The Political Economy of Tax Enforcement: A Look at the IRS from 1978-2010

THE POLITICAL ECONOMY OF TAX ENFORCEMENT:

A LOOK AT THE IRS FROM 1978–2010

August 14, 2013

Abstract

This paper looks at whether political ideology matters for enforcement of the nation’s tax laws.

An analysis of the resources available to the IRS suggests that the party affiliation of the President

makes no difference to the overall level of IRS resources. However, there are significant increases in

the share of the IRS budget devoted to detecting tax fraud and the number of IRS employees devoted

to criminal investigation under Democratic administrations. Audit probabilities for corporations are

also significantly higher under Democratic administrations. This increase in audit probability raises

the effective tax rate for corporations by about three percent.

Keywords: Tax enforcement, Tax compliance, IRS, Corporate audits

JEL Codes: H25 (Business Taxes and Subsidies), H26 (Tax Evasion)

1

1 Introduction

The gross tax gap is defined as the amount of true tax liability faced by taxpayers that is not paid on

time. The Internal Revenue Service (henceforth, IRS)’s most recent estimates of the gross tax gap for

the year 2006 are $450 billion or 16 percent of the true tax liability.1 One potential avenue for reducing

the tax gap is through increased spending on enforcement. However, in the budget agreement reached

between the Republican-controlled House and the Democratic-controlled Senate in April 2011, funding

on enforcement-related activities by the IRS was reduced by over $700 million relative to the $6 billion

requested by the agency for FY 2012. At the time of the bipartisan agreement reached between both

chambers of Congress, Speaker John Boehner’s office released a statement on the issue: “The Obama

administration has sought increased federal funding for the [IRS] – money that could be used to hire

additional agents to enforce the administration’s agenda on a variety of issues. This increased funding

is denied in the agreement.”2 Episodes like these have caught the attention of pundits. In a post in

The New York Times Economix blog, Bruce Bartlett, who held senior policy positions in the Reagan

and George H.W. Bush administrations wrote:

“Unfortunately, Republicans have been treating the I.R.S. like a political punching bag

for years, cutting its personnel and restricting its ability to do its job. The number of I.R.S.

employees fell to 84,711 in 2010 from 116,673 in 1992 despite an increase in the population

of the United States of 53 million over that period” (Bartlett, 2012).

The action by the Speaker of the House alluded to in the previous paragraph would not be an

isolated example of political actors attempting to influence the course of direction at the agency. In

1997 during the presidency of Bill Clinton, the IRS came under scrutiny when a series of conservative

nonprofit organizations like the Heritage Foundation and Citizens Against Government Waste were

the targets of audits, prompting a bipartisan effort in the Senate to investigate the accusation that

a political motivation was behind such audits (Time Magazine, March 24, 1997). More recently, the

revelation by IRS exempt organizations division chief, Lois G. Lerner, that IRS employees singled out

applications from “Tea Party” groups for 501c(4) tax-exempt status for additional scrutiny, has raised

questions about the degree to which political influences affect the agency. News stories such as these

suggest the possibility that the President (or the Congress) may try to alter the priorities of the agency

in order to have it serve his (or their) political goals.

In spite of the significant tax gap of $450 billion and the anecdotal evidence presented above that

political ideology impacts the operation of the IRS, there has been no systematic examination of the

extent to which partisan control influences the resources allotted to the IRS. The lack of empirical lit-

erature on this topic motivates this paper. I seek to answer a number of questions: Do party ideologies1http://www.irs.gov/newsroom/article/0„id=252038,00.html, Accessed 06/12/20122http://www.speaker.gov/Blog/?postid=235069, Accessed 02/23/2012

2

matter for tax administration? Do Democratic administrations spend more resources on tax adminis-

tration and enforcement than Republican administrations? In particular, do Democratic administra-

tions audit more corporate income tax returns? Finally, is a higher audit probability associated with

greater corporate tax collections by the Treasury?

Before seeking to answer these questions, it is worth taking a step back and asking a more fun-

damental question of why political actors would choose to achieve their policy goals through a change

in enforcement resources rather than through a change in the tax code. As Kopczuk (2006) points out,

changing the tax code typically requires politically costly tax reform, and “tax avoidance – letting well

enough alone” may be all that is possible. Kopczuk’s insights suggest that it may be politically infea-

sible for policy makers to explicitly reduce tax rates. Instead they may achieve similar outcomes by

reducing enforcement or increasing avoidance opportunities. Although reduced enforcement may not

be the most efficient way of reducing the tax burden, that might be all that is possible given political

gridlock.

This is the basic intuition for the current paper: When faced with political gridlock, Presidents

may affect the level of tax collection by varying the budget of the IRS, the administrative agency in

charge of tax collection and enforcement, and by altering how its resources are allocated across various

activities. In particular, if we assume that Republican Presidents want to reduce effective corporate tax

rates (either because of innate preferences or because that represents the views of their constituents)

and if we accept the premise that it is more difficult to change the tax code (e.g. the statutory rates)

than to reduce the probability of audits, then we arrive at a simple conclusion: Republican Presidents

will try to reduce effective tax rates by reducing the probability of audits – potentially through starving

the IRS of resources.

The paper examines whether the political affiliation of the incumbent in the White House af-

fects the size of the overall IRS budget and workforce and its allocation among various activities and

whether it affects audit probabilities for corporations. It finds that although there is no evidence that

the overall IRS budget or workforce is larger during Democratic administrations, party affiliation of

the President makes a difference to the share of those resources that are allocated towards enforce-

ment. In particular, the share of IRS budget devoted to detecting tax fraud and the number of criminal

investigators as a share of the IRS workforce are both significantly higher during a Democratic Presi-

dency. It also finds that over the period 1978–2010, corporate audit probabilities were about 40 percent

lower on average under Republican administrations than under Democratic administrations. These

findings strongly suggest that although the President has limited or no influence on budgetary levels

and resources which require the approval of Congress, he does have an influence on the allocation of

these resources to various activities.

The paper is laid out in five sections. Section 2 offers a brief literature review and Section 3

describes the data and the empirical specifications used for formally testing the hypotheses. Section 4

3

is the heart of the paper, presenting the results obtained on IRS budgets and personnel resources and

audit probability for corporations. Section 5 concludes.

2 Literature Review

This paper relates to a broad literature that documents the effects of politics on decision-making in

the economic realm. For example, Poterba (1994) finds that in the late 1980s when regional economic

downturns and increased expenditure demands led to substantial state budget deficits, political factors

played an important role in the adjustment process. Deficit adjustment was much faster in states

where the state house and governorship were controlled by the same party than when control was

divided. Furthermore, tax increases and spending cuts were both significantly smaller in gubernatorial

election years than at other times in the face of substantial state budget deficits. In addition to these

state institutions, state fiscal institutions also appeared to have real effects on the speed and nature of

fiscal adjustment to unexpected deficits (Poterba, 1994).

There is also a large body of work that examines the responsiveness of bureaucratic agency be-

havior to political actors at different levels of the government and to various economic and social

conditions. The literature mainly finds that bureaucratic agencies are responsive to the preferences of

Presidents (Olson, 1995; Scholz and Wood, 1998; Wood, 1990; Wood and Waterman, 1991, 1993) and /

or Congress (Olson, 1995, 1996; Weingast and Moran, 1983; Wood, 1992).

Focusing more narrowly on the issue of tax administration, few papers have looked at the question

of whether political ideology matters for enforcement of the nation’s tax laws by the IRS. In an early

examination of the issue, Scholz and Wood (1998) look at the ratio of corporate to individual audits at

the state level over the period 1974–1992 and find that the odds of corporate versus individual audits

change with different presidential administrations and increase with increased Democratic control

over Congress.

Another paper that has examined the effects of political influences on tax administration is Young,

Reksulak, and Shughart (2001). The authors use panel data for 33 IRS districts over a 6-year period

from 1992–1997 and find evidence that “the fraction of individual tax returns audited is significantly

lower in districts that are important to the President electorally and that have representation on key

congressional committees” (Young, Reksulak, and Shughart, 2001, pp. 201). Although the paper is

relevant for our analysis, their conclusions must be tempered by a number of considerations. First, the

study looks at a relatively short period from 1992 to 1997 and during five of these six years, President

Bill Clinton was in the White House. Thus the evidence of executive branch pressure on the IRS could

be unique to his administration. Second, during these six years, one party controlled both chambers.

During the first three years, 1992 through 1994, Democrats controlled both the Senate and the House,

whereas in the second half, from 1995 through 1997, control of Congress switched to Republican hands.

4

Thus, using this limited period of time from 1992–1997 does not let us identify the effects that stem

from controlling only one chamber of Congress independently of the effects of unified congressional

control by a single party.

Compared to the literature cited above, the present work offers a number of advantages. First,

it analyzes changes in tax administration and enforcement over a significantly longer period of time

than has been considered in any previous study. The period spanning fiscal years 1978–2010 analyzed

in the paper spans three Democratic (Carter FY 1976–1980, Clinton FY 1993–2000, and Obama FY

2009–2010) and three Republican administrations (Reagan FY 1981–1988, Bush I FY 1989–1992,

and Bush II FY 2001–2008). It also encompasses a number of significant political movements; for

example, the Reagan revolution of the 1980s, the Republican control of the House after a gap of forty

years in 1994 following the “Contract with America”, the closely divided electoral landscape of the

early 2000s, and the Democratic triumph of the 2006 and 2008 electoral cycles. In addition, barring

Scholz and Wood (1998) which looks at the ratio of corporate to individual audits, no previous study

has investigated the effect of political ideology on audit rates of corporations. Even though corporate

income tax revenue averaged only around 23 percent of personal income tax revenue over the sample

period 1978–2010, the amount of revenue obtained through audits of corporate income tax returns

was over half the amount generated directly through all audits and significantly higher than that

generated through audits of personal income tax returns.3 For example, in FY 2010, individual income

taxes amounted to $1.176 trillion whereas corporate income taxes amounted to $277 billion, less than

24 percent of the individual income tax liability. In contrast, in the same year, of the total recommended

additional taxes and penalties of $44.8 billion, $26.2 billion or a full 58 percent came from audits of

corporate income tax returns.4

The dominance of taxes and penalties from audits of corporations holds for the entire sample

period and is not unique to 2010. For 1978, the starting year of the sample period, recommended

taxes and penalties from audits amounted to $3.3 billion or about 53 percent of the $6.3 billion in

recommended taxes and penalties from all audits. Thus, if one is to examine enforcement activity at

the IRS and see whether political influences are operative on audit activity, audit rates of corporate

income tax returns need to be at the front and center of that analysis. This paper addresses that

need by focusing on the variation in the audit probabilities for corporations of different sizes over

time. Lastly, it also takes a further step and demonstrates that the variation in audit rates affects the

amount of taxes collected by the Treasury.3This does not capture the deterrent effects of audits.4In 2010, audits of individual income tax returns yielded only $15.1 billion or 34% in additional recommended taxes, with the

balance 8% coming predominantly from audits of estate and trust income tax returns and employment tax returns.

5

3 Empirical Methodology and Data

3.1 Description of Data

The primary source of data for this paper is the annual IRS Data Books. These data books offer a

picture of the IRS’ operations and provide information on revenue collections and enforcement activity.

I obtain data on IRS operating costs and number of IRS personnel (Full Time Equivalents, or FTEs)

from the data books. These books also provide data on the resources devoted by the IRS to enforcement-

related activity. The IRS does not break down these overall numbers down to the level of individuals

vis-a-vis corporations. Thus, these numbers used pertain to resources geared towards all entities

served by the IRS: Individuals, corporations, partnerships, estates, trusts, and others.

Data on the overall IRS budget and personnel resources are available over the entire period from

1978–2010, as is the data on the number of criminal investigators.5 However, given the changes in

the IRS operating structures and reporting around fiscal year 2001 following the passage of the IRS

Restructuring and Reform Act of 1998, the size of the budget dedicated to detecting tax fraud is only

available over the period 1978 through 2000.

In addition to obtaining data on IRS’ budgetary and personnel resources, I am also able to obtain

data on the percent of corporate tax returns audited in any given year. It is defined as the ratio of

the number of corporate returns audited to the number of returns filed in the previous year, because

generally audit activity is associated with returns filed in the prior calendar year. A note on availability

of corporate audit data: Although aggregate data regarding audit rates of corporations is available over

a longer period of time, data disaggregated based on the size of the corporation (assets held), is only

available for the period from fiscal year 1978 onwards; hence the choice of 1978 as the starting point

for all our analysis. The number of asset classes in which the IRS reports information changes from

year to year. However, it is possible to construct an integrated time series for the percent of returns

audited for corporations in four asset classes for the entire period from 1978–2010: Those with assets

less than $1 million, assets between $1–10 million, assets between $10–100 million, and assets in

excess of $100 million. In addition, we also have data on all corporations for which the size of their

assets is not known.

The political variables included in the analysis pertain to the partisan control of the White House

and the two chambers of Congress. In alternate specifications, I also control for whether it is a Pres-

ident’s first or second term in office. Details regarding the data sources are provided in Appendix

A.

One important aspect is the choice of the appropriate lag structure in the specification. Budgets

and priorities for a given fiscal year are generally set in the prior fiscal year. Thus audit rates, IRS

budgets and personnel, and their allocation to various activities in fiscal year t can only be ascribed to5The choice of time period analyzed is influenced by the period for which data on audit activity of corporations is available.

6

decisions reached in the previous fiscal year (t-1) which are a function of the political environment at

that point of time.6 Hence, I consider the first lags of all exogenous variables. I provide the summary

statistics in Table 1 below.

[Table 1 about here]

3.2 Empirical Approach

3.2.1 Effect of political ideology on IRS resources and allocation to enforcement-related

activity

When I look at the level of overall budgetary resources made available to the IRS, I examine the IRS

budget, normalized by overall federal outlays over the period 1978–2010. Although the IRS budget has

in fact grown in nominal terms over time, what one should care about is the size of the IRS budget rel-

ative to all federal outlays since that ratio better reflects the extent to which different administrations

prioritize enforcement of the nation’s tax laws. Thus to the extent that Republicans and Democrats

have different spending priorities, normalizing the size of the IRS budget by the size of overall federal

outlays helps us get at the causal impact of the party affiliation of the President on the resources spent

on tax administration and enforcement. In order to control for time trends that result from changing

macroeconomic conditions and tax laws, I take a non-parametric approach and include decade fixed ef-

fects in the above specification. It also makes the identification more credible since in each of the three

decades between 1978 and 2010, we have had periods of Democratic and Republican rule. The effects

of partisan control on audit probability are therefore being estimated by comparing the years under

Democratic administrations with years under Republican administrations within the same decade.7

The specification used is:

Scaled IRS budgett = β0+β1∗Party Presidentt−1+β2∗Party Senatet−1+β3∗Party Houset−1+µt+εt (1)

In the above specification, Party President is a dummy variable for the party affiliation of the

President, coded 1 when a Democrat is in the White House and 0 otherwise. Likewise, Party Senate

and Party House are also dummy variables, coded 1 when Democrats are in charge of the Senate and

the House respectively. Lastly, µt represents decade fixed effects.

I use specifications similar to (1) when I analyze the number of FTEs or the share of budgetary or

personnel resources that are devoted to detecting tax fraud. In all of these specifications, I control for6The first step in setting the federal budget involves the President presenting a budget proposal for the coming Fiscal Year

to Congress on or before the first Monday in February. This is then debated in Congress which is expected to complete action onbudget resolution by April 15 (Keith, 2008). Although the April 15 deadline is frequently not met, it generally remains the casethat the decisions regarding priorities for a given fiscal year are taken in the prior fiscal year.

7Recall that the data provided by the IRS is for a fiscal year and that political variables are being lagged by one year (seefootnote 6). Thus, for the 1980s, we have 2 years of Democratic administration followed by 8 years of Republican administration.Likewise, for the 1990s, we have 4 years of Republican administration followed by 6 years of Democratic administration, andlastly for the 2000s, we have 2 years of Democratic administration followed by 8 years of a Republican administration.

7

the partisan affiliation of both chambers of Congress and in alternate specifications, I also control for

whether the President is in his first or second term in office. An alternative approach to controlling for

time trends through using decade fixed effects is to use a linear time trend. Results presented in Table

2 through Table 4 use both these approaches.

3.2.2 Effect of political ideology on corporate audit probabilities

Using the IRS data books, I construct the audit probabilities for five different nominal asset classes,

also over the period 1978–2010. In order to examine the hypothesis of whether Democratic adminis-

trations audit more corporate returns when in office, the specification used is:

Audit probabilityi,t = β0+β1∗Party Presidentt−1+β2∗Party Senatet−1+β3∗Party Houset−1+αi+µt+εt

(2)where, Audit probabilityi,t is the percent of returns audited for corporations in asset class i in fiscal

year t. In alternate specifications, I also control for whether a President is in his first or second

term in office since it has been suggested that Presidents who are eligible for re-election audit less

intensively than Presidents who do not face the prospects of re-election (Rios, 2003). As before, in

order to account for changing macroeconomic conditions and tax laws, I include decade fixed effects. To

address concerns about serial correlation, I cluster the standard errors by asset class (Bertrand, Duflo,

and Mullainathan, 2004).8

In addition to looking at the percent of returns audited, I also examine the variation in the absolute

number of returns audited. In these specifications, I introduce the dependent variable in logarithmic

form to reduce heteroskedasticity (Wooldridge, 2008, pp. 274). I control for the log of number of returns

filed in the prior calendar year in that asset class as one would expect that the number of returns that

are audited would go up with an increase in the number of returns filed. One disadvantage to using the

figures on percent of returns audited aggregated by asset class, is that it treats the audit probability

as a step function of asset size, while it is most likely a relatively smooth function defined over asset

size. This is likely to overestimate (underestimate) the probability of audits for corporations on the

lower (upper) bound of an asset class. In order to get around this issue, we can, using the data at hand,

approximate what a smooth function defined over asset size might look like. Through a series of steps

(described in detail in Appendix B), I construct a smoothed audit function matching the empirically

observed audit probabilities for different asset sizes. I use the generated audit functions to predict

what the audit probabilities would be at pre-defined asset sizes. Given that I have four asset classes8Audit coverage ratios for a given asset class are a measure of enforcement and they may respond to non-compliance by

corporations in that asset class, raising the issue of endogeneity between enforcement activity and non-compliance. I exploredthe use of a 3 SLS framework in which audit probabilities are influenced by various measures of non-compliance such as thelevel of recommended taxes and penalties per audit and the percent of audits which did not result in a change. Those regressionsindicated that audit rates for a given asset class were not impacted by past levels of non-compliance for that asset class as bestas I could tell from the measures of non-compliance at hand. Hence the decision to use a more parsimonious specification andproceed with an OLS framework as opposed to a more involved 3SLS framework.

8

for which I know the size of their assets (“Less than $1 million”, “Between $ 1–10 million”, “Between

$10–100 million”, and “Over $100 million”), the asset sizes at which I calculate the audit probabilities

are $1 million, $10 million, and $100 million. Finally, I use the three predicted audit probabilities that

are generated from this exercise and use a specification identical to that in (2) to examine the extent

of partisan influence on these predicted audit probabilities.

3.2.3 Effect of audits on effective tax rate and voluntary compliance

One implication from a model of a corporation trying to maximize its expected post-tax profits is that

following an increase in the audit probability, it should reduce the extent of underreporting of income.

As true corporate profits can be assumed exogenous to the tax instruments such as the audit proba-

bility, a reduction in the extent of underreporting of corporate income should result in an increase in

the net revenue collected by the Treasury. In order to examine this hypothesis, I analyze variations

in the effective tax rate paid by corporations. The effective tax rate is defined as the ratio of the net

corporate tax collections, as reported by the IRS to a measure of corporate profits from the Bureau of

Economic Analysis (BEA).9 I subsequently regress the effective tax rate on the percent of returns au-

dited to examine whether the effective tax rate is responsive to variations in the audit probability. As

the effective tax rate is also affected by the highest marginal corporate tax rate,10 I control separately

for that in the regressions.

In addition to examining the effective tax rate, I also construct a measure of voluntary compliance,

defined as the ratio of average net revenue collected per return filed to the average recommended taxes

and penalties per audit. One would expect that as the audit probability goes up, this measure of vol-

untary compliance should also go up as the net revenue collected per return goes up and recommended

taxes and penalties per audit comes down.11

4 Results

4.1 Results on IRS budget and personnel

Figures 1 through 4 and Tables 2 through 4 provide the results of the analyses in which I exam-

ine variations in the IRS budget, its workforce, and the emphasis on detecting tax fraud over the9The best measure to be used in the denominator in the calculation of the effective tax rate would be the income of

corporations that is actually subject to income tax per Statistics of Income (SOI), IRS. Such data is available from theIRS for the period 1960-2007 but not beyond 2007. Since I would like a series which runs for the entire period from1978–2010, I choose to go with the numbers for profits earned by all domestic corporations made available by the BEA.(http://www.bea.gov/iTable/index_nipa.cfm) The correlation coefficient between the two different measures of corporate prof-its, that from the IRS and from the BEA is 0.9821. Given this high correlation coefficient between the two measures of profits, Iuse the numbers from BEA because it is available over the entire period of time from 1978–2010.

10I include the top statutory marginal tax rate as a proxy for the statutory tax rate faced by corporations since the vastmajority of corporate profits in any given year is earned by corporations that are subject to the top marginal corporate tax rate.

11Looking at the level of taxes and penalties, I indeed find consistent evidence that taxes and penalties are lower, both inabsolute terms and on a per audit basis, when we have a Democratic administration in place.

9

period 1978–2010. As was noted before, the IRS data on its overall resources or their allocation to

enforcement-related activity is not broken out by individuals and corporations but includes all re-

sources that are geared towards serving all of the various classes of taxpayers. The dependent vari-

ables have been rescaled in Tables 2 through 4 so as to avoid having unnecessary zeros in the estimated

coefficients. The economic and statistical significance of the results is unchanged if we avoid rescaling

the dependent variable in these tables.

4.1.1 Overall IRS resources

Figure 1 presents the underlying data while Panel A of Table 2 presents the results of estimating

specification (1) with the IRS budget, normalized by all federal expenses, as the dependent variable of

interest.

[Figure 1 about here]


As Figure 1 indicates and results in columns (1) through (4) of the first row of Panel A of Table 2

confirm, party affiliation of the President does not appear to make an impact on the overall size of the

IRS budget. The results in the second row suggest that Democratic control of the Senate raises the IRS

budget. For example, when the control of the Senate switches from Republican hands to Democratic

hands, the coefficient in column (1) suggests an increase in the size of the IRS budget (expressed as

a fraction of overall federal outlays) by 0.0389 percent. Given that the average value of IRS budget

(again expressed as a fraction of overall federal outlays) is 0.427 percent, an increase of 0.0389 percent

represents a proportional increase of over nine percent. Surprisingly perhaps, when the Democrats

are in charge of the House, the coefficients in the third row suggests that the IRS budget is smaller.

That, however, turns out to follow from the fact that denominator of the dependent variable, federal

outlays are significantly higher when Democrats are in control of the House. In absolute terms, the

IRS budget is not significantly smaller (or larger) when Democrats are in control of the House.

Overall federal outlays are comprised of outlays on defense and non-defense items. For robustness,

I also re-estimate specification (1) above by normalizing the IRS budget by total federal non-defense

outlays. The results are similar to what we observe in Panel A of Table 2; Democratic control of the

White House and the House makes no difference to the level of IRS resources whereas, Democratic

control of the Senate raises the resources available to the IRS by about eight percent of its mean value.

Results are available on request.

There is a potential problem with the use of overall federal outlays (or federal outlays on non-

defense items) in the denominator of the dependent variable in that federal outlays are susceptible to

partisan influence, similar in nature to the kinds of partisan influences that might affect the level of

resources available to the IRS. If Democratic Presidents have preferences for higher levels of federal

10

government spending in general, then that might account for the non-results in Panel A of Table 2,

in which it turns out that Democratic control of the White House does not raise the IRS budget once

normalized by all federal outlays. Panel B of Table 2 addresses this issue of possible endogeneity of

federal outlays explicitly. Instead of normalizing the IRS budget by total federal outlays, I use GDP

for normalization.

The results in Panel B of Table 2 are consistent to what we see earlier in Panel A but with the dif-

ference that in this case, Democratic control of the House is no longer seen as reducing the IRS budget.

As before, party of the President has no significant influence on the IRS budget whereas Democratic

control of the Senate raises the IRS budget by about six percent of the mean of the dependent variable.

A story similar to that for the IRS budget emerges from looking at the size of the IRS workforce.

Two approaches of looking at the size of the IRS workforce are adopted: First, I normalize it by the size

of the overall federal workforce and second, I normalize it by the size of the overall civilian labor force.

Data on the federal workforce is available only from 1981 onwards, so the regressions are estimated

over a 30-year period from 1981 to 2010, rather than being estimated over the entire 33-year period

from 1978 through 2010, as has been considered in the paper thus far. Figure 2 and Panel A of Table

3 present the results using the number of federal FTEs for normalization, while Panel B of Table 3

presents results using the size of the civilian labor force for normalization.



Looking at Figure 2 and Panel A of Table 3, I conclude that the party of the President has no

influence on the size of the IRS workforce whereas Democratic control of the Senate raises the IRS

workforce (normalized by the size of the federal workforce) by about 0.6 percent or 12 percent of the

mean. Even though Panel A of Table 3 gives the impression that Democratic control of the House

reduces the size of the IRS workforce, that observation follows from the fact that Democratic control

of the House raises the size of the federal workforce by more than the size of the IRS workforce. In

absolute terms, the size of the IRS workforce does not vary based on Democratic or Republican control

of the House.

Panel B of Table 3 examines the size of the IRS workforce as a fraction of the civilian workforce

and confirms the observations from Panel A of this table. Party affiliation of the President appears

not to matter. If anything, the results go in the opposite direction as columns (3) and (4) of Panel

B suggest that Democratic administrations reduce the size of the IRS workforce as a fraction of the

civilian workforce. Moreover, the coefficients in the second row suggest that if the Senate were to

switch from Republican to Democratic hands, then the size of the IRS workforce goes up by about 10

percent of the mean of the dependent variable. Once I normalize the IRS workforce by the size of the

civilian labor force, which is less susceptible to partisan control of Congress than the federal workforce,

11

the preliminary observation from Panel A of Table 3 that Democratic control of the House reduces the

size of the IRS workforce does not hold. This is reminiscent of the pattern of results I find in Table 2

when considering the IRS budget and alternative ways of normalizing it.

Thus, the overall picture that emerges from looking at Figures 1 and 2 and Tables 2 and 3 is that

the party of the President has little to no influence on the overall size of the resources available to

the IRS. Congress, on the other hand, has a more significant role to play in influencing the size of

the budget. In particular, I find strong and consistent evidence that Democratic control of the Senate

raises the resources available to the IRS by anywhere between 6 to 12 percent, depending on whether

we consider the IRS budget or the size of the IRS workforce and on how either of these variables are

normalized.

4.1.2 Allocation of IRS’ resources toward enforcement

The theoretical framework suggested in the Introduction indicates that the President would choose

the least cost instrument available to him to achieve his political goals. If, as our discussions suggest,

Republican Presidents would like to minimize the tax burdens for corporations, then one possible

way of achieving that would be to cut the overall IRS budget or workforce and thereby influence the

effort that IRS employees are able to devote to enforcement and compliance-related activity. However,

the coefficients on the variable for party affiliation of the President in Tables 2 and 3 suggest that

Republican Presidents do not reduce the funding level of the IRS and conversely, Democratic control

of the Presidency does not raise the level of resources at the disposal of the IRS either.

Given that IRS budgets are also directly influenced by Congress through the appropriations pro-

cess, an alternative approach that the President might choose to have his priorities be implemented is

to change the activities prioritized by the agency. In that case, it may not be very meaningful to look at

overall budgetary or personnel resources but to look at disaggregated data to see if IRS employees are

shifted from one function to another to better align with the administration’s priorities. Moreover, if

administrations hire more IRS employees in order to provide high levels of customer-service and be re-

sponsive to the needs of taxpayers, then the total number of FTEs may not be the appropriate measure

to look at in any case. In either scenario, there may not be a perceptible change in the total number of

employees but in the tasks they are assigned to and the number of returns they can audit. Since my

focus is on the extent to which Democratic administrations crack down on tax evasion and avoidance,

in the remainder of the analysis in this sub-section, I examine whether Democratic administrations

increase the resources that are dedicated to specifically detecting tax evasion and avoidance.

One way of looking at this question is to look at the budget for the Criminal Investigations division

and number of criminal investigators that the IRS hires since it is this division that investigates two

broad categories of cases: tax violations and money laundering violations (Dubin, 2004). The 2000 data

book of the IRS also suggests that if we are to focus on resources devoted to ensuring tax compliance,

12

it would be worthwhile to focus on Criminal Investigation:

“IRS Criminal Investigation’s primary resource commitment is to develop and investigate

Legal Source tax investigations. Legal Source tax investigations involve legal industries and

legal occupations and more specifically, legally earned income, in which the primary motive

or purpose is the violation of tax statutes: Title 26 (tax violations) and Title 18 (tax related)

of the U.S. Code. . . . The prosecution of Legal Source Tax Crimes cases is key to promoting

voluntary compliance with the tax laws.”

Therefore, the next two figures, Figures 3 and 4 and Table 4 specifically examine partisan influ-

ences on the share of resources that are dedicated to enforcement instead of looking at aggregate level

of resources available to the IRS. In Figure 3 and Panel A of Table 4, I examine the share of the IRS

budget that is dedicated towards detecting tax fraud and in Figure 4 and Panel B of Table 4, I look at

the number of criminal investigators, normalized by the size of the IRS workforce. Given the changes

in the IRS operating structures and reporting around fiscal year 2001 following the passage of the IRS

Restructuring and Reform Act of 1998, it is not possible to construct an uninterrupted series for the

size of the budget dedicated to detecting tax fraud over the entire period from 1978–2010. The data

available permits an analysis only over the period 1978 through 2000. Data on the number of criminal

investigators is however available over the entire period from 1978–2010.




The results in Panel A of Table 4 suggest that the share of the IRS budget dedicated to detecting

tax fraud goes up by 11 percent of the mean value of the dependent variable when a Democrat is

in charge of the White House. Partisan control of the Senate and House appear to have opposing

effects, with Democratic control of the House increasing the budget dedicated to detecting tax fraud

and Democratic control of the Senate decreasing the same. The surprising result that Democratic

control of the Senate reduces the IRS budget dedicated to detecting tax fraud can be explained by the

fact that Democratic control of the Senate raises the overall IRS budget significantly. In fact, when I

consider the absolute size of the budget dedicated towards detecting tax fraud, I find no evidence that

Democratic control of the Senate reduces it, whereas Democratic control of the White House and the

House raises it.

Looking at Figure 4 and the coefficients in the first row of Panel B of Table 4, I conclude that

having a Democrat in the White House raises the number of criminal investigators, normalized by the

IRS workforce, by about five percent of the mean value of the dependent variable and accounts for

about two-thirds of the variation in this variable. The overall picture that emerges from Figures 3 and

13

4 and both panels of Table 4 is that administrations of different political dispositions are indeed able

to influence the allocation of budgetary resources and personnel to tax enforcement. The results are

consistent with the theoretical framework outlined in the Introduction which suggests that incumbent

governments prefer to choose the least cost instrument available to them to influence policy.

4.2 Audits of Corporations

Having examined the resources available to the IRS, I present the results from the analysis of political

influence on the audit probability for all corporations for the period 1978–2010 using specification (2)

in Table 5. The dependent variable in columns (1) and (2) of Table 5 is the percent of returns audited,

and that in columns (3) and (4) is the log of number of returns audited controlling for the log of number

of returns filed in the prior calendar year.


The results in Table 5 indicate that the percent of corporate returns audited is higher during

Democratic administrations. For example, looking at the coefficient in column (2) on party of the

President, (which is a dummy variable coded 1 when a Democrat is in the White House), and dividing

it by the percent of returns audited, averaged across the five asset classes, we see that the increase

of 1.66 percent is 11 percent of the mean value of this variable (15.5 percent) and 8 percent of the

standard deviation, across all asset classes (20.1 percent).

I also use a specification identical to the one used above for examining the variation in predicted

audit probabilities, generated using the approach outlined in Appendix B. The coefficients that I obtain

for party affiliation of the President, 9.5 and 16.7 are very similar to the coefficients I obtain in the first

two columns of Table 5. Since the smoothed audit functions are constrained such that the predicted

audit probability generated using the function lies between 0 and 1, I also explore the use of a double-

censored tobit, instead of OLS. I obtain coefficients of 7.2 and 15.9 that are similar in magnitude

and statistical significance to those estimated by OLS. Full results are available on request. To the

best of my knowledge, this is the only paper in the literature that generates a smooth audit function

(rather than use a stepwise audit function) and uses predicted audit probabilities from the smooth

audit function to estimate the effect of political ideology on audit probability.

Select robustness checks I conduct a number of robustness checks to explore the sensitivity of the

result that audit probabilities are higher under Democratic administrations. To conserve space, I only

present the coefficient on the key variable of interest, “Party of the President” in Appendix Table A1.

Full results are available on request.

1. Robustness check 1 (RC 1) – Alternative specifications:

(a) Fixed Effects versus Random Effects: The choice between a random effects and fixed ef-

fects specification is made on the basis of the Hausman test. The Hausman test does not

14

reject the null hypothesis of no correlation between the unobserved asset class effects and

the explanatory variables and hence the use of a Random Effects specification. However, I

also examine the robustness of the results to Fixed Effects specification since Griliches and

Hausman (1986) stress that observing consistent estimates across alternative panel data

estimation techniques supports the absence of serious errors in variables problems. The

results reported in Table 5 are generally robust to Fixed Effects specifications.

(b) Additional controls for the intensity of Democratic control in both chambers of Congress: In

the regressions above, I control for the partisan composition of Congress by simply intro-

ducing a dummy variable for whether the Democrats control the Senate and / or the House.

An alternative approach would be to also control for the extent of Democratic control, since

one might expect the effects of a Democratic majority to be different if they enjoy a narrow

edge in Congress versus an overwhelming advantage. To operationalize this idea, I include

additional terms for the share of seats held by Democrats in each chamber of Congress

and interact those with the dummy variables for partisan control in the respective cham-

ber. The coefficients corresponding to the party affiliation of the President are substantively

unchanged.

(c) Introducing additional interactions between the party of President and term of the Presi-

dent: Different parties may need to cater to different political bases and Republican Presi-

dents may be especially reluctant to audit corporations in their first term in office. To allow

for this possibility, I introduce an interaction term between party of the President and the

term of the President. There appears no support for this hypothesis that the effect of hav-

ing a Republican in the White House is different between the two terms; in both terms,

Republican administrations audit significantly less corporate returns than Democratic ad-

ministrations.

2. RC 2: Additional lags for number of returns filed: I explore alternate specifications in which I

re-define the audit probability as the ratio of number of returns audited in a given fiscal year to

the average number of returns filed in the three-year period prior to the given fiscal year. This

is done because “audits completed in the current year include a mix of returns filed during the

previous three years.” (Scholz and Wood, 1998, pp. 152) Likewise, when I examine variation in

the log of number of returns audited, I introduce the log of the number of returns filed in all three

calendar years prior to the fiscal year for which data is being reported as control variables. The

results are very similar to those in Table 5.

3. RC 3: Dropping corporations which belong to the largest asset class: One concern with the results

above might be that for the largest firms, audit probabilities are close to 1 and there is limited

variation in the percent of returns audited over time. That is not entirely true since in this

15

sample, for the asset class that includes the largest firms, viz. those with assets in excess of $100

million, audit probabilities average 0.47 and vary between a low of 0.21 in 2009 to a high of 0.86

in 1985. In any case however, the results are robust to the exclusion of firms that belong to this

asset class.

4. RC 4: Issues with drift across nominal asset classes: The audit probability figures constructed

using the IRS data books is in terms of nominal asset classes, e.g. corporations with assets be-

tween $1 and $10 million, corporations with assets between $10 and $100 million, etc. where

the thresholds are not adjusted for inflation. One possible issue with the use of asset probability

based on these nominal thresholds is that over time, as the average asset size of corporations

increases because of inflation, more and more corporations will fall in asset classes that corre-

spond to (nominally) larger thresholds. However, given constraints on the IRS budget, fewer and

fewer of such corporations that belong to (nominally) larger asset classes would be audited re-

sulting in a general decline in audit probability over time for those asset classes. This decline in

audit probability could then be attributed to a change in partisan control of the Presidency and

the Congress possibly biasing us in favor of a spurious positive finding between political ideology

and audit probability. To rule out this alternative explanation, I introduce an additional control

variable, viz. the fraction of returns that are filed by small firms in any given year. In doing so, I

define small firms as all corporations that have assets less than $10 million, the same definition

as is used by the IRS. The results obtained in Table 5 are robust to the inclusion of this additional

control variable.

4.3 Effect of audits on effective tax rate and voluntary compliance

The evidence presented thus far suggests that partisanship of the President and Congress matters

for the allocation of IRS resources to enforcement-related activity. It would be interesting to note

whether this variation in enforcement-related resources manifests itself in the form of a variation

in the effective tax rate that is paid by corporations. This section takes steps in that direction. I

analyze the effect of the variation in audit probability on the effective tax rate and a measure of

voluntary compliance and present the results in Table 6.12 Panel A of Table 6 does not control for the

party affiliation of the President and Congress, whereas in Panel B, I add them to the specification as

additional controls.


One potential issue with the regressions involving the measure of voluntary compliance estimated

in columns (3) and (4) is that percent of returns audited shows up on both the LHS and the RHS as the12The measure of voluntary compliance used is the ratio of average net revenue collected per return filed to the average

recommended taxes and penalties per audit.

16

dependent variable in these regressions also involves the percent of returns audited.13 To get around

this issue, I use an Instrumental Variables approach in columns (5) and (6) to instrument for the

percent of returns audited in a given fiscal year. The instrument I choose is the number of corporate tax

returns filed two years prior to the fiscal year for which I look at this measure of voluntary compliance.

For example, when I examine the voluntary compliance for fiscal year 1988, the instrument I use

is the number of tax returns filed in that asset class in calendar year 1986. The measure of IRS

enforcement is the ratio of audits completed in this year divided by the number of returns filed in the

prior calendar year. My choice of instrument relies upon the fact that variation in percent of returns

audited is induced by both the denominator and numerator in this ratio – variations in the number of

returns audited by the IRS each year (which is determined by the IRS, and arguably influenced by the

President and Congress) and variations in the number of returns filed (the denominator, which is likely

to be the result of exogenous factors).14 The instrument in question, the number of tax returns filed

two years prior to the current fiscal year, is exogenous to any individual corporation’s choices and yet

impacts the audit probability since the number of returns filed in two consecutive years for any given

asset class are strongly correlated.15 The value of the F-stats in the first stage of the IV, corresponding

to the two specifications in columns (5) and (6) of Panel A is 116.46 and 34.37, and those in Panel B

are 57.42 and 24.31. All of these are well in excess of 10 which is used as a rule of thumb to exclude

weak instruments (Stock and Watson, 2002). Results from the first stage regressions are available on

request. Results using the IV approach are presented in columns (5) and (6) of Table 6.

The results in both panels of Table 6 suggest that an increase in audit probability has the intended

effect of increasing the effective tax rate and improving voluntary compliance. To provide a sense of

magnitude of these effects, the coefficient on percent of returns audited in column (1) of Panel A of

2.236 implies that raising the audit probability from 1.33 percent (25th percentile in the time series

data from 1978–2010) to 2.67 percent (75th percentile in the data) increases the effective tax rates,

on average, by three percent, which amounts to a 12 percent increase in effective tax rates.16 Taking

this a step further, I note that the average audit probability for all corporations over this period is

2.03 percent under Republican administrations and 3.34 percent under Democratic administrations.17

This increase of 1.31 percent in the audit probability as we move from a Republican to a Democratic13Measure of voluntary compliance = (Net corporate tax collected per return filed)/(Recommended taxes and penalties per

audit) = (Net corporate tax collected / Recommended taxes and penalties) * (Percent of returns audited / 100)14An instrument, similar to that described, has been used in Hoopes, Mescall, and, Pittman, 2012.15In the context of this example, the number of tax returns filed in 1987 is strongly correlated with the number of tax returns

filed in that same asset class in 1986.16Given that audits are typically completed over a number of years, I explore the robustness of the results to controlling for

the lagged audit probability and find coefficients that are very similar in magnitude and significance to those reported in Table6.

17This is the average audit probability for all corporations. If I were to average the audit probability across the five assetclasses, then I get an average of 15.5 percent as has been reported in the summary statistics in Table 1. The difference can beexplained by the fact that the first measure puts equal weight on all corporations, whereas the second measure puts equal weighton all asset classes, in effect, weighing corporations with assets more than $100 million, same as corporations with smallerasset sizes. Given that smaller corporations are far more numerous, the audit probability averaged across all corporations issignificantly less than the audit probability averaged across the five asset classes.

17

administration translates to an increase in the effective tax rate of just under three percent. Given

that corporate profits were $1.418 trillion in 2010, an increase of three percent in the effective tax

rate translates to an annual increase in corporate tax collections of about $43 billion. Furthermore,

comparing the coefficients on percent of returns audited in Panels A and B, we find that the increase in

effective tax rate and voluntary compliance largely stems from the increase in audit probability under

Democratic administrations, rather than the direct impact of having a Democrat in the White House.

4.4 Summary of results

I summarize the results of the empirical exercise conducted thus far in Table 7.


The tentative conclusion that emerges from the table is that the President is not able to influence

the overall level of resources commanded by the IRS. In contrast, however, the President is able to

exert more subtle influences on the direction of the agency through changes in the share of the agency’s

budget that is dedicated towards detecting tax fraud or the share of IRS employees who are hired as

criminal investigators. These subtle influences also manifest themselves in the fact that corporations

get audited less under Republican administrations.

Congress also emerges as a powerful actor. There is strong and consistent evidence that Demo-

cratic control of the Senate increases the level of budgetary and personnel resources available to the

agency. The influence of the House is mixed. In some respects, its effect on the variables of interest is

similar to the effect of the Presidency in that Democratic control of the House is also associated with

a higher share of the budget dedicated towards detecting tax fraud and a higher audit probability for

corporations.

5 Conclusions

Analyzing the extent of partisan influence on audits of corporate tax returns is central to an under-

standing of the effect of political ideology on tax administration and enforcement. Even though cor-

porate income tax collection averaged only around 23 percent of personal income tax collection over

the period 1978–2010, the amount of revenue obtained through audits of corporate income tax returns

was over half that generated directly through all audits and significantly higher than that generated

through audits of personal income tax returns.

This paper helps provide a framework for understanding why an administration might care about

audit rates in the first place and offers some evidence that Republican administrations are “soft” on

corporations when it comes to corporate audits. Audit probabilities are higher under Democratic ad-

ministrations and corporations respond to this increased audit probability by reporting their tax lia-

18

bility more accurately when Democrats are in charge of the White House. The enhanced probability

of an audit is also reflected in the effective tax rate borne by corporations, which is higher by about

three percent when a Democratic administration is in place. It is remarkable that this increase in the

effective tax rate comes about with no change in the tax code, in particular, without any change in the

statutory tax rate but simply from the greater emphasis on detecting corporate tax avoidance under

Democratic administrations.

In addition to the emphasis on corporate audits and the extended period of analysis, this paper

also provides evidence that relates the variation in audit rates to subtle variations in the allocation of

IRS resources to enforcement-related activity. Although I find little evidence suggesting that the party

affiliation of the President makes a difference to the overall level of IRS funding or the number of IRS

employees, there is consistent evidence across specifications which suggest that when the control of

the Senate switches from Republican to Democratic hands, overall IRS resources goes up by between

6 to 12 percent. The lack of results at the aggregate level for the Presidency masks the fact that there

are significant increases in the share of the IRS budget devoted to detecting tax fraud and the number

of IRS employees devoted to criminal investigation under Democratic administrations. Democratic

Presidents increase the share of the IRS budget dedicated to detecting tax fraud and the share of

criminal investigators hired at the IRS by about 11 percent and 5 percent respectively. The overall

evidence presented paints the picture of an agency which is not entirely immune to political influences.

Finally, although the examination of variation in the corporate audit probability was motivated

by the goal of understanding the effect of political ideology on tax administration and enforcement,

the underlying intuition that motivated the analysis is much more broadly applicable. When faced

with gridlock, politicians are likely, as rational actors, to make use of the least-cost instrument at

their disposal to realize their desired policy goals. If gridlock makes it costly for politicians to use

certain policy instruments, they may attempt to circumvent it with the use of other instruments at

their disposal that do not involve the active cooperation and support of other stakeholders. Future

research should explore the use of such less obvious instruments in other contexts as well.

19

Table 1

Summary statistics for the period 1978–201018

Units Mean StandardDeviation Minimum Maximum

Data on IRS budgets & headcount:IRS budget dedicated todetecting tax fraud

In Millions ofdollars 268 94 121 402

IRS budget In Millions ofdollars 6,736 3,110 1,962 12,353

Federal expenses In Billions ofdollars 1,593 820 459 3,517

IRS budget dedicated todetecting tax fraud / IRSbudget

In percent 5.44 0.65 4.39 6.56

IRS budget / Federalexpenses In percent 0.43 0.047 0.33 0.51

IRS Budget / GDP In percent 0.0892 0.00837 0.0775 0.108Criminal Investigators 2,815 193 2,517 3,348IRS FTEs 99,105 10,565 83,756 117,945Total federal workforce 19 In millions 1.975 0.149 1.737 2.174Total civilian workforce In millions 129.2 16.6 99 154.3Criminal Investigators /IRS FTEs In percent 2.86 0.24 2.37 3.22

IRS FTEs / Federalworkforce In percent 5.1 0.54 4.04 5.78

IRS FTEs/ Civilianworkforce In percent 0.0777 0.0113 0.0592 0.0964

Data on corporate audits:Audit probability (averagedacross all five asset classes) In percent 15.45 20.08 0.239 86.26

Revenue per corporate taxreturn filed (net of refunds) In dollars 58,410 37,205 15,130 163,299

Taxes and penalties peraudit 20 In dollars 402,317 338,085 22,391 1,111,757

Effective Tax Rate In percent 25.0 6.1 12.7 38.9Data on political variables:

Party President 0 = Republican;1 = Democratic 0.394 0.496 0 1

Party in charge of Senate 0 = Republican;1 = Democratic 0.515 0.508 0 1

Party in charge of House 0 = Republican;1 = Democratic 0.636 0.489 0 1

Term of the President 1 = 1st term;2 = 2nd term 1.36 0.489 1 2

18Data on the IRS budget dedicated to detecting tax fraud and data on the total federal workforce are available only for theperiod 1978-2000 and 1981-2010. Thus, summary statistics for measures involving these variables are for the period 1978-2000and 1981-2010 respectively.

19The federal FTE count does not include the number of workers employed by the Postal Service.20It may appear surprising that the average level of recommended taxes and penalties per audit is much higher than the

20

Table 2

Partisan influences on IRS budget (as a share of all federal outlays or GDP) over the period 1978–2010

(1) (2) (3) (4)

Panel A:

Dependent variable: IRS Budget / All federal outlays (As a percent) (X 100)

Party President 1.177 1.183 -0.212 -0.514

(0 = Republican; 1 = Democratic) (0.64) (0.65) (-0.12) (-0.29)

Party in charge of Senate 3.890** 4.296** 3.384* 3.840**

(0 = Republican; 1 = Democratic) (2.38) (2.74) (1.82) (2.56)

Party in charge of House -5.873** -6.144** -5.325* -5.953**

(0 = Republican; 1 = Democratic) (-2.22) (-2.30) (-1.97) (-2.07)

Term of President 1.210 2.053

(1 = 1st term; 2 = 2nd term) (0.85) (1.34)

Constant 45.98*** 44.71*** 39.36*** 37.83***

(12.8) (12.4) (15.7) (13.3)

Number of observations 33 33 33 33

R2 0.23 0.24 0.57 0.60

Panel B:

Dependent variable: IRS Budget / GDP (As a percent) (X 10,000)

Party President -1.735 -1.775 -35.01 -38.33


Party in charge of Senate 55.03* 52.20* 50.26* 55.28**


Party in charge of House 18.29 20.17 2.624 -4.283

(0 = Republican; 1 = Democratic) (0.43) (0.46) (0.065) (-0.097)

Term of President -8.423 22.57

(1 = 1st term; 2 = 2nd term) (-0.32) (0.89)

Constant 870.8*** 879.6*** 840.7*** 823.8***

(15.2) (13.6) (23.5) (21.3)


R2 0.20 0.20 0.68 0.70

Linear Time Trend Y Y N N

Decade Fixed Effects N N Y YRobust t statistics, reported in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

average net revenue collected per return filed. This is because audits rates are significantly higher for the largest corporationsand hence the "average" corporation which is audited is significantly larger than the "average" corporation which files a taxreturn.

21

Table 3

Partisan influences on IRS workforce (as a share of the overall federal workforce or the civilian laborforce) over the period 1981–201021/ 1978–2010

(1) (2) (3) (4)

Panel A:

Dependent variable: IRS Workforce / Federal workforce (As a percent) (X 10)

Party President 1.227 1.259 -0.861 -1.501

(0 = Republican; 1 = Democratic) (0.58) (0.58) (-0.41) (-0.74)

Party in charge of Senate 6.687** 7.263*** 5.254** 5.688***


Party in charge of House -9.215*** -9.601*** -8.300*** -9.100***


Term of President (1 = 1st term; 2 =

2nd term)

1.615 2.899*

(1.39) (1.82)

Constant 54.66*** 53.01*** 50.50*** 48.56***

(14.2) (14.6) (18.4) (17.5)


R2 0.51 0.53 0.69 0.75

Panel B:

Dependent variable: IRS Workforce / Civilian workforce (As a percent) (X 1,000)

Party President -2.008 -2.002 -4.942* -5.184*


Party in charge of Senate 6.922** 7.376** 8.425*** 8.792***


Party in charge of House -1.162 -1.464 -5.295 -5.800


Term of President 1.352 1.650

(1 = 1st term; 2 = 2nd term) (0.52) (0.61)

Constant 89.94*** 88.52*** 67.70*** 66.46***

(17.7) (14.2) (13.4) (12.4)


R2 0.66 0.66 0.82 0.82



21The federal workforce numbers are available from the OMB from 1981 onwards and exclude all federal workers employedby the Postal Service.

22

Table 4

Partisan influences on share of the IRS budget dedicated to detecting tax fraud over the period1978–2000 and the number of criminal investigators (as a share of the total IRS workforce) over the

period 1978–2010

(1) (2) (3) (4)

Panel A:

Dependent variable: Budget dedicated to tax fraud / IRS Budget (As a percent) (X 10)

Party President 5.742*** 5.318*** 7.399*** 5.503***


Party in charge of Senate -7.097*** -7.174*** -6.900*** -5.215***


Party in charge of House 7.926*** 7.276*** 9.839*** 5.416***


Term of President -1.502 -4.423***

(1 = 1st term; 2 = 2nd term) (-1.23) (-4.24)

Constant 56.67*** 59.24*** 45.91*** 53.70***

(18.80) (16.80) (27.70) (25.20)


R2 0.89 0.90 0.90 0.96

Panel B:

Dependent variable: Criminal Investigators / IRS Workforce (As a percent) (X 10)

Party President 1.378* 1.371* 1.586* 1.914**


Party in charge of Senate -1.303 -1.809** -1.308 -1.803**


Party in charge of House 0.527 0.864 0.381 1.064


Term of President -1.503** -2.230***

(1 = 1st term; 2 = 2nd term) (-2.10) (-2.98)

Constant 29.49*** 31.07*** 29.17*** 30.83***

(23.90) (19.80) (21.50) (19.20)


R2 0.24 0.32 0.44 0.60



23

Table 5

Partisan influences on audit probability over the period 1978–2010

(1) (2) (3) (4)

Dependent variable: Dependent variable: Log of

Percent of returns audited (X 10)22 number of returns audited

Party President 10.92* 16.62*** 0.164* 0.214**


Party in charge of Senate -8.317 -16.67 -0.027 -0.0863


Party in charge of House 34.06* 40.90* 0.265** 0.320***


Term of President -35.33** -0.261***

(1 = 1st term; 2 = 2nd term) (-2.00) (-3.04)

Log of number of returns filed 0.499** 0.604***

(2.57) (3.54)

Constant 40.62 71.77 2.101 1.055

(1.10) (1.45) (1.00) (0.56)

Specification used23 RE RE RE RE

Decade Fixed Effects Y Y Y Y

Number of observations 165 165 165 165Robust t statistics, adjusted for clustering by asset class, reported in parentheses * p < 0.10, ** p < 0.05, *** p <0.01

22The dependent variable in columns (1) and (2) of Table 5 have been rescaled so as to avoid having unnecessary zeros in theestimated coefficients. The economic and statistical significance of the results is unchanged if we avoid rescaling the dependentvariable in these tables.

23The choice of whether to go with a random effects (RE) or a fixed effects (FE) specification is made in each case on the basisof the Hausman test conducted. Results are available on request.

24

Table 6

Effect of audit probability on effective tax rate and a measure of voluntary compliance over the period1978–2010

(1) (2) (3) (4) (5) (6)

Dependent variable: Dependent variable:

Effective Tax Rate Measure of voluntary compliance

Panel A: Without controlling for party affiliation of the President and Congress

Percent of returns

audited

2.236*** 3.682*** 191.7*** 109.2*** 169.1*** 130.1***

(3.35) (4.63) (6.47) (5.75) (5.33) (7.25)

Top marginal Tax Rate -0.484** -0.877*** -2.912 -4.458 -4.552 -9.449**

(-2.13) (-4.07) (-0.51) (-1.32) (-0.92) (-2.54)

Constant 37.72*** 38.25*** -116.7 99.35 -10.74 245.1**

(4.60) (5.50) (-0.5) (1.00) (-0.06) (2.10)

R2 0.32 0.48 0.85 0.95 0.81 0.92

Panel B: Controlling for party affiliation of the President and Congress

Percent of returns

audited

2.414** 3.892*** 169.3*** 100.5*** 152.2*** 113.9***

(2.57) (4.36) (5.29) (8.30) (4.16) (5.16)

Top marginal Tax Rate -0.577* -1.255*** 2.743 -5.105* -1.169 -9.624*

(-1.91) (-4.41) -0.41 (-1.72) (-0.22) (-2.01)

Party President 0.526 4.659** 47.47 87.92*** 43.93 85.13***

(0 = Republican;

1 = Democratic)

(0.24) (2.50) (1.34) (3.38) (1.45) (2.84)

Party in charge of

Senate

-1.371 -3.725 62.65 15.14 28.54 -4.344

(0 = Republican;

1 = Democratic)

(-0.52) (-1.70) (1.01) (0.59) (0.61) (-0.16)

Party in charge of House -0.805 4.538 -141.9*** -60.76* -106.5** -50.5

(0 = Republican;

1 = Democratic)

(-0.19) (0.84) (-2.83) (-1.82) (-2.75) (-1.66)

Constant 41.80*** 45.69*** -235.8 91.8 -60.45 243.3

(3.46) (4.35) (-0.93) (0.96) (-0.31) (1.43)

R2 0.34 0.57 0.88 0.98 0.86 0.97

Linear Time Trends Y N Y N Y N

Decade FE N Y N Y N Y

Econometric approach OLS OLS OLS OLS IV IV

Number of observations 33 33 33 33 32 32Robust t statistics reported in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

25

Table 7

Summary of results presented in Tables 2 through 5

Effect of a switch from a Republican to a Democratic

President Senate House

IRS budget

Normalized by all federal expenses No effect ↑ ↓

Normalized by GDP No effect ↑ No effect

IRS personnel (FTEs)

Normalized by overall federal workforce No effect ↑ ↓

Normalized by total civilian workforce No effect/↓ ↑ No effect

Fraud budget, normalized by IRS budget ↑ ↓ ↑

Criminal investigators, normalized by IRS FTEs ↑ No effect /↓ No effect

Audit probability for corporations ↑ No effect ↑↑: Increase; ↓: Decrease

26

Figure 1: IRS budget normalized by federal expenses (1978 - 2010)

27

Figure 2: IRS workforce normalized by federal workforce (1981 - 2010)

28

Figure 3: Budget dedicated to detecting tax fraud normalized by IRS budget (1978 - 2000)

29

Figure 4: Fraction of IRS workforce employed as criminal investigators (1978 - 2010)

30

APPENDIX A

Data sources:

Data on the size of the IRS budget and personnel and its allocation across functions is obtained from

the IRS data books available on the IRS website at: http://www.irs.gov/taxstats/article/0„id=226923,00.html

(Accessed 14th December, 2011).

The numbers for federal outlays and the size of the federal workforce are drawn from the website of the

Office of Management and Budget at: http://www.whitehouse.gov/omb/budget/Historicals/ (Accessed

14th December, 2011).

The corporate income tax rates are available at: http://www.jct.gov/publications.html?func=startdown&id=3719

& http://www.irs.gov/pub/irs-soi/02corate.pdf (Accessed 14th December, 2011).

Data on corporate profits are available at: http://www.bea.gov/iTable/index_nipa.cfm (Accessed 29th

December, 2011). I use Tables 6.16 B, 6.16 C, and 6.16 D to construct a comprehensive time series of

corporate profits over the period 1978–2010.

APPENDIX B

Construction of the data on predicted audit probabilities:

I assume that corporation asset size (denoted by A) is distributed as a Pareto distribution (Axtell,

2001). Thus,

Probability(A ≥ ai) = (ao/ai)α, α > 0 (A1)

and a0 is the minimum asset size for a corporation. I take logs on both sides of the above equation to

obtain:

log(Probability[A ≥ ai]) = α(ln(ao)− ln(ai)) (A2)

The IRS makes available the number of returns filed by corporations in four different asset classes,

for which asset size is known (A fifth asset class is “Assets not reported.”) Using the data on number

of returns filed, we can generate probabilities for three different threshold values (or, ai’s referring to

the notation above): Corporations with assets more than $1 million, those with assets more than $10

million, and those with assets more than $100 million. I then use the audit probabilities available

for each year of the sample (1978–2010) from the IRS data books and estimate regressions along the

lines of (A2) to obtain estimates of α and ao for each year individually. The average value of α and

ao are 0.6652 and $37,146 respectively. While the average values of these variables are not used in

the estimation, they are provided to give a sense of magnitudes. Using these parameter estimates, I

calculate for each asset class and for each year, an asset size which corresponds to the median asset

size for that asset class. For example, for the smallest asset class, “Assets less than $1 million”, I set

up the equation:

Probability(A ≤ $w million) = 1/2 ∗ Probability(A ≤ $1 million) (A3)

Similarly, for the next two asset classes, “Assets between $1 – 10 million” and “Assets between $10

– 100 million”, I set up the equations:

Probability[Aε($1 million, $x million)] = 1/2 ∗ Probability[Aε($1 million, $10 million)] (A4)

&

Probability[Aε($10 million, $y million)] = 1/2 ∗ Probability[Aε($10 million, $100 million)] (A5)

Finally, for the asset class, “Assets more than $100 million,” I set up the equation:

Probability(A ≥ $z million) = 1/2 ∗ Probability(A ≥ $100 million) (A6)

I use the parameter estimates of α and a0 obtained for each individual year and equations along

the lines of (A3)–(A6), to obtain values of w, x, y and, z for each year separately. At the end of this step,

I have data on the IRS reported audit probabilities for each asset class (from the data books) and an

asset size corresponding to that asset class (constructed from the exercise above). I then use a cubic

polynomial for each year separately to fit the audit probabilities reported by the IRS for each asset

class. If I were to graph the average predicted audit probabilities for each year that the two parties

were in charge of the Presidency, we would obtain graphs along the lines of that in Figure A1.

[Figure A1 about here.]

Finally, I calculate audit probabilities at three asset sizes for each year: $1 million, $10 million,

and, $100 million using the cubic polynomial that I obtain in the previous step. It is these predicted au-

dit probabilities that are used in the estimation procedure when I regress audit probabilities on party

affiliation of the President and party affiliation of both chambers of Congress. In the first specification

for which coefficients are reported in the text, I do not control for the term of the President, whereas

in the second specification I do.

32

Figure A1

Variation in predicted audit probability with party affiliation of the President

33

Table A1

Summary of robustness checks on partisan influence on audit probability over the period 1978–2010

(1) (2) (3) (4)

Coefficients below are that for the variable, “Party of the President”

Base specification (from Table 5)10.92* 16.62*** 0.164* 0.214**

(1.86) (2.79) (1.73) (2.03)

RC 1 (a): Using Fixed Effects

specifications

10.92 16.62** 0.175 0.234*

(1.86) (2.79) (1.86) (2.25)

RC 1 (b): Controlling for intensity of

Democratic control

9.305* 13.43** 0.145 0.210*

(1.70) (2.19) (1.44) (1.69)

RC 1 (c): Interacting party of the

President with term of the President

14.81* 0.228

(1.79) (1.57)

RC 2: Introducing additional lags for

number of returns filed

0.131* 0.192*** 0.158* 0.190*

(1.84) (2.75) (1.67) (1.80)

RC 3: Dropping corporations which

belong to the largest asset class

13.46** 16.49** 0.199** 0.253**

(1.97) (2.13) (2.07) (2.27)

RC 4: Controlling for drift across

nominal asset classes

13.03** 16.53*** 0.178* 0.215**

(2.35) (2.76) (1.85) (2.03)

Controlling for term of President N Y N YRobust t statistics reported in parentheses, * p < 0.10, ** p <0.05, *** p <0.01

34

References:

Axtell, Robert L., 2001. “Zipf Distribution of U.S. Firm Sizes.” Science 293(5536), 1818–1820.

Bartlett, Bruce, 2012. “The Tax Gap.” The New York Times Economix blog, January 10.

Dubin, Jeffrey A., 2004. “Criminal Investigation Enforcement Activities and Taxpayer Noncompliance.”

Internal Revenue Service, Washington DC.

Griliches, Z., and J.A. Hausman, 1986. “Errors in variables in panel data.” Journal of Econometrics

31(1), 93–118.

Internal Revenue Service Data Books 1978–2010. Internal Revenue Service, Washington, DC.

Hoopes, Jeffrey, D. Mescall, and J. Pittman, 2012. “Do IRS Audits Deter Corporate Tax Avoidance?”

The Accounting Review 87(5), 1603–1639.

Keith, Robert, 2008. “CRS Report for Congress: Introduction to the Federal Budget Process.” Congres-

sional Research Service, Washington, D.C. Available online: http://opencrs.com/document/98-721/2008-

03-07

Kopczuk, Wojciech, 2006. “Tax simplification and tax compliance: An economic perspective.” In Saw-

icky, Max (ed.) Bridging the Tax Gap: Addressing the Crisis in Tax Administration, 111–143, Economic

Policy Institute, Washington DC.

Olson, Mary, 1995. “Regulatory Agency Discretion among Competing Industries: Inside the FDA.”

Journal of Law, Economics, and Organization 11(2), 379–405.

––– 1996. “Substitution in Regulatory Agencies: FDA Enforcement Alternatives.” Journal of Law,

Economics, and Organization 12(2), 376–407.

Poterba, James, 1994. “State Responses to Fiscal Crises: The Effects of Budgetary Institutions and

Politics.” The Journal of Political Economy 102 (4), 799–821.

Rios, Valentin E. 2003. “Liberals, Conservatives and Your Tax Return: Partisan Politics and the En-

forcement Activities of the IRS.” University of Chicago, Working Paper.

Scholz, John T., and B. Dan Wood. 1998. “Controlling the IRS: Principals, Principles, and Public

Administration.” American Journal of Political Science 42(1), 141–162.

Stock, James H., and Mark W. Watson. 2002. Introduction to Econometrics. 2nd Edition. Addison

Wesley.

Weingast, Barry R., and Moran, Mark J., 1983. “Bureaucratic Discretion or Congressional Control?

Regulatory Policymaking by the Federal Trade Commission.” The Journal of Political Economy 91(5),

765–800.

Wood, B. Dan. 1990. “Does Politics Make a Difference at the EEOC?” American Journal of Political

Science 34(2), 503–530.

––– 1992.“Modeling Federal Implementation as a System.” American Journal of Political Science 36(1),

40–67.

Wood, B. Dan, and Richard Waterman. 1991. “The Dynamics of Political Control of the Bureaucracy.”

35

American Political Science Review 85(3), 801–828.

––– 1993. “The Dynamics of Political-Bureaucratic Adaptation.” American Journal of Political Science

37(2), 497–528.

Wooldridge, Jeffrey M., 2008. “Introductory Econometrics.” Cincinnati, South-Western College Pub.

Young, Marilyn, Michael Reksulak, and, William F. Shughart II. 2001. “The Political Economy of the

IRS.” Economics & Politics 13(2), 201–220.

36

Documents

The Political Economy of Tax Enforcement: A Look at the IRS from 1978-2010