Induction, Deduction Abduction:
Three Legitimate Approaches to Organizational
Research
Paul E. Spector
University of South Florida
November 6, 2015
Roadmap
• Nature of inference
– Exploratory, Confirmatory, Explanatory
•
•
•
•
History
Current state of research approaches
Winds of change
Guide to inductive/abductive methods
–
–
–
–
Data mining/data science
Inductive methods
Writing inductive papers
Evaluating inductive papers
• The way forward
Three Types of Inference
• Deduction
– Reasoning from true premises
– Confirmatory research
• Theory/hypothesis testing
• Induction
– Generalizing from observations
– Exploratory research
• Abduction
– Inference to best explanation
– Explanatory research
Deduction
• Reasoning from true premises
1. All students have laptops
2. Mary is a student
3. Therefore: Mary has a laptop
• If 1 and 2 are true, 3 must follow
• 2 and 3 can be true even though 1 is not
Deduction: Theory Testing
• Confirmatory theory testing
1. If theory is true (X leads to Y)
2. X relates to Y
• If 1 is true, 2 follows
• Finding 2 is support (confirms) 1
• 2 can be true even if 1 is false
Deduction: Model Testing
1.
X1
M
Y
X2
2. Data will fit predetermined pattern
• If 1 is true, 2 follows
• 2 can be true even if 1 is false
– Model equivalence problem
Deductive Inference
• Conclusion rests on truth of premises
• If premises are true, the conclusion is true
• What if premises are false?
– Mary has a laptop even though not all students do
– Mary lied about being a student
• Confirmation suggestive not conclusive
Deduction, Confirmation, Disconfirmation
• Deductive test of theory
– Collect data to confirm
– Uncertainty that additional cases disconfirm
• Theory: All swans are white
– Confirmation requires checking every swan
– Disconfirmation requires finding only one black
swan
Inductive Inference
• Generalizing from observed cases
– All students in my class have laptops
– Therefore all students have laptops
• Assumption that future cases same as
observed
• Assumes uniformity of nature
• Future cases not necessarily the same
– Boundary conditions
– Population differences
– Type 1 errors
Inductive Study
• Collect data on a phenomenon
• Generalize conclusions
• Observe pattern of sample relationships
• Conclude pattern exists in population
Features of Inductive Research
•
•
•
•
•
Exploratory
Not explanatory
No hypothesis/theory to test
Has purpose/question to address
Need to define scope of generalization
– What is the population?
– Boundary conditions
• Discovery rather than confirmation
• Allows serendipity
Abduction
• Inference to the best explanation
1. We observe X
2. We reason that the best explanation is A
3. We conclude that A is therefore true
•
•
•
•
Beyond inductive generalization
Attributions
Theory based on observation
What is criterion for “best”?
Criterion for Best Explanation
• Testable
• Parsimony
• Explains most variance
– Most accurate prediction
• Explains most phenomena
– Relates to the most variables
• Makes most sense: Subjective
Integrated Approach
• Deduction
– Theory test
• Induction
– Exploratory discovery
• Abduction
– Theoretical explanation for observations
• Ideal order
– Induction  Abduction  Deduction
History of Inference in Organizational and
Related Sciences
Exploratory Phase
•
•
•
•
•
Prior to 1980s
Behaviorism
Emulate physical science
Focus on observables
Limited theory
Transition Phase
•
•
•
•
•
1980s-1990s
Rise of cognitive science
Rejection of behaviorism
Focus on unobservables
Emphasis on models and theory
Confirmatory Phase
•
•
•
•
2000s
Rejection of exploratory methods
Hypothesis-driven research
Requirement of theory
Our Current State?
Deduction  Deduction  Deduction
Quick Journal Analysis
• Journal of Applied Psychology
• 1971 Issue 1 vs. 2015 Issue 1 & 4
– 15 articles from each
• Number of paragraphs
–
–
–
–
Introduction
Method
Results
Discussion
• Have hypotheses
Year
Intro Method Results Discussion
Hypoth.
1971
4.7
10
6.7
6.8
27%1
2015
24
18.5
10.6
14.9
93%2,3,4
1Three-fourths
specific tests of theory: Equity and Herzberg.
2One paper tested theory without specifying hypotheses as such.
3Some use inductive language (explore)
4Hypotheses not tests of specific theories.
Folly of Deductive Exclusiveness
• “Deduction orders and rearranges our
knowledge without adding to its content.”
– More confidence in what we thought we knew.
• “Induction can amplify and generalize our
experiences, broaden and deepen our
empirical knowledge.”
– New knowledge
Vickers, 2014
Pure Forms
• Deduction
– Hypotheses derived directly from theory
– Hypothesis tests confirm/disconfirm theory
• Induction
– Generalizations made from observations
• Abduction
– Explanation/theory made from inductive
generalization
Current Pseudo-Deduction
•
•
•
•
Data collected
Hypotheses chosen to match data
Theories argued to support hypothesis
Hypotheses not tightly linked to theory
– “Theory informed the hypothesis”
– “Theory supports the hypothesis”
• More abductive than deductive
Science Based on Disclosure
• Scientific report should describe what was
done
– What was assessed and when
– When hypotheses were derived relative to data
collection and analysis
– Inductive, deductive, or abductive?
Where Now?
Inductive/Abductive Forces
Prominent Scholars
• Donald Hambrick, 2007
– “Too much of a good thing”
– Why should facts await theories?
• Edwin Locke, 2007
– “The case for inductive theory building”
– Critique of deductive approach
– Examples of successful induction-based theories
Evolving Journal Practices
• Special journal issues
–
Journal of Business and Psychology 2014
•
–
Rogelberg, Ryan, Schmitt, Spector, Zedeck
Human Resource Management Review
•
–
Woo, O’Boyle, Spector
Journal of Organizational Behavior
point/counterpoint: Theory
• Journals allowing inductive papers
–
Academy of Management Discoveries
•
–
Andrew Van de Ven
Journal of Business and Psychology
Research Integrity
•
•
•
•
Confirmation bias concerns
HARKing
P mining (Simmons et al. 2011)
Replication crisis
Data Science
– Data mining
•
•
•
Finding meaningful/useful data patterns
Exploratory
Inductive
– Business intelligence
– HR analytics
– “Big data”
Challenges To Pendulum Moving
• Lack of knowledge
– People only know current practice
• Authors
• Editors
• Reviewers
• Belief in deductive approach
• Resistance to change
Deductive Study
•
•
•
•
•
•
•
Need premises to confirm
State theory
Derive hypotheses directly from theory
Design study to test hypotheses
Collect data
Test hypotheses
Draw conclusion
– Confirm/disconfirm theory
Inductive Study
•
•
•
•
•
•
Exploratory
State purpose/questions
Design study to address purpose
Collect data
Analyze data
Interpret/generalize results
Abduction
• Finding best explanation for inductive
observations
• Explain results with existing theory
• Explain results with new theory
Doing and Reporting Inductive Research
Inductive Introduction
•
•
•
•
Clearly state purpose/question
Review relevant background literature
Note gaps in knowledge study will fill
Is extensive literature review needed?
– Not in health and natural sciences
• Introductions state purpose and define terms
• Limited background literature
Heart failure is an extremely prevalent syndrome in developed and developing countries (Go et al., 2014). Outcomes associated with heart failure are poor, with patients
experience increasing symptoms as the syndrome progresses (Lam and Smeltzer, 2012). Symptoms are responsible for a decline in functional status, deteriorating healthrelated quality of life (HRQL), repeated hospitalization, and early demise (Falk et al., 2013, Murthy and Lipman, 2011 and Song et al., 2010). A growing body of
research demonstrates that self-care by patients can improve these outcomes (Ditewig et al., 2010, Jones et al., 2012 and Tung et al., 2013). Self-care involves a process
of maintaining physiological stability by monitoring symptoms, adhering to the treatment regimen (self-care maintenance), and promptly identifying and responding to
symptoms (self-care management) (Riegel and Dickson, 2008). Clinicians are challenged by the difficulties associated with engaging patients in self-care (Gardetto,
2011).
Self-care requires that patients have knowledge and skills, both of which are influenced by cognition (Dickson et al., 2008). Cognitive impairment is recognized as an
issue in 25–50% of adults with heart failure (Dodson et al., 2013, Gure et al., 2012 and Pressler, 2008). Most studies investigating the relationship between cognitive
impairment and self-care have demonstrated that heart failure patients with cognitive impairment are poor at self-care. For example, Harkness et al. (2013) found that
patients with mild cognitive impairment scored significantly lower in self-care management than those without cognitive impairment. Alosco et al. (2012) found that
patients with lower (worse) scores on the Mini Mental Status Examination were more likely to fail in medication management, a dimension of self-care. Smeulders et al.,
2010a and Smeulders et al., 2010b found that heart failure patients with better cognitive status benefitted more from a chronic disease self-management program
compared to patients with poorer cognitive function. Only one investigative team found no relationship between cognition and self-care (Cameron et al., 2009), although
in a subsequent study the same investigators (Cameron et al., 2010) found that heart failure patients with mild cognitive impairment exhibited lower self-care
management and self-care confidence than patients without cognitive impairment. Hajduk et al. (2013) found no association between overall cognitive status and selfcare in heart failure patients, but when they analyzed specific domains, impaired memory was associated with poor self-care while executive function and processing
speed were not. This inconsistency suggests that cognition is not a direct predictor of self-care behaviors in patients with heart failure.
Mechanisms or pathways through which cognitive impairment affects self-care are worth clarifying because they may suggest targets for intervention. One potential
mechanism is self-efficacy or task-specific confidence, which has been defined as confidence in the ability to perform the various self-care behaviors (e.g. confidence in
one's ability to follow a low salt diet) (Riegel and Dickson, 2008).
The situation specific theory of heart failure self-care (Riegel and Dickson, 2008 and Riegel et al., 2015) specifies that self-care maintenance (monitoring of heart failure
symptoms and adhering to treatments) influences self-care management (the response of patients to signs and symptoms of a heart failure exacerbation) and that both are
influenced by self-care confidence. Therefore, the overall purpose of this study was to test self-care confidence as mediating the relationship between predictors of selfcare and the self-care behaviors of maintenance and management (Riegel and Dickson, 2008).
To date, few researchers have investigated the role of confidence in influencing self-care behaviors. Cene and colleagues (2013) found that perceived emotional and
informational support was associated with better self-care maintenance and possibly better self-care management in a sample of heart failure patients. A similar result
was found by Sayler et al. (2012) who showed that self-care confidence mediated the relationship between social support and self-care maintenance and between social
support and self-care management. In a mixed method study Dickson et al. (2008) found that patients with lower self-care confidence and impaired cognition had lapses
in self-care behaviors and were classified as “inconsistent” while those with higher self-care confidence and better cognition were better in self-care and were classified
as “experts”. Building on prior studies illustrating that cognitive impairment and self-care confidence are both predictors of self-care behaviors and building on the
situation specific theory of heart failure self-care that proposes confidence as a mediator of the relationship between self-care behaviors and its predictors, the specific
objective of this study was to evaluate whether self-care confidence mediates the relationship between cognition and the self-care behaviors. Specifically we
hypothesized that cognition affects self-care maintenance and management only indirectly by its influence on self-care confidence (Fig. 1).
Inductive designs
•
•
•
•
Basic designs much the same
Collect more variables
Bigger n to allow cross-validation
Use methods that capture large amounts of
data
Inductive Analysis
• Broader range of analyses in a study
– More descriptive statistics
– More inferential statistics
– Fully explore data
– Utilize graphical methods
– Numerical analysis
– Text analysis
Data Mining
• Family of techniques to find patterns in data
• Knowledge discovery
• Finding novel patterns
– Forecasting
– Understanding
• Computer science + Statistics
• New and old techniques
Four Data Mining Problems
• Data Matrix
– Records by attributes
– People by variables
• Analysis of columns (attributes)
1. Association
2. Classification
• Analysis of rows (records)
3. Clustering
4. Outlier analysis
Association
– Relationships among attributes
• Correlations among continuous variables
• Exploratory factor analysis: Underlying
structure
– Association rules mining (ARM)
•
•
•
•
Items that go together
If A, probability of B
Support: Prevalence of A and of B
Confidence: Co-occurrence of A and B
Classification
• Supervised: One column target attribute
– Predict target attribute from other attributes
– Criterion and predictors
• Data used to “train”
– Develop predictor model
– Regression
• Forward, Backward, Stepwise
– ANOVA/MANOVA
– Multilevel modeling
Clustering
•
•
•
•
Analysis of Records (Rows)
Unsupervised
Records sharing attributes
Cluster analysis
–
–
–
–
Based on attribute similarity
Each case in one cluster
Case more similar to own cluster
Predict cluster membership
Text Clustering
• Based on word similarity
• Vector space model
–
–
–
–
–
–
–
Word by document matrix
List of all words in all documents
Delete too common and too rare
Each document has vector of words
Vector similarity determines cluster membership
Compile most common within-cluster words
Interpret: Based on 10 most common words
Gupta, 2014
Outlier Analysis
• Records that are different from others
• Generated by different mechanism
• Used to detect deviance
– Crime
– Fraud
• Statistical criteria
– Univariate: 3 SD from Mean
– Multivariate Distance
• Euclidian distance
• Standardized distance: Mahalanobis D
Limitations To Inductive Approach
Type 1 Error Problem
• Likelihood increases with more tests
• Solution
– Cross-validation/replication
– Supporting evidence in literature
• Others found same results
Interpretation Difficulties
• Patterns might not be coherent
– No framework to interpret results
• Results disconnected list of facts
• Solution
– Abduction: Finding best explanation
• Existing theory
• Modifying theory
• New theory
What Makes a Good Inductive Paper?
• Sound research question
– Important
– New
– Unanswered
•
•
•
•
Rigorous methodology
Multiple methods
Cross-validated
Linked to existing literature
Contribution To a Scientific Discipline
• Testing old theory
– Too many theories and not enough tests
– Modal number of theory tests in AMR = 0
Kacmar & Whitfield 2000
• New findings: Exploration and Discovery
• Andrew Van de Ven’s anomalies
• Presenting new theory
• Solution to practical problems
• Academic-practice divide
Concluding Thoughts
• Authors should submit inductive papers.
– Editors are more open than people assume.
• Reviewers should evaluate papers on
contribution beyond theory.
• Data Mining literature has useful tools.
• Exploratory research foundation of scientific
progress.
References
Aggarwal, C. C. (2015). Data mining : The textbook. New York City: Springer.
Douven, I. (2014). Abduction. The Stanford encyclopedia of philosophy (Spring 2014
edition). August 13, 2015. Retrieved from
http://plato.stanford.edu/archives/spr2014/entries/abduction/
Gupta, G. K. (2014). Introduction to data mining with case studies 3rd ed. Delhi, PHI Learning.
Hambrick, D. C. (2007). The field of management's devotion to theory: Too much of a
good thing? Academy of Management Journal, 50(6), 1346-1352.
doi:10.5465/AMJ.2007.28166119
Kacmar, K. M., & Whitfield, J. M. (2000). An additional rating method for journal articles in
the field of management. Organizational Research Methods, 3(4), 392-406.
doi:10.1177/109442810034005
Locke, E. A. (2007). The case for inductive theory building. Journal of Management,
33(6), 867-890. doi:http://dx.doi.org/10.1177/0149206307307636
Okasha, S. (2002). Philosophy of science: A very short introduction. Oxford, UK: Oxford
University Press.
Payne, D., & Trumbach, C. C. (2009). Data mining: Proprietary rights, people and
proposals. Business Ethics: A European Review, 18(3), 241-252. doi:10.1111/j.14678608.2009.01560.x
Rosenberg, A. (2012). Philosophy of science: A contemporary introduction (3rd ed.). New
York City: Routledge.
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology:
Undisclosed flexibility in data collection and analysis allows presenting anything as
significant. Psychological Science, 22, 1359-1366.
Vickers, J. (2014). The problem of induction. The Stanford encyclopedia of philosophy
(Spring 2014 edition). August 13, 2015. Retrieved from
http://plato.stanford.edu/archives/spr2014/entries/induction-problem/