Introduction
Drug testing for welfare policies has been proposed in nearly every state, a trend that began following the 2008 economic recession (Ledford Reference Ledford2018; NCSL 2017).Footnote 1 In 2009 alone, over 20 states considered proposals to make drug testing a requirement of assistance (NCSL 2017). Two states passed drug testing legislation in 2009, one in 2010, two in 2011, three in 2012, one in 2013, three in 2014, one in 2015, and one in 2016, for a total of 15 states with drug testing for welfare laws.Footnote 2 Moreover, Figure 1 shows proposal rates across US states from 2009 to 2018. All but four states – Delaware, Idaho, Virginia, and Wyoming – proposed at least one drug testing for welfare bill. The states with the most proposals in the sample were New York, with proposals in eight years in the sample, and Illinois, with a proposal every year in the sample.
Figure 1. State proposal rates 2009–2018.
Existing research shows a significant causal relationship between state racial minority populations, state proportions of racial minority welfare recipients, state levels of racial resentment, and the adoption of punitive welfare policies (Fellowes and Rowe Reference Fellowes and Rowe2004; Ledford Reference Ledford2018; Soss et al. Reference Soss, Schram, Vartanian and Brien2001; Volden Reference Volden2006). This article contributes to the extant literature by evaluating state-level racial factors and the diffusion of recent drug testing for welfare policies across US states. The author argues that these proposals represent racially motivated efforts to restrict public assistance eligibility. In that way, drug testing laws are modern iterations of the government’s legacy of neglecting and controlling Black Americans. Moreover, drug testing laws perpetuate the erroneous stereotype that most welfare recipients are deviant and undeserving of assistance due to immoral behavior and that people who test positive for drugs either do not deserve governmental assistance or must meet other conditions for aid.
In this case, the assumption is that a significant number of recipients of public assistance are abusing drugs despite a 1996 National Institute of Alcohol Abuse and Alcoholism study that found no statistically significant differences in levels of illegal drug use between welfare recipients and non-welfare recipients (Grant and Dawnson Reference Grant and Dawson1996). Evidence from states with drug testing laws reinforces the NIAAA’s findings. Results from seven states which require drug tests for welfare applicants (Tennessee, Montana, Texas, Maine, Michigan, Mississippi, and West Virginia) found that less than 1% of applicants tested positive for drugs. In contrast, according to the National Survey of Drug Use and Health conducted by the National Institute on Drug Abuse, about 10% of adult Americans are estimated to regularly use illegal drugs (The Center for Law and Social Policy 2015; National Institute on Drug Abuse 2019). A prime example of the mismatch between perceptions and reality is Oklahoma, where in 2016, a little over 17,000 people applied for Temporary Assistance for Needy Families (TANF) benefits and 3,856 of which were given a Substance Abuse Subtle Screening Inventory test (SASSI) in order to determine if there was a “reasonable suspicion” of drug abuse (Covert and Israel Reference Covert and Israel2017).Footnote 3 About 1 in 3 of those screened were ordered to undergo a drug test for benefits due to the results of the SASSI. However, less than 9%, or 101 applicants, tested positive (Covert and Israel Reference Covert and Israel2017). While the cost of Oklahoma’s 2016 drug testing program was $668,818.48, state officials decided to keep the program (Covert and Israel Reference Covert and Israel2017).
This article examines the factors that influence the diffusion of drug testing for welfare policies across states. More specifically, state-level racial variables’ effect on the likelihood of the proposal of drug testing legislation is examined. While numerous studies have examined the diffusion of welfare policies, including some focusing on racial factors, only one study has evaluated drug testing proposals. However, this study only examines proposals from 2008 to 2014 and omitted variables accounting for the influence of drug-related variables on the diffusion of these policies.
Literature review
Various scholars have studied the diffusion of policies relating to cash assistance. Critical in many of these analyses are variables related to race, most often the percentage of TANF recipients who are racial minorities in a state and/or the percentage of racial minorities comprise a state’s population (Fellowes and Rowe Reference Fellowes and Rowe2004; Ojeda et al. Reference Ojeda, Whitesell, Berkman and Plutzer2019; Owens and Smith Reference Owens and Smith2012; Soss et al. Reference Soss, Schram, Vartanian and Brien2001; Soss, Fording, and Schram Reference Soss, Fording and Schram2008). For example, Soss et al. (Reference Soss, Schram, Vartanian and Brien2001) analyzed state policy adoptions following the passage of PRWORA. They found evidence that family caps and time limits were more likely in states with higher proportions of Black and Latino AFDC recipients (Soss et al. Reference Soss, Schram, Vartanian and Brien2001, 386). The authors concluded that ideology and race primarily drove state welfare policies (Soss et al. Reference Soss, Schram, Vartanian and Brien2001, 389).
Similarly, Fellowes and Rowe (Reference Fellowes and Rowe2004, 365) examined the diffusion of TANF policies across states between 1997 and 2000 and found that race plays an essential part in the diffusion of welfare policies across states. Their research shows that state populations that included higher percentages of Blacks and Latinos were more likely to adopt stricter eligibility policies. In contrast, those with a higher proportion of Black residents on welfare were more likely to adopt harsher work requirements and lower levels of cash benefits (Fellowes and Rowe Reference Fellowes and Rowe2004, 367–68).Footnote 4
Owens and Smith (Reference Owens and Smith2012) also analyzed state policy adoptions after the 1996 welfare reform law passed. According to the authors, the proportion of racial minorities on TANF in a state has a threshold effect on the likelihood of states’ reformation of the federal prohibition against drug felons receiving TANF and food stamps (552). As the percentage of racial minority recipients increased, the time until states reformed the federal drug felon ban shortened. However, as the proportion of racial minority recipients increased in states with small/medium racial minority caseloads, states were faster to reform the ban (Owens and Smith Reference Owens and Smith2012, 552).
Little research has examined the diffusion of drug testing laws across states, apart from Ledford (Reference Ledford2018), who found that the number of Black TANF recipients and higher levels of state aggregate symbolic racism positively affected the likelihood of a state proposing drug testing laws for welfare. While Ledford (Reference Ledford2018) analyzed the effect of racial factors – including the racial composition of TANF recipients, state Black populations, and levels of state symbolic racism – on the propensity of states to propose drug testing laws, his analysis stopped at 2014 and did not include any measures evaluating the effect of drug use on diffusion. Moreover, Ledford’s analysis utilized multi-level logistic models instead of event history analysis to keep states that propose laws in the dataset instead of dropping them once they experience “failure” (that is, adoption). Ledford (Reference Ledford2018, 520) argued that since failure, in this case, does not mean that the event can never occur again – unlike more traditional failures in EHA analysis, such as death – states should not be dropped from the model after experiencing an “event.”
Of course, Ledford (Reference Ledford2018) is not the first to connect feelings of racial prejudice and the formation of welfare preferences (Avery and Peffley Reference Avery, Peffley, Soss and Fording2003; Gilens Reference Gilens1995, Reference Gilens1999). The relationship between feelings of racial resentment and more restrictive welfare policies relies on racial stereotypes suggesting that Black people are irresponsible, lazy, and get more than they deserve (Henry and Sears Reference Henry and Sears2002; 2003; Ledford Reference Ledford2018, 513; Sears et al. Reference Sears, Hetts, Sidanius, Bobo, Sears, Sidanius and Bobo2000).
Soss, Fording, and Schram (Reference Soss, Fording and Schram2008) proposed the Racial Classification Model (RCM) to explain how race affects welfare policymaking. The model combines elements of social classification and Schneider and Ingram’s (Reference Schneider and Ingram1997) social constructions. In short, the authors seek to explain why policymakers may react to race by adopting more punitive welfare policies. First, Soss et al. argued that legislators are just as prone to racial stereotyping as other Americans. Moreover, legislators use these stereotypes and information on the racial composition of policy clientele to conclude the traits and characteristics of that population (Soss, Fording, and Schram Reference Soss, Fording and Schram2008, 540). These conclusions regarding the population’s relative social desirability or self-sufficiency are then used to make decisions about the policies affecting these groups. Finally, Soss et al. explained that when racial cues as more predominant, i.e., when racial minority populations become a larger proportion of the target population, legislators’ racial stereotypes are activated (Reference Soss, Fording and Schram2008, 541).
Fellowes and Rowe (Reference Fellowes and Rowe2004) suggested that the effect of racial factors they observe in state welfare policymaking can be explained by combining four types of pressures: constituent, institutional, paternalistic, and resource. The theory regarding constituent pressure claims that as constituents become more favorable toward welfare programs – or, as the authors state, constituents are “more liberal, less racist, or less class-biased” – state policymakers respond to this change by passing more liberal welfare policies (Fellowes and Rowe Reference Fellowes and Rowe2004, 363). However, the opposite is also the case – when constituents become less liberal, more racist, and more class-biased, they pressure policymakers to implement more restrictive welfare policies.
The second type of pressure involves institutional factors like party control of government and government ideology. As expected, the institutional pressure theory posited that democratically controlled governments and more liberal governments pass more lenient welfare policies than those implemented by both Republican-controlled governments and conservative governments (Fellowes and Rowe Reference Fellowes and Rowe2004). The aforementioned Soss et al. (Reference Soss, Schram, Vartanian and Brien2001) evaluation of policy changes spurred by the 1996 welfare reforms also found that restrictive policies were adopted at significantly higher rates in states with conservative governments (386). Likewise, Karch (Reference Karch2007) also studied the causal mechanisms responsible for the diffusion of time limits and family caps in welfare policies across states and found that conservative states were more likely to adopt time limit and family cap policies (49).
There is also evidence that both the number of racial minority representatives and female representatives in a state legislature have a significant effect on the types of welfare policies states consider and adopt (Preuhs Reference Preuhs2006; Reingold and Smith Reference Reingold and Smith2012). An extensive scholarship chronicles the influence women have on the legislative process at the local, state, and federal levels. Broadly, women legislators are more likely to hold more liberal welfare preferences, introduce welfare legislation, and oppose punitive criminal policies (Owens and Smith Reference Owens and Smith2012; Poggione Reference Poggione2004; Reingold Reference Reingold2008; Reingold and Smith Reference Reingold and Smith2012). Because of these general differences, scholars have included measures of female incorporation in the legislature in analyses of the diffusion of welfare policies. However, as Reingold and Smith (Reference Reingold and Smith2012) noted, past research found contradictory results on women in the legislature’s effect on the kinds of welfare policies states pursue. For example, a study by Cowell and Langbein (Reference Cowell-Meyers and Langbein2009) on the likelihood of state adoption of specific TANF policies found higher numbers of women legislators “is just as likely to increase the odds of adoption as it is to decrease the odds, as it is to have no effect at all” (Reingold and Smith Reference Reingold and Smith2012, 134).
Similarly, scholars have noted the impact of incorporating racial minority legislators on the content of state welfare laws. Broadly, racial minority representatives are more likely than their White counterparts to propose and vote for legislation that reflects the interest of racial minority groups. This has been supported by research on both Black and Latino legislators (Haynie Reference Haynie2001; Owens Reference Owens2005; Preuhs Reference Preuhs2005; Santoro Reference Santoro1999; Tatalovich Reference Tatalovich1995). For example, Preuhs (Reference Preuhs2007) found that increases in the number of Latino representatives have a positive relationship with spending on welfare, the generosity of benefits, and the type of welfare benefits available to recipients (287). However, Preuhs (Reference Preuhs2007) and other authors have noted the possibility that the incorporation of racial minority legislators may not have much effect on policymaking for various reasons, including structural barriers facing Black/Latino representatives, the prior existence of substantive representation of the interests of racial minorities by non-descriptive representatives, and the paradox of representation created by the emergence of minority-majority districts (Hero and Tolbert Reference Hero and Tolbert1995; Klinkner and Smith Reference Klinkner and Smith1999; Lublin Reference Lublin1997; Preuhs Reference Preuhs2007, 278–79; Swain Reference Swain1995).
While the aforementioned Owens and Smith (Reference Owens and Smith2012) study found no significant relationship between the proportion of female or Black legislators and the likelihood of reforming the federal drug ban on TANF benefits, Reingold and Smith (Reference Reingold and Smith2012) found the increased presence and influence of women legislators resulted in relaxed welfare eligibility requirements (552). Moreover, the authors found instances where the work of women of color and men of color affected welfare policy, such as the passage of more flexible work requirements (Reingold and Smith Reference Reingold and Smith2012, 143).
Karch (Reference Karch2007) also found evidence supporting the third type of pressure described by Fellowes and Rowe (Reference Fellowes and Rowe2004) – resource pressure. Resource pressure argues that states with fewer financial resources will resort to less generous welfare policies and that neighboring states will “race to the bottom” to avoid attracting welfare recipients (Fellowes and Rowe Reference Fellowes and Rowe2004). Karch (Reference Karch2007, 47) found that wealth, or the slack resources hypothesis, positively and significantly affected time limits and family cap policies.
Finally, Fellowes and Rowe (Reference Fellowes and Rowe2004) detailed paternalistic pressures in which governments attempt to execute a moral agenda through various policy changes. According to this theory, more restrictive policies may be enacted if policymakers observe an increase in “immoral” behavior. In the case of drug testing laws, the presumed immoral behavior would be drug use. Drug use has traditionally been publicly perceived as a racialized issue, with media over-representing minorities in coverage of drug use (Gilens Reference Gilens, Schram, Soss and Fording2003).
Due to the influence drug policy may have on the diffusion of drug testing for welfare policies, it is also necessary to evaluate extant literature on recent trends in drug use. According to the Kaiser Family Foundation, White Americans account for 82–85% of opioid overdoses in the US compared to Black Americans, who accounted for 6–8% from 2009 to 2015. While White Americans remain the highest proportion of those who die from opioid overdoses from 2016 to 2018, Black Americans’ overdose rates have increased to between 10% and 13% (Kaiser Family Foundation 2020). Importantly, opioid use is perceived to be an overwhelmingly White problem. Furthermore, media coverage of White opioid addicts categorically differs from that of Black addicts (McGinty et al. Reference McGinty, Stone, Kennedy-Hendricks, Sanders, Beacham and Barry2019; Netherland and Hansen Reference Netherland and Hansen2017; Shachar et al. Reference Shachar, Wise, Katznelson and Campbell2020).
Content analyses of US media coverage of drug epidemics and opioid abuse have changed in recent years (McGinty et al. Reference McGinty, Stone, Kennedy-Hendricks, Sanders, Beacham and Barry2019; Netherland and Hansen Reference Netherland and Hansen2016; Shachar et al. Reference Shachar, Wise, Katznelson and Campbell2020). For example, Netherland and Hansen (Reference Netherland and Hansen2017) analyzed press articles between 2001 and 2011 to examine the differences in coverage of White non-medical opioid users and racial minority heroin users. The authors find that media portrayals were consistently more sympathetic toward White opioid users than racial minority heroin addicts. Netherland and Hansen describe this distinction as reminiscent of the distinction made in the 1980s and 1990s between crack cocaine (a “Black drug”) and powder cocaine (a “White drug”) (Reference Netherland and Hansen2017). On the other hand, McGinty et al. (Reference McGinty, Stone, Kennedy-Hendricks, Sanders, Beacham and Barry2019) found that news coverage of opioid use before 2007 was focused on criminal justice. However, that focus veered toward a prevention and treatment model between 2007 and 2009 (McGinty et al. Reference McGinty, Stone, Kennedy-Hendricks, Sanders, Beacham and Barry2019). The authors argue that this evolution can be attributed to increased opioid use among White Americans.
Another content analysis of news coverage of the opioid epidemic from 2016 to 2017 and the crack cocaine epidemic from 1988 to 1989 found that articles about opioid addiction were more likely to frame the issue as a public health concern rather than a criminal justice issue, while stories on the cocaine epidemic predominantly focused on criminalization and penalization (Shachar et al. Reference Shachar, Wise, Katznelson and Campbell2020). In addition, the authors studied articles referencing heroin use in three periods – 1988–1989; 1992–1993; and 2016–2017 – and determined that the alteration of media framing and government policy responses cannot be solely explained by changes in public views about substance abuse but, instead, can be attributed to racial factors (Shachar et al. Reference Shachar, Wise, Katznelson and Campbell2020). The authors explain that racial cues affect society’s perceptions of crime severity (Shachar et al. Reference Shachar, Wise, Katznelson and Campbell2020).
Using the work of Schneider and Ingram (Reference Schneider and Ingram1997) as well as Soss, Fording, and Schram’s (Reference Soss, Fording and Schram2008) RCM, it is likely than White opioid users are classified by policymakers as “dependents,” while Black illicit drug users are deemed “deviants.” State legislators and policymakers are not impervious to the influence of racial resentment and can rely on negative racial stereotypes to design policy (Ingram and Schneider Reference Ingram, Schneider, Schneider and Ingram2005; Ledford Reference Ledford2018, 515; Schneider and Sidney Reference Schneider and Sidney2009; Soss, Fording, and Schram Reference Soss, Fording and Schram2008). Moreover, scholars have noted that while welfare recipients have negative social constructions, Black welfare recipients have an even more negative construction. Many groups labeled deviant – welfare mothers, poor people, and drug users – are stereotypically associated with Black welfare recipients (Schneider and Sidney Reference Schneider and Sidney2009). The over-representation of Black men as criminals by the media, the criminal justice system, and government officials since the War on Drugs has resulted in the widespread stereotype that Black Americans are drug users (Welch Reference Welch2007).Footnote 5
In other words, it may be that because White opioid users are perceived as being more sympathetic, they are therefore seen to be more deserving of help than Black drug users. It should be noted that race’s influence on perceptions of crime severity and government response is not limited to drug use. For example, Heitzeg (Reference Heitzeg2015) notes the pattern of the medicalization of Whiteness and the criminalization of Blackness in various crimes, including violent crimes, sexual assaults, mass shootings, and so forth. Owens and Smith (Reference Owens and Smith2012) describe how the social and political rights of drug felons, aka “deviants,” have repeatedly been infringed on by the government, including the federal ban on government cash and food assistance to those convicted of drug crimes.
Theory
The observation of proposals precipitates a deeper understanding of the diffusion of policy interest and policy adoption. Moreover, there is reason to believe that looking at proposals has explanatory utility. For example, Hall (Reference Hall1998) explained how legislators have limited resources – both time and money. Therefore, sponsoring/introducing a bill can indicate policy priorities (Hall Reference Hall1998).
Scholars often argue that emulation plays a considerable role in policy diffusion (Berry and Berry Reference Berry and Berry1990; Boehmke and Witmer Reference Boehmke and Witmer2004; Karch Reference Karch2007; Volden Reference Volden2006). The theoretical reasoning for this is straightforward: the more successful a policy, the more likely that policy is to be pursued elsewhere. Importantly, Karch (Reference Karch2007) notes that this process is often non-partisan and, instead, relies on the relative effectiveness of the policy in question. However, policy emulation cannot be the primary mechanism propelling the diffusion of drug testing policies. For the most part, these policies have failed to achieve either of their stated goals – to identify a substantial number of public assistance beneficiaries who use drugs and restrict them from receiving benefits and, subsequently, to lower the overall costs of welfare programs in the state. Although drug testing laws have failed to find a significant proportion of welfare applicants using drugs, and the process of drug testing applicants has proven to cost more money than states save from eliminating drug users from welfare rolls, states continue to propose or pass such requirements. Therefore, there must be another mechanism or mechanisms driving policy diffusion.
As described above, researchers have generally found that the size of racial minority populations and the number of racial minority welfare beneficiaries in a state significantly influence the diffusion of welfare policies. For these reasons, the author expects that higher percentages of Blacks in a state’s population and the number of Blacks on TANF will increase the likelihood of the proposal of drug testing for welfare laws. Relatedly, the author theorizes that aggregated estimates of state-level racial resentment will have a positive relationship with proposals.
Hypothesis 1: Higher percentages of Blacks receiving welfare and higher percentages of Black populations will significantly increase the likelihood of drug testing for welfare proposals.
Hypothesis 2: Higher estimates of racial resentment will significantly increase the likelihood of drug testing for welfare proposals.
Volden (Reference Volden2006) and Fellowes and Rowe (Reference Fellowes and Rowe2004) both concluded that ideology also influences a state’s likelihood of adoption. Evidence also suggests that conservative states and those with unified Republican control are more likely to enact punitive welfare policies which restrict eligibility – such as those policies which mandate passing a drug test in order to receive assistance (Fellowes and Rowe Reference Fellowes and Rowe2004; Soss et al. Reference Soss, Schram, Vartanian and Brien2001; Volden Reference Volden2006). Therefore, I expect states with higher levels of conservatism and those controlled by the Republican party will be more likely to propose drug testing policies. As Gilardi (Reference Gilardi2010) suggests, politicians preoccupied with reelection may be more concerned with the political success of a policy rather than its effectiveness. Therefore, success, in the context of drug testing for welfare, may not be categorized by policy success but rather by political success.
Hypothesis 3: Republican party control and higher levels of conservative ideology will significantly increase the likelihood of drug testing for welfare proposals.
Finally, because of the racial disparities between opioid use and widespread illicit drug use and the perception of the opioid epidemic, I expect significant differences between the effect of state-level drug use and opioid overdoses on the likelihood of drug testing for welfare proposals.
Hypothesis 4: Higher levels of opioid overdose deaths will decrease the likelihood of drug testing for welfare proposals, while higher levels of illicit drug use will increase the likelihood of drug testing for welfare proposals.
Research design
For the past two decades, event history analysis (EHA) has been the traditional method of studying the diffusion of policies across states (Boehmke Reference Boehmke2009, 229). EHA entails a logistic or probit analysis wherein the dependent variable is a state-year binary variable representing whether a state adopted a policy. In other words, the dependent variable in hazard models has two parts, the event indicator variable – whether a state proposed drug testing for welfare – and a measure of time from the baseline time to the event. EHA is a pooled cross-sectional time-series analysis that allows researchers to observe the effect of independent variables that fluctuate yearly because of the analyses’ inclusion of longitudinal variation (Karch Reference Karch2007, 40). This method includes a state in the analysis until the policy in question is proposed. After one state in the model proposes a policy, all states in the model are classified as being at risk (Boehmke Reference Boehmke2009, 231; Karch Reference Karch2007, 205). Following a state’s policy proposal, it is dropped from the model (Boehmke Reference Boehmke2009, 231; Karch Reference Karch2007, 205).
Since the data are discrete-time data indicating whether a proposal or law occurred each year, the author uses an adapted Cox Proportional Hazard Model for recurrent events with states clustered and robust standard errors. This allows states to remain in the model even after their first proposal. As shown in Table 1, variables representing the major theories of the mechanisms that drive policy diffusion are included in the analysis for all fifty states from 2009 to 2018. The analysis begins in 2009 as it was the first year a state adopted a drug testing policy.
Table 1. Causal mechanisms of welfare policy diffusion and analysis variables
Theoretically, after the first passage year, all other states become at risk of proposal. In the final year, the author has data for 2018, and six states proposed drug testing legislation.
The dependent variable is a state-year dummy variable for each year indicating whether a state proposed drug testing for welfare (1) or did not (0) in that year. The main independent variables of interest are the proportion of Black families receiving TANF in a state, the percentage of Black Americans in a state population, and state-level estimates of racial resentment.Footnote 6 The racial resentment variable is extrapolated data from 2008, 2012, and 2016 state-level estimates of racial resentment taken from Smith, Kreitzer, and Suo (Reference Smith, Kreitzer and Suo2020). The authors used multi-level linear regression with poststratification weighting (MRP) to connect three variables – census data, American National Election Study (ANES) survey data measuring levels of racial resentment, and data on state-level variables that are predicted to influence racial attitudes – to create state-level estimates of racial resentment (Smith, Kreitzer, and Suo Reference Smith, Kreitzer and Suo2020).Footnote 7
The author accounts for the potential influence of institutional pressures by including party control of the government, government ideology, proportion of women in the legislature, and a South binary variable. The party control variable measures whether Democrats or Republicans control the state government, while the state government ideology variable is Berry et al.’s (Reference Berry, Ringquist, Fording and Hanson1998) measure which accounts for yearly ideological scores for five state governmental actors: the governor and the two major parties in both chambers of the state legislature. The Berry et al. data, updated to 2018, permits the examination of whether states with conservative governors and more conservative legislators are more likely to propose drug testing for welfare policies.
The correlation between this measure of state government ideology with other variables in the analysis, such as the other political variable, control of government, as well as total revenue, and the South region measure, would result in multicollinearity, potentially affecting the relative significance of the variables as well as enlarging the confidence intervals in the analysis. However, the correlation between the government ideology measure and the region variable, total revenues, and total population were all below reasonably low levels. State ideology and party control of government, however, were highly correlated at .80, indicating that multicollinearity may be a problem. Cronbach’s Alpha between party control of government and state government ideology was not sufficiently high enough to justify combining the variables. It should be noted that while the party control of government and state government ideology variables capture similar concepts, they are distinct. The former notes which party holds the governorship are the majority in the state legislature, while the latter accounts for the ideological distribution within state governmental actors.
Also included are variables that capture resource pressures. State total revenue, total debt, total population, per capita personal income, the average number of TANF recipients per month, and state unemployment rates are included to account for resources/economic influences on proposing or passing drug testing for welfare. Furthermore, since there is evidence that states compete in a race to the bottom, it may be that neighboring states track the eligibility rules of nearby states. The geographic proximity variable is operationalized as the proportion of border states that adopted drug testing policies. This statistical proxy is commonly used to represent geographic proximity in diffusion research (Berry and Berry Reference Berry and Berry1990, Reference Berry and Berry1992; Haider-Markel Reference Haider-Markel2001; Mintrom Reference Mintrom1997, Reference Mintrom2000). Finally, the author accounts for the potential impact of paternalistic pressures with the inclusion of estimates for monthly illicit drug use in a stateFootnote 8 and opioid deaths.Footnote 9
Results
Testing assumptions
Because the analysis uses a Cox Proportional Hazard Model, covariates are examined to identify violators of the model’s proportionality assumptions. To test for proportionality, the author evaluates Schoenfeld residuals for each variable in the analysis. Following the Cox Model, there are several violators of the proportionality assumption. To avoid model overfitting, which can occur when too many indicators relative to the number of failures in the sample are included in the model, only the significant variables from Model 1 are included in Model 2. Moreover, to account for the potential that statistically significant violators of the proportionality assumption in Model 1 are time-dependent, government ideology is included in the main model and as a time-varying variable in Model 2.
Drug testing proposals Model 1Footnote 10
Table 2 shows that while the coefficient for the percentage of Blacks in a state’s population is in the hypothesized direction, it is not statistically significant. Moreover, the number of Blacks on TANF is predicted to have a negative but statistically insignificant effect on proposals. Therefore, Hypotheses 1 and 2 are not supported. However, Model 1 finds that the largest substantive influence on the likelihood of proposal is state-level racial resentment estimates – for every standard deviation increase in state racial resentment levels, the proposal’s likelihood is predicted to increase by a factor of 2.35. In other words, as states become more racially resentful, the estimated rate of proposal increases by a substantial 135%. This supports my hypothesis that higher levels of racial prejudice increase the likelihood of the proposal of drug testing laws.
Table 2. Factors associated with the proposal of drug testing for welfare laws
Model 1: N = 342, number of failures = 123; Model 2: N = 342, number of failures = 123, robust standard errors reported in parentheses. Significance levels: ***p ≤ .01, **p ≤ .05, *p ≤ .10.
Table 3. Support of hypotheses
The author also finds support for Hypothesis 4. Both estimates related to drug use – illicit drug use and opioid overdose death rates – are statistically significant. The higher the percentage of estimated illicit drug users in a state, the higher the likelihood of a state proposing a law (p ≤ .10). Conversely, as more people die of opioid overdoses, the probability of proposing a law decreases (p ≤ .01). For every one percent increase in opioid overdoses, the rate of proposal decreases by 41%, while every one percent increase in illicit drug use estimates increases the rate of proposal by 16% (Figure 2).
Figure 2. Probability of proposal, Model 1.
Government ideology is the only political variable with a statistically significant influence on proposal rates. Contrary to what was predicted in Hypothesis 3, as states become increasingly liberal, the predicted likelihood of proposal increases by 92% (p ≤ .01). The author theorizes that this is because drug testing bills are more likely to pass through the legislative process and be signed into law sooner in conservative states, while drug testing bills are repeatedly proposed by the minority party in liberal states. Interestingly, while the proportion of women in the state legislature does have a negative relationship with the likelihood of proposal, it is not statistically significant.
The economic variable of per capita personal income significantly decreases the likelihood of proposal. For every dollar increase in per capita income in a state, the rate of proposal decreases by 31% (p ≤ .01). Unemployment rates have a positive and statistically significant effect on proposals. Every one percent increase in a state’s unemployment rate increases the predicted rate of proposal by 94% (p ≤ .01). The overall number of TANF recipients significantly decreases the estimated rate of proposal by 21% (p ≤ .10). The author suspects that this negative relationship is a manifestation of the electoral incentives of elected officials – elected officials with more welfare recipients in their constituencies may experience electoral motivations to keep the barriers to receiving welfare assistance low.
Finally, geographic proximity decreases the hazard rate. The more states in the region that have adopted drug testing laws, the less likely a state is to propose a drug testing law (p ≤ .01). This could be explained in various ways. First, it could be due to state drug testing laws’ various legal challenges.Footnote 11 These legal difficulties could discourage neighboring states’ efforts from pursuing similar legislation. Second, this phenomenon could be explained by the failure of drug testing laws to accomplish their stated goals – to find a significant number of drug-using welfare recipients and to reduce government spending. Failing to accomplish these policy objectives may discourage neighboring states from proposing drug testing legislation.
Drug testing proposals Model 2
To avoid overfitting, Model 2 includes only those significant variables in Model 1, and the government ideology is included in the main model and as time-varying due to its violation of the proportionality assumption in the first model. State estimates of racial resentment continue to have a positive and substantive relationship with the likelihood of proposal. Every one-unit increase in a state’s racial resentment estimate increases the likelihood of proposal by 113% (p < .01).
The hazard ratios for illicit drug use and opioid overdoses change very little in Model 2. Illicit drug use continues to significantly increase the rate of proposal, this time by 24% (p < .01). Alternatively, an increase in the age-adjusted opioid overdose death rate is predicted to decrease the rate of proposal by 40% (p < .01). The main government ideology variable and the interaction between government ideology and time are statistically significant in Model 2. The predicted rate of proposal increases by 84% for every one-unit increase in government ideology (p ≤ .05). Moreover, the effect of government ideology on proposals is multiplied by 8% for every one-year increase in time (p ≤ .05).
While per capita income does not have a statistically significant effect on proposals in Model 2, unemployment continues to have a statistically significant and positive effect. The rate of proposal is estimated to increase by 128% as unemployment rates increase by one percent (p < .01). Finally, the geographic proximity variable has a negative and statistically significant relationship with drug testing proposals – the rate of proposal decreases by 38% for every one unit increase in the geographic proximity variable (p < .01). In other words, states become less likely to consider drug testing welfare applicants as the proportion of bordering states which have adopted this policy increases.
Discussion
As shown in Table 3, the author finds support Hypotheses 2, 3, and 4. While the proportion of Black families on TANF and the percentage of Blacks in a state does not significantly increase the likelihood of proposal, there is still evidence that race plays an essential role in the diffusion of drug testing for welfare policies. The relationship between state-level estimates of racial resentment and predicted proposal rates in both models is substantively large and statistically significant. Higher estimates of racial resentment are predicted to increase the likelihood of proposal by 135% in Model 1 and 113% in Model 2. Supporting Hypothesis 4, higher estimates of illicit drug use increase the likelihood of proposal, while higher numbers of opioid overdose deaths have the opposite effect in both models. Because of the theorized relationship between public perception and social classification/constructions of those abusing opioids and those abusing other illicit drugs (Santoro and Santoro Reference Santoro and Santoro2018), the author suspects these results also suggest that race is an important determinant of proposal.
Drug testing laws are more likely to be proposed in states with liberal governments in both Model 1 and Model 2. The author believes that proposals are more likely in states with liberal governments than in states with conservative governments because adoption rarely occurs in states with more liberal governments. Despite this, Republican state legislators in these states repeatedly propose drug testing. In more conservative states, adoption is more likely, and hence there are fewer opportunities to propose bills repeatedly. While the author intends to collect data on the party identification of every sponsor and co-sponsor of the proposals in the data, for now, the author finds some support for this theory in data collected for three states: Illinois, New York, and Massachusetts.
Illinois, for example, has the most proposal years in the sample data – the Illinois state legislature had at least one drug testing bill proposed every year from 2009 to 2018. Illinois is the seventh most liberal state in the data according to the state government ideology variable. Multiple bills were proposed in six of the ten years – in both 2015 and 2017, seven individual drug testing for public assistance bills were introduced in the state legislature. Of the 28 legislators who sponsored or co-sponsored the total of 33 proposals between 2009 and 2018, 25 were Republicans, and three were Democrats.
The state of New York – which ranks amongst the 11 most liberal states according to the Berry et al. measure of government ideology – saw proposals of legislation requiring drug testing for public assistance in eight of the 10 years in the analysis, the second highest number of years in the sample. For five of those years, multiple bills relating to drug testing for public assistance were introduced, for a total of 21 bills. Of the 43 state legislators who sponsored or co-sponsored drug testing bills in New York between 2009 and 2018, 98.71% were Republican, and 1.30% were Democratic.
Finally, Massachusetts, the most liberal of the 50 states from 2009 to 2018 according to the Berry et al. measure of state government ideology, had drug testing for public assistance policies introduced in five of the 10 years in the analysis. Of the 24 state legislators who sponsored or co-sponsored such legislation, 70.83% were Republicans and 29.17% were Democrats. The results from these three states indicate that over 70% of sponsors and co-sponsors on legislation relating to drug testing for public assistance in Illinois, New York, and Massachusetts – liberal states with a higher number of proposal years in the sample – were members of the Republican Party. This supports the idea that Republican state legislators repeatedly propose drug testing legislation in liberal states, making liberal state governments more likely than their conservative counterparts to have drug testing policies proposed.
The economic variable of state unemployment rates has a positive and significant relationship with drug testing proposals in both Model 1 and Model 2. The author theorizes that this is due to efforts on behalf of state governments to restrict welfare eligibility in times of economic hardship to reduce government spending. This theory is also somewhat supported by the results in the proposal Model 1, in which higher per capita incomes significantly negatively affect the rate of proposals.
In short, racial/constituent, paternalistic, resource, and institutional factors influence the likelihood of drug testing for welfare proposals. Moreover, paternalistic/racial pressures – operationalized by both drug variables – and institutional pressures – operationalized through government ideology – were the most consistent and significant predictors of proposal.
Nevertheless, there is much more to examine regarding drug testing for welfare. Future research should focus on qualitative analyses of the states that passed drug testing laws. These studies could examine differences in the severity of including drug treatment conditions to receive assistance and lifetime bans by state. Relatedly, scholars should study how states that have passed drug testing laws implement them and how these policies affect the lives of welfare recipients. Finally, content analyses of media coverage of drug testing proposals could provide insight into framing techniques and how the media and policymakers present drug users.
Data availability statement
Replication materials are available on SPPQ Harvard Dataverse at https://dataverse.harvard.edu/dataverse/sppq/.
Funding statement
The authors received no financial support for the research, authorship, and/or publication of this article.
Competing interest
The author declares none.
Author biography
Dr. Liza Gordon-Rogers is a research associate at the Union of Concerned Scientists (UCS), where she studies election administration. Prior to joining UCS, Dr. Gordon-Rogers served as a Visiting Assistant Professor of Political Science at Gettysburg College for two years. Dr. Gordon-Rogers received her master’s and doctoral degrees from West Virginia University.
