Which level of research focuses on larger social entities such as the nation?

Analytical sociologists share James Coleman's outlook that “[t]he principal task of the social sciences lies in the explanation of social phenomena, not the behavior of single individuals” (1990, p.2). Looking at empirical research in sociology, however, it is clear that while “[s]ocial theory continues to be about the functioning of social systems of behavior, (…) empirical research is often concerned with explaining individual behavior” (Coleman, 1990, p.1). The analysis of individual behavior is necessary for a complete understanding of systemic outcomes (a position known as ‘methodological individualism’). As analytical sociologists stress, however, it is not generally sufficient for this understanding (e.g., Boudon, 1979). The reason for the lack of sufficiency lies in the potentially complex ways in which individual behaviors interact to produce macro-level outcomes. The explanation of a macro-level outcome in terms of its generative interdependent individual actions, is variously known as the ‘micro-macro link approach’ or the ‘social mechanisms approach’. In the words of Hedström (2005, p.24) “[t]he core idea behind the mechanism approach is that we explain not by evoking universal laws, or by identifying statistically relevant factors, but by specifying mechanisms that show how phenomena are brought about.” Hedström offers a definition of mechanisms as consisting of “(…) a constellation of entities and activities that are organized such that they regularly bring about a particular type of outcome” (2005, p.25).

The potential complexity of social mechanisms has long been acknowledged (e.g. Boudon, 1977; Lindenberg, 1977). Coleman (1990, p.22), for instance, stresses that the macro-level outcomes to be analyzed frequently involve true interdependencies of individual behavior, going beyond mere “aggregation”. Hedström (2005, p.26) similarly emphasizes that “(…) the same entities (individuals actors) strung together in different ways can be expected to regularly bring about different types of outcomes”, suggesting that the “stringing together” (i.e., the way in which individual actions interact) is an important factor determining macro outcomes. Obviously, not all social mechanisms are complex in this sense, and ‘easy cases’ can be found where simple aggregation explains macro outcomes. Generally, however, it is safe to say that analytical sociology has a lot to offer when explaining macro-level outcomes produced by interdependent individual actions. For the explanation of such phenomena, we “(…) have to look at the system of interaction between individuals and their environment, that is, between individuals and other individuals or between individuals and the collectivity” (Schelling, 2006, p.14). To understand this system of interaction, analytical sociological theories of social mechanisms are comprised of at least two parts: (i) a theory of individual behavior (the “micro foundations”, cf. Wittek et al., 2013, p.5), and (ii) a theory of how individual behaviors combine, under specific rules (i.e., norms, institutions), to produce macro-level outcomes (Coleman, 1990).

The focus on social mechanisms implies that in empirical research at least some of the hypotheses (and according to analytical sociologists the quintessentially “sociological” among them) take human collectivities such as groups, teams or societies as their units of analysis. This is true for most studies in the domain of ‘experimental behavioral game theory’ (e.g., Camerer, 2011), on which we focus in this paper and in which groups of participants make strategic choices. The interest of such games (e.g., public good games, coordination games, volunteer's dilemmas, etc.) lies not only in the behavior of individuals, but also in the group-level outcomes (e.g., whether the group's public good is successfully produced). Treating experimental groups as the unit of analysis for at least some hypotheses seems to imply that a great many participants must be recruited and tested, which is potentially very onerous, as statistical power of tests using macro-level outcomes is generally low with a small number of groups. Therefore, most studies from this domain focus on testing the micro-level hypotheses implied by their theories, which increases statistical power because of a larger number of observations.

Focusing on the explanation of individual behavior works at least to some extent in the simple cases in which the macro-level outcome is an aggregate (such as a count, a sum or a mean) of individual actions (or their consequences), for which a sampling distribution can be derived. For instance, in an experimental linear public goods game with several treatments (e.g. Fehr and Gachter, 2000), mean contributions to the public good by participants can be computed per treatment, and treatments can be statistically compared with regard to this mean. But even in this simple case the researcher has very little information about the sampling distributions of macro-level explananda (e.g., total contribution at the group level) for different treatments, due to the low number of groups.

From an analytical sociology perspective, testing only individual-level hypotheses fails altogether when the macro-level outcome depends on interdependent individual actions in more complex ways. In such cases, the sampling distribution of aggregated individual behavior (in the sense of counts, sums or means) does not equal the sampling distribution of the macro-level outcome. Moreover, the latter distribution (which is necessary for statistical inference concerning macro-level hypotheses) frequently cannot be analytically derived. For instance, even in a simple step-level public goods game (e.g., Kerr, 1992, Van de Kragt et al., 1983), where the public good is produced if and only if at least a certain minimum number of group members contribute, a higher number of contributors across all groups does not imply a higher number of ‘successful’ groups (the relevant macro-level outcome). The number of successful groups critically depends on the distribution of contributors across groups, and the aggregate sum of contributors contains limited information about this. Hence, one is again left with taking observed groups as the unit of analysis for inference about the macro-level outcome, resulting in low statistical power unless a considerable number of groups is available.

In this paper we offer an alternative to the standard solutions employed in experimental research of either exclusively focusing on individual behavior or settling for low power tests and propose and illustrate a methodology allowing statistical inference concerning the micro-macro link central to analytical sociology (Buskens et al., 2014; Hedström and Swedberg, 1998), using micro-level data only. Our method allows statistically more powerful inferences about social mechanisms by exploiting all the information about macro-level outcomes the data contain. Since we make all the data and R-scripts used for this research publicly available on the Open Science Framework (https://osf.io/scfx3/), our methodology can be employed by any researcher whose data have a certain structure and whose data and theory meet two assumptions.

The data structure to which our method applies is as follows. A set of N micro-level units (typically, (the behavior of) individual people) is distributed across T treatments, with each micro unit assigned to a single treatment (i.e., in a between-subjects design). In the remainder we will refer to such a distribution of micro units across treatments as a permutation. Within each treatment t the micro units are distributed across Kt groups, with each micro unit assigned to a unique group. In the remainder we will refer to such a set of groupings (one per treatment t) as a partition. The two core assumptions of our methodology are that (i) the micro-macro link or social mechanism is formally expressible as a function of the micro-level data (‘individual behavior’), and (ii) the macro-level units (‘groups’) in the data are randomly formed and do not lead to statistically dependent individual behavior (i.e., individual behavior is statistically independent of group membership).1

Data from behavioral game theory experiments often satisfy these two assumptions. First, such research typically involves explicit ‘rules’ mapping individual behaviors to group outcomes and these rules thus codify the experiment's social mechanism. Meeting this assumption forces the researcher to explicate the social mechanism at the core of her theory. Quite apart from statistical inference, this assumption is a conditio sine qua non for analytical sociology. Second, experimental macro units (such as groups or networks) are very often artificially created in the laboratory, and hence do not involve many of the statistical dependencies real-world groups have. This renders the second assumption more plausible. Note, however, that experiments involving communication between individuals before the behavior of interest almost inevitably violate the second assumption. This is not to say our method is principally unsuited to observational research, but the purest illustrations are likely found in experiments. Therefore, we draw our two examples from a published experimental paper in the field of experimental behavioral game theory. The paper used for illustration is the Dijkstra and Bakker (2017) publication on Step-level Public Goods (SPG).

In the next section we state the purpose of our method and give its outline in terms of six steps that have to be completed in order to apply it. The subsequent section illustrates with two examples. A conclusion and discussion section closes the paper.

Cited by (0)

Recommended articles (6)

• Research article

  Fluctuating courses and constant challenges: The two trajectories of black-white earnings inequality, 1968–2015

  Social Science Research, Volume 77, 2019, pp. 30-44

  Show abstractNavigate Down

  While rich literature exists on black-white earnings disparity, few studies differentiate the separate challenges facing different socioeconomic segments of the black population. Based on the Panel Study of Income Dynamics (PSID), this study explores the evolvement of black-white inequality in early-career earnings separately among college-educated and non-college-educated Americans from 1968 to 2015. The results indicate that while college-educated African Americans underwent fluctuating progress toward racial earnings equality, contingent on larger political and economic climates, African Americans of limited educational attainment experienced persistently wide earnings gaps with their white counterparts. By confirming the significance of social class on African-American economic outcomes, this study provides evidence in support of class-specific policy interventions toward racial equality. It also opens the door for subsequent research to explore specific mechanisms through which class-based racial disadvantages are operationalized.

• Research article

  Response styles confound the age gradient of four health and well-being outcomes

  Social Science Research, Volume 78, 2019, pp. 215-225

  Show abstractNavigate Down

  Accurate representation of the association of health and well-being outcomes over age can inform us about how well the population is doing, where segments of the population may be in need, and allow hypothesis generation about correlates and causes of observed gradients. In this paper, we examine the possibility that response styles can impact associations between respondent age and four common, self-report variables: overall health; life satisfaction; pain intensity; and, fatigue level. Response styles (RSs) are defined as tendencies for people to use certain locations on response scales (e.g., extremes, middle) regardless of scale content. Although RSs have been shown in some circumstances to bias self-reports, often little attention is paid to this source of bias.

  A sample of 2000 Internet panelists completed an online questionnaire that included a set of heterogeneous items to compute measures of the following RSs: acquiescence, disacquiescence, midpoint responding, and extreme responding.

  A total of 1839 participants formed the analytic sample and they had an average age of 54.4 (range: 21–89). RSs were associated with age and with the outcome variables. Moreover, the age pattern of the health and well-being variables was considerably altered when RSs were controlled. An alternative, item-response theory-based method of deriving measures of RSs confirmed many of the associations from the primary method of computing RS variables. Across the different self-report outcomes, disacquiescence had the greatest effect on the age patterns, and pain and fatigue were the outcomes most impacted by controlling for RSs.

  In accord with prior research, we recommend greater consideration be given to RSs in aging research.

• Research article

  Till multiculturalism do us part: Multicultural policies and the national identification of immigrants in European countries

  Social Science Research, Volume 77, 2019, pp. 88-100

  Show abstractNavigate Down

  How do multicultural policies affect immigrants' identification with the country of destination? Theory suggests that these policies may have two opposite effects and either widen or diminish the gap between the national identification of natives and immigrants. In addition to these opposite effects, I expect that the effects of multicultural policies are also diverse depending on immigrants' cultural and social distance from the host society. In this study, immigrants are categorised based on generations and origins. Using newly constructed measurements for multicultural policies, as well as European Social Survey Round 7 with 20 European countries, I conduct a multilevel analysis. The results indicate that multicultural policies diminish the gap between the national identification of natives and immigrants. However, these effects are evident only for non-European immigrants and not for European immigrants. Furthermore, I find no evidence that the effects differ for the first and second generations.

• Research article

  Entropy Generation and Natural Convection of Air under a Magnetic Quadrupole Field in a Square Enclosure

  Energy Procedia, Volume 105, 2017, pp. 5073-5078

  Show abstractNavigate Down

  The natural convection and entropy generation of air are numerically investigated in a two-dimensional square enclosure under a magnetic quadrupole field. In this study. The finite-volume method is applied to solve the coupled equation for flow, energy and entropy generation. Simulations are conducted to obtain solutions in terms of streamlines, isotherms, local entropy generation contours, total entropy generation, Bejan number for various ranges of the magnetic force parameter and Rayleigh number in a nongravitational field. Two symmetric cells with the same strength but opposite rotation directions are observed with the flow filed. Total entropy generation increases with increasing the magnetic force number. However, the average Bejan number holds the reverse trend. The magnetic field has a broad spectrum of applications in engineering areas to control natural convection. The local and total entropy generation presented in this study can be useful addition where one wants to minimize the irreversibility of the system.

• Research article

  How great and how good?: Third places, neighbor interaction, and cohesion in the neighborhood context

  Social Science Research, Volume 77, 2019, pp. 68-78

  Show abstractNavigate Down

  Though Ray Oldenburg's (1989) notion of “third places”, or places conducive to sociality outside of the realms of home and work, has received both scholarly and popular attention over the past several decades, many of the author's central claims remain empirically untested. The present study considers the association between neighborhood third places, cohesion and neighbor interaction. Drawing on various literatures regarding interaction in public space and neighborhood use-value, we consider how the role of third places might vary according to neighborhood socioeconomic context. Using data from Wave I of the Los Angeles Family and Neighborhood Study (LAFANS) and data on third places from the point-based business data of ReferenceUSA, we test the effect of third places on cohesion and neighbor interaction across neighborhood poverty strata. We find support for the hypothesis that third places are associated with greater cohesion and neighbor interaction, and that neighbor interaction mediates the relationship between third places and cohesion in poor neighborhoods.

• Research article

  Proximity and the principle-policy gap in white racial attitudes: Insight from views of student assignment policies in Wake County, North Carolina

  Social Science Research, Volume 78, 2019, pp. 95-103

  Show abstractNavigate Down

  We use a survey of residents of Wake County, North Carolina to test a proximity explanation for what scholars call the “principle-policy gap” in whites' views of government action on race. The derived hypothesis is confirmed when underlying broad views of race are represented by ideology. We show that whereas liberals are materially more supportive of racial diversity in student bodies than are moderates and conservatives, this difference is reduced to statistical insignificance as respondents’ personal situations are more directly affected by the policy—a condition denoted by having a child of school age. Liberals with school-aged children are particularly wary of racial integration when they are asked to support the diversification of classrooms as opposed to schools. Having a school-aged child has no material conditioning effect on views of diversity when attitudes are represented by a more explicit measure of racial views.

  Which level of research focuses on larger social entities such as a nation?

  At the macro level, sociologists examine social structures and institutions. Research at the macro level examines large-scale patterns.

  What is a level of analysis that involves the study of large

  Macrosociology is a large-scale approach to sociology, emphasizing the analysis of social systems and populations at the structural level, often at a necessarily high level of theoretical abstraction.

  Which specific form level of social analysis would study large groups and societies?

  Macro-level sociology looks at large-scale social processes, such as social stability and change. Micro-level sociology looks at small-scale interactions between individuals, such as conversation or group dynamics.

  What does meso level mean?

  Meso level In general, a meso-level analysis indicates a population size that falls between the micro and macro levels, such as a community or an organization. However, meso level may also refer to analyses that are specifically designed to reveal connections between micro and macro levels.