Signal Fires: The Use of Diversity Value Signaling in Local Government Job Advertisements Following the Summer of Racial Reckoning

Following the events of the summer of 2020, public and private sector organizations have engaged in a variety of actions related to promoting, valuing, and celebrating diversity, equity, and inclusion (DEI). Perhaps due to their role as direct providers of public safety, local governments are along the front lines of DEI efforts and values. Using a novel dataset of International City/County Management Association job advertisements from July 1, 2020–November 1

F or 8 minutes and 46 seconds, a police officer kneeled on the back of a citizen's neck. Onlookers pleaded for the officer to stop. The man underneath the officer's knee (as well as three other officers), George Floyd, did not survive the encounter. The video taken by one of the onlookers would soon go viral, sparking protests and unrest around the globe and in cities across the United States, especially in the City of Minneapolis, Minnesota where this event took place. The protests, unrest, and calls for justice and accountability would endure through the following months despite the ongoing presence of a global pandemic. In short, the summer of 2020 was marred by disease, fire, and social unrest. For some, the flames would indicate the destruction of property and rage, but for others, the flames signaled a cry for help, especially from marginalized communities across the globe.
The summer of 2020 would come to be referred to as the Summer of Racial Reckoning (Chang, Martin, and Marrapodi 2020), and, in the wake of these events, many public sector organizations were compelled to as-sess (or reassess) their roles in contributing to the continuation of historical and systemic inequities. Among these organizations, local government administrations stand out as particularly important. Local governments make hiring decisions, identify employees eligible for promotion, and sanction employees when necessary. They are also direct service providers to the public in a multitude of ways, ranging from providing public safety to building public parks and spaces. Moreover, through their police powers, local law enforcement agencies may simultaneously deprive an individual of his or her liberty while protecting another's right to protest (Fisk et al. 2020).
Local governments, when acting as an employer, have a variety of options relative to the language used in their recruitment materials. For some organizations, diversity, equity, and inclusion (DEI) is on the minds of decision-makers with such language reflected in the advertisement itself. In this way, a commitment to DEI is signaled alongside traditional information such as required skills, educational requirements, salary, and Signal Fires | 87 benefits, as well as expected years of experience. For others, however, the emphasis of DEI within job advertisements is more muted with employers including a notation that they are an equal opportunity employer. Finally, cities and counties may also prioritize other organizational values and avoid verbiage relative to DEI or equal opportunity.
Decisions about the language used in job advertisements are paramount for both the employer and the potential employee. For job seekers, the job advertisement fills an information gap and enables the applicant to gain information relative to his or her perceived fit within the organization (Andreassen 2021). From an organizational perspective, Schmidt, Chapman, and Jones (2015) explain that job advertisements are a vehicle for organizations to present information about their ideal candidate's knowledge, values, skills, and abilities. If they fail to do so, the employer risks losing out on applicants who self-select out of the recruitment process, thus reducing the quantity and potential quality of the applicant pool. Additionally, job advertisements, when they signal something about values or skills, are more likely to make that job attractive to someone who shares the desired values or skills (Ashraf et al. 2016;Gaucher, Friesen, and Kay 2011). Yet, even before applicants can decide on their perceived fit, they must first receive, process, and evaluate information in a crowded labor market, meaning the choice of language used in the advertisement can matter a great deal (Lazear and Oyer 2012, 494). This signaling process and applicant assessment, according to Keppeler and Papenfuß (2021), is not fully understood by students of public administration.
Thus, the purpose of this study is to determine the degree to which local government contexts explain variation in the use of diversity value signaling language. To do so, we examine the language utilized in job postings for local government administrators posted via the International City and County Management Association (ICMA) job center web portal from July 2020-November 2020. Each job advertisement (n = 113) is examined for language germane to DEI and analyzed via logistic regression to determine the degree to which community characteristics and government structures are associated with diversity value signaling. This time frame offers an important window into how organizations respond to changes in their environment and the degree to which social movements serve as impetus for change.

Theoretical Background
Public administration scholarship has long sought to understand how and why public sector organizations make the decisions they do, especially as it relates to diversity, equity, and inclusion. In general, researchers have found that organizations make decisions to mitigate the effects of uncertainty in their environment and to take advantage of environmental conditions (Aldrich and Mindlin 1978;Downey, Hellriegel, and Slocum 1975;Galbraith 1973). Scholarship also notes that for organizations to survive, they adapt and be nimble enough to effectively respond to an evolving environment (Aldrich and Mindlin 1978;Boyne et al. 2011;Galbraith 1973). However, organizations confront and understand their environments differently, and, as such, adopt and implement differing strategies to respond to their environment. Perceptions of the environment, we note, may also guide the behavior and decision-making of an organization (Aldrich and Mindlin 1978;Galbraith 1973). In fact, in some instances, perceptions can be more valuable than the objective data that surrounds the organization (Duncan 1972;Pfeffer and Salancik 1978).
A subset of this scholarship has focused on organizational decision-making and outcomes relative to diversity, equity, and inclusion (Pitts 2009;Cole, Haun, and Silvera 2022). One vein of DEI scholarship elucidates the differences and relationships between diversity management (and other proactive and voluntary management efforts) from more legalistic approaches. Ng and Burke (2005) identified three key differences. First, diversity management involves the recruitment and subsequent selection of qualified employees from disadvantaged groups, especially those individuals who would have been "missed" through more traditional recruitment methods. Second, diversity management goes beyond quantitative targets and toward celebrating/capitalizing on employees' cultural and experiential differences. Third, intentional and institutionalized efforts are needed to realize the benefits of diversity management and may include new policies, programs, and organizational support (see also Ely and Thomas 2001;Pless and Maak 2004). In a similar effort, Fisk and colleagues (2019) categorized diversity, equity, and inclusion efforts as fitting within a 2 × 2 typology based on the degree of institutionalization and the degree to which diversity management principles were practiced within an organization.
Scholarship has also examined the outputs and mech-anisms of DEI programming. While noting the potential for increased intergroup conflict and bias (Ashikali and Groeneveld 2015), recent scholarship has observed positive relationships between diversity management efforts and increases in the number of minority employees (Groeneveld and Verbeek 2012), job satisfaction (Pitts 2009), turnover (Groeneveld 2011), employee loyalty and commitment (Groeneveld 2011), and employee performance and outputs (Choi and Rainey 2010). Ashikali and Groeneveld (2015, 149) add, "in practice, diversity management should aim to foster the positive cognitive effects of diversity and to mitigate the negative affective effects that arise from intergroup biases. . . . Diversity management should therefore include policies, programs, and management activities that address the diversity in the organizational workforce to bring about its potentially positive effects." Noting the positive benefits of DEI, Pitts, Hicklin, and Hawes (2010), unpacked the factors that distinguish more effective DEI efforts from those that are less effective. Their results pointed to environmental uncertainty, that is, efforts to promote stability and mitigate ambiguity as well as environmental favorability, including ample organizational resources, time, and a supportive intergroup culture as key explanatory factors that shape how well organizations implement their DEI programming. According to Pitts et al. (2010), organizations with greater resources and expertise are more likely to implement DEI programs effectively and, as such, are more likely to realize performance dividends. Similarly, organizations operating in an uncertain environment are also more likely than others to adopt and effectively implement diversity management programs as a way of promoting stability. Efforts at unpacking the factors that shape the recruitment and selection of qualified candidates from historically disadvantaged groups have also taken on increased importance. Drawing from Chapman et al. (2005), researchers have observed the importance of an applicant's perceived fit within the organization and suggest that it is among the strongest predictors of an applicant's decision to apply (Cable and Judge 1996). Building on this foundation, Linos (2018) utilized a field experiment to understand the utility of various job advertisement messages and the impact of such messages on the recruitment of new police officers. Results demonstrated that those job advertisements with messaging that addressed personal benefits were "three times as effective as the control and are particularly effective for applicants of color and women. Importantly, public service messages alone do not seem to attract more people to the police than the control group" (Linos 2018, 68-70).

Environmental Factors
Local government decision-making scholarship has also begun to unpack why some local governments are moving in the direction of adopting DEI polices/values while others are not.
Research has observed a variety of need-related factors. Kerr and Mladenka (1994), for example, found a direct connection between increasing percentages of minority residents and the willingness of those communities to support minority employment programs. Hur and Strickland (2012) also observed relationships between DEI efforts and city-level characteristics via a sample of local governments in North Carolina. They found a larger population, a higher percentage of African American residents, and increases in affluence/wealth were associated with increased likelihood of DEI policy adoption. They also noted that governmental structure (i.e., local form of government) is influential in a locality's willingness to promulgate DEI policies. Similarly, based on a sample of local governments in Oregon, Nishishiba (2012) observed an association between the adoption of diversity management programs and changing demographics. Cooper and Gerlach (2019) found that as communities become more diverse, they are more likely to support the creation of a chief diversity officer position within city hall.
Finally, Fisk et al. (2020) reported a strong relationship between the percentage of foreign-born residents and the adoption of three measures aimed at inclusion (i.e., offering cultural competency training, translation services, and a preference for multlingual job applicants).
In short, organizations tend to be more responsive to their external environment, especially when it is an environment that is favorably inclined to DEI programming (Pitts, Hicklin, and Hawes 2010). Based on previous work, we offer the following hypothesis: H1: Local governments with more non-white populations are more likely to signal diversity values in job advertisements. H 1a: Local governments with higher economic inequality are more likely to signal diversity values in job advertisements.
The literature also suggests that organizations exist in a dynamic environment and that they are responsive to that environment (Hur and Strickland 2012;Kerr and Mladenka 1994). Therefore, we hypothesize that local governments are responsive to their environment. We also anticipate that communities whose populations are becoming diverse are more likely to signal diversity values in job advertisements while those communities that are losing minority populations will not. The aspects of the community that are examined in this study to represent environmental change is the change in the percentage of the non-white population by directions (i.e., increase or decrease).
H2: Local governments in changing environments are responsive to their environment.
H 2a: Local governments that are losing nonwhite populations will be less likely to include diversity values in job advertisements.
H 2b: Local governments that are gaining nonwhite populations will be more likely to include diversity values in job advertisements.

Institutional Structure and Capacity
Previous research also points to the important role of institutional structure and capacity in terms of engendering more inclusive policies and procedures (Huang and Liu 2018). Researchers have demonstrated that the institutional structure of local government matters and shapes commitment to inclusionary practices (Fisk et al. 2020;Huang and Liu 2018;Hur and Strickland 2012). The presence of strong elected executives have contributed to the adoption of climate policies (Krause 2011), immigration (Huang and Liu 2018), and economic development incentives. Because our data involves primarily city and county managers, we turn to the institutional differences in city councils. Local government council members are generally elected as either an at-large or districted representative. Atlarge members are based on the number of votes cast for each member and typically, a specific number of candidates are selected to serve on the council regardless of which parts of the city or county they live in. It could be the case, for example, that the three winners of an election could all live within one neighborhood. District-based elections, on the other hand, elect council members based on geographic districts, and members are elected for each district. Geographic boundaries for districts often come along with demographic characteristics (for example, race and income) (Forest 2018). Therefore, the local government's job post may reflect more diversity values when the city/county's so-ciodemographic characteristics are more districted. Based on the literature, we hypothesize that as the percentage of districted councilmembers increases, the greater the likelihood of including diversity value signaling.
H3: Local governments with more councilmembers elected by geographic district are more likely to signal diversity values in job advertisements.
For capacity, many researchers have adopted population as a measure of institutional capacity and found a relationship between the capacity and the adoptions of inclusive practices and policy innovations (Fisk et al. 2020;Huang and Liu 2018;Krause 2011;Sullivan 2002). Sullivan (2002), for example, utilized the natural log of population and found that local governments with larger population size tended to engage in more policy innovation. Huang and Liu (2018) also found that localities with a larger population were more likely to adopt pro-immigrant policies. Similar results were reported by Fisk et al. (2020). Here, local governments with larger populations were associated with an increased likelihood of adopting inclusionary policies. Given these findings, we utilize the population size as a proxy for local government capacity and hypothesize that local governments which are home to a larger number of citizens will be more likely to signal diversity values in job advertisements.
H4: Local governments with a higher population size are more likely to signal diversity values in job advertisements.

Descriptive Statistics
The statistical analysis is based on data acquired from the ICMA job center website (https://icma.org/job-center) during the observation period of July 1, 2020 through November 1, 2020. A keyword search was administered within the website to locate job advertisements for major leadership positions only (i.e., City Administrator and Assistant/Deputy Chief Administrator). This created a total sample of 113 job advertisements (n = 113). A keyword search was then conducted within each job advertisement. In our analysis, keywords included: diverse environment, diversity, equity, inclusion and/or their abbreviations (Ng and Burke 2005). We did not include EEO as a measure of diversity signaling.
The dependent variable was dichotomized (1 = includes a keyword, 0 = does not include a keyword). As outlined in Table 1, 25 of 113 job posts included at least one DEI signaling keyword (22% of all job ads). For the 25 advertisements that signaled DEI, the frequency of signaling varied as well. The majority (16) included a single mention of DEI value signaling (64%), but seven ads mentioned DEI twice (28%), and two signaled more than 3 times (8%). Among those 113 observations, 86 were at the city level and three at the county level. The rest of the observations consisted of 21 townships and three census-designated places (CDP). Of those that used signaling, 22 are city governments, 2 are townships, and 1 is a CDP. We did not observe any diversity keywords in the county job advertisements.
Most of the job advertisements included only one DEI keyword and it was located near the EEO statement or at the beginning of the introduction of the jurisdiction. The city of Brawley City, CA, for example, starts the text by introducing the city "with a diverse and growing population of over 26,000." The rest of the content, however, does not talk about any DEI related concern. The job advertisement posted by City of Fort Collins, CO is the only one that includes four diversity signaling words. Starting with the summary of the job opportunity, the ideal candidate should "embrace our commitment to equity and inclusion" and "bring their own set of diverse experiences that add value and perspective." It then states the city government's commitment to high-quality public service and desire to "attract and employ diverse and competitive talent." It ends with a formal statement of EEO and "values diversity at all levels of its workforce." By comparing contents, we find that even though there are just three more DEI keywords identified in the Fort Collins job posts compared to the Brawley City posts, the gap of DEI information conveyed between the two localities is huge. For most of the advertisements with only one keyword, they seem to just comply with the public value and include the statement of DEI as a formality. For those with more keywords, the governments seem to put efforts on not only indicating their commitment to DEI but also on recruiting candidates who commit to the same values. For this reason, we conduct statistical analysis on not only whether local governments signal DEI values in their job advertisements, but also the magnitude of the signal.
Key independent variables for each of the 113 communities relative to context were collected for 2018 demographic characteristic (NonWhite %), an economic inequality indicator (Gini), and demographic change variables between 2010 and 2018 1 (NonWhite change, Absolute NonWhite Change). These demographic and economic data were derived from the American Community Survey (ACS). The demographic variable (Non-White %) ranges from 0 to 80.339%, with an average of 19.839% across the observations. Gini index is a measure of income equality and ranges from 0 to 1, where a "0" suggests absolute equal distribution of income and a "1" suggests absolute inequality. The Gini index in this sample group ranges from .061 to .558, with a mean of .427. The change variables capture the non-white population changes between 2010 and 2018. NonWhite change measures the population change with directions (i.e., increase or decrease). This measure captures whether the community is losing or growing minority population and ranges from -23.707% to +19.227% in this sample. The Absolute NonWhite Change calculates the absolute values of the non-white change variable, measuring the magnitude

Signal Fires | 91
of the non-white population change without counting for the direction. The internal context variables include the size of the council (Council Size), as well as the percent of the council membership who are districted as opposed to being members elected at large (District %). This council information was captured from the jurisdictions' websites in 2020. The smallest council size includes 5 committee members while the largest contains 15. Some local government's council members are elected totally at large (0%) while others' council members are elected totally by districts. On average, 35.69% of the council members are elected by districts across the 113 communities. Control variables include the natural log of population (Log Pop) in 2018, which was also collected from the ACS. Descriptive statistics of all variables are available in Table 2.

Analytic Strategy
To analyze the data, maximum likelihood estimation (MLE) is utilized to examine the relationship between diversity signaling language and the covariates in our primary analysis. We utilized logistic regression as our dependent variable is dichotomous, that is, whether job advertisements include diversity signaling language ("1" if included and "0" else). The first model is a static model where independent indicators are static (i.e., no percentage or amount changes of the variables). The second model substitutes the NonWhite % variable with Absolute NonWhite Change in order to measure whether the change in non-white population has an impact on diversity signaling language, regardless of the direction of changes. The last two models include the same variables utilized in the static model but consider the direc-tion of changes. For each significant predictor, examples of predicted probabilities are presented as graphs. For the robustness model, Poisson regressions are conducted to estimate the variation of diversity signaling word count with the same independent variables and the same conditions as in the logistic regression. In the Poisson regression, the dependent variable is a count variable measuring the number of diversity keywords.

Results
Results from the four logistic regression models are presented in Table 3. Models 1 and 2 include the static percentage of non-white population and the absolute change of non-white, respectively. Models 3 and 4 include the non-white change variable, which is non-negative (Non-White change ≥ 0) and negative (NonWhite change < 0), respectively. In general, all models present strong statistical power in the estimation (LR X 2 , p ≤ .05). Variables approximating population size, non-white population and percentage of district produce statistical significance, whereas variables representing economic inequality, absolute non-white change, and council size present no relationship with diversity signaling.
The population size (Log Pop) is positively associated with diversity signaling across all but the last model: Model 1 (.541, p ≤ .05), Model 2 (.729, p ≤ .001), Model 3 (.620, p ≤ .05). In Model 1 for example, a 1% increase in population results in it being 1.718 (exp(.541)) times more likely for local governments to include diversity signaling language in their job advertisements. Figure 1 shows the predicted probabilities of including diversity signaling word as the population changes based on Model 1. We can see from Figure 1 that across the scale of population, the predicted probability of signaling diversity in job posts increases from .013 to .72 when the natural log of population changes from 4 to 15, holding all else equal. The non-white population (NonWhite %) is also positively associated with diversity signaling in Model 1 (.034, p ≤ .05). In effect, this means that local governments are 1.035 (exp(.034)) times more likely to include diversity signaling language in their job posts for every 1% increase in non-white population. Figure  2 shows the predicted probabilities of including diversity signaling languages over the different percentages of non-white population. The probability of a local government signaling diversity in job advertisements increases from .126 to .558 when localities have no non-white citizens as compared to those with an 80% non-white population.
The percentage of district (District %) has a negative relationship with diversity signaling across the static model (-.023, p ≤ .01), the absolute change model (-.020, p ≤ .01), and the non-negative and negative change models (-.019, p ≤ .05 and -.039, p ≤ .05). In other words, local governments are 2% (exp(-.023)) less likely to include diversity signaling language in their job posts when there is a one unit increase in the percent of council districts in the absolute change model. Figure 3 shows the predicted probabilities of diversity signaling keywords based on Model 1. There is .342 probability for local governments whose council members are elected at-large (present as "0" in the District % axis) to signal diversity in their job posts. For governments whose council members are totally elected by district (present as "100" in the District % axis), the predicted probability decreases to .091. Signal Fires | 93

Robustness Model
An additional analysis was executed to determine the degree to which the results were consistent in instances in which diversity values were strongly signaled. Of these, the majority of job advertisements mentioned diversity signaling language once or not at all. But in several instances, diversity value signaling language is repeated throughout the job advertisement. To account for this, a count variable that measures the word count of diversity keywords as a dependent variable was created. There were two options for estimating models with a count dependent variable: Poisson Regression and Negative Binomial Regression (NBR). When the count dependent variable was over dispersed (i.e., its mean does not equal to variance), NBR would be utilized for the estimation because it provided less restriction on the assumption that the mean of the variable must be equal to its variance in Poisson regression (Long 1997, 230-238). To determine the appropriateness of the regression model, we conducted a Poisson goodnessof-fit test and found no overdispersion in the distribution of the dependent variable. Therefore, Poisson regression was appropriate, and several Poisson regressions were conducted using the same variables in the main analyses. Results in Table 4 showed the same directions of re-lationship as the logistic regression models. In the static model (Model 1 in Table 4) for instance, natural log of population (.412, p ≤ .001) and the percent of non-white population (.017, p ≤ .05) are positively associated with the number of diversity keywords, while the percent district (-.009, p ≤ .05) are negatively associated with the word count. However, the percent district variable is not statistically significant in Models 2 and 3 in the Poisson regression, and is only significant at the 95% confidence intervals (99% in the logistic regression) in Model 1. In addition, the percentage of non-white population becomes statistically significant in the Negative Non-White Change model (Model 4) in Poisson regression and is positively associated with the diversity word count (.034, p ≤ .05). This new significance indicates that those local governments with a higher percentage of non-white residents are more likely to signal diversity values in the governmental job posts when they are facing the problem of losing minority population. The percentage of district is no longer significant in Models 2 and 3.

Discussion
The results of this study paint a complicated picture of DEI signaling and offer mixed support for our hypoth-

Figure 3. Effect of District % on Diversity Signaling Language
Signal Fires | 95 eses. Results showed partial support for Hypothesis 1, which linked the use of DEI signaling language to diverse populations and economic inequality. In general, as the percentage of the non-white population increased, so did the likelihood that a job advertisement would include DEI signaling language. We also observed support for Hypothesis 4, suggesting a relationship between population size and diversity value signaling. However, we saw limited support for Hypothesis 2, which tested for the role of changing environments on diversity value signaling, and our data did not show support for our third hypothesis. To test the robustness of our findings, we also assessed whether the frequency of diversity value signaling language is associated with % non-white and population size. Similar to Pitts, Hicklin and Hawes (2010), the use of diversity signaling language appears largely related to organizational capacity and environmental favorability, that is, those local governments located in environments that are inclined to support DEI efforts. While data cannot pinpoint precise reasons, there are still several possibilities that help explain our findings. We begin with unpacking our population variable. One possibility is that population size is driving uncertainty as opposed to the uncertainty being driven by growth of specific demographic groups. Similarly, it may be that as the city's population has grown, the local government is responding to changing preferences, values, and needs of citizens, which are reflected in the job advertisement. Finally, population size may be unrelated to needs or values, but rather may be a function of capacity and resources. In other words, larger local governments have more specialized staff and in-house expertise, which again may be reflected in the choice of language within the advertisements (Fisk et al. 2020).
We also note the relationship between the percentage of non-white residents and diversity signaling. One possible explanation is that the local government is responding to the needs of its diverse citizens via a job advertisement. Conversely, this relationship may not be in response to a need, but rather a preference, that is, communities with a greater percentage of non-white residents are seeking out an administrative leader who shares the community's values. Another alternative is that hiring authorities may decide to signal diversity in order to intentionally seek out individuals who share values related to DEI without regard to the external environment but driven instead by their internal needs or normative beliefs. In sum, the results provide some indications that local governments are responding to environmental favorability and uncertainty.

Conclusion
The results of this examination indicate that both external and internal characteristics are associated with diversity value signaling in local government administrations. An additional test of robustness confirms these findings, as they find that stronger diversity value signaling in the job advertisement is related to the same characteristics. The present study is not without limitations and sets the stage for a long line of future research. Data used for the study were collected July 2020 through November of 2020 and examined the immediate response of local government administrations to the events of the summer of 2020. The specificity of the dataset captures the short-term response but that specificity limits our study's generalizability to the present status of DEI value signaling. Future research should examine these questions over a longer duration to test the durability of these changes and should also consider pre/post analysis as to whether such advertisements contribute to a more diverse applicant pool.