Strategies to integrate qualitative and quantitative data in mixed methods research and evaluation. See an overview of the different types of mixed methods research; how NVivo handles combining qualitative and quantitative data and how specific analytical techniques can be used on any project to synthesize and summarize mixed methods data.
Conducting Integrated Mixed Methods Research and Analysis Using NVivo
1. Presenters:
Lisa LeRoy, MBA, PhD
Lisa_leroy@abtassoc.com
Anna Jefferson, PhD
Anna_jefferson@abtassoc.com
Stuart Robertson, Ed.D.
stuart@robertsoneducational.com
September 2017
Conducting Integrated Mixed
Methods Research and Analysis
Using Software
2. Abt Associates | pg 2
Presentation Overview
When are mixed methods most useful and how can
we better integrate qualitative and quantitative
methods? (Lisa LeRoy)
Case Study: Understanding Consumers Financial
Decision Making? (Anna Jefferson)
Mixed Methods Analytical and Reporting Capabilities
of Qualitative Data Analysis Software (Stuart
Robertson)
3. Abt Associates | pg 3
Mixed Methods Evaluation:
Conditions
When different evaluation questions require different
methods, or when a single evaluation question requires
more than one method to answer all components
When different methods are used to answer the same
elements of a single question, increasing confidence in
the validity and reliability of the evaluation results
When the results from one method are used to help
design future phases of the evaluation using other
methods
U.S. AID. (June 2013). Technical Note: Conducting Mixed-Methods Evaluations. Washington, DC.
4. Abt Associates | pg 4
Mixed Methods Considerations
Degree of interaction/independence between
methods
Priority/precedence of each method
Timing; sequential, concurrent, phased
Sample; same sample for different methods, subset,
different samples
Juncture; points at which methods will be mixed
(design, data collection, analysis, etc.)
Creswell and Plano Clark 2011
5. Abt Associates | pg 5
The bottom line…
Use the methods that can best answer the research
and evaluation questions
However, the options are expanding
“Mixing” can occur at each stage of research:
– Study objectives - Analysis
– Design - Reporting
– Data collection - Dissemination
– Interpretation - Reflection and refinement
6. Abt Associates | pg 6
Why the increased attention on
mixed methods?
7. Abt Associates | pg 7
Why the increased attention on
mixed methods?
Emphasis on best available evidence
Technological advances in collecting and analyzing
data (e.g. social media, social network analysis, GIS)
Quantity of available data is increasing
Desire to improve the intervention with evaluation
learnings
Desire to understand intervention “inner workings”
(implementation science)
Understand how changing conditions affect an
intervention (e.g. environment)
8. Abt Associates | pg 8
Parallel versus Integrated
Mixed Methods
Evaluation of program using RCT
Qualitative method with a sample of program
participants or implementers
Results analyzed and reported separately
9. Abt Associates | pg 9
Integrated Mixed Methods
Qualitative and quantitative methods remain
independent
Qualitative and quantitative methods are used in a
study but linkages between them are absent
One type of evidence is privileged over another
Research is driven by political agendas
Burch P and Heinrich C (2016) Mixed Methods for Policy Research and Program Evaluation
10. Abt Associates | pg 10
What are the useful mixed methods
frameworks?
The Qualitative Report - http://nsuworks.nova.edu/tqr
11. Abt Associates | pg 11
Mixed Methods Frameworks
(selected)
A. Tashakkori & C. Teddlie
P. Bazeley (analysis)
J. Creswell & V. Plano Clark (design)
S. Hesse-Biber & B. Johnson
A. Onwuegbuzie & K. Collins (sampling)
P. Burch & C. Heinrich (implementation)
J. Maxwell (theory/philosophy)
14. Abt Associates | pg 14
Case Study: Understanding Consumers
Financial Decision Making
Started with the qualitative data (different than many
mixed methods studies)
Both qualitative and quantitative data were collected
on every participant – full sample for both data types
16. Abt Associates | pg 16
Study Context
Focus group study for Consumer Financial
Protection Bureau (2013-14)
Goal: understand consumer decision-making about
four topics
– Credit reports and scores
– Auto financing and budgeting
– Comparison shopping for financial products
– Financial rules of thumb
17. Abt Associates | pg 17
Data Sources
Qualitative: 32 consumer focus groups
– Boston, St. Louis, Atlanta, and Seattle
– 308 consumers
Quantitative/Close-ended: Survey data
comprised of screening data and questionnaires
– Complete screening data for all 308; self-
administered questionnaires for 299 participants
– 100 data points
– Topics ranged from income/assets to financial
products, financial experiences, information sources,
and psychographics (e.g., stress about money)
18. Abt Associates | pg 18
Two Levels of Analysis
Interpretive analysis: identifying themes based on
transcripts
Attributes analysis: using questionnaire data to
identify patterns and segments among consumers.
Mixed-methods goal: How do thematic findings
intersect with participants’ prior experiences,
attitudes, and characteristics?
19. Abt Associates | pg 19
Analysis Walk-Through
Thematic analysis was done in NVivo 10 for each
topic and cutting across topics
20. Abt Associates | pg 20
Analysis Walk-Through
Transcripts were auto-coded in NVivo so each
participant was a node
Note: all names are pseudonyms.
21. Abt Associates | pg 21
Analysis Walk-Through
Survey data allowed us to associate each participant
node with the participants’ survey data
Note: all names are pseudonyms.
22. Abt Associates | pg 22
Analysis Walk-Through
Auto-coding by participant in the transcripts allowed
every statement by a participant to also be classified
by its 100 close-ended data points
This structure allowed for robust queries in NVivo
that identified patterns based on thematic
statements, participant characteristics, experiences,
and/or attitudes
24. Abt Associates | pg 24
Final Product: Consumer Segments
Defined on behavioral dimension
– Some segments further divided on motivational, situational,
or perceptual factors
Utilized themes from qualitative data and participant
survey data
Process:
– Qualitative analysts conducted first analysis, identified
themes and initial segments; qualitative and quantitative
analysts conferred regularly about findings, confirming
resonance between focus group and survey data, including
statistical analysis (outside NVivo)
25. Abt Associates | pg 25
Sample Thematic Findings Supporting
Segments: Credit Reports and Scores
Respondents self-reported high levels of knowledge of
and engagement with credit reports and scores
– Awareness of credit scores’ implications for lending (generally),
employment, insurance, and housing
– However, many were often confused or incorrect in their
understanding of credit reports and scores
– “Checking credit” means various things to respondents and
may include over-reporting
Consumers may cycle among different styles
– Expected purchases, negative financial events, or other life
changes
– Negative events trigger some people to become proactive
monitors while others continue to be more reactive
26. Abt Associates | pg 26
Sample Thematic Findings Supporting
Segments: Credit Reports and Scores
Binary thinking around credit being “good” or “not so good”
Some more report-driven (i.e., to look for errors) while
others score driven (i.e., to get “good” credit)
Many reported identifying and disputing errors on their
reports
– Dispute process usually considered frustrating and confusing
Confusion, frustration, and uncertainty were common
feelings associated with credit reports and scores
– Most do not feel in control of their credit reports
– Unsure what advice to trust about how to build credit
27. Abt Associates | pg 27
Sample Consumer Segments:
Credit Reports and Scores
Consumer segments for credit reports and scores
Frequency of checking
– Segment 1a: Never checked
– Segment 1b: Past checkers
– Segment 1c: Frequent checkers
How information is sought:
– Segment 1d: Passive (vs. active) checkers
Editor's Notes
MM Designs (Creswell and Plano)
Convergent parallel
Explanatory sequential
Exploratory sequential
Embedded
Transformative
Multiphase
Node classification was done for each participant, whose name became a node under the “respondent” node step shown above. In the panel called “node classifications” here, it is showing all the variables associated with each speaker (all 100 of them—it keeps going). The tab shows the participant names/nodes in the first column, then shows the values for each of their responses. These were uploaded as an excel worksheet. In this shot, it’s showing participants’:
age,
comfort with their level of savings,
comfort with their debts,
preferred mode of paying for everyday transactions,
most important factor to them in choosing a credit card, and
whether they feel stressed by money (coded as confident or vulnerable). This became one of our most interesting variables for understanding variations irrespective of people’s financial situations (based on self-reported income, debt levels, and savings) and prior experience with the financial system (both how experienced they were and if their experiences were predominantly positive or negative).
**NOTE this is confidential because the dataset hasn’t been de-identified yet (working on that with QSR).
This query, for example, shows the same output visualized two different ways, for the query of how people with different levels of financial experience (Level 1-3) and different amounts of stress about money (confident/vulnerable) preferred to receive information about financial products:
In column 1 that participants were classified into three levels of experience (from low to high amounts, based on screening questions), AND that we used their screening questions to split them into money-stressed or not money-stressed (vulnerable/confident for shorthand)
In columns A through D, it shows how people in those groups reported preferring to receive their information about financial products (red is lower counts to green with higher counts). Each cell is showing how many people nodes that’s associated with.
In the top panel, you see the results by the total number of people in each cell. The N’s for each one matter here:
L1 con = 15
L1 vuln = 11
L2 con = 46
L2 vuln = 27
L3 con = 38
L3 vuln = 28
In the bottom panel, you see the row percentage—that is, what percentage of low-experience but low money stress people preferred to receive information in each mode.
One thing that jumps out here is the overall strong preference across these characteristics for talking to company representatives (column D), but that it’s most pronounced for low experience, highly money stressed people (money stressed = “I become stressed about making decisions about money,” not about their objective financial situation). However, moderately experienced but stressed folks were kind of outliers on preferring to get information online while low experience but not stressed folks preferred mobile/apps.
With your data as thoroughly coded (thematically and attributes-wise) as ours was, you could reduce or expand the complexity of the queries an almost infinite amount. However, we weren’t going on a fishing expedition but were using close-ended traits as differentiators that either suggested themselves through prevalent qualitative themes or the literature (i.e., on self-efficacy).