NR-585AT · Week 5 of 8 · Baseline, analysis and the numbers you will defend

NR-585AT Week 5 Baseline and Analysis: How to Write It

The short answer

Nothing exposes a weak plan faster than a director asking, at the end of a presentation, how you will know whether it worked. The unprepared answer is that outcomes will be monitored. The prepared answer names the baseline period, the metric with its denominator, the comparison, the analysis that will be run and what size of change would count as real rather than as noise. NR-585AT Week 5 is where those commitments get written down before any data exists, which is the only point at which they are method rather than convenient hindsight.

Your section may print this as NR 585AT or NR585AT; it is the same course. Chamberlain publishes no syllabi outside Canvas. The placement here is our teaching judgment from the course's catalog arc; your section's rubric decides what your week actually asks.

NR 585AT Week 5 grading scale at Chamberlain, the criterion levels this assessment is scored on, from Chamberlain Tutors
How Chamberlain grades NR 585AT Week 5, visualized by Chamberlain Tutors.

What NR-585AT Week 5 asks for

An analysis plan answers four questions in order, and writing it in that order prevents most of the trouble. What will you describe? What will you test? What assumptions does each test carry and how would you check them? And what would count as an answer, stated in advance?

Descriptive statistics come first because they are how a reader meets the data. Means and standard deviations for interval and ratio variables that are roughly symmetric, medians and interquartile ranges where the distribution is skewed, counts and percentages for categorical variables. Skew is the rule rather than the exception in operational data: length of stay, wait times, contact attempts and turnaround times all pile up at one end and stretch out into a long tail, and a plan promising means for every variable has not looked at the shape of what it will collect. In departmental reporting this matters practically as well as statistically, because a mean turnaround time that includes three catastrophic outliers describes a month nobody experienced.

Inferential choices follow from the level of measurement you set in the previous stage and from the structure of the comparison. Two independent groups with a continuous outcome point toward a t test; the same measure taken twice from the same units points toward a paired test. More than two groups point toward analysis of variance. Two categorical variables point toward a chi-square test. Several predictors of one continuous outcome point toward multiple regression. Two continuous variables point toward correlation, with the standing reminder that correlation describes association and never establishes cause. Non-parametric alternatives exist for all of these and are the right choice when assumptions fail, which in departmental samples they often do.

Baseline deserves separate attention because leadership work lives or dies on it. A baseline is not a single number from last month; it is a defined period long enough to show ordinary variation, described with both a central value and a measure of spread. Without spread you cannot tell a real shift from a normal fluctuation, which is how organizations end up celebrating improvements that were always within the usual range. Say how long your baseline period is, why that length, and what ordinary variation looked like inside it.

The deliverable at this stage is usually an analysis and data management section of roughly 1,000 to 1,400 words, sometimes with a table pairing each question with its planned analysis. Expect a rubric row on data handling and protection alongside the statistics themselves.

The NR-585AT Week 5 method, step by step

Six moves for writing an analysis plan a decision maker could hold you to.

  1. Tabulate each question against its variables and their levels

    One row per question: variables involved, level of measurement of each, number of groups, and whether observations are independent or repeated. Most of test selection falls out of that table before any statistical reasoning happens.

  2. Define the baseline period and characterize its variation

    State the length, the reason for that length, the central value and the spread. Twelve months captures seasonal pattern; three months may not. A baseline reported without spread cannot support any later claim that something changed.

  3. Select each test from the structure and name its alternative

    Choose the parametric test that fits, then name the non-parametric alternative you would use if assumptions fail, along with the condition that would switch you between them. Writing both demonstrates reasoning that a single test name conceals.

  4. State assumptions and how each would be examined

    Normality, equality of variance, independence of observations, adequate expected cell counts. Say how you would check each: histograms, a formal test, an inspection of counts. This paragraph separates a plan from a list of test names.

  5. Commit to a meaningful difference before you have any data

    Give the alpha level, and more importantly say what size of change would matter operationally. Six minutes off a turnaround time may be detectable and irrelevant. Naming the threshold in advance is what stops a small statistical win from being presented as a success.

  6. Write the cleaning, missing data and storage rules now

    How incomplete records are handled, how out-of-range values are checked, who has access, how identifiers are separated from the analytic file, how long records are kept. Decisions made before data exists are method; the same decisions made afterwards are choices no reader can verify.

A layout and word budget for an analysis and data section

Our frame for an analysis section of roughly 1,200 words. It is our own outline rather than anything the university issues, and your week's rubric outranks it wherever they disagree.

SectionWhat belongs in itWord target
Baseline definitionPeriod length with a reason, central value, measure of spread, and what ordinary variation looked like.180 to 220
Preparation and cleaningEntry and range checks, how errors are detected, who verifies, and any double-entry step.140 to 180
Descriptive planWhich summary statistic applies to which variable, with the variables you expect to be skewed named.170 to 210
Inferential plan by questionEach question paired with a named test, its alternative, and the reason the test fits the data structure.250 to 300
Assumptions and thresholdsAssumptions with their checks, the alpha level, and the change that would count as operationally meaningful.200 to 240
Missing data and stewardshipHandling rules for incomplete records, storage location, access control, retention and destruction.190 to 230

Evidence craft for analysis plans

Cite a methods or statistics source when you name a test. Test selection is a methodological claim like any other, and one citation at first mention converts a preference into a supported decision.

Keep statistical language exact. Significant means a result was unlikely under the null hypothesis, not that it was large or important, and a non-significant result is not evidence of no difference. In a research methods course this precision is graded directly, and in a leadership setting it is the difference between an honest report and an overstated one.

Ask for effect sizes alongside every test. Say which effect size measure you would report and why it helps a decision maker. A plan reporting only p values produces an output nobody can act on and nobody can price.

Keep the analysis consistent with the design you already described. If your comparison is between two units at one point in time, a paired test has no place in the plan. Internal contradictions between sections are among the easiest errors for a grader to catch, because both sections are in the same document.

Handle subgroup analysis honestly. Examining many subgroups until one reaches significance produces findings that will not replicate. Name in advance any subgroup you intend to examine and say why it was specified beforehand, particularly where equity across populations is the reason.

Write de-identification as procedure rather than promise. Say how the link between identifiers and study numbers is held, who holds it, and when it is destroyed. Data will be kept confidential is an intention; a separated key file with named access is a method.

Five mistakes that cost points in this week's territory

  • A test that does not fit the data. Proposing a t test for a categorical outcome is the most common mismatch and it usually traces back to a variable whose level of measurement was never stated.
  • A baseline of one number. Without a period and a spread, no later comparison can distinguish a change from ordinary fluctuation.
  • Assumptions never mentioned. A plan naming three tests and no assumptions has skipped the part that demonstrates understanding.
  • Statistical significance treated as importance. Without an effect size and a stated operational threshold, the plan cannot say what result would justify the resource.
  • Missing data ignored. Every operational dataset has gaps, and a plan that never mentions them is describing an idealized study rather than the one proposed.

Before you submit

  • The baseline has a stated period, a reason for its length, a central value and a spread
  • Each question is paired with a named analysis and a reason it fits
  • Levels of measurement from the previous stage drive every test choice
  • Assumptions are named with the method that would check them
  • An alpha level and an operationally meaningful difference are both stated
  • Any subgroup analysis is specified in advance with its rationale
  • Storage, access and de-identification are written as procedures with named responsibility

Writing the analysis plan for NR-585AT?

Send the rubric out of Canvas with your questions and variable list. A premium original draft comes back in 24 to 48 hours with the baseline defined, each question matched to a defensible analysis and the stewardship plan written as procedure, and revisions run until the grade lands.

Questions students ask about this stage

My statistics course was a long time ago. How much do I need to remember?
Enough to match a test to a data structure and explain the match in two sentences. You are not being asked to derive anything or to run an analysis. Build the question-by-variable table first, because once you can see that you have two independent groups and a continuous outcome, the test names itself, and the same holds for every other combination you are likely to propose. Then write in plain language what the test would tell you: whether the average wait differs between the two periods by more than would be expected from ordinary variation. Graders reward accurate plain language over decorative vocabulary, and in a compressed session the plain version is also considerably faster to write.
Should I plan to look at results by patient subgroup?
Often yes, and the reason is not statistical curiosity. An intervention that improves an average while leaving one group behind, or while working only for patients who already had the easiest access, is a result a department needs to know about before it scales anything. Specify the subgroups in advance, say why each was chosen, and acknowledge that subgroup analyses have less power than the main comparison and are therefore exploratory. That last sentence is what keeps the analysis honest. What you must avoid is examining group after group after the fact and reporting whichever comparison happened to reach significance, because that practice produces findings that disappear the next time anybody looks.
How do I set a threshold for what counts as a meaningful change?
Use three anchors and say which one you relied on. The first is the literature: if published studies report the size of difference considered clinically important for your outcome, cite it and adopt it. The second is your own baseline variation: a change smaller than the ordinary month-to-month spread is unlikely to be distinguishable from noise, whatever a test says. The third is the operational one: the smallest change that would actually alter a decision, such as freeing enough hours to close a gap in coverage or reducing delays enough to affect the next admission. Writing the threshold down before data exists protects you twice, once against overclaiming a small win and once against dismissing a real effect because it looked modest.

Keep going

Online now