How Cultural Meanings of Occupations in the U.S. Changed During the Covid-19 Pandemic

Abstract

Social research highlights the stability of cultural beliefs, broadly arguing that population-level changes are uncommon and mostly explained by cohort replacement rather than individual-level change. We find evidence suggesting that cultural change may also occur rapidly in response to an economically and socially transformative period. Using data collected just before and after the outbreak of Covid-19 in the U.S., we explore whether cultural beliefs about essential and non-essential occupations are dynamic in the face of an exogenous social and economic shock. Using a sample of respondents whose characteristics match the U.S. Census on sex, age, and race/ethnicity, we fielded surveys measuring cultural beliefs about 85 essential and non-essential occupations using the evaluation, potency, and activity (EPA) dimensions from the Affect Control Theory paradigm. We expected that EPA ratings of essential work identities would increase due to positive media coverage of essential occupations as indispensable and often selfless roles in the pandemic, while EPA ratings of non-essential identities would decline. Our findings show patterns that are both clear and inconsistent with our predictions. For both essential and non-essential occupations, almost all statistically significant changes in mean evaluation and potency were negative; activity showed relatively little change. Changes in evaluation scores were more negative for non-essential occupations than essential occupations. Results suggest that pervasive and persistent exogenous events are worth investigating as potential sources of episodic cultural belief change.

Introduction

In late December 2019, a novel coronavirus made the jump from an animal host to a human in Wuhan, China. By the end of May, over 100,000 people in the United States had died of Covid-19, the disease caused by the virus (CDC, 2020). In March, Covid-19 was declared a national emergency and an international pandemic; American states began issuing stay-at-home orders mandating the closure of non-essential businesses (AJMC, 2020). Employees that could work from home began to do so; people who worked in non-essential occupations and who could not work remotely were often unable to work at all. By April, over 23 million Americans were left unemployed as the unemployment rate increased by 10.3 percent in a single month to 14.7 percent, the highest rate and largest over-the-month increase since the Bureau of Labor Statistics began keeping records in 1948 (BLS, 2020).

Did this massive dislocation in the U.S. work force bring changes to the cultural meanings of occupational identities? Did doctors and nurses who cared for the desperately ill in hospital ICUs become heroes? Did grocery store stock clerks and package delivery people who provided citizens with essential goods become more respected? Did the truck drivers who kept the food supply chain functional seem better, more powerful, or more active? Did people thrown out of work suffer the usual fate of victims, and get derogated for their misfortune, reducing their perceived goodness and powerfulness? (Lerner & Melvin, 1971; Lincoln & Levinger, 1972).

Affect control theory (ACT) offers a clear, precise framework for assessing cultural belief change. In ACT studies, cultural sentiments associated with identities are measured on three affective dimensions: ratings of evaluation (good to bad), potency (powerful to powerless), and activity (lively to inactive). These “EPA ratings” capture important aspects of the affective meanings that respondents from a culture share about occupational titles and other identities. EPA values represent fundamental cultural sentiments in the context of the theory and more broadly in psychological research (Scholl, 2013).

Prior work shows that these sentiments are widely shared within a national language culture (i.e., by those who share both a language and a nationality) and usually very stable over time (see Heise, 2010 for a comprehensive review). This stability is consistent with evidence from longitudinal analyses of probability samples from the U.S. population that show the general stability of cultural beliefs (e.g., Kiley & Vaisey, 2020). It also enables ACT’s theoretical model of behavior, in which actors compare the EPA impressions of the identities and behaviors involved in an event to their fundamental EPA cultural meanings when choosing a subsequent behavior. Stability in shared meanings makes possible a “control system” that guides actors to choose behaviors and reactions that they find circumstantially and culturally sensible.

In spite of prevailing evidence for stability, a few ACT researchers have also shown that EPA sentiments may change in periods of novel macrostructural conditions or after successful social movements (e.g., MacKinnon & Luke, 2002; Schneider & Schröder, 2012). The results of these studies implicitly support another model of cultural evolution: one where extended periods of social upheaval or substantial, persisting alteration of environmental conditions can induce systematic cultural change (Swidler, 1986). Such “unsettled times” may cause the beliefs of individuals within the affected culture to shift—especially beliefs about aspects of social life made salient by the environmental shock. Transformative events like the Covid-19 pandemic present a novel opportunity to understand whether cultural beliefs can shift when substantial social change impacts a national culture for an extended period of time.

Our study investigates whether EPA values of essential and non-essential occupations changed during the social upheaval surrounding the Covid-19 pandemic. We collected data on 650 occupational titles in the fall and winter of 2019–2020, before the pandemic struck the U.S., and used quota sampling to match the U.S. population on sex, age, and race/ethnicity, and education. Three months after the Covid-19 outbreak in the U.S., we collected data again on 41 occupations officially designated as essential by state governments and policy research centers, and 44 non-essential comparison occupations. We explore (1) whether mean EPA ratings for these occupational identities changed across waves, and (2) whether ratings for occupations designated as essential and non-essential changed differently.

Culture: Stable, but Changing

Culture is often seen as a set of features or interpretive lenses shared by members of a group or society, and it tends to change slowly over time (Goodenough, 1961; Romney, 1991; Romney et al., 1987). We consider language, religious beliefs, types of cuisine, or behavioral tendencies part of a group’s culture if they distinguish that group from some other groups and are shared by members of the group. More recent conceptualizations of “culture in action” conceive of culture as a toolkit of strategies used pragmatically in situational contexts (Swidler, 1986), or values that motivate—and scripts used to justify—patterns of consistent behavior (Vaisey, 2009). This diverse array of perspectives all emphasize the shared meanings associated with cultural forms and the relative stability of those meanings across situations through time.

Of course, cultures do change. Decades of research on social change using the General Social Survey (Marsden, 2012; Marsden et al., 2020), for example, documents changing cultural patterns using repeated measures from a probability sample of the U.S. population. When the survey began in the early 1970s, there was important variation in whether Americans thought that Black and White people should go to school together, whether women should work outside the home, and whether women were psychologically unfit to occupy political office. These items were eventually dropped from the survey because they had no variance. The culture had changed.

Population-wide cultural changes such as these are rarely considered the result of individuals changing their views. Ryder (1965) famously proposed that cohort replacement can serve as a key mechanism for society-level changes in attitudes, beliefs, and behavioral patterns. This perspective has found support in many empirical studies where researchers using longitudinal data show that features of culture can be so stable within individuals that societally held beliefs change primarily when older members die and younger cohorts with new socialization experiences replace them (Davis, 1992; Kiley & Vaisey, 2020; Schwadel, 2011).

Yet some theoretical work suggests that population-level cultural belief change may also occur because individuals in society change their beliefs in a collective way. Proposed mechanisms for individually-driven systematic cultural change tend to involve substantial shifts in a community’s environment or network structure (Centola et al., 2018; Lizardo & Strand, 2010; Swidler, 1986). While students of culture and cognition often debate whether actors process major environmental changes by either fundamentally retooling their beliefs or making sense of them in a more discursive way (Lizardo & Strand, 2010), they agree that substantial environmental change can cause poor fit between one’s existing cultural beliefs and the conditions of their new environment.

Hypotheses about transient or durable forms of cultural change are difficult to test. Environmental changes that may induce either form of cultural change must be substantial and punctuated enough to lead to noticeable friction between a set of salient cultural beliefs held by actors within a culture and the realities of their new environmental conditions, and last for long enough that these moments of friction cannot be explained away as anomalous. The Covid-19 pandemic represents a substantial, punctuated, and persistent environmental change. As such, the resulting social dislocation represents ideal circumstances to assess the plausibility of cultural change driven by members of a culture collectively resituating some of their beliefs.

One Lens on Culture: EPA

There are many ways to measure shared cultural meanings. We employ a well-known system developed by Osgood et al. (1957), which uses three discrete dimensions to measure the fundamental affective responses that members of a culture share about actors and behaviors: evaluation (good-bad), potency (powerful-powerless) and activity (lively-quiet) (EPA). Osgood (1962, 1964) established that these three dimensions could be used to characterize a large array of concepts (from inanimate objects to social positions) and were useful in a wide range of cultures. Heise (1979, 2007, 2010) used these dimensions to develop a geometric space within which cultural meanings for identities, behaviors, emotions, personal characteristics, and social settings could all be simultaneously located. The positions of concepts within EPA space reflect fundamental sentiments about those concepts in a given language culture, which people act to maintain through social interaction.

Heise (2010) documented that EPA sentiments are widely shared within national language cultures. Much of the within-culture variation in EPA ratings of individual concepts is the result of measurement error or unique individual experience (e.g., rating the concept of “mother” immediately after having a negative conversation with your mother on the phone might be different from our general cultural understanding of what mothers should be and are usually). Recently, Freeland and Hoey (2018) demonstrated that shared fundamental EPA sentiments about occupational identities were a good indicator of the stable, widely shared status structure of occupations within the United States. The scale that they created using EPA profiles of occupations and an ACT-derived understanding of occupational status performed better than more traditional occupational prestige measures in predicting traditional survey measures of job satisfaction, happiness, and feelings of being respected at work.

The ACT tradition views fundamental EPA sentiments as shared and stable reference signals that society maintains because people recreate their culture through interaction, but some researchers highlight that EPA values can change over time. MacKinnon and Luke (2002) provide an excellent example of EPA measurements as an indicator of social change. While they found very high stability of EPA means in Canadian culture in surveys at two points in time, they also identified significant change in views about homosexuals, religious identities, and some political identities between 1981 and 1995. These findings are noteworthy because the sentiment changes coincided with social movements related to the identities for which views changed over a relatively short period of time. Though the study analyzed independent samples from two points in time, the findings suggest that the EPA sentiment differences could represent changes not fully explained by cohort effects. Similarly, Schneider and Schröder (2012) found that the meaning of manager had changed in the U.S. and Germany at the same time as macro-level changes in business relations and leadership styles. These studies echo Swidler’s (1986) template for the different ways one might expect cultural meanings to evolve in “unsettled times.”

Social Upheaval and Punctuated Change

We use the tremendous changes that occurred in the occupational system in early 2020 to address the question: do EPA values change in a collective, consensual way during a natural disaster that brings material and structural changes to social life? Beginning in mid-March 2020, much non-essential economic activity in the U.S. was stopped or dramatically curtailed as a result of the spread of Covid-19. Unemployment rose sharply as businesses and other activities that required face-to-face interaction closed to prevent viral spread. Medical facilities were filled to capacity as hospitals and associated services attempted to deal with the infected population. State and local governments declared some functions to be essential and mandated their continued operation, while others were deemed non-essential and ordered to shut down for varying lengths of time. Supply chains and delivery services struggled to help people stay fed and supplied with needed goods. First responders and other government service providers tried to continue their essential services while protecting themselves from infection. During this time, there was routine talk of essential workers as “heroes” risking their health and safety to serve the broader population. People gathered on porches to applaud medical personnel when they changed shifts.

We examine these events as a potential case of punctuated change, asking several related questions: are the massive changes in work engagement and occupational activities surrounding the Covid-19 pandemic associated with changes in the affective meanings of work identities? Did the people who kept working during these difficult times gain esteem and power in the larger culture? If words like “essential” and “hero” are used to refer to people who work in normally lower status jobs like maintenance workers, home health care workers, or grocery clerks, does this punctuated change translate into higher EPA values for those work identities?

Conversely, do people occupying positions that are deemed non-essential and who are visibly thrown out of work lose stature on EPA? We know that persons who experience negative actions at the hands of others tend to lose evaluation. We tend to derogate victims, even when they suffer through no fault of their own (Heise, 2007). Similarly, just by being acted upon (as opposed to being the author of their own actions), people seem less powerful and less lively. Therefore, we expect a decline in the EPA values of occupational identities linked with work positions that suffered large displacement and unemployment.

We need to make two features of these predictions very clear. First, our expectation that the EPA values of essential occupations will go up, and that the EPA values of non-essential occupations will go down are drawn from our general understanding of structural changes in the economic sector during the Covid-19 crisis and our general application of ACT principles about how events affect EPA sentiments. We do not call these expectations “hypotheses” because they are not logically derived from the structure of the theory itself. We highlight our expectations so that the reader will be able to distinguish patterns in the data that we anticipated from those we did not (and are as a result interpreting in a post-hoc manner).

Second, we want to establish that we are studying a potential social change while it is occurring. The impact of the Covid-19 crisis in the occupational sector was assessed in June and July of 2020. While some businesses were re-opening during this period, few would argue that the economic system had regained a new, stable equilibrium. The U.S. economy had not returned to its earlier state or gained a “new normal.” It is far too early to know whether any changes that we observed could be considered permanent shifts in fundamental sentiments—like MacKinnon and Luke’s (2002) post-hoc interpretations of changes in the EPA ratings of homosexual, religious, and political identities—or whether they are the result of a large population experiencing a transient shift in their impressions simultaneously.

The difference between these two outcomes has important implications for the future. If differences exist and represent permanent cultural change, people will view certain types of workers in new ways and seek to maintain these new cultural meanings through their social interactions. However, if differences represent shared transient impressions as a result of collectively experienced events, then we will see social interactions that re-establish the old meanings as the economy re-opens and recreates its old structure. More time must pass to answer this deeper question, an issue which we revisit in the discussion. In the present study, we only ask whether collective belief changes among individuals in a culture accompany a punctuated and substantial shift in the environment. We make no assumption about whether these changes will be circumstantially constrained or long-lasting.

Data

Participants and Design

To answer our questions, we analyze the EPA meanings associated with 85 essential and non-essential occupational identities before and after the Covid-19 outbreak. Data for the study were collected through a pair of online surveys. Pre-pandemic data were collected between May of 2019 and March¹ of 2020, as part of an ACT sentiment study about occupational prestige and social deference. The pre-pandemic survey included EPA ratings for 650 occupations, including all 570 occupations listed in the 2010 US Census.

After the Covid-19 outbreak, we replicated the previous study for 85 occupational identities from the pre-pandemic survey (see Table 1 for a list of stimuli). To develop the stimulus list, we began by compiling a list of essential occupations identified in two sources: state COVID-19 executive orders² and the Center for Economic and Policy Research’s demographic study of workers in frontline industries (Rho et al., 2020). Because these sources often listed industries rather than specific occupations, we built a list of Census occupations within these industries, and flagged all occupational identities in the pre-pandemic survey that matched this list of essential workers.

Table 1.

Weighted Two-Sample T-Tests for Selected Occupations in the Analysis Sample.

		Evaluation			Potency			Activity
		Weighted		BFC	Weighted		BFC	Weighted		BFC
		Pre	Post		Pre	Post		Pre	Post
Non-Essential	Actor	2.81	0.72 **		1.37	1.65		1.58	1.47
	Bartender	3.62	1.40 ***	*	0.59	0.19 **		2.23	1.86 **
	Blackjack dealer	2.33	0.39*		0.82	0.70		1.08	1.01
	Cafeteria server	1.14	1.45		−0.38	−0.61		1.26	1.19
	Casino manager	3.61	0.20***	*	2.20	1.95		1.80	1.15**
	Cook	2.58	1.69*		0.97	0.78		2.00	1.62**
	Cosmetologist	1.29	1.47		0.48	0.17		0.20	0.23
	Designer	2.98	1.17**		1.10	1.04		0.41	0.27
	Door-to-Door salesman	−0.16	0.06		−0.40	−0.42		0.51	1.21*
	Dietitian	0.50	1.59		0.39	0.60		−0.38	0.02
	Event planner	2.61	1.67**		1.04	1.21		1.40	1.37
	Jewelry maker	1.97	1.36*		0.42	0.53		−0.44	−0.95*
	Librarian	0.78	2.20		0.22	−0.07		−1.44	−1.72*
	Musician	4.30	1.59***	**	1.39	0.84***	**	2.32	2.12
	Piano tuner	−0.86	1.58		0.49	0.02 *		−0.15	−0.33
	Real estate agent	1.11	1.08		1.07	1.13		0.81	0.70
	Sign painter	0.60	1.18		−0.06	−0.12		0.13	−0.20
	Telemarketer	−1.14	−1.10		−1.04	−0.94		1.34	1.32
	Ticket taker	0.88	1.13		−0.33	−0.38		0.09	−0.08
Essential	Bank manager	1.44	1.04**		2.07	1.87		0.33	0.01*
	Bulldozer mechanic	1.14	1.10		1.38	1.42		1.74	1.64
	Bus driver	1.72	1.47*		0.48	0.11**		0.76	0.42*
	Elder care aide	2.14	1.85		0.58	0.52		−0.21	−0.28
	Elementary school teacher	2.57	2.41		1.30	0.58***	***	1.46	1.20
	Farm owner and operator	2.37	1.98*** *		1.74	1.42*		2.06	1.84
	Fast food worker	0.84	1.05		−0.72	−0.85		1.67	1.80
	Forklift driver	0.88	1.06		1.13	0.63 *		1.73	1.42
	Grocery bagger	1.43	1.68		−0.75	−1.09		0.89	0.81
	Home health aide	1.84	1.86		0.44	0.44		0.25	0.36
	Licensed practical nurse	2.32	2.40		1.26	1.37		0.98	1.53 *
	Mailman	1.92	1.92		0.42	0.32		0.82	0.77
	Medical assistant	2.19	2.08		0.78	0.88		0.89	1.06
	Pharmacist	1.77	2.01		1.60	1.47		−0.02	0.16
	Physician	2.39	1.88 ***	**	2.47	2.17**		1.19	0.76**
	Registered nurse	2.65	2.44*		1.78	1.44*		1.91	1.50**
	School counselor	1.91	1.77		1.24	0.69**		0.12	−0.05
	Subway operator	1.20	1.17		1.34	0.84*		1.43	0.94*
	Truck driver	1.19	1.12		1.13	0.47**		1.36	1.11

N Occupations = 85 (40 shown).

***p < .001; **p < .01; *p < .05; . p < .1 (two-tailed tests). BFC columns show the number of significant mean differences after applying a Bonferroni correction to tests within each dimension.

Notes. See Table A in the online supplement for an expanded version of this table containing all 85 occupational identities. The N for pre-Covid means ranged from 99 to 1132 depending on the module of the survey; the N for post-Covid means was 198 for all occupations.

All five authors then used a blind voting system to select candidates from the 650 occupations in the pre-pandemic survey who were considered essential and non-essential, to ensure that occupations included in our study were both officially identified as essential and would likely be seen as essential by the general public. Occupations that received three or more votes as “essential” in addition to being flagged as essential occupations in the previous step became the 41 essential occupations in our analysis sample. Non-essential occupations included 44 identities for which none of the five authors voted in favor of classification as an “essential” occupational identity. The post-outbreak survey included ratings of these 85 occupations, and was fielded in June and July of 2020.

Acting as cultural informants, 2726 unique respondents rated the 85 occupations. To mitigate respondent fatigue, each participant was assigned to a module that included questions on a subset of identities (between 31 and 59 identities per module³). This study includes responses from 14 modules in total—12 from the pre-pandemic survey, and two from the post-outbreak survey. 31 occupations in the “core module” of our pre-pandemic survey were rated by 1131 respondents⁴, while all other pre-pandemic modules included 99 to 132 valid responses. Each of the two post-outbreak modules was rated by 191 to 192 participants, for a total of 383 unique respondents. In all cases, the number of raters per stimulus is above the number usually used in ACT sentiment studies to achieve reliabilities over .90 for estimated means (Heise, 2010).

Respondents to both surveys were U.S. citizens drawn from Qualtrics panels. A quota sampling method was used to approximately match the sample of respondents who completed each survey module to the population proportions reported by the 2010 Census in its categorical measures for age, race and ethnicity, level of education, and gender in the pre-pandemic survey, and for age, race and ethnicity, and gender in the post-outbreak survey. The marginal distributions from the Census for the four matched variables in the pre-pandemic survey were then used to construct sample weights via iterative proportional fitting for both surveys. Respondents who failed attention checks were not included in the analyses or Ns reported above.

Survey Protocol

We administered both surveys through the Qualtrics XM platform. Recruited respondents first answered demographic questions to determine eligibility based on quota sampling criteria. Those who qualified to proceed were shown an interactive demonstration of how to rate an identity stimulus using three sliding scales representing ACT’s EPA dimensions of affective meaning. They then provided ratings of each occupational identity in their assigned module using these three scales.

We randomized the order occupations appeared within each module, as well as the order of the EPA dimensions being rated for each occupation. Each rating screen showed an occupational identity as a stimulus situated above a continuous sliding scale.⁵ The left and right ends of the scale were labeled with anchor words representing an EPA dimension: the evaluation scale was anchored with “bad/awful” and “good/nice;” potency with “powerless/little” and “powerful/big;” activity with “slow/quiet/inactive” and “fast/noisy/active.” Scales were labeled at nine intervals with language indicating the magnitude of affective meaning. Each end of the slider was labeled “infinitely,” the middle position was labeled “neither/equally,” and the positions between were labeled “extremely,” “quite,” and “slightly.” The survey concluded with additional demographic questions and a few items about work and politics from the General Social Survey.

Measures

For each of our 85 occupations, we calculate two means and variances: one for the ratings before the Covid-19 outbreak, and one for the post-outbreak data. Ratings were recorded as continuous positions (to the 10th decimal place) along the slider bar from 4.0 to −4.0 (with 0 in the center). Each observation was assigned a sample weight based on marginal distributions of age, gender, race/ethnicity, and education level from the 2010 Census through iterative proportional fitting in R. Weights were used when computing all mean EPA scores presented in our analyses to both account for minimal differences in the samples⁶ and to capture national cultural sentiments to the best of our ability.

Analysis

To assess whether EPA values of essential and non-essential occupational identities changed after the outbreak of Covid-19 in the U.S. and the resulting social and economic dislocation, we first look for change within each of the three affective dimensions of interest. In our first analysis, we treat pre-outbreak and post-outbreak mean EPA ratings of occupations as matched pairs and perform two-sample unpaired t-tests. We use a Bonferroni correction to adjust for the large number (3 × 85) of tests conducted, and use the corrected result to determine whether to reject the universal null hypothesis—that the mean differences within each dimension did not change with the advent of the pandemic. Our second analysis is a global paired t-test, which allows us to summarize the net direction and magnitude of change in each dimension using the weighted means from the first set of tests.⁷ We also assess stability using the correlations of pre-outbreak and post-outbreak mean EPA ratings. Finally, we use a linear mixed modeling strategy to test whether mean EPA ratings for essential and non-essential occupations in our sample changed in a significantly different way across the two waves.

Results

Did EPA Values Change?

Table 1 summarizes the results of a randomly selected subset⁸ of 85 unpaired Welch-adjusted⁹ two-sample t-tests within each of the three EPA dimensions for all occupational identities in the study. Within each dimension, we use a Bonferroni correction to account for the higher family-wise error rate expected when conducting multiple t-tests. The presence of any statistically significant p-values in a given dimension that persist after the correction provides evidence in favor of rejecting the universal null hypothesis for the family of tests¹⁰—or that there are no significant differences in the pre-outbreak and post-outbreak means for an EPA dimension of interest (Lee & Lee, 2018).

A quick glance at Table 1 shows a clear pattern, but not the one that we expected. Almost all uncorrected significant differences in the families of pairwise tests for mean evaluation and potency are negative. Uncorrected mean differences in evaluation and potency that trend in the positive direction are both small and almost always non-significant. After the Bonferroni correction,¹¹ persisting significance in the evaluation and potency t-test families allow us to reject the universal null hypothesis within these dimensions, and provides strong evidence that evaluation and potency means changed—and changed negatively—with the advent of the pandemic. Both essential and non-essential occupations contribute to the trend of significant and negative change in the evaluation and potency t-test families.

While the analysis does not allow us to make post-hoc inferences about specific occupations, it is clear that some essential and non-essential occupations were viewed as significantly less good and less powerful after the Covid-19 outbreak than before it began. The family of unpaired t-tests capturing mean differences in the activity dimension, however, are comparatively stable. In contrast to evaluation and potency, none of the uncorrected significant p-values are significant at alpha level of .001, and some of the significant changes in mean activity ratings are in fact positive. No p-values in any of these tests exceed the Bonferroni-corrected critical value. As such, the family of tests does not provide evidence of significant change in the activity dimension.

How Large Were the Changes?

Table 2 shows the results of three paired t-tests that communicate the overall direction and size of estimated mean changes for each of the three EPA dimensions. In each test, we compare the pre-outbreak and post-outbreak estimated means for all occupations (N = 85),¹² and penalize the resulting p-values with a Bonferroni correction to account for the multiple tests.

Table 2.

Paired T-Test Comparing Weighted Pre-Outbreak and Post-Outbreak Occupation Means for EPA Dimensions.

Dimension	Pre-Covid Mean	Post-Covid Mean	Mean of the Differences	P
Evaluation	1.576	1.406	.170 ***	<.001
Potency	.833	.666	.167 ***	<.001
Activity	.908	.768	.140 ***	<.001

N = 85.

***p < .001; **p < .01; *p < .05; . p < .1 (two-tailed tests).

The tests in Table 2 show that mean differences before and during the pandemic are consistently negative in all three affective dimensions (p < .001). The largest mean changes with the lowest p-values appear in the evaluation and potency dimensions, with respective mean differences of .170 and .167. Our results show that activity means also tend to be significantly lower after the Covid-19 outbreak, with a mean difference of .140. While individual pairwise tests within the activity dimension from Table 1 do not provide evidence that mean activity differences for at least one of the occupations were conclusively significant, the paired test in Table 2 shows that mean activity score differences were clearly negative in aggregate for activity as well.

While there is evidence that mean EPA values for occupations in our study changed—especially evaluation and potency—it is important to note that the system of cultural meanings in the work institution remained largely stable during the socially tumultuous period of inquiry. The correlation between pre-outbreak and post-outbreak means in each dimension for the 85 work identities is exceptionally high.¹³ Correlations of pre- and post-outbreak values are above .91 in all three EPA dimensions. There is a clear, strong and positive association between pre- and post-outbreak mean ratings for both essential and non-essential identities. The correlation value for evaluation is slightly lower than that for potency simply because there are very few occupational identities that are not positively evaluated in U.S. culture, so variation is limited on that dimension.

In other words, although the Covid-19 crisis appears to have brought with it less positive, less powerful, and less active cultural sentiments about occupational identities, it did not upend society’s understanding of the status structure. Grocery baggers did not suddenly become equal to firefighters in evaluation, even though both were essential workers. Mail carriers did not suddenly become higher potency than professional athletes simply because they kept working during the pandemic while athletes were sidelined.

Did “Essential Occupation” Status Moderate EPA Value Changes?

Are changes in the mean score for each EPA dimension after the outbreak different for essential and non-essential workers? Our final question involves whether “essential worker” status moderates the relationship between mean EPA value and post-outbreak status.¹⁴ We speculated that change in EPA for essential occupations might be more positive compared to non-essential occupations, because of their central role in keeping the economy and medical system running. In retrospect, given the results in Tables 1 and 2, we might expect the fact that essential workers were forced to keep working under often dangerous conditions may have made them more likely to decline in EPA—especially in evaluation and potency.

To assess this possibility, we specify a set of linear mixed models to estimate post-outbreak mean EPA values (Table 3). This “mixed effects” approach allows us to capture differences in mean ratings for each EPA dimension from pre-outbreak to post-outbreak within essential and non-essential occupational groups (the “fixed” components), while accounting for variation in overall mean rating differences between occupations (the “random” component¹⁵).

Table 3.

Linear Mixed Models Predicting EPA Mean Occupation Scores.

	Model 1 (Evaluation)		Model 2 (Potency)		Model 3 (Activity)
Coefficient	Estimate	P	Estimate	p	Estimate	p
(Intercept)	1.502***	<.001	0.584***	<.001	0.725***	<.001
Post-Outbreak Status	−0.245***	<.001	−0.111 **	.008	−0.110 **	.008
Essential Status	0.154	.241	0.516**	.001	0.379*	.050
Post-Outbreak x Essential	0.156**	.004	−0.116 .	.051	−0.062	.287
N (85 × 2)	170		170		170
Marginal Pseudo R^2	.058		.071		.044
Conditional Pseudo R^2	.924		.957		.956

***p < .001; **p < .01; *p < .05; . p < .1.

Note. Model uses weighted means of the 85 occupational identities in the analysis sample.

Each occupation appears in the analysis dataset for the mixed model of a given EPA dimension twice, once for each point in time. The intercept of each model is a random parameter representing the value of the average pre-outbreak mean value of the relevant EPA dimension for non-essential occupations.

The binary indicator “Post-outbreak status” identifies whether an observation contains a pre-outbreak (0) or post-outbreak (1) mean. The coefficient associated with this variable in Table 3 is the estimated effect of the post-outbreak transition (i.e., of Covid-19) on the mean value of a given dimension rating when essential status is zero, or for non-essential occupations. Essential status is another binary fixed parameter that identifies whether an occupation was classified as essential (0 = no; 1 = yes). Coefficients for “Essential Status” in Table 3 represent the estimated effect of essential status on the difference in evaluation, potency, or activity scores when post-outbreak status is zero, or before the pandemic. Note that any significance in this coefficient would be driven by pre-pandemic differences in the EPA scores of the occupations included in our analysis sample (a feature of our study design), rather than inherent differences in how U.S. culture evaluates essential and non-essential occupations.

The final term in each model is an interaction between post-outbreak status and essential status. We use this term to estimate the difference in mean evaluation, potency, and activity differences for essential and non-essential occupations. Interpreting the associated coefficients allows us to assess whether the effect of transitioning to the pandemic on changes to mean evaluation, potency, or activity is moderated by essential worker status of an occupation—that is, whether changes in cultural sentiments were different for essential and non-essential occupations.

For non-essential occupations, Covid-19 had a substantial negative effect on mean evaluation (−.245, p < .001), and an effect in the same direction for both potency (−.111, p < .01) and activity (−.110, p < .01). Before the pandemic, the essential occupations in our analysis sample had slightly higher mean evaluation, potency, and activity ratings. The models indicate that the difference between essential and non-essential pre-Covid means are significant for potency (p < .01) and activity (p < .05), but not for evaluation. Again, it is important to note that these differences are a feature of the occupations in our analysis sample, which contains purposively selected occupational identities categorized as essential and non-essential. We preserve the coefficients for readers to understand this design aspect, but do not interpret them as real differences between the population of essential and non-essential workers from before the pandemic.

For essential occupations, Covid-19 had a different effect on some mean EPA values. Post-outbreak mean evaluations for essential occupations are significantly more positive (.156) than non-essential occupations (p < .01). In other words, mean evaluation scores for non-essential occupations dropped almost three times as much (−.245) as mean evaluation scores for essential occupations (−.089). Broadly speaking, it seems that essential occupations were largely protected from the decline in evaluation that accompanied U.S. culture’s transition to the pandemic. The same cannot be said for the moderating effect of essential status in the other two dimensions. The difference in pre-outbreak and post-outbreak mean potency values may be slightly more negative (−.116) for essential occupations, though this interaction only meets criteria for marginal significance (p = .051). The interaction term coefficient is also negative for activity (−.062) in Model 3, though its p-value does not meet our criteria for significance. Figure 1 conveys the interactions for evaluation and potency, using the equations derived from Model 1 and Model 2 to estimate predicted values for the difference in post-outbreak mean scores for both essential and non-essential occupations.

Figure 1.

Estimated change in mean evaluation and potency scores by essential status of occupations. Notes: Plots (a) and (b) show the estimated effect of the interaction between the essential status and post-outbreak status indicators for evaluation (plot a) in Model 1 (p < .01) and potency (plot b) in Model 2 (p < .1). Post-outbreak status is a binary variable, but we plot the different relationships between post-outbreak status and predicted evaluation and potency values as lines to illustrate the meaningful difference in slopes that the interactions represent. Confidence intervals reported for each of the means include values that fall within one standard error of the estimate. Y-axes differ to capture ranges of each scale in which the estimates fall, but both capture a full one-unit change in the −4/+4 scale for comparability. Differences in predicted values for pre-outbreak scores are driven by our process of selecting occupations for the study, and should not be interpreted as fundamental differences between essential and non-essential occupational identities.

Regardless of significance, accounting for differential EPA ratings of essential and non-essential occupations that may have developed in the transition to the pandemic explains a relatively small amount of variance. Even in the evaluation and potency dimensions, the sizeable main effect of post-outbreak status shows that essential and non-essential occupational identities changed in largely similar ways. Mean ratings at the occupation level for evaluation, potency, and even activity decreased during this period of economic and social upheaval.¹⁶

Conclusions

Our results show a clear pattern that, to some extent, we did not anticipate. At least some of the individual occupations we studied declined significantly in both evaluation and potency, and the average mean values in all three dimensions for occupations in our sample were significantly more negative. Some occupations declined more substantially on these dimensions than others, but both when correcting for multiple unpaired tests and when using paired comparisons on the population of means, we found no statistically significant EPA change in the positive direction. Essential status of occupations interacted with the overall negative post-outbreak shift in mean evaluation scores, with most of the decline in mean evaluation ratings explained by a decline in ratings for non-essential occupations after the Covid-19 outbreak. While the effect size is small, its presence is noteworthy—on average, essential occupation status may have protected this class of occupations from post-outbreak declines in evaluation.

In the context of ACT’s theoretical framework, the overall negative effect of the pandemic outbreak could be viewed as victim derogation. This interpretation is consistent with the notion that workplace dislocation has made occupational identities victims of the pandemic. Essential occupations have been especially salient as socially necessary and worthy of public praise during the pandemic, which may explain the lower decline in mean evaluation scores for these identities compared to non-essential occupations. While essential workers are clearly impacted by the pandemic, public discourse about their value and importance to our societal functions could attenuate victim derogation.

On the other hand, we also find that essential occupation status may come with a slightly larger decrease in potency compared to non-essential occupations. While this trend is both marginally significant (p = .051) and inconsistent with our pre-specified expectations, it fits with a cultural narrative that emerged during the pandemic conveying essential workers as individuals whose occupational identities force them to be “on the front lines” where they are frequently exposed to members of the public who may transmit the virus. Perhaps being forced to work during dangerous times makes one seem to be the object of coercive employment structures, not merely a hero, compared to jobs where more people were laid off or allowed to work from home.¹⁷

Discussion

Both ACT and theories of cultural change (Heise, 2010; Lizardo & Strand, 2010) broadly propose (but rarely test) that systematic changes in the beliefs of individuals can come from large, punctuated, and persistent structural changes to social environments, especially where cultural meanings and sentiments about certain stimuli from before the structural shift no longer map well to the same stimuli after the shift. We can confidently say that the Covid-19 pandemic represents a large, punctuated, and persistent structural change. In line with current theory, we suggest this shift may have caused members of U.S. culture to process structural changes to work events and job-related signals in ways that led them to view occupational identities differently during the pandemic.

ACT’s systematic approach to measuring cultural meaning assumes that EPA sentiments are widely shared within a culture, and that circumstances inducing deflection (here, emotional tension about one’s own occupational identity in the context of a set of work situations or events) or measurement error within individuals near the time of data collection (such as having an argument with a relative or getting a promotion at work) will cancel each other out. This allows the mean value to represent the cultural standard that all of the cultural informants have more or less adopted toward a stimulus (Heise, 2010). However, an extended environmental shock representing “unsettled times” calls these assumptions into question. Very few people in the U.S. have been unaffected by the Covid-19 pandemic and its ramifications for work and the economy. Even those who are out of the labor force have interacted with essential workers still on the job or heard about essential workers’ contributions and general unemployment trends in the media. By viewing a large number of work-related events shaped by the Covid-19 pandemic, the impressions our respondents have for occupational identities could have been deflected in a consistently negative direction. Whether the cultural changes observed in our research reflect transient sentiments or new fundamental sentiments—and whether the measurable shift in cultural beliefs will disappear or persist after the pandemic—remains to be seen. The answer to this question has important implications for ACT scholars and other researchers studying culture and social behavior.

On one hand, the changes we observe could be temporary. If EPA values for the occupations in our sample are restored to their pre-outbreak means after the pandemic, this would suggest that fundamental sentiments and cultural meanings are quite stable, even in the presence of a major long-term shock to our social environment. This outcome fits within the existing architecture of ACT’s theoretical model, which posits that persistent deflection induced by differences between transient and fundamental impressions of occupations during the pandemic are resolved by people interacting with or labeling these occupations differently until the pandemic ends (i.e., making sense of occupations discursively). In ACT, signs praising medical workers and truck drivers as heroes, people clapping from balconies as shifts of workers change, and customers saying “thank you for being here” to grocery baggers and delivery people are attempts to resolve the cultural meaning dislocation that has occurred.

On the other hand, Covid-19 could result in a punctuated move to a new equilibrium for cultural sentiments about occupations. If so, we would expect to see this change appear in more negative post-pandemic EPA values for occupations in general, or an asymmetric rebound in evaluation and potency for essential and non-essential occupations, compared to pre-outbreak EPA values. These patterns would suggest that cultural meanings can be shaped durably by “unsettled times” and are malleable in the face of large-scale structural change. In the context of ACT, this outcome would provide evidence that fundamental sentiments can shift (i.e., people can retool their beliefs) when persistent transient sentiments about a cultural object lead people to experience an extended period of deflection in a systematic direction. The new post-pandemic EPA values would represent new cultural beliefs that persist into “settled times,” and operate as a new reference signal in the control system that ACT describes.

New fundamental sentiments would affect social interactions, shaping behaviors of workers toward each other and of workers with customers, clients, and other interaction partners. It might also shape the structure of individuals’ identities: if one’s occupational identity becoming more negative shifts it away from their understanding of “myself as I really am,” pressures might lead that individual to deemphasize their work activities in their self-concept compared to other more self-actualizing identities (MacKinnon & Heise, 2010). If EPA ratings of occupations become durably more negative, going to work might generate affective deflections (here, emotional tension about one’s own occupational identity in the context of a set of work situations or events) that might need to be repaired through the pursuit of more positive activities in other institutions to maintain a stable sense of self. Persistence in comparatively more positive evaluations of essential occupations may lead to different behaviors. Since ACT predicts that the typical emotions for an identity occupant will be roughly similar to the EPA meaning of the identity in normal, identity-maintaining interactions (Heise, 2007; MacKinnon, 1994), new and lower evaluation and potency values would indicate that many workers would feel less pleasant, efficacious emotions while working after the outbreak of Covid-19.

Our findings show that EPA meanings for occupations have changed during the pandemic: they have declined in evaluation (primarily for non-essential occupations), potency (perhaps more so for essential occupations), and to some extent activity. However, our current data are not equipped to assess whether these changes are the result of a transient but systematic deflection away from stable fundamental sentiments in U.S. culture, or reflect a more durable cultural shift to a new equilibrium. Researchers will need to wait until the economic effects of the pandemic on the economy have stabilized and re-measure the EPA of these occupational identities to answer this question.

Author Biographies

Joseph M. Quinn is a PhD Candidate in the Department of Sociology at Duke University. He investigates how cognitive processes and features of social environments combine to shape emotions, behaviors, and patterns of inequality between groups.

Robert E. Freeland is a Visiting Research Fellow at the Duke Network Analysis Center. His research centers on integrating subjective elements of the social world including identity, status, and cultural meaning with structural inequality by race, class, and gender.

Jesse Hoey is a Professor of Computer Science at the University of Waterloo. He studies emotional intelligence by attempting to build computational models of some of its core functions, and applying these models to domains with social and economic impact.

Kimberly B. Rogers is an Associate Professor of Sociology at Dartmouth College. Her research examines how inequalities are produced, maintained, and resisted through behavior and emotion dynamics in social interactions.

Lynn Smith-Lovin is Robert L. Wilson Professor of Arts and Sciences at Duke University. She studies identity and emotion, and has received lifetime achievement awards from four ASA sections.