ImpactMojoMixed Methods 101www.impactmojo.in

ImpactMojo 101 Series · Free Forever

Mixed
Methods
101

Intentionally Combining Quantitative & Qualitative Evidence in One Study — a Foundational Course for Development Researchers & MEL Practitioners in South Asia

Research-BackedSouth Asia Focus100 SlidesFree Access

ImpactMojoMixed Methods 101www.impactmojo.in

Agenda

What We Cover

What Mixed Methods Is

Slides 3–11

Why Mix? Five Purposes

Slides 12–20

Paradigms & Pragmatism

Slides 21–29

The Core Designs

Slides 30–38

Choosing a Design

Slides 39–46

Sampling in Mixed Methods

Slides 47–55

Integration: the Heart

Slides 56–64

Quality & Validity

Slides 65–73

Worked Examples

Slides 74–82

Common Pitfalls

Slides 83–91

Practice & Tools

Slides 92–99

ImpactMojoMixed Methods 101www.impactmojo.in

Section One

What Mixed Methods Is

ImpactMojoMixed Methods 101www.impactmojo.in

Definition

Mixing is a choice, not an accident

Mixed methods research

Research that intentionally collects, analyses and integrates both quantitative and qualitative data within a single study or programme of inquiry — so the combination answers questions neither approach could answer alone.

The operative word is integration. Running a survey and some interviews in the same project is not yet mixed methods — it is two studies in a trench coat.

Not mixed methods	Mixed methods
A survey, and separately some interviews	Interviews chosen because of what the survey showed
Two chapters that never refer to each other	A finding neither strand could reach alone
Quotes added to illustrate a percentage	Quotes that explain why the percentage is what it is
Two teams, one report	One design, specified before fieldwork

The distinguishing feature is integration, not the presence of two data types. Most studies described as mixed methods in this sector are the left-hand column, and Section 7 is where that gets tested.

The choice has to be made in advance. Deciding to add interviews after the survey disappoints is a salvage operation, not a design — and it shows in what the interviews can be asked to do.

ImpactMojoMixed Methods 101www.impactmojo.in

More Than Both

Doing both is not the same as mixing both

Multi-method (parallel)

A quant strand and a qual strand sit side by side. Each is written up separately. They never meet. The reader does the integrating — if at all.

Mixed methods

The strands are deliberately connected: one informs the other, or they are merged and compared. Integration is designed in, and reported as a finding in its own right.

Ask of any 'mixed' study: where do the numbers and the words actually talk to each other?

Test of whether it is really mixed	Answerable from
Did one strand influence the other’s design?	The sampling plan; the instrument
Is there a finding that needs both?	The conclusions section
Was there a point where they met?	A joint display, or its absence
Was divergence reported?	Whether any disagreement appears at all

The fourth row is diagnostic. Two strands examining the same reality almost always disagree somewhere; a report in which they agree on everything has usually not looked, or has quietly dropped what did not fit.

Apply these four to the last “mixed methods” report you read. Most fail the third, which is why the joint display in Section 7 is the single most useful tool in this deck.

ImpactMojoMixed Methods 101www.impactmojo.in

The Promise

1 + 1 = 3

The case for mixing rests on a simple idea: integrated quantitative and qualitative findings yield insight greater than the sum of the parts. Numbers show the pattern; words explain the mechanism; together they make a claim each alone could only gesture at.

The combination of qualitative and quantitative methods can produce a more complete picture — the whole is greater than the sum of the parts.

— paraphrasing the '1 + 1 = 3' principle, mixed-methods literature

Alone	Together
“Uptake is 44%”	“44%, and the 56% cite a specific fee nobody recorded”
“Women say the timing is impossible”	“…and attendance drops 30 points in harvest months”
“The programme had no measured effect”	“…because in half the sites it was never delivered”

The third row is the classic case for mixing in evaluation. A null result is uninterpretable on its own — the theory could be wrong, or the programme could never have happened — and only implementation data distinguishes them.

1 + 1 = 3 is a promise, not a guarantee. Badly mixed work gets 1 + 1 = 1.5: two thin strands where one done well would have been stronger. Section 10 covers how that happens.

ImpactMojoMixed Methods 101www.impactmojo.in

Complementary Strengths

What each strand brings

Quantitative

Scale, prevalence, magnitude
Comparison across groups
Generalisable to a population
Tests hypotheses; measures effect

Qualitative

Meaning, process, context
Why and how, in whose words
Depth on the unexpected
Generates hypotheses; explains effect

Each method's weakness is the other's strength. Mixing is offsetting one set of blind spots with another.

ImpactMojoMixed Methods 101www.impactmojo.in

A Short History

From 'wars' to a third paradigm

1960s–80s: quant dominates; qual fights for legitimacy

→

1980s–90s: 'paradigm wars' — are the two even compatible?

→

1989: Greene et al. map five purposes for mixing

→

2000s: Creswell & Plano Clark codify designs; a 'third paradigm'

Mixed methods is now widely treated as a distinct methodology — not a truce, but a framework with its own logic.

Era	Position
1970s–80s	“Paradigm wars” — the two traditions held to be incompatible
1990s	Purposes and designs formalised; Greene et al. 1989
2000s	Pragmatism adopted; a “third methodological movement”
Now	Standard in evaluation; journals and reporting standards exist

The wars mattered and are over in practice. Nobody funding an evaluation today argues that combining the two is incoherent; the live disputes are about how to integrate well, which is a craft question rather than a philosophical one.

The philosophy is still worth an hour. Section 3 exists because a researcher who cannot say what their methods assume will mix them badly — usually by treating qualitative data as a weaker form of the same thing.

ImpactMojoMixed Methods 101www.impactmojo.in

In Development

Why MEL practitioners reach for it

Development questions are stubbornly mixed. Did the programme work? is quantitative. For whom, why, and how? is qualitative. An impact figure without a mechanism is hard to act on; a rich story without scale is hard to defend.

Explain why an effect is large in one district, absent in another
Surface unintended consequences a survey never asked about
Give voice and context to a hard outcome number
Build instruments grounded in how people actually talk

MEL question	Why one strand is not enough
Did it work?	Quant answers this; cannot say why
Why did it work here and not there?	Needs the mechanism, not the average
Was it delivered as designed?	Implementation data, usually qualitative
What did participants think it was for?	Only answerable by asking
Should we scale it?	Needs effect size and conditions together

The last row is why donors increasingly require mixing. A scale-up decision needs both the magnitude of the effect and the conditions under which it appeared, and a study answering only the first will be scaled into contexts where it does not hold.

The third row is the cheapest addition to any evaluation. A short process strand costs a fraction of the impact study and prevents the most expensive mistake in the field — concluding a theory failed when a programme was never delivered.

ImpactMojoMixed Methods 101www.impactmojo.in

What It Is Not

Three honest cautions up front

Not a quality guarantee. Two weak strands make one weak study, not a strong one.
Not 'qual to decorate quant'. A few quotes around a regression is not integration.
Not free. Mixing roughly doubles the skills, time and coordination required.

Mix because the question demands it — never because 'mixed methods' sounds rigorous in a proposal.

Caution	What follows
Not automatically better	A well-run single-method study beats a badly mixed one
Not a fix for a weak strand	Two weak strands do not make one strong finding
Not cheaper or faster	Costs more, in money and in skills

The second row is the most common misuse in this sector. Adding interviews to a survey with a poor sampling frame does not repair the frame; it produces a report where the qualitative material quietly carries claims the quantitative strand cannot support.

Budget honestly. Mixed designs need people who can do both, or two people who will actually talk to each other, plus time for integration that is nobody’s deliverable and therefore never scheduled.

ImpactMojoMixed Methods 101www.impactmojo.in

Roadmap

How this course is built

Foundations

Why mix; paradigms & pragmatism
The four core designs & notation
Choosing, sampling, integrating

Practice

Quality, validity & meta-inference
Worked development examples
Pitfalls, teams, tools, writing

Pairs with ImpactMojo's Qualitative Methods and Data Literacy 101 decks — this one is about combining them.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Two

Why Mix? Five Purposes

ImpactMojoMixed Methods 101www.impactmojo.in

The Framework

Greene, Caracelli & Graham's five purposes

The most-cited answer to 'why mix?' comes from a 1989 review by Jennifer Greene, Valerie Caracelli and Wendy Graham. They distilled mixed-methods studies into five distinct purposes — still the field's working vocabulary.

Triangulation

Complementarity

Development

Initiation

Expansion

Name your purpose before your design. The 'why' should drive the 'how'.

ImpactMojoMixed Methods 101www.impactmojo.in

Purpose 1

Triangulation: do they agree?

Triangulation

Using two methods to study the same phenomenon, seeking convergence or corroboration — if both point the same way, confidence rises.

Survey data say child immunisation rose; FGDs with mothers independently describe easier access and more outreach visits. Two routes, one conclusion — the finding is more credible for it.

Watch the trap: when the two disagree, that is not a failure to fix — it is a finding to explain (see 'initiation').

Triangulation requires	Or it is not triangulation
Both strands address the same construct	Two different questions cannot corroborate
Both are strong enough to be believed	A weak strand agreeing proves nothing
They are genuinely independent	Same respondents, same day, same enumerator is one source
Disagreement was possible	If nothing could have diverged, nothing was tested

The third row is violated constantly. Interviewing the same households immediately after surveying them, through the same enumerator, produces two accounts sharing every source of bias — and agreement between them is not corroboration.

Triangulation is the most claimed and least practised purpose. It is the word people reach for when they mean “we did both”, and the four conditions above are what separate the claim from the thing.

ImpactMojoMixed Methods 101www.impactmojo.in

Purpose 2

Complementarity: a fuller picture

Here the methods examine different facets of the same issue, each elaborating the other. The aim is not agreement but richness — depth layered onto breadth.

A survey measures how many adolescent girls dropped out; interviews reveal the texture — menstruation, distance, safety, marriage. Neither replaces the other; together they explain the dropout.

Strand	Its facet of the same phenomenon
Survey	How many households changed practice, and by how much
Interviews	What changing meant to them, and what it cost
Observation	What actually happens at the point of delivery

Complementarity differs from triangulation in what agreement would mean. Here the strands are not checking each other; they are describing different facets, so they cannot corroborate and are not expected to.

This is the most common legitimate purpose in programme evaluation, and naming it as complementarity rather than triangulation prevents a reader from expecting a corroboration the design cannot deliver.

ImpactMojoMixed Methods 101www.impactmojo.in

Purpose 3

Development: one method builds the other

Qual first: interviews surface local categories & language

→

Use them to design a valid, grounded survey instrument

→

Quant then: measure prevalence at scale

→

Each strand sequentially informs the next

Development is sequential by definition: findings from method A shape method B — sampling, instruments or implementation.

One strand builds the other	Example
Qual → quan	Interviews generate the items for a survey instrument
Quan → qual	Survey identifies which cases are worth interviewing
Qual → sampling	Fieldwork reveals a subgroup the frame omitted
Quan → instrument	Pilot data shows which questions do not discriminate

The first row is how good instruments get built in unfamiliar contexts, and skipping it is why imported scales so often behave strangely here — the items were generated from a different population’s account of the construct.

Development is necessarily sequential, which has a scheduling consequence: the second strand cannot be tendered, staffed or costed precisely until the first reports. Plan a decision point, not a fixed second phase.

ImpactMojoMixed Methods 101www.impactmojo.in

Purpose 4

Initiation: provoke new questions

Initiation deliberately looks for contradiction and paradox between strands — the friction that reframes the question and sparks fresh insight.

Survey: satisfaction with the clinic is high. Interviews: people are quietly furious about waiting times. The dissonance is the discovery — perhaps 'satisfaction' measured low expectations, not good service.

Initiation happens when	Response
The strands contradict each other	Treat as a finding; investigate why
A strand surfaces a group nobody had considered	Revise the framework
The construct turns out to mean something else locally	Rebuild the measure

Initiation is the purpose nobody plans for and several studies end up needing. It is what happens when mixing works: the collision produces a question neither strand was designed to ask.

It is also the purpose most likely to be suppressed. A contradiction discovered late is inconvenient for a report already promised, and the path of least resistance is to footnote it — which is Section 10’s sixth pitfall.

ImpactMojoMixed Methods 101www.impactmojo.in

Purpose 5

Expansion: extend the breadth

Expansion uses different methods for different components of an inquiry — commonly quant for outcomes and qual for process — to extend the study's range, not to study one thing two ways.

Classic in evaluation: measure the impact (quant) and evaluate how implementation actually unfolded (qual) within the same study.

Strand	Question it covers
Quantitative	Outcomes: did indicators move, and by how much
Qualitative	Process: how it was delivered, and what participants did
Together	A broader account than either scope alone

Expansion is the weakest form of mixing and the most common in evaluation. The strands do not interact; they cover different questions in one report, which is legitimate and should be named honestly rather than described as triangulation.

It is where integration most often fails. Because the strands answer separate questions, nothing forces them to meet — so an expansion design needs a deliberate integration step or it becomes the parallel play of Section 10.

ImpactMojoMixed Methods 101www.impactmojo.in

Side by Side

The five purposes at a glance

Purpose	Core question	Strands
Triangulation	Do independent methods converge?	Same phenomenon
Complementarity	Can one enrich the other?	Overlapping facets
Development	Can one build the other?	Sequential, A → B
Initiation	Where do they contradict?	Tension sought
Expansion	Can we widen the scope?	Different components

Source: Greene, Caracelli & Graham (1989), Toward a Conceptual Framework for Mixed-Method Evaluation Designs.

Purpose	Strands address	Agreement expected?
Triangulation	The same construct	Yes — that is the test
Complementarity	Different facets	Not applicable
Development	One informs the other’s design	Not applicable
Initiation	Whatever collides	No — disagreement is the point
Expansion	Different questions	Not applicable

Read the third column before writing your conclusions. Only triangulation makes agreement meaningful; in the other four, reporting that the strands “confirmed each other” is a claim the design does not support.

A study can have more than one purpose, and most good ones do — development in phase one, complementarity in phase two. Name each, and say which strand serves which.

ImpactMojoMixed Methods 101www.impactmojo.in

Using It

From purpose to a design decision

Triangulation → collect concurrently, then compare
Development → collect sequentially, A informs B
Complementarity / Expansion → embed one within the other
Initiation → stay open; let contradiction redirect you

The purpose points at the design; the next sections turn that into concrete blueprints.

Purpose	Design it implies
Triangulation	Convergent parallel — both strands, independently
Complementarity	Convergent, or embedded
Development	Sequential, in the order the building requires
Initiation	Any — but with room in the schedule to follow up
Expansion	Parallel, with a deliberate integration point

The purpose determines the design, not the other way round, which is why this section comes before Section 4. Choosing “explanatory sequential” because it is familiar, then working out what it is for, is how studies end up with a second phase that answers nothing the first raised.

The fourth row is a scheduling instruction. If you want initiation to be possible, the timeline must contain slack after the first results — otherwise a contradiction can only be noted, never pursued.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Three

Paradigms & Pragmatism

ImpactMojoMixed Methods 101www.impactmojo.in

Why It Comes Up

Methods carry worldviews

Paradigm

A shared set of beliefs about what reality is (ontology) and how we can know it (epistemology) — the worldview a research approach is built on.

Quantitative work grew from post-positivism (one measurable reality); much qualitative work from constructivism (realities are socially constructed). Mixing seems to ask one study to hold two worldviews at once.

A method assumes	Which matters when
What kind of thing reality is	You ask whether a construct exists independently of measurement
What counts as knowing it	You decide whether one respondent’s account is evidence
What the researcher’s role is	You judge whether closeness to participants is bias or access

The third row causes most of the friction on mixed teams. To one strand, an interviewer who has built a relationship over months has contaminated the data; to the other, they have finally earned access to what people actually think.

You do not have to resolve this to work together, and you do have to notice it — otherwise the disagreement gets attributed to personalities and gets managed instead of discussed.

ImpactMojoMixed Methods 101www.impactmojo.in

Two Worldviews

Post-positivism vs constructivism

	Post-positivism	Constructivism
Reality	One, measurable	Multiple, constructed
Knower & known	Separate, objective	Intertwined, situated
Goal	Explain, predict, generalise	Understand, interpret meaning
Typical method	Survey, experiment	Interview, ethnography
Logic	Deductive, test theory	Inductive, build theory

Stated this starkly, the two look irreconcilable — which is exactly the objection mixed methods had to answer.

	Post-positivism	Constructivism
Reality	One, knowable imperfectly	Multiple, socially constructed
Knowledge	Approximated, provisionally	Co-produced with participants
Aim	Explanation, generalisation	Understanding in context
Quality	Validity, reliability	Credibility, transferability

The last row has practical consequences on any mixed team. Applying “is it reliable?” to a qualitative strand, or “is it credible?” to a survey, imports the wrong standard — and Section 8 keeps them separate deliberately.

These are caricatures, and useful ones. Almost no working researcher holds either position in the pure form on this slide; the point is to recognise which pole a disagreement is coming from.

ImpactMojoMixed Methods 101www.impactmojo.in

The Objection

The incompatibility thesis

Incompatibility thesis

The claim that quantitative and qualitative methods rest on opposed, irreconcilable paradigms — so combining them in one study is philosophically incoherent.

If methods are welded to worldviews, the argument runs, you cannot honestly mix them. This was a central charge in the 'paradigm wars' of the 1980s.

ImpactMojoMixed Methods 101www.impactmojo.in

The Answer

Pragmatism: the home of mixed methods

Mixed-methods scholars — notably Tashakkori & Teddlie — answered with pragmatism: choose methods by what works for the research question, not by allegiance to a paradigm.

The research question is more important than either the method or the philosophical worldview that underlies it.

— the pragmatist stance, after Tashakkori & Teddlie

Pragmatism asks first	Rather than
What is the question?	What is my paradigm?
What would answer it?	Which method am I trained in?
What follows for action?	What is true in the abstract?

Pragmatism is the working philosophy of most mixed-methods research, and its appeal is obvious: it licenses the combination without requiring anyone to abandon a position.

The objection is worth knowing. Critics argue “whatever works” smuggles in an unexamined view of what counts as working, and that a philosophy which avoids the question is not the same as one that answers it.

ImpactMojoMixed Methods 101www.impactmojo.in

Pragmatism

What pragmatism actually claims

Knowledge is judged by its consequences and usefulness
The forced choice between objective and subjective is false
Methods are tools — pick the fit-for-purpose one
What 'works' to answer the question is the test of a good design

Pragmatism does not deny paradigms exist — it denies they must dictate your toolkit. The question is sovereign.

Pragmatism holds	It does not hold
The question drives the method	That method choice is arbitrary
Knowledge is judged by consequences	That truth is whatever is convenient
Both strands can inform one inquiry	That their assumptions are identical
Inquiry is always situated	That situatedness excuses poor method

The right-hand column is what pragmatism gets accused of, usually by people who have met it as a one-line justification in a methods chapter rather than as a philosophical position.

The useful test: can you say what would count as being wrong? A pragmatist answer that cannot is doing the thing the critics describe.

ImpactMojoMixed Methods 101www.impactmojo.in

Stances

More than one way to hold the paradigms

A-paradigmatic / pragmatist

Set the philosophy aside; let the question and the practical payoff decide. The mainstream mixed-methods position.

Dialectical

Deliberately hold opposing paradigms in tension, mining the friction for insight — close kin to 'initiation'.

Other researchers adopt a transformative stance, foregrounding equity and the standpoint of marginalised groups throughout the design.

Stance	Position
Purist	The paradigms are incompatible; do not mix
Pragmatist	Choose by the question; set the dispute aside
Dialectical	Hold both, and use the tension productively
Transformative	Start from justice; design to shift power

The dialectical stance is the most interesting for evaluation. Rather than dissolving the disagreement, it treats the friction between strands as a source of insight — which is exactly the initiation purpose in Section 2.

The transformative stance is the one most relevant to this sector. It asks whose interests the study serves and whether participants have any say in the questions — the same ground as ImpactMojo’s Decolonising Development course.

ImpactMojoMixed Methods 101www.impactmojo.in

Common Ground

The two strands share more than they fight over

Both seek to reduce error and rule out alternative explanations
Both rely on systematic, transparent procedures
Both can be done well or badly — rigour is method-agnostic
Both ultimately serve better understanding for decisions

Once you see the shared commitments, combining the methods looks less like a contradiction and more like good sense.

Both traditions accept
Evidence should be systematically collected and documented
Claims must be checkable by someone else
The researcher’s position affects what is produced
Sampling decisions determine what can be said
Findings can be wrong, and the design should allow that to show

The shared ground is larger than the disputes, which is worth saying on a mixed team where the two strands are usually staffed by people trained to see each other as sloppy or naive.

The last row is the common standard. A qualitative study designed so that no evidence could contradict its framing has the same problem as a trial with no comparison group, and both traditions would say so.

ImpactMojoMixed Methods 101www.impactmojo.in

Takeaway

You can mix — with eyes open

The incompatibility thesis is answered, not ignored. You may mix methods coherently — provided you can say why your question needs both and how you will reconcile what they tell you.

Philosophy earns its keep here: it forces you to be explicit about your stance, so reviewers can judge the study on its own terms.

Mixing with eyes open means	Concretely
Naming your stance	One sentence in the methods section
Holding each strand to its own standard	Section 8’s three quality questions
Not treating qual as weak quant	No “only 12 interviews, so indicative”
Expecting disagreement	And planning what you will do with it

The third row is a tell worth watching for in your own writing. Describing a qualitative strand as “indicative” or apologising for its sample size applies a sampling logic it never claimed — the strand was chosen for depth, and depth has no n.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Four

The Core Designs

ImpactMojoMixed Methods 101www.impactmojo.in

A Shared Language

The notation of mixed methods

Creswell & Plano Clark gave the field a compact notation for describing designs. Learn these four marks and you can write any design in a line.

Symbol	Meaning
QUAN / QUAL (caps)	The dominant, prioritised strand
quan / qual (lower)	The secondary, supporting strand
+ (plus)	Strands run concurrently, at the same time
→ (arrow)	Strands run sequentially, one then the next

Two questions define every design: timing (+ or →) and priority (which is in caps).

Notation	Reads as
QUAN + QUAL	Both at once, equal priority
QUAN → qual	Quant first and dominant; qual follows, secondary
QUAL → quan	Qual first and dominant; quant follows
QUAN(qual)	Qual embedded inside a dominant quant design
Capitals	The dominant strand; lower case the supporting one

The notation is worth learning because it forces two decisions that otherwise get made by accident: which strand leads, and whether they run together or in sequence. Writing QUAN → qual commits you in a way that “mixed methods” does not.

Put it in the proposal. A reviewer can see immediately what you intend, and a team can see what they have agreed — which prevents the drift Section 10 ends with.

ImpactMojoMixed Methods 101www.impactmojo.in

Design 1

Convergent parallel: QUAN + QUAL

Both strands run at the same time, with equal weight; results are merged and compared at interpretation. Best for triangulation.

ImpactMojoMixed Methods 101www.impactmojo.in

Design 2

Explanatory sequential: QUAN → qual

Quant comes first and leads; qualitative follow-up explains the numbers — surprising results, outliers, group differences. The most common design in evaluation.

ImpactMojoMixed Methods 101www.impactmojo.in

Design 3

Exploratory sequential: QUAL → quan

Qualitative comes first and leads; it builds a grounded instrument or typology the quantitative strand then tests at scale. Matches the 'development' purpose.

ImpactMojoMixed Methods 101www.impactmojo.in

Design 4

Embedded: one strand inside another

A supportive strand is nested inside a dominant design — classically a qualitative process evaluation embedded within an experiment, asking a different but related question.

ImpactMojoMixed Methods 101www.impactmojo.in

Four Designs

The core designs, compared

Design	Notation	Timing	Best for
Convergent parallel	QUAN + QUAL	Concurrent	Triangulation, fuller picture
Explanatory sequential	QUAN → qual	Sequential	Explaining numbers
Exploratory sequential	QUAL → quan	Sequential	Building instruments
Embedded	QUAN(qual)	Either	Adding a different sub-question

Source: Creswell & Plano Clark, Designing and Conducting Mixed Methods Research. Real studies often combine or extend these.

Design	Use when	Main risk
Convergent QUAN + QUAL	You want corroboration or a fuller picture	No real integration at the end
Explanatory QUAN → qual	You have results that need explaining	Phase two under-planned and rushed
Exploratory QUAL → quan	You need to build a measure or know the terrain	Phase one runs long; phase two shrinks
Embedded	One strand answers a subordinate question	The embedded strand is treated as decoration

Each design has a characteristic failure, and each is a scheduling failure. Convergent designs fail at the join; sequential designs fail because the second phase inherits whatever time the first left.

Sequential designs need a protected budget for phase two, agreed before phase one begins. Otherwise the follow-up interviews that were the whole point become six rushed calls in the final week.

ImpactMojoMixed Methods 101www.impactmojo.in

Beyond the Four

Complex & multiphase designs

The four are building blocks. Large programmes chain them into multiphase designs — an exploratory phase to build tools, a convergent phase to evaluate, an explanatory phase to interpret — across years.

You are not picking one design forever. You are choosing the right pattern for this question, then connecting patterns as a programme matures.

Complex design	Where it appears
Multiphase	A programme evaluated across several rounds and years
Mixed within a case study	Several sites, each with both strands
Mixed within an experiment	An RCT with an embedded process strand
Mixed within participatory work	Community-set questions, mixed evidence

Most real evaluations are the third or fourth row rather than a textbook four-square design, and that is fine — the four core designs are a vocabulary for describing what you are doing, not a menu you must choose from.

The more phases, the more integration points, and each one needs a named person and a scheduled week. Complexity multiplies the number of places where the strands can quietly stop talking.

ImpactMojoMixed Methods 101www.impactmojo.in

Read a Notation

Practice: what does this study do?

QUAL → QUAN

Sequential, equal weight, qual first. Explore to build a survey, then weight both strands equally at interpretation.

quan + QUAL

Concurrent, qualitative-led, with a smaller supporting quantitative strand running alongside.

Caps = priority, position = order, the connector = timing. Three marks, the whole design.

Notation	What the study does
QUAL → QUAN	Qualitative first, both equally weighted — e.g. build a scale, then validate it
quan → QUAL	A small survey to select cases, then the main qualitative study
QUAN(qual)	A trial with an embedded process evaluation
QUAL + quan	An ethnography with a supporting descriptive survey

The second row is a design many teams run without naming. Using a short survey purely to identify whom to interview is a legitimate and efficient sequential design — and describing the survey as a findings source overstates what it was built for.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Five

Choosing a Design

ImpactMojoMixed Methods 101www.impactmojo.in

Start Here

The question chooses the design

Do not start from 'I want to do mixed methods'. Start from the question, and let three sub-questions point you to a design: which order, which priority, which point of contact.

What do I actually need to know?

→

Which method leads — and does the other follow or run alongside?

→

Which strand carries more weight for this claim?

→

Where will the strands meet?

If the question is	The design is
“Did it work, and why?”	QUAN → qual
“What matters here, and how common is it?”	QUAL → quan
“Do these two sources agree?”	QUAN + QUAL
“Was it delivered as designed?”	QUAN(qual)
“What do people here think a good outcome is?”	QUAL, possibly alone

The last row is the honest one. Sometimes the answer is a single strand done well, and a course on mixed methods should say so — adding a survey to a question that does not need one is Section 10’s fifth pitfall.

Write the question as one sentence first. If it contains “and why” or “and how common”, the design follows from the conjunction, and the order follows from which half you already know.

ImpactMojoMixed Methods 101www.impactmojo.in

Decision 1

Which order? Timing of the strands

Concurrent ( + )

Collect both at once. Faster; suits triangulation. But the strands cannot inform each other's design.

Sequential ( → )

One finishes before the next begins, so A shapes B. Slower; suits building and explaining.

If you need one strand to design the other, you must go sequential — time it into the workplan early.

	Concurrent	Sequential
Time	Shorter overall	Longer — phases add up
Staffing	Two teams at once	Can reuse one team
Can one inform the other?	No	Yes — that is the point
Risk	Strands drift apart	Phase two gets squeezed

Timing is usually decided by the calendar rather than the question, which is the wrong way round but is the reality of commissioned work. If the deadline forces concurrent, say so in the methods rather than describing it as a design choice.

Concurrent designs need a scheduled meeting, not just a shared deadline. Two teams working simultaneously with no contact until the report stage is the most reliable route to Section 10’s parallel play.

ImpactMojoMixed Methods 101www.impactmojo.in

Decision 2

Which priority? Weighting the strands

Priority is which strand carries the main analytic load — signalled by the CAPS in the notation. It follows from the question, not from which method you find easier or fund more.

QUAN-led: the headline claim is about magnitude or impact
QUAL-led: the headline claim is about meaning or process
Equal: both claims matter and stand together

A frequent failure is unstated priority: a qual strand quietly demoted to decoration. Decide weighting openly, in advance.

Priority is signalled by	Which is why
Which strand the research question centres	The question decides, not the budget
Which gets more of the budget and time	Stated intent loses to resourcing
Which drives the conclusions	Visible in the report’s structure
Which is described first and at length	Readers infer priority from space

Priority is set by resourcing whether or not anyone decides it. A design declared equal-weight, with 80 per cent of the budget on the survey, is a QUAN(qual) design that has not admitted it — and the qualitative strand will be thin for reasons nobody wrote down.

Decide it explicitly and write it in the notation. Equal priority is the hardest to sustain and the most often claimed; if you cannot protect the budget for both, choose which leads.

ImpactMojoMixed Methods 101www.impactmojo.in

Decision 3

Where will the strands meet?

Every mixed design must name its point(s) of integration — where numbers and words actually connect. A design with no such point is multi-method, not mixed.

At design: one strand builds the other's sampling/instrument
At data: one dataset is transformed to join the other
At analysis / interpretation: merge, compare, joint display

Point of interface	What happens there
Design	One strand shapes the other’s instrument or sample
Data collection	Same respondents, linked identifiers
Analysis	Joint display; data transformation
Interpretation	Meta-inferences drawn across both

The second row is the one that has to be planned at the start and cannot be added later. If the qualitative sample cannot be linked to survey records, most of the interesting integration in Section 7 becomes impossible — and the linking key must exist before either strand collects anything.

Name the interface points in the proposal. “The strands will be integrated at analysis” is not a plan; “a joint display comparing survey uptake with interview-reported barriers, by wealth quintile” is.

ImpactMojoMixed Methods 101www.impactmojo.in

A Chooser

If your aim is… pick…

Your aim	Likely design
Corroborate a finding two ways	Convergent parallel (QUAN + QUAL)
Explain surprising survey results	Explanatory sequential (QUAN → qual)
Build a context-valid instrument	Exploratory sequential (QUAL → quan)
Understand how a trial played out	Embedded (QUAN with qual process strand)
Provoke new questions from tension	Concurrent, dialectical stance

Aim	Pick	Because
Explain a surprising result	QUAN → qual	You need the result before you know whom to ask
Build an instrument for this context	QUAL → quan	Items must come from local accounts
Check two sources against each other	QUAN + QUAL	Independence requires simultaneity
Understand delivery inside a trial	QUAN(qual)	The process question is subordinate
Identify whom to study in depth	quan → QUAL	A small survey is a sampling tool

This table is the deck compressed into a decision. Find your aim on the left; the notation on the right is what belongs in the proposal, and the reason column is what you say when a reviewer asks why.

ImpactMojoMixed Methods 101www.impactmojo.in

Constraints

Be honest about feasibility

Time: sequential designs need two field rounds — budget the gap
Team: do you have both quant and qual skills, or must you hire?
Money: mixing roughly doubles instruments, training, analysis
Access: can you return to the same site for the second strand?

An elegant design you cannot resource becomes a poor one in practice. Match ambition to capacity.

Constraint	What it rules out
Six-month timeline	Sequential designs with a real phase two
One researcher	Concurrent strands; equal priority
No qualitative analysis skill on the team	QUAL-dominant designs
Fixed donor indicators	Exploratory instrument-building
Fieldwork window in one season	Anything needing two visits

Choosing a design you cannot resource is the commonest planning error here, and it does not fail visibly — it fails as a second phase that shrinks until it is six phone calls, reported as though it were the design.

A smaller design done properly is the better answer. An embedded qualitative strand of fifteen well-conducted interviews, integrated at analysis, is worth more than an ambitious sequential design that ran out of time.

ImpactMojoMixed Methods 101www.impactmojo.in

Write It Down

Specify the design before fieldwork

Pin the design in your protocol: the notation, the purpose, the priority, the timing and — above all — the planned point of integration. This is your defence against drift.

A one-line design statement — 'QUAN → qual explanatory, integrated via joint display' — tells a reviewer in seconds what you are doing and why. And document deviations as they happen: an honest account of how the design evolved beats a too-tidy one.

Specify before fieldwork	Example
Purpose	Complementarity, plus development in phase one
Notation	QUAN → qual
Priority and why	Quant leads; the commission is an outcome question
Interface points	Sampling, and a joint display at analysis
Who integrates, and when	Named person; a scheduled week
What happens if they diverge	Agreed in advance, not negotiated late

The last two rows are what separate a design from an intention. Integration has no deliverable of its own, so unless a person and a week are named, it is the task that gets absorbed by whatever is late.

Six lines in the proposal. They cost an afternoon, they answer the questions a reviewer will ask, and they are the document the team returns to when the strands start to drift.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Six

Sampling in Mixed Methods

ImpactMojoMixed Methods 101www.impactmojo.in

Two Logics

Sampling pulls in opposite directions

Quant sampling

Probability, larger n, aims for representativeness so findings generalise to a population.

Qual sampling

Purposive, smaller n, aims for information-richness — cases chosen because they illuminate the question.

Mixed methods must hold both logics at once — and decide how the two samples relate.

	Quantitative logic	Qualitative logic
Goal	Represent a population	Illuminate a phenomenon
Selection	Random, or probability-based	Purposive — chosen for what they can show
Size driven by	Precision required	Saturation, and depth
Bad sample looks like	Unrepresentative	Uninformative — a different failure

The logics genuinely conflict, which is why sampling is the hardest part of a mixed design. Randomly selecting interviewees from a survey sample satisfies one logic and defeats the other — you get a representative dozen who may have nothing informative to say.

The fourth row is the one to hold in your head. Judging a qualitative sample by representativeness applies the wrong standard, and it is the standard a quantitatively trained reviewer will reach for by default.

ImpactMojoMixed Methods 101www.impactmojo.in

Key Choice

Concurrent vs sequential sampling

Concurrent: draw both samples independently, around the same time — they need not overlap
Sequential: draw the second sample from the results of the first — the order creates the link

Sequential sampling is where mixing earns its keep: the first strand tells you whom the second should talk to.

Concurrent sampling	Sequential sampling
Two samples drawn independently	Second sample drawn from the first
Preserves independence for triangulation	Enables explanation of specific results
Cannot link at the individual level	Linked, if an identifier was kept
Suits QUAN + QUAL	Suits QUAN → qual

The third row is a decision that must be taken before the survey runs. Individual-level linking requires an identifier and a consent that covers follow-up; neither can be added afterwards, and without them the strands can only be compared in aggregate.

Independence and linkage pull against each other. Nested sampling gives explanatory power and costs the independence triangulation needs — which is another reason to settle the purpose in Section 2 before the sampling plan.

ImpactMojoMixed Methods 101www.impactmojo.in

Nested Samples

Drawing the qual sample from the quant

Nested sample

A sample for one strand drawn as a subset of the other strand's sample — e.g. interviewing a purposive selection of survey respondents.

In an explanatory design, you might survey 1,200 households, then interview 30 chosen from them — deliberately spanning high and low outcomes to explain the spread.

Nesting lets you link a person's numbers to their words — the tightest form of integration at the case level.

Requirement for nesting	Consequence if missing
A retained identifier	No individual-level joining
Consent covering recontact	Follow-up is not permissible
Contact details, kept securely	Cannot locate the selected cases
Selection criteria decided in advance	Cases chosen to confirm what you already believe

The fourth row is the integrity requirement. Choosing follow-up cases after seeing which would make the best story is the qualitative equivalent of the forking paths problem, and it is invisible in the write-up unless the criteria were stated first.

The second and third rows are data-protection questions, not just logistics. Retaining names and numbers for recontact changes what your dataset is and what obligations attach to it — see ImpactMojo’s Digital Ethics course.

ImpactMojoMixed Methods 101www.impactmojo.in

Selecting Cases

Whom to follow up — on purpose

Extreme cases: the biggest gainers and the non-responders
Typical cases: households near the median, to ground the norm
Confirming / disconfirming: cases that test the emerging story
Maximum variation: span the range deliberately

The survey gives you a sampling frame for the interviews that no convenience approach could match — use it.

Selection strategy	Answers
Extreme cases — highest and lowest outcomes	What distinguishes success from failure
Typical cases — near the median	What the ordinary experience is
Deviant cases — those the model got wrong	What the model is missing
Maximum variation	The range of experience across the sample
Confirming and disconfirming	Whether an emerging explanation holds up

The third row is the most productive and the least used. Households the quantitative model predicted badly are precisely where an unmeasured factor is operating, and interviewing them is the fastest route to finding what the survey left out.

State the strategy in the methods section. “Twenty interviews were conducted” tells a reader nothing; “deviant cases, selected as the twenty households with the largest residuals” tells them what the strand can support.

ImpactMojoMixed Methods 101www.impactmojo.in

Sizing

How big should each strand be?

Quant: powered

Size for the precision or power the claim needs — margin of error, effect size, subgroups.

Qual: saturation

Size for saturation — keep sampling until new interviews stop yielding new themes.

Do not judge the qual strand by quant logic. 'Only 25 interviews?' is the wrong question if those 25 reached saturation.

Strand	Size driven by	Typical range
Survey	Precision needed for the smallest subgroup you will report	Hundreds to thousands
In-depth interviews	Saturation — when new cases stop adding	Often 15–40
Focus groups	Coverage of relevant groupings	Often 6–12 groups
Case studies	Depth achievable within the budget	A handful

The ranges are conventions, not rules, and saturation is a judgement made during fieldwork rather than a number set in advance. Committing to “30 interviews” in a proposal is a budgeting necessity and should not be mistaken for a methodological standard.

The first row is where subgroup ambition bites. A survey powered for a district estimate cannot support the caste-by-gender breakdown someone will ask for later, and adding it after the fact is not possible.

ImpactMojoMixed Methods 101www.impactmojo.in

Frame Matters

Mind the gap between the two frames

If the survey frame and the interview frame differ — different villages, different eligibility — your integration compares apples and oranges. Make the relationship between frames explicit.

Best practice in development MEL: sample the qualitative cases from within the quantitative sites, so context and outcomes line up.

Frame mismatch	Effect on integration
Survey covers households; interviews cover individuals	The unit differs, so findings cannot be joined directly
Survey lists registered beneficiaries; interviews include non-users	Different populations; comparison is invalid
Survey in one season; interviews in another	Seasonal difference read as a strand difference
Different geographic boundaries	Aggregates do not line up

The third row produces false divergence and is easy to miss. Two strands describing the same district in different seasons will disagree about food adequacy, work and migration — and a team looking for an explanation of the disagreement will find one that is not there.

Record both frames explicitly in the methods, including dates. When the strands diverge, the first thing to check is whether they were describing the same population at the same time.

ImpactMojoMixed Methods 101www.impactmojo.in

Equity Lens

Sample to see the excluded

A representative survey can still drown out small, marginalised groups. Purposive qualitative sampling is how you deliberately over-listen to those the averages bury — Dalit, Adivasi, disabled, migrant respondents.

Mixed methods can correct for whom the numbers miss — but only if you design the qual sample to do exactly that.

Who a standard design misses	How to reach them
Households absent at survey time	Revisit at different hours; ask neighbours who is away
Non-participants and dropouts	Sample deliberately from those who left
Women, where a male head answers	Separate interviews, separate space
People with disabilities	Accessible formats; do not rely on proxy report
Migrants	Ask at origin and destination, or accept the gap in writing

The second row is the one mixed designs are best placed to fix. A programme survey covers participants by construction; a qualitative strand can go to the households that stopped coming, which is where the most useful evaluation evidence usually sits.

Write down who you could not reach. It costs a sentence, it is the only record the gap will ever have, and it prevents a later reader from taking your sample as the population.

ImpactMojoMixed Methods 101www.impactmojo.in

Plan It

A sampling plan for both strands

State each strand's frame, method and target size, and the rationale
State whether the samples are independent, nested or identical
If sequential, state the rule for selecting the second from the first
Name how the two samples will be linked at analysis

The sampling plan states	For each strand
Population and frame	And where they differ
Selection method	Random, stratified, purposive — and which purposive strategy
Size, and what determined it	Precision, or saturation
Relationship between the two samples	Nested, or independent
Who will be missed	Stated, not discovered later

One page, written before fieldwork. The fourth row is the one unique to mixed methods and the one most often left implicit — and it determines whether the integration in Section 7 is possible at all.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Seven

Integration: the Heart

ImpactMojoMixed Methods 101www.impactmojo.in

The Core Idea

Integration is what makes it mixed

Integration

The deliberate point(s) at which the quantitative and qualitative strands are brought together — connected, merged or embedded — so that the combination yields more than the strands apart.

If you can remove either strand and the conclusions barely change, you never really integrated. Integration is the test of a genuine mixed-methods study.

Without integration you have	With it you have
Two studies in one binding	One study with two kinds of evidence
Findings that sit side by side	Findings that could not be reached alone
Conclusions from the dominant strand	Conclusions that survived a second source

Integration is the defining act, and it is the step most likely to be skipped — because it belongs to nobody’s workplan, produces no deliverable of its own, and happens at the point in a project where time has already run out.

Which is the argument for naming a person and a week. Everything else in this deck is design; integration is the only part that is purely a scheduling decision.

ImpactMojoMixed Methods 101www.impactmojo.in

Three Ways

Connecting, merging, embedding

Approach	How it works	Typical design
Connecting	One strand's results lead into the next	Sequential designs
Merging	Bring both datasets together to compare	Convergent parallel
Embedding	One strand nested to support the other	Embedded design

These map onto the designs: the design you chose already implies how you will integrate.

Way	What happens	Fits
Connecting	One strand’s results feed the other’s design or sample	Sequential designs
Merging	Both datasets brought together for joint analysis	Convergent designs
Embedding	One strand sits inside the other throughout	Embedded designs

Merging is the hardest and the most valuable. Connecting happens almost automatically in a sequential design; merging requires someone to sit with both datasets and construct the comparison, which is what the joint display on the next slide is for.

ImpactMojoMixed Methods 101www.impactmojo.in

Tool 1

The joint display

A joint display is a single table or figure that arrays quantitative and qualitative results side by side around a common organising principle — the workhorse of integration.

Sub-group	QUAN: outcome	QUAL: why (illustrative)	Meta-inference
High-uptake block	Uptake ~80%	"The ASHA visits every week"	Frontline contact drives uptake
Low-uptake block	Uptake ~35%	"The centre is two hours away"	Distance, not awareness, is the barrier

The right-most column — the meta-inference — is the integration. The numbers and quotes only set it up.

Finding	Survey	Interviews	Read together
Uptake	44%	“The fee is small but it is cash”	An unrecorded cost gates access
Timing	Attendance −30pp in harvest	“Nobody can come in those weeks”	Confirms; suggests a calendar fix
Satisfaction	82% satisfied	Persistent complaints about staff	Divergence — investigate the question wording

A joint display is a table with a row per finding and a column per strand, plus a final column for what the two say together. It is the single most useful integration tool in this deck, and it takes an afternoon.

The last column is the work. Filling the strand columns is transcription; writing what the row means jointly is where the meta-inference happens — and a display whose last column restates the other two has not integrated anything.

ImpactMojoMixed Methods 101www.impactmojo.in

Tool 2

Following a thread

Following a thread starts from a striking finding in one strand and traces it deliberately through the other — weaving a single line of argument across both datasets.

Spot a surprising survey result (a sharp gender gap)

→

Pull the thread into the interviews on that theme

→

Return to the data to test what the interviews suggest

→

Report one integrated narrative, not two

Step	Example
Pick a finding from one strand	Survey: women in the poorest quintile attend least
Follow it into the other	Interviews: what did those women say?
Follow it back	Does the reason they gave show up as a variable?
Report the thread, not just the ends	Including where the trail went cold

Following a thread is the narrative version of a joint display, and it is often easier to write. It also surfaces the useful negative result: a reason given repeatedly in interviews that has no corresponding variable in the survey is a gap in the instrument.

Do it for three or four findings, not everything. A thread per major finding is enough; attempting it exhaustively produces a report nobody finishes.

ImpactMojoMixed Methods 101www.impactmojo.in

Tool 3

Data transformation

Quantitising: turn qualitative codes into counts — 'how many interviewees raised distance?' — to merge with survey data
Qualitising: turn quantitative profiles into narrative typologies — 'the reluctant adopter' — to compare with cases

Powerful but lossy: counting themes can strip their meaning. Transform with care, and keep the original alongside.

Transformation	What it does	Risk
Quantitising	Count how many interviewees raised a theme	Implies a sampling logic the strand never had
Qualitising	Build narrative profiles from survey clusters	Reads more into a cluster than it holds

Quantitising is common and frequently misleading. “12 of 20 interviewees mentioned the fee” invites the reader to treat 60 per cent as an estimate, when the twenty were purposively selected and the eight who did not mention it were never asked directly.

If you count themes, say what the count is not. One sentence — “these are counts within a purposive sample and are not prevalence estimates” — keeps the useful pattern and blocks the wrong inference.

ImpactMojoMixed Methods 101www.impactmojo.in

When They Agree

Convergence builds confidence

When both strands point the same way, you have convergence — the strongest, most reassuring outcome of integration. The claim now rests on two independent legs.

Report convergence explicitly: 'survey and interviews independently identified distance as the binding constraint' is a stronger sentence than either finding alone.

Convergence is meaningful when	And weak when
The strands were independent	Same respondents, same visit, same enumerator
Disagreement was possible	Both instruments framed the question the same way
Both strands were well executed	One is thin and simply echoes the other

Convergence increases confidence in proportion to how surprising it is. Two strands that could not have disagreed have not corroborated anything, and a report presenting that as triangulation has overstated its own evidence.

Say what would have counted as divergence. A sentence naming the disagreement you were prepared to find makes the agreement you did find worth something to a reader.

ImpactMojoMixed Methods 101www.impactmojo.in

When They Clash

Divergence is a finding, not a failure

When strands disagree, resist the urge to pick a winner or bury the conflict. Dissonance often exposes a flaw in one measure — or a deeper truth the averages hid.

Re-examine both measures: which question were they really answering?
Look for the subgroup or context that reconciles them
Report the tension honestly — it is frequently the best insight

Possible reason for divergence	Check
The strands measured different things	Compare the exact wording and the construct
Different populations or periods	Compare the frames and the dates
Social desirability in one strand	Who asked, in whose presence?
One strand is wrong	Check execution before concluding anything
Both are right; reality is complex	The most interesting case — report it

The satisfaction case is the standard illustration. Surveys routinely record high satisfaction while interviews record sustained complaint, and the usual explanation is the third row: a stranger with a form, asked in a public space, gets the answer that is safe to give.

Divergence is a finding, and it is the finding most likely to be dropped. Report it, work through these five, and say which explanation you settled on — that paragraph is usually the most valuable in the report.

ImpactMojoMixed Methods 101www.impactmojo.in

The Payoff

Meta-inferences: the 1 + 1 = 3

Meta-inference

An overall conclusion drawn by integrating the inferences from both strands — a claim that neither the quantitative nor the qualitative analysis could have reached alone.

The meta-inference is the deliverable of mixed methods. If your write-up has none — only a results section per strand — the integration has not happened yet.

A meta-inference is	It is not
A conclusion drawn across both strands	A summary of each in turn
Something neither could support alone	The quant finding with a quote attached
Explicitly labelled as such	Left for the reader to assemble

Meta-inferences are the deliverable of a mixed design. If your conclusions section could be written from one strand alone, the mixing produced nothing, whatever the methods section says.

Write them as their own subsection. Naming them — “drawing on both strands” — makes the integration visible to a reader and to a reviewer, and it forces you to check that you actually have any.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Eight

Quality & Validity

ImpactMojoMixed Methods 101www.impactmojo.in

A Harder Bar

Three quality questions, not one

A mixed study must satisfy quality on three fronts: the quant strand on its own terms, the qual strand on its own terms, and — uniquely — the quality of the integration itself.

Is the QUAN strand rigorous? (validity, reliability, power)

→

Is the QUAL strand rigorous? (credibility, transferability)

→

Is the MIXING rigorous? (does integration add genuine value?)

Question	Standard applied
Is the quantitative strand sound?	Its own: sampling, measurement, analysis
Is the qualitative strand sound?	Its own: credibility, transferability, audit trail
Is the mixing sound?	Legitimation — the third bar, unique to mixed work

The third question is the one reviewers skip and the one that decides whether the study was mixed at all. A report can pass both strand-level checks and still be two studies stapled together, which no single-strand standard would detect.

All three bars are higher than a single-method study, which is the honest cost of mixing: you must satisfy two traditions and a third standard that neither tradition teaches.

ImpactMojoMixed Methods 101www.impactmojo.in

Each Strand

Hold each to its own standard

Concern	Quantitative term	Qualitative term
Truth value	Internal validity	Credibility
Applicability	External validity	Transferability
Consistency	Reliability	Dependability
Neutrality	Objectivity	Confirmability

Do not impose quant criteria on the qual strand or vice versa. Each tradition has its own well-developed standards — use them.

Quantitative asks	Qualitative asks
Is the sample representative?	Were the cases well chosen for the question?
Is the measure valid and reliable?	Are the accounts credible and well evidenced?
Do the results generalise?	Is enough context given to judge transferability?
Are the analyses reproducible?	Is there an audit trail from data to claim?

The rows are parallel and not translations. Transferability is not generalisability with a softer name: it places the judgement with the reader, which is why a qualitative report owes thick context rather than a confidence interval.

The last row is under-supplied in practice. An audit trail — how codes were derived, who coded, how disagreements were settled — is the qualitative equivalent of a reproducible script, and it is omitted far more often.

ImpactMojoMixed Methods 101www.impactmojo.in

The Mixing Standard

Legitimation

The validity framework specific to mixed methods (Onwuegbuzie & Johnson): the extent to which combining the strands yields credible, defensible meta-inferences.

Where single-method work asks 'is this valid?', mixed methods adds: 'is the integration legitimate?' — a distinct question Tashakkori & Teddlie also pushed the field to formalise.

Legitimation asks	Failure looks like
Were the samples compatible?	Comparing populations that do not overlap
Did integration actually happen?	Two sets of findings, no joint claim
Were meta-inferences warranted?	A conclusion neither strand supports
Was divergence handled?	Contradictions absent from the report

Legitimation is the mixing-specific quality standard, and the term is worth using precisely because it names something the two traditions have no word for — the soundness of the join rather than of either side.

Four questions, checkable by a reader. Run them over your own draft before submission; the fourth is the one most reports fail, and fixing it usually means adding a paragraph you were tempted to leave out.

ImpactMojoMixed Methods 101www.impactmojo.in

Legitimation Types

Ways integration can go wrong

Sample integration: do the two samples support a joint claim?
Weakness minimisation: does each strand cover the other's gaps?
Conversion: is quantitising / qualitising done faithfully?
Political: are conflicting findings reconciled honestly, not by fiat?

Each is a place to interrogate your own study before a reviewer does.

Threat	What goes wrong
Sample integration	The two samples are not comparable, so joint claims do not hold
Sequential	Phase one’s errors are carried into phase two’s design
Conversion	Quantitising or qualitising asserts more than the data holds
Weakness minimisation	A weak strand’s limits are ignored because the other is strong
Paradigmatic	Assumptions of the two traditions are never reconciled or acknowledged

The fourth is the most damaging in evaluation. A strong survey lends borrowed credibility to a thin qualitative strand, and the report’s conclusions rest partly on evidence nobody scrutinised because it arrived in good company.

The second is why sequential designs need a decision point. If phase one had a sampling problem, phase two built on it inherits the problem and compounds it — and by then the design is fixed.

ImpactMojoMixed Methods 101www.impactmojo.in

Two Dimensions

Design quality vs interpretive rigour

Design quality

Is the design suitable, adequate and faithfully implemented? Did the strands actually run as planned, with the integration point intact?

Interpretive rigour

Do the meta-inferences truly follow from the integrated evidence? Are they consistent, plausible and not over-reaching?

After Tashakkori & Teddlie's inference quality: a good design badly interpreted — or vice versa — still fails.

Design quality	Interpretive rigour
Was the design suitable for the question?	Do the conclusions follow from the evidence?
Was it implemented as planned?	Are alternative explanations addressed?
Were the procedures adequate?	Is each claim traceable to a strand?

The two dimensions fail independently. A well-designed study can be over-interpreted, and a compromised design can be reported with exemplary care — and the second is more useful to a reader than the first.

The third row on the right is the practical test. Go through your conclusions and mark each against the strand supporting it; anything unmarked is either a meta-inference to label or an assertion to remove.

ImpactMojoMixed Methods 101www.impactmojo.in

Strengthen It

Practices that raise quality

Pre-specify the design, priority and integration point in a protocol
Use joint displays so integration is visible and checkable
Triangulate — and report divergence, do not hide it
Member-check qualitative findings; report quant uncertainty
Keep an audit trail of decisions across both strands

Practice	Raises
Specify the design before fieldwork	Design quality; prevents drift
Build a joint display	Legitimation; makes integration visible
Report divergence and its resolution	Interpretive rigour
Have both strands read each other’s draft	Catches misused evidence early
Take findings back to participants	Credibility, and it is rarely done

The fourth row is nearly free and disproportionately effective. A quantitative analyst reading the qualitative draft catches over-claims from purposive samples; a qualitative researcher reading the survey chapter catches constructs that do not mean what the variable name suggests.

The fifth is the one this sector should do and does not. Returning findings to participants is both a credibility check and an obligation — see ImpactMojo’s Decolonising Development course on why it is almost never budgeted.

ImpactMojoMixed Methods 101www.impactmojo.in

Reporting

Report the mixing, not just the methods

Reporting guidance (e.g. GRAMMS — Good Reporting of A Mixed Methods Study) asks you to justify the design, describe each strand, and crucially show where and how you integrated.

Common reviewer complaint: a paper reports two strands but never the integration. State the meta-inference explicitly, or the mixing is invisible.

Report	Because a reader cannot infer it
Why you mixed — the purpose	It determines what agreement means
The design, in notation	Order and priority are otherwise invisible
Where the strands met	Otherwise integration is assumed, not shown
What diverged, and how you resolved it	Its absence reads as its non-existence
What each strand could not support	Prevents the wrong conclusion being drawn

Reporting standards for mixed methods exist — GRAMMS is the best known — and they amount to roughly these five points. Most reports describe both methods in detail and none of the mixing.

Five sentences in the methods section. They are what distinguishes a report a reviewer can assess from one they can only take on trust.

ImpactMojoMixed Methods 101www.impactmojo.in

Bottom Line

Quality is mostly about integration

A mixed study can have a flawless survey and superb interviews and still be weak — if the two never genuinely meet. The distinctive quality question is always: did mixing add value, defensibly?

Judge the join, not just the parts. That is what separates mixed methods from two studies stapled together.

If you have time to check one thing	Check
In a proposal	Is there a named integration point, with a person and a week?
In a draft report	Is there a joint display, or a thread followed across strands?
In conclusions	Could this be written from one strand alone?

Quality in mixed methods is mostly about integration, because the strand-level standards are already well taught and well policed. The join is the part nobody owns, and it is the part that makes the design worth its extra cost.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Nine

Worked Examples

ImpactMojoMixed Methods 101www.impactmojo.in

Example 1

An RCT with an embedded process evaluation

A cash-transfer programme is tested with a randomised trial (QUAN). Nested inside is a qualitative process evaluation (qual): how was the cash spent, who in the household decided, what got in the way?

Notation: QUAN(qual), embedded. The trial measures whether it worked; the embedded strand explains why, and flags whether the mechanism matched the theory of change.

Element	Detail
Notation	QUAN(qual)
Purpose	Complementarity, plus explanation of the null result
Quant strand	The trial: did the outcome move?
Qual strand	Process: was it delivered, and how was it received?
Interface	Sampling (sites), and analysis (joint display by site)

The embedded process strand is what makes a null result interpretable. Without it, “no measured effect” is compatible with a wrong theory and with a programme that never happened, and the two call for opposite decisions.

It has to be embedded from the start. Adding process work after a disappointing result is a salvage exercise, and it cannot recover what was happening during the trial period.

ImpactMojoMixed Methods 101www.impactmojo.in

Example 1, cont.

Why the qual strand earns its place

Trial alone says

'Consumption rose by X; the effect was larger for female-headed households.' True — but mute on mechanism.

Embedded qual adds

Women described controlling the transfer for the first time — explaining the heterogeneity the trial only detected.

The meta-inference — 'the effect ran through women's control of the cash' — needed both strands.

What the qual strand can establish	Which the trial cannot
Whether the intervention was delivered	The trial measures outcomes, not fidelity
What participants understood it to be	Compliance can hide misunderstanding
What else changed in the sites	Contamination and co-interventions
Why some sites did better	Heterogeneity has to be explained, not just measured

The fourth row is where the scale-up decision is actually made. Knowing an average effect tells you little about whether it will hold in the next district; knowing which site conditions produced it tells you where to go next.

A process strand costs a fraction of a trial and is the difference between a result and a usable finding. It is also the first line cut when budgets tighten, which is exactly backwards.

ImpactMojoMixed Methods 101www.impactmojo.in

Example 2

A survey explained by follow-up interviews

A district health survey (QUAN) finds institutional delivery is high overall but stubbornly low in three blocks. Purposive follow-up interviews (qual) in those blocks explain the gap — an explanatory sequential design.

Element	Detail
Notation	QUAN → qual
Purpose	Explanation
Case selection	Deviant cases — largest residuals from the model
Interface	Nested sampling; joint display by subgroup
Requirement set at survey stage	Retained identifier; consent to recontact

The last row is the one that has to be got right months earlier. An explanatory sequential design is impossible if the survey did not keep a way to find people again, and that decision is made before anyone knows what will need explaining.

ImpactMojoMixed Methods 101www.impactmojo.in

Example 2, joint display

Reading the two together

Institutional delivery by block (%) — illustrative

Illustrative, patterned on district HMIS-style data

Interviews (illustrative) attributed the three low blocks to a closed sub-centre and night-time transport fears — the meta-inference the bars alone could not give.

ImpactMojoMixed Methods 101www.impactmojo.in

Example 3

Building an empowerment scale, locally

An exploratory sequential study (QUAL → quan): in-depth interviews first surface how women in a region actually describe agency — in mobility, money and voice. Those categories build a survey scale that is then validated at scale.

The payoff: an instrument grounded in local meaning, not imported from another context — the 'development' purpose in action, answering a validity problem head-on.

Phase	What it produces
Qualitative: what does empowerment mean here?	Domains and language from local accounts
Item generation	Draft items in the words respondents used
Cognitive testing	Items that are understood as intended
Quantitative: administer at scale	Reliability, structure, and a usable measure

This is the strongest case for exploratory sequential design in this sector. Imported empowerment and wellbeing scales behave oddly here precisely because their items were generated from a different population’s account of the construct.

The third row is skipped most often and costs least. Sitting with ten respondents while they think aloud through a draft instrument catches more measurement error than any amount of subsequent statistical adjustment.

ImpactMojoMixed Methods 101www.impactmojo.in

Example 4

Triangulating a corruption estimate

A convergent parallel study estimates leakage in a welfare scheme three ways at once: a household survey of benefits received, administrative records of benefits disbursed, and qualitative interviews on how diversion happens.

Where the three converge, confidence is high. Where the survey and records diverge, the interviews explain the gap — turning a discrepancy into the finding.

Source	Bias it carries
Household survey self-report	Under-reporting; fear of consequences
Official records	What was recorded, by those involved
Interviews with intermediaries	Their position in the transaction
Direct observation	Observer effects; small coverage

Triangulation earns its name here because each source is biased in a different direction, so convergence between them is genuinely informative — unlike the same-respondent case on slide 62.

Expect divergence and plan for it. Sources with opposite biases will not agree on a magnitude; the useful output is usually a bounded range plus an account of which source is likely to under- and over-state, not a single number.

ImpactMojoMixed Methods 101www.impactmojo.in

RealWorld Evaluation

Mixing under real constraints

Michael Bamberger and colleagues (RealWorld Evaluation) argue mixed methods is often the most practical response to development reality: tight budgets, short timelines, no baseline, and political pressure.

Use existing secondary data to substitute for a missing baseline
Add a small qual strand to interpret a thin quant one (or vice versa)
Triangulate to compensate when no single method can be done fully

Constraint	Realistic response
No baseline	Reconstruct with recall and records; state the limits
No comparison group	Use qualitative counterfactual reasoning explicitly
Budget for one strand	Embed a small second strand rather than none
Weeks, not months	Fewer sites, more depth; say what was not covered

The RealWorld Evaluation approach exists because textbook designs meet real budgets. Its contribution is not lowering standards but making the compromises explicit — naming what was given up, rather than reporting as if nothing was.

The third row is the practical instruction. Fifteen good interviews embedded in a survey-only evaluation is a large improvement over none, and it is affordable in almost any budget that funds the survey.

ImpactMojoMixed Methods 101www.impactmojo.in

Common Thread

What the examples share

In every case the value lives in the join: the trial needed the mechanism, the survey needed the explanation, the scale needed the local voice, the estimate needed corroboration. None is two studies side by side.

When you sketch your own study, write the meta-inference you hope to reach first — then check both strands are actually needed to get there.

All four examples
State a purpose before a design
Name where the strands meet, and plan it early
Depend on a decision taken months before it mattered
Produce a claim neither strand could make alone

The third row is the one worth carrying away. Every example turns on something settled at design stage — a retained identifier, an embedded strand, a case-selection rule — that could not be added later at any price.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Ten

Common Pitfalls

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 1

Parallel play: no integration

The commonest failure: a quant chapter and a qual chapter that never speak. Two methods were used; nothing was mixed. The reader is left to do the integration the authors avoided.

Diagnostic: delete one strand. If the conclusions survive intact, you had parallel play, not mixed methods.

What the report has	What integration would add
Chapter 4: survey results	A joint display putting each survey finding beside the interview evidence on the same construct
Chapter 5: interview themes	A statement of what the two together support that neither supports alone
Discussion citing both	A meta-inference naming where they agreed, where they did not, and why

The delete-one-strand test is blunt and it works. Cover the qualitative chapter and reread the conclusions. If nothing in them becomes unsupported, the qualitative strand was never load-bearing.

Parallel play is often a writing failure rather than a design failure — the analysis was integrated in the analyst's head and never got onto the page. That is still a failure: the reader cannot check reasoning they cannot see.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 2

Qual as decoration

A regression table dressed up with three illustrative quotes is not integration — it is garnish. The qualitative strand here adds colour but no independent evidence and no meta-inference.

Quotes should change what you conclude, not merely illustrate a number you already had. Otherwise the qual strand is doing no analytic work.

Use of a quote	Doing analytic work?
Illustrates a coefficient already estimated	No — the conclusion is unchanged without it
Explains why the effect is larger in one district	Yes — supplies a mechanism the survey cannot
Shows respondents read the survey item differently	Yes — challenges the measure itself
Opens the section as colour	No — rhetorical, not evidential

Ask what the quote would have to say for you to change the finding. If no possible quote could have changed it, the qualitative data was never evidence in this study — it was illustration.

This is the most common form in evaluation reports written for donors, where quotes are read as warmth rather than argument. The incentive is real; the methodological cost is that the qualitative fieldwork is paid for and then not used.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 3

Unequal weighting by accident

Priority should be a decision, not a side-effect of which strand had more money or a more confident analyst. Too often the survey silently dominates and the interviews are demoted at write-up.

If the qual strand cost a third of the budget but gets one paragraph, ask whether priority was chosen — or just happened.

Symptom	What it usually means
Qual strand summarised in one paragraph	Priority defaulted to the survey at write-up
Interviews used only to design the survey	Sequential exploratory — legitimate, if stated
Sample sizes reported for one strand only	The other strand is not being held to a standard
Qual analyst not in the design meetings	Priority was decided before anyone chose it

Write the priority into the protocol in notation (QUAL→quan, QUAN+qual) before fieldwork. A stated priority can be defended; a drifted one cannot be distinguished from neglect.

Unequal weighting is not itself a fault — most good mixed designs are unequal. The fault is weighting by accident, where budget, seniority or confidence decided what should have been a design question.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 4

Under-estimating skills & time

Skills gap

Few researchers are equally strong in both traditions. A team weak on one side produces a lopsided study with a token strand.

Time gap

Sequential designs need two field rounds and the gap between. Budgets and timelines built for one method under-resource the second.

Plan for both costs up front — or the weaker strand becomes the decoration you did not intend.

Resource	Single-method plan	What mixing actually needs
Fieldwork rounds	One	Two, plus the analysis gap between them
Analysis staff	One tradition	Both, overlapping rather than sequential
Calendar	Fieldwork → analysis	Round 1 → analyse → redesign → round 2
Integration time	None	Weeks — joint displays are analysis, not formatting

The gap between rounds is the part budgets omit. A sequential design where round two starts before round one is analysed has the schedule of a mixed study and the logic of two parallel ones.

A team strong in one tradition and weak in the other does not produce a weaker second strand — it produces a token one, because the weak side cannot tell the difference between thin data and sufficient data.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 5

Forcing methods the question doesn't need

Sometimes a single method answers the question best. Bolting on a second strand 'because mixed methods is rigorous' wastes resources and dilutes focus — methodolatry, not method.

The honest test: can you state a question that genuinely needs both? If not, do one method well rather than two methods thinly.

Question	Needs both?
Did the transfer raise consumption?	No — a well-powered trial answers it
Why did uptake stall in two blocks?	No — qualitative fieldwork answers it
Did it raise consumption, and why did some households refuse?	Yes — neither strand reaches the other half
Does the scale measure what respondents think it measures?	Yes — cognitive interviews plus psychometrics

State the question that needs both, in one sentence, before writing the proposal. If the sentence needs an "and also" to include the second strand, the second strand is probably decoration.

Funders increasingly ask for mixed methods by default, which makes methodolatry a rational response to a bad incentive. Naming that openly in the proposal — and defending a single strong method — is usually received better than the token second strand.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 6

Ignoring divergence

When the strands disagree, the temptation is to quietly trust the numbers and shelve the interviews (or the reverse). Suppressing divergence throws away the study's most informative result.

Reviewers notice a too-tidy convergence. Report disagreement, and what you think it means — that is rigour, not weakness.

Divergence	Bad response	Good response
Survey shows gains; interviews describe none	Report the survey, footnote the interviews	Ask whether the measure captures what respondents value
Interviews report exclusion; admin data shows coverage	Treat interviews as anecdote	Check who the administrative denominator excludes
Strands agree suspiciously well	Report as confirmation	Check whether both drew on the same informants

Convergence from two strands sharing a sampling frame is not convergence — it is one measurement reported twice. Independent error is what makes agreement informative.

Divergence is the finding most likely to be right and least likely to be published. Recording it in the protocol as an expected and reportable outcome makes it harder to quietly drop later.

ImpactMojoMixed Methods 101www.impactmojo.in

Pitfall 7

Vague design, mid-study drift

Without a pre-specified design, priority and integration point, a study drifts: the planned sequential link is skipped, the joint display never gets built, and 'mixed methods' becomes a label, not a method.

A one-line design statement in the protocol is the cheapest insurance against drift — write it before you collect a single data point.

Written before fieldwork	Cost if left implicit
Design in notation	Timing and priority get decided by whoever analyses first
The integration point	Joint display never gets built — no one owned it
What links the strands	Round two is designed without round one's findings
What divergence would mean	Disagreement becomes an inconvenience, not a result

One line in the protocol is the whole intervention: "QUAN→qual explanatory; qualitative sampling drawn from the lowest-scoring quartile; integration by joint display at analysis."

Drift is invisible from inside the study, because each individual shortcut is reasonable under time pressure. The pre-specified line is what lets you notice you took them.

ImpactMojoMixed Methods 101www.impactmojo.in

Avoiding Them

A pre-flight checklist

Can I name the question that needs both strands?
Have I stated the design, priority and timing in notation?
Have I named the integration point — before fieldwork?
Does the team have real skills in both traditions?
Is there time and budget for the second strand, properly?
Will I report divergence as openly as convergence?

Check	Fails if
A question needing both strands	You can answer it with one method well
Design stated in notation	Timing or priority is still undecided at fieldwork
Integration point named	Named after data collection ends
Skills in both traditions	One strand has no specialist in the design meetings
Budget for the second strand	The second strand is funded from what is left over
Divergence reportable	No plan for what to do if the strands disagree

Every item is answerable before any data is collected, which is the point — each one becomes considerably more expensive to fix once fieldwork has started.

Reviewers of mixed-methods submissions reject most often for insufficient integration. Five of these six checks are integration questions asked early enough to still act on.

ImpactMojoMixed Methods 101www.impactmojo.in

Section Eleven

Practice & Tools

ImpactMojoMixed Methods 101www.impactmojo.in

Build the Team

Mixing needs mixed skills

Few people do both traditions equally well, so most strong mixed-methods work is team work — pairing quant and qual specialists who genuinely respect each other's craft.

Bring both specialists into design, not just analysis
Agree the integration point together, early
Make space for the strands to challenge each other's findings

Stage	Both specialists needed?
Framing the question	Yes — this is where priority is really set
Sampling design	Yes — nested sampling needs a retained identifier
Instrument design	Yes — each strand shapes what the other can ask
Fieldwork	Separately, usually
Integration	Yes — a joint display built by one strand is not joint

The common failure is bringing the qualitative specialist in at analysis. By then the sampling frame is fixed and the identifier that would have let you link the strands was never collected.

"Respect each other's craft" is not a courtesy. A team where one tradition privately regards the other as unserious will resolve every divergence in one direction, and no protocol prevents that.

ImpactMojoMixed Methods 101www.impactmojo.in

Software

Tools for each strand — and the join

Tool	Good for	Note
R / Python	Quant analysis, reproducible workflows	Free, scriptable
Stata / SPSS	Survey analysis with weights	Common in MEL shops
NVivo / ATLAS.ti	Coding qualitative data	Support joint-display matrices
Dedoose	Built for mixed-methods integration	Links codes to variables
KoboToolbox / ODK	Collecting both data types in the field	Free, offline-capable

No tool integrates for you. Software stores and links the data; the meta-inference is still yours to make.

Task	What to check before committing
Linking codes to variables	Can the tool export the link, or is it locked in?
Joint-display matrices	Does it produce a display you can publish, or a screenshot?
Team coding	Licence count and whether coding is merge-able
Reproducibility	Is the analysis a script or a sequence of clicks?

No software integrates for you. Dedoose and NVivo make the join easier to record; the decision about what a code and a variable jointly mean is analysis, and it stays with the analyst.

For teams in South Asian NGOs the constraint is usually licence cost and offline fieldwork, not features — KoboToolbox plus R plus a shared spreadsheet joint display is a complete and defensible stack.

ImpactMojoMixed Methods 101www.impactmojo.in

Writing It Up

Structuring a mixed-methods report

State the design and rationale in notation, up front
Report each strand clearly — but do not stop there
Devote a dedicated integration section to joint displays
Lead the discussion with the meta-inferences
Be explicit about convergence and divergence

If a reader can find your meta-inference in thirty seconds, the write-up has done its job.

Section	What belongs there
Methods	Design in notation, priority, timing, integration point, rationale
Results — strand	Each strand on its own terms, judged by its own standards
Results — integration	Joint displays; where the strands met and where they did not
Discussion	Meta-inferences first, strand-level findings as support

The dedicated integration section is what distinguishes a mixed-methods report from two reports. Without it, integration exists only in the discussion, where a reader cannot audit it.

Test the draft on a reader from each tradition. If the quantitative reader skips the qualitative results and loses nothing, the integration section is not doing its work.

ImpactMojoMixed Methods 101www.impactmojo.in

Publishing

Getting mixed work past reviewers

Mixed-methods papers are often rejected for insufficient integration or for a strand judged thin by single-method reviewers. Pre-empt both: justify the mixing and show the join.

Use a reporting standard such as GRAMMS, and consider a journal or section that understands mixed designs — reviewers matched to the method judge it fairly.

Reviewer objection	Pre-empt it by
"The strands are not integrated"	A named integration section with joint displays
"The qualitative sample is too small"	Stating the sampling logic — information power, not representativeness
"The survey is underpowered"	Reporting power for what the survey is actually claiming
"Why mixed methods at all?"	The one-sentence question that needs both strands

GRAMMS is a reporting standard, not a quality standard. It makes your design legible to a reviewer from either tradition, which removes the objections that come from misreading rather than disagreement.

Two of the four objections above come from a reviewer applying one tradition's standards to the other strand. Naming the standard you are applying, in the methods section, is the cheapest defence.

ImpactMojoMixed Methods 101www.impactmojo.in

Reading List

The foundational texts

Creswell & Plano Clark — Designing and Conducting Mixed Methods Research (the designs & notation)
Greene — Mixed Methods in Social Inquiry (purposes & the dialectical stance)
Tashakkori & Teddlie — Handbook of Mixed Methods (pragmatism, legitimation, inference quality)
Bamberger et al. — RealWorld Evaluation (mixing under development constraints)

Text	Read it for
Creswell & Plano Clark	The core designs, notation and worked procedures
Greene	The five purposes and why paradigm stance matters
Tashakkori & Teddlie	Pragmatism, legitimation, inference quality
Bamberger et al.	Mixing under real budget and time constraints

Start with Creswell and Plano Clark if you need to design a study this quarter, and with Greene if you need to justify why you are mixing at all. They answer different questions.

Bamberger is the one written for the conditions most readers of this deck work in — incomplete baselines, compressed timelines, and a design that has to survive contact with a real programme.

ImpactMojoMixed Methods 101www.impactmojo.in

Go Deeper

Where to keep learning

Journal of Mixed Methods Research — the field's home journal
BetterEvaluation — practical mixed-methods guidance for MEL
3ie & JPAL — evaluations that embed qual in trials
GRAMMS reporting criteria — a checklist for your write-up

Pair this deck with ImpactMojo's Qualitative Methods, Data Literacy and Theory of Change 101 courses.

Resource	Best for
Journal of Mixed Methods Research	Current debates on integration and legitimation
BetterEvaluation	Practical guidance written for MEL practitioners
3ie & J-PAL	Examples of qualitative work embedded in trials
GRAMMS criteria	A checklist to run over your own draft

Read published joint displays before building your own. The format varies widely and the good ones are the fastest way to see what a display has to make visible.

Pair this deck with Qualitative Methods 101 for the strand most readers are weaker on, and Theory of Change for the framework that most often supplies the integration point.

ImpactMojoMixed Methods 101www.impactmojo.in

The Takeaways

If you remember five things

Mix on purpose — intentional combination, not 'both at once'
Let the question choose the design — timing & priority
Integration is the heart — joint displays, meta-inferences
Divergence is a finding — never suppress it
1 + 1 = 3 only when the strands genuinely meet

ImpactMojoMixed Methods 101www.impactmojo.in

Mixed Methods 101 · Complete

Now make the
numbers and words
talk to each other.

Mixed methods is not doing both — it is designing the point where quantitative and qualitative evidence meet, so the whole exceeds the sum of its parts. Explore the rest of the ImpactMojo 101 Series, free forever.

More 101 Courses Explore ImpactMojo Dataverse

Continue your learning

Participatory Methods Studio Survey Design Studio MEL Flagship

CC BY-NC-ND 4.0·Free Forever·ImpactMojo 101 Series