MPH 550 Week 6 Survey Design Proposal Example

Big Enough to Answer the East Side Question: Designing a Two-Year County Survey of Diabetes, Activity and Access to Care

[Student Name]

University of Phoenix

MPH/550: Public Health Statistics

Week 6 Assignment

[Instructor Name]

[Date]

The county survey, its sample sizes, budget, response assumptions and timeline are composites written for a model paper; research findings come from the sources cited.

The county's first household survey found diagnosed diabetes at 15.2% on the east side and 11.2% elsewhere, but the comparison was inconclusive: the 95% interval for the difference ran from about minus 2 to plus 10 points, and power to detect a real 4-point gap was only about a quarter. The health board asked for a survey that could settle the question. This paper proposes that design.

Objectives

The primary objective is to estimate the difference in diagnosed diabetes prevalence between east side adults and adults in the rest of the county, with enough precision to detect a 4-point difference. Secondary objectives are to estimate physical inactivity, blood pressure diagnosis, insurance coverage and usual source of care in both areas, and to track change between waves.

Why a Survey

Other designs were considered. Clinic records would miss adults without a usual source of care, who are more common on the east side. Modeled PLACES estimates cannot separate neighborhoods within the county. Only a population survey with an area identifier can compare the two areas directly with known precision, which is why the board's question calls for one.

Hypotheses

The null hypothesis states that diabetes prevalence is equal in the two areas; the alternative states that it differs. The test is two-sided, because a finding in either direction would matter for planning.

Target Population

The target population is adults aged 18 and older living in households in the county. People in institutions, such as prisons and nursing homes, and people without a fixed address are excluded from the household frame; the proposal notes this gap and recommends separate outreach studies for them.

Sampling Frame

The frame is a list of residential mailing addresses, which covers nearly all households regardless of telephone type. A BRFSS pilot comparing a mail survey of sampled addresses with telephone sampling found the address approach could achieve higher response in low-response states, reached households without landlines and offered cost savings (Link et al., 2008).

Sampling Design

Addresses will be stratified into two areas, east side addresses and all other county addresses, and sampled at different rates so each area yields the same number of completed interviews. Within each sampled household, one adult will be selected at random using the next-birthday method to avoid over-representing whoever answers first.

Sample Size

To detect prevalence of 15% versus 11% with a two-sided alpha of 0.05 and 80% power, the standard formula for two proportions requires about 1,110 completed interviews per area. With a design effect of 1.5 from unequal weights, the target rises to about 1,660 per area, or 3,320 in total.

Making It Affordable

One annual wave of that size exceeds the department's budget. The proposal instead runs two annual waves of 1,700 completed interviews each, 850 per area per year, and pools them for the primary comparison. Pooling reaches the target while producing yearly county estimates with margins of about 2.5 points for the whole county.

Checking the Assumptions Behind the Size

The sample size depends on the expected prevalences, which come from the first survey and carry uncertainty. If the true difference is 5 points, power rises above 90%; if it is 3 points, power falls to just under 60%. The proposal reports these scenarios so the board understands that the design is sized for a difference it considers meaningful, not for any difference at all.

Response Assumptions

Assuming a 30% response rate among eligible sampled addresses and about 10% of addresses vacant or ineligible, each wave needs about 6,300 sampled addresses, weighted toward the east side where response was lower in the first survey.

Measures

Questions will come from the BRFSS core, including the diabetes, physical activity, blood pressure and insurance items, so results can be compared with state and national estimates. A systematic review judged BRFSS estimates in line with other self-report surveys but further from surveys that take physical measurements, and it found the depth of validation uneven across health topics (Pierannunzi et al., 2013). The proposal acknowledges that self-reported diagnosed diabetes misses undiagnosed cases.

Translation and Testing

The questionnaire will be professionally translated into Spanish and back-translated. Staff will conduct cognitive interviews with 12 English-speaking and 12 Spanish-speaking residents to check that questions are understood as intended, then pilot the full process with 200 addresses.

Contact Procedures

Contacts follow trial evidence. A Cochrane review of postal questionnaire trials found that monetary incentives roughly doubled the odds of response, with an odds ratio of 1.87; unconditional incentives, shorter questionnaires and a second copy at follow-up also helped (Edwards et al., 2009). Five dollars in the first envelope buys more responses than any reminder. Each household will receive a letter with a five-dollar bill and a web link, a reminder postcard, a paper questionnaire with a return envelope and a final reminder.

Nonresponse Bias

Response rates alone predict nonresponse bias poorly; what matters is whether the survey topic relates to who responds (Groves, 2006). A health survey may attract healthier or more engaged residents. The proposal will compare respondents with census characteristics, compare early and late responders and use follow-up of a subsample of nonrespondents by phone where numbers can be matched.

Weighting

Weights will adjust for selection probabilities, including the stratum sampling rates and the number of adults in each household, then be calibrated to census totals by age, sex, ethnicity and area.

Analysis Plan

The primary analysis compares weighted diabetes prevalence between areas using design-based standard errors, reporting the difference and 95% confidence interval. A secondary logistic regression will adjust for age, sex and income to ask whether any difference persists. All analyses are specified now, before data collection, and additional analyses will be labeled exploratory.

Secondary Analyses

Secondary objectives will be analyzed the same way, with each outcome compared between areas and reported with its interval. Because several secondary outcomes will be tested, results will be presented as descriptive support for the primary finding, and no single secondary result will be used to claim a difference between areas. Change between waves will be reported for the county as a whole.

Data Quality

Staff will double-enter a 10% sample of paper questionnaires, check web responses for straight-lining and out-of-range values and review item nonresponse by question.

Ethics

The proposal will go to an institutional review board. The survey is voluntary and anonymous; responses are stored without names, and results are reported only in aggregate. Respondents receive information on local diabetes screening regardless of their answers.

Budget and Timeline

Each wave costs about $118,000, mainly printing, postage, incentives, data entry and analyst time, for about $236,000 over two years. Pilot and cognitive testing take three months, each wave's field period ten weeks and the pooled analysis two months after the second wave.

Limitations

The design excludes people without fixed addresses and institutional residents. Self-report misses undiagnosed diabetes. Pooling assumes prevalence changes little between waves. Nonresponse bias may remain after weighting.

Conclusion

The proposed survey is sized to answer the question the first survey could not. A power calculation sets the target, pooling two waves makes it affordable, an address-based frame reaches all households, validated items allow comparison, contact procedures follow trial evidence and a prespecified analysis protects against searching for results.

References

Edwards, P. J., Roberts, I., Clarke, M. J., DiGuiseppi, C., Wentz, R., Kwan, I., Cooper, R., Felix, L. M., & Pratap, S. (2009). Methods to increase response to postal and electronic questionnaires. Cochrane Database of Systematic Reviews, 2009(3), Article MR000008. https://doi.org/10.1002/14651858.MR000008.pub4

Groves, R. M. (2006). Nonresponse rates and nonresponse bias in household surveys. Public Opinion Quarterly, 70(5), 646-675. https://doi.org/10.1093/poq/nfl033

Link, M. W., Battaglia, M. P., Frankel, M. R., Osborn, L., & Mokdad, A. H. (2008). A comparison of address-based sampling (ABS) versus random-digit dialing (RDD) for general population surveys. Public Opinion Quarterly, 72(1), 6-27. https://doi.org/10.1093/poq/nfn003

Pierannunzi, C., Hu, S. S., & Balluz, L. (2013). A systematic review of publications assessing reliability and validity of the Behavioral Risk Factor Surveillance System (BRFSS), 2004-2011. BMC Medical Research Methodology, 13, Article 49. https://doi.org/10.1186/1471-2288-13-49