small n analysis in policy research && innovation
TRANSCRIPT
Beyond the quantitative-qualitative divide
OPENINNOVA::David Lopez
Large-N && Small-N Analysis in Innovation
studies
Creative Commons Attribution ShareAlike 3.0 (http://creativecommons.org/licenses/by-nc-sa/3.0/ )
OpenInnova:David López
Quantitative versus Qualitative Analysis in Social Sciences
Two main approaches to social science analysis:
Qualitative
analysis
Quantitative
analysis
OpenInnova:David López
Large N approach. Provides answers to: what causes revolutions ?
Quantitative analysis:◦Large-N approach (extensive use of cross-
sectional data)
Theory
Empirical Data
Multivariate
Analysis
Model
OpenInnova:David López
Small-N approach. Provides answers to: what caused the French revolution ?
Qualitative analysis:◦Small-N approach (qualitative comparisons of
cases)
Cases
Qualitative analysis
Causal chains
“Soft Model”
OpenInnova:David López
Small-N analysis in policy research (innovation is about policy after all)
When it comes to policy analysis, such as innovation regimes and R&D, statistical inference is not enough, holistic approaches are needed in order to:◦ Explore several combinations and their consequences.◦ Conduct context-specific assessments.
For instance: What It takes to avoid poverty ?◦ Does college education make a difference for married white
males from families with good incomes ?◦ And college education for unmarried black females from low-
income families ?
Moreover, what about scenarios with limited data such as OECD innovation database for instance ?
OpenInnova:David LópezSmall-N Analysis by example: Avoiding poverty (I)
Small-N approaches consider cases as combinations of causally relevant conditions
College Educated
High Parental income
Parent College
educated
High AFQT Score
Poverty avoidanc
e
Number of cases
1 0 0 0 0 0 30
2 0 0 0 1 0 3
3 0 0 1 0 ? 4
….
16 1 1 1 1 1 23
OutcomeCausal conditions
Goal of the analysis: Identify different combinations of case characteristics explicitly linked to poverty avoidance.
OpenInnova:David LópezSmall-N Analysis : Fuzzy-set approach
But…. what do we mean by “High parental income” ???
How strong is the inference: CollegeAvoiding poverty
Charles Ragin suggests Fuzzy-set logic.
High parental income
Low parental income
High parental income
Low parental income
OpenInnova:David LópezSmall-N Analysis: Sufficiency && Consistency
Xi
Y
)()),(min()( iiiii XYXYXyConsistenc
Xi→Y
OpenInnova:David LópezSmall-N Analysis: Necessity && Consistency
Y
Xi
)()),(min()( iiiii YYXXYyConsistenc
Y →Xi
OpenInnova:David LópezSmall-N Analysis: Sufficiency && Coverage
Xi1
Y
)()),(min()( iiiii YYXYXCoverage
X1* X2 →Y
Xi
Y
Xi→Y
Xi2
Xi
Y
Xi→Y
OpenInnova:David LópezfzQCA in Action: Democracies in interwar Europe (1918-1936)
Data from 18 European countries (Ragin 2008).
COUNTRY SURVIVED BREAKDOWN DEVELOPED URBAN LITERATE INDUSTRIAL STABLEAustria 0,01 0,99 0,74 0,14 0,98 0,76 0,35Belgium 0,98 0,02 0,99 0,89 0,96 0,98 0,96Czech 0,85 0,015 0,42 0,96 0,97 0,91 0,87Estonia 0,12 0,88 0,15 0,07 0,96 0,02 0,87Finland 0,64 0,36 0,43 0,03 0,98 0,09 0,51France 0,98 0,02 0,97 0,02 0,97 0,83 0,93Germany 0,01 0,99 0,85 0,83 0,98 0,96 0,23Greece 0,03 0,97 0,05 0,1 0,11 0,38 0,35Hungary 0,41 0,59 0,08 0,2 0,81 0,08 0,09Ireland 0,91 0,09 0,62 0,04 0,96 0,02 0,93Italy 0,01 0,99 0,25 0,11 0,38 0,49 0,51
Netherland 0,98 0,02 0,97 0,99 0,99 0,94 0,99Poland 0,12 0,88 0,03 0,22 0,55 0,02 0,02Portugal 0,01 0,99 0,02 0,01 0,02 0,12 0,02Romania 0,25 0,75 0,02 0,03 0,15 0,02 0,78Spain 0,03 0,97 0,04 0,41 0,08 0,22 0,14Sweden 0,98 0,02 0,93 0,15 0,99 0,7 0,87UK 0,98 0,02 0,98 0,98 0,99 0,98 0,96
Outcome
OpenInnova:David LópezfzQCA in Action: Democracies in interwar Europe (1918-1936)
0.5
DEVELOPED URBAN LITERATE INDUSTRIAL STABLE number survivedconsist1 1 1 1 1 3 0.8843371 0 1 0 1 1 0.7738101 1 0 0 0 0 0.7413791 1 0 0 1 0 0.7368421 1 0 1 0 0 0.7272731 1 0 1 1 0 0.7272731 0 1 1 1 2 0.7253521 1 1 0 1 0 0.7128711 1 1 0 0 0 0.6947370 1 1 1 1 1 0.6754971 0 1 0 0 0 0.6745560 1 1 0 0 0 0.6279070 1 1 0 1 0 0.6200001 0 0 0 0 0 0.5945951 0 0 0 1 0 0.5890410 1 0 0 1 0 0.5890411 0 0 1 1 0 0.5774651 0 0 1 0 0 0.5774650 1 0 1 1 0 0.5522390 1 1 1 0 0 0.5288460 0 1 0 1 2 0.5081970 0 1 0 0 2 0.5061730 1 0 1 0 0 0.4933330 1 0 0 0 0 0.4846151 1 1 1 0 1 0.3928571 0 1 1 0 1 0.3793100 0 1 1 1 0 0.3706290 0 1 1 0 0 0.3706290 0 0 0 1 2 0.3069770 0 0 1 1 0 0.2878790 0 0 1 0 0 0.2483660 0 0 0 0 3 0.225543
OpenInnova:David LópezfzQCA in Action: Democracies in interwar Europe (1918-1936)
Consistent enough
Non consistent
0.75
DEVELOPED URBAN LITERATE INDUSTRIAL STABLE number survivedconsist
1 1 1 1 1 3 1 0.884337
0 0 0 0 0 3 0 0.2255431 0 1 0 1 1 1 0.7753520 0 1 0 1 2 0 0.5081970 0 1 0 0 2 0 0.5061730 0 0 0 1 2 0 0.3069771 0 1 0 1 1 0 0.7738100 1 1 1 1 1 0 0.6754971 1 1 1 0 1 0 0.3928571 0 1 1 0 1 0 0.379310
OpenInnova:David López
Least parsimonious solution (remainders not considered)
If all five conditions are present then is sufficient for a democracy to survive.
But NOTurban and NOTindustrial does not really seem to matter much:
DEVELOPED*LITERATE*STABLE SURVIVALfzQCA in Action: Democracies in interwar Europe (1918-1936)
OpenInnova:David López
Considering Breakdown instead of Survival as the outcome of interest:
Two sufficient conditions emerge:
NOT DEVELOPED * NOT URBAN * NOT INDUSTRIAL BREAKDOWN
DEVELOPED*LITERATE*INDUSTRIAL*NOT STABLE BREAKDOWN
fzQCA in Action: Democracies in interwar Europe (1918-1936)
OpenInnova:David López
Nested analysis approach (Evan S. Lieberman) The best of both worlds ? Just adding extra workload ?
OpenInnova:David López
Nested analysis approach (Evan S. Lieberman) The best of both worlds ? Just adding extra workload ?
Triangulation