To our knowledge, this is the first study to examine sociodemographic characteristics associated with RTRM non-response using a BRFSS sample. Overall, 15.3% of survey respondents did not respond to the RTRM. Patterns of non-response varied across population subgroups. We observed heterogeneity in non-response across RTRM items where differential treatment in work (54.8%) and healthcare settings (26.9%) had the highest levels of non-response. There were significant differences in non-response by age, race/ethnicity, socioeconomic status and employment status.
In this SC-BRFSS sample, response patterns to the RTRM varied across population subgroups. Individuals who identified as Hispanic, indicated unreported income, and were uninsured had higher percentages of non-response to the RTRM, while individuals who were older and retired had lower percentages of non-response. These categories are similar to the groups that are typically non-responders to the BRFSS core. BRFSS makes efforts to weight the sample data by age, gender and race/ethnicity, with more recent administrations of the BRFSS employing a raking weighting methodology that additionally considers marital status and socioeconomic status (e.g., educational attainment and property owner/rental status) to reduce bias and improve representativeness not only in the BRFSS core but potentially in optional modules.
RTRM item-specific non-response ranged from 16.3 to 54.8%, with the highest levels of non-response observed among the questions inquiring about differential treatment in work and healthcare settings. Reasons for the high level of non-response for differential treatment in work settings may be attributed to labor market participation. In our sample, only 55.4% of the population reported being currently employed. The other portion of respondents comprised of those who reported being unemployed, retired, homemaker/student, or unable to work, and may have decided to not answer the question because of their work status. The question querying respondents about differential treatment in healthcare settings also had relatively increased levels of non-response (overall, 26.9%). We do not suspect this is a function of health insurance status since many respondents reported having some type of health insurance. However, research assessing specific psychometric properties of these questions may be needed to further refine this important question about one’s lived experience.
Racial/ethnic differences were observed in item-specific non-response. In general, Black and White respondents had relatively similar levels of non-response across RTRM items. Hispanic respondents had the lowest levels of non-response for the questions about differential treatment in work settings. Labor market participation and age of Hispanic respondents may partially explain these differences. Moreover, Hispanic respondents had the highest levels of non-response to the questions about socially-assigned race, race-consciousness, and emotional and physical reactions to race-based treatment. Willingness and motivation to respond to the aforementioned questions may be dictated by several factors. Hispanics respondents may experience a unique constellation of structural disadvantages (e.g., immigration legal status and language proficiency) that intersect with the traditional axis of inequality that emphasizes race which may influence actual and perceived encounters with negative and differential treatment [25]. A higher non-response to these questions may also be a function of the classification of Hispanics/Latinos as an ethnicity rather than a racial group. Prior research documents that directives established by the Office of Management and Budget (OMB), which mandates the standards and provides guidance for the collection of race and ethnicity data in the US, may not provide relevant options for describing Hispanic racial identity and the options can be confusing because it may not cohere with Hispanics’ understanding of race [26, 27]. The extent to which differential responses to these questions and for people of different racial background, ethnic identity, or nativity, which are all important correlates of one’s racialized lived experience, is not clear and needs to be further explored. It is possible that RTRM items carry sociopolitical implications and can evoke associations, feelings, or different judgments in the minds of respondents that give rise to particular interpretations or function differently by race/ethnicity [28]. Notably, Hispanics only represented 5% of respondents, thus, the observed levels of non-response need to be interpreted with caution. Future studies should assess differential item functioning of the RTRM items by race/ethnicity with sufficient sample sizes.
To our knowledge, BRFSS does not report response rates of optional modules in the annual data quality report. We are aware of one prior study that assessed non-response to a BRFSS optional module [23]. Crouch et al. examined the SC-BRFSS Adverse Childhood Exposures (ACE) optional module and documented differences between responders and non-responders by sex, age, race/ethnicity, education, income, and rurality in bivariate analyses. Overall, there were some similarities in the range of sociodemographic factors (e.g., age, race/ethnicity, and income) associated with non-response as evidenced from our bivariate analyses. However, unlike Crouch et al. (2018), we did not observe differences in response type by sex, educational attainment or geographic region of residence for the overall RTRM. Yet, for few RTRM items, sex and educational attainment were related to non-response. For example, differential treatment in work settings, females were over-represented among non-respondents. Further, we identified retirees and uninsured respondents overrepresented among overall RTRM non-responders. Moreover, in adjusted analyses, Hispanic respondents, those with unreported income, and retirees were most consistently associated with non-response in the RTRM and across RTRM items. This suggest that when studying race-based differential treatment and health, associations may be weaker among these groups and are likely to be conservative. It is not well known how non-response to BRFSS optional modules are affected by sociodemographic factors. Future studies estimating and assessing factors influencing RTRM and other optional module response levels may help to improve understanding of the extent of bias and its implication for interpreting results.
It is imperative that studies using these data consider employing analytic techniques that can accommodate patterns of missingness during the analysis stage. Typical approaches to compensate for variations in probability of selection and overall nonresponse, including weighting the sample data, may not fully address item-specific nonresponse. Prior research has shown that developing weights for a subset of respondents answering specific modules, for example, is not practical [29]. For data missing at random, considering approaches such as multiple imputation [17, 29, 30], full information maximum likelihood estimation [17, 30], and inverse probability weighting [31, 32] during the analysis stage are potential strategies to reduce bias and improve validity of the estimates. Further, RTRM nonresponse may also be a function of order effects. In South Carolina, the optional module is administered after core component questions and are often placed towards the end of the questionnaire.
As with any study using the BRFSS, we acknowledge that the data is limited by its retrospective, self-reported nature. Similar to questions about sensitive and stigmatizing life circumstances (e.g., abuse/neglect,) or health conditions (e.g., mental health, substance use), responding to experiences of race-based differential treatment may be subjected to recall and social desirability bias. While these results may be comparable to other states in the American South, they are not generalizable nationally given the limited sample size of Asians, Native Americans, and Hawaiian/Pacific Islanders. The RTRM was developed specifically for public health surveillance of racialized lived experiences in a US context. Despite these limitations, our paper has some important implications. Valid and reliable measures of experiences of racial discrimination in population-based surveys may be critical for guiding anti-racist action in public health practice and programs and informing anti-racist, data-driven policy implementation. This public health surveillance data is valuable for illuminating sources of inequities between specific populations. Additional research is needed to consider whether the observed factors influencing non-response in RTRM are consistent across other states with different proportions of Hispanics, Asians, and Native Americans, and multiracial populations groups. However, to provide this type of insight, it is incumbent among more states to administer this optional module as part of the BRFSS to provide additional insight regarding the contribution of racial discrimination to health and well-being.