national surveys. Publicly available data from larger surveys provides an essential resource for medical sociology.
Experimental and quasi-experimental methods of data collection are less common in sociology but still provide useful information for medical sociologists. Experimental methods have been used to good effect in, for instance, vignette studies in which researchers randomly manipulate characteristics of patients and study the conditions under which physicians treat them differently (Stepanikova 2012). Quasi-experimental methods exploit real-world interventions that “randomly” affect some groups of people more so than others. For instance, studies of the effect of education on health have exploited the uneven roll-out of changes in compulsory schooling laws and studies of the effect of trauma on mental health have exploited random variation in the communities exposed to a natural disaster, such as a tsunami (Courtin et al. 2019; Frankenberg et al. 2012).
Ethnography, participant/observation, and in-depth interviews are common techniques for creating the rich data necessary to understand social phenomena. Researchers who enter the field to collect qualitative data gain access to a more expansive social world than researchers who rely solely on questionnaires to gather information. Qualitative data collection allows for the observation of the serendipitous social action that so often goes unobserved in other data collection strategies and is especially good at integrating the context of social life. Qualitative researchers are themselves the data collection instruments, so these approaches require that researchers constantly reflect on how their own social positions affects what they observe (Fine and Hancock 2017).
“Found” data is another source of data that’s having a growing influence on social science research. A creative researcher can develop unique datasets with information from administrative records, historical documents, social networks, and other publicly available data that is increasingly located online. For example, Cotti et al. (2015) merged the BRFSS with publicly available data on the Dow Jones Industrial Average, a stock index, and found that large drops in the stock market were associated with poorer mental health and higher levels of smoking, binge drinking, and fatal car accidents involving alcohol. Government websites in particular can have treasure troves of data for researchers willing to wade into the deep waters of internet data collection and organization. While offering an interesting new approach, inefficiency lurks. Data collection can go on ad infinitum if the researcher hasn’t adequately conceptualized the key components of their research question.
METHODS FOR ANALYZING DATA
Assuming a researcher has data in hand, they have a wide variety of research methods at their disposal to explore data in an effort to answer their research questions. In the sections that follow, we review the principal methods being used to understand data in ways that have helped advance medical sociology.
Quantitative Methods
If a researcher has numerical data, then they’ll use quantitative methods for the analysis. A numerical array of data typically consists of columns representing study variables and rows containing observations of variables for each case/participant. For example, if we collected information about years of schooling from a sample of 100 adults, then each of the 100 rows would contain the education of a single participant. Variables are chosen through a process called operationalization (or simply measurement) in which researchers choose how to empirically represent concepts from their research questions. Because of the diversity of perspectives in medical sociology, most peer-reviewed journals require researchers to provide rigorous justifications for their measurement choices. This makes careful operationalization a critical feature of quantitative research in medical sociology (Aneshensel 2002; Link 2002; McAlpine et al. 2018).
Measurement
Operationalization involves two independent stages, identifying (or designing) a valid measure of a concept and then determining how exactly to incorporate that measure into a statistical model. The first stage involves a good deal of perseverance and creativity from researchers. Operationalizing complex sociological concepts generally requires a detailed review of prior studies in light of one’s specific research question. The second stage is more circumscribed and can often depend on the availability of measures and the degree to which measurement models will help inform one’s research question.
Researchers often take one of the following approaches to incorporating measures into statistical models: (1) single variable, (2) multiple variables, or (3) latent variable (Bollen et al. 2001). In some cases, the second step in operationalization is relatively straightforward. For example, if we are interested in estimating the extent to which age is related to alcohol use in a population, then we would look for data that includes respondents’ age and some assessment of alcohol consumption. This would be a single variable approach because we only used one variable to represent each concept. We should note, however, that even in this simple example prior research offers multiple potential measures of age (e.g. in years, meaningful age ranges) and alcohol use (e.g. drinks on average, binge drinking, any drinking, etc.). In other cases, operationalization is less clear. Socioeconomic status (SES) is a recurring term in health disparities research, but despite its relative cohesiveness as a theoretical concept, researchers uses a wide variety of variables to operationalize SES. Generally, medical sociologists use a multiple variable approach to measuring SES by incorporating two or more variables related to educational, economic, and occupational attainments (Wolfe 2015). Although the single variable approach would be easier to interpret, the complexity of using multiple variables to represent SES is usually offset by the greater amount of information gained from results.
Working with latent variables requires an entirely different approach to measurement (Bollen 1989). For single and multi-variable operationalizations, researchers assume that sociological concepts can be directly observed. When we investigate variables such as years of age or schooling, we assume that people accurately know and report that information (with some random error being acceptable). Suppose, however, we’re interested in the amount of depression in the general population. Do people always know if they’re depressed? Does everyone have the same definition of depression? This is a trickier situation. Fortunately, the Center for Epidemiologic Studies Depression (CESD) scale, which originally included 20 items (Radloff 1977), is a well-established measure of depression. There are a number of statistical approaches to combining indicators of depression like the CES-D items into a single scale (e.g. Payton 2009; Perreira et al. 2005), but they all assume that we can create a latent variable for depression that avoids the measurement error we would encounter with a single question asking directly about depression.
Measurement concerns might seem pedantic at times but taking measurement issues seriously can generate important research. For example, Montez et al. (2012) evaluated models of US mortality with 13 different measures of education derived from Hummer and Lariscy (2011). The preferred functional form of education included a linear decline in mortality risk from 0 to 11 years, a notably larger reduction in mortality risk after high school completion, and a steep linear decline continuing after high school completion. Although their primary aims were methodological in nature, searching for the optimal form of education revealed an interesting theoretical insight – educational attainment benefits survival through both human capital accumulation and socially meaningful credentials (Collins 1979).
Research on health lifestyles offers another example of the importance of measurement (Cockerham 2005; Cockerham et al. 2020). Health lifestyles refer to meaningful combinations of health behaviors that people adopt. We can imagine a lifestyle involving regular exercise, a nutritious diet, and abstention from smoking and heavy drinking. Alternatively, we can also imagine a largely sedentary lifestyle with limited concerns about a nutritious diet. And one could continue with several other possible clusters of health behaviors that coalesce into recognizable health lifestyles. To investigate these potential lifestyles and their relationship to adult health, Cockerham et al. (2020) identified latent classes of different health lifestyles, and their results revealed
