Survey design
On June 6, 2017, the Seattle City Council passed Ordinance 125324 imposing a tax on distributing sugary beverages in Seattle. Tax implementation began on January 1, 2018. In Seattle, large distributors now pay a 1.75 cents per ounce tax on sugary beverages. Taxed beverages include drinks that have added sugar. The tax does not include diet beverages, 100% fruit juices, or milk products. The tax was intended to address equity issues, as revenues from the tax are being invested in programs that increase access to healthy and affordable food, expand early education for pre-school aged children, and help high school graduates enter college [11]. We designed a survey to investigate Seattle residents’ perceptions about the tax itself and their views on the potential positive and negative health and economic impacts of the tax. These analyses focused on survey questions that addressed perceptions about: (1) the tax itself (5 items); (2) the health and economic impacts of the tax (6 items); and (3) demographic characteristics (12 items).
Data collection
Data were collected by a survey research firm, Ironwood Insights Group, LLC, between October and December 2017, using a mixed-mode (telephone and online) survey. The telephone survey was conducted by trained interviewers and standard data quality assurance checks were employed (e.g. live monitoring of approximately 10% of all calls over the survey period). Participants were contacted up to 6 times via telephone and 2 times on average. They were not compensated for participation.
Phone survey participants were selected using a stratified random sampling approach, further described below, sampling from databases of all working landline and cell phone numbers for Seattle. Participants who completed the survey online were selected from several existing panels comprised of a large sample of individuals who either completed online surveys or opted in to participate in online surveys in the past. All Seattle residents age 18 and older were eligible for inclusion. Those refusing to answer the screener questions on income and race/ethnicity or who did not speak or read English or Spanish or read Vietnamese were ineligible.
We designed the survey to be able to test whether opinions about the tax were different for lower-income versus higher-income populations. Lower-income was defined as < 260% of the Federal Poverty Line (FPL). We estimated that in order to detect a 10 percentage point difference in tax approval between higher- and lower-income participants, with a power of 80% and a 5% probability of Type I error, assuming a 60% tax approval rating, we would need a sample of 356 participants per income group. We successfully recruited 456 higher-income participants and 395 lower-income participants. We also aimed to recruit a sample that had the same race/ethnic distribution as the population of Seattle, based on the 5-year American Community Survey (ACS) sample (2012–2016).
We recruited a total of 851 participants (46% completed by phone and 54% completed online). Similar to response rates in national-level random digit dial surveys [15, 16] and a recent evaluation of the sugary beverage tax in Philadelphia, [5] our survey had a response rate of 3.6% for participants contacted via landline and 6.7% for participants contacted via cellphone, estimated using the American Association for Public Opinion Research Response Rate Number 4 (see Additional file 1: Table S1) [17]. We were not able to estimate a response rate for our online sample; however, the demographic characteristics were mostly similar by mode (telephone and online) across income and race/ethnicity (Additional file 1: Table S2).
The phone and web versions of the survey were offered in English and Spanish and we also offered the web version of the survey in Vietnamese. The University of Washington School of Public Health Institutional Review Board determined that this study was exempt.
Description of key variables
Primary independent variables
Participants were categorized as having incomes below <260% FPL or ≥ 260% FPL based on their self-reported total annual household income and given household size, defined using the annual federal poverty guideline (see Additional file 3 for survey questions) [18]. We created mutually exclusive categories for race/ethnicity, based on self-reported answers to two separate questions about race and ethnicity. Based on responses, individuals were categorized as follows: non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, non-Hispanics of “other” race, or Hispanic. Native Hawaiian or Other Pacific Islanders, American Indian and Alaska Natives, and individuals who reported two or more races, were categorized as non-Hispanic of an “other” race because Native Hawaiian or Other Pacific Islanders and American Indian and Alaska Natives each make up a very small segment of the population in Seattle when categorized separately.
Additionally, participants were asked to indicate which of five categories reflected their education (some high school, completed high school, some college or university, completed graduate or professional degree). Participants were also asked to indicate their gender (male, female, self-identify), their age (18–30 years old, 31–40 years old, 41–50 years old, 51–64 years old, ≥ 65 years), annual household income (<$30,000, $30,000–$59,999, $60,000–$89,999, $90,000–$120,000, >$120,000), political affiliation (Democrat, Republican, Independent, Other), and whether they had heard of the tax prior to participating in the survey (yes, no, don’t know). In addition, participants were asked about their consumption of sugary beverages during the prior 30 days, using a modified version of the NHANES Dietary Screener Questionnaire (none or < 1 per week, 1 per week, 2–6 per week, 1 per day, ≥ 2 day, don’t know) [19].
Primary dependent variables
The primary dependent variables included participants’ opinion about the tax and participants’ perceptions regarding the potential impact of the tax on: child well being, public health, cross-border shopping, small businesses, the Seattle economy, job loss, family finances, tax effects on people with low-income and people of color, and autonomy to choose what beverages one drinks. Questions were queried as four-category variables and there was also an option to report “don’t know” or “refused” (see Additional file 3). In this survey, we described the tax itself (e.g. 1.75 cents per oz), explained what the tax revenue would be used for in Seattle (e.g. increased access to healthier food) and then asked participants’ opinion about the tax itself using a four-category Likert scale, with response options of strongly approve, somewhat approve, somewhat disapprove, and strongly disapprove. Then, to assess participants’ perceptions about the health and economic impacts of the tax, each participant was read two statements and asked to indicate if the first statement was much or somewhat closer or the second statement was much or somewhat closer to her/his belief. For example, participants were asked whether the statement I will travel to another city to buy sugary drinks so I don’t have to pay the tax was somewhat or much closer to their own view as compared to the statement I will not travel to another city to buy sugary drinks because of the tax. For simplifying the reporting of the results, we collapsed the responses from four-category variables to two-category variables, since preliminary analyses indicated that the direction of the associations and statistical significance were similar when using a two- or four-category variable. “Strongly” and “somewhat” agree were collapsed into “agree”, “strongly” and “somewhat” disagree were collapsed into “disagree”. Similarly, “somewhat” and “much” closer for the matched pair statements were collapsed to capture respondents’ agreement with that statement. In our analyses, we also report the “don’t know” responses, but “refusals” were coded as missing values.
In addition to examining individual survey items, we created a score to summarize overall perceptions of perceived health and economic impacts of the tax (henceforth referred to as the tax impacts score). The tax impacts score was comprised of the nine questions related perceived tax impacts on: child well being, public health, cross-border shopping, small businesses, the Seattle economy, job loss, family finances, impacts of the tax on low-income people and people of color, and autonomy over beverage choice. A participant received a 1 if they responded that the impact of the tax would be positive/beneficial (e.g. tax will improve public health), a 0 if they responded that they “don’t know”, and a − 1 if they responded that the impact of the tax would be negative/detrimental (e.g. tax will not improve public health) (score range − 9 to + 9). A higher score indicated that the tax was perceived to have more positive impacts on health and economic factors.
Statistical analysis
Analyses of specific questions
We first estimated participants’ perceptions of the impacts of the tax for the entire sample. Based on prior evidence that documents differences in consumption by race/ethnicity and income [20, 21, 31, 32], we hypothesized that perceptions about the tax would also vary by these factors. Therefore, we then used chi-square tests to test for differences in participants’ response to survey questions by high- versus low-income (3 × 2 chi-square) and across all race/ethnicities (3 × 5 chi-square)(e.g. were there any statistically significant differences in the proportion of respondents who reported “agree”, “disagree”, or “don’t know”, comparing all race/ethnicities to each other). We did not test for statistical differences between each racial/ethnic group in order to avoid excessive statistical testing and because we were not powered to do so. In supplemental analyses, we also describe tax support by gender, age, education level, political affiliation, participants’ prior knowledge of the tax (i.e. had they ever heard of the tax) and consumption (none or < 1 per-week, ≥ 1 per-week), but we do not test for statistical differences between each group to avoid excessive statistical testing.
Analyses of tax impact score
In addition, we aimed to better understand the association between demographic characteristics and overall perceptions of the tax, as measured the tax impact score, on a continuum. We first used unadjusted linear regression models, with robust standard errors, to estimate whether income was associated with the tax impacts score. In a second, separate model, we estimated whether race/ethnicity was associated with the tax impacts score. We then used linear regression models to further explore the association between income and race/ethnicity and the tax impacts score, in one model that mutually adjusted for income (<260% FPL, ≥ 260% FPL) and race/ethnicity, while also controlling for education, sex, age, and political affiliation.
All results presented are based on analyses using survey weights, constructed using the raking method, an iterative proportional weight (see Additional file 2) [22]; weights adjusted results to the known City of Seattle population totals (as determined by the 5-year ACS) for race/ethnicity, gender, age, and annual median household income. All statistical analyses were performed using Stata 15.1 (StataCorp LP, College Station, Texas). Both chi-square and linear regression results were adjusted for multiple comparisons using the Holm-Bonferroni Sequential Correction, in order to determine statistical significance (p < 0.01 in these analyses) [23]. First, results for both chi-square and the linear regressions were each ordered from the smallest p-value to the largest p-value. Second, the second smallest p-value is corrected with a Bonferroni approach ([number of tests − order of test + 1] × p-value). The correction procedure stops when the first non-significant test is obtained.