18. Integrative Analysis & Interpretation

Cross-method dialogue

Integrative analysis is structured dialogue between data streams. The dialogue surfaces three relations:

• Convergence — methods point at the same finding from different angles
• Divergence — methods point in different directions
• Complementarity — methods illuminate different facets that combine to a fuller picture

A rigorous integrative analysis reports all three. Reporting only convergence inflates confidence; reporting only divergence inflates ambiguity; reporting only complementarity hides where the methods genuinely agree or disagree.

The discipline is to look for each. If you find only one of three, look harder.

Building the dialogue

The most common failure mode is integration that happens only in the discussion paragraph. The transdisciplinary practice integrates during analysis, with documented artifacts.

A useful sequence:

1. Analyze each strand independently, producing strand-specific findings.
2. Build alignment artifacts (matrices, joint displays) that lay strand findings against each other on a shared dimension.
3. For each row of the artifact, ask: where do strands converge? diverge? complement?
4. Develop integrated inferences for each row.
5. Generate meta-inferences that span rows.

The artifact is the audit trail. A study that reports integrated inferences without showing the artifacts is asking for trust where it should be showing work.

Convergence

Convergence is when methods agree on a finding. It's the easy case but worth being careful about:

• Genuine convergence: methods independently produce the same finding from different measurement bases.
• Apparent convergence: methods produce findings that look similar but measure different things; the agreement is illusory.

A test for genuine vs. apparent convergence: do the methods operationalize the construct differently? If both methods use self-report on the same instrument's items, convergence is barely interesting. If one method measures behavior, another measures self-report, and a third measures third-party observation, convergence is much more informative.

Divergence

Divergence is informative when interpreted, not glossed. When methods disagree, ask:

• Different facet? The methods might be measuring different aspects (reported behavior vs. revealed behavior; current state vs. retrospective).
• Different time-frame? A snapshot survey and longitudinal interviews may pick up different temporal slices.
• Sample noncomparability? The strands may have sampled differently.
• Construct slippage? The construct may not be the same across methods.
• Genuine contradiction? Sometimes the methods just disagree, and the disagreement is the finding.

Report the divergence and your interpretation of it. "Mixed evidence" is a fine result if the mixing is interpreted.

Complementarity

Complementarity is the relation most distinctive to mixed-methods work. Methods don't agree or disagree; they cover different territory, and the combined picture is more complete.

Patterns of complementarity:

• Quant gives scale; qual gives meaning.
• Quant gives correlation; qual gives mechanism.
• Quant identifies subgroups; qual differentiates within them.
• Qual identifies a construct; quant measures its prevalence.

Naming complementarity explicitly — "method A captured X; method B captured Y; together they describe the phenomenon at both layers" — is more honest than treating complementary findings as if they converged.

From raw findings to wisdom

The data–information–knowledge–wisdom (DIKW) hierarchy is a useful audit of integrative write-ups:

• Data — raw observations (interview transcripts, survey responses, lab values)
• Information — data organized into meaningful structure (themes, descriptive statistics, codebook outputs)
• Knowledge — information situated in context and theory (claims about what the patterns mean)
• Wisdom — knowledge translated into judgment for action under uncertainty (recommendations, decision implications)

A common imbalance: studies rich in information but thin on knowledge and wisdom. Pages of tables, paragraphs of themes, and a one-paragraph conclusion gesturing at "implications for practice."

The transdisciplinary discipline is to do all four. Move up the hierarchy explicitly in the write-up: here is what we measured (data); here is what we found when we organized it (information); here is what we believe it means in context (knowledge); here is what we recommend or where action is premature (wisdom).

Visual integration strategies

Tools that force integration:

• Joint displays — tables aligning strand findings (covered in Lesson 13)
• Integration matrices — themes × strand × inference, organized for synthesis
• Causal maps — visual representation of mechanisms supported by combined evidence
• Trajectory plots — quant trends with overlaid qualitative narrative milestones
• Stakeholder synthesis maps — findings organized by stakeholder relevance

Each tool forces a particular kind of integration. Pick the tool whose forcing function matches your analysis.

Practical implications

The last step of integrative analysis is translation — what would a reader do with this finding?

A useful template:

• For practitioners: what should change tomorrow?
• For policy: what does this suggest for resource allocation or rule change?
• For research: what's the next study?
• For community: what would the people studied want to know?

Some findings support all four; some support only one or two; some are explicitly premature for action. The honest write-up names the level of action the evidence supports and the level it doesn't.

A worked vignette

A study of a youth mentorship program combines quant outcomes (academic performance, attendance) with qual interviews (mentor and mentee experience).

Strand-independent findings:

• Quant: small but significant increase in attendance, no detectable change in GPA at 1 year
• Qual: mentors report deep relational investment; mentees report ambivalence — appreciation mixed with surveillance feeling

Integrative analysis:

• Convergence: both strands suggest the program "matters" to participants
• Divergence: quant shows behavioral change in attendance but not academics; qual identifies a relational dimension the outcomes don't capture
• Complementarity: quant gives the behavioral signal; qual gives the experiential layer that explains why academics didn't shift (mentees felt monitored rather than supported, which doesn't show up in attendance data)

Meta-inference: the program produces relational impact and modest behavioral effects; academic outcomes are likely to require a longer time horizon, or a redesign that addresses the surveillance feeling.

Wisdom-level translation: continue program, redesign monitoring practices to reduce surveillance feel, evaluate at 24 months. Premature recommendations on scaling.

Closing

Cross-method dialogue surfaces convergence, divergence, and complementarity. Integration happens during analysis with artifacts, not in the discussion paragraph. The DIKW hierarchy is a useful audit. Translation to action — including naming where action is premature — is part of integrative analysis.

Next: reflexivity and positionality — the researcher as primary instrument, and decolonizing approaches in interpretive work.