Abstract
This article develops a cultural and contextual approach to studying agency that attends to variation in how people narrate their experiences. Drawing on the large-scale, nationally representative American Voices Project data, the article uses computational methods to test patterns in agentic expression and qualitative methods to examine how respondents narrate agency and passivity as they describe their lives. This analysis captures agentic moments, widespread narratives through which people emphasize their agentic selfhood as they recount specific situations. Moreover, individuals use narrative moves—such as shifting their focus and drawing on subtypes of agency—to craft agentic moments despite constraints. We argue that narratives of agency are variable, situational, and often co-occurring with narrative passivity, which enables people to narrate themselves as agentic even in challenging situations.
The concept of agency broadly encapsulates the self in action as people construct goals, plan for the future, make decisions, and develop pathways of action to achieve their objectives (Emirbayer and Mische 1998; Markus and Kitayama 2003, 4). Across disciplinary fields, scholars have investigated the many ways that an individual’s sense of agency is linked to their well-being and ability to overcome challenges. An increased sense of agency is associated with better educational and health outcomes (Andersson and Hitlin 2022; Bandura 1997; Werner and Smith 2001; Duckworth et al. 2007) and can affect subjective well-being in ways comparable to income (Hojman and Miranda 2018). Contrastingly, a sense of lack of agency, or passivity, is associated with uncertainty, anxiety, and emotional distress (Beck 1992; Robinson et al. 1991).
Some scholars have studied agency as a quality that individuals have based on their ability to manipulate their environments, entertain a wide array of choices, and pursue their goals. Research stemming from rational choice and psychological models of human behavior define agency as a personal quality that people do or do not have based on their ability to make choices and further their interests (Gray, Gray, and Wegner 2007; Hedstrom and Swedberg 1996; Jenks 1998). In this literature, agency is assumed to directly contrast with passivity (Jenks 1998). Scholars often use this definition to examine the relationship between agency and stratification, concluding that those in more privileged social positions have access to a wider range of choices, are more easily able to change their circumstances, exert more power over social structures, and thus have more agency. They assert that those in lower social positions face more constraints, are unable to manipulate their environments, are powerless in the face of hardships, and thus have less agency (Boroditsky and Ramscar 2002; Mirowsky and Ross 2007; B. Silver et al. 2021).
In contrast, other scholars have theorized that agency exists on a continuum or have shown that people’s subjective understandings and expressions of agency may not directly correspond to the choices available to them or the constraints they face. Some scholars have argued that agency can vary throughout the lifecourse and shift in degree based on contextual features (Abebe 2019; Fuchs 2001). Others have conceptualized agency as part of a fluid process of sensemaking connected to cultural structures (Frye 2012; Silva 2012; Snibbe and Markus 2005). This body of research emphasizes people’s subjective sense of agency, which depends on cultural frameworks that influence how people understand their options, choices, and pathways to achieve their goals (Hitlin and Elder 2007; Markus and Kitayama 2003; Silva and Corse 2018). Thus less privileged individuals can creatively respond to the barriers they face and feel a sense of control over their lives despite limited resources (Edin and Schaefer 2015). This sense of agency is critical for individuals to achieve dominant models of worth and value (Lamont 2019; Ho 2023). However, scholarship that studies, measures, and theorizes agency has yet to fully investigate the strategies that people use to position themselves as agentic, even as they narrate challenging contexts, as well as how subjective understandings of agency and passivity interact as individuals describe their experiences.
In this article, we ask how people express their sense of agency as they recount their experiences. Moreover, how do narrations of agency and passivity interact in different moments of people’s lives? We draw on the sociological literature on narratives and use computational and qualitative methods to explore narrations of agency within the American Voices Project, a nationally representative large-scale interview dataset.1 This dataset is uniquely suited for this analysis because of the wide-ranging scope of the interviews, which purposely cover many different aspects of respondents’ lives. We first draw on computational methods to test large-scale patterns in agentic expression drawing on scholarship on agency and stratification. We explore agentic expression between individuals and investigate how agentic expressions can vary within the same person. Subsequently, to understand variations in narrations of agency, the interaction between narrative agency and passivity, and how people can narrate agency in constraining contexts, we perform a qualitative analysis on a subsample of data.
Through this multimethod approach, we develop the concept of agentic moments, or discrete moments in which individuals across social groups narrate agency in different contexts and despite facing constraints that may be thought of as limiting choice and agency. Agentic moments encompass specific situations, defined as actors and their settings, in which people narrate their agentic capacity when facing concrete challenges (Berger and Luckmann 1966; Garfinkel 1967; Tavory 2018). Our concept of agentic moments foregrounds the intersections and co-occurring nature of agency and passivity and recognizes individuals’ abilities to narrate agency in challenging situations. Furthermore, we identify common narrative moves: planned or unplanned narrative strategies that shift the focus of narratives or draw on different ways of claiming agency to support agentic moments. Thus we argue that people narrate themselves as agentic in ways that are variable, situational, and often co-occur with narrative passivity.
By developing the concept of agentic moments and identifying the narrative moves that enable these moments, we contribute a cultural and contextual model that captures the variability of agency, its co-occurrence with passivity, and how people use narratives to present themselves agentically. Furthermore, we use a mixed-methods analysis of an unprecedented nationally representative dataset of wide-ranging interviews to demonstrate the applicability of our concepts across diverse people, contexts, and themes. Thus, we extend work that has shown that people tend to present themselves as agentic (Silva and Corse 2018; Ho 2023), even in particularly challenging or constraining circumstances (Enck and McDaniel 2015; Charmaz 2002), as well as work that has shown that a person’s narrative of self can become more (or less) agentic over time (Polkinghorne 2007).
THEORETICAL DEVELOPMENTS IN THE STUDY OF AGENCY
We first explain research stemming from rational choice and psychological models of human behavior that define agency and passivity as contrasting qualities that individuals have based on their ability to exercise personal choice and control. We then explain research that challenges this conception through a cultural and contextual account of agency.
Agency and Passivity
In a tradition extending back to Aristotle, scholars have differentiated between people who act and people who are acted upon (Reader 2007). Traditionally in the social sciences, a rational choice and psychological model of agency dominated. In this model, agency is seen as a discrete quality possessed and enacted by people who act purposefully, intentionally, and methodically to pursue their personal interests (Fuchs 2001; Hedstrom and Swedberg 1996; Kahneman 2003; Parsons 1951). Related concepts across disciplines include self-efficacy (Bandura 1997), locus of control (Wallston and Wallston 1982), autonomy (Sayer 2011), and capabilities (Sen 1999). In contrast, passivity has been conceptualized as the opposite of agency to describe a lack of control or constraints imposed by external forces that limit choice and command over one’s environment (Jenks 1998). In this literature, passive individuals are those who receive, feel, desire, or experience in ad hoc, uncontrolled, or overly structured ways (Gray, Gray, and Wegner 2007; Gray and Wegner 2009; Jenks 1998; Parsons 1951). Passivity has been associated with the concepts of patiency (Ash, Stammbach, and Tobia 2021; Burkitt 2016), learned helplessness (Peterson, Maier, and Seligman 1995), perpetual waiting or resignation (Duncan 2015), and unreflective feeling or experience (Schweitzer and Waytz 2021). The literature depicts passive individuals as unable to work toward or achieve goals, overcome challenges, or express control over their life circumstances. Passivity has been less developed as a concept and is most frequently conceptualized as the lack of agentic qualities. Scholarship theorizes the relationship between agency and passivity (Burkitt 2016; Reader 2007), but has yet to empirically study interactions between the two concepts.
The conceptualization of agency as a quality held and enacted, contrasted with passivity, and connected to individual action, choice, and interests has become a central component of research on agency and stratification. In this research, people with access to more resources or privilege have more choice and control in acting on their preferences, planning and achieving their futures, and affecting their environments (Kohn 1989; Kraus, Piff, and Keltner 2009; Pattillo 2013). In contrast, those with less privilege or access to resources face more constraints, fewer choices, and less control over their environments (Edin and Schaefer 2015; Desmond 2012; Kraus, Piff, and Keltner 2009; Lachman and Weaver 1998; Silva 2012). From this theoretical vantage point, disadvantage and hardship impose increased constraints that limit people’s choices and agency. For example, research on social class suggests that differences in agentic dispositions begin to develop in childhood: middle-class children learn to actively express and reach their desires, goals, and needs, whereas lower-class children learn a sense of constraint (Lareau 2002). As this example shows, researchers studying agency suggest that privileged groups have more agency, whereas less privileged groups face more constraints that limit their agency.
Agency as Continuous, Variable, and Cultural
This rational choice and psychological conceptualization of agency and passivity as qualities individuals possess has been challenged in a number of ways. First, the theoretical literature has conceptualized agency as highly variable and existing on a continuum with passivity. Second, cultural scholars have emphasized the importance of understanding people’s subjective understandings and narrations of agency. We build from these bodies of literature to develop a narrative and situational conceptualization of agency.
First, some scholarship has suggested that agency and passivity may be fluid—both existing on a continuum and highly variable between contexts and people. Theoretical scholarship argues that agency and passivity are relative and interdependent; consequently, they often occur to varying degrees (Abebe 2019; Burkitt 2016; Fuchs 2001) and in relation to one another (Gomart and Hennion 1999; Reader 2007). As a result, people’s agency varies based on contextual features, shifting over time and between contexts. Moreover, researchers have shown that agency is variable even between individuals with the same demographic characteristics. For example, Anthony Jack (2016) shows that lower-income undergraduates at an elite university, because of different high school environments, were more or less proactive in engaging with professors and seeking help at their university. Although this body of research theorizes agency and passivity as existing on a continuum—rather than as a binary—and recognizes the variability of agency, it does not examine how agency and passivity in fact co-occur and interact in specific contexts.
Second, beyond considering agency as a quality or capacity that people have, some scholars have demonstrated the significance of people’s subjective understandings and expressions of agency. In this research, people construct and express agency through their interpretations of who they are and what they can do, which ascribes meaning to their realities (Bruner 1991; Ewick and Silbey 2003; Fuchs 2001; Correll 2004; Silva and Corse 2018). People can interpret their abilities to manipulate their circumstances, make decisions about their lives, and influence social structures (Sewell 1992); they can interpret themselves as having choice or being guided by fate or luck (Abend 2018; Sauder 2020; Stephens, Fryberg, and Markus 2011). People generally draw on their resources, skills, and knowledge to narrate an interpretive understanding of themselves as agentic and able to affect the course of their lives (Ho 2023; Silva 2012). Subjective understandings and expressions of agency have been studied through different cultural structures, including frames (Fuchs 2001), accounts (Scott and Lyman 1968; Damaske 2013), imaginings (Frye 2012; Mische 2009; Zilberstein, Lamont, and Sanchez 2023), and feelings or judgments (C. Silver et al. 2021). In line with the theorization of agency as existing along a continuum, people can interpret varying levels of control over different spheres of life, issues, time-horizons, or scales (Hitlin and Elder 2007; Mische 2009). However, this literature does not study passivity or the relationship between the two concepts.
Finally, scholarship highlights how subjective understandings of agency in relation to individual choice, goals, and decisions are reinforced within Western, neoliberal societies that ascribe value to an autonomous self (Lamont 2019; Markus and Kitayama 2003; Markus 2017). American culture assigns moral value to this conception of agency—individuals should be autonomous and independent because they are less worthy people when they are constrained or dependent (Burkitt 2016; Reader 2007). As a result, individuals strive to find opportunities to construct themselves as agentic in order to achieve a sense of value and worth (Lamont 2019; Ho 2023). In contexts of constraint, individuals reinterpret their environments by focusing on what is within their control and pursuing the criteria for social worth that are available to them (Anderson 1999; Willis 1977). A cultural and contextual conceptualization of agency challenges the assumption that people in disadvantaged positions understand and describe themselves as having less agency simply because they face increased structural barriers (Edin and Lein 1997; Edin and Schaefer 2015) or because they may narrate agency and choice differently than more advantaged people (Abend 2018; Markus and Kitayama 2003; Stephens, Fryberg, and Markus 2011). The importance of constructing an agentic self to claim moral worth highlights the need to understand strategies through which people narrate agency even in the face of constraints.
FROM AGENCY TO AGENTIC MOMENTS
To understand how people express agency in various aspects of their lives, we analyze how people use narratives (McAdams 1993; Somers 1994; Ewick and Silbey 2003; Polletta et al. 2011) to position themselves as agentic in specific situations (Berger and Luckmann 1966; Garfinkel 1967; Tavory 2018). We inductively develop the joint concepts of agentic moments and narrative moves, through which we argue that people’s narratives of agency are variable, situational, and often co-occurring with narrative passivity, enabling people to narrate themselves as agentic even in challenging situations.
We turn to narratives to examine people’s expressions of agency because narratives are structured stories through which people interpret their lives (Ewick and Silbey 2003; Polletta et al. 2011). Through stories, individuals can remember and retell events from their lives in ways that allow them to feel as though they had control over their circumstances, acted intentionally, and fulfilled agentic norms. Thus narratives are ideal for analyzing how people express their agency (and passivity) in particular contexts. Moreover, narratives are crucial to constructing a sense of identity (McAdams 1993; Somers 1994). Whereas past work shows how people assert virtuous identities by narrating agentic imagined futures (Frye 2012; Ayala-Hurtado 2021; Zilberstein, Lamont, and Sanchez 2023) or justifying past experiences in interaction (Scott and Lyman 1968; Damaske 2013), we use narratives to understand how people recount their pasts as agentic to cultivate an agentic presentation of self (Enck and McDaniel 2015; Goffman 1959). Through these stories, individuals construct agentic identities, allowing them to claim moral worth and sustain a basis of self that they can extend into their imagined futures (Frye 2012; Somers 1994).
Furthermore, we draw on theoretical research that presents agency as contextual and on a continuum to understand people’s narrations of agency as both situational and intertwining with passivity. First, we analyze specific situations—defined as actors and their settings (Berger and Luckmann 1966; Garfinkel 1967; Tavory 2018)—in which people narrate aspects of their lives more or less agentically. Second, just as agency and passivity may shift along a continuum based on context (Abebe 2019; Burkitt 2016; C. Silver et al. 2021), people may also narrate varying levels of agency and passivity, and even both together in different discrete situations. For instance, an individual might experience constraints, such as poverty or medical diagnoses, that lead them to describe themselves as generally powerless within a given situation; however, these same challenges may also position individuals to narrate themselves as agentically developing creative solutions for the problem at hand. Thus the stories people tell about their lives may incorporate different levels and expressions of agency within specific contexts and may even fluidly incorporate descriptions of agency and passivity within the same context.
We thus inductively develop a situational conceptualization of people’s narrations of agency, which we term agentic moments. Agentic moments function as narratives in which people describe themselves as acting agentically in particular situations.2 These moments can be brief, can be composed of different ways of claiming agency, and can be closely intertwined with descriptions of passivity. The concept of agentic moments is not applicable in instances where individuals narrate a situation entirely passively.
Agentic moments are often facilitated by narrative moves through which individuals adapt narrations of their realities, and particularly of situations in which they encounter constraints, in ways that help them envision themselves as the protagonists of their lives. The concept of narrative moves explains how people construct agentic selfhood through narration. These planned or unplanned narrative strategies manifest in at least two ways. First, when considering their ability to affect their environments, individuals may shift the focus of their narrative—whether spatially, temporally, or socially—to create opportunities to enact an agentic self (Sanchez, Lamont, and Zilberstein 2022; Ayala-Hurtado 2021). Second, people may draw on different ways of claiming agency to emphasize their agentic selves. Agency can range widely, including planning, deciding, or acting in relation to one’s goals, but these types of agency have seldom been considered separately. When individuals find that some types of agency are unavailable to them, other types can help them maintain a sense of agentic selfhood. For instance, in situations marked by concrete challenges that effectively eliminate choice, individuals may still deploy narrative moves to claim agentic selfhood by narrating their active decision to do nothing (Willis 1977). Consequently, even amid a general lack of control, individuals can identify moments through which they can maintain a narrative of the agentic self.
In summary, we investigate narrativized agentic moments, focusing on how individuals narrate themselves as agentic as they describe situations across many spheres of life—including difficult situations in which individuals confront serious constraints. Although structural inequalities undoubtedly shape the choices and resources available to people, we show how people across socioeconomic strata, ethnoracial categories, and other dimensions of privilege narrate agentic moments, and we provide examples of narrative moves that enable individuals to maintain these agentic narratives.
DATA AND METHODS
The American Voices Project, a large-scale dataset designed to enable researchers to study people’s experiences and perspectives in the United States across a wide range of topics, is an ideal dataset to study people’s narrations of agency (for project details, see Edin et al. 2024, this issue). It boasts a nationally representative sample through cluster sampling across census tracts and includes survey and interview data. Interviews in the dataset were collected between 2019 and 2022. The narrative structure of the interviews provides data about how people understand their actions and situate themselves in relation to a variety of domains of life—including work, family, personal finances, health, interpersonal relationships, and more. Given their tremendous scope, these data provide unique analytical leverage that research on agency has not had access to. The data reveal how respondents narrate their agency and passivity without specific reference to either concept (Ahearn 2001; Fausey et al. 2010), which would be more likely to elicit socially desirable responses, that is, descriptions of a more agentic self (Tavory 2020), rather than reveal the resources and strategies used to narrate agentic selfhood.
We iteratively combined computational and qualitative methods to leverage the breadth and depth of the data and verify our findings (Chakrabarti and Frye 2017; Nelson 2020, 2021). Powerful computational methods provided tools to reveal large-scale patterns in individuals’ expressions along an agency-passivity continuum and narrow the analytic focus to the level of the situation, while qualitative methods allowed us to dig deeper into the narrative data (Voyer et al. 2022).
We began by inductively reading forty randomly selected interviews to familiarize ourselves with the data and identify themes related to agency and passivity for further analysis. We paid particular attention to the ways that respondents described their agency and passivity and the parts of the interview in which these themes arose. Given the varied contexts and sections in which relevant themes occurred and the idiosyncratic structure of the semi-structured interviews, we decided to analyze the entire interview texts, allowing us to assess agentic narrations in a variety of situations.
Concept Mover’s Distance Model
We then turned to a computational analysis, developing a measure of agency for the interviews using the concept mover’s distance (CMD) method (Taylor and Stoltz 2020; Stoltz and Taylor 2019). This method has recently been used to discern cultural schemas and frames in a variety of contexts (Taylor and Stoltz 2020; Carbone and Mijs 2022; Daenekindt and Schaap 2022; Voyer et al. 2022). CMD relies on word embeddings, a popular method in the computational social sciences (Kozlowski, Taddy, and Evans 2019; Durrheim et al. 2022; Boutyline and Soter 2021). Word embedding algorithms use massive, digitized text corpuses to create high-dimensional vector spaces in which each word in the training corpus is represented as a vector within the space. Thus each word is positioned distinctively within the vector space. Words that frequently co-occur, appearing in semantically similar contexts and understood to have related meanings, are positioned near one another in the vector space, whereas those with more dissimilar meanings are positioned further apart. If, for example, the words agency and control appear in semantically similar contexts in the training corpus, they would be positioned near one another in the vector space.
CMD builds on this technique and “word mover’s distance” (Kusner et al. 2015) with the intention of measuring concept engagement in texts. To do so, it calculates the similarity between a given document’s words and a concept, operationalized as an ideal pseudo-document that includes only words related to that specific concept. For instance, if the pseudo-document contains only the word agency, some documents’ words would be closer in the vector space to this one-word document and thus expected to more strongly invoke the cultural idea of agency, while those of other documents would be further away. An important characteristic of CMD is that the word or words in the concept-specific pseudo-document (say, agency) do not need to explicitly appear in the documents to measure their distance to the concept. For example, if agency and control, as well as other similar words, are positioned near one another, the inclusion of words like control in the analyzed documents would signal a closeness to the concept of agency, even if the word agency does not appear in the document. For this reason, CMD is a more robust method than a dictionary method that depends on the specific words researchers choose to include in the dictionary.3
Because of the complexity of the concept of agency, we drew on the semantic directions subtype of CMD (Taylor and Stoltz 2021), which creates a pseudo-document based on the average of a researcher-compiled list of words rather than a single word.4 The semantic directions method uses antonym pairs rather than words, such as liberal-conservative (Taylor and Stoltz 2020). This method has the added benefit of allowing us to measure agency and passivity along a continuum, as some theorizations of agency have suggested (Abebe 2019; Fuchs 2001; C. Silver et al. 2021). We created a list of twenty-one antonym pairs (for example, decisive-indecisive) that were linked to the extremes of an agency-passivity continuum; the majority of the words on the agency side were categorized as firmly agentic in previous research (see especially Ash, Stammbach, and Tobia 2021, table 1; Sayer 2011; Schweitzer and Waytz 2021, appendix A), while the rest were developed inductively from our qualitative reading of the data (see table A.1). Documents with higher numbers on this measure had words closer to the concept of agency, while those with lower numbers were closer to the concept of lack of agency, or passivity.
Model Descriptive Statistics
To run the models, we read all interviews into R and isolated speech labeled interviewee, interviewee 1, or participant to capture relevant respondent text and exclude text spoken by the interviewer. Our sample includes 1,521 interviews.5 We also segmented interviews into two-hundred-word chunks with twenty-five-word overlaps to account for variation within interviews, more easily identify the most meaningful parts of the texts, and account for the arbitrariness of segment boundaries (Chakrabarti and Frye 2017; Sherin 2013; Voyer et al. 2022).6 We dropped segments with less than 15 words, resulting in 95,813 segments.7 For both full and segmented datasets, we cleaned the text to remove numbers, symbols, whitespace, punctuation, and meaningless words.8 To generate the CMD models, we used the fastText word embedding tool common in social science research, trained on billions of web pages from the Common Crawl (Stoltz and Taylor 2019; Taylor and Stoltz 2020; Voyer et al. 2022).9
We ran a CMD semantic directions analysis on the full interviews to obtain an overall ranking of respondents along the continuum of agency to passivity, as well as on the segments to reveal differences within respondents. This method produces a number for each interview or interview segment that represents how closely the text aligns with the concept of agency, positive numbers, or passivity, negative numbers (for descriptive statistics, see table 1). We also calculated the number of words respondents spoke to account for their talkativeness. We used quantitative methods to analyze the distribution of expressions of agency and passivity and identify patterns.10 In doing so, we extend theories that conceptualize agency as a continuum to study narrations of agency (Abebe 2019; Burkitt 2016; C. Silver et al. 2021).
To date, no standardized validation measures for CMD models have been established. Therefore, like earlier research (Voyer et al. 2022), we relied on a mixture of qualitative and interpretive checks to ensure that the model captured the concepts of agency and passivity. First, we qualitatively analyzed the thirty most and the thirty least agentic segments. Overall, the segments made sense to the researchers and captured themes and language expected to be more or less associated with agency or passivity. Second, we compared the one hundred most frequent words in the top 5 percent (most agentic) and bottom 5 percent (most passive) of the sample. The results show that segments from the top 5 percent frequently incorporated words we would expect to be associated with agency, such as change, use, and savings, that are related to intentional and goal-driven action.11 Segments from the bottom 5 percent more frequently included words we would expect to be closely associated with passivity, such as pain, depressed, and couldn’t, that respondents used to describe constraints or feelings of helplessness. We understood these checks as indicating that the model appropriately captured agency and passivity.
Qualitative Analysis
To understand the meaning of agency in specific situations, we qualitatively coded a subsample of interview segments. A qualitative analysis of segments allowed us to find and access respondents’ narratives. Using the results of the semantic directions model, we randomly selected five hundred segments from the top 5 percent (most agentic), five hundred from the bottom 5 percent (most passive), and three hundred from the middle 5 percent.12 The segments in the top and bottom 5 percent had scores higher than 1.5 or lower than -1.5, which previous studies using CMD considered substantial (Carbone and Mijs 2022; Voyer et al. 2022). This sampling strategy allowed us to analyze how agency and passivity manifest in respondents’ narratives in a variety of situations. Situations could extend over multiple segments, and segments could contain multiple situations. We coded the segments in NVivo; we focused on coding narrations of agency or passivity. We also coded for different spheres of life, like career, family, health, or politics; social scale, like individual, interpersonal, community, or society; and specific constraints, such as discrimination, substance use or addiction, poverty or financial hardship, or incarceration, among others.
While conducting the qualitative coding, we were typically unaware of the metadata attached to each segment that labeled the text as more agentic or passive; this information was stored separately from the narrative segments. Nonetheless, throughout the analysis, we were surprised by how frequently we identified brief narratives of agency in segments that the CMD model had labeled as highly passive, or brief narratives of passivity in segments the CMD model had labeled as highly agentic. In these cases, our in-depth qualitative analysis was able to identify narratives of agency and passivity within segments that by and large were representative of the other end of the continuum. For instance, a segment in which a respondent described her medical condition might be computationally labeled as passive and might seem largely passive on close reading because the respondent was describing hardships and blocked opportunities due to a medical disability. At the same time, our in-depth analysis could identify brief moments of agency as the respondent recounted learning how to manage this medical condition. Because the computational and qualitative analyses focus on distinct analytic structures (the general use of language in the segment versus the narrative meaning of words in the situation), and because the CMD model assigns a numeric value to each segment along a continuum, it does not invalidate our analysis for a segment categorized as highly passive by the computational analysis to still include agentic language. These insights led us to a theorization and analysis of the interactions and coexistence of agency and passivity (Timmermans and Tavory 2012). We draw quotes primarily from the interview segments in our subsample, as well as occasionally drawing on contextual information from the full interviews around those segments.
FINDINGS
We first draw on quantitative methods to understand the distribution of expressions of agency and passivity across the sample—including both between respondents and within respondents. We then turn to a qualitative analysis of select interview segments to develop the concept of agentic moments and identify narrative moves.
Agency and Demographics: A Quantitative Analysis
We drew on a quantitative analysis to understand how expressions of agency and passivity were distributed along the agency-passivity continuum within our sample—including both among the full-text interview transcripts and among the two-hundred-word interview segments. We employed MLE regression methods on the sample of full interviews with demographic variables such as gender, age, race-ethnicity, and educational attainment as independent variables and the CMD score for each interview as the dependent variable (see table 2).13 The results demonstrate that some privileged characteristics, specifically, having a college education or identifying as a man, are significantly correlated with higher values along the agency-passivity continuum (p < .05 for gender (man = 1) and p < .001 for education), suggesting that people with these characteristics are likely to use more agentic language to describe their experiences. Meanwhile, other characteristics associated with inequality, including race and income, are not significantly correlated with different values. These results are in line with some previous scholarship about the relationship between privilege and agency, notably work on education and gender (Duncan 2015; Lareau 2002; Mirowsky and Ross 2007). However, the results are inconsistent with scholarship on other forms of privilege, especially regarding income (Kohn 1989; Kraus, Piff, and Keltner 2009).
Regression of CMD Score and Demographic Characteristics
Given our interest in exploring variable and fluid narrations of agency, we then drew on descriptive statistics to further understand patterns in expressions of agency for individual people. We investigated whether there was more variance in agency-passivity CMD scores between different respondents (inter-interview variance) or between different segments of the same respondent’s interview (intra-interview variance). To do so, we calculated the intraclass correlation coefficient using the segments as observations and the respondents as clusters. This analysis produced a value of 0.15883. This value indicates that there is much more variance in agency-passivity CMD scores within clusters—that is, between interview segments within the same respondent—than between clusters—that is, between different respondents. This finding is in line with the research on cultural and contextual studies of agency that recognizes variation in experiences of agency over the life course or based on contextual factors (Abebe 2019; Frye 2012), showing that expressions of agency are also variable. Such a finding is not unexpected because respondents are discussing a wide range of topics in the data; however, these results also bolster our intuition that analyzing expressions of agency at the level of the individual masks variation of within-individual expressions of agency, including how agency manifests in constraining situations and in relation to passivity. We turn to the qualitative data to conduct an interpretive analysis that allows us to understand how agency functions at the level of the situation and develop the concept of agentic moments.
Qualitative Findings
We develop the concept of agentic moments by showing instances in which narratives of agency vary within individuals and in which narratives of agency are evident in constraining situations. First, we demonstrate instances of within-individual variation in the use of agentic (and passive) narratives in different situations. Second, we assess and analyze the co-occurring nature of descriptions of agency and passivity within a respondent’s narrative of a single situation. Finally, we highlight two narrative moves that facilitate agentic moments in the face of obstacles that may be thought of as limiting agency.
Intra-Individual Agentic and Passive Moments
Respondents often narrated substantial agency in some situations while narrating substantial passivity in other situations, emphasizing the importance of understanding narratives of agency as situationally dependent. We demonstrate this finding by drawing from respondents with more than one segment in our qualitative subsample, where one segment was quantitatively coded as the top 5 percent on the agency-passivity continuum and qualitatively coded as agentic, and the other was quantitatively coded as the bottom 5 percent on the continuum and qualitatively coded as passive. Sixty of the 788 respondents included in the subsample—7.6 percent—belonged in this most extreme group.14
For example, Melissa, a White woman whose household total income was low and who did not have a bachelor’s degree, primarily used passive language as she described abusive relationship dynamics with her former partner and the father of her children. In the segment coded as highly passive, Melissa described an incident of domestic violence: “I screamed for somebody to call the cops but nobody [did]. My daughter, she was [a toddler]. She was sitting right there on the edge of the bed. It was a small room. There was the bed, the entertainment center… . and that’s where he threw me. She was on the edge of the bed watching the whole thing, screaming.” In this segment, Melissa primarily narrated a moment of passivity—she described herself as thrown around, as unable to get away from her partner, and as incapable of protecting her screaming daughter who was a painful witness to the event—which Melissa later described as having had long-lasting repercussions for her child. Although she narrated a brief example of agency through her cry for help, she overwhelmingly described the situation using passive language.15
Although Melissa generally narrated situations relating to her abusive relationship passively, she narrated situations related to some other aspects of her life highly agentically. In particular, she repeatedly voiced her profound determination to make a better life for her children, especially with regard to education. She described her persistence in advocating on her children’s behalf, which she said had earned her a reputation at school. In the segment coded as highly agentic, Melissa described her relationship with the school principal, with whom she was frustrated due to his lack of action in fixing issues at the school: “I’ve been in [the principal’s] office a couple times talking to him… . If I feel like you’re not taking care of my kids, right, then I’m going to come up there and we’re going to have a problem… . I’ve gone up there and gone off how many times. They get to the point where they know me. They know my voice.” Melissa remained steadfast in voicing these issues to the principal; her dissatisfaction with the school did not lead her to narrate herself as resigned or as believing the conditions were unalterable. She described how she directed her communications to a specific actor as a catalyst for problem-solving. She proudly noted that her children were doing well in school and earning distinctions as honor roll students.
Not only was Melissa capable of narrating different facets of her life agentically and passively, but the two facets were interconnected through their relationship to her children. In the passive segment, Melissa narrated a violent experience where she had felt unable to act, focusing intently on her daughter’s presence and negative reaction, and, implicitly, her inability to protect her. In the active segment, Melissa described the intense energy she devoted to caring for her daughter and her other child, invoking powerful language that reclaimed her agency: “if … you’re not taking care of my kids … we’re going to have a problem.”
Other respondents similarly revealed how expressions of agency and passivity are situationally dependent. For example, Natalie, a more privileged respondent—a biracial woman whose household income was high—described herself as a workaholic unhappy with her job. However, she narrated her imagined career trajectory in a highly agentic way: “whatever I do, it’s gotta be with helping [others]. Only cause of where I come from … I just see it to be different to where I’m not stressing out trying to make it to a job that really doesn’t care about their employees … when my business open up, it will be a totally different atmosphere because I’ve been through what people go through. So, I’m hoping that five years from now or a year from now that I will be closer to that.” Natalie narrated her professionally oriented agency through her belief in a better professional future for herself in which she would be able to make a difference and support others at work. Implicitly, she connected this agency to her own previously precarious work trajectory.
In contrast, Natalie expressed passivity when describing her relationship with her father, referring to it as a trauma. In the segment coded as highly passive, she described her feelings when her father was arrested: “I said, ‘This is some bullshit. He got arrested?’ … And we wound up going down to the Central Booking. It’s so embarrassing. It’s not as bad as it is hurtful. Right, so, painful, hurtful, however you want to describe it … we sitting in the courts, in the middle of the fucking night. When my father came, I was so embarrassed about how he looked. And we waited.” Natalie’s repeated descriptions of the hurt, pain, and embarrassment she felt, as well as her experience waiting, indicate a narrative of passivity. She felt powerless and unable to intervene in the situation. These segments exemplify the ways in which people narrate some experiences as passive and others as agentic, even sometimes, as in Melissa’s case, using the two experiences as counterpoints.
Co-Occurrence of Agentic and Passive Moments
Furthermore, we show that narratives of agency and passivity are not only drawn on by the same respondent in different situations, but frequently co-occur and enable each other in the same situation. We turn to segments narrated by respondents that the prior literature or our computational analysis would associate primarily with expressions of passivity because of characteristics associated with increased constraints (low education or socioeconomic status). Our analysis finds concurrent narratives of agency and passivity, even as respondents narrated difficult circumstances. Our concept of agentic moments captures the simultaneous and co-constitutive nature of agentic and passive narratives and the ways in which constraining challenges often enable agentic responses.
For example, Dan, a White man with no college degree and a low household income, described an encounter with the police in which he felt powerless: “[I] just was wandering on a walk and what happened is that … police swarmed me. They tasered me but God protected me. The taser did connect, I felt the electrical current but didn’t flop around like a fish, didn’t feel pain, you know. I just felt a tingling sensation. God was definitely protecting me. The cop just approached me.” Violent police encounters are situations in which civilians frequently experience limited choices, powerlessness, and pressures to submit, qualities associated with passivity. Additionally, given the association between educational attainment and agency-passivity in our regression results and in previous research, Dan’s lack of a four-year college degree would make crafting an agentic identity and experience less accessible. Indeed, Dan narrated the instant that the police tasered him using passive language. His description of the unexpected nature of the encounter, in which “the cop just approached me” when he was “wandering on a walk” enhanced his narrativization of himself as passive: lacking knowledge about why the event occurred and control over its occurrence. Further, his invocation of God credited his sensations to an external entity, locating agency and protection outside of the self, and emphasizing a lack of ability to change or control the situation.
However, Dan simultaneously expressed agency as he described his ability to resist police action, narrated the situation as an instance of unlawful arrest, and revealed plans to pursue legal action. In the moment of the police contact, he described how he “wrestled with the rest of [the police].” He positioned himself as exerting effort to actively change the course of the event. He continued to describe the situation as a “kidnapping incident with the police.” His reflection on the event as not his fault and out of his control prompted him to proactively plan to take legal action. He explained how he “already has an attorney and they are suing” to address the instance in which he “resisted unlawful arrest.” Dan’s narration of the situation as an instance of resistance and fault on the part of the more powerful actors demonstrates his effort to craft an agentic narrative in relation to his experience of constraint.
Another example of a respondent who narrated both agency and passivity in response to a constraining situation was Molly, a biracial woman with less than a high school degree and a low household income. When asked about her financial situation, Molly voiced passivity, saying, “it’s not the best. I mean I’m broke as hell.” She continued responding to this question just after the segment ended: “I would like to improve it, it’s just it’s hard trying to get a job when you don’t have a [degree], and you don’t have reliable transportation, and you don’t have much of a work experience.” She noted her difficult financial situation and the structural obstacles standing in her way as she contemplated the possibility of a better life.
Nonetheless, when imagining how she would respond if her financial situation unexpectedly improved, Molly was able to plan and prioritize tasks in an agentic manner. She voiced agency when describing how she would organize the use of the money: “It depends on the urgency. If a bill needs to be paid or something like that, or if we needed food or whatever, that would be the main thing that I would think of first, either a bill or food-wise or if we just need your basic everyday stuff like soap or lotion, stuff like that … if that is taken care of, [I’d] see if the kid may need anything. After that, I would most likely just save it or use it for myself or keep it for myself.” Despite her difficult financial situation and lack of education, Molly described herself as capable of handling such a situation with ease, prioritizing and making decisions about how the money should be used.
In other instances, respondents narrated intertwined moments of passivity and agency despite being in privileged positions associated with fewer experiences of constraint. Sarah, a White woman with a degree and a high household income, described her attempts to manage anxiety and depression exacerbated by disruptions from the COVID-19 pandemic. As a social worker for children, she found disruptions to her routines challenging and anxiety provoking. Anxiety at work affected her general well-being and emotions. In response to questions about her mental health she recounted, “I struggle with bursts of anxiety for sure. I definitely lean on the more anxious side of things … being kind of stuck in the same place brings on symptoms of depression just by the sheer fact of not being able to leave and experience new things.”
Sarah described pandemic-related disruptions as contributing to her feeling a lack of control, predictability, or ability to plan or prepare for the immediate or long-term future. In her narrative, these experiences of passivity led to general feelings of depression and anxiety. In response to these emotional changes, Sarah also recounted her attempts to manage her depression and anxiety: “I’m proud of how I’m doing emotionally. I’ve made it a priority of myself over anything… . I guess going back to a routine, finding ways to incorporate some sense of normalcy or newness into your days is important. I would say I’ve struggled with bouts of both [anxiety and depression] … but I’ll try to focus on managing it.” Sarah narrated agency through her prioritization of mental health goals, her efforts to find routines to plan for and predict the immediate future, her attempts to restore a sense of normalcy, and her focus on managing her problems through deliberative and reflective action. She described how she was “not as worried as I was before” about work by the time subsequent waves of pandemic disruptions occurred. She suggested that her ability to anticipate responses to disruptions and her development of tools to manage them meant that “it wasn’t as bad, you know, that is comforting. That shows that there’s been improvement.” Sarah’s experience shows how people can narrate events as simultaneously producing experiences of passivity and agency. Furthermore, passive narratives provide opportunities for respondents to craft agentic narratives as they describe how they develop strategies and routines to overcome their challenges. Instances of resistance thus serve to increase respondents’ sense of agency as they draw on those past experiences to confront subsequent obstacles.
Another example of a respondent who voiced both agency and passivity was Tom, a White man with a degree and a high household income. Despite being male and having financial and educational resources associated with expressions of increased choice and agency, Tom recounted his many run-ins with the Social Security system that presented challenges and constraints. He recounted his experience having to go to the Social Security office because he had received a lower amount than anticipated on a recent check: “Here’s the hard thing about that. You have to go to the office, and you can wait half to three-quarters of a day before you get in to see someone to talk about your issue. Even then, part of the time, you can’t get it settled, so you have to come back again or you have to wait on them or you have to provide more information. It’s not the easiest process … I have to go see them and find out what has happened here, and I haven’t had time.” Tom narrated passivity as he described the many obstacles in his path to getting the help he needed (“the hard thing about that”). He described the length of time he would have to wait in the office to even communicate with someone. He explained the difficulty of the process, suggesting that he might be forced to return to the office more than once to get help. At the same time, he narrated agency as he described his determined and persistent actions to resolve the issue: “I have to go talk to them about [it]”; “I have to go see them.”
Narrative Moves that Enable Agentic Moments in Constraining Situations
Beyond showing that the same respondents can narrate passive moments in some situations and agentic moments in others, as well as that they can narrate passive and agentic moments within the same situation, we also identify common narrative moves respondents use to position themselves as having agency despite the constraints they face, and that thus facilitate agentic moments. We focus on two kinds of moves: how respondents narrow or expand the focus of their narratives and how respondents highlight specific types of agency to facilitate agentic moments.
One narrative move that allowed respondents to claim a sense of agency in unexpected circumstances was shifting their focus to maintain a sense of control. Emily, a White woman with a degree and a high household income, described how she dealt with her mother’s health challenges:
her health is going downhill very quickly … and I’m here and I can’t really do anything. I feel guilty that I’m not [there] and I can’t help her, but I also feel [that] thank God I am [here], because if I was [there] I would have to help her with things constantly and it would probably be a very stressful situation for me. Being miles away is … good because I’m very separated from it… . I kind of just pretend it’s not happening most of the time so it doesn’t affect my day-to-day life or work or anything even though I feel like I should be worrying a lot more than I do.
Facing her mother’s debilitating illness, Emily described little agency in addressing the course of the disease. Instead of only narrating passivity in response to a situation outside of her control (“I can’t really do anything”), Emily was able to narrate an agentic moment by focusing on her everyday routines, work life, and immediate environment. She expressed an implicit choice to shift her focus to her day-to-day life rather than focusing on her inability to help her mother. In this instance, by reducing the spatial and interpersonal scale of her focus, Emily crafted a narrative of agency and control and was able to regulate stress and worry.
Contrastingly, Ashley, a Black woman whose household income was low, facilitated an agentic moment by shifting her narrative to expand her focus. She discussed her difficulty controlling her temper and quick emotional reactions, initially describing herself using passive language, as out of control: “I know I have a messed up attitude. I got a messed up temper. I go from zero to a hundred real quick.” However, she continued:
I try to just kind of stay to myself. I try to treat people how I want to be treated, talk to people how I want to be talked to… . I think I’m doing real good. Everybody get depressed, everybody living with different things that’s going on in their lives and it makes them depressed. But I don’t be depressed… . If I’m sad or if I’m going through anything nobody would ever known about it unless I opened my mouth and tell them because I carry it real well. I still smile. I still talk to people.
In contrast to Emily’s shrinking her focus to retain a sense of control, Ashley expanded her focus beyond herself to narrate her emotional reactions as a deliberate choice. First, by acknowledging that “everybody living with different things,” she compared her emotional state with that of other people in society, recognizing alternative challenges and ways of reacting. Then, by reflecting on how she wanted to be treated and thinking about how her behaviors and emotional responses affected others, she narrated her conscious and deliberate choices to control her emotions and self-presentation. In this instance, shifting to expand her focus led Ashley to identify multiple ways to react and narrate herself as an agent who consciously chose a specific self-presentation.
Another common narrative move was for respondents to draw on different ways of claiming agency that facilitated agentic moments in trying situations, particularly highlighting their ability to make difficult decisions or overcome obstacles. For instance, Pam, a Black woman with a degree and a low household income, faced chronic health struggles that were outside her control. Despite her very difficult health circumstances, Pam described herself as determinedly overcoming obstacles, making decisions, and maintaining her self-reliance to construct an agentic self-narrative. She described her strategies to treat her pain:
I said, well, Lord, I just have to deal with some of this pain. I said as long as it’s not to the point where it’s excruciating, if it gets excruciating then I have to go somewhere, but it’s just like annoying, very annoying, aching. At night it gets really, really terrible, but when I take that Tylenol PM … it helps me sleep. It kind of calms it down … but anything is manageable. I think you have to put your mind to anything. I don’t worry about it. Back when [the doctor] did have me on those medicines I wouldn’t take so much. I would probably take half a piece of the pill and I said well, Lord, if I can get through with that I can make it with this.
Although curing her pain remained outside her ability, Pam crafted an agentic moment by focusing on her ability to manage her health condition and relying on her own remedies over stronger medications or professional help. She learned many of these techniques, such as taking Tylenol at night, through her past experiences dealing with the challenge, which helped support her belief that she could continue to manage the pain and “get through,” overcoming any obstacles that came her way. Additionally, she differentiated between “annoying” and “excruciating” pain, each of which required a different approach to management. By delineating strategies for managing these types of pain, she was able to narrate herself as an actor able to make decisions about how to address her health condition.
In a different context, Ben, a biracial man with no degree and a low income, discussed two complicated recent experiences, one where he “got in trouble” for protecting neighborhood children from the police, and another when his identity was stolen. Discussing the identity theft, Ben voiced a passive moment, saying that he had trouble finding a lawyer, and one had even hung up on him: “a lot of them have probably been a little hesitant because I’ll be taking on the state at this point… . I haven’t had the best of luck yet.” He thus narrated his lack of options and difficulty moving forward. Nonetheless, he positioned himself as actively “working on it,” saying, “I’ve got a couple ideas left.” His focus on his determination to overcome obstacles allowed him to narrate himself as agentic despite his frustrating experiences. Moreover, despite his previous negative experiences with the police, he narrated an agentic moment in his most recent experience with them: “I ran into some nonsense with cops and stuff like that, but I usually just play my role, play it cool, and just ask real, like personal questions. I’m an asshole sometimes. I like to be sarcastic and stuff, if I feel bold.” By demoting his experience with the cops as nonsense and foregrounding his performance in front of them, he similarly highlighted his ability to skillfully overcome obstacles to describe himself as agentic.
DISCUSSION AND CONCLUSION
In this article, we examine how people narrate agency—their ability to construct goals, make decisions, plan for the future, and develop pathways of action toward their objectives in their everyday lives as they face obstacles and constraints. By analyzing the American Voices Project data using computational methods, we show that respondents’ narratives of agency at the level of the individual align with some research concluding that social disadvantage constrains agency, but that this level of analysis fails to reveal the variation in narrations of agency between the many situations an individual may encounter. By performing a qualitative analysis of interview segments, we find that respondents narrate agency in some situations and passivity in others, can narrate a single situation in a way that is simultaneously agentic and passive, and use several maneuvers to describe themselves as agentic, even in challenging situations. Based on this analysis, we develop the concept of agentic moments, highlighting individuals’ capacity to narrate agency in constraining situations. Agentic moments result from narrative moves individuals employ to support narratives of agentic selfhood. Thus we argue that people’s narratives of agency often co-occur with narrative passivity and are variable and situationally dependent. We build on research to propose a cultural and contextual model of agency that captures how people are able to narrate an agentic self, regardless of their levels of privilege or the constraints they encounter.
Our analysis contributes to research on agency by proposing agentic moments as a narrative and situational concept that illuminates how people craft agentic narratives, including in challenging contexts. The concept of agentic moments captures the co-occurrence of agency and passivity, and the ways in which people can narrate both. As we empirically demonstrate, agency and passivity often interact—experiences of hardship create openings for agentic moments. People narrate agentic pasts to achieve ideals of an agentic self despite challenges. Importantly, our mixed-methods analysis of a nationally representative interview-based dataset enables us to demonstrate that people across socioeconomic strata, ethnoracial categories, and other social dimensions craft agentic moments to narrate themselves as the protagonists of their lives in a multitude of different contexts. We thus extend the research tradition demonstrating that people, including people in disadvantaged positions, draw on cultural tools to present themselves as agentic and thus craft a socially worthy identity for themselves (Anderson 1999; Lamont 2019; Edin and Schaefer 2015), particularly in a highly individualistic American society (Lamont 2019; Markus and Kitayama 2003; Markus 2017). Future scholarship should continue to examine the meanings and manifestations of agency to explain how individuals narrate and understand their abilities to overcome challenges (Markus and Kitayama 2003). Research could develop explanations for why and under what circumstances participants blend high- and low-agency narrations, including by examining how cross-national cultural repertoires of the worthy self may influence people’s narrations in different national contexts, or analyze how agentic moments manifest across different spheres of life (that is, in relation to work, family, health), scales, or temporal horizons. Further, future research is necessary to unpack concepts, experiences, and narratives of passivity, including how and why individuals may narrate passive moments.
Relatedly, we identify narrative moves as strategies that people deploy to craft agentic moments and which illuminate how people can construct agentic selfhood through narration. We show that individuals can shift their focus within their narratives, expanding or restricting their scope, to maintain a sense of control. For example, people may narrate more agency by focusing on the here-and-now rather than on personal challenges that seem intractable, like chronic health issues, or broader challenges, like climate change. Thus, along with temporal frames (Mische 2009), perceptions of scale play a role in how people work to position themselves as agentic. Additionally, we show that people can narrate agency by focusing on different ways of claiming agency, including overcoming challenges, making decisions, and espousing self-reliance. Future research is necessary to identify additional narrative moves and investigate how narrative moves are patterned. Furthermore, scholars should explore more deeply the multiple ways that people conceive of agency, including by developing a typology of agency.
Last, we contribute to mixed-methods research in showing the benefits of combining computational text analysis and qualitative analysis of interviews to explore large-scale datasets. Powerful computational methods allowed for a breadth of focus and attention to detail that would otherwise be impossible or invisible to a human reader (Voyer et al. 2022). Meanwhile, qualitative methods allowed verification of computational findings and interpretive analysis of the specific ways in which narratives of agency and passivity manifest. Nonetheless, we primarily drew on methods used to study published texts, like books, lyrics, or internet pages. Future research should continue to develop strategies specifically designed to analyze large-scale interview transcript datasets such as the American Voices Project.
Our use of interview data and focus on narratives allow us to reveal how people describe their abilities to affect their environments and overcome challenges. Given these data, we cannot draw conclusions about how people enact agency in interaction with others or tangibly affect their environments. Additionally, we recognize that interview settings may encourage respondents to talk about their experiences using more agentic language as they are asked to reflect on their lives and share their perspectives (Tavory 2020). Nonetheless, narratives structure how people understand their past, present, and future actions, and inform how they conceive of their sense of self and capabilities. Developing a complex understanding of agency is critical to elucidating the relationship between structural constraints and cultural narratives of the self.
Our research dovetails with the work of others in this double issue that speak to the complexity of agency by considering the interplay between subjective understandings and social forces (Rocha Beardall, Mueller, and Cheng 2024; Cramer, Youngling, and Rooker 2024; Sauder, Shi, and Lynn 2024, this issue). Other work in this double issue shows the utility of the concept of agentic moments indirectly by identifying narratives in which individuals make choices, formulate plans, and express judgments (Abramson et al. 2024, this issue; Chu and Lee 2024, this issue; Hiebert, Kahris, and Seefeldt 2024, this issue; Jackson 2024). In concert with these articles, we leverage the breadth and depth of the AVP data to illuminate the experiences of the American people, focusing on their narrations of agency as they navigate the challenges they encounter and pursue the futures they desire.
Appendices
Antonym Pair Word List Used in CMD Model
Covariate Variables
FOOTNOTES
↵1. We use interviews as evidence of shared frameworks and classification systems that shape people’s understandings of their environments and actions (Lamont and Swidler 2014; Pugh 2013), rather than as evidence of internal states or feelings (C. Silver et al. 2021; Moore 2016). Within our approach, interviews always involve an aspect of presentation of the self (Goffman 1959), raising questions about the resources, norms and ideals people draw on to interpret and express their identities and actions. We thus analyze people’s narrative depictions of their environments and capabilities, in line with established research using interviews and narrative analysis (Bruner 1991; Ewick and Silbey 2003; Ho 2023; Silva 2012).
↵2. Although we do not discuss them specifically in this article, passive moments are narratives in which people describe themselves as experiencing a lack of agency in particular situations, in which their choices are limited, and they experience constraints.
↵3. Other methods are possible; for example, Almog Simchon, Britt Hadar, and Michael Gilead (2023) operationalize agentive language based on the use of active versus passive voice in statements. CMD has the benefit of foregrounding the meaning of language more broadly, taking into account how respondents describe specific actions, rather than focusing narrowly on verb tenses.
↵4. We decided not to use another subtype of CMD, centroid analysis, which also allows the inclusion of multiple initial terms, because in a centroid CMD model words referring to alternatives to a focal concept (like hate to love or passivity to agency), will be located more closely to words related to the focal concept than if they were completely unrelated (Stoltz and Taylor 2019). This makes these models less suitable for our purposes because the lower boundary of the range of agency has no meaning or direction. We ran centroid models of agency and passivity. Using a qualitative test for comprehension and quantitative tests for consistency, we found that these models captured the top end of agency or passivity but did not capture the range of experiences of agency and became less comprehensible the farther from the most extreme edge of the cases. For these reasons, we found the centroid models unsuitable for analysis.
↵5. Ninety-seven interviews were dropped from the sample because they were mislabeled or not fully translated into English.
↵6. We ran models with segments of different length and found that two-hundred-word chunks yielded the most coherent and succinct segments.
↵7. Respondents’ narratives often extend over more than one segment. This does not pose a problem to our analysis since our analysis is not dependent on counting agentic-passive incidents.
↵8. We used the standard English Stopword package and customized it to our dataset. We added back in 107 words that were meaningful in our analysis, such as pronouns (she, they), common verbs (think, can, do) and adverbs (immediately, likely). We cleaned an additional fifty-nine words specific to the corpus such as “ah”, “hum” and “mis.” We tested the sensitivity of models to different Stopword lists and found that this list produced the most coherent results.
↵9. We chose fastText over word2Vec, more commonly used in sociology because fastText is a newer method that breaks down words more granularly, the creators of CMD and previous scholars using CMD used fastText (Taylor and Stoltz 2020; Voyer et al. 2022), and because of the available pre-trained dictionary trained on the Common Crawl.
↵10. We use demographic variables as coded by AVP (see table A.2). We recode college completion as a binary between those with and without a college degree, because having a college degree is significantly related to various socioeconomic and life outcomes (Autor 2014; Case and Deaton 2021; Dickson and Harmon 2011; Hout 2012). Our analysis holds when using a more fine-grained measure of educational attainment.
↵11. In the case of savings, we saw repeatedly as we analyzed our data that both low- and high-income respondents routinely referred to building their savings or gradually saving money in agentic ways to plan for their futures, expand their alternatives, and achieve specific goals.
↵12. These 1,300 segments came from 788 respondents in the sample.
↵13. We controlled for the length of each text to account for more and less talkative respondents who would, overall, have varying opportunities to describe agency or passivity. We included sampling weights and employed listwise deletion.
↵14. To protect confidentiality, we use broad racial (White, Black, or biracial), educational (bachelor’s degree or not), and household income (low = $35,000 per year or less, high = $100,000 per year or more) categories.
↵15. We discuss the co-occurrence of narratives of agency and passivity, which was very common across the interviews, in the following section.
- © 2024 Russell Sage Foundation. Zilberstein, Shira, Elena Ayala-Hurtado, Mari Sanchez, and Derek Robey. 2024. “The Self in Action: Narrating Agentic Moments.” RSF: The Russell Sage Foundation Journal of the Social Sciences 10(5): 118–40. https://doi.org/10.7758/RSF.2024.10.5.05. We thank Dustin Stoltz, Bart Bonikowski, Jason Robey, Bo Yun Park, Michèle Lamont, Steve Worthington, who provided data science support at the Institute for Quantitative Social Science, Harvard University, and the AVP technical team and editors. Direct correspondence to: Shira Zilberstein, at szilberstein{at}fas.harvard.edu, 33 Kirkland Street, Cambridge, MA 02138, United States; Elena Ayala-Hurtado, at eayalahurtado{at}g.harvard.edu; Mari Sanchez, at mjsanchez{at}g.harvard.edu; Derek Robey, at derekrobey{at}g.harvard.edu.
Open Access Policy: RSF: The Russell Sage Foundation Journal of the Social Sciences is an open access journal. This article is published under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
REFERENCES
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.