Attitudes Towards Taxation and Self-Serving Bias

ATTITUDES TOWARDS TAXATION ANANDI MANI, SHARUN MUKAND AND DANIEL SGROI

Todays Talk Preview Motivation I: self-serving bias and theory Motivation II: luck vs. ability and attitudes towards taxation Experimental Design (including screen shots) Results

Preview (the idea in a nutshell) People bias their recall to bolster their self-image. They would rather think of themselves as hardworking than lazy: blame bad luck when they fail, attribute success to high effort/ability. People are more/less sympathetic towards those who are like themselves. To be precise: the deserving rich are more likely to want to see lower taxes on effort than the lucky rich unless the lucky rich can fool themselves into thinking luck was not an issue when they succeeded: essentially attitudes are a function of own-experiences but crucially, also of your perceptions of what happened and these perceptions can be biased. We show this is true in a controlled lab environment. And have field tests to come

Motivation I The theory of self-serving bias (SSB)

Memory is Malleable I have done this, says my memory. I cannot have done that, says my pride, remaining inexorable. Finally memory yields. [Friedrich Nietzsche, Beyond Good and Evil; from B&T QJE2002] I had during many years followed the Golden Rule, namely, that whenever a published fact, a new observation or thought came across me, which was opposed to my general results, to make a memorandum of it without fail and at once; for I had found by experience that such (contrary and thus unwelcome) facts and thoughts were far more apt to escape from memory than favorable ones. [Charles Darwin in The Life of Charles Darwin, by Francis Darwin; from B&T QJE2002].

Bias in Assessing Luck vs. Ability? "For almost two centuries, Spain has hosted an enormously popular Christmas lottery. Based on payout, it is the biggest lottery in the world and nearly all Spaniards play. In the mid 1970s, a man sought a ticket with the last two digits ending in 48. He found a ticket, bought it, and then won the lottery. In a subsequent interview, he argued that luck had nothing to do with his choice of lottery and he was so intent on finding that particular number, because he replied, I dreamed of the number seven for seven straight nights. And 7 times 7 is 48. -- Quoted in Stanley Reisner (1977) Suggestive Evidence of a Self-Serving Bias (SSB) in Beliefs about reasons for Own Success.

Theory in a Nutshell Put (very) simply, the idea is that we move from maximising utility by choosing actions or gambles to being able to bias our recall (or information set) in a way that boosts our self-image which forms part of our utility. E.g. We may have had an argument yesterday (we cannot forget that fact) but we can easily bias our memory of whether we won the argument or not to boost our present utility. This is a self-serving bias or SSB since the bias in memory is designed to boost our own utility. We can also consider Projection Bias wherein you might think of others as like yourself so you project your beliefs about your own situation onto others ( I got rich through hard work, you are poor, so you must have been lazy... )

Key Papers Key papers: Benabou & Tirole: QJE2002 Self- confidence and personal motivation , JPE 2004 Willpower and personal rules , QJE2006 Belief in a just world and redistributive politics Carillo & Mariotti: REStuds2000 Strategic ignorance as a self-discipline device

Motivation II How beliefs about the importance of luck vs. ability in life can influence attitudes towards taxation

What Shapes Attitudes HOMO ECONOMICUS EFFECT: An individual s attitude is a function of effect of redistribution on his net income. (linear redistributive scheme all individuals with income less than average, will favor a higher tax rate) [Roberts, 1977; Varian, 1979] EFFECT OF SOCIAL PREFERENCE: Individuals are endowed with a Social Preferences over resource allocations to all individuals in society (Arrow, 1963). So there need not be a direct link between level of individual income and support for redistribution. 1. 2.

Source of Income may affect Attitudes to Redistribution Given Social Preferences (Efficiency, Fairness) Decisions made behind a Veil of Ignorance need not favor high taxes (or full equality), if incentives matter for output. If individuals think that effort should be rewarded, they may favor lower taxes. Social Learning environment re: role of Hard work vs. Luck can also affect attitudes towards tax-spend policies Piketty, (QJE 1995).

Beliefs about Luck vs. Effort and Attitudes to Redistribution Attitudes towards redistribution differ both within and across countries .(World Values Survey, 2001) America Europe Poor people trapped in poverty 29% 60% Poor are lazy and lack willpower 54% 30% In the long run, hard work brings better life 59% 34-43% .and are correlated with differences in attitudes towards taxation and spending

Social Expenditure and Beliefs (Alesina and Glaeser, 2001)

Reasons for Heterogeneous Beliefs re: Role of Luck vs. Effort in Life Outcomes Different underlying Production Functions: Some individuals are in occupations where effort is more/less important for outcomes. Identical Production Function but Learning Bias Due to Social Learning Environment (Piketty) Due to Self-Serving Bias 1. 2. 1. 2.

Summing Up Some ways that SSB and Projection can work: Beliefs of the Rich: Effort/ability main driver of life income outcomes, i.e. He/she got rich (mainly) because of own effort. (SSB) Other rich individuals also got rich (mainly) because of own effort. And poor are mostly lazy (Projection Bias) Beliefs of the Poor: Luck is the main driver of life income outcomes.

Related Literature Taxation: Piketty and Saez (QJE2011), Diamond and Saez (JEP2011) Attitudes towards redistribution: Alesina and Glaeser, (2001). Alesina and Angeletos (AER2005), Piketty (QJE1995), Di Tella, et al (QJE2006); Giuliano- Spilimbergo(2009) Psychology. Blind Bias Spot. Pronin (2006, 2009)

Experimental Design

Our Questions BEHAVIOR (Motivation I) Do individuals have a SSB/Projection Bias in how they update and process information (on the role of Luck versus Ability in shaping Life outcomes)? PUBLIC ECONOMICS (Motivation II) What are the behavioral underpinnings of Attitudes towards Tax and Redistribution?

Three Hypotheses 1: When choosing what tax rate to set, what you are taxing matters (luck or ability/effort). 2: When choosing what tax rate to set, your own life experiences matter (have you been lucky or displayed high ability/effort). 3: People will make use of self-serving biases when they can to bolster their self-perceptions. Also we examine supplementary issues such as the importance of gender, background, prior political beliefs, etc.

Why the Lab? We are hoping to produce a causal link between attitudes and choices Moreover we are looking at a prior link between experiences/information and attitudes The former could potentially be done in a field/survey context (albeit with loose incentives) The later requires more control: only really feasible in a lab where we can endow people with different information and experiences (essentially gift them with luck or not)

Outline of Key Features Individuals undertake a task that involves both effort and luck (a lottery), one of which turns out to matter. Task completion results in a payout: they become Rich or Poor . Rich can be of two types: Deserving and Undeserving (Lucky) Rich. Poor can be of two types: Lazy Poor and Unlucky Poor. Subjects choose Tax Rates under Two Information Treatments: Partial Information about sources of Own Income (Scope to deceive oneself about role of luck, depending on own outcome). Full Information about sources of Own Income (harder to deceive oneself). Examine Tax rates chosen to test our hypotheses.

Timeline Registration and log-in. Introductory questionnaire (1) Additions task (2): 5 minutes to add up a series of 5 2-digit numbers) and if they exceed a threshold they are allocated as HIGH effort (equivalently LOW effort). Random Lottery (3): each subject endowed with HIGH or LOW lottery payouts with 50:50 (known) odds. Partial or full information on their performance is revealed. Tax choices (4): 2 screens designed to pick-up how they would tax luck and effort . Payment and Debriefing

Logistical Details Recruitment and registration fully anonymous. We need a good number of subjects to cover our hypotheses (we will be comparing across subgroups) To that end, we obtained 452 subjects in total that is a large number for a lab experiment. Fully computerized on-screen terminals (complete privacy). Fully incentivized: 5 show-up fee with the potential to win up to 20 for 45 minutes work.

(1) Questionnaire Content Initial questions include: Gender; Age; Subject/Maths ability; Background (parents occupation, type of school state or private, student loan); Political views & attitudes to luck and hard work Brief probability question.

(2) The Effort Task Subjects have 5 minutes to undertake some simple additions questions. Each question consists of Five 2-digit numbers, e.g. 36 + 41 + 84 + 72 + 92 = ? Typically we saw between 5 and 20 correct additions in the pilot. We set a threshold of around 15, above which they are categorized (by the software) as HIGH effort, below which LOW effort.

(3) The Wealth Lottery Subjects face a simple probability (with a prior of 50:50) that they are awarded a LOW lottery win or HIGH lottery win. They know the density but not necessarily their own outcome of the lottery (depends upon their information treatment).

Rich and Poor Total earnings follow a simple scheme: Either the LOTTERY or ADDITIONS task is chosen at random to be the important task. If the lottery is deemed important then subjects are allocated to be RICH if they scored HIGH in the lottery or POOR if they scored LOW. If the additions task is deemed important then subjects are allocated to RICH if they scored above the threshold (X) and POOR otherwise. Common Info (Prior to addition task): Payment Scheme, Distribution of threshold (X), Expected Distribution of Rich vs. Poor outcomes . A sensible choice of X is based on Pilot data (varies around 15). It is not determined for sure to avoid certainty in the minds of the subjects about whether they have made it or not.

Information Treatments Roughly half the subjects receive Full information (FI) about their performance, the rest Partial Information (PI). Full information (FI) includes Wealth Lottery Outcome Number of correct Additions Threshold X Outcome: Rich or Poor (pre-tax) Partial Information (PI) includes only outcome: Rich or Poor

(4) Tax Preferences All subjects asked to set a tax rate (for the rich) on luck (income if obtained via the lottery) and a tax rate on effort (income if obtained via the additions task). A key variable we will examine later is the difference between the two tax rates. One person per session randomly chosen as tax setter and his/her tax is used. (Subjects know this). Two Cases (each applied in half the sessions, known to subjects): Tax Case 1: Chosen Tax rate is applied to all subjects based on their total (pre-tax) earnings. Tax Revenue is redistributed equally among all subjects. Tax Case 2: As in (1) above except the tax-setter is not subject to taxation or redistribution (they are instead paired with another tax-setter). 1. 2.

Results

Methodology We have just finished collecting data so this will be a rough overview We will focus on the tax-setter excluded results (the tax-setter included are very similar except rates tend to be lower among the rich for obvious (HOMO ECONOMICUS) reasons. The key variable of interest will be the difference between the tax rate on luck (the lottery tax) and on effort (the additions tax). With 452 subjects we can afford to split the sample into rich/poor, or full information/partial information when required.

Hypothesis 1 Does the source of wealth of the person to be taxed matter? Here we just need to check if tax rates are chosen to be higher for luck than ability across all possible subjects information type does not matter here since they know for sure the source of the person being taxed; so the full sample can be used. Easily done through eye-balling the raw averages and running t-tests.

Tax Rates (setter excluded) Overall Full Information Partial Information No. of obs. Addition Tax Lottery Tax No. of obs. Addition Tax Lottery Tax No. of obs. Addition Tax Lottery Tax Lucky Rich 108 26.3 43.4 39 29.42 40.89 29 24.8 39.8 Deserving Rich 68 19.3 40.3 47 15.28 44.28 61 23.9 42.2 176 23.6 41.5 86 23.01 42.43 90 22.7 40.5 RICH Unlucky poor Low effort poor 116 43.3 69 65 50.8 72.16 51 44.9 73.9 160 47.9 72.9 78 42.01 66.23 82 45.2 71.7 276 46 70.7 143 46.81 68.93 133 45.1 72.5 POOR

Interpretation Eye-balling makes it clear that hypothesis 1 is right, and in fact the difference between tax rates is very high. This is supported by a battery of t-tests suggesting with very high confidence that the tax on luck is higher than the tax on effort/ability across the board. Note: this also coincides with optimal tax ideas (we should tax something that is not under our control...like luck ) and not de-incentivize hard work. However, while of some interest, this is not very surprising nonetheless without this result hypotheses 2 & 3 would not make sense...

Hypothesis 2 Does own-source of income matter? Our underlying hypothesis is that it does and that those who know they got rich through effort will behave differently from those who know they got rich through luck. We need to restrict the sample to those who know their source of income (the full information treatment) which will reduce the number of observations. Then we can regress the tax rate on effort and luck on what matters for the tax-setter s income (luck or effort), together with any controls that seem to matter (from the questionnaire).

Regression Variables Effort Income: the income gained by subjects from effort (a dummy = 1 if they were high effort types. Female: gender variable (1 = female, 0 = male). Politics (political spectrum measure, 1-7: higher indicates more right-wing in the initial questionnaire). Constant: a direct measure of the average tax rate, or average difference between rates. Finally, the number of observations reflects the different subgroups (rich vs. poor, full info only).

Regressions on The Tax Difference by Source of Income Regression No Sample 1 2 3 4 5 6 FI RICH FI RICH FI RICH Luck Tax - Effort Tax FI POOR FI POOR FI POOR Luck Tax - Effort Tax Effort Tax Luck Tax Effort Tax Luck Tax Inc Source is Effort -14.143*** [4.213] 3.388 [6.251] 17.000*** [6.043] -3.972 [5.829] 0.813 [3.190] 11.008 [11.645] 86 0.11 8.792 [5.436] -5.938 [5.044] -13.685** [5.664] -14.443** [5.722] 1.687 [2.456] Female Politics 29.426*** 40.894*** [3.538] 86 0.11 42.015*** 72.169*** 31.431*** [3.907] [3.576] 143 143 0.02 0.01 Constant [4.253] 86 0.00 [11.356] 143 0.1 Observations R-squared Robust standard errors in brackets; * sig at 10%; ** sig at 5%; *** sig at 1%

Interpretation Powerful support for hypothesis 2 coming from the rich: The rich who got rich through effort are significantly more generous when setting the effort tax pulling down the rate by over 14%. The difference (luck effort) widens for the high effort rich though the affect is coming through the effort tax rate falling not the luck tax rate rising. The results from the poor are exactly opposite (which again confirms hypothesis 2). Gender matters: females seem to care about ex post inequality regardless of income source as compared to males. Political stance unimportant.

Hypothesis 3 To what extent to people self-delude when forming their attitudes? The key here is to compare the partial vs. full information treatments. Do those who do not know why they were successful behave differently from those who do and do they do so in a direction that suggests self-serving biases are in place. We can eyeball the original table to see something interesting is in place and then confirm with some regressions. The controls from the questionnaire also suggest some interesting gender effects, as well as a role for political leanings and private schooling.

New Regression Variables Full Info: Set equal to 1 if they were in the FI treatment. Interaction: Set equal to 1 only if their income source was effort (the additions task) and they knew it (FI treatment).

Regressions on The Tax Difference by Information Regression No Sample 7 8 9 10 FI & PI RICH Luck Tax - Effort Tax 1.51 [5.795] FI & PI RICH Luck Tax - Effort Tax 0.459 [5.350] -5.419 [4.031] -4.032* [2.103] -4.288 [5.025] 17.310** [7.998] 34.600*** [8.790] 176 0.08 FI & PI POOR Luck Tax - Effort Tax -2.595 [6.327] FI & PI POOR Luck Tax - Effort Tax -1.615 [6.109] -14.042*** [4.233] 3.236* [1.840] 0.154 [6.600] -11.776 [8.318] 24.883*** [9.012] 276 0.08 Inc Source is Effort Female Politics -4.401 [5.015] 16.021* [8.201] 15.869*** [3.432] 176 0.06 1.095 [6.722] -12.135 [8.544] 29.059*** [5.009] 276 0.03 Full Information Interaction Constant Observations R-squared

Interpretation Again, support for hypothesis 3 (SSBs matter): A significant interaction term indicates that if I am deserving rich and I know it I become more sympathetic towards the rich who obtained their income through effort. Again, significance falls off on the poor which makes sense (the taxes are on the rich, not the poor, so projection is less relevant here) but the signs make sense (if I find out I am a low- performing poor then I am not especially sympathetic towards the rich who obtained their income through effort). Gender again is important but now so too is political leaning: the more right wing (or female) I am the more I want to shrink the gap between tax rates if I am rich. If I am poor this reverses for the right wing, so they become keener on being relatively tough (lenient) on luck (effort).

Conclusions