PUBH 8048 Week 9 Analysis Plan Example

An Analysis Plan That Follows a Claim About Levels: A Multilevel Model for Social Capital and Shutoff Preparedness Among Older Adults Who Live Alone With Powered Equipment, With Predictions Named Before Data Arrive

Student Name

Walden University

PUBH 8048: Research Applications of Theories, Models, and Conceptual Frameworks in Public Health

Instructor Name

Month Day, Year

An Analysis Plan That Follows a Claim About Levels: A Multilevel Model for Social Capital and Shutoff Preparedness Among Older Adults Who Live Alone With Powered Equipment, With Predictions Named Before Data Arrive

Why the Framework Requires a Multilevel Model

The social capital framework used in this study places resources at two levels: in a person's network of ties and in the cohesion of the neighborhood where the person lives (Portes, 1998; Kawachi et al., 1997). Residents of the same neighborhood share its cohesion and are likely to resemble one another in other ways, so their preparedness is not independent. An ordinary regression that treated each resident as independent would understate uncertainty and would blur the distinction between a person's own ties and the neighborhood's character. A multilevel model, with residents nested within neighborhoods, estimates both kinds of association at their proper levels. Neighborhood effects research has been urged to specify what about neighborhoods matters and to analyze area and individual factors at their own levels (Diez Roux, 2001), which is what this plan does.

Data Structure

Participants are residents aged 65 and older who live solo and use electrically powered medical equipment in Calder County, the course's composite setting, recruited within census block groups in parts of the county where the utility cuts power during fire weather. The target is 420 residents in about 60 block groups, roughly seven per block group. Neighborhood cohesion is measured by asking all participants, plus a random sample of other adult residents in each block group, about neighbors' willingness to help and trust, then averaging within block groups.

The Model

The primary outcome is the preparedness index, a count from 0 to 3. The model is a multilevel Poisson regression with a random intercept for block group. Individual-level predictors are bonding and bridging tie scores and a binary indicator of any linking tie, excluding utility program enrollment, which is part of the outcome. The block-group-level predictor is mean neighborhood cohesion. Covariates are age, sex, income category, a self-rated health measure and type of equipment, included because the conceptual diagram shows income and health as possible causes of both ties and preparedness.

Named Predictions

Before data arrive, the plan states what the framework predicts, so results can be judged against it:

1. Each additional practical tie will be associated with higher preparedness, with bonding ties showing the largest association for backup power and bridging ties for written plans.

2. Residents with any linking tie will be more likely to be fully prepared.

3. After accounting for individual ties, higher neighborhood cohesion will still be associated with higher preparedness.

4. Part of the association between cohesion and preparedness will run through individual ties, so the cohesion coefficient will shrink when tie scores are added.

5. The variance in preparedness between block groups will be modest, as most variation lies among individuals.

Testing the Path From Cohesion to Ties

Prediction 4 requires a mediation analysis: comparing the cohesion coefficient in models with and without individual tie scores and estimating the indirect path from cohesion to ties to preparedness. Because data are cross-sectional, this tests consistency with the framework's claim, not causation.

Sensitivity Analyses

Three sensitivity analyses address known weaknesses. First, block groups with fewer than four participants will be combined with adjacent ones to stabilize cohesion estimates. Second, the model will be rerun with fully prepared status as a binary outcome. Third, linking ties will be redefined to exclude any contact initiated by the utility, testing whether the association reflects institutions reaching residents rather than residents reaching institutions.

Power

With 420 residents in 60 clusters and an assumed intraclass correlation of 0.05, the design effect is about 1.3, giving an effective sample of roughly 320 for individual-level associations. That is adequate for moderate associations but limited for detecting small neighborhood-level effects, which the report will state.

Handling Missing Data

Some residents will not complete every item, especially the network questions, which ask people to count others and can be tiring. If fewer than 10% of values are missing on any variable, the analysis will use complete cases and report how many were dropped. If more are missing, multiple imputation will be used, with the imputation model including all variables in the analysis model and block group. Missing cohesion values from the mail sample do not threaten the individual analysis, since cohesion is aggregated, but block groups with fewer than eight cohesion respondents will be flagged.

Software and Checks

Models will be fit in standard statistical software with adaptive quadrature for the random intercept. Before interpreting estimates, the analyst will check for overdispersion in the preparedness count, since a Poisson model assumes the variance equals the mean; if overdispersion is present, a negative binomial or ordinal model will replace it, and the change will be reported. Residuals will be examined by block group to identify neighborhoods the model fits poorly.

Exploratory Analyses

Two exploratory analyses are labeled as such in advance. One examines whether the association between bonding ties and preparedness differs for residents with family nearby versus far away. The other compares residents who use oxygen with those using other devices, since oxygen users may face sharper deadlines when power fails. Neither will be used to support the main conclusions; both may suggest questions for later work.

Reporting

Estimates will appear in the order of the predictions, with each estimate shown alongside the prediction it tests, so readers can see where the framework held and where it did not. A prediction that fails will be reported as prominently as one that holds. The intraclass correlation will be reported to show how much preparedness varies between neighborhoods.

Limits Set in Advance

The plan does not test whether preparedness protects residents during actual shutoffs, which would require outcome data on health events that are rare. Nor does it estimate causal effects of ties; the design is observational. National estimates show that hundreds of thousands of people whose home medical devices need electricity live in the community, most often older adults (Molinari et al., 2017), so even an observational description of which ties go with preparedness can guide outreach.

References

Diez Roux, A. V. (2001). Investigating neighborhood and area effects on health. American Journal of Public Health, 91(11), 1783-1789. https://doi.org/10.2105/AJPH.91.11.1783

Kawachi, I., Kennedy, B. P., Lochner, K., & Prothrow-Stith, D. (1997). Social capital, income inequality, and mortality. American Journal of Public Health, 87(9), 1491-1498. https://doi.org/10.2105/AJPH.87.9.1491

Molinari, N. A. M., Chen, B., Krishna, N., & Morris, T. (2017). Who's at risk when the power goes out? The at-home electricity-dependent population in the United States, 2012. Journal of Public Health Management and Practice, 23(2), 152-159. https://doi.org/10.1097/PHH.0000000000000345

Portes, A. (1998). Social capital: Its origins and applications in modern sociology. Annual Review of Sociology, 24, 1-24. https://doi.org/10.1146/annurev.soc.24.1.1