to saving, directly linking position in social space to an important financial outcome.
Because we identify the clusters empirically through observation of emergent variation in survey data rather than imposing group boundaries a priori, we are able to explore how contextuallydependent groups of households approach saving and to potentially generate new hypotheses for future work that extend beyond the productive but limited focus on single variables. The more holistic clustering approach also allows us to identify groups of households that are seemingly unique and might not otherwise be considered together but that are comparably situated in social space and thus have common behavior.
In what follows, we first elaborate on prior research on saving and the utility of a casecentered approach to social behavior. We then use data from the Consumer Expenditure Survey (CES) to identify financial types based on information about household spending, an important indicator that simultaneously reflects household-level and aggregate-level traits, identities, and behaviors. Spending is any purchase of goods or services that are essential (e.g., rent, mortgage, groceries), discretionary (e.g., entertainment, consumer electronics, vacations), or a combination of each (e.g., clothing, dining out). In American consumer culture, spending (e.g., to support a political or religious group, wear certain clothing, get a tattoo) is an excellent summary indicator of preferences, habits, and identity. Spending also reflects important elements of structural position, class status, and geographic location because race and ethnicity, education, income, and related traits (e.g., area of residence) can affect both the availability and cost of goods and services and the way households allocate resources across possible expenditures. For instance, living in a food or retail desert inflates the cost of groceries and other items, forcing households to allocate a disproportionate share of total income to basic needs than if affordable goods and services were available nearby (Moore and Roux 2006; Myers et al. 2011). After we identify financial types, we use the CES to explore differences across types of households in saving.
Explaining Saving Behavior A growing body of research acknowledges that saving and accumulated assets are fundamental to household well-being and thus to research on inequality, and work in this area has identified many household traits that correlate with saving. Race and ethnicity have attracted considerable attention, and it is now well-known that African Americans and Latinos have lower levels of household saving than whites (Campbell and Kaufman 2006; Oliver and Shapiro 2006;
Taylor et al. 2011). There is also evidence that some ethnic groups accumulate assets more rapidly than others following immigration, a pattern that appears to reflect elements of selection into immigration (e.g., by education) and the conditions that prevail when the immigrants arrive in the U.S. (Hao 2007). Educational attainment and self-employment are strongly correlated with saving and accumulating assets, although it is still unclear whether these achieved traits can overcome inheritance in producing adult outcomes (Hansen 2013). Although gender effects are difficult to identify for married households, unmarried women approach saving differently than men and ultimately tend to save less, with the important exception that younger women are now outpacing their male counterparts (Chang 2010; Raffalovich, Monnat, and Tsao 2009;
Yamokoski and Keister 2006). Similarly, marriage has advantages for saving that divorce erases (Zagorsky 2005); and religious beliefs can affect saving by, for example, diverting funds to tithing rather than allowing it to grow in personal bank accounts or other investments (Keister 2011). Aggregate processes attract less research attention, but there is evidence that structural barriers to accessing banks and other organizations in the formal economy, recessions, and other aggregate processes also affect saving (O'Brien 2012; Shapiro 2004; Wolff, Owens, and Burak 2011).
From Variables to Households Despite the many strengths of research on saving, prior work clearly focused on single traits (other factors controlled) that correlate with saving. To move beyond studying isolated marginal effects, we propose that the factors that influence saving interact in nuanced and complex ways to create types of households with distinct and identifiable traits and lifestyles. By moving from a variable-centered analysis to a case-centered analysis, we aim to identify household positions in social space (Bourdieu 1990) and to capture some of the unique practices related to each social position that promote saving. Our approach turns on the idea that information from a behavior or set of related behaviors (e.g., spending) can be used to identify groups of households that are similarly-situated in social space and who face similar constraints and opportunities. This approach has a long history in other fields, such as marketing, where identifying well-defined social types is important (Weiss 1988, 2000), and its use in the social sciences is becoming increasingly common and generative (Brint, Riddle, and Hanneman 2006;
Garip and Asad 2014; Martin 2011).
The goal of the case-centered approach is to reorient analysis toward the complete package of behaviors associated with a particular position in social space and, then, to use information about social position to provide a more holistic explanation of social behavior.
Variants on this approach include the nearly complete endogenous reproduction of social position such as Bourdieu’s notion of habitus1 and recent extensions of field and network theory (Martin 2003; Padgett and Powell 2012) as well as pragmatic attempts to identify coherent aggregates that react similarly to (typically unobservable) forces related to occupancy of a socialAs an acquired system of generative schemas, the habitus makes possible the free production of all the thoughts, perceptions, and actions inherent in the particular conditions of its production – and only those” (Bourdieu 1990).
ecological niche (Amato, Kane, and James 2011; Brint, Riddle, and Hanneman 2006; Garip and Asad 2014; Kalleberg and Moody 1994). Our use of cluster analysis is more akin to the second, more pragmatic, strategy but shares traits with approaches that reproduce social position as well.
The conceptual insight underlying these techniques is familiar to most social scientists as interaction effects, in which the meaning of one variable is conditional on the meaning of another. For example, we know that African Americans living in majority white neighborhoods accumulate savings differently than those living in majority black neighborhoods because housing values vary and appreciate differently across these neighborhoods (Shapiro 2004). The logical, but methodologically cumbersome, extension of the interaction-effect strategy is to allow the effects of all variables to vary simultaneously by every other variable; this would be equivalent to asking how occupants of any point in an n-dimensional space defined by the variables behave. This approach to empirical understanding is the foundation of good ethnography: rather than identifying how outcomes vary by a single dimension, the ethnographer observes deeply-contextualized behavior and identifies unique insights into how situations interact with numerous traits to shape behavior. Although cluster-analytic approaches cannot reach the level of rich detail attained by good ethnography, the goal is to take seriously the idea that contextualized effects are better representations of life situations than single variables while retaining the powerful generalizability of large-scale surveys (Abbott 1998).
Our approach identifies the position that a case (often individuals, but our focus is households) occupies in a wider social field (Martin 2011). The set of behaviors centered on a particular topic defines a multidimensional space in which cases occupy positions. We expect this space to be clumpy—to admit to clusters of cases that form unobserved groups and that act similarly because they are faced with similar resources and constraints and have internalized similar ways of acting. The logic here is akin to that of Bourdieu, where habitus describes how a household’s structural position affects its members’ behaviors, lifestyles, and habits through internalized preferences, tastes, and cultural understandings (Bourdieu 1990). Since this activity includes social interaction and, in the case of saving and spending, interdependent market forces, history, and social organization conspire to make the distribution of the behavior space irregular or clumpy, allowing observers to identify similarities among cases by their proximate positions.