• Is Private Campaign Finance a Good Thing?

Is Private Campaign Finance a Good Thing?
Estimates of the Potential Informational Bene…ts
November 7, 2005
Abstract
What would happen if the current US campaign …nance system, mostly based
on private donations, were replaced by a public funding scheme of the same mag-
nitude? It has been argued that public funding would deprive voters of useful
information, but this can only be true if private donations are somehow targeted
to ‘
better’ candidates. Using a survey-based dataset about the e¤ectiveness of
state legislators in North Carolina, we ask what voters can learn about the char-
acteristics of a legislator from the amount and pattern of contributions received
during the campaign. The total amount that a candidate receives is a positive,
but weak, predictor of that candidate’ e¤ectiveness. However, the sum of con-
s
tributions below a given threshold ($2,000) is a positive and strong signal of
e¤ectiveness, while the sum of contributions above such threshold is a negative
signal of e¤ectiveness. We also …nd that only contributions from organizations
1
 (rather then individuals, parties, or own money) convey a positive signal. In sum,
our evidence contradicts the informational argument in favor of private funding
when contributions are large or when they come from individuals and parties.
1 Introduction
Campaign …nance regulation is a controversial topic. Supporters of stricter rules usually
argue that the mixing of money and politics may cause a number of negative outcomes:
politicians spend too much time fundraising, they modify their policy stance in order to
attract donations, or they give preferential treatement to donors, thus skewing political
outcomes in favor of the wealthy and the organized. Another charge is that campaign
…nance exacerbates the incumbency advantage, which in turn makes elected leaders
less accountable to voters.1
Instead, supporters of weak or no regulation usually make use of three lines of
argument: a pragmatic one (it is better to have legal campaign contributions rather
than illegal ones), a constitutional one (limits on campaign …nance constitute a vio-
lation of the First Amendment on free speech),2 and an informational one (campaign
…nance is an e¤ective instrument to provide voters with useful political information).
A forceful summary of the informational argument is provided by Bradley A. Smith, a
commissioner with the Federal Election Commission:3
1
See Levitt [9] for a critical survey of campaign …nance reform proposals.
2
In Buckley v. Valeo, the U.S. Supreme Court distinguished between campaign contributions
and campaign expenditures, ruling that limits on the latter do violate free speech and are therefore
unconstitutional, while limits on the former are constitutional.
3
Congressional testimony, February 27, 1997, House Subcommittee on the Constitution. Available
2
 “[...] the fact of the matter is that, more than ever in American society,
communicating in the political realm requires the expenditure of money.
Money is not an evil in politics – it is a source of information to voters.
E¤orts to regulate the ‡ of money in politics over the past 20 years have
ow
done much more than money ever did to distort the political system and
create a public distrust of government. It is now time to try a new approach
–that is, it is time to deregulate politics. There is simply no a priori method
to say what is fair or not fair –how much groups should be able to spend,
or what kind of advocacy they can spend it on. The bureaucracy that
has been established to regulate politics is sti‡ grassroots advocacy and
ing
political communication.”
The informational argument in favor of campaign …nance is based upon two logical
steps. First, as Bradley states, money must be “a source of information to voters.”
But this is not su¢ cient, because private contributions could be replaced by a system
of public funding in which candidates received the same amounts, subject to the same
restrictions on spending (if any), that they would have received under the private
system. This shift can potentially keep the source of information to voters but it would
avoid the alleged negative side e¤ects highlighted above. Many countries, especially
in Europe, rely on generous public campaign …nancing. In the U.S. public …nancing
plays a large role in presidential campaigns and in some states (e.g. Minnesota and
from: http://www.cato.org/testimony/ct-bs022797.html. Bradley joined the FEC in 2000.
3
New Jersey).4
A second logical step is therefore necessary to complete the informational argument:
a given amount of private contributions must be more e¤ective than the same amount
of public funding in conveying information to voters. In a public system, the distri-
bution of campaign funds among candidates could only be based on pre-determined
criteria. Instead, with private contributions, the allocation of campaign money is at
the discretion of donors. If donors are more likely to give money to high-quality can-
didates –whatever the de…nition of quality may be –then a private campaign …nance
system has a potential informational advantage over a public one. If, however, private
contributions do not tend to go to better candidates, then shifting to a public funding
system that provides the same overall amount of money would not deprive voters of
useful information.
Hence, a key question to evaluate the strength of the informational argument is:
What does the amount of campaign money that a candidate receives say about that
candidate’ characteristics? In particular, do “better”candidates receive more money?
s
One can also specialize the question to di¤erent classes of campaign funds. For regula-
tory purposes, it would be particularly interesting to know whether the informational
value of campaign money varies according to the size of the donation made and to the
characteristics of the donor.
4
Obviously, replacing private funds with public funds would impose an additional tax burden.
However, the overall amount of campaign spending in the U.S. is extremely low relative to the size of
government (Ansolabehere, de Figuereido, and Snyder [1]).
In the North Carolina dataset we use, replacing all private contributions in a two-year electoral cycle
would cost less than 0.01% of the aggregate income of North Carolina. See footnote 19 for details.
4
 To the best of our knowledge, this set of empirical questions has not yet been
addressed. The vast empirical literature on campaign …nance has not attempted to
estimate the informational bene…t of campaign …nance. It has mostly focused on the
policy distortions caused by campaign …nance (is there a link between private contribu-
tions and policy making?) and on the electoral e¤ectiveness of campaign money (are
candidates more likely to win if they receive more money?). Our empirical exercise
is inspired by the theoretical literature on campaign …nance with rational voters (e.g.
Potters et al. [11], Prat [13], [14], Coate [4], [5], Gerber [6] and Ashworth [3]). This
body of work identi…es a trade-o¤ between a policy distortion and an informational
bene…t, both due to the presence of private campaign …nance. In equilibrium, high-
quality candidates receive more contributions than low-quality candidates, and this
money is used to provide voters with information about candidates’quality. However,
candidates may need to distort their policy choices (away from voters’ interests) in
order to attract private donations. The overall welfare e¤ect of allowing private do-
nations depends on the relative magnitudes of the policy cost and the informational
bene…t. The goal of our paper is to provide a …rst assessment of the positive side of
this trade-o¤.5
We use a dataset collected by the North Carolina Center for Public Policy Research
(NC Center), which attempts to measure the “e¤ectiveness” of North Carolina state
legislators. The Center asks about 500 legislators, lobbyists, and journalists to assess
5
In the conclusion, we will return in more detail to the connection between our …ndings and the
existing literature on campaign …nance.
5
the e¤ectiveness of each legislator in the current legislative session. Every respondent
is asked to provide a ranking of all legislators. Our data covers six electoral cycles,
from 1990 to 2000. The North Carolina e¤ectiveness score can be taken to represent
the consensus view of political insiders on legislators’quality. We also have information
on campaign contributions for every North Carolina legislator.
We put ourselves in the shoes of North Carolina voters who are trying to evaluate
the e¤ectiveness of their representatives, which we take to be proxied by the NC Center
ranking. We suppose that voters do not observe the ranking directly, but they may have
other kinds of information about incumbents. We consider two classes of voters and
four campaign …nance information scenarios. The two classes of voters di¤er according
to their knowledge of biographical and professional information about the incumbent
candidates. Uninformed voters are only aware of easily observable characteristics of
representatives, i.e. their gender and their party a¢ liation, while informed voters, on
top of gender and party a¢ liation, also know age, race, profession, tenure of their
representatives, and whether the representative belongs to the majority party and/or
holds key o¢ ces in the legislature. The campaign …nance information scenarios are:
(1) the voter has no information about campaign contributions; (2) the voter knows
the total amount of contributions that each representative receives; (3) the voter also
knows what amount comes from small contributions (less than $2,000) or large (over
$2,000); (4) the voter knows the source of contributions received (candidate’ own
s
money, party money, funds from organizations, funds from individuals).
We analyze every combination of voter types and campaign …nance information
6
scenarios, and we obtain four main results. First, the total amount of contributions
that a candidate receives is a useful predictor of the candidate’ e¤ectiveness ranking
s
both for the Senate and the House, but only in the eyes of uninformed voters. How-
ever, this e¤ect is small. To increase her ranking (in the eyes of uninformed voters) by
one percentage point, a House representative must spend an additional $16,000 (the
median amount of contributions received by a House representative is just $24,600).
The equivalent …gure is around $32,000 for a senator (median amount of contributions:
$37,500). On the other hand, in the case of informed voters, the correlation between ef-
fectiveness and total contributions received, albeit positive, is not statistically di¤erent
from zero.
Second, the ability to disaggregate between small and large contributions (scenario
3) is particularly useful, for both chambers and for both classes of voters. The informa-
tional bene…t of moving from scenario 2 to 3 tends to be greater than the informational
bene…t of moving from scenario 1 to 2. Small contributions are a strong positive pre-
dictor of e¤ectiveness: House representatives only need an additional $5,500 in small
money to increase (again in the eyes of uninformed voters) their ranking of one per-
centage point. Senators only need $6,100. Instead, the striking result is that large
contributions are negatively related to candidate’ e¤ectiveness. This is true for both
s
chambers and for all classes of voters. Only in the case of an uninformed voter trying
to evaluate the e¤ectiveness of his House representative, is the correlation with large
contributions still negative, but not signi…cantly di¤erent from zero.
Third, it is extremely useful to disaggregate total contributions according to their
7
source (scenario 4). In particular, contributions from organizations (both pro…t and
nonpro…t ones) are strongly and positively correlated with legislator e¤ectiveness. This
holds for both chambers and for all classes of voters. In the eyes of uninformed voters,
House representatives only need an additional $2,400 in funds from organizations in
order to increase their ranking by one percentage point. The corresponding …gure for
Senators amounts to $3,400. On the other hand, the e¤ectiveness of a representative in
the House is negatively related to the amount of donations from party sources and from
private individuals, with the former having a larger e¤ect than the latter. In general,
these negative correlations are not statistically signi…cant in the case of Senators.
Fourth, the explanatory power of campaign …nance information is lower than the
explanatory power of the other forms of voter information that we consider. The
informational value of passing from the uninformed voter case to the informed voter case
(i.e. learning about the candidate’ race, age, tenure, profession, key o¢ ces held, and
s
whether she belongs to the majority party) appears to be larger than any informational
bene…t that can be gained through campaign …nance information.
The results about the size and the source of campaign funds are clari…ed by par-
titioning total contributions along these two classi…cation grids at the same time, i.e.
by looking at small and large contributions from the four di¤erent sources mentioned
above. It turns out that small contributions from organizations are strongly and pos-
itively correlated with legislator e¤ectiveness for both chambers and both classes of
voters, while large contributions from organizations are positively and signi…cantly
correlated with legislator e¤ectiveness only in the eyes of uninformed voters in the
8
House. On the other hand, both small and large contributions from political parties
are signi…cantly and negatively correlated with e¤ectiveness in the case of the House,
while for the Senate this holds only for large contributions. Finally, for both cham-
bers, only large contributions from private individuals are signi…cantly and negatively
correlated with legislator e¤ectiveness.
Our empirical exercise on the informativeness of campaign …nance is based on the
likely unrealistic assumptions that voters are able to: (1) observe the amounts of con-
tributions that all incumbent candidates receive, and in some cases the actual sources
of these contributions; and (2) use that information e¢ ciently to make inferences about
candidates’e¤ectiveness. This assumption stacks the deck in favor of an informational
role for private campaign …nance. If voters only observed imperfect signals of contribu-
tions (such as the number of TV ads they see) or if they were boundedly rational in the
way they process this information, then we should expect the amount of information
generated by private campaign spending to be lower than under our assumptions. In
this sense, our estimates should be seen as upper bounds on the informational value of
private campaign …nance.6
The main overall lesson of our exercise is that, at least according to our evidence,
certain types of private funding provide no informational bene…t (and perhaps an
informational cost). This is true for large contributions and for contributions from
organizations, individuals, and own money. For these types of funds, there appears to
6
Of course, one should also keep in mind that our results may underestimate the informational
bene…t because our measure of e¤ectiveness may contain noise and because voters may value other
aspects of a candidate’ personality, such as morality or ideology.
s
9
be no trade-o¤ between informational bene…t and policy cost. It is hard to see why
one would not want to replace these funds with public funds of the same amount.
The paper is organized as follows. Section 2 discusses our data sources and Section
3 reports the results of our analysis. Section 4 relates our …ndings to the existing
literature and concludes
2 Data
2.1 Some institutional background
The North Carolina legislature (the General Assembly) consists of two chambers: a
House of Representatives with 120 members and a Senate with 50 members. All mem-
bers are elected every two years, for two-year terms. The General Assembly is typically
described a hybrid -a partially professional, partially amateur legislature7 . In 2001 each
member received a salary of $13,951 plus a $104 per diem for living expenses. Leg-
islative leaders earned substantially more -e.g., the Speaker of the House was paid a
salary of $38,151 together with an expense allowance of $16,956.
Regular legislative sessions are biennial and convene in January following each elec-
tion. The Democratic Party dominated the North Carolina General Assembly until
recently. During the 1990-1999 period Democrats held 61% of all state legislative
seats. However, in 1994 Republicans won control of the state House for the …rst time
7
In fact, in 1986-88 the North Carolina legislature was ranked 22nd by Squire’ [15] index of
s
legislative professionalism.
10
in 100 years. They won again in 1996, but then lost control in 1998.
Internally, the legislature is mainly organized along party lines. The majority party
controls all committee chairs8 , but some vice-chairs and subcommittee chairs are left
to the minority. Electorally, party organizations in North Carolina are stronger than in
most other southern states, but still rank just below the U.S. average (see, e.g., Cotter,
et al., 1984). Finally, Morehouse (1981) classi…ed North Carolina as a state in which
pressure groups are strong.
2.2 Data on Legislator E¤ectiveness
As mentioned in the introduction, we take the data on legislator e¤ectiveness from the
North Carolina Center for Public Policy Research (NC Center), an independent non-
pro…t organization. At the end of each regular legislative session, the NC Center asks
state legislators, lobbyists and legislative liaisons, and capital news correspondents to
rate each member of the General Assembly according to their “e¤ectiveness” The NC
.
Center has continuously conducted this survey since 1977. The sample of insiders be-
ing interviewed comprises all 170 legislators, all lobbyists registered in the state capital
and residing in North Carolina (250-325 lobbyists), and all journalists who regularly
cover the General Assembly (35-45 journalists): this amounts to a total sample size of
475-550. Every two years The NC Center publishes a handbook (Article II: A Guide to
the N.C. Legislature), which contains the ranking of legislators, as derived from these
8
Apart from the question on legislator e¤ectiveness, in its survey the NC Center asks respondents
to name the …ve or six “most powerful” committees in both houses. Such list almost always included
Appropriations, Finance, Judiciary I, Rules, and Education (the latter one as of 1989).
11
ratings.
Here follows a quote from the handbook, which well explains how -according to the
NC Center- respondents should interpret the e¤ectiveness concept:
“[...] Ratings were to be based on their participation in committee work,
their skill at guiding bills through ‡ debate, their general knowledge and
oor
expertise in special …elds, the respect they command from their peers, the
enthusiasm with which they execute various legislative responsibilities, the
political power they hold (either by virtue of o¢ ce, longevity, or personal
attributes), their ability to sway the opinion of fellow legislators, and their
aptitude for the overall legislative process.” (From Article II: A Guide to
the 1991-1992 N.C. Legislature, p. 212.)
Similarly to what done by Padró i Miquel and Snyder [10], we invert the original
ranking measure, so that higher values of our dependent variable correspond to greater
e¤ectiveness. Also, in order to make results more easily comparable across the two
chambers, we normalise our e¤ectiveness measure on a common 0-100 scale.
2.3 Data on campaign …nance
We match the e¤ectiveness measure with data on campaign …nance contributions re-
ceived by legislators running for reelection to the General Assembly. Such data comes
from the North Carolina State Board of Elections9 . Here, detailed pieces of information
9
See the URL: http://www.sboe.state.nc.us/, under the “data and statistics” heading.
12
regarding contributions received by candidates running for the General Assembly are
available in electronic format, beginning with the 1990 elections.
We aggregate this data to create a set of campaign …nance …gures for each candidate
in each election year. All …gures are in real terms, i.e. they are translated into 2000
dollars. First, we calculate the total amount of contributions received by candidate
j in election t. To provide a cleaner measure of this, which should be independent
of the timing of received contributions, we exclude all bank transactions from the
computation, in particular interest rate payments.
Second, we calculate the total amounts of contributions which are respectively above
and below a …xed threshold of $2,00010 . When doing this, we take into account the
fact that some contributors may in fact donate in separate installments: if the total
amount donated to candidate j by contributor i for election t is above the threshold,
our procedure classi…es this as a large contribution, even if each installment, separately
considered, is below the $2,000 threshold.
Third, we distinguish total campaign …nance contributions according to their source.
We consider four categories of donors: the candidate herself, her political party, orga-
nizations (both pro…t and nonpro…t ones) and individuals.
Table 1 presents some summary statistics about this campaign …nance data. We
separately show …gures about the Lower House and the Senate. As we are matching
campaign …nance data with information on legislator e¤ectiveness, the tables refer to
incumbent candidates, i.e. those for which an e¤ectiveness score is available. In all
10
Varying this threshold ($1,500, $2,500 and $3,000) indeed produces qualitatively similar results.
13
cases (total contributions, small and large contributions, total contributions classi…ed
according to the source) and for both chambers the distribution of funds received by
candidates is positively skewed, as witnessed by the fact that for each category the
average contribution is larger than the median one. The median amount of total con-
tributions to an incumbent candidate in the House is about $24,600, while the average
one is about $47,000. The median sum of contributions received by an incumbent can-
didate in the Senate is slightly below $37,500, with the average one almost reaching
$77,000.
2.4 Other data
When constructing the information set enjoyed by di¤erent class of voters, we make
use of data on gender, age, race, party a¢ liation, tenure of incumbent legislators, and
key o¢ ces held by them. These pieces of information are taken from the NC Center’
s
Article II guides and various editions of the North Carolina Manual. In particular, the
tenure variable is de…ned as the number of terms a given representative has continuously
served in the House or Senate. The count restarts in case a representative switches
from the Lower House to the Senate (or vice versa).
We use an ordinal variable to measure the relevance of the posts held by a candidate
in the current legislature. This variable takes on a value of …ve if the representative
is Speaker or Speaker pro tempore, a value of four if she is President pro tempore or
Majority Leader, three if she is Minority Leader, two if she is Majority or Minority
14
Whip, one if she is chair of some top committee, and zero otherwise.
3 Results
We analyze both chambers of the North Carolina State Assembly. The basic speci…ca-
tion is as follows:
0 0
Rjt = cjt + xjt + j + jt
where Rjt is the normalised e¤ectiveness ranking given to the incumbent candidate j
at time t (regarding her performance during the past legislature), xjt is a vector of
candidate characteristics that possibly vary across time, j is an individual e¤ect and
jt is an idiosyncratic error. The focus of our analysis is on cjt , which is a vector
containing information on contributions received by candidate j during the electoral
campaign taking place at time t. We consider four di¤erent scenarios regarding the
campaign …nance information available to voters:
1. No information about campaign contributions;
2. Total amount of contributions received by each representative;
3. Total amount of small contributions (less than $2,000) and total amount of large
contributions (at least $2,000);
4. Total amount of contributions disagreggated by source (own money, party funds,
funds from organizations, all other contributions).
15
 The exact form of the cjt vector varies across these scenarios.
Our results, based on a random e¤ects speci…cation, are shown in Tables 2 and 3.
Table 2 corresponds to the case of an uninformed voter - i.e. a voter who observes
only gender and party a¢ liation of representatives. Table 3 is for the case of an
informed voter - i.e. one who observes what the uninformed type knows, plus age,
race, profession11 , tenure, key o¢ ces held in the past legislature, and whether the
representative’ party had the majority. In both tables we present estimates for both
s
the Lower House and the Senate.
Rather than discussing each regression individually, it is more instructive to com-
pare the role of the same variables across treatments. First, consider the coe¢ cient
on total campaign contributions in Scenario 2. For both chambers and both classes
of voters, the estimated coe¢ cient is positive. However, its strength decreases when
voters become more informed. It is also lower for the Senate. For informed voters in
both chambers, the coe¢ cient is not signi…cantly di¤erent from zero.
The estimates imply that an increase of $1,000 in the campaign chest of a House
representative would lead uninformed voters to predict an increase of 0.062 percentage
points in her ranking. Put di¤erently, a House representative who wishes to go up
one percentage point in the ranking of uninformed voters must …nd an additional
$1;000
0:062
= $16; 100. To put this in perspective, the median amount of contributions
received by a House representative is $24,600. The equivalent …gure for a senator is
11
In practice, it turns out that the key distinction is whether the representative is a lawyer or not.
We thus simplify the exposition by replacing profession with a dummy variable for lawyers.
16
$32,250 (and the median contribution amount is $37,500).
When we move to scenario 3, the coe¢ cient on small money is signi…cant at the
1% level for both classes of voters and for both chambers. As before, its estimate is
higher for uninformed voters and for the House. Now, money is much more informative
than in scenario 2. A House representative who wishes to go up a percentage point in
the ranking of uninformed voters must …nd only an additional $5,500 in small money
(the median amount of small money received by a House representative is $21,000).
The equivalent …gure for a senator is $6,100 (median amount of small money: $32,000).
Even in the case of informed voters is the magnitude of the correlation sizeable: in order
to increase her e¤ectiveness ranking by one percentage point, a House representative
needs an additional $9,200 in small money, while a Senator would need almost the
same amount, i.e. $9,500.
Instead, the coe¢ cient on large money is negative (and signi…cant for all cases
except uninformed voters in the House). The point estimate of the amount of large
money needed to make the ranking of a representative go down of a percentage point
ranges from $12,000 (for an uninformed voter in the Senate) to $23,800 (for an informed
voter in the House).
In scenario 4, there is a strong and positive correlation between the ranking measure
and the amount of funds from organizations. Such correlation is statistically signi…cant
at the 1% level for all class of voters and both chambers, and the estimated coe¢ cient
is large in magnitude. If one focuses on uninformed voters, it is su¢ cient for a House
representative to raise an additional $2,400 in funds from organizations in order to
17
increase her ranking by one percentage point (the median amount of funds from orga-
nizations received by House representatives is $13,000). The corresponding …gure for a
Senator is $3,400 (median amount of organizations’money: $19,800). Even in the case
of an informed voter is the correlation estimated to be large: the necessary increase in
organizations’funds to obtain a percentage point increase in the e¤ectiveness ranking
amounts to $5,700 and $6,000, in the House and in the Senate respectively.
Party funds are signi…cantly and negatively correlated with legislator e¤ectiveness
for both classes of voters in the Lower House, while such correlation is mildly sta-
tistically signi…cant for uninformed voters in the Senate. The point estimate of the
amount of party funds needed to make the ranking of a House representative go down
of a percentage point ranges from $3,000 (uninformed voters) to $5,600 (informed
ones). Similarly, contributions from individuals are signi…cantly and negatively cor-
related with legislator e¤ectiveness for both classes of voters in the House, while the
correlation, albeit negative, is never statistically signi…cant in the case of Senators.
The size of the correlation is smaller than the one found for party money: in the eyes
of uninformed voters, the amount of private individuals’money needed to decrease the
ranking of a House representative by one percentage point amounts to $19,20012 .
Finally, the coe¢ cients on own funds are not signi…cantly di¤erent from zero in any
of the four cases being studied.
12
As more thoroughly discussed in section 3.1, it turns out that are large (i.e. above $2,000) contri-
butions from parties and private individuals signi…cantly and negatively correlated with e¤ectiveness
for both chambers and both classes of voters.
18
 A clear pattern has emerged. Total contributions have a mild positive relation
with e¤ectiveness, which is the composition of two e¤ects with opposite signs. Small
contributions have a strong positive e¤ect, while large contributions have a negative
e¤ect. The mild positive relation of total contributions with e¤ectiveness can be also
analysed by looking at the partition of funds according to the source: funds from
organizations are strongly and positively correlated with the e¤ectiveness measure,
while funds from political parties and private individuals are negatively correlated,
signi…cantly so for the House.
How much information on e¤ectiveness do voters gain when they have campaign
…nance information? Table 4 reports 16 R-squared values (the …gures written in large,
regular characters) corresponding to the two classes of voters and four campaign …nance
information scenarios for the House and for the Senate. Obviously, the (unadjusted)
R-square coe¢ cient increases when one adds information, either by looking at a more
informed class of voters within the same scenario or by examining a more informative
scenario for the same class of voters. It must be however noted that scenario 3 is not
nested within scenario 4; on the other hand, scenario 2 is nested within both scenario
3 and 4 (which correspond to alternative partitions of total contributions received by
candidates).
Every shift downwards or rightwards in Table 5 corresponds to adding one or more
right-hand variables to the previous regression, again with the exception of the move
from scenario 3 to scenario 4. Hence, it is possible to perform an F-test between
each pair of neighboring cells, and between scenario 2 and scenario 4. The …gures
19
written in small italic font correspond to P-values for the F-statistics computed for
a pair of neighboring cells. For example, look at informed voters in the Senate and
compare scenario 2 (R2 = 52:49%) and scenario 3 (R2 = 55:29%). The P-value of the
F-statistics is 0.001. Regarding the P-values on the left of scenario 4, they correspond
to the F-tests between scenario 2 and 4.
Table 4 highlights several patterns. Firstly, moving between the two classes of
voters is, in general, more useful from an informational viewpoint than moving between
campaign …nance information scenarios. All 8 F-statistics for vertical comparisons are
highly signi…cant, while some of the 12 F-statistics for horizontal comparisons are not.
The most powerful informational treatment appears to be the shift from uninformed
to informed voters. This is particularly striking in the House, where in scenario 1, the
R-squared for uninformed voters is 3:57% while that for informed voters is 43:53%.
Campaign …nance information can make up for only a small fraction of this gap: the
R-squared for uninformed voters in scenario 4 is 20:89%. The patterns are similar in
the Senate, but less pronouced. These …ndings put an upper bound to the importance
of campaign …nance as an informational channel.
When we focus on horizontal comparisons, we notice that the di¤erence between
scenarios 2 and 3 is always signi…cant at the 1% level. The di¤erence between scenarios
2 and 4 is always statistically signi…cant, at the 1% level for both classes of voters in
the House and uninformed ones in the Senate; it is signi…cant at 5% level for informed
voters in the Senate. Instead, the di¤erence between scenarios 1 and 2 is not signi…cant
for informed voters in both chambers.
20
 To sum up, the overall picture emerging from Table 4 is: (1) campaign …nance
information is useful, but less so than other (basic) types of information; and (2) the
most informative part of campaign …nance is the disaggregation of total contributions
according to the size and the source.
Tables 2, 3 and 4 are based upon a random e¤ects speci…cation. Tables 6 and 7 show
results for uninformed and informed voters respectively (i.e. they parallel Tables 2 and
3), but adopt a …xed e¤ects design. The signs and signi…cance levels of coe¢ cients are
broadly similar across the two designs.13 In addition, we examined the purely cross-
sectional variation in the data using the “between”estimator, and …nd similar patterns
but with even larger coe¢ cients. In the interest of space we do not report these results,
but they are available upon request.
3.1 Size and source of contributions compared
As discussed above, the partition of total contributions according to size and source
gives voters a valuable amount of information about legislator e¤ectiveness. Two re-
lated questions stem from this …nding:
1. What is the relationship between the results about the size of contributions (sce-
nario 3) and the ones about the source thereof (scenario 4)?
2. Which of the two partitions provides voters with “more” information about the
unobserved ranking of legislators?
13
The only notable exception occurs with organizations’money for informed voters in the Senate,
which is no longer statistically signi…cant with a …xed e¤ects speci…cation.
21
 In order to answer these questions, we classify total contributions according to the
size and the source at the same time (i.e. we obtain eight campaign contributions
…gures for each incumbent candidate in each year). Again considering both classes
of voters and both chambers, we re-run the regressions with this new scenario about
campaign …nance information. Results are presented in Table 5, where each column
corresponds to a di¤erent class of voters in the Lower House and in the Senate.
A consistent pattern emerges here, which helps explain the results regarding sce-
nario 3 and scenario 4. Firstly, small contributions from organizations are strongly
and positively correlated with legislator e¤ectiveness. For both classes of voters in the
House and for uninformed ones in the Senate the coe¢ cients are signi…cantly di¤erent
from zero at the 1% level, and the size of the e¤ect is remarkably large. An incumbent
candidate in the House, who wants to increase her ranking by one percentage point in
the eyes of uninformed voters, needs only an additional $1,800 of small contributions
from organizations. An additional $3,800 is needed to achieve the same one percent-
age point increase in the ranking if one considers the perception of informed voters.
Regarding the Senate, the needed increase in small money from organizations ranges
from $3,100 (uninformed voters) to about $7,500 (informed ones). In fact, large contri-
butions from organizations are signi…cantly (and positively) correlated with legislator
e¤ectiveness -at the 10% level- only for uninformed voters in the House.