Does District Magnitude Matter? The Case of Taiwan

Transcript

1. The Question
   Background
   Data and Methods
   Results
   Does District Magnitude Matter?
   The Case of Taiwan
   Carlisle Rainey
   Department of Political Science
   Department of Statistics
   Florida State University
   January 13, 2012
   Rainey District Magnitude
   

   View Slide

2. The Question
   Background
   Data and Methods
   Results
   Table of Contents
   1 The Question
   2 Background
   Theory
   Critique
   3 Data and Methods
   Data
   Posterior Probabilities
   Bayesian Model Averaging
   4 Results
   Posterior Densities
   Summary
   Rainey District Magnitude
   

   View Slide

3. The Question
   Background
   Data and Methods
   Results
   The Key Question
   Does proportional representation cause higher turnout?
   Rainey District Magnitude
   

   View Slide

4. The Question
   Background
   Data and Methods
   Results
   The Key Question
   Does proportional representation cause higher turnout?
   • Some authors say “Yes” (Powell 1982, Franklin 2002, Selb
   2009).
   Rainey District Magnitude
   

   View Slide

5. The Question
   Background
   Data and Methods
   Results
   The Key Question
   Does proportional representation cause higher turnout?
   • Some authors say “Yes” (Powell 1982, Franklin 2002, Selb
   2009).
   • Some authors say “Maybe” (Blais and Dobrzynska 1998, Blais
   and Aarts 2005, Fornos et al. 2004).
   Rainey District Magnitude
   

   View Slide

6. The Question
   Background
   Data and Methods
   Results
   Theory
   Critique
   Why Might Turnout be Higher Under PR?
   • Disproportional systems leave some voters without
   representation (Powell 2000, Karp and Banducci 2008).
   Rainey District Magnitude
   

   View Slide

7. The Question
   Background
   Data and Methods
   Results
   Theory
   Critique
   Why Might Turnout be Higher Under PR?
   • Disproportional systems leave some voters without
   representation (Powell 2000, Karp and Banducci 2008).
   • Proportional rules gives parties incentive to disperse across the
   ideological space, making narrow appeals to voters (Cox
   1990).
   Rainey District Magnitude
   

   View Slide

8. The Question
   Background
   Data and Methods
   Results
   Theory
   Critique
   Why Might Turnout be Higher Under PR?
   • Disproportional systems leave some voters without
   representation (Powell 2000, Karp and Banducci 2008).
   • Proportional rules gives parties incentive to disperse across the
   ideological space, making narrow appeals to voters (Cox
   1990).
   • Proportional rules create “nationally competitive districts” in
   which parties have an incentive to mobilize everywhere
   (Powell 1982, Cox 1999, Selb 2009).
   Rainey District Magnitude
   

   View Slide

9. The Question
   Background
   Data and Methods
   Results
   Theory
   Critique
   Why Might Turnout be Higher Under PR?
   • Disproportional systems leave some voters without
   representation (Powell 2000, Karp and Banducci 2008).
   • Proportional rules gives parties incentive to disperse across the
   ideological space, making narrow appeals to voters (Cox
   1990).
   • Proportional rules create “nationally competitive districts” in
   which parties have an incentive to mobilize everywhere
   (Powell 1982, Cox 1999, Selb 2009).
   In summary, under PR, citizens should be more likely to...
   1 feel represented (H2),
   2 feel close to a party (H3),
   3 be contacted (H4),
   ...and then vote (H1).
   Rainey District Magnitude
   

   View Slide

10. The Question
    Background
    Data and Methods
    Results
    Theory
    Critique
    Some Problems
    Theoretical
    • PR reduces the decisiveness of elections (Jackman 1987)
    • Disproportionality might actually increase the incentive to
    mobilize (Schram and Sonnemans 1996, Rainey, under
    review).
    Rainey District Magnitude
    

    View Slide

11. The Question
    Background
    Data and Methods
    Results
    Theory
    Critique
    Some Problems
    Theoretical
    • PR reduces the decisiveness of elections (Jackman 1987)
    • Disproportionality might actually increase the incentive to
    mobilize (Schram and Sonnemans 1996, Rainey, under
    review).
    Methodological
    1 Most research has focused on testing whether PR increases
    turnout, not adjudicating among explanations.
    2 The methods used in this literature cannot assess the evidence
    in favor of a null effect.
    3 PR systems are different in many ways from majoritarian
    systems, making it difficult to estimate an unbiased effect.
    Rainey District Magnitude
    

    View Slide

12. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Data
    Taiwan offers an especially useful case for testing theoretical
    arguments
    • substantial variation in district magnitude
    • national political context held constant
    Rainey District Magnitude
    

    View Slide

13. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Data
    Taiwan offers an especially useful case for testing theoretical
    arguments
    • substantial variation in district magnitude
    • national political context held constant
    Survey (CSES) conducted during the 2001 legislative election
    • the magnitude of each respondent’s district
    • self-reported turnout
    • questions tapping whether respondents feel represented, feel
    close to a party, and were contacted by a party
    Rainey District Magnitude
    

    View Slide

14. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Assessing Evidence for the Null: Posterior Probabilities
    Given a set of models M = {M1, ..., Mn
    }, one of which generated
    the data D, it is possible to assign probabilities to each model.
    p(Mi
    |D) =
    p(D|Mi
    )p(Mi
    )
    n
    k=1
    p(D|Mk
    )p(Mk
    )
    These quantities allow us to...
    1 assess the relative quality of the model
    2 assign probabilities to hypotheses (as opposed to p-values)
    • P(βmagnitude
    = 0|D)
    3 average estimates across models
    Rainey District Magnitude
    

    View Slide

15. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Model Uncertainty
    The Problem of Model Specification
    • I have little prior belief about what the correct model
    specification is.
    Rainey District Magnitude
    

    View Slide

16. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Model Uncertainty
    The Problem of Model Specification
    • I have little prior belief about what the correct model
    specification is.
    • Fifteen possible control variables, all occurring causally prior
    to effect of district magnitude.
    Rainey District Magnitude
    

    View Slide

17. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Model Uncertainty
    The Problem of Model Specification
    • I have little prior belief about what the correct model
    specification is.
    • Fifteen possible control variables, all occurring causally prior
    to effect of district magnitude.
    • Including too many variables can make estimates less precise,
    including too few can lead to omitted variable bias.
    Rainey District Magnitude
    

    View Slide

18. The Question
    Background
    Data and Methods
    Results
    Data
    Posterior Probabilities
    Bayesian Model Averaging
    Model Uncertainty
    The Problem of Model Specification
    • I have little prior belief about what the correct model
    specification is.
    • Fifteen possible control variables, all occurring causally prior
    to effect of district magnitude.
    • Including too many variables can make estimates less precise,
    including too few can lead to omitted variable bias.
    • Inferences are somewhat sensitive to model specification.
    (215 = 32, 768 possible combinations of control variables.)
    I explicitly average across inferences, weighting by the posterior
    probability of each model.
    Rainey District Magnitude
    

    View Slide

19. The Question
    Background
    Data and Methods
    Results
    Posterior Densities
    Summary
    A Look at One Model in Detail:
    Logit Coefficient Estimates for Model of Turnout
    −0.10 0.05 0.20
    0.0 0.6
    Age
    −0.3 0.0 0.3
    0.0 0.6
    Age.Squared
    −0.8 −0.2 0.4
    0.0 0.6
    Male
    −0.1 0.1 0.3
    0.0 0.6
    Education
    0.0 1.0
    0.0 0.6
    Married
    −0.5 0.5
    0.0 0.6
    Union.Member
    −0.2 0.0 0.2
    0.0 0.6
    Household.Income
    −1.0 0.0
    0.0 0.6
    Rural
    −1.0 0.0 1.0
    0.0 0.6
    Small.Town
    −0.4 0.2 0.8
    0.0 0.6
    Suburb
    −0.5 0.5
    0.0 0.6
    White.Collar
    −1.0 0.0
    0.0 0.6
    Worker
    −1.5 −0.5 0.5
    0.0 0.6
    Hakka
    −0.5 0.5
    0.0 0.6
    MinNan
    −1.0 0.5 1.5
    0.0 0.6
    Mainlander
    −0.4 0.0
    0.0 0.6
    log.Magnitude
    Rainey District Magnitude
    

    View Slide

20. The Question
    Background
    Data and Methods
    Results
    Posterior Densities
    Summary
    Estimate of the Coefficient for District Magnitude
    Rainey District Magnitude
    

    View Slide

21. The Question
    Background
    Data and Methods
    Results
    Posterior Densities
    Summary
    Summary of Results
    Results for the other outcomes are similar to turnout. (Number in
    parentheses is percent chance that βmagnitude
    = 0, conditioning on
    the data.)
    1 feel represented (H2, .02),
    2 feel close to a party (H3, .03),
    3 be contacted (H4, .03),
    ...and then vote (H1, .03).
    Rainey District Magnitude
    

    View Slide

22. The Question
    Background
    Data and Methods
    Results
    Conclusion
    • The literature on electoral rules has asserted that PR
    increases turnout, but this argument has theoretical
    contradictions and empirical difficulties.
    • Using posterior probabilities and the useful case of Taiwan, I
    find strong evidence that proportional electoral rules do not
    increase turnout.
    • I find strong evidence for the null for each of the three
    mechanisms I test.
    Rainey District Magnitude
    

    View Slide