Happy Voters. Exploring the Intersections between Economics and Psychology. Federica Liberini 1, Eugenio Proto 2 Michela Redoano 2.

Transcription

1 Exploring the Intersections between Economics and Psychology Federica Liberini 1, Eugenio Proto 2 Michela Redoano 2 1 ETH Zurich, 2 Warwick University and IZA 3 Warwick University 29 January 2015

2 Overview Main Hypothesis When individuals are happier are they more likely to support the Incumbent Parties? SWB measures can be used to proxy for utility and explain variation in voting intention that goes beyond the variation captured by the standard financial and economic indicators? Are voters rewarding only the increase of wellbeing the government is directly responsible for, or are they also sensible to shocks exogenous to the government action?

3 Overview Main Results Yes, individuals when they are happier are more likely to support the Incumbent Parties. Yes, SWB can explain variation in voting that goes beyond the variation captured by the standard financial and economic indicators No, shocks in happiness that can reasonably be considered exogenous affects the support for the incumbent parties

4 Overview Outline 1 Overview 2 Background Literature 3 Data 4 Baseline Model and Results 5 Difference in Difference Model and Results 6 Binomial Probit Model and Results 7 Conclusions

5 Background Literature Retrospective voting Effect of financial and economic outcomes on voting decisions: Kramer 1971; Fiorina 1979, 1981; Kinder and Kiewiet 1981;Markus 1988; Lewis-Beck 1988 Happiness as a proxy for Utility : Kahneman and Thaler (1991); Rabin (1998); Layard et. al. (2008); Benjamin et al. (2012) Partnership effect and reverse causality: Di Tella and MacCulloch (2005); Dolan, Metcalfe, and Powdthavee, (2008)

6 Background Literature Voters rationality Blind Retrospection: Achen and Bartels (2004); Healy, Malhotra and Hyunjung Mo (2010); Wolfers (2009) Why Blame?: Gurdal et al. (2013). Exogenous shocks influences political support: Powdthavee and Oswald (2010, 2014), Giuliano and Spilimbergo (2014)

7 Data Dataset British Household Panel Survey (BHPS). Data from 1991 to ,000 British residents (as well as their families ) on yearly basis. The database contains, beside well being questions, information on political orientation and participation, voting behavior and intentions, personal information on finances, jobs, family status as well as region of residence.

8 Data Measuring Well Being and Voters Intentions Main measure of well being based on the responses to the question: How dissatisfied or satisfied are you with your life overall? There are seven possible categories, from #1 (not satisfied at all), to #7 (completely satisfied). To construct our measure of voting intentions we use the answers relative to the following questions: If there were to be a General Election tomorrow, which political party do you think you would be most likely to support?

9 Data Partisan Voters We use two questions from BHPS to identify non ideological voters: 1. Generally speaking do you think of yourself as a supporter of any one political party? 2. Do you think of yourself as a little closer to one political party than to the others? To the extend that the respondent replies no to both questions (1) and (2) he/she is classified as non partisan.

10 Data Summary Statistics Variable Obs Resp. Mean Std. Dev. Min Max Support Incumbent Life Satisfaction Times Resp. is Non Partisan married Widowhood Income (ln) Female Age Fin. situation = Better Fin. situation = Worse Note: data used for these descriptive statistics include the balanced sample of all individuals observed consecutively for all years between 1996 and Respondents dropped from the sample include those who were below the age of 16 in 1991, as well as the individuals in the top percentile of the income distribution and of the age distribution.

11 Data Summary Statistics, cont d Average Life Satisfaction, conditional on Political Ideology Labour Partisan Non Partisan Conservative Partisan Strong Medium Weak Weak Medium Strong Conservative (1.558) (1.435) (1.306) (1.337) (1.147) (1.307) (1.435) Labour (1.582) (1.362) (1.296) (1.320) (1.182) (1.284) (1.491) Labour (1.421) (1.323) (1.269) (1.316) (1.151) (1.201) (1.339) Labour (1.438) (1.274) (1.217) (1.282) (1.102) (1.222) (1.279)

12 Baseline Model and Results The Baseline Model Does life satisfaction matter for voting decisions (intentions)? Model 1 (traditional model) Support Incumbent it =γfinancial it + δx it + ηy i ε it Model 2 (wellbeing model) Support Incumbent it =βwellbeing it + δx it + ηy i ε it Model 3 (full model) Support Incumbent it =β WellBeing it + γ Financial it + δ X it + ηy i ε it Estimators: OLS (with FE), Probit (with RE-Chamberlain)

13 Baseline Model and Results Baseline Equation, Linear Probability Models on full sample of respondents Dependent Variable: Financial Situation Only Life Satisfaction Only Financial Situation and Life Satisfaction 1 If Supporting Incumbent [1] [2] [4] [5] Party [3] Financial Situation: *** *** *** Better (0.0046) (0.0046) (0.0046) Worse *** *** ** (0.0046) (0.0046) (0.0046) Family Income *** *** *** *** *** (0.0039) (0.0039) (0.0039) (0.0039) (0.0039) Satisfaction with Life: Satisfied [5,6,7] *** *** (0.0051) (0.0051) Satisfaction with Life: [1,2,...,7] *** *** (0.0020) (0.0020) Observations 48,432 48,432 48,432 48,432 48,432 R-squared Number of pid 4,882 4,882 4,882 4,882 4,882 Note: Baseline model looking at determinants of the probability of supporting the incumbent party. Models are estimated using an FE LPM. Sample: 4,882 respondents observed since All specifications include auxiliary control variables (a dummy for married individuals, age, age squared, and a dummy for female respondents). Region and wave dummies are also included. The variable lfsato from BHPS was used to define the level of life satisfaction. It is equal to seven different levels of life satisfaction, varying from completely satisfied (=7) to not at all satisfied (=1). For Model [2] and Model [4], the variable is recoded as a dummy identifying whether the individual is satisfied (>4), whereas for Model [3] and [5], life satisfaction is used as a continuous variable. Standard errors are clustered by respondent and reported in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

14 Baseline Model and Results Baseline Equation, Linear Probability Models on restricted sample of non partisan voters Dependent Variable: Financial Situation Only Life Satisfaction Only Financial Situation and Life Satisfaction 1 If Supporting Incumbent [1] [2] [4] [5] Party [3] Financial Situation: Better (0.0089) (0.0090) (0.0089) Worse (0.0088) (0.0088) (0.0088) Family Income * * * ** ** (0.0074) (0.0039) (0.0073) (0.0073) (0.0074) Satisfaction with Life: Satisfied [5,6,7] *** *** (0.0087) (0.0087) Satisfaction with Life: [1,2,...,7] *** *** (0.0034) (0.0034) Observations 12,926 12,926 12,926 12,926 12,926 R-squared Number of pid 1,520 1,520 1,520 1,520 1,520 Note: Baseline model looking at determinants of the probability of supporting the incumbent party. Models are estimated using an FE LPM; the coefficients reported represent the marginal effects. Sample: 1,520 respondents who are classified as swing voters. All specifications include auxiliary control variables (a dummy for married individuals, age, age squared, and a dummy for female respondents). Region and wave dummies are always included. The variable lfsato from BHPS was used to define the level of life satisfaction. It is equal to seven different levels of life satisfaction, varying from completely satisfied (=7) to not at all satisfied (=1). For Model [2] and Model [4], the variable is recoded as a dummy identifying whether the individual is satisfied (>4), whereas for Model [3] and [5], life satisfaction is used as a continuous variable. Standard errors are clustered by respondent and reported in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

15 Difference in Difference Model and Results Identification of the effect of happiness on Party Support So far we can only say that higher levels of life satisfactions affect positively the probability of supporting the ruling party. But we are not able to separate the effect of government policies from the effect of government-unrelated outcomes on life satisfaction. So we cannot tell if voters support the incumbent because they directly attribute their increased well being to public policies, or they are just bias towards the status quo when they are happy(ier). How can we identify this effect?

16 Difference in Difference Model and Results Identification Strategy Consider the death of the spouse as an exogenous shock of (un)happiness; Find a matching sample of individuals that did not experience the shock but have the same ex-ante probability of experiencing it (matched control group); Compare before and after the shock changes in political support responses of affected individuals to changes in political support responses of unaffected individuals.

17 Difference in Difference Model and Results Propensity score matching Treated Group: 363 respondents whose spouse dies between Propensity Score: estimated probability that a respondent experiences widowhood anytime between 1992 and 2008, given his/her characteristics in Estimates used for one-to-one matching with a confidence interval of Covariates in PS Equation: gender, age, empl. status (also of spouse), income, health indicators, household composition. Matched Control group: individuals whose ex-ante probability (as in 1991) of experiencing widowhood is the same as for those who did loose their spouse between 1992 and Final sample: 460 individuals, 230 ( 153 females and 77 males) from the treated group and 230 from the matched control group.

18 Difference in Difference Model and Results Propensity score matching, cont d Bias Mean Equality of Means Sample Treated Control % % Reduced t-test p>t Age in 1991 Unmatched Matched Female Unmatched Matched In Working Age in 1991 Unmatched Matched Dummy: 1 if ever smoked in life Unmatched Matched Dummy: 1 if had permanent job in 1991 Unmatched Matched Dummy: 1 if employed full time in 1991 Unmatched Matched Dummy: 1 if Spouse/Husband was employed in 1991 Unmatched Matched

19 Difference in Difference Model and Results Propensity score matching, cont d Bias Mean Equality of Means Sample Treated Control % % Reduced t-test p>t ln (Household Income) in 1991 Unmatched Matched Dummy: 1 if in good health in 1991 Unmatched Matched Dummy: 1 if visited GP more than twice in 1991 Unmatched Matched Dummy: 1 if ever hospitalized in 1991 Unmatched Matched Dummy: 1 if ever used Alternative Medicine Unmatched Matched Dummy: 1 if regularly check blood pressure Unmatched Matched Dummy: 1 if regularly does chest x-ray Unmatched Matched Dummy: 1 if regularly check cholesterole Unmatched Matched

20 Difference in Difference Model and Results Voting after Happiness Shock determined by the spouse death Probability of Supporting Incumbent Treated Group Probability of Supporting Incumbent Years of Treatment* Prob. (Support Incumbent) All Other Years Years of Widowhood* Prob. (Support Incumbent) Control Group Treatment Group Average Life Satisfaction Treated Group Average Life Satisfaction Years of Treatment* Average Life Satisfaction All Other Years Years of Widowhood* Average Life Satisfaction Control Group Tratment Group * "Years of Widowhood" refer to year of decess and two following years

21 Difference in Difference Model and Results Difference in Difference set up: estimation Support Incumbent it = α + β 1 treated i +β 2 after it treated i + β 3 after it +γ X it +δ t +u it treated = dummy for treated group (widow) after =dummy for after the treatment (death of the spouse) β 2 =estimated coefficient of interest, the effect of the treatment on the treated group, compared to the control group.

22 Difference in Difference Model and Results Difference in Difference on full matched sample Dependent Variable: Support Incumbent (1) (2) (3) (4) (5) (6) Treated (0.0405) (0.0406) (0.0422) (0.0422) (0.0451) (0.0451) After*Treated ** (0.0396) (0.0371) (0.0354) After*Treated*Female * ** (0.0457) (0.0438) (0.0434) After*Treated*Male (0.0598) (0.0583) (0.0577) Observations 3,162 3,162 2,543 2,543 1,862 1,862 R-squared Note: sample composition is 230 treated and 230 matched control individuals; OLS estimate, where after it is set to 1 in the years after spouse death. Models (1) and (2) restrict to four years before and after spouse death; models (3), (4) to only three years before and after spouse death; models (5) and (6) to only two years before and after spouse death. All specifications also include auxiliary control variables (a dummy for married individuals, the natural logarithm of yearly Household Income, Age, Age squared). Region and wave dummies are also always used. Standard Errors are clustered by respondent and reported in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

23 Difference in Difference Model and Results Difference in Difference on matched sample of female Dependent Variable: Support Incumbent (1) (2) (3) (4) (5) (6) Treated (0.0507) (0.0527) (0.0553) (0.0579) (0.0528) (0.0580) After*Treated * ** (0.0486) (0.0457) (0.0442) Treated*(1,2) year before spouse death (0.0362) Treated*(0,1,2) year after spouse death (0.0539) Treated* (2) year before spouse death (0.0376) Treated* (1) year before spouse death (0.0412) Treated* year of spouse death ** (0.0465) (0.0544) Treated*(1) year after spouse death ** * (0.0520) (0.0597) Treated*(2) year after spouse death (0.0540) (0.0625) Treated*(3) year after spouse death (0.0595) (0.0670) Constant ** 0.985** 0.860** 0.861** 0.866** (0.395) (0.424) (0.447) (0.426) (0.426) (0.427) Observations 2,079 1,669 1,218 1,669 1,669 1,669 R-squared

24 Difference in Difference Model and Results DiD on Full Matched Sample, Effect of Labour Legislatures Dependent Variable: Support Incumbent [1] [2] [3] [4] [5] [6] Treated (0.0408) (0.0428) (0.0459) (0.0459) (0.0459) (0.0485) After*Treated * (0.0735) (0.0730) (0.0693) After*Treated*Labour (0.0793) (0.0794) (0.0764) Observations 3,162 2,543 1,862 1,862 1,862 1,862 R-squared Note: Sample composition is 230 treated and 230 matched control individuals; Models [1] and [2] further restrict, respectively, to four and three years before and after spouse death; Models [3] to [6] restrict to only two years before and after spouse death. OLS estimates are based on the regression showed in equation 4 (SupportInc it = α + λ1xtreated i + λ2xafter it xtreated i + λ3xafter it + X it γ + δt + u it ), where after it is set to 1 in the years after spouse death. All specifications include the same control variables used in previous DID models, but in addition they introduce an interaction with the dummy Labour, which is 1 for all years when the Labour party held power. Standard errors are clustered by respondent and reported in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

25 Binomial Probit Model and Results The Biprobit Recursive Model Support Incumbent it = βwellbeing it + δx it + ηy i u it (1) WellBeing it = γwidowood i,t + δ X it + η Y i v it (2) WellBeing i,t = [0, 1] widowood i,t = year individual i became widow

26 Binomial Probit Model and Results Results frome Biprobit Recursive Model Full Sample Labour Legislations Only Model (1) Model (2) Dependent Variable: Support Incumb. Satisfied Support Incumb. Satisfied Satisfied with Life [lfsato=5,6,7] *** ** (0.2317) (0.2473) Widowhood *** *** (0.0831) (0.0865) Constant *** *** * (0.1421) (0.0930) (0.1689) (0.0997) Observations 48,432 44,149 Log-Likelihood Rho (0.1432) (0.1490) Wald Test (rho=0) Note: sample composition for Model (1) is all respondents observed since 1996; Model (2) restricts this sample to survey waves collected during Labour Legislations only. Models are estimated using a recursive bivariate probit, where the probability of supporting the incumbent depends on life satisfaction, which, in turn, is affected by widowhood. All specifications also include auxiliary control variables (a dummy for married individuals, the natural logarithm of yearly Household Income, Age, Age squared), region and wave dummies. Standard Errors are in parenthesis. *** p<0.01, ** p<0.05, * p<0.1

27 Conclusions Conclusions the happier the individual becomes the more she supports the Incumbent Parties. SWB can explain variation in voting that goes beyond the variation captured by the standard financial and economic indicators shocks in happiness that can reasonably be considered exogenous affects the support for the incumbent parties