by Stian Reimers¤
University College London
Abstract
Taxation policy is driven by many factors, including public opinion, but little research has examined the strength and
stability of the public’s taxation preferences. This paper demonstrates one way in which preferences for progressiveness
depend on the framing of the question asked. Participants indicated how they would share a fixed tax burden between
two individuals who earned different amounts of money, either by adjusting the amount of tax paid by the two individuals,
or by adjusting the amount of post-tax income retained. The units in which tax was described — amount of money
or percentage tax rate—were manipulated orthogonally. There was a strong metric effect: Participants favored progressiveness
more when tax was described as a percentage rather than amount. However, there was also a clear interaction:
for amounts, participants favored progressiveness significantly more when considering post-tax money retained rather
than tax paid; for percentages, no such effect was found.
Keywords: tax, framing, psychology, judgment, heuristics, biases.
1 Introduction
Progressiveness in tax is a topical issue. Recently in
the United States and United Kingdom there has been
a move to increase the progressiveness of taxation: One
of U.S. President Obama’s campaign pledges was to increase
the highest rate of income tax to just under 40%.
Similarly, the United Kingdom government has recently
announced it will introduce a new higher tax band of
50% for people earning over £150,000 ($225,000) a year.
Conversely, several countries, like Russia, Ukraine, and
the Baltic States, have in recent years introduced versions
of flat taxes, where all citizens pay the same proportion
of their income no matter what its absolute value.
(These “flat” taxes often come with significant tax-free
allowances, meaning the overall tax system is still slightly
progressive. For a review of flat taxes, see Keen, Kim, &
Varsano, 2008.)
The structuring of income tax systems generally has
multiple — often conflicting — goals. The most obvious
goal is raising revenue to spend on public services:
balancing the potential extra revenue taken by high tax
rates with the disincentive to work and temptation to illegally
evade taxes that high tax rates induce. Another key potential of taxation is distributional: using, for example,
utilitarian or egalitarian criteria to redistribute wealth
between the rich and poor (see Musgrave & Musgrave,
1989, for a review).
Although the issue provokes strong opinions about the
optimal progressiveness of a tax regime, one area that is
often ignored is the extent to which the public believe
progressive taxation is fair and appropriate, and the ways
in which psychological mechanisms for representing fairness
affect tax progressiveness preferences. Existing data
give a mixed picture. Generally, when asked whether
those on higher incomes should pay a higher marginal
rate of tax, people agree (e.g., Edlund, 2003; see Kirchler,
2007, for a review). However, people’s progressiveness
preferences appear to depend on the way in which questions
are posed. For example, Roberts, Hite, and Bradley
(1994) found that when participants were asked abstract
questions about progressiveness — in the form of “are
progressive tax rates more or less fair than flat tax rates”
— students with some tax education indicated strongly
that progressive taxation was more fair, in line with other
studies. However, when asked concrete questions—how
much more tax somebody earning $40,000 should pay
relative to somebody earning $20,000 — the same participants
indicated a preference for flat taxes, in this case
the dominant response being that the former should pay
double. On the other hand, Edlund (2003) showed that
Swedish taxpayers favored progressiveness for both abstract
and concrete questions, a finding he attributed to
Swedes’ better awareness of the welfare state and other provisions paid for by taxation. Similarly, Lewis (1978)
found that participants who gave preferences for taxation
at different incomes favored progressiveness.
This inconsistency is perhaps not surprising given the
large body of work that shows that preferences are often
unstable, and the way in which options are framed
can affect people’s preferences in a large number of areas
(e.g., Kahneman & Tversky, 1979; Tversky & Kahneman,
1981; De Martino et al., 2006). Framing effects
have been specifically found in tax preferences (for reviews
see McCaffery & Baron, 2005, 2006a). For example,
McCaffery and Baron (2004) demonstrated several
framing effects in participants who chose tax regimes
for different family situations. Among other things, they
found participants indicated that regimes in which the
taxation system was described in terms of tax bonuses for
certain groups rather than tax penalties for other groups
were fairer, even though the situations were numerically
identical. They also found what they called a “metric” effect
in which participants favored more progressive taxation
when taxation levels were described as percentages
of gross income rather than absolute number of dollars.
Finally, they found a modest status quo effect: Participants
were drawn towards the initial taxation structure.
One intriguing finding of Roberts et al. (1994) was that
participants were less inconsistent between abstract and
concrete questions when concrete questions were framed
in terms of the amount of money left after tax rather than
the amount of tax paid. Furthermore, the authors note in
passing that their pattern of results “indicates that changing
the reference point from taxes to residual income produced
a shift away from regressive taxes and toward flat
and progressive taxes” (p. 182), although the assertion
was not tested statistically.
This framing effect deserves further investigation, for
at least two reasons. First, although asking people about
the amount of tax that should be paid by people on different
incomes seems the most direct way to assess preferences,
asking people about money left after tax may be
more grounded in people’s experience of paying tax, and
hence have a greater stability and validity: It is likely that
people are more able to report how much money they take
home after tax — not least because that amount appears
on their bank statements — than the amount of money
they pay in tax in a given month. Second, psychological
wellbeing appears to be associated with the—relative or
absolute – amount of consumption that a person is able
to do, or, crudely, the amount of money they have left after
tax and other unavoidable drains on resources are met
(e.g., Alpizar, Carlsson, & Johansson-Stenman, 2005).
This suggests that attempts to determine people’s preferences
for taxation should use a frame in which residual
income is foregrounded. The experiment reported here examines the effects of
two variables, in a completely between-subjects factorial
design: the effect of “tax paid” versus “money left” framing;
and the effect of describing tax as percentages or
amounts — the “metric effect” investigated by McCaffery
and Baron (2004), and interactions between the two
variables. The surface features of the task are similar to
those used by Roberts et al. (1994), using two taxpayers
who have different gross incomes. However, the response
method is different. In Roberts et al.’s paper, people were
asked about ratios of taxes, in the form “if Andy earns
$80k and Bob earns $40k, how much more tax should
Andy pay than Bob?”, and then marking on a scale with
labels of “the same amount,” “twice as much,” and so on,
which may bias people towards responding heuristically
with “twice as much,” the most available and justifiable
ratio. What I do is introduce a similar scenario, but inform
participants that the total amount of money to be
taken from Andy and Bob together will be fixed, and have
participants move a slider to indicate how the tax burden
should be split.
2 Method
2.1 Participants
A total of 384 participants were recruited using the
ipoints scheme (www.ipoints.co.uk), a UK-based internet
rewards scheme in which members accrue points for
making purchases and completing surveys. These can be
exchanged for products — such as CDs and DVDs —
and services—such as cinema and theme park tickets. A
subset of members who had opted-in to complete surveys
received an email from ipoints with a link to the experiment.
Participants were paid 50 ipoints (which had a
value of approximately 25 pence / USD 0.40 to the respondent)
for completing the two-minute experiment.
2.2 Design and procedure
The experiment was written in Adobe Flash (Reimers &
Stewart, 2007) and run online. The progressiveness question
was described in one of four ways, with a 2-factor,
2-levels-per-factor between-subjects design. The factors
were framing (money left after tax vs. money paid in tax)
and units (tax/residual income described as a percentage
vs. tax/residual income described as an amount). Each
participant completed a single trial, which used one of
the four possible descriptions, randomly allocated to each
participant. In all conditions, participants were told that
Andy earned £50,000 before tax and Bob earned £20,000
before tax, and that the total amount of tax to be paid between
the two of them would be £19,000. (This is close to the actual amount of tax they would pay between them
in the UK.) A screenshot from one condition is shown
in Figure 1, and the experiment itself is available to
try at http://www.newresearchshows.co.uk/tax.html. Instructions
for the task were the same in all conditions,
except for the penultimate paragraph. Participants in the
“money left” framing condition were given the following
line:
How much do you genuinely believe Andy and
Bob should each have once they’ve paid tax?
Participants in the “tax paid” framing condition were
given the following line:
How much do you genuinely believe Andy and
Bob should pay respectively?
In all 4 conditions, participants responded in the same
way, by moving a slider to set the distribution of tax
burden between Andy and Bob (as shown in Figure 1).
In the “money left” framing condition, “Andy pays . . .
in tax” and “Bob pays . . . in tax” were replaced with
“Andy keeps . . . after tax” and “Bob keeps . . . after tax”,
displayed in the same way, and numbers were recalculated
appropriately. In the “percentage” units condition amounts in pounds were replaced with integer percentages,
representing the same tax take. All other text remained
the same.
Participants were instructed to click and drag on a 250-
pixel gray bar. To avoid anchoring effects based on the
initial position of the slider, the slider was initially invisible:
participants clicked on the grey bar to make the
slider, which moved along the bar, visible at the location
where they first clicked. They could then drag the slider
between the extremes. At the left extreme, Bob paid all of
the £19,000 tax, and Andy nothing; at the right extreme,
Andy paid all of the tax and Bob nothing. The scale
was linear, and values updated as the slider was dragged.
Across all conditions a given position of the slider represented
the same tax split. Only the description of the split
varied.
3 Results
The following data were excluded: Duplicate submissions
(2 cases); participants who spent less than 45 seconds
in total reading the instructions for the scenario
and choosing their response (63 cases). Exploratory data
analyses conducted separately for each condition identified
total of 8 data points which were more than 3 in terquartile ranges away from the median for that condition,
which were excluded as outliers. Of the remaining
participants, 34.8% were male, and mean age was 42.2
years (SD = 13.1 years).
Overall, participants favored a mildly progressive tax
regime, with mean tax rate for Andy of 29% (£14,600
of £50,000) and for Bob of 22% (£4,400 of £20,000).
This is almost exactly the current rate of tax Andy and
Bob would pay under the current UK regime, perhaps
demonstrating a status quo bias. Means and standard
errors for participants across conditions can be seen in
Figure 2. A two-factor ANOVA, with factors units (percentage,
amount) and framing (money left, tax paid) revealed
a main effect of units, F(1, 307) = 24.3, p < .001,
with participants’ preference for more progressive taxation
stronger when taxes were described as percentages
rather than sums of money. There was no significant
effect of framing, but there was an interaction between
framing and units, F(1, 307) = 15.0, p < .001.
An analysis of simple main effects showed a marginal
effect of framing for tax rates described as percentages
F(1, 307) = 3.60, p = .059, with progressiveness favored
slightly more in the “tax paid” condition. But there was a
highly significant effect of framing for tax rates described
as amounts F(1, 307) = 12.4, p = .001, in which participants
favored significantly more progressiveness when the tax regime was described as money left after tax than
amount of tax paid.
This analysis emphasizes the “tax paid” nature of the
task. An alternative analysis could emphasize money retained,
for example by using the ratio of income remaining
between Bob and Andy, as a dependent measure, with
higher values indicating more progressiveness (in other
words, Bob’s post-tax income gets closer to Andy’s as the
ratio increases). The two dependent variables, (1) proportion
of tax paid by Andy and (2) ratio of Bob’s income remaining
to Andy’s income remaining, are monotonically
but not linearly related.
Unsurprisingly, the pattern of results obtained using
ratio of residual income is similar to that for proportion
of tax paid by Andy: the amount of money Bob retains
after tax, as a proportion of the amount of money
Andy retains after tax, is .469 (Tax Paid/Percentage), .451
(Money Left/Percentage), .407 (Tax Paid/Amount), and
.443 (Money Left/Amount). A two-factor ANOVA using
this measure as a dependent variable, rather than proportion
of tax paid by Andy, as above, showed the same
pattern of significant results: A main effect of units, F(1,
307) = 24.7, p < .001, no significant effect of framing,
and an interaction between framing and units, F(1, 307)
= 14.6, p < .001. The simple main effects were also as
before: A marginally significant effect of framing for percentages,
F(1, 307) = 3.50, p = .062, and a highly significant
effect of framing for amounts F(1, 307) = 12.0, p =
.001. The second part of the analysis was to examine the effects
of gender, age, education, and income on tax preferences.
One unexpected but notable trend was a quadratic
relationship between income and tax preference (Figure
3), suggesting that people on incomes significantly below
those of both Andy and Bob, had less of a preference
for progressiveness than those whose income is approximately
that of Bob’s, or between that of Andy and Bob.
To examine the significance of this effect, a multiple
regression analysis was used, with proportion of tax paid
by Andy as the dependent variable, and gender, age, education,
income, and income-squared as the independent
variables. To account for group differences across the 4
framing-units cells, experimental condition was recoded
as a 4-level dummy independent variable. The model was
a significant fit, F(8,281) = 6.76, p < .001, and as well as
significant effects of the experimental conditions, there
was a significant effect of income, (¯ = -0.60, t = 2.57, p
= .01), income-squared, (¯ = -0.71, t = -3.05, p = .002),
and a marginal effect of age (¯ = 0.09, t = 1.67, p = .096),
suggesting older participants favored more progressiveness.
One potential explanation for unexpected effect of income
might be due to other factors being confounded
with income. Most saliently, many of the people on low
incomes might be retired, and so have significant assets,
and a history of high earning, yet with a current low income.
To examine this, the analysis was repeated excluding
all participants aged 60 or over (in the UK few
people retire before the age of 60). The same general
pattern of results was seen, and most importantly, the effect
of income-squared in the regression remained significant.
This suggests retirement is not what drives the
curvilinear association between income and progressiveness
preferences. However, there may be other variables,
which were not examined in this post hoc analysis, that
could underlie this result, so conclusions from this relatively
uncontrolled sample must be tentative.
4 Discussion
People’s preferences for tax progressiveness were affected
both by framing the scenario in terms of income
retained versus tax paid, and by describing the tax paid by
individuals as percentages versus amounts. People preferred
more progressive taxation when the scenario was
described as percentages of income rather than absolute
amounts of money. There was also an interaction: When
taxation was described as absolute amounts of money,
people preferred more progressiveness when thinking
about post-tax income retained rather than amount of tax
paid. There was no such effect for taxes described as percentages.
The metric effect — people preferring more progressiveness
when tax is described as percentages of income
rather than amounts — replicated previous work by Mc-
Caffery and Baron (2004), and reflects the fact that describing
tax as an amount can give the impression of progressiveness
even when the percentage tax rate is low. For
example, if Andy and Bob in the experiment each paid
20% tax, Andy would pay £10,000 a year and Bob would
pay £4,000.
The framing effect for amounts was perhaps the most
important novel finding of this experiment, building on
the observation of Roberts et al. (1994). Participants preferred
more progressive taxation when the magnitude of
tax burden was described in terms of amount of money
Andy and Bob had left after tax rather than the amount of
money paid in tax.
One possible explanation is that different framing conditions
activate different notions of fairness. Framing
the task as “tax paid” could lead participants to consider
what constitutes a fair contribution to society. Framing
the task as “money left” could lead participants to consider
what is a fair amount of money for an individual to
live on. It may also affect reference points: “Tax paid”
framing could plausibly make people evaluate post-tax
income relative to gross income, whereas “money left”
could make people evaluate post-tax income relative to
zero income.
Another interesting finding was the fact that the metric
effect was substantially smaller in the money left condition
than the tax paid condition. Clearly this may be a
finding specific to the values used in this experiment. On
the other hand, if people are more consistent across units
in the money left condition, it might be because they find
that condition easier to represent, and exhibit more stable
and firmly-held preferences in that condition.
Research on attitudes towards progressiveness has
largely used questions focusing on the amount of tax paid.
There are obvious reasons for this, but if what matters to
people — and what is most readily represented — is the
amount of money they have in their pay check after tax
has been deducted, it may suggest that asking about people’s
preferences described as money left rather than tax
paid may reduce “isolation” effects (McCaffery & Baron,
2006b), focusing attention on the bottom line implications
of different degrees of progressiveness, grounded
in experience, rather than activating an abstract notion of
what a fair amount of tax to pay is. Using “money left”
framing could give a better insight into genuine preference,
and possibly reduce metric effects and other framing
or contextual biases.
Custom Search
