Bulls Eye or Ricochet? Ethnically Targeted Campaign Ads in ...
Transcript of Bulls Eye or Ricochet? Ethnically Targeted Campaign Ads in ...
Bulls Eye or Ricochet? Ethnically Targeted Campaign Ads in the 2008 Election
Matt A. Barreto University of Washington
Victoria M. DeFrancesco Soto Northwestern University [email protected]
Jennifer L. Merolla
Claremont Graduate University [email protected]
Ricardo Ramirez
University of Southern California [email protected]
Author names are presented alphabetically, authorship is equal
Prepared for Presentation at the 2009 Chicago Area Behavioral Workshop
Evanston, IL, May 8, 2009
Introduction
In an effort to court the Latino community, both political parties made a point to
campaign aggressively for Latino “swing” votes through advertisements and get-out-the-vote
(GOTV) drives starting in 2000 (Segal 2006; Martinez-Ruiz-Velasco 2007). The increased
commitment to utilize Spanish-language advertising and direct contact with Latino voters
since the 2000 election1 has yet to be followed by consistent evidence that this targeted
outreach has worked. The two main questions that result are, do targeted ethnic appeals
influence the vote choice of Latinos and what is the effect on non-Latinos?
The rich literature on campaign effects, especially in the form of political
advertisements has almost exclusively focused on non-Latino voters and English-language
materials. Yet the realities of the political campaign have changed markedly over the last
decade. Today, no viable candidate for President or for offices in states with a sizable Latino
population can compete without a “Hispanic Outreach Coordinator” and at least one
Spanish speaking member of their advertisement staff. Gone are the days of Gerald Ford’s
tamale eating incident. Instead, candidates and campaigns today understand they need to be
savvy to win Latino votes. What they many not understand is the means to achieve this goal.
Recently candidates have relied on their Portuguese wife (Kerry), their half-Mexican nephew
(Bush), and their own immigrant story (Schwarzenegger) to try to connect with Latino
voters. The delivery of messages in Spanish or a combination of Spanish and English has
also characterized the recent push for the Latino electorate. However, we still do not know
if these approaches work. Are ethnically targeted campaigns effective in mobilizing Latinos
or is it simply seen as pandering? More specifically, what cues matter for Latinos? How
1 Although Latinos had been active in presidential campaigns prior to 2000, such as the Viva Kennedy clubs in 1960, many scholars point to 2000 as the first election to implement a well organized and sustained Latino outreach and Spanish language advertisement campaign.
2
important is the use of Spanish in the targeted outreach? How important are the
endorsements of co-ethnics for Latino vote choice? If either Spanish language or Latino
endorsements are utilized in political advertisement, what direct or indirect effects are there
among non-Latino voters? Lastly, are these ricochet effects constant across all non-Latino
racial groups?
The purpose of this project is to offer a more nuanced understanding of the effects
of the diversification in targeted campaign strategies that are increasingly making use of
Spanish-language and Latino ethnic appeals by presidential candidates, by answering the
questions noted above. We do this by drawing on the marketing, campaign effects, and
Latino politics literature to advance an account of the effects of language and ethnic cues on
Latino and non-Latino voter preferences, and explore possible mediating effects. More
important, we are able to discern the individual and combined effects of language and
targeted ethnic group appeals. The findings of our field experiment challenge conventional
wisdom about consistency of effects across racial groups and have important implications
for future outreach by candidates and parties. They suggest that candidates and campaigns
must further refine their approach to Latino voters, and at the same time, improve their
ability to micro-target their ads to the intended audience.
The remainder of the paper is divided into five parts. First, we review the extant
literature on campaign effects generally, as well as the emerging literature on mobilization
and persuasion or Latino voters. Second, given that this is the first study to directly measure
individual-level effects of Latino-targeted versus mainstream appeals, we present the
plausible expectations and relevant hypotheses of the measurable effects. We then review
the experimental design and relevant variables that will be considered. The fourth section
presents the results, and the final section concludes with an eye towards the considerations
3
that candidates and campaigns must undertake as they decide whether or not to separately
target Latinos only in Spanish-language media or whether it makes sense to use limited
language and ethnic cues within their general media campaign.
Campaign Effects
Scholars of political behavior have long explored the effects of political
advertisements and messages on voters. With respect to political ads, early work by
Patterson and McClure (1976) focused on the effects that campaign ads had on the voting
preferences of citizens. These scholars found that campaign ads did little to change the
minds of voters, a finding that echoed the more general findings of minimal campaign
effects in the literature (e.g., Campbell et al. 1960; Lazarsfeld, Berelson and Gaudet 1944).
However, more recent work (e.g., Shaw 1999; West 2005) has found that ads have an impact
on evaluations of the candidates and vote choice. Furthermore, many studies have
demonstrated that ads can have an impact on voter learning about the candidates,
perceptions of candidates’ chances of winning, voter turnout, and by altering the standards
by which candidates are judged (e.g., Alvarez 1997; Ansolabehere and Iyengar 1995; Brians
and Wattenberg 1996; Finkel and Geer 1998; Johnston et al. 1992).
Within this general literature, scholars have only explored variation by the tone and
targeting of the message. For example, there is a debate on whether negative or positive
advertisements increase the likelihood of turnout (e.g., Ansolabehere and Iyengar 1995;
Finkel and Geer 1998; Freedman and Goldstein 1999; Geer 2006; Goldstein and Freedman
2002). Others have examined whether certain groups of voters are more receptive to
targeted messages (Clinton and Lapinski, 2004). However, they have not explored the
4
effects of these messages on Latinos, nor on how Latino targeted messages might affect
Latinos and non-Latinos.
Although there are few studies about campaign effects on Latino voters, we are not
the first to take note of the new campaign strategies targeting Latinos. A handful of research
projects and articles focus specifically on Spanish-language and ethnically targeted campaigns
(Shaw et. al. 2001; Michelson 2003; Ramírez 2005; 2007; DeFrancesco Soto and Merolla
2006). From a variety of perspectives, these studies all tend to conclude that ethnic-based
appeals are more successful at mobilizing Latinos to vote. Just as the English-language
studies fall short of providing data for Latinos, the Latino politics research does not
investigate the effects of ethnically targeted ads for non-Latinos. Moreover, the Latino
politics research in this domain has tended to focus exclusively on turnout.
This research is among the first to consider the effects that general and ethnically-
targeted campaign ads have on Latinos, as well as among non-Latinos, and to extend this
research to vote choice. Nuño (2007) finds that ethnic partisan mobilization significantly
affected candidate preference among Latino voters in the 2000 election. However, it is
unclear what the content of the mobilization message was, given that Nuño relies on self-
reported mobilization measures from a telephone survey. In fact, he speculates that Latino
canvassers used different, more effective campaign appeals, making it difficult to compare to
the Anglo canvassers. In contrast, by using a controlled experiment, this study can examine
the effect of the actual campaign ads and messages.
In the past, the two streams of research were conducted in isolation of one another.
General studies examined the effect of “mainstream” campaign appeals on the general
electorate, providing no insight into effect on Latino voters. Similarly, the aforementioned
Latino mobilization research has limited its inquiry into how Latino voters respond to ethnic
5
campaign appeals. We argue that the new demographic realities of the American electorate
make it much harder to neatly segregate Latino and non-Latino voters. Campaign billboards
high above California freeways prominently advertise their candidate in English and in
Spanish for all potential voters to see. Bilingual radio spots run on many popular English
and Spanish-language radio stations hoping to attract bilingual Latino listeners. Spanish-
language or bilingual campaign mailers to zip codes with a high percentage of Latino
households may inadvertently be received by non-Latinos voters. Finally, some campaigns
may experiment with using “Spanglish” in their television commercials to reach out to
Latino voters.
In 1994 when Republican Pete Wilson ran for re-election as Governor of California
his campaign was coordinated with the divisive anti-immigrant Proposition 187. He ran
multiple television ads depicting Latino immigrants as intruders and criminals, running
across the Mexican border with no regard for U.S. laws. The goal was to mobilize
conservative anti-immigrant voters, without regard for Latino voters who were viewed as an
inconsequential segment of the electorate. Twelve years later Republican Arnold
Schwarzenegger ran for re-election as Governor of California and campaigned heavily for
Latino votes, emphasized his own immigration story, and employed a Spanish-
language/bilingual campaign in print, radio and TV, even while Pete Wilson served as an
advisor. Campaign strategies have clearly changed, however due to the complexity of
modern campaigns scholars have not appropriately categorized the types and effects of
campaign appeals when the Latino electorate is in play.
Approaches to Campaigning for Latino Votes
Over time, marketing research began to more carefully craft messages to target
particular audiences. Studies in both marketing research and political science have
6
demonstrated that targeted appeals are more effective than non-targeted appeals among the
targeted population, though the effects vary with respect to the non-targeted audience. For
example, in a controlled experiment, Aaker et. al. (2000) examined the effect of targeted
commercials on the intended audience as well as the “non-target market.” Although their
treatment ads are related to soft drinks and movies, their results are quite instructive. They
find targeted appeals are received favorably with the intended audience and unfavorably by
the non-target market. Similarly, Brumbaugh (2002) compared the effect of receiving ads
which cued themes from the dominant (Anglo) culture and from a subculture (i.e. targeted).
She found members of the dominant and subculture to be receptive to the dominant culture
ads (i.e. mainstream), but that only members of the subculture were receptive to the targeted
ads. The subculture targeted ads were not perceived negatively by the dominant culture, but
rather had no effect. With respect to political science research, Clinton and Lapinski (2004)
found that targeted ads in the 2000 presidential election were more effective than non-
targeted ads in increasing turnout, while Merolla and DeFranceso (2006) found that ads
targeted to Latinos had a stronger effect in increasing the likelihood of turnout. However,
neither study focused on the effects of targeted messages on those from the non-targeted
group.
Most recently, White (2007) compares exposure to racial cues in political messages
about the war in Iraq and welfare among both Whites and Blacks. The basic premise of
White’s research fits well with our expectations, that racial priming has significantly different
effects on the in-group and out-group. Through a controlled experiment, White finds that
both explicit and implicit racial priming activates different racial attitudes among Whites and
Blacks. While overt racial primes result in in-group identification among Blacks, the same
treatments result in out-group resentment among Whites. In a comparison of Latinos and
7
non-Latinos, we might expect the cue of Spanish-language materials to elicit similar
responses.
The studies above inform some of our expectations as we turn to the question of
how political candidates can appeal to the Latino population. We identify four general
approaches to campaigning for Latino voters that candidates might employ. First, they may
rely on their traditional, English campaign ads and simply rely on these general ads for both
their Latino and non-Latino outreach. Second, they may take their traditional campaign ads
and translate them into Spanish or make them bilingual, without changing the content.
Third, they may design a new targeted campaign ad for Latino voters with images of Latino
families, children, or prominent Latino endorsements, while keeping the ads in English.
Fourth, they may translate their targeted ad, with Latino specific content, into Spanish or
make it bilingual. Although there are dozens of variations to campaign ads (i.e. tone,
content), these four general categories capture the substantive differences between each, and
provide a framework for scholars interested in the effects of Latino targeted messages.
Below, we explore a 2 x 2 typology of campaign ads, and the expected effects for
both Latino and non-Latino voters when compared to a control group. We turn first to
potential effects of these different types of appeals among Latinos. In general, drawing from
the literature discussed above, we expect Latino targeted messages to resonate more with
Latino voters (relative to the control), with bilingual targeted messages potentially having the
most noticeable effects. There is some empirical support to this claim with respect to
Latinos. Studies have shown that individuals are more likely to be mobilized by a co-ethnic
(Fraga 1988; Shaw, de la Garza and Lee 2000, Michelson 2003, and Ramírez 2005; 2007) or
given exposure to targeted ads (Merolla and DeFrancesco Soto 2006). As a result, we
specifically hypothesize that ethnically targeted ads will have a positive effect.
8
H1: Targeted messages affect the preferences of Latinos.
H1a: This effect is more pronounced for the most targeted appeal (Latino Bilingual).
Table 1: Typology of Campaign Ads and Expected Effects on Voters
English Spanish/Bilingual
Traditional “Mainstream”
Latino: limited to null effects
Non-Latino: significant effects
Latino: significant effects
Non-Latino: H1: no effect H2: negative effect
H3: positive effect
Ethnically Targeted
Latino: significant effects
Non-Latino: H1: no effect H2: negative effect
H3: positive effect
Latino: significant (and potentially stronger) effects
Non-Latino: H1: no effect H2: negative effect H3: positive effect
Traditional English ads may also have a positive effect on Latino vote choice,
although we expect the effect to be somewhat limited (Brumbaugh 2002).
H2: Non-targeted messages have null to minimal effect on the preferences of Latinos.
In contrast, we expect non-Latino voters to be most persuaded by the Traditional
“Mainstream” English-language campaign ads. These are the ads non-Latino voters are
accustomed to seeing, and in reality, are designed to target non-Latino voters.
H3: Non-targeted messages affect the preferences of non-Latinos.
We are agnostic as to what the effects might be for the targeted appeals. First, if
non-Latinos perceive a message to be targeted to a different group (i.e. in Spanish/bilingual,
Latino endorser), then the message may have null effects (Brumbaugh 2002). Second, non-
Latinos may react negatively to the Latino-targeted ad (Aaker et al. 2000), especially if there
are implicit racial or ethnic cues. This expectation is consistent with work by Valentino
9
1999, Valentino, Hutchings, and White 2002 and Mendelberg 2001 who find that white
voters react negatively to implicit racial campaign images. Third, and equally plausible, is
that non-Latinos may react positively to an ad clearly targeting Latinos and view this as
evidence of a candidate’s appeals to racial inclusion (Fraga and Leal 2004). This plausible
explanation of symbolic outreach efforts, or “ricochet pandering” for minorities, but
specifically Latino voters, has not been empirically tested. Given the lack of definitive
patterns in the literature, we consider all three hypotheses as plausible.
H4a: Targeted messages have a null effect on non-Latinos. H4b: Targeted messages have a positive effect on non-Latinos. H4c: Targeted messages have a negative effect on non-Latinos. Experimental Design
We implemented our study online from September 9th to the 16th with a sample of
registered voters in Los Angeles County who provided their email addresses when they
registered to vote. Subjects were randomly assigned to a control or treatment group. Those
in the treated groups were exposed to either an advertisement from Barack Obama or John
McCain, which was either in English or was Bilingual, and was either a mainstream ad with
an Anglo endorser or a targeted ad with a Latino endorser. The subjects filled out a pre-
survey, watched the advertisement, and then filled out a post-test which contained evaluation
questions about both candidates.
Participants and Design
We obtained the email addresses of about 675,000 registered voters in LA County
who provided this information when they registered to vote. While those who provided
their email addresses closely resembled the full sample of registered voters on almost all
dimensions, there were many more people ages 30 and younger. Furthermore, our goal was
10
to have a large enough sample of Latinos to look at the effects of targeted ads on the target
population. To achieve both objectives, we selected a random stratified sample by age of
75,000 Latino surnames and 75,000 non-Latino surnames to receive our email solicitation.
All 150,000 individuals received an email solicitation in which we asked them to participate
in a study on the 2008 Presidential election.2 We randomized incentives in the email
solicitation as part of another study.3 The total number of subjects recruited was 5343, and
4672 had a computer that was capable of handling the study materials.
With respect to demographic and socioeconomic factors, the average respondent
was 35, graduated from college, and had an income level between 60,000 and 80,000. 50% of
the sample was male; 34% identified as Latino; 51% identified as white; 10% identified as
Asian; 4% as African-American; and, 78% were born in the U.S. With respect to
partisanship, 16% identified as Republican, 55% as Democrat, and the remaining 19% as
Independent. In terms of ideology, our mean respondent was slightly liberal. These sample
characteristics compare reasonably well with the population of registered voters in Los
Angeles County. We have a slight over representation of Democrats and a slight under
representation of Republicans; there are 51% of Democrats and 20% of Republicans in the
voter file. Our sample is also slightly younger than the full voter file sample, even though we
stratified our sampling by age. While we are pleased that the sample closely approximates
the electorate in Los Angeles County, it is not meant to be a representative sample of all
2 Those with a Spanish surname received the invitation in English and in Spanish. 3 The control group was not given any monetary compensation, while the treatment groups had some form of monetary compensation for the first 250 people in each condition to respond to the survey. Three conditions were some form of a $10 dollar gift card to a store, to give to a charity of their choice, or for a choice between the two. Two conditions were a lottery for a 1 in 50 chance of winning a $100 or $250 dollar visa gift card. The response rate by condition varied from a low of 2.47% for those in the control group, which had no monetary compensation, to a high of 3.74% for those who received the 10 gift card for a store.
11
voters in L.A. because we are most interested in the effects of the experiment on registered
voters.
Subjects were randomly assigned to the control group or one of eight political
advertisements by the computer: Obama English Miller; Obama Bilingual Miller; Obama
English Mendoza; Obama Bilingual Mendoza; McCain English Miller; McCain Bilingual
Miller; McCain English Mendoza; and, McCain Bilingual Mendoza. An investigation of
whether subjects were evenly distributed across conditions, for each study, according to
potentially relevant variables (i.e., invitation compensation condition, party identification,
ideology, gender, education, age and income) revealed a significant difference only in the
distribution of gender (p=0.09).4
Procedures
Subjects who wanted to take part in the study clicked on a link to the survey, which
they could take in English or Spanish. Subjects did an audio visual test to make sure that
they could hear and see a test video. If they were not able to see or hear the test video, they
were thanked for being willing to participate and were excused from the study (on-screen).
Among those who were able to watch and hear the test video, they then proceeded to fill out
a brief survey which asked basic demographic information, group identity evaluations, as
well as political predisposition questions. Treated groups were then exposed to one of the
eight advertisements. All groups then filled out a post-treatment survey that contained
questions that would allow for a manipulation check, evaluations of the candidates, and
behavior relevant to the upcoming election.
4 To be certain that slight differences in partisanship did not impact the final results, the data were weighted to party identification across each of the eight treatment cells to match the party identification in the control group, within each racial group. This way we can be certain that the exact same proportion of democrats and republicans are found in each of the eight treatment groups.
12
Treatments As stated previously, subjects in the treated groups received one of eight
advertisements. The advertisements all contained the same script, with variations only with
respect to our key conditions: the candidate being sponsored, the language of the
advertisement, and the presence of Latino content. The language manipulation and
candidate manipulation were straightforward. The same script was used (see Appendix), but
certain phrases were repeated in Spanish and we varied the candidate’s name and picture.
For example, the first line in the script (which is replicated in the appendix) noted: “John
McCain/Barack Obama, he’s a once in a generation leader, un lider unico (added for
bilingual), with a bold agenda.” The first image to appear on the screen was of John McCain
or Barack Obama. We chose images of the candidates that had similar backgrounds; and, in
some instances, we were able to swap just the candidates heads onto the same background.
[table 1 about here]
We varied the presence of Latino content in two ways. First, we had either an Anglo
male, Harold Miller, or a Latino male, Hector Mendoza, endorse the candidate. Second, we
scattered the phrase Latino throughout some sections of the script. For example, the Anglo
endorser was President of the National Advocacy Alliance, while the Latino endorser was
President of the Latino Advocacy Alliance. One limitation to this design is that we can not
say specifically whether it is the endorser or the use of the word Latino throughout the script
that may drive differences for this condition. However, the current structure of the Latino
content more closely mirrors the approach used in Latino targeted ads, so the ad should be
more similar to what voters would see in actual elections, making it higher in external
validity.
13
Manipulation Check The programming of the on-line survey forced treated respondents to watch the
advertisement, so they could not just skip it. However, we wanted to test whether subjects
were paying attention to the political advertisements. One indicator we can use to see if
subjects were paying attention to the ads is whether they recalled the endorser. In the post-
test, we asked respondents to identify the individuals who have endorsed each candidate
from a list of seven names, which included the endorsers in the ads as well as other
individuals who endorsed the candidates during the election, such as Ted Kennedy
endorsing Barack Obama and George Bush endorsing John McCain. If respondents paid
attention to the ad content, then we should find that those who received the Miller condition
(either English or Bilingual) should be more likely to select Miller as endorsing the given
candidate relative to those who did not receive the Miller ad. The same logic would apply
for those in the conditions with the Mendoza endorsement. In figure 1, we present the
proportion of individuals who said Miller or Mendoza endorsed each candidate by those
who received an ad with or without the given endorsement.
[Insert Figure 1 about here]
The first thing that is clear for both candidates is that individuals were far more likely
to say that Miller or Mendoza endorsed the given candidate when they were exposed to a
political advertisement with that endorser. For example, 60% of individuals identified Miller
as endorsing Obama when they saw an ad with him as the endorser and only 1.95% of those
who were not exposed to such an ad said Miller endorsed the candidate. All of these
differences in identifying the endorser for those in and not in the condition are statistically
significant according to difference in proportions tests. Furthermore, the pattern of results
is similar across all of the racial and ethnic groups in our sample. However, the percentage
14
of respondents who correctly identified the endorser they received does appear to vary
across the conditions. A lower percentage of those in the Miller condition were able to
correctly identify Miller as endorsing McCain, 40.56%, or Obama, 60%, relative to those
who were exposed to the Mendoza conditions, where 80.98% correctly identified him as
endorsing Obama and 73.67% correctly identified him as endorsing McCain. These
differences are statistically significant according to difference in proportions tests. Thus, it
appears that recall of the endorser was much higher when the endorser was Mendoza for
both candidates. This pattern of higher recall for Mendoza was similar across all racial and
ethnic groups in the sample.
As another test of whether individuals paid attention to the content of the
advertisements, we asked them to rate (on a seven point scale) how much importance the
candidates placed on four issues with higher values being more important: education, the
economy, health care, and immigration. The first three issues were discussed in the
advertisements, with education receiving the most coverage and health care the least amount
of coverage. Thus, we should find that those exposed to the given candidate ads think the
candidates place more importance on these issues relative to the control group. We did not
discuss the issue of immigration, so we should not observe differences between the control
and the groups exposed to the given candidate ad. We plot the mean importance
respondents thought each candidate placed on the four issues for the control group and
those exposed to the ads of each candidate in Figure 2. For Obama, we see that individuals
exposed to the ad did think he placed more importance on the issues of education and the
economy relative the control group, as expected. However, the differences are not
significant for the issue of health care. Meanwhile, as expected, there are no significant
differences on the issue of immigration. A similar pattern emerges for those individuals
15
exposed to the McCain ad, who think he places more importance on the issues of education
and health care relative to those in the control group, while they perceive no differences on
the issue of immigration. However, they do not perceive any significant differences on the
issue of the economy for McCain. Overall, it seems that respondents linked the content in
the ads to the importance candidates placed on the issues, with the exception of health care
for Obama and the economy for McCain. However, health care was mentioned very briefly
in the ad, and McCain was being criticized for his position on the economy around the time
of our study.
[Insert Figure 2 about here]
Vote Choice
Close to the end of our survey, preceding questions of political sophistication,
education, and income, respondents were asked their vote choice intention. Survey
participants were asked: “If the 2008 election were being held today, and you had to decide,
would you vote for Barack Obama, John McCain, or somebody else?” The overwhelming
majority, 70% indicated that they would cast their ballot for Obama, while 21% indicated
that they would cast their ballot for the Republican candidate. The remaining 9% of the
participants either did not know, 4%, or would cast their ballot for someone else, 5%.
In order to begin to address our hypotheses we examine how vote choice was (or
was not) influenced by the exposure to the different advertisement treatments. In this
analysis we will focus on the main effects of the advertisements. As we will elaborate in the
conclusion, having examined the ad main effects on vote choice and other relevant political
measures we will turn to the incorporation of moderators. However, below we will
concentrate on how the different ads alone influenced both Latinos and non-Latinos.
16
Results for Latino Voters
We first examine the ad effects among the Latino sub-sample, considering whether
the ethnically targeted ads hit their “bulls eye.” With regards to vote choice, we are
interested in whether the targeted ads for each of the candidates, Obama and McCain, had a
positive effect on their respective levels of support. In other words, we expect that Latinos
who were exposed to targeted Obama ads would be more likely to vote for Obama than
those not exposed to an advertisement; we hold the same expectation in reference to
McCain and his targeted ads. Looking first at the vote choice for Obama, we see that in the
control group the percent of Latinos voting for Obama is 66%. Each of the Obama ad cells
sees an increase in the percentage voting for Obama, but only two are statistically significant,
according to a difference of proportions test – both of the bilingual ads. The level of
support for Obama was 75% for both the Bilingual Miller and the Bilingual Mendoza
treatments, an increase of over ten percentage points. The trend among the different groups
indicated that Obama’s ethnically targeted ads that employed Spanish had a positive and
significant mobilizing effect on Latinos, however there was no added effect for the Mendoza
bilingual ad, and the English Mendoza had no statistically significant effect at all. These
results are presented in figure 3A.
[Insert figures 3A & 3B about here]
Turning to McCain we see a much lower baseline level of support, 17% of Latinos
voting for McCain in the control. Once again, we find a similar pattern where all of the
treatments register an increase in support for McCain but only the two bilingual ethnically
targeted McCain ads are statistically significant increases. McCain’s support increases to
22% in the bilingual Miller ad and 25% in the bilingual Mendoza ad (see figure 3B). The
English Mendoza ad did not have a statistically significant effect.
17
Results for White Voters
Among non-Latino Whites a new pattern emerges with both the Obama and the
McCain ads. The generic “mainstream” ads have no statistically significant effect, and the
overtly ethnically targeted ads – bilingual Mendoza do not cause an increase or decrease in
support. However, the two remaining Latino targeted ads seem to backfire completely. The
English ad featuring a Latino endorser, and the bilingual ad featuring an Anglo endorser
both cause a demobilizing effect for White voters – across both candidates. For Obama ads,
his support among Whites starts at 70% in the control and decreases to 64% with both the
English Mendoza and the bilingual Miller treatments. These ads, which we describe as
“mixed message” ads do not resonate with White voters, and seem to go so far as to reduce
their level of support for the candidate. However, it is not outright hostility towards Latinos
driving this, but perhaps the sense of competition, or spillover that the mixed-message ad
creates. The overtly targeted ad has no negative effect, however that ad may be sidestepped
completely by the White voter, easily dismissing, or approving of it as a Latino targeted ad.
The mixed-message ads are not as easy to dismiss. With the control adjusted downwards
about 45 points for McCain, the exact same pattern emerges whereby the two mixed-
message ads lead to reduced support for McCain while the clearly targeted ad does not have
a negative effect. The fact that this pattern is consistent across both Obama and McCain ads
suggests this is not simply mixed partisan cues or expectations coloring the outcome, but
rather a deeper response to the Latino outreach ads that cross over into mainstream ads.
[Insert figures 4A & 4B about here]
18
Results for Black Voters
For African Americans, the election of Barack Obama was truly historic. Blacks,
long the largest and most significant minority group in the United States were
hypermobilized by Obama’s candidacy. At the same time, there was a sense that Obama
would not be able to run as a “Black” candidate, for fear of inciting backlash from White
voters, so expectations were not overly high that Obama would run an aggressive Black
targeted campaign, in particular in television ads. However, Latinos were often discussed as
an important coalition partner, perhaps even a reluctant coalition partner, and thus
considerable attention would need to be paid to Latino outreach. How would African
American voters respond to Obama targeting Latino voters? Our results indicate a backlash
by Black voters to Latino targeted ads. Not surprising, in the control group, Obama
received 98% of the Black vote, and after viewing the four ads, each treatment group did
decrease in support for Obama. Even the “mainstream” English Miller ad saw a reduction
in support for Obama from 98% to 89%, though the drop was not statistically significant.
More telling were the sharp decreases witnessed after Black voters viewed the three Latino
targeted ads. The bilingual Miller ad saw support for Obama decrease to 81%, while the two
ads featuring Mendoza both saw support decrease further, down to 78%. The bilingual
Mendoza ad seemed to have the most negative effect whereby the votes lost by Obama were
transferred directly over to McCain. In contrast, in the other two Latino targeted ads a large
number transferred to the undecided or other category instead of directly to McCain. With
regard to the McCain ads, too few African Americans supported McCain to observe any
meaningful data across the different treatments.
[Insert figure 5 about here]
19
Results for Asian Voters
Finally, we turn to Asian American voters. Asian voters in our dataset reflect the
diversity of the Asian American community in Southern California, with large proportions of
Chinese, Filipino, Korean, and Japanese ancestry. However, sample size considerations
prevent us from being able to drill down into each national origin group, and instead we
examine the data for Asian Americans as a collective minority. Nonetheless, some
interesting results were observed. With respect to the ads for Obama, the two bilingual ads
resulted in an eight point increase in support. Asians in the control group supported Obama
at 56% compared to 64% in each the bilingual Mendoza and the bilingual Miller treatments
(see figure 6). No change is noted in vote support in the English Mendoza or the English
Miller ads. Two possible explanations exist. First, skeptics may argue that Filipinos are
more likely to have an affinity for the Spanish language outreach as descendants of a former
Spanish colony where some Spanish words are common. Further, the name Mendoza is also
a commonly occurring Filipino surname. However this does not appear to be the
explanation for a couple reasons. First, Filipinos are only one part of the larger Asian
sample, approximately 20%, leaving 80% of non-Filipino Asians in the sample. Second, any
confusion over Mendoza should have also registered in the English Mendoza ad, which was
not statistically different from the control. Third, when we pull Filipinos out and estimate
Asian vote choice again, the same results are present. Instead, we think a deeper aspect is at
play related to shared minority status. In particular, Asians and Latinos share their status as
linguistic minorities who both have strong use of their native language. While the ads are
not Asian language outreach, they do represent an effort by the candidate to reach out in a
non-English language to a minority group. Perhaps some Asian voters see this as a future
20
promise of outreach to their own community as the number and influence of their electorate
grows.
[Insert figure 6 about here]
Possible Mediating Effects
The results presented thus far are interesting, but only scratch the surface of what is
happening in terms of ethnically targeted campaign ads. Do all Latinos or Whites respond
the same, or do some mediating independent variables create significant cleavages in how
subgroups respond to the ads? In particular, we are interested in how factors related to race,
racial identity, and immigration may interact with the exposure to the treatment and create
the positive or negative mobilizing effects noted above.
For Latinos, our first cut at the analysis included language preference and ethnic
identity. Language usage is an obvious variable to consider given that the two targeted ads
that worked were both the Spanish/bilingual ads. It may be that Latinos who consume
Spanish media are driving this effect and it is simply a communication barrier, not
necessarily a symbolic mechanism in outreach. To this extent we subdivided the Latino
sample into those who reported usage of Spanish media (TV, Radio, Newspaper or Internet)
and those who reported no such Spanish media usage. For the Spanish dominant group the
bilingual ads performed well, leading to higher support for Obama than either of the two
English ads. However, the effect was even stronger for English dominant Latinos were the
increase over the control for the bilingual ads was over 12 points. Thus, regardless of
language usage, the bilingual ads were popular among Latinos. Likewise, we noted high
support for the two bilingual ads for Latinos across varying degrees of ethnic identity.
Understandably, the level of ethnic awareness and identity is naturally high among Latinos in
21
Los Angeles, however comparing lower and higher ethnic identifiers, similarly high levels of
response to the bilingual ads was found.
Among Whites and Blacks more distinguishable patterns emerged. For Whites,
cutting the sample by their degree of racial identity, we find those who state their racial
identity is important witnessed a statistically significant decrease of 15 points in support for
Obama after viewing the mixed-message Latino ads. However, similar to the full sample,
even this subgroup of Whites did not have a negative reaction to the overtly Latino targeted
ad (see figure 7A). In contrast, Whites who stated their racial identity is not important to
them did not have a negative reaction to the mixed-message ads, in fact none of the
treatments were statistically different than the control among this subgroup.
[Insert figures 7A and 7B about here]
Among African Americans, attitudes towards immigration seem to have the largest
mediating effect on the Latino targeted treatments. When the Black sample is split in half,
those reporting below or above the mean on an immigrant feeling thermometer question,
significant differences emerge (see figures 8A/8B). Black voters who hold lower than
average feelings towards immigrants are driving the negative results for the two bilingual ads.
It seems the introduction of Spanish into the Obama ad creates discontent among those
who are anti-immigrant. Remarkably, the intended vote rate for Obama goes from a perfect
100% in the control to only 50% after viewing the bilingual Mendoza ad, for Black voters
with low opinions of immigrants. Meanwhile, African American who have a positive view of
immigrants show no significant decrease in support for Obama after viewing the bilingual
ads.
[Insert figures 8A and 8B about here]
22
Discussion
The Latino targeted advertisement treatments do not have a uniformly positive or
negative effect across our sample. An understanding of these effects requires the
differentiation of racial and ethnic sub-groups; a non-Latino grouping is too blunt of a
categorization and the examination of Latino targeted effects requires disaggregation not
only based on ethnicity but also race. Non-Latinos is simply too heterogeneous of a
category and based only on this category we would conclude that the Obama targeted ads
had no effect on non-Latinos and that only one of McCain’s targeted ads had a negative
effect.
A different picture emerges once we break down non-Latinos into racial sub-groups.
A clear pattern of the negative effect of ethnically targeted ads emerges. This effect is most
visible among blacks, but is also seen among Whites with mixed-message ads. The ad
treatment that had a consistently statistically negative effect was the bilingual Miller
treatment. The vote shares for both Obama and McCain among whites decreased in moving
from the control to their respective targeted bilingual Miller treatment. Among blacks,
Obama’s bilingual Miller ad also had a negative effect. Based on this analysis we can
tentatively conclude that ethnically targeted ads will not have a positive effect on the non-
targeted group. The expectation should instead be narrowed down to either null or negative
effects.
The bilingual ads had a different effect when we look at Latinos. Obama’s bilingual
ads were both significant, seeing close to a 10 percentage increase in support for Obama.
Among Latinos, McCain’s bilingual treatments also had a small mobilizing effect. Our initial
hypothesis that ethnically targeted ads would have a positive effect on Latinos is partially
supported. Not all of the ethnically targeted ads have a positive effect. In fact, the bilingual
23
ads seem to be the only ones resonating with Latinos, however we did not find any negative
effects with ethnically targeted ads.
It is interesting that the most consistently significant treatment across groups was not
the most ethnically targeted, combining both Spanish and a Latino endorser. Instead the ad
treatment that featured Miller in a bilingual narrative elicited the most consistent and
significant effect. Across groups the language of the ad appears to be the most mobilizing or
demobilizing aspect. Close to half of the Latinos in the sample are English dominant, 30%
are Spanish dominant, and the remainder bilingual. Based only on the main effects we can
initially conclude that the effectiveness of the Bilingual ad was not so much a result of
comprehension facilitation but of targeting based on linguistic symbolism. We posit that the
cultural symbol of Spanish exerts a powerful mobilizing effect among Latinos.
Among non-Latinos the bilingual Mendoza ad (Obama), the most ethnically targeted
ad, only had an effect among African Americans. The lack of a larger backlash effect from
non-Latinos suggests a number of possible explanations. First, an ad that is mixed—either a
combination of Latino content or Spanish—may appear as pandering while an ad that is
both targeted in its language and content may appear as a more genuine multicultural
outreach effort. These are just two of many potential explanations of why different types
and levels of ethnically targeted ad treatments differing effects across groups and by
candidates.
The other consistent effect across groups was that the non-targeted English Miller
ads did not have any effect. A crude take home point for a campaign strategist based on
this analysis would be that trying to kill two birds with one stone is not possible. The
general ads are not focused enough to have an effect on any group, while the ethnically
24
targeted ads can elicit support among Latinos but then this comes at the cost of a negative
backlash among certain non-Latino segments.
Conclusion
In the closing weeks of this election cycle, the Latino mass outreach was capped by a
unique political ad where the Democratic candidate for president, Barack Obama fully
conveyed the message himself and did so in Spanish. Did Obama’s time and effort in
rehearsing his message in Spanish actually pay off? Based on the survey experiment we
conducted it is very possible that the symbolic gesture of using Spanish to communicate his
message potentially had a positive effect in increasing Obama’s support. While McCain did
not narrate a complete Spanish language ad, his use of bilingual and translated ads also likely
had positive effects among the Latino electorate. However, these Spanish language and
bilingual ads if seen by non-Latinos had the potential of hurting of each of the candidates.
Campaigns seek to maximize their effectiveness while minimizing their cost and
effort. As a result of the diversity of the present day electorate one-size-fits all messages are
no longer feasible and candidates must engage in increasing micro-targeting efforts. Latino
targeted messages as seen in the above experiment results and recent studies do in fact have
positive results. Candidates seeking out the support of Latinos can effectively reach this
group through careful targeting but they must also be vigilant that exposure to these ads is
limited in its reach beyond Latinos. A key set of considerations of ethnically targeted ads,
which we did not consider here but will do so in the next iteration of our analysis is that of
individual level racial/ethnic identity. Latinos vary in their level of ethnic group
identification and this variance is likely to moderate the receipt of Latino targeted messages.
At the same time non-Latinos vary in their own racial group identity and feelings of
25
ethnocentrism both of which we must also include.
26
Table 1: Experimental Design
McCain ObamaEnglish/Anglo Harold Miller
English; no Latino appeals Harold Miller
English; no Latino appeals Bilingual/Anglo Harold Miller
Bilingual; Latino appeals Harold Miller
Bilingual; Latino appeals English/Latino Hector Mendoza
English; no Latino appeal Hector Mendoza
English; no Latino appeal Bilingual/Latino Hector Mendoza;
Latino appeals Hector Mendoza; Latino appeals
Control No ad given No ad given
27
Figure 1: Percentage identifying Miller or Mendoza as Endorsing Obama and McCain, by Experimental condition
0
10
20
30
40
50
60
70
80
90
Miller condition Not Miller Condition Mendoza Condition Not Mendoza Condition
Experimental Condition
ObamaMcCain
28
Figure 2: Mean Importance Candidate Place on Issues, by Condition
3
4
5
6
7
Education Economy Health Care Immigration
Obama ControlObama AdMcCain ControlMcCain Ad
29
Figure 3A
Vote for Obama by Treatment - LATINOS
73.5
69.9
66.7
75.275.1
62
64
66
68
70
72
74
76
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl ( Pr=.771 ) ( Pr=.297 ) ( Pr=.020) ( Pr=.051)
Figure 3B
Vote for McCain by Treatment - LATINOS
21.7
24.7
18.317.8
21.1
12
14
16
18
20
22
24
26
JM Bi MillerJM Bi MendozaJM En MillerJM En MendozaControl ( Pr=.272 ) ( Pr=.513 ) ( Pr=.076) ( Pr=.087)
30
Figure 4A
Vote for Obama by Treatment - WHITES
70.1
69.2
64.664.2
70.5
62
64
66
68
70
72
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl ( Pr=.087 ) ( Pr=.519 ) ( Pr=.228) ( Pr=.096)
Figure 4B Vote for McCain by Treatment - WHITES
21.8
26.8
20.4
25.125.4
16
18
20
22
24
26
28
JM Bi MillerJM Bi MendozaJM En MillerJM En MendozaControl ( Pr=.032 ) ( Pr=.256 ) ( Pr=.087) ( Pr=.066)
31
Figure 5
Vote for Obama by Treatment - BLACKS
81.3
77.7
88.8
98.0
78.2
72
76
80
84
88
92
96
100
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl ( Pr=.051 ) ( Pr=.255 ) ( Pr=.068) ( Pr=.089)
Figure 6
Vote for Obama by Treatment - ASIANS
64.164.1
57.5
53.6
56.3
50
52
54
56
58
60
62
64
66
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl ( Pr=.684 ) ( Pr=.116 ) ( Pr=.084) ( Pr=.077)
32
Figure 7A
Vote for Obama by Treatment - WHITES{ Who state their racial identity is important }
71.5
52.2
69.5
53.8
67.2
48
52
56
60
64
68
72
76
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl
Figure 7B
Vote for Obama by Treatment - WHITES{ Who state their racial identity is NOT important }
68.669.669.171.671.8
48
52
56
60
64
68
72
76
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl
33
Figure 8A
Vote for Obama by Treatment - BLACKS{ Who have below average rating of Immigrants }
50.0
88.9100.0
87.5
66.7
40
50
60
70
80
90
100
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl
Figure 8B
Vote for Obama by Treatment - BLACKS{ Who have above average rating of immigrants }
70.0
100.090.9
100.0 100.0
40
50
60
70
80
90
100
BO Bi MillerBO Bi MendozaBO En MillerBO En MendozaControl
34
Appendix: Script Senator Barack Obama/John McCain, he’s a once-in-a-generation leader/UN LIDER UNICO, with a bold agenda. In times like these, we need a leader who is willing to make the right decisions to lead our country back to prosperity. Images: Picture of candidate alone and in crowd That is why Harold Miller/Hector Mendoza, president of the National/Latino Advocacy Alliance, endorses Barack Obama/John McCain for President. Image 3: Picture of endorser alone with title An increasing number of children/Latino children can’t read at their grade level. So every child can succeed, he’ll fight for education reforms hailed as the most sweeping in a decade/ LUCHARA PARA REFORMAS EDUCATIVAS Images: School House, Children in Class Senator Obama/McCain has a history of reaching across the aisle and will do so to tackle the pressing problems that confront our community/Latino community/nuestra comunidad/nuestra comunidad Latina such as soaring health care costs, predatory mortgage lenders, and our dependence on foreign oil. Images: Congress in session, Gas Prices, home foreclosures, oil rig, As President, Senator Obama/McCain, will fight for all the people/PARA TODOS, not the powerful or the special interests. Image 16: Candidate in crowd, Group of people Keeping the best solutions, doing what’s right. HE’S ON OUR SIDE/(EL ESTA A NUESTRO LADO). That is why Harold Miller/Hector Mendoza supports Barack Obama/John McCain for President. Images: Candidate with workers, candidate alone
35
References Aaker, Jennifer, Anne Brumbaugh, and Sonya Grier. 2000. “Nontarget Markets and Viewer
Distinctiveness: The Impact of Target Marketing on Advertising Attitudes.” Journal of Consumer Psychology. 9(3): 127-140.
Ansolabehere, Stephen, Shanto Iyengar, Adam Simon, and Nicholas Valentino. 1994. “Does
Attack Advertising Demobilize the Electorate?” American Political Science Review 88: 829-38.
Ansolabehere, Stephen, and Shanto Iyengar. 1995. Going Negative: How Political Advertisements
Shrink and Polarize the Electorate. New York: The Free Press. Ansolabehere, Stephen, Shanto Iyengar, and Adam Simon. 1999. “Replicating Experiments
Using Aggregate and Survey Data: The Case of Negative Advertising and Turnout.” American Political Science Review 93: 901-909.
Brians, Craig Leonard and Martin P. Wattenberg. 1996. “Campaign Issue Knowledge and
Salience: Comparing Reception from TV Commercials, TV News and Newspapers.” American Journal of Political Science. 40(1):172-193.
Brumbaugh, Anne. 2002. “Source and Nonsource Cues in Advertising and Their Effects on
the Activation of Cultural and Subcultural Knowledge on the Route to Persuasion.” Journal of Consumer Research. 29(Sept): 258-269.
DeFrancesco Soto, Victoria M. and Jennifer L. Merolla. 2006. “Vota por tu Futuro: Partisan
Mobilization of Latino Voters in the 2000 Presidential Election.” Political Behavior 28:285–304
DeFrancesco Soto, Victoria 2006. “Partisanship with Accents: When Does Ethnicity Trump
Partisanship?” Paper presented at American Political Science Association annual conference. Philadelphia, PA.
Forehand, Mark and Rohit Desphandé. 2001. “What We See Makes Us Who We Are:
Priming Ethnic Self-Awareness and Advertising Response.” Journal of Marketing Research. 38(Aug): 336-348.
Fraga, Luis R. and David Leal. 2004. “Playing the ‘Latino Card’: Race, Ethnicity, and
National Party Politics,” Du Bois Review: Social Science Research on Race, v1: 297-317. Fraga, Luis R. 1988. “Domination through democratic means: Nonpartisan slating groups in city electoral politics.” Urban Affairs Quaterly. 23(4): 528-555. Jamieson, Kathleen Hall. 1992. Dirty Politics: Deception, Distraction, and Democracy. New York:
Oxford University Press. Mendelberg, Tali. 2001. The Race Card: Campaign Strategy, Implicit Messages, and the Norm of
Equality. Princeton, NJ: Princeton University Press.
36
Nuño, Stephen. 2007. “Latino Mobilization and Vote Choice in the 2000 Presidential
Election.” American Politics Research. 35 (March): 273-293. Ramírez, Ricardo. 2005. “Giving Voice to Latino Voters: A Field Experiment on the
Effectiveness of a National Non-Partisan Mobilization Effort.” Annals of the American Academy of Political and Social Science. 601:66-84.
Ramírez, Ricardo. 2007. “Segmented Mobilization: Latino Non-Partisan Get Out The Vote
Efforts in the 2000 General Election.” American Politics Research. 35(2):155-175.. Segal, Adam. 2006. Total Spanish-Language TV Spending by Market in the 2004 Presidential
Election. Baltimore: Hispanic Voter Project, Johns Hopkins University. Shaw, Daron, de la Garza, Rodolfo O. Lee, Jongho. 2000. “Examining Latino turnout in
1996: a three-state, validated survey approach (California, Florida, Texas).” American Journal of Political Science. 44:2 338-46.
Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White. 2002. “Cues that Matter:
How Political Ads Prime Racial Attitudes during Campaigns.” American Political Science Review 96: 75-90.
Valentino, Nicholas A. 1999. “Crime News and the Priming of Racial Attitudes During Evaluations of the President.” Public Opinion Quarterly. 63: 293-320. Wattenberg, Martin P. and Craig Leonard Brians. 1999. “Negative Campaign Advertising:
Demobilizer or Mobilizer?” American Political Science Review. 93(4):891-899. White, Ismail K. 2007. “When Race Matters and When It Doesn’t: Racial Group Differences
in Response to Racial Cues.” American Political Science Review. 101(May): 339-354.
37