PHLT 8064 Week 8 Analysis Plan Example

Course	PHLT 8064 Dissertation Seminar III
Week	Week 8
Paper type	Analysis plan
Length	About 1,040 words, 6 pages
Format	APA 7 student paper
School	Walden University
Program	PhD in Public Health
Updated	October 2026

Free sample paper for PHLT 8064 Week 8

Analysis Plan by Research Question: Estimands, Models, a Causal Diagram and Table Shells for a Registry Cohort on Drive Time and Documented Diagnostic Hearing Evaluation, Written Before Any Record Is Released

Student Name

Walden University

PHLT 8064: Dissertation Seminar III

Instructor Name

Month Day, Year

What this page is doingThe title lists the four parts each question receives and stresses the timing, because a plan written after the data arrive is a description, not a plan.

How the Plan Is Organized

This plan is organized by research question rather than by statistical technique. Each question receives the same four parts in the same order: the estimand stated in words, the model that estimates it, the covariates and how they were chosen, and a table shell showing exactly what will be reported. Organizing this way makes a missing part obvious. Lundberg et al. (2021) argue that quantitative research should state its target quantity before its method, because the estimand is what links a statistical result to the question a reader cares about. Everything below is fixed before the registry extract is released.

What this page is doingCommitting to one structure for every question lets a reader see at once whether any question has been shortchanged.

Research Question 1

Is a longer road trip to the closest site offering pediatric diagnostic audiology linked to lower odds that a referred infant in the composite state has a diagnostic evaluation on record by the age of three months?

Estimand

The odds of documented evaluation by 90 days for infants with a given drive time, compared with the odds for infants whose drive time is 30 minutes shorter, among all eligible infants, holding the confounders in the causal diagram fixed. A banded version compares each drive-time band with the under-30-minute band.

Model

Logistic regression of documented evaluation on drive time, fitted with generalized estimating equations that cluster infants within birth hospitals and use an exchangeable working correlation. Infants born at the same hospital share screening equipment, staff and referral routines, so their outcomes are unlikely to be independent. Generalized estimating equations give valid standard errors for the population-averaged association under that clustering (Zeger & Liang, 1986).

Covariates

Covariates are chosen from a directed acyclic graph printed in the appendix, not by stepwise selection or by changes in the estimate. The graph places drive time, the outcome and each recorded variable according to their assumed causal relationships. The adjustment set, chosen by the rules Greenland et al. (1999) set out, is the minimal set that blocks every backdoor path from drive time to evaluation without conditioning on a variable caused by drive time. The resulting set is maternal age, maternal education, payer, maternal race and ethnicity, and gestational age. Birth hospital is handled by clustering rather than adjustment, because hospital choice is partly caused by where families live and adjusting for it would remove part of the association under study.

Table 1

Shell for Research Question 1: Odds of Documented Evaluation by 90 Days

Drive time	Infants	Evaluated by 90 days, percent	Unadjusted odds ratio (95% CI)	Adjusted odds ratio (95% CI) \|
Per 30 minutes, continuous	All eligible	To be entered	To be entered	To be entered \|
Under 30 minutes	To be entered	To be entered	1.00 (reference)	1.00 (reference) \|
30 to 59 minutes	To be entered	To be entered	To be entered	To be entered \|
60 to 89 minutes	To be entered	To be entered	To be entered	To be entered \|
90 minutes or more	To be entered	To be entered	To be entered	To be entered \|

Note. Adjusted estimates use the covariate set derived from the causal diagram; CI = confidence interval.

Research Question 2

Is the link between drive time and a documented evaluation by three months different for infants whose births Medicaid paid for than for infants with private coverage?

Estimand

The odds ratio for documented evaluation per 30 minutes of drive time within each payer group, and the difference between them on the multiplicative and additive scales.

Model

The first question's model, extended with an interaction between drive time and payer. Results will be reported as Knol and VanderWeele (2012) recommend: odds ratios within each payer stratum, the ratio of odds ratios for the multiplicative interaction and, on the additive scale, RERI, the relative excess risk owed to interaction, which is more relevant to where a program should direct help.

Covariates

The same adjustment set, with payer removed from the set and entered as the modifier.

Table 2

Shell for Research Question 2: Drive Time Association by Payer

Measure	Medicaid	Private coverage \|
Infants, n	To be entered	To be entered \|
Evaluated by 90 days, percent	To be entered	To be entered \|
Adjusted odds ratio per 30 minutes (95% CI)	To be entered	To be entered \|
Ratio of odds ratios (95% CI)	To be entered	Reference \|
Relative excess risk due to interaction (95% CI)	To be entered	Reference \|

Note. Infants with other or unknown payer are reported descriptively and excluded from this model.

Missing Data

Maternal education is expected to be missing for a small share of birth certificates, and missingness may be related to payer and age. Complete-case analysis would drop those infants and could bias estimates if missingness relates to the outcome. The plan commits to multiple imputation using chained equations and 20 completed datasets, an imputation model that includes the outcome, drive time and all covariates, and estimates combined by Rubin's rules (Rubin, 1987). Sterne et al. (2009) warn that imputation models omitting the outcome bias associations toward the null, which is why the outcome is included. Exposure and outcome are not imputed: infants without a geocoded address are excluded as described in the sampling plan, and infants with unknown status are handled by sensitivity analysis.

What this page is doingExplaining why the outcome enters the imputation model anticipates the question a careful committee member is most likely to ask.

Sensitivity Analyses

These analyses are labeled as sensitivity analyses in advance so that none can be promoted to the main result after the fact.

1. Unknown statuses recoded as not evaluated, then as evaluated.

2. Outcome window extended to 180 days.

3. Restriction to infants whose later registry records show the same address as the birth certificate.

4. Restriction to births after the registry's 2018 software change.

5. An E-value for the main adjusted estimate.

What the Plan Leaves Open

The plan does not specify what would count as a meaningful association for the program; that judgment will be made with program staff and reported as theirs. It also leaves the exact routing software version to be recorded at the time drive times are computed. Any analysis added after the data are released will be reported as exploratory and kept out of the abstract.

References

Greenland, S., Pearl, J., & Robins, J. M. (1999). Causal diagrams for epidemiologic research. Epidemiology, 10(1), 37-48. https://doi.org/10.1097/00001648-199901000-00008

Knol, M. J., & VanderWeele, T. J. (2012). Recommendations for presenting analyses of effect modification and interaction. International Journal of Epidemiology, 41(2), 514-520. https://doi.org/10.1093/ije/dyr218

Lundberg, I., Johnson, R., & Stewart, B. M. (2021). What is your estimand? Defining the target quantity connects statistical evidence to theory. American Sociological Review, 86(3), 532-565. https://doi.org/10.1177/00031224211004187

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. Wiley.

Sterne, J. A. C., White, I. R., Carlin, J. B., Spratt, M., Royston, P., Kenward, M. G., Wood, A. M., & Carpenter, J. R. (2009). Multiple imputation for missing data in epidemiological and clinical research: Potential and pitfalls. BMJ, 338, b2393. https://doi.org/10.1136/bmj.b2393

Zeger, S. L., & Liang, K.-Y. (1986). Longitudinal data analysis for discrete and continuous outcomes. Biometrics, 42(1), 121-130. https://doi.org/10.2307/2531248

What the PHLT 8064 Week 8 instructions ask for

Week 8 asks candidates to write the analysis plan for their dissertation in enough detail that the analysis could be carried out without further decisions. Faculty expect the plan to proceed question by question, linking each to a statistical model appropriate to the outcome's type and the data's structure. Covariate selection must be justified; many instructors now expect a causal diagram rather than stepwise methods. Missing data, clustering and sensitivity analyses should be addressed explicitly, and table shells showing what will be reported are often required. The plan should be written before data access and say so. Length varies from four to seven pages, and sources on causal inference, regression for clustered data and missing data methods are usual. Programs differ on software detail, so follow your template.

Inside the PHLT 8064 Week 8 example

The paper opens by explaining why it is organized by research question, citing Lundberg and colleagues on stating the estimand first. The first question receives an estimand phrased as odds per 30 minutes, a model using generalized estimating equations because infants share birth hospitals, and a covariate set derived from a directed acyclic graph following Greenland and colleagues; birth hospital is handled by clustering rather than adjustment, and the reason is given. Table 1 is a shell for that question. The second question adds a product term and reports within-stratum, multiplicative and additive measures, with Table 2 as its shell. Sections on missing data and sensitivity analyses follow, the latter numbered and labeled in advance. A last section states what remains undecided and how post hoc analyses will be marked exploratory.

Reading the PHLT 8064 Week 8 rubric

Rubrics for the analysis plan reward correctness and completeness first. Each research question must map to a model suited to a binary outcome and clustered data, and the sample earns that credit by explaining why generalized estimating equations fit its hospital clustering. Covariate selection by causal diagram scores higher than selection by significance, and graders check that variables on the causal path are not adjusted for. The interaction plan is assessed against current reporting guidance. Missing data handling earns marks when the method and its key settings, the number of imputations and the contents of the imputation model, are stated. Prespecified sensitivity analyses demonstrate rigor. Table shells score for showing exactly what will be reported. Clear writing and accurate citations of methods literature are weighed in the final rows.

PHLT 8064 Week 8 help: mistakes that cost marks

Analysis plans most often fall short by naming techniques without connecting them to questions. A paragraph that says logistic regression will be used leaves unclear which question it answers, what the comparison is and what will be adjusted for. Covariate lists chosen because they are available, or because they were significant in a bivariate screen, invite committee challenge. Clustering is easy to forget in health system data, and ignoring it produces standard errors that are too small. Missing data plans that default to complete-case analysis without a reason are another weak spot, as is an imputation model that omits the outcome. Write the estimand in words before you choose a model. If your study involves survey weights, survival time or counts, a custom analysis plan fitted to that design is available free through the form on this page.

Get a PHLT 8064 Week 8 example written to your instructions

Send your research questions, the variables your data will provide and the Week 8 plan instructions with the rubric. The plan works question by question, states each estimand in words, justifies covariates with a causal diagram and supplies empty table shells for the results. There is no charge for a first plan, delivered in 24 to 48 hours. Its hospitals and variables are hypothetical.

More PHLT 8064 papers and PhD in Public Health samples

PHLT 8064 Week 8 questions, answered

What is an estimand, and why state it in words?

An estimand is the precise quantity a study sets out to estimate, such as the difference in odds of evaluation between infants in the farthest and nearest drive-time bands. Stating it in words before naming a model forces clarity about the comparison and the population. The example gives one per research question. Committees increasingly ask for estimands because a model can be run correctly and still estimate something nobody intended.

Why use a directed acyclic graph to choose covariates?

Because it makes the reasons for adjustment explicit. A directed acyclic graph draws the assumed causal relationships among exposure, outcome and other variables, which shows which variables confound the association and which lie on its causal path and should not be adjusted for. The example prints its graph so a committee can challenge the assumptions directly. Selecting covariates by statistical significance offers no such transparency.

Where can I find a free PHLT 8064 Week 8 sample paper?

This page holds the complete Week 8 analysis plan for a registry cohort on drive time and infant hearing evaluation, with an estimand, model, covariate set and table shell for each research question. Missing data and sensitivity plans are included, and every method is cited to a real source.

Why use generalized estimating equations here?

Infants born at the same hospital share staff, equipment and referral habits, so their outcomes are not independent. Generalized estimating equations account for that clustering and give valid standard errors for the population-averaged association, which is the quantity the sample's first question asks about.

Why include the outcome in the imputation model?

Leaving the outcome out makes the imputed values unrelated to it, which pulls estimated associations toward zero. Sterne and colleagues warned about this in their guidance on multiple imputation, and the sample follows that advice by imputing maternal education with the outcome, exposure and all covariates in the model.