Integration and Reporting

The Codebook: making qualitative evidence reproducible

1. Short definition (one sentence). Written so another researcher can apply the code without asking for clarification. Vague definitions produce inconsistent coding. The Care Team entry reads: responses that credit a coach, nurse, or care coordinator, and their ongoing contact, as what helped the patient manage their condition.

2. Full description. Extended explanation of what the theme captures, who typically expresses it, and what range of responses it covers. Care Team’s full description specifies that it captures any mention of the clinical support relationship: a health coach, nurse, care coordinator, or provider who checked in, set goals, noticed setbacks, listened, or otherwise stayed personally connected to the patient.

3. Inclusion rules. When this code applies. For Care Team: apply when the response names or clearly refers to a coach, nurse, coordinator, or provider and credits that person’s contact, attention, or guidance. Include weekly check-ins, follow-up calls, goal-setting with a coach, someone noticing when the patient slipped.

4. Exclusion rules. When this code does not apply. For Care Team: do not apply when the support comes from family, a spouse, or other patients (that is Support). Do not apply to statements only about the patient’s own confidence or control (that is Confidence). Do not apply to barriers like cost, transportation, or feeling overwhelmed.

5. Borderline cases. Responses that could plausibly fit more than one code, with the resolution rule. If a response credits both a coach and a family member, Care Team is the primary code when the care-team relationship is the driver and the family is mentioned incidentally.

Quantitative reliability is built into the measurement (Cronbach’s alpha, test-retest). Qualitative reproducibility is built into the codebook. A theme without a codebook entry is not wrong, but it cannot be audited, cannot be replicated by a second coder, and cannot be defended when a reviewer asks how the code was applied. Writing the codebook is not administrative work; it is the documentation that earns the findings their credibility.

Theme by Group: where the 10-of-11 pattern becomes visible

The theme-by-group analysis here uses clinic as the grouping variable, which is the grouping the quantitative tests also used, making it the natural comparison. The 13 of 16 Access responses from Lakeview (81%) tells you that the barrier theme is heavily concentrated at the clinic with lower improvement rates. The deeper finding, 10 of 11 non-improvers coded Access, comes from splitting by the A1C improvement variable rather than by clinic. Both views tell the same story: the people who did not improve are the people describing structural barriers. The clinic split makes this visible because Lakeview has nine non-improvers and Northside has two.

Trustworthiness: the audit trail and the one-coder limitation

MM Studio builds the audit trail automatically from project history: when the project was created, when the data was linked, when themes were built, when the codebook was saved, when merge steps ran. A researcher who followed the 19-step workflow has a timestamped record of when each decision was made without keeping a separate log. That record is what a second researcher would follow to audit the work.

The Rigor Dashboard (Lesson 12) will flag this study’s only Missing item: coding agreement has not been computed. A single coder coded all fifty responses. This is common in small applied studies and in teaching demonstrations, but it is a limitation that requires explicit acknowledgment. Inter-rater reliability, having a second coder independently code a sample of responses and computing percent agreement or Cohen’s kappa, is the standard way to address it.

In the manuscript: qualitative coding was completed by one researcher; a second coder was not available for this demonstration study; future work should include inter-rater reliability checks using a sample of at minimum 20% of responses. The rigor dashboard’s Missing flag for this item is honest reporting, not a defect to hide.

Beyond the audit trail, trustworthiness in qualitative work includes member checking (did participants review the themes?), negative case analysis (did you actively look for responses that challenged the emerging interpretation?), and reflexivity notes (what assumptions did the researcher bring?). MM Studio provides space for these; none are automated. For a teaching demonstration, the audit trail and the codebook together carry the primary credibility weight.

Merge and Compare: what has to be true before the joint display opens

Care Team, Confidence, and Support are all paired with the self-efficacy t-test. They are confidence and relationship themes, and the quantitative variable they speak to is the self-efficacy scale. Access is paired with the A1C chi-square. Access is an outcome theme: the barrier it describes is associated with whether patients improved clinically, not with how confident they felt. Overwhelm and Transportation are also paired with the chi-square for the same reason.

Pairing every theme to the same test would be a mistake. It would imply that all six themes speak to the same quantitative phenomenon, which they do not. Pairing requires the researcher to know what each theme is actually about, and to match it to the result that addresses that construct.

The joint display, View 1: the theme display

Patients who coded Care Team describe a clinical relationship that helped them manage their condition. The t-test to which it is linked shows Northside scored higher on self-efficacy, and the Theme by Group analysis showed that 9 of 12 Care Team responses came from Northside. The strands agree: care team contact correlates with higher confidence.

Patients who coded Access describe structural barriers: cost, insurance gaps, food as medicine. The chi-square to which it is linked shows Lakeview had significantly lower A1C improvement. The Theme by Group showed 13 of 16 Access responses came from Lakeview. The sentiment is the only negative lean in the table. Before anyone reads the merge interpretation, these three facts (different linked test, negative sentiment, Lakeview-heavy distribution) are already visible in the theme display. Integration is not reading off the screen; it is seeing those three facts together.

Access appears in 32% of responses, more than any other theme. That frequency matters less than where those responses sit. The theme display cannot show you the 10-of-11 pattern. It can show you that Access has a different linked test and a different sentiment profile. A student who reads only the frequency column and concludes that Access is most important because 16 is bigger than 12 has not integrated; they have counted. Integration requires reading across all four columns, then comparing what the theme display shows with what the case matrix will show next.

The joint display, View 2: the case matrix

The theme display works at the group level. Group-level agreement can hide individual-level contradiction. Two clinics can differ on average while substantial numbers of individuals inside each clinic behave like patients at the other. The case matrix is where you find those people, and where the study stops being about statistical categories and becomes about the person who lost insurance mid-program.

L06 (Lakeview, score 3.00, positive narrative) is visible in the next row of the matrix. A score of 3.00 is more than two standard deviations below the Lakeview mean. The narrative is positive. That discordance, low number and good story, is exactly what the discordant case view exists to surface and examine before interpretation is locked in.

The joint display, View 3: discordant and negative cases

Self-efficacy 3.00, more than two SDs below the mean. Narrative positive: by the end I trusted myself to adjust my routine on my own. The number says this patient struggled to feel capable; the words say they found confidence despite everything. This is not a data error. It is a patient whose self-report at the moment of the survey and their narrative of what the program did for them are telling different stories. Both are real. The discordant flag is the prompt to ask which is more informative for the finding.

Self-efficacy 6.40, nearly 1.5 SDs above the mean, in the strong zone. Narrative: healthy food is expensive, and some weeks I had to choose what to buy. Access coded, negative sentiment. This patient is the most important row in the study for interpretation: a good number, and a structural barrier that the number does not show. This case is what stops the conclusion that Lakeview patients were less engaged. This patient was engaged enough to score 6.40, and still faced a barrier that confidence alone could not clear.

The Contextual Lens: what must not be concluded

The Consequence frame for Access in this study produces the sentence that is the difference between a finding that helps a clinic and one that harms patients: do not frame Lakeview patients as non-compliant. The joint display already shows that non-improvers described barriers. The Contextual Lens makes sure the interpretation does not walk back to the blame narrative through careless word choice in the report.

Convergence and Divergence: naming where the strands agree and where they split

When you mark aggregate-versus-experience divergence, MM Studio sends you back to the contextual lens on purpose. That is the dangerous moment in mixed methods: the number looks like a group failure; the words say resources. Naming the split without explaining it is how blame sneaks back in.

Meta-inferences: what only the merge can claim

A meta-inference is not a re-statement of the quantitative result with a quote attached. Saying that self-efficacy differed significantly and patients said care team contact mattered is proximity, not integration. A true meta-inference makes a claim that requires both strands as evidence: the program’s confidence effect is concentrated where care team relationships are strongest, and blocked where structural barriers interrupt adherence. That claim cannot be made from the t-test alone, and it cannot be made from the Care Team and Access themes alone. Both are required.

Evidence Strength: thin themes and what to do about them

Transportation (3 responses) and Overwhelm (5 responses, borderline) are flagged. The decision in this study is to retain both, as thin themes with explicit disclosure rather than merging them into Access and losing the distinction between structural barriers (Access) and experiential barriers (getting to appointments, early program difficulty). The report will flag both as thin themes with insufficient evidence for confident claims, note that they warrant follow-up, and recommend a study with targeted data collection for these barriers. Keeping them visible is more honest than absorbing them into a larger category that buries the distinction.

Rigor Dashboard: plain-language readiness before the final report

The rigor dashboard turns the project’s state into plain language so the researcher, and ultimately any reviewer, can see exactly what was done and what was not. Researchers who use it this way arrive at the report step with no surprises and no gaps to hide.

Report Builder: assembling one defensible document

The Report Builder assembles sections in standard research order: executive summary, abstract, methods (quantitative and qualitative), quantitative findings, qualitative findings, integrated findings, implications, and limitations and trustworthiness. Each section draws from a corresponding step in the 19-step workflow. The researcher can insert whole sections or individual blocks, then edit within the report.

The Report Builder includes an AI assist option in the editing bar. If the researcher uses ReliCheck Intelligence to draft or revise any section of the report, that use must be disclosed. MM Studio generates a disclosure statement for the report’s limitations section: noting that AI-assisted language generation was used in preparing the report, specifying which sections were AI-drafted, and confirming that all statistical interpretations and qualitative judgments were reviewed and approved by the researcher.

The integrated report claims four meta-inferences, supported by both strands. It does not claim causation. It does not claim that removing access barriers would immediately close the improvement gap. It does not claim the qualitative findings are generalizable beyond these fifty patients. It does note that the findings are consistent across multiple levels of analysis (group statistics, individual case matrix, discordant case review) and that the one-coder limitation warrants a replication with inter-rater reliability checks. That is the scope of what this study can honestly say.

Exit ticket: defend the integration

Two clinics, same program, different outcomes. The data preparation found a clean dataset with two decisions to document. The quantitative strand found a large confidence gap and a significant outcome gap. The qualitative strand found six themes, Access loudest and the only one with a negative lean. The integration found that the people whose numbers stayed flat are the people telling you the barrier was access, not effort. That finding did not come from the numbers alone or the words alone. It came from putting both strands in the same table and reading what they said about the same patients.

Reporting language for integration, limitations, and AI disclosure

Integration followed a convergent parallel design. Access diverged from the aggregate pattern: of the 11 patients who did not improve their A1C, 10 provided responses coded to the Access theme, describing barriers including medication cost, insurance gaps, and food insecurity. Five themes (Care Team, Confidence, Support, Overwhelm, Transportation) converged with the quantitative findings: patients coding these themes showed higher self-efficacy scores on average and improved at higher rates. The integrated finding is that the clinic outcome gap reflects differential access to the resources required to adhere to the program, not differential effort or program quality.

Twenty-three cases were flagged as discordant or carrying negative narratives, including five cases of strong discordance. One patient (L06) scored more than two standard deviations below the group mean while describing confident self-management. One patient (L03) scored above the mean while describing food-cost barriers. Both cases were retained and are reported as qualifications to the aggregate pattern: high confidence scores do not guarantee the absence of structural barriers.

Qualitative coding was completed by a single researcher. Inter-rater reliability was not computed. Future replication should include a second coder with at least 20% of responses independently coded. Two themes, Transportation (n = 3) and Overwhelm (n = 5), had insufficient response counts for confident claims and are reported as preliminary observations requiring targeted follow-up. All findings are descriptive and associational; the non-random clinic assignment prevents causal inference.

ReliCheck Intelligence (AI-assisted language generation) was used in the preparation of the executive summary and integrated findings sections of this report. All statistical interpretations, qualitative judgments, coding decisions, and meta-inferences were reviewed and approved by the researcher prior to publication. The AI-assisted text was edited for accuracy and consistency with the underlying analysis.

Guide 01 (Introduction and Data Prep) and Guide 02 (Quantitative and Qualitative Analysis). ReliCheck MM Studio · Mixed Methods Teaching Guide 03 of 03. Part 3: Integration and Reporting · Study: Northside vs Lakeview Diabetes Coaching · N = 50.