Nash equilibrium approach to ML estimation with...
Transcript of Nash equilibrium approach to ML estimation with...
![Page 1: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/1.jpg)
Nash equilibrium approach to ML estimation withapplication to voting models
Alexei Zakharov
State University - Higher School of Economics, DECAN laboratory, Moscow,Russia
Central Institute for Mathematics and EconomicsPrepared for 2009 Australasian Public Choice conference
December 22, 2009
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 2: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/2.jpg)
Brief description of my work
I propose a new methodology for ML estimation of multinomialchoice models. It will be used to re-estimate the probabilisticvoting model for several countries. In particular, one can analyzesurvey data to tell
How much effect do policy programs of political parties haveon voters?
What do political parties maximize?
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 3: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/3.jpg)
Brief description of my work
Consider estimating an econometric model of discrete choice fromsurvey data, such as voting behavior: What party to vote for,depending on the policy positions of the parties, and votercharacteristics such as gender, income, etc.?
Existing methodology What values of the model parameters bestexplain the observed survey response?
Proposed methodology What values of the model parameters bestexplain the observed survey response AND the policypositions of the parties, given our assumptions on
1 The objective functions of the parties.2 The information available to the parties.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 4: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/4.jpg)
Brief description of my work
Consider estimating an econometric model of discrete choice fromsurvey data, such as voting behavior: What party to vote for,depending on the policy positions of the parties, and votercharacteristics such as gender, income, etc.?
Existing methodology What values of the model parameters bestexplain the observed survey response?
Proposed methodology What values of the model parameters bestexplain the observed survey response AND the policypositions of the parties, given our assumptions on
1 The objective functions of the parties.2 The information available to the parties.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 5: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/5.jpg)
Brief description of my work
Consider estimating an econometric model of discrete choice fromsurvey data, such as voting behavior: What party to vote for,depending on the policy positions of the parties, and votercharacteristics such as gender, income, etc.?
Existing methodology What values of the model parameters bestexplain the observed survey response?
Proposed methodology What values of the model parameters bestexplain the observed survey response AND the policypositions of the parties, given our assumptions on
1 The objective functions of the parties.2 The information available to the parties.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 6: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/6.jpg)
Some related works
Clinton and Meirowitz (2003)
The quantal response equilirium: McKelvey and Palfrey(1995).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 7: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/7.jpg)
Formal models of political competiton and the empiricalpuzzle
1 Hotelling (1929), Downs (1957), and Black (1958) — medianvoter policy coincidence.
2 More than one dimension — equilibrium does not exist indeterministic models. McKelvey (1976, 1979), Plott (1968),McKelvey and Schofield (1987).
3 Policy platforms in real elections do, in fact, diverge.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 8: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/8.jpg)
Formal models of political competiton and the empiricalpuzzle
1 Hotelling (1929), Downs (1957), and Black (1958) — medianvoter policy coincidence.
2 More than one dimension — equilibrium does not exist indeterministic models. McKelvey (1976, 1979), Plott (1968),McKelvey and Schofield (1987).
3 Policy platforms in real elections do, in fact, diverge.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 9: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/9.jpg)
Formal models of political competiton and the empiricalpuzzle
1 Hotelling (1929), Downs (1957), and Black (1958) — medianvoter policy coincidence.
2 More than one dimension — equilibrium does not exist indeterministic models. McKelvey (1976, 1979), Plott (1968),McKelvey and Schofield (1987).
3 Policy platforms in real elections do, in fact, diverge.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 10: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/10.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 11: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/11.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 12: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/12.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 13: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/13.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 14: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/14.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 15: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/15.jpg)
Probabilistic voting models
Hinich, Ledyard, and Ordeshook (1972), Hinich (1977,1978)— first formal models. “Mean voter theorem”.
Lindbeck and Weibull (1987,1993), Coughlin (1992), Banksand Duggan (2005) — more on MVT.
Lin, Enelow, and Dorussen (1999), Schofield (2007) — severalplayers.
McKelvey and Patty (2006) — quantal response equilibrium.
Patty (2005, 2007) — voteshare maximizers vs. probability ofwin maximizers.
Zakharov (2009) — generalized VN-M utility.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 16: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/16.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 17: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/17.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 18: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/18.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 19: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/19.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 20: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/20.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 21: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/21.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 22: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/22.jpg)
Estimation of probabilistic voting models
Poole and Rosenthal (1984), Alvarez and Nagler (1995) —US Presidential elections.
Rabinowitz and MacDonald (1989), Merrill and Grofman(1999), Iversen (1994) — directional and proximity models.Ansolabahara, Rodden, Snyder (2008)— “issue voting”.
Alvarez and Nagler (1998) — Britain 1987, Netherlands 1994.Strategic voting.
Adams, Dow, and Merrill (2006), Adams and Merrill (2008),Thurner and Eymann (2000), Plane and Gershtenson (2004),Peress (2005) — abstention in spatial models.
Quinn, Martin, Whitford (1998) — Bayesian MNP estimation.
Zakharov and Fantazzini (2008), Thurner and Eymann(2000), Hellwig (2008) — variable salience.
Merrill and Adams (2001), Schofield (2007), Schofield andSened (2006), Schofield and Zakharov (2010) — equilibriumanalysis of formal models with estimated parameters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 23: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/23.jpg)
What are the results?
Estimation utilizes logit or probit discrete-choice models
All works show that positions on main policy issues to somedegree affect the choice of voters
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 24: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/24.jpg)
Models of discrete choice — an overview
Consider a problem of estimating a model of individual choice.
1 A dataset with i = 1, . . . ,N observations, each correspondingto an individual.
2 For each observation, a vector of personal characteristicsxi ∈ R
M1 .
3 and a choice variable di ∈ {1, . . . , J}.
4 The utility of individual i choosing an alterantive j is
uij = u(xi , αj , β, j) + ǫij = uij + ǫij , (1)
where αj ∈ RM2 is a vector of choice-specific parameters, andβ ∈ RM3 is a vector of choice-independent parameters.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 25: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/25.jpg)
Models of discrete choice — an overview
Consider a problem of estimating a model of individual choice.
1 A dataset with i = 1, . . . ,N observations, each correspondingto an individual.
2 For each observation, a vector of personal characteristicsxi ∈ RM1 .
3 and a choice variable di ∈ {1, . . . , J}.
4 The utility of individual i choosing an alterantive j is
uij = u(xi , αj , β, j) + ǫij = uij + ǫij , (1)
where αj ∈ RM2 is a vector of choice-specific parameters, andβ ∈ RM3 is a vector of choice-independent parameters.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 26: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/26.jpg)
Models of discrete choice — an overview
Consider a problem of estimating a model of individual choice.
1 A dataset with i = 1, . . . ,N observations, each correspondingto an individual.
2 For each observation, a vector of personal characteristicsxi ∈ RM1 .
3 and a choice variable di ∈ {1, . . . , J}.
4 The utility of individual i choosing an alterantive j is
uij = u(xi , αj , β, j) + ǫij = uij + ǫij , (1)
where αj ∈ RM2 is a vector of choice-specific parameters, andβ ∈ RM3 is a vector of choice-independent parameters.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 27: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/27.jpg)
Models of discrete choice — an overview
Consider a problem of estimating a model of individual choice.
1 A dataset with i = 1, . . . ,N observations, each correspondingto an individual.
2 For each observation, a vector of personal characteristicsxi ∈ RM1 .
3 and a choice variable di ∈ {1, . . . , J}.
4 The utility of individual i choosing an alterantive j is
uij = u(xi , αj , β, j) + ǫij = uij + ǫij , (1)
where αj ∈ RM2 is a vector of choice-specific parameters, andβ ∈ RM3 is a vector of choice-independent parameters.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 28: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/28.jpg)
Models of discrete choice — an overview
We make some assumptions about the distribution of the randomvariables ǫij — usually independence for different values of i . Letd ∈ JN denote the choices of all individuals, and x ∈ RM1N thepersonal characteristics of all individuals. Our goal is to estimatethe values of the parameters α = (αj) ∈ RM2J , β given ourobservations (x , d).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 29: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/29.jpg)
An example
Assume that ǫij are distributed independently with a Type 1extreme value distribution:
P(ǫij ≤ h) = e−e−h
. (2)
Then, the likelihood of observation i would be
Pi =e uidi
∑Jk=1 e uik
, (3)
and of the whole sample —
L(x , d , α, β) =
N∏
i=1
Pi . (4)
Maximizing L will give us the maximum-likelihood estimates of αand β.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 30: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/30.jpg)
An example
Assume that ǫij are distributed independently with a Type 1extreme value distribution:
P(ǫij ≤ h) = e−e−h
. (2)
Then, the likelihood of observation i would be
Pi =e uidi
∑Jk=1 e uik
, (3)
and of the whole sample —
L(x , d , α, β) =
N∏
i=1
Pi . (4)
Maximizing L will give us the maximum-likelihood estimates of αand β.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 31: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/31.jpg)
An example
Assume that ǫij are distributed independently with a Type 1extreme value distribution:
P(ǫij ≤ h) = e−e−h
. (2)
Then, the likelihood of observation i would be
Pi =e uidi
∑Jk=1 e uik
, (3)
and of the whole sample —
L(x , d , α, β) =
N∏
i=1
Pi . (4)
Maximizing L will give us the maximum-likelihood estimates of αand β.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 32: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/32.jpg)
Assumptions
Assumption 1. There exist K player agents. Each player agent k
can choose some action yk from a finite strategy set Sk .Put S = ×Sk . Let y ∈ S denote an action profile for the playeragents. For any k and y ∈ S , let y−k be the actions of all playeragents other than k .
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 33: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/33.jpg)
Assumptions (cont.)
Assumption 2. The are N individual (or non-player) agents. Thepayoff to an individual i choosing an alternative j ∈ J depends onthe actions of the player agents:
uij = u(xi , αj , β, y , j) + ǫij = uij + ǫij . (5)
Individuals observe y before making their choices. d and x areknown to the observer.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 34: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/34.jpg)
Assumptions (cont.)
Assumption 3. Every realization of d defines a payoff Uk(d , y) toevery player agent k , for every y . The player agents know the truevalues of the parameters (α, β) and x , but cannot observe ǫijs.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 35: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/35.jpg)
Expected payoffs to player agents
Assuming that ǫij are independent, the expected payoff of playeragent k is
Uk(x , α, β, y) =∑
δ∈JN
(
N∏
i=1
piδi(xi , α, β, y)
)
Uk(δ, y). (6)
where δ runs through all possible choice profiles, and piδiis the
probability that individual i chooses alternative δi .
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 36: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/36.jpg)
The final assumption
Assumption 4. The observed actions y are a Nash equilibrium in agame with players 1, . . . ,K , strategy sets Sk , and utilities
Uk = Uk + ǫk , (7)
where ǫk are independent random variables. The values ǫk areknown to all player agents, but not to the observer.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 37: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/37.jpg)
Consider two agent action profiles, y and some y ′. Denote by
Pj(x , α, β, y , y ′) = P(Uk(x , α, β, y) ≥ Uk(x , α, β, (y ′k , y−k)) (8)
the probability that agent k choses action yk over action y ′k , given
that all other agents choose y−k .
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 38: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/38.jpg)
Suppose that S = ×Sk is a set of action profiles, with y ∈ S .Suppose that agent k knows that all other agents will choose y−k .As ǫk are independent, the likelihood that he chooses action yk
from Sk is the probability that any pairwise comparison between yk
and any other action y ′k ∈ Sk is in favor of yk .
The likelihood of observing y ∈ S is then
LP(x , α, β, y , S) =K∏
k=1
∏
y ′
k∈Sk−{yk}
Pj(x , α, β, y , (y ′k , y−k)). (9)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 39: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/39.jpg)
Suppose that S = ×Sk is a set of action profiles, with y ∈ S .Suppose that agent k knows that all other agents will choose y−k .As ǫk are independent, the likelihood that he chooses action yk
from Sk is the probability that any pairwise comparison between yk
and any other action y ′k ∈ Sk is in favor of yk .
The likelihood of observing y ∈ S is then
LP(x , α, β, y , S) =K∏
k=1
∏
y ′
k∈Sk−{yk}
Pj(x , α, β, y , (y ′k , y−k)). (9)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 40: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/40.jpg)
Suppose that S = ×Sk is a set of action profiles, with y ∈ S .Suppose that agent k knows that all other agents will choose y−k .As ǫk are independent, the likelihood that he chooses action yk
from Sk is the probability that any pairwise comparison between yk
and any other action y ′k ∈ Sk is in favor of yk .
The likelihood of observing y ∈ S is then
LP(x , α, β, y , S) =K∏
k=1
∏
y ′
k∈Sk−{yk}
Pj(x , α, β, y , (y ′k , y−k)). (9)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 41: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/41.jpg)
The definition of the ML estimator
Definition
Let 0 < γ ≤ 1, and S be a set of alternatives for player agents.
The weighted Nash equilibrium maximum likelihood estimator of
(α, β) maximizes the weighted likelihood function
L = LP(x , α, β, y , S)γL(x , d , α, β)1−γ . (10)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 42: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/42.jpg)
Example — a probabilistic voting model
Suppose that individuals represent a representative sample ofvoters. Note that as K = J, I will use subscript j to index playeragents.
xi ∈ RM1 — personal characteristics (such as age or income)
vi ∈ RM4 — individual’s preferences with respect to thepolicies that will be carried out by the winning party in theelection
The choice variable di represents the index of the politicalparty that the individual intends to vote for in the upcomingelection.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 43: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/43.jpg)
Example — a probabilistic voting model
Suppose that individuals represent a representative sample ofvoters. Note that as K = J, I will use subscript j to index playeragents.
xi ∈ RM1 — personal characteristics (such as age or income)
vi ∈ RM4 — individual’s preferences with respect to thepolicies that will be carried out by the winning party in theelection
The choice variable di represents the index of the politicalparty that the individual intends to vote for in the upcomingelection.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 44: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/44.jpg)
Example — a probabilistic voting model
Suppose that individuals represent a representative sample ofvoters. Note that as K = J, I will use subscript j to index playeragents.
xi ∈ RM1 — personal characteristics (such as age or income)
vi ∈ RM4 — individual’s preferences with respect to thepolicies that will be carried out by the winning party in theelection
The choice variable di represents the index of the politicalparty that the individual intends to vote for in the upcomingelection.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 45: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/45.jpg)
Example — a probabilistic voting model
Suppose that individuals represent a representative sample ofvoters. Note that as K = J, I will use subscript j to index playeragents.
xi ∈ RM1 — personal characteristics (such as age or income)
vi ∈ RM4 — individual’s preferences with respect to thepolicies that will be carried out by the winning party in theelection
The choice variable di represents the index of the politicalparty that the individual intends to vote for in the upcomingelection.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 46: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/46.jpg)
Example — continued
Let the utility functions of the individuals be given by
uij = aj + αTj xi + β‖vi − yj‖
2 + ǫij = uij + ǫij , (11)
where
aj is a party-specific constant,
αj ∈ RM1 is a party-specific vector of parameters,
β is a parameter, ‖ · ‖ is the Eucledian norm,
yj ∈ RM4 is the policy program of party j ,
ǫij is an independent random variable.
Values xi are usually the socio-economic characteristics of thevoter (age, religion, etc).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 47: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/47.jpg)
Example — continued
Let the utility functions of the individuals be given by
uij = aj + αTj xi + β‖vi − yj‖
2 + ǫij = uij + ǫij , (11)
where
aj is a party-specific constant,
αj ∈ RM1 is a party-specific vector of parameters,
β is a parameter, ‖ · ‖ is the Eucledian norm,
yj ∈ RM4 is the policy program of party j ,
ǫij is an independent random variable.
Values xi are usually the socio-economic characteristics of thevoter (age, religion, etc).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 48: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/48.jpg)
Example — continued
Let the utility functions of the individuals be given by
uij = aj + αTj xi + β‖vi − yj‖
2 + ǫij = uij + ǫij , (11)
where
aj is a party-specific constant,
αj ∈ RM1 is a party-specific vector of parameters,
β is a parameter, ‖ · ‖ is the Eucledian norm,
yj ∈ RM4 is the policy program of party j ,
ǫij is an independent random variable.
Values xi are usually the socio-economic characteristics of thevoter (age, religion, etc).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 49: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/49.jpg)
Example — continued
Let the utility functions of the individuals be given by
uij = aj + αTj xi + β‖vi − yj‖
2 + ǫij = uij + ǫij , (11)
where
aj is a party-specific constant,
αj ∈ RM1 is a party-specific vector of parameters,
β is a parameter, ‖ · ‖ is the Eucledian norm,
yj ∈ RM4 is the policy program of party j ,
ǫij is an independent random variable.
Values xi are usually the socio-economic characteristics of thevoter (age, religion, etc).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 50: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/50.jpg)
Example — continued
Let the utility functions of the individuals be given by
uij = aj + αTj xi + β‖vi − yj‖
2 + ǫij = uij + ǫij , (11)
where
aj is a party-specific constant,
αj ∈ RM1 is a party-specific vector of parameters,
β is a parameter, ‖ · ‖ is the Eucledian norm,
yj ∈ RM4 is the policy program of party j ,
ǫij is an independent random variable.
Values xi are usually the socio-economic characteristics of thevoter (age, religion, etc).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 51: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/51.jpg)
Example — continued
Let the variables ǫij be distributed according to
P(ǫij ≤ h) = e−e−h
. (12)
Then the probability of individual i voting for party j given by
pij =e uij
∑Jh=1 e uih
. (13)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 52: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/52.jpg)
Example — continued
Let the variables ǫij be distributed according to
P(ǫij ≤ h) = e−e−h
. (12)
Then the probability of individual i voting for party j given by
pij =e uij
∑Jh=1 e uih
. (13)
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 53: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/53.jpg)
Assume that the payoff of a political party is equal to the expectednumber of votes that it will receive in the elections times aconstant µj . We have
Uj(x , α, β, y) = µj
N∑
i=1
pij . (14)
For each j , this value is a function of individual characteristics x ,the parameters α, β, and the policy platforms y .One can define a game between the J parties, where the strategyof party j is yj ∈ RM4 , and the payoff is (6).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 54: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/54.jpg)
Assume that the payoff of a political party is equal to the expectednumber of votes that it will receive in the elections times aconstant µj . We have
Uj(x , α, β, y) = µj
N∑
i=1
pij . (14)
For each j , this value is a function of individual characteristics x ,the parameters α, β, and the policy platforms y .One can define a game between the J parties, where the strategyof party j is yj ∈ RM4 , and the payoff is (6).
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 55: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/55.jpg)
Example — continued
Let ǫk be distributed as ǫij . We have
LP(x , α, β, y , S) =K∏
k=1
eUj (x ,α,β,y)
∑
y ′
j∈Sj−{yj}
eUj (x ,α,β,(y ′
j,y−j ))
. (15)
Let the weighted log likelihood function for this problem be
L = wV LV + wPLP , (16)
where LV is the likelihood of the observed voting profile, and wV ,wP are weights.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 56: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/56.jpg)
Example — continued
Let ǫk be distributed as ǫij . We have
LP(x , α, β, y , S) =K∏
k=1
eUj (x ,α,β,y)
∑
y ′
j∈Sj−{yj}
eUj (x ,α,β,(y ′
j,y−j ))
. (15)
Let the weighted log likelihood function for this problem be
L = wV LV + wPLP , (16)
where LV is the likelihood of the observed voting profile, and wV ,wP are weights.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 57: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/57.jpg)
Estimating the model: the data
A dataset for 1996 Israel Knesset elections. Pre-electionsurvey, N = 922
Policy preferences: 2 dimensions, based on 23 questions.
Security: Talks with PLO, settlements, equal rights for Jewsand Arabs, Palestine state, Oslo accords.
Religion: religious laws vs. democracy, gov. spending onreligious institutions.
Party positions estimated by experts, same factor weightsapplied.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 58: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/58.jpg)
Estimating the model: the data
A dataset for 1996 Israel Knesset elections. Pre-electionsurvey, N = 922
Policy preferences: 2 dimensions, based on 23 questions.
Security: Talks with PLO, settlements, equal rights for Jewsand Arabs, Palestine state, Oslo accords.
Religion: religious laws vs. democracy, gov. spending onreligious institutions.
Party positions estimated by experts, same factor weightsapplied.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 59: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/59.jpg)
Estimating the model: the data
A dataset for 1996 Israel Knesset elections. Pre-electionsurvey, N = 922
Policy preferences: 2 dimensions, based on 23 questions.
Security: Talks with PLO, settlements, equal rights for Jewsand Arabs, Palestine state, Oslo accords.
Religion: religious laws vs. democracy, gov. spending onreligious institutions.
Party positions estimated by experts, same factor weightsapplied.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 60: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/60.jpg)
Estimating the model: the data
A dataset for 1996 Israel Knesset elections. Pre-electionsurvey, N = 922
Policy preferences: 2 dimensions, based on 23 questions.
Security: Talks with PLO, settlements, equal rights for Jewsand Arabs, Palestine state, Oslo accords.
Religion: religious laws vs. democracy, gov. spending onreligious institutions.
Party positions estimated by experts, same factor weightsapplied.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 61: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/61.jpg)
Estimating the model: the data
A dataset for 1996 Israel Knesset elections. Pre-electionsurvey, N = 922
Policy preferences: 2 dimensions, based on 23 questions.
Security: Talks with PLO, settlements, equal rights for Jewsand Arabs, Palestine state, Oslo accords.
Religion: religious laws vs. democracy, gov. spending onreligious institutions.
Party positions estimated by experts, same factor weightsapplied.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 62: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/62.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 63: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/63.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 64: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/64.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 65: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/65.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 66: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/66.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 67: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/67.jpg)
Estimating the model: the data
Two-dimensional spatial model analyzed in Schofield (2007)and Zakharov and Fantazzini (2008).
Take αj = 0.
Two largest parties (Likud and Avoda) are player agents;payoff is equal to vote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 600γ.
Hence, γ = 0 corresponds to the traditionalmaximum-likelihood estimation; for γ = 1, only the likelihoodof player agents (political parties in this case) is considered
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 68: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/68.jpg)
Estimation: the results
γ = 0 γ = 0.5 γ = 0.8Likud 0.7778 0.6135 2.3978Labor 0.9901 0.6552 1.8475Mafdal -0.6270 -1.0018 -0.8998Modelet -1.2595 -0.8874 1.7995Third Way -2.2916 -2.4721 -0.3101Shas -2.0239 -2.5701 -3.0521β -1.2075 -1.9050 -3.6245
Log likelihood (voters) -776.95 -823.0 -1,204.5Log likelihood (parties) -1,444.1 -1,338.0 -1,172.8
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 69: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/69.jpg)
Estimation: the results
γ = 0 γ = 0.5 γ = 0.8Likud 0.7778 0.6135 2.3978Labor 0.9901 0.6552 1.8475Mafdal -0.6270 -1.0018 -0.8998Modelet -1.2595 -0.8874 1.7995Third Way -2.2916 -2.4721 -0.3101Shas -2.0239 -2.5701 -3.0521β -1.2075 -1.9050 -3.6245
Log likelihood (voters) -776.95 -823.0 -1,204.5Log likelihood (parties) -1,444.1 -1,338.0 -1,172.8
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 70: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/70.jpg)
Estimation: Nash equilibrium simulation
−1.5 −1 −0.5 0 0.5 1 1.5
−1
−0.5
0
0.5
1
1.5
Avoda
Likud
Shas
3rd way
Meretz
Mafdal
Modelet
(a) Actual positions
−1.5 −1 −0.5 0 0.5 1 1.5
−1
−0.5
0
0.5
1
1.5
LikudAvoda
3rd way
Meretz
Mafdal
Modelet
Shas
(b) Nash equilibrium, γ = 0
−1.5 −1 −0.5 0 0.5 1 1.5
−1
−0.5
0
0.5
1
1.5
Likud
Avoda
Shas
Mafdal
Modelet
Meretz
3rd way
(c) Nash equilibrium, γ = 0.5Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 71: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/71.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 72: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/72.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 73: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/73.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 74: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/74.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 75: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/75.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 76: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/76.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 77: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/77.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 78: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/78.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 79: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/79.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).1977 Netherlands pre-election dataset.N = 529.Policy preferences: 2 dimensions, based on 7 questions:income distribution, reaction to terrorism, nuclear energy,state-owned enterprises, environment, multinationalcorporations, abortion.2 dimensions interpreted as economic left-right and scope ofgovernment.Party positions estimated from party elite survey. Choiceamong 4 parties: PvdA, CDA, VVD, and D66.Take αj = 0.PvdA and CDA are player agents; payoff is equal to vote share.For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 400γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 80: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/80.jpg)
Estimation: 1977 Netherlands dataset
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9−1.6
−1.5
−1.4
−1.3
−1.2
−1.1
−1
−0.9
−0.8
−0.7
−0.6
γ
β
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 81: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/81.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 82: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/82.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 83: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/83.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 84: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/84.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 85: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/85.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 86: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/86.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 87: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/87.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 88: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/88.jpg)
More estimation: Eurobarometer surveys
Originally used in Quinn, Martin, and Whitrofd (1998).
1979 UK pre-election dataset.N = 426.
Policy preferences: As with Netherlands, 1977. Party positionsestimated from party elite survey.
Choice among 4 parties: Labor, Conservative, LiberalDemocrats.
Take αj = 0.
Labor and Conservative are player agents; payoff is equal tovote share.
For each party, the strategy set has five elements: theobserved policy position, and four deviations (plus or minus 1on each dimension).
Take µ1 = µ2 = 5/N. Let the weights be wV = (1 − γ) andwP = 300γ.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 89: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/89.jpg)
Estimation: 1979 UK dataset
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9−0.7
−0.65
−0.6
−0.55
−0.5
−0.45
−0.4
−0.35
−0.3
−0.25
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 90: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/90.jpg)
Robustness of estimation
How robust are the estimated parameters with respect to ourchoice of strategy sets? Other exogenous parameters — µ, γ.
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
Each strategy set contains observed policy position and 5 or10 draws from a normal distribution.
100 trials for each |Sk | = 6 or |Sk | = 11 and σ = 0.5 or σ = 1.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 91: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/91.jpg)
Robustness of estimation
How robust are the estimated parameters with respect to ourchoice of strategy sets? Other exogenous parameters — µ, γ.
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
Each strategy set contains observed policy position and 5 or10 draws from a normal distribution.
100 trials for each |Sk | = 6 or |Sk | = 11 and σ = 0.5 or σ = 1.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 92: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/92.jpg)
Robustness of estimation
How robust are the estimated parameters with respect to ourchoice of strategy sets? Other exogenous parameters — µ, γ.
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
Each strategy set contains observed policy position and 5 or10 draws from a normal distribution.
100 trials for each |Sk | = 6 or |Sk | = 11 and σ = 0.5 or σ = 1.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 93: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/93.jpg)
Robustness of estimation
How robust are the estimated parameters with respect to ourchoice of strategy sets? Other exogenous parameters — µ, γ.
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
Each strategy set contains observed policy position and 5 or10 draws from a normal distribution.
100 trials for each |Sk | = 6 or |Sk | = 11 and σ = 0.5 or σ = 1.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 94: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/94.jpg)
Robustness of estimation
−2.5 −2 −1.5 −10
0.5
1
1.5
2
2.5
3
3.5
4σ = 0.5, |S | = 6σ = 1, |S | = 6σ = 0.5, |S | = 11σ = 1, |S | = 11
Figure: Distributions of β estimates, γ = 0.5.Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 95: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/95.jpg)
Interpretation
Larger size of S results in less dispersion of estimated β.
Larger σ results in larger |β|.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 96: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/96.jpg)
Size of strategy set
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
For each party, the strategy set has 1 + 4M elements: theobserved policy position, and four deviations (plus or minus 1on each dimension) repeated M times.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 97: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/97.jpg)
Size of strategy set
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
For each party, the strategy set has 1 + 4M elements: theobserved policy position, and four deviations (plus or minus 1on each dimension) repeated M times.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 98: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/98.jpg)
Size of strategy set
1996 Israel dataset. Likud and Avoda are player agents.
Take γ = 0.5, µ1 = µ2 = 5/N. Let the weights bewV = (1 − γ) and wP = 600γ.
For each party, the strategy set has 1 + 4M elements: theobserved policy position, and four deviations (plus or minus 1on each dimension) repeated M times.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 99: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/99.jpg)
Effect of strategy set size
0 5 10 15 20 25 30−2.25
−2.2
−2.15
−2.1
−2.05
−2
−1.95
−1.9
−1.85
M
β
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 100: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/100.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 101: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/101.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 102: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/102.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 103: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/103.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 104: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/104.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 105: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/105.jpg)
Asymptotic properties
S are user-defined, but the estimates α, β have asymptoticproperties.Fix x , d .Suppose that Dk are the k distributions formwhich the members of the strategy sets Sk are drawn.Define thelikelihood of draw yi as
f (yi |α, β) =K∏
j=1
Pj(x , α, β, y , yi ) × L1−γ
γ (17)
the probability that the observed actions y are chosen by all playeragents, given alternative yi , times the probability that nonplayeragents choose d .Given M draws we have the likelihood function
L(α, β|Y ) =∏
f (yi , α, β). (18)
As we know, the MLE estimator has asymptotic properties, such asconsistency plimM→∞β = β(x) and asymptotic normality.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 106: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/106.jpg)
A question for further research
What happens to the estimates when both N → ∞ and N → ∞?
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 107: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/107.jpg)
Some questions that can be analyzed.
1 Are parties risk-averse?
2 Do parties maximize voteshare or value some specific policy?
3 Are parties forward-looking with respect to coalition formationin cabinet?
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 108: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/108.jpg)
Some questions that can be analyzed.
1 Are parties risk-averse?
2 Do parties maximize voteshare or value some specific policy?
3 Are parties forward-looking with respect to coalition formationin cabinet?
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 109: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/109.jpg)
Some questions that can be analyzed.
1 Are parties risk-averse?
2 Do parties maximize voteshare or value some specific policy?
3 Are parties forward-looking with respect to coalition formationin cabinet?
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 110: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/110.jpg)
Some problems
There are restrictions on model specification. Consider, forexample, the following utility function for political party j :
Uj(y) = ηVj(y) − (1 − η)φ(‖yj − aj‖), (19)
where η ∈ [0, 1], Vj is j ’s vote share, aj is j ’s preferred policy, andφ(·) is an increasing function.Then likelihood is maximized at η = 0 or η = 1.
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 111: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/111.jpg)
The Bayesian approach.
Computationally intensive
Bayes ratio test can be used to compare models that are notnested
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting
![Page 112: Nash equilibrium approach to ML estimation with ...mathecon.cemi.rssi.ru/seminar/2009/Zakharov_2009_presentation.pdf · Consider estimating an econometric model of discrete choice](https://reader034.fdocuments.us/reader034/viewer/2022050109/5f46d1ca9f519e71561851b6/html5/thumbnails/112.jpg)
Thank you
Alexei Zakharov Nash equilibrium approach to ML estimation with application to voting