Presentation at special event "To Explain or To Predict?" at Tel Aviv University, July 9, 2012. Event co-organized by the Israel Statistical Association and Tel Aviv University's Department of
Presentation at special event "To Explain or To Predict?" at Tel Aviv University, July 9, 2012. Event co-organized by the Israel Statistical Association and Tel Aviv University's Department of Statistics and OR.
Galit Shmuéli Ij Israel Statistical Association & Tel Aviv University July 9, 2012 To Explain or To Predict?
Points for discussion: goo.gl/gcjlNTwitter: #explainpredict
Road MapDefinitionsExplanatory-dominated social sciencesExplanatory ≠ predictive modeling Why? Different modeling paths Explanatory vs. predictive powerSo what?
DefinitionsExplanatory modeling:Theory-based, statistical testing ofcausal hypothesesExplanatory power:Strength of relationship in statisticalmodel
DefinitionsPredictive modeling:Empirical method for predicting newobservationsPredictive power:Ability to accurately predict newobservations
Statistical modeling in social science researchPurpose: test causal theory (“explain”) Association-based statistical models Prediction nearly absent
Explanatory modeling à-la social sciencesStart with a causaltheoryGenerate causalhypotheses onconstructsOperationalize constructs → Measurable variablesFit statistical modelStatistical inference → Causal conclusions
In the social sciences,data analysis is mainly used for testing causal theory. “If it explains, it predicts”
“Empirical prediction alone is un-scientific”Some statisticians share this view: The two goals in analyzing data... I prefer to describe as “management” and “science”. Management seeks profit... Science seeks truth. - Parzen, Statistical Science 2001
52 “predictive” articles among 1,072in Information Systems top journals
Why Predict? for Scientific Research new theory develop measures compare theories improve theory assess relevance predictabilityShmueli & Koppius, “Predictive Analytics in IS Research”(MISQ, 2011)
“A good explanatory model will alsopredict well”“You must understand the underlyingcauses in order to predict”
Philosophy of Science“Explanation and prediction have thesame logical structure” Hempel & Oppenheim, 1948 “It becomes pertinent to investigate the possibilities of predictive procedures autonomous of those used for explanation” Helmer & Rescher, 1959 “Theories of social and human behavior address themselves to two distinct goals of science: (1) prediction and (2) understanding” Dubin, Theory Building, 1969
Four aspects Y=F(X) E(Y)=f(X)1. Theory – Data2. Causation – Association3. Retrospective – Prospective4. Bias - Variance
“The goal of finding models that arepredictively accurate differs from thegoal of finding models that are true.”
Point #1Best explanatory model ≠ Best predictive model
Four aspects Y=F(X) Y=f(X)1. Theory - Data2. Causation – Association3. Retrospective – Prospective4. Bias - Variance
Predict ≠ Explain “we tried to benefit from an extensive set of attributes describing each of the movies in the dataset. Those attributes certainly carry a significant signal and + can explain some of the user behavior. However… they could not help at all ? for improving the [predictive] accuracy.” Bell et al., 2008
Predict ≠ ExplainThe FDA considers two productsbioequivalent if the 90% CI of therelative mean of the generic to brandformulation is within 80%-125%“We are planning to… develop predictive models for bioavailabilityand bioequivalence” Lester M. Crawford, 2005 Acting Commissioner of Food & Drugs
Goal Design & Data EDADefinition Collection PreparationVariables? Model Use &Methods? Evaluation, V Reporting alidation & Model Selection
Study design & data collectionObservational or experiment?Primary or secondary data?Instrument (reliability+validity vs. measur accuracy)How much data?How to sample? Hierarchical data
Data Preprocessing missing reduced- feature models partitioning
Data exploration & reduction Interactive visualization PCA SVD
Which Variables? endogeneity ex-post availabilitycausation associations Multicollinearity? A, B, A*B?
Methods / Models Blackbox / interpretable Mapping to theory variance biasShrinkage models ensembles
Model fit ≠ Validation Explanatory powerTheoretical Empirical Data model model Evaluation, Validation & Model SelectionEmpirical Training data Over-fitting model Holdout data analysis Predictive power
Model Use test causal theory Inference Null hypothesisnew theoryDevelop measurescompare theories Predictive performanceimprove theory Naïve/baselineassess relevance Over-fitting analysispredictability
Point #2Explanatory Predictive Power ≠ PowerCannot infer one from the other
out-of-sample interpretationp-values prediction accuracy PerformanceR2 costs Metrics Training vs.goodness-of-fit holdout type I,II errors over-fitting
Predictive Power Explanatory Power
The predictive power of anexplanatory model has importantscientific valueRelevance, reality check, predictability
In “explanatory” fieldsPrediction underappreciatedDistinction blurredUnfamiliar with predictivemodeling/assessment “While the value of scientific prediction… is beyond question… the inexact sciences *do not+ have…the use of predictive expertise well in hand.” Helmer & Rescher, 1959
How does all this impact Scientific Research?
What can be done? acknowledgeincorporate prediction into curriculum
What happens in other fields? Epidemiology Engineering Life sciencesWhat about “predictive only”fields? http://goo.gl/gcjlN
Shmueli (2010), “To Explain or To Predict?”, Statistical ScienceShmueli & Koppius (2011), “Predictive analytics in IS research”, MISQ