We used publicly available data from the 79th Michigan State Quarterly Survey (SOSS) conducted in May 2020. The survey was a stratified random sample of 1086 non-institutionalized Michigan English-speaking adults who could be reached by cell phone or fixed. . About 35% of the interview sample came from recontacts. The remaining 65% of the sample came from a random sample of phone numbers in the state. Nonresponse adjustments were made to ensure that the sample was representative of the state’s adult population. The sample was matched to a sample frame of 1,000 respondents, constructed from the 2016 American Community Survey. Matching was based on gender, age, race, and education and weighted against to the sampling frame using propensity scores [22, 23].
Interviews were conducted using the Computer-Assisted Telephone Interviewing (CATI) system of the Office of Survey Research (OSR) of the Institute for Public Policy and Social Research (IPPSR). [22, 23]. CATI is a survey modality in which interviewers follow a script on the computer to conduct interviews over the phone. The OSR CATI system uses built-in logic that allows sequential movement from question to question and automatic skip patterns based on answers. . The dataset analyzed in this study is available on the Michigan State University website at http://ippsr.msu.edu/survey-research/state-state-survey-soss/soss -data/soss-79b-spring-2020 .
The purpose of this study was to assess the association between seeking information, worries/fears about the pandemic, trust, political ideology, and willingness to participate in contact tracing in Michigan.
Factors Contributing to Willingness to Participate in Contact Tracing
Given the importance of contact tracing in public health efforts to mitigate the spread of COVID-19, we examined factors associated with willingness to participate in contact tracing. We were inspired by literature on the use of information technologies such as apps, preventive health behaviors and the context of COVID-19 at the time.
Our primary outcome variable was a composite measure of willingness to participate in contact tracing efforts derived from responses to three questions. Responses were measured on a seven-point scale ranging from “not at all true”  to “Very true” . The composite index for each respondent was the average measure of the following three questions: (i) “I would feel comfortable reporting people I have been in contact with to the local or state health department if I had symptoms of COVID-19” (ii) “I would be comfortable using a computer or phone app that shares my symptom information with my local or state health department” and (iii) “I am willing to give my local or state health department personal information to help limit the spread of COVID 19” (Cronbach’s Alpha: 0.87 (CI=0.85, 0.88).
Independent variables: predictors of willingness to participate in contact tracing
We included variables capturing the following factors potentially associated with willingness to participate in contact tracing: (1) seeking information, (2) concerns and fears, (3) trust/mistrust, (4) political ideology and (5) demographic factors (Tables 1 & 2). Unless otherwise stated, responses were measured on a seven-point scale rating “how true” people felt a series of statements were. Responses ranged from “not true at all”  to “Very true” .
Information search was captured by questions about where people got information about COVID-19, trusted sources of information, frequency of seeking information, number of sources, and whether they believed the media accurately depicted the severity of the pandemic.
Survey respondents were asked to indicate the frequency (regularly, occasionally, rarely, never) and level of trust (not at all (1) – very much (7)) in sources of information, including public health institutions (CDC, state and local health). departments), national media sources, local media sources, health care (health care providers, insurance companies) and social networks (friends and family). We aggregated responses on national media sources to capture the frequency of and trust in right- and left-wing centrist sources, categorized as such based on a spectrum defined by the Pew Research Center . The set of questions is provided in Table 1.
To assess the total number of information sources, we created dummy variables equal to 1 if the respondent reported obtaining information occasionally or regularly from an information source and zero otherwise and summed over the all 21 sources to calculate the total number of sources of information on COVID-19. 19 for each respondent. Table 1 summarizes the variables generated from several questions and their internal consistency (Cronbach’s alpha).
We asked three questions about various concerns and fears related to the pandemic. First, we asked about the perceived threat of the coronavirus to personal health. Second, to assess concerns about the harmful effects of misinformation, we asked whether or not the statement “I fear misinformation about COVID-19 will make people less sure” is true. Third, we asked them if they were concerned about private information being used against them.
Trust and mistrustwere measured using a composite measure of general trust in the healthcare system based on three questions (Cronbach’s alpha: 0.87, 95% CI: 0.85, 0.88) (see table 1). To assess distrust, or beliefs about the integrity of the state and the federal government, we asked people to indicate the extent to which they believed two statements to be true: “I think the governor’s office has an agenda that keeps them from telling the whole story to the public” and “I think the feds have an agenda that prevents them from telling the whole story to the public” (emphasis added).
Political ideology [conservative, moderate, liberal, or other] was also measured given the politicized nature of the pandemic throughout its development. Respondents reported demographic factorsage, gender, race/ethnicity [White, Black, Other (not reported)]and education [less than high school, high school graduate, some college, college graduate or higher]that we included in our statistical analysis.
The variables analyzed were ordinal variables with non-normal distributions. Quantile-quantile (QQ) plots of all variables appeared linear, but variable distributions were negatively or positively skewed . The Shapiro-Wilk and Anderson Darling normality tests all gave p-values less than 0.05, implying that the assumptions of normality were not satisfied. Spearman’s rank order correlation and Mann Whitney’s U tests were used to analyze the data. Both methods are nonparametric methods that do not rely on the assumption of normality and are suitable for the analysis of continuous and ordinal data.
The data contained 12% missing data, which was imputed by hot deck imputation. Hot deck imputation is appropriate when missing data is random , which we assessed by visual inspection. Visual inspection showed arbitrary or unstructured missing data patterns with no obvious mechanism, suggesting that the missing was ignorable.
A Spearman rank order correlation analysis was conducted to determine the strength and direction of the association between the dependent variable (willingness to participate in contact tracing) and each continuous independent variable (Table 3). We also performed an exploratory bivariate analysis to examine the relationship between the dependent variable and each categorical independent variable (Table 2). The dependent and independent variables used in the regression analysis were standardized to allow simple comparisons of effect sizes and to facilitate interpretation.
The construction of our final model followed a structured approach. Based on exploratory bivariate analysis (Mann Whitney and Spearman rank correlations), we submitted candidate predictors with significant relationships (pp
Based on the results of the stepwise regression, we ran a multiple linear regression model to investigate the relationship between willingness to participate in contact tracing and the independent variables identified in the stepwise regression. Residual analysis was used to verify model assumptions. The assumptions of normality of residuals and homogeneity of variance were not violated. There were few potential outliers, but the Cook’s distance values calculated to assess the effect of outliers were less than 0.28, below the influential point threshold of one or more. . A multicollinearity test showed that the variance inflation factors were within acceptable limits ranging from 1.1 to 2.6 indicating that there was no substantial multicollinearity in the model . Thus, the final model matches the data.