Loss of Work Due to Illness
RANDS during COVID-19
The Research and Development Survey (RANDS) is a platform designed for conducting survey question evaluation and statistical research. RANDS is an ongoing series of surveys from probability-sampled commercial survey panels used for methodological research at the National Center for Health Statistics (NCHS). RANDS estimates were generated using an experimental approach that differs from the survey design approaches generally used by NCHS, including possible biases from different response patterns and increased variability from lower sample sizes. Use of the RANDS platform allowed NCHS to produce more timely data than would have been possible using our traditional data collection methods. RANDS is not designed to replace NCHS’ higher quality, core data collections. Below we provide experimental estimates of loss of work due to illness with coronavirus for two rounds of RANDS during COVID-19. Data collection for the first round occurred between June 9, 2020 and July 6, 2020 and data collection for the second round occurred between August 3, 2020 and August 20, 2020. Information needed to interpret these estimates can be found in the Technical Notes.
NCHS included a question about the inability to work due to being sick or having a family member sick with COVID-19. The National Health Interview Survey (NHIS), conducted by NCHS, is the source for high-quality data to monitor work-loss days and work limitations in the United States. For example, in 2018, 42.7% of adults missed at least one day of work in the previous year due to illness or injury and 9.3% of adults were limited in their ability to work or unable to work due to physical, mental, or emotional problems.
The experimental estimates on this page are derived from RANDS and show the percentage of U.S. adults who did not work for pay at a job or business, at any point, in the previous week because either they or someone in their family was sick with COVID-19.
- Loss of Work Due to Illness
- Telemedicine
- Reduced Access to Care
Use the drop-down menus to show data by selected groups (age, race and Hispanic origin, sex, education, urbanization, and chronic conditions).
Technical Notes
Data Source
The National Center for Health Statistics (NCHS) Division of Research and Methodology (DRM) contracted with the National Opinion Research Center (NORC) to conduct the Research and Development Survey (RANDS) during COVID-19. The sample for this study was drawn from NORC’s AmeriSpeak® Panel (amerispeak.norc.orgexternal icon) using a stratified sample design to obtain a random, representative sample of U.S. adults aged 18 and over, where sampling strata were defined by race and Hispanic origin (grouped by 1) Hispanic, 2) black non-Hispanic, and 3) white non-Hispanic or other non-Hispanic), age (18–34, 35–49, 50–64, and 65 and over), sex (male or female), education (Bachelor’s degree or less or Bachelor’s degree or above), and income level (less than $75,000 or more than $75,000). NORC performed sampling independently within each of the 96 strata using simple random sampling. The RANDS survey was conducted in English using web and telephone administration. Survey administration mode was determined by the preference of the panelists.
RANDS during COVID-19 included two rounds of data collection (round 1 and round 2). In round 1, NORC invited 8,663 randomly selected panelists to participate in RANDS between June 9, 2020 and July 6, 2020. At the completion of round 1, 6,800 interviews were completed for an overall completion rate of 78.5%. Of the 6,800 completed interviews, 6,390 (94.0%) were completed via web administration and 410 (6.0%) were completed via telephone. In round 2, NORC invited 8,651 of the panelists from round 1 to participate in round 2 of RANDS during COVID-19. Responses were collected from August 3, 2020 to August 20, 2020. At the completion of round 2, 5,981 interviews were completed for an overall completion rate of 69.1%. Of the 5,981 completed interviews, 5,559 (92.9%) were completed via web administration and 422 (7.1%) were completed via telephone. Overall, 5,452 panelists participated in both rounds 1 and 2.
Survey Questions
The questionnaire included questions to assess U.S. adults’ loss of work due to illness with COVID-19, use of telemedicine, and access to healthcare during the COVID-19 pandemic. Questions related to loss of work due to personal or family illness with COVID-19 were asked of those who did not work for pay at a job or business in the previous week. The survey question on RANDS related to this topic included the following:
Loss of Work
Were you unable to work because you or a family member was sick with the coronavirus?
The estimates reported are the percentage unable to work in the last week due to a personal or family member illness with coronavirus among all adults.
Weighting and Estimation
NORC provided sample weights for RANDS during COVID-19 round 1 and round 2 which account for the sample design and are calibrated to U.S. population counts to account for differential nonresponse and under coverage of some groups on the sample frame. The sample weights developed by NORC were based on the panel weights from the AmeriSpeak® Panel and the RANDS survey-specific sampling weights for each round. The panel weights account for the sampling probability from the AmeriSpeak® Panel, calculated as the inverse probability of selection from the NORC National Sample Frame, which are adjusted for non-response and for a subsample of housing units that have a nonresponse follow-up. These panel weights are adjusted using raking to align the panel weights with external population totals obtained from the U.S. Census Bureau Current Population Survey (CPS). From the panel weights, the RANDS survey-specific sampling weights are derived using the probability of selection into RANDS associated with a sampled panel member and an adjustment for nonresponse to the RANDS survey. The RANDS survey-specific sampling weights for each round of RANDS during COVID-19 are raked to general population totals, trimmed for extreme weights, and re-raked to the same population totals to form the final NORC-provided sample weights.
NCHS implemented an additional weighting step to calibrate the NORC-provided weights to the National Health Interview Survey (NHIS), an established core household survey conducted by NCHS. NORC-provided sample weights for round 1 and round 2 were adjusted through an additional post-stratification step in which the marginal totals of RANDS from each round were raked to the marginal population totals of the 2018 NHIS sample adult file (n = 25,417), associated with the following demographic and health characteristics: age, sex, race and Hispanic origin, education, income, Census region, marital status, diagnosed high cholesterol, diagnosed asthma, diagnosed hypertension, and diagnosed diabetes. Characteristics with missing values in RANDS and in NHIS were included in the raking procedure, while characteristics with missing values in one of the surveys were removed for the raking step with an adjustment on the weights for the non-missing values. The calibrated weights were proportionally adjusted to sum to the total number of RANDS respondents for each round (n = 6,800 for round 1, n = 5,981 for round 2). All analyses reported are based on the NCHS calibrated sample weights for the respective round. It is important to note that RANDS during COVID-19 is a probability-sampled panel survey. Although the NORC-provided sample weights are designed to provide nationally representative estimates, the additional calibration further adjusts the RANDS data for differences in health and demographic factors between RANDS and NHIS due to possible differences in response propensities, coverage, and sample variability. The application of sample weights to the data is required to produce results with meaningful population representativeness and to accurately assess the sampling error of statistics based on the survey data.
For each outcome reported, the sample size, percentage estimate, and standard error estimate are shown for round 1 and round 2. Standard error estimates were generated using a Taylor series linearization approach. SUDAAN’s PROC CROSSTAB was used to calculate all estimates. PROC CROSSTAB accounts for the complex survey design variables, including the sampling strata and sampling weights. Missing values were excluded from the reported estimates. Estimates that did not meet the NCHS Data Presentation Standards for Proportions are suppressed (https://www.cdc.gov/nchs/data/series/sr_02/sr02_175.pdfpdf icon).
Each set of estimates is reported for the total overall (at the national level) and by selected factors including demographics and diagnosed chronic conditions. Selected factors included age group (18–44, 45–64, and 65 and over), race and Hispanic origin (white non-Hispanic, black non-Hispanic, other non-Hispanic, Hispanic), sex (male or female), education (high school graduate or less, some college, or Bachelor’s degree or above), urbanization (metropolitan or non-metropolitan), and chronic conditions (one or more chronic conditions, diagnosed diabetes, diagnosed hypertension, or current asthma). For subgroup analyses, sex is identified by the response to the survey item “Please tell us your gender.” Urbanization is assigned by zip code, where metropolitan includes metropolitan and micropolitan areas and non-metropolitan includes all other designations. One or more chronic conditions is defined as a diagnosis of one or more of the following: hypertension, also called high blood pressure; high cholesterol; coronary heart disease; current asthma; chronic obstructive pulmonary disease (COPD), emphysema, or chronic bronchitis; cancer or a malignancy of any kind; or diabetes excluding pre-diabetes and borderline diabetes. Estimates are accompanied by bar charts displaying comparisons within the selected factors between the two rounds.
Comparisons between Rounds
Reported statistical testing results identify statistically significant differences in the experimental estimates between rounds 1 and 2. Since the sampled panelists from round 1 of RANDS during COVID-19 were also invited to participate in round 2, the percentage estimates for the selected indicators are not independent. Of the 6,800 respondents to round 1 and 5,981 respondents to round 2, 5,452 panelists participated in both rounds. To account for the correlation in the repeated measurements, point estimates provided for rounds 1 and 2 were statistically compared using a correlated two-sample t-test. The variance of the difference between point estimates was approximated using the variance of the estimates from rounds 1 and 2 and a correction factor to account for the sample of panelists responding to both rounds. The correction factor incorporated the design effects from both rounds and estimated correlation using the weighted Pearson correlation coefficient. Correction factors were estimated separately for each indicator within subgroups. Correlations between rounds differed by indicator and for categories within subgroups. The resulting test statistic was assumed to be approximately normally distributed. The reported significance is based on the two-sided test with a significance level of 0.05 and does not account for multiple comparisons. Significance testing was not reported in cases where either the round 1 estimate or round 2 estimate were suppressed.
Limitations and Data Use
The information presented here is based on data collected by commercial vendors that maintain groups of respondents, called panels, who agree to participate in surveys, typically in exchange for payment or prizes. Panel surveys can be implemented much quicker and can be less expensive to conduct than typical NCHS surveys that draw new samples from nationally representative frames and collect information by phone, online, or in person. Panel surveys also typically have more sample bias and less accuracy than traditional survey methods.
RANDS is an ongoing NCHS research program designed to investigate whether and, if so, how panels with more biases can be used in conjunction with other, higher quality data collections to increase the scope of information collected, the timeliness of data collection and the sample size, and to expand the scope and granularity of NCHS statistical products. The RANDS was not designed to replace NCHS’ higher quality, core data collections but to be used in conjunction with those surveys especially during periods when standard data collection methods face challenges.
The COVID-19 pandemic offers an opportunity to further explore the collection of new information more quickly than is possible using the Federal Statistical System’s core data collection approaches. Building on past work, NCHS took advantage of the RANDS platform to collect information regarding aspects of the public health emergency not currently being captured in sufficient detail in government or nongovernment surveys, including access to health care for non-COVID-19 conditions. In addition, measurement research that may be instructive for interpreting other Federal and non-Federal surveys during the pandemic will be conducted and calibration research will continue. Experimental national estimates are only provided for selected variables. The other variables are being collected specifically for research purposes. The experimental national estimates for the inability to work due to illness with COVID-19, telemedicine before and during the pandemic, and problems accessing specific types of health care due to the pandemic have been calibrated using information from NCHS’ NHIS in an effort to correct for some portion of the potential bias in the panel relative to the NHIS, as described above. Information on the effect of calibration for past data collections and for this round of data collection will be presented on the RANDS website. As research is underway to both improve the calibration method and understand potential sources of measurement error, estimates may be updated based on the results of that research. These estimates should be considered experimental. On August 28, 2020, NORC provided a revised file for the round 1 respondents and the calibrated weights were updated accordingly. The reported estimates reflect this update. In the event of future updates to the weights or the calibration approach, the change will be documented and tables will be updated accordingly.
Statistically significant differences between experimental estimates from rounds 1 and 2 identified using the reported p-value approach should be interpreted with caution. Small estimated design effects for subgroups may impact the calculated test statistic and the normality assumption may be violated in the case of small sample sizes. The American Statistical Association (ASA) has released a statement on statistical significance based on p-values (https://amstat.tandfonline.com/doi/full/10.1080/00031305.2016.1154108external icon). The ASA’s published guidance indicates that the p-value should not be the only factor used when making decisions about scientific findings. NCHS will release a future methodological report with additional details and results from the correlated two-sample t-test on the RANDS during COVID-19 published indicators.
Note that a statistical test of the difference between the RANDS during COVID-19 estimates using the published estimates and their standard errors and assuming independence between rounds is conservative for positively correlated estimates and may not identify a statistically significant difference that would be identified if the dependence were considered. However, differences between rounds 1 and 2 that are identified as statistically significant under the assumption of independence will be significant when accounting for the correlation. Statistical testing results in the published tables accounted for the correlation.