DNP Capstone Data Analysis Help — Choosing the Right Methods and Interpreting Your Findings

What Level of Statistical Analysis Is Expected in a DNP Capstone?

The expected level of statistical analysis in a DNP capstone is determined by the project design, the outcome data type, and the sample size — not by a desire to demonstrate statistical sophistication. Most DNP capstone data analyses consist of: (1) descriptive statistics for all variables (frequencies, means, standard deviations, medians, ranges, and confidence intervals as appropriate), reported for the full sample and for pre-implementation and post-implementation data separately; and (2) one primary inferential test that answers the PICOT question — paired t-test, Wilcoxon signed-rank, chi-square, McNemar, or a statistical process control chart analysis. This is not a minimalist or insufficiently rigorous analysis. It is the methodologically correct analysis for a practice improvement project with a single primary outcome measure and a pre-post design.

Doctoral-level analytical competence in a DNP capstone is demonstrated by: (a) correctly identifying the data type of the outcome variable (continuous vs categorical), (b) checking the statistical assumptions of the selected test before applying it, (c) correctly interpreting both statistical significance (p-value) and clinical significance (Minimal Clinically Important Difference), (d) acknowledging the limitations of the analysis (small sample size, absence of a control group, convenience sampling) in the discussion section, and (e) drawing only those conclusions that the data support — not overstating results from a small, single-site QI project as evidence of universal treatment efficacy.

Running an ANCOVA, MANOVA, or logistic regression without justification — when a paired t-test is the correct test — does not demonstrate greater statistical competence. It demonstrates a misunderstanding of test selection criteria and inflates the complexity of the analysis section without adding analytical value. Committees notice this. The statistical analysis plan approved in Chapter 3 of the proposal is the plan that must be implemented — post-hoc changes to the analysis approach require committee notification and explanation in the manuscript methods section.

Pre-Post Comparison Analysis for QI and EBP DNP Projects

Pre-post comparison is the most common analytical design in DNP capstone projects — one group of patients, staff, or outcomes measured at baseline (before implementation) and again after implementation. The appropriate test for this design depends on the data type and the normality of the distribution.

Paired t-test: Use when the outcome variable is continuous (a number that can take any value within a range — HbA1c level, pain NRS score, PACU length of stay in minutes, PHQ-9 score), the same subjects or units are measured twice (pre and post), and the distribution of difference scores is approximately normal. Normality is checked with the Shapiro-Wilk test — if p greater than 0.05, normality is not rejected and the paired t-test is appropriate. The paired t-test produces four outputs that must all be reported: the mean difference (M_diff), the 95% confidence interval for the difference, the t-statistic with degrees of freedom, and the p-value. Example results sentence: "The mean HbA1c decreased from 9.2% (SD=0.8) at baseline to 8.4% (SD=0.7) post-implementation, a statistically significant reduction of 0.8 percentage points (t(29)=4.82, p<0.001, 95% CI [0.45, 1.15])."

Wilcoxon Signed-Rank Test: The non-parametric alternative to the paired t-test. Use when the normality assumption is violated (Shapiro-Wilk p≤0.05) or when the sample size is small (N below 30), making the central limit theorem insufficient to justify normality assumptions. The Wilcoxon signed-rank test produces the median difference, Z-statistic, and p-value. It is slightly less statistically powerful than the paired t-test when normality holds, but it makes no distributional assumptions and is appropriate for any continuous or ordinal data type. Example: "The median PHQ-9 score decreased from 12 at baseline to 7 post-implementation, a statistically significant improvement (Z=-3.42, p=0.001)."

McNemar Test: Use when the outcome variable is categorical (binary — yes/no, compliant/non-compliant, screened/not screened), the same subjects are classified at two time points (pre and post), and the comparison is between the two paired categorical outcomes. McNemar is appropriate for: "percentage of patients who received a PHQ-9 screening (Yes/No) before versus after protocol implementation," or "percentage of catheter days with documented necessity assessment (Yes/No) before versus after the nursing education intervention." The McNemar test applies to a 2×2 contingency table and produces a chi-square statistic (for N greater than 25) or Fisher's exact (for small N) with a p-value. Example: "Documentation of daily catheter necessity assessment improved from 42% (50/119 eligible catheter days) at baseline to 87% (104/120 eligible catheter days) post-implementation (McNemar's chi-square = 38.2, p<0.001)."

Chi-Square Test of Independence: Use when both the outcome variable and the grouping variable are categorical, and the samples are independent (not the same subjects measured twice — a different group pre vs post, or two separate groups receiving different interventions). Requires a minimum expected cell count of 5 in each cell of the contingency table. When any expected cell count is below 5, use Fisher's Exact Test instead. Chi-square is less commonly used in DNP capstone projects than McNemar because most DNP designs compare the same population pre and post (paired data), not two independent groups.

Assumption check decision tree: Is the outcome variable continuous or categorical? → Continuous: Is N ≥ 30? → Yes: Run Shapiro-Wilk. → Shapiro-Wilk p>0.05 (normal): Use paired t-test. → Shapiro-Wilk p≤0.05 (non-normal): Use Wilcoxon signed-rank. → Is N <30: Use Wilcoxon signed-rank regardless of normality. → Categorical: Are subjects the same pre and post? → Yes: Use McNemar. → No (independent groups): Use chi-square (or Fisher's exact if expected cell count below 5).

Statistical Process Control Charts for DNP Quality Improvement Projects

Statistical process control (SPC) charts are used in QI projects to track process performance over time and distinguish common-cause variation (the expected, stable variation in any process) from special-cause variation (signals of a real change in the process). SPC charts are the preferred data presentation format for QI data that is collected weekly or monthly across the implementation period — they show not just whether the rate changed from pre to post, but when the change began, whether it is sustained, and whether the change exceeds expected variation.

P-chart (Proportion chart): Use for outcome measures expressed as proportions — the percentage of events per denominator when the denominator is approximately constant across time periods. Examples: hand hygiene compliance rate (% of opportunities observed where hand hygiene was performed), PHQ-9 completion rate (% of eligible visits where PHQ-9 was completed), catheter necessity assessment documentation rate (% of eligible catheter days where assessment was documented). The P-chart requires a minimum of 20 data points for reliable control limits (upper control limit, lower control limit, centre line at the mean proportion). Data points that fall outside the control limits, or that show 8 consecutive points on one side of the centreline, signal a special cause — either improvement (desired) or deterioration (requiring investigation).

U-chart (Rate chart): Use for outcome measures expressed as rates where the denominator varies across time periods. Examples: CAUTI rate per 1,000 catheter days (the number of catheter days varies each month with census fluctuations), falls per 1,000 patient days, central line-associated bloodstream infections per 1,000 central line days. The U-chart recalculates the control limits for each data point to account for the varying denominator — this is the critical difference from the P-chart, which assumes a constant denominator.

I-MR Chart (Individuals and Moving Range chart): Use for individual continuous measurements that are not expressed as proportions or rates — weekly mean pain NRS score, monthly average PACU length of stay in minutes, weekly mean patient satisfaction score. The I-MR chart pairs an Individuals chart (each data point is one measurement) with a Moving Range chart (the difference between consecutive measurements, which assesses process stability). Appropriate when the subgroup size is 1 — each week produces one aggregate measurement rather than a sample from which a mean is calculated.

Run chart: The simplest alternative to SPC charts when fewer than 20 data points are available. A run chart plots data over time with a centreline at the median (not the mean). It uses run chart rules to identify signals of special cause: 8 or more consecutive data points above or below the median (a shift), 6 or more consecutive data points trending consistently up or down (a trend), or 14 or more data points alternating up and down (a sawtooth pattern, indicating two alternating causal factors). Run charts require no statistical software — they can be constructed in Microsoft Excel with a median formula. For small-N DNP projects with 8 to 15 data points, a run chart is methodologically appropriate and does not require the 20-point minimum for SPC chart control limits.

SPC software: QI Macros for Excel (Excel add-in — most user-friendly for clinical QI projects, produces publication-quality SPC charts), Minitab (more comprehensive statistical software — typically available through university licences), Excel with manual formulas (requires knowledge of the UCL and LCL calculation formulas for each chart type), SPSS (does not produce SPC charts natively — QI Macros or Excel required for SPC).

Qualitative Data Analysis for DNP Projects: Thematic and Content Analysis

Qualitative data analysis is used in DNP capstone projects that collect open-ended text data — staff focus group transcripts, patient experience interview transcripts, open-ended survey responses, or field notes from observational data collection. It is most common in PMHNP capstones examining patient experience with mental health services, Nurse Executive capstones evaluating staff perceptions of a leadership program, Population Health capstones examining community health needs, and Nursing Informatics capstones assessing user experience with technology. It may also appear as the qualitative strand of a mixed-methods program evaluation design.

Thematic Analysis (Braun & Clarke, 2006, revised 2019): The most widely used qualitative analysis method in DNP capstones. Braun and Clarke's six-phase process: (1) Familiarisation: read and re-read all data, take initial notes, develop an initial sense of patterns; (2) Generating initial codes: systematically code all data segments that are relevant to the research or evaluation question — one code per meaningful unit; (3) Searching for themes: group codes into candidate themes based on conceptual similarity; (4) Reviewing themes: check candidate themes against the full data set — do they accurately represent the data? Are any themes too broad or too narrow? (5) Defining and naming themes: produce a clear definition for each theme and a name that reflects its content — not a vague label but a concise statement of the theme's meaning; (6) Writing up: write the thematic narrative with supporting quotes from the data. Each theme must be illustrated with at least two to three representative quotes; quotes are anonymised (Participant 1, Participant 2).

Content Analysis: A more quantitative approach to qualitative data — codes are counted as well as described, producing frequency tables alongside the thematic narrative. Appropriate when the research question asks "how often" as well as "in what way" — for example, how frequently staff mentioned workload as a barrier versus how frequently they mentioned lack of training. Content analysis produces both a frequency table of code occurrence and a narrative description of the coded content.

NVivo: The most commonly used qualitative software in nursing doctoral programs. Organises codes, memos, data extracts, and themes in a searchable project file. Allows the researcher to run word frequency queries, code co-occurrence analyses, and visualise theme maps. Required by some universities; recommended by most. Access is typically available through university library software licences.

Reflexivity: Required in thematic analysis — a statement in the methods section acknowledging the researcher's positionality (clinical background, relationship to the clinical setting, prior assumptions about the phenomenon) and how this may have influenced the coding and theme development process. "The DNP student conducted all data analysis from the dual position of practitioner and researcher. Reflexive memos were maintained throughout the coding process to document analytical decisions and manage the potential influence of clinical preconceptions on theme development." Without a reflexivity statement, the thematic analysis chapter is incomplete regardless of how rigorously the coding was conducted.

Statistical Significance vs Clinical Significance in DNP Practice Change

Statistical significance and clinical significance are distinct concepts that both must be addressed in the discussion section of a DNP capstone manuscript. Reporting only statistical significance (p-value) without addressing clinical significance is the single most common analytical error in DNP manuscripts — it demonstrates a fundamental misunderstanding of what statistical testing proves.

Statistical significance (p<0.05): The probability that a result as extreme as the observed result would occur by chance if the null hypothesis (no difference between pre and post) were true is less than 5%. A p-value does not indicate the size or clinical importance of the effect. With a large enough sample, any difference — no matter how trivially small — will be statistically significant. A 0.1 HbA1c reduction in a sample of 500 patients will be highly statistically significant (p<0.001), but a 0.1 HbA1c reduction is clinically meaningless.

Clinical significance — Minimal Clinically Important Difference (MCID): The smallest change in an outcome measure that a patient would feel is worthwhile or that clinicians regard as reflecting a meaningful change in health status. MCIDs are published in the psychometric validation literature for validated instruments and in clinical practice guidelines for quality metrics. Key MCID values for common DNP capstone outcome measures: PHQ-9 depression scale MCID = 5 points (Kroenke et al.); GAD-7 anxiety scale MCID = 4 points; NRS pain scale MCID = 2 points (Farrar et al.); HbA1c MCID for type 2 diabetes management = 0.5 percentage points (ADA standards). Discussion section language: "The observed reduction of 0.8 percentage points in mean HbA1c from 9.2% to 8.4% exceeds the MCID of 0.5 percentage points established in the ADA Standards of Medical Care in Diabetes, indicating that the change is both statistically significant and clinically meaningful."

Effect size: A standardised measure of the magnitude of the effect, independent of sample size. Cohen's d for t-tests: d=0.2 (small), d=0.5 (medium), d=0.8 (large). Cramer's V for chi-square: V=0.1 (small), V=0.3 (medium), V=0.5 (large). Effect sizes must be calculated and reported alongside p-values in any published or formally presented DNP capstone results — they allow readers to assess the practical magnitude of the change regardless of sample size.

Small sample limitation: Many DNP capstone projects have small samples (N below 30) — one unit, one semester, all eligible patients. A small sample reduces statistical power (the ability to detect a real effect if one exists), meaning that a non-significant p-value (p above 0.05) does not prove the intervention was ineffective — it may simply mean the sample was too small to detect a real but small effect. The discussion section must address this explicitly: "The absence of a statistically significant difference in the secondary outcome (30-day readmission rate) may reflect insufficient statistical power rather than the absence of a true clinical effect, given the small sample size (N=28) and the relatively short implementation period (10 weeks)."

Presenting Your Data: Tables, Run Charts, and Figures in APA 7 Format

All data presentations in the DNP capstone manuscript follow APA 7th edition student paper format. Tables are numbered and titled above the table; figures (including run charts, SPC charts, and PRISMA diagrams) are numbered and titled below the figure. No vertical lines appear in APA tables; no shading is used unless it is light grey for alternating rows. All table and figure numbers must be cited in the text before the table or figure appears.

Pre-post comparison results table: Rows = outcome variables; Columns = Pre-implementation M (SD) or %, Post-implementation M (SD) or %, Difference (95% CI), Test statistic (t or Z with df), p-value. The table is titled above: "Table 1. Pre- and Post-Implementation Outcome Measures." A note below the table cites the statistical tests used: "Note. Paired t-tests were used for continuous normally distributed outcomes. Wilcoxon signed-rank tests were used for non-normal or ordinal outcomes. *p<0.05. **p<0.01. ***p<0.001."

Run chart or SPC chart as Figure 1: Titled below the figure: "Figure 1. Weekly CAUTI Bundle Compliance Rate (P-chart), October to December 2026." The centreline (mean proportion), UCL, LCL, and the implementation start date (marked with a vertical line or arrow) must be clearly labelled. The caption includes a note explaining any special cause signals: "Note. The vertical line at week 3 indicates the start of the nurse-driven daily necessity assessment protocol. Data points outside the control limits after week 3 indicate special-cause variation consistent with a sustained process improvement."

Demographics table (Table 1): Always the first table in the results section — presents sample characteristics (age range, gender distribution, diagnosis categories, unit characteristics) before any outcome data. This establishes who the project population was and allows readers to assess the applicability of findings to other settings.