HS840 Unit 8 Stepped-Wedge Data Analysis Plan Example

Course	HS840 Advanced Health Research and Analysis Methods
Unit	Unit 8
Paper type	Data analysis plan
Length	About 880 words, 6 pages double-spaced
Format	APA 7 student paper
School	Purdue University Global
Program	Doctor of Health Science (DHSc)
Updated	October 2026

Free sample paper for HS840 Unit 8

Fixed Before the First Record Is Pulled: A Unit 8 Data Analysis Plan for a Stepped-Wedge Study of Same-Day Contraceptive Insertion

[Student Name]

Purdue University Global

HS840: Advanced Health Research and Analysis Methods

Unit 8 Data Analysis Plan

[Instructor Name]

[Date]

The network and study are composite; the statistical methods and their sources are real.

What this part is doingStating that the plan is fixed in advance tells the reader why the plan's choices can be trusted later, whatever the results show.

The Estimand

The primary estimand is the difference, on the odds ratio scale, in the probability that a request for an implant or IUD is followed by insertion within 30 days, between clinic-periods with same-day service and clinic-periods without it, among all eligible requests, regardless of whether the patient was offered or accepted same-day insertion. Requests are assigned to the clinic where the request was made, not where the device was placed, so that a patient who travels from a control clinic to an intervention clinic counts toward the control clinic. This is an intention-to-treat estimand at the clinic-period level, matching the decision leadership faces: whether to offer the service, not whether patients use it.

Descriptive Statistics

The first table will report, for each clinic and period, the number of requests, the number and percentage completed within 30 days, the median days from request to insertion, the age distribution of requesting patients, and the share requesting an implant rather than an IUD. A second table will report the same by condition, pooled across clinics. A figure will plot each clinic's completion proportion by period, marking the period in which it crossed over, so that readers can see the raw pattern before any model.

The Primary Model

The outcome is binary for each request: insertion within 30 days or not. The primary analysis uses a generalized linear mixed model with a logit link. Fixed effects are an indicator for whether the clinic was offering same-day service in that period and a categorical effect for each of the seven periods, the standard specification for this design (Hemming et al., 2015). A random intercept for clinic accounts for correlation among requests in the same clinic.

The period effects are not optional: because more clinics are exposed as time passes, any change over calendar time would otherwise be credited to the intervention.

Period effects also soak up the month-five formulary change, provided its impact is roughly uniform across the network. The model will be fitted by maximum likelihood with adaptive quadrature. With only twelve clinics, standard errors from mixed models can be too small, so the test of the intervention effect will use the Kenward-Roger adjustment.

What this part is doingNaming the small-sample correction in advance addresses the most common technical criticism of trials with few clusters before a reviewer raises it.

Checking the Correlation Structure

The simple model assumes that the correlation between requests in the same clinic is the same whether they occur in the same period or a year apart. Methods literature on stepped-wedge designs shows that this assumption can bias the standard error when correlation decays over time (Li et al., 2021). As a prespecified sensitivity analysis, the model will be refitted with an additional random effect for clinic by period. If the intervention estimate or its confidence interval changes materially, both results will be reported and the more conservative interpreted.

Secondary Analyses

Time since crossover. A model replacing the single exposure indicator with categories for the first and later periods after crossover will test whether the effect grows as clinics gain experience, which is plausible for a procedure-based service.

Days to insertion. Among requests completed within 90 days, time from request to insertion will be modeled with a mixed model on the log of days plus one, since this distribution is strongly skewed.

Any method started. The proportion of requests followed by any contraceptive method within 30 days will be modeled like the primary outcome, to show whether gains in insertion displace other methods or add to them.

Counseling experience. The proportion of survey respondents rating all four items of the Person-Centered Contraceptive Counseling scale excellent (Dehlendorf et al., 2021) will be modeled with the same structure. Because response is voluntary, response rates by condition will be reported, and the analysis will be presented as secondary.

Missing Data

Missing records and missing surveys are different problems and are handled differently. Records are not missing in the usual sense; a request without an insertion is a valid outcome of no. The risk is misclassification, addressed by the chart review in the instrument plan. Surveys will be missing for many patients. The plan uses multiple imputation for the counseling outcome, with twenty imputed datasets, using clinic, period, condition, age, method requested and completion status as predictors, under the assumption that response depends on these observed characteristics. A complete-case analysis will be reported alongside.

Hypotheses Mapped to Tests

1. Same-day service increases completion within 30 days: odds ratio for exposure in the primary model, two-sided test at 0.05.

2. The effect grows with time since crossover: comparison of early and later exposure categories.

3. Same-day service shortens time to insertion: exposure coefficient in the log-time model.

4. Same-day service does not reduce top-box counseling ratings: exposure coefficient in the counseling model, interpreted against a prespecified margin of 5 percentage points.

What this part is doingMapping each hypothesis to its exact test keeps the analysis from expanding after the data are seen.

What the Plan Does Not Allow

No subgroup analyses beyond age group and method type will be reported as findings; any others will be labeled exploratory. The plan will be posted in a public registry before the first crossover.

References

Dehlendorf, C., Fox, E., Silverstein, I. A., Hoffman, A., Campora Pérez, M. P., Holt, K., Reed, R., & Hessler, D. (2021). Development of the Person-Centered Contraceptive Counseling scale (PCCC), a short form of the Interpersonal Quality of Family Planning care scale. Contraception, 103(5), 310-315. https://doi.org/10.1016/j.contraception.2021.01.008

Hemming, K., Haines, T. P., Chilton, P. J., Girling, A. J., & Lilford, R. J. (2015). The stepped wedge cluster randomised trial: Rationale, design, analysis, and reporting. BMJ, 350, h391. https://doi.org/10.1136/bmj.h391

Li, F., Hughes, J. P., Hemming, K., Taljaard, M., Melnick, E. R., & Heagerty, P. J. (2021). Mixed-effects models for the design and analysis of stepped wedge cluster randomized trials: An overview. Statistical Methods in Medical Research, 30(2), 612-639. https://doi.org/10.1177/0962280220932962

Reading the HS 840 Unit 8 instructions

The eighth unit generally asks learners to write a statistical analysis plan for their proposed study, specifying the estimand or primary comparison, descriptive statistics, the primary model, secondary models, handling of missing data and the test for each hypothesis. Six to eight pages is common. Each choice should be justified by the design, the outcome's type and its distribution. For clustered designs, instructors expect the model to account for clustering and, in stepped-wedge studies, for calendar time. Missing data deserve their own section, with methods matched to the type of missingness. Many prompts ask for a list linking hypotheses to tests, and some ask learners to state which results will be treated as exploratory.

How the HS 840 Unit 8 example is put together

The estimand section defines the comparison, population and handling of patients who travel between clinics. Descriptive tables and a clinic-by-period figure are specified before any model. The primary model is described in words, with the highlighted sentence explaining why period effects are required. A section on correlation structure adds a clinic-by-period random effect as a sensitivity analysis. Four secondary models follow, each with its model. The missing data section separates misclassified records from missing surveys and specifies twenty imputations with named predictors. A numbered list maps four hypotheses to tests, including a 5-point margin for counseling ratings. A final section limits subgroup claims and commits to registration.

HS840 Unit 8 rubric: what earns full points

Analysis plans are typically graded on alignment of methods with the question and design, correctness of the statistical approach, treatment of missing data, transparency and prespecification, and clarity. Alignment often carries the most weight; a plan that ignores clustering or calendar time in a stepped-wedge design loses heavily. Correctness rewards models suited to the outcome type, such as logistic models for binary outcomes. Missing data points depend on matching the method to the mechanism and reporting a comparison analysis. Prespecification earns credit through a hypothesis-to-test map and limits on subgroup claims. Clear writing that a statistician could implement without asking questions completes the score. Instructors also check that every hypothesis in the proposal appears in the plan with a named test.

Common HS840 Unit 8 mistakes, and how to avoid them

The most frequent error is planning a simple comparison of proportions that ignores clustering, which overstates precision. Another is omitting period effects in a stepped-wedge analysis. Some learners delete incomplete cases without considering bias. Others list many secondary tests with no plan for interpreting them. Plans that describe software steps instead of models confuse the reader. Treating satisfaction scores as normally distributed when they cluster at the top is another common slip. If your study has a different outcome or design, such as a continuous measure in a two-arm trial, send your question and design, and a plan can be prepared that ties each analysis to them. Leaving the estimand undefined makes it impossible to tell what the primary result actually means.

Get an HS840 Unit 8 example written to your instructions

Analysis plans rest on everything before them: the design, measures and hypotheses from earlier units, plus the Unit 8 prompt, the rubric and any software your course requires. With those, the plan defines its estimand, derives each test from design and data, and treats missing values by mechanism with a sensitivity check. 24-48h, and free for a first custom sample. The paper above is an original model document written by our desk, not a submitted student paper and not an official Purdue University Global document.

More HS840 papers and DHSc samples

HS840 Unit 8 questions, answered

What is an estimand, and do I need one?

It is a precise statement of the quantity a study aims to estimate: population, comparison, outcome, and how events such as stopping the intervention are handled. Many doctoral methods courses now expect one, following the ICH E9(R1) guidance on trials. The sample's estimand includes all enrolled clinicians regardless of use, which decides the primary analysis before any data exist.

Is deleting incomplete cases ever acceptable?

Occasionally, when very little is missing and the missingness plausibly has nothing to do with the outcome. More often it biases results and wastes information. The sample uses methods that retain partial data and states its missing-at-random assumption. If your plan uses complete cases, justify it with the expected missingness and add a sensitivity analysis.

Where can I find a free HS840 Unit 8 sample paper?

The complete Unit 8 data analysis plan is posted here, specifying a mixed logistic regression with period terms for a stepped-wedge study of same-day contraception.

Why are period effects required in a stepped-wedge analysis?

More clinics are exposed as time passes, so any change over calendar time, such as a formulary update, would otherwise be credited to the intervention.

How does the plan handle missing survey responses?

With multiple imputation using clinic, period, condition, age, method requested and completion status as predictors, with complete-case results reported alongside.