The Hispanic Immigrant Voter and the Classic American Voter: Presidential Support in the 2012 Election

Abstract

In their classic 1960 work, Angus Campbell and his colleagues offer a model to explain political behavior. They posit a funnel of causality, whereby the causal flow moved from remote long-term forces, such as socio-demographics and party identification, to more immediate short-term forces, such as issues and candidates, finally arriving at the vote choice itself. This explanation has withstood the test of time in studies of the United States and other democracies. The question at hand in this article is how Latin American immigrants comport themselves in the national political environment of the United States. Can the political preferences of Hispanic immigrants be explained pretty much the way the political preferences of native-born Americans can be explained? In other words, does the funnel of causality apply to them? Our findings, based on analysis of 2012 American National Election Study and Latino Immigrant National Election Study survey data, indicate that it does.

The American Voter (TAV) established a paradigm for studying political behavior. Angus Campbell and his colleagues (1960) posited a funnel of causality to explain presidential voting behavior, whereby the causal flow moved from remote long-term forces, such as socio-demographics and party identification, to more immediate short-term forces, such as issues and candidates, finally arriving at the tip of the funnel, the vote choice itself. That paradigm has not gone unchallenged, and other approaches have competed for its place in the theoretical spotlight, for example, rational choice, political geography, historical-institutional, media and communications, campaign strategy. Even when TAV is acknowledged as the dominant approach, the extent of its reach faces challenges. For instance, if its explanation held for the originally studied presidential elections of 1952 and 1956, does it hold now? To answer that question, “Thousands of journal articles and conference papers have been published and presented on the subject of voting behavior in the decades since 1960, pieces that have reconsidered the original work of Campbell, Converse, Miller and Stokes” (Lewis-Beck et al. 2008, x). A more recent example includes a special issue of Electoral Studies, dedicated to revisiting TAV, its evidence and theories (Lewis-Beck 2009). Thus, the body of published work following TAV precepts has become vast, and some of this research has examined its empirical reach. Michael Lewis-Beck and his colleagues, revisiting TAV theory and applying it to the U.S. presidential elections of 2000 and 2004, draw an unambiguous conclusion: “Does the social-psychological explanation of presidential vote choice, developed in The American Voter and symbolized in its famous theory of the ‘funnel of causality,’ still hold? The essential answer drawn from our revisit must be yes” (2008, 427). Whether it applies to other advanced democracies poses a question explored early on for the British case (Butler and Stokes 1969). The funnel idea still guides much British electoral research, including refined points over the exogeneity of party identification (Clarke et al. 2004, 2009; Whiteley et al. 2013).

The Michigan model, as it is sometimes called, has found ground beyond Anglo-Saxon shores. A current cutting-edge volume, The Nordic Voter, by Åsa Bengtsson and his colleagues explicitly acknowledges the debt: “The general idea behind the sequencing of chapters is well known to most students of electoral behavior: the ‘funnel of causality’” (2014, 10). A recent work on French presidential elections provides an acknowledgment “To Philip E. Converse,” and claims, “Our approach is straightforward, drawing on the founding ‘Michigan model’ of political behavior … or at least the French variant” (Lewis-Beck, Nadeau, and Bélanger 2012, 12). They apply such a model to French voters in four election surveys from 1988 to 2007. In the first national election survey investigation of vote choice in Austria, Sylvia Kritzinger and her colleagues state in their section on theory, “we organize the work around the funnel of causality, first introduced by Campbell et al.” (2013, 26–27).

None of the works cited thus far, however, examine cultures that are not clearly Western and have both lower incomes and more fragile democratic institutions. Unfortunately, systematic political behavior research based on scientific election surveys has been relatively scarce in these parts of the world. That picture has begun to change, especially in Latin America, where a considerable number of well-crafted public opinion polls have now been administered, primarily through the good offices of the Latin American Public Opinion Project at Vanderbilt University (Carlin, Singer, and Zechmeister 2015; Seligson and Zechmeister 2012; Zechmeister and Corral 2013). A question, then, is how these citizens behave at the ballot box. Does the funnel of causality argument help account for their choices? One recent study, which explicitly fits a Michigan-style model to an election survey data pool from twelve Latin American countries finds, encouragingly, that almost half the variance in national vote choice can be explained (Lewis-Beck and Ratto 2013).

That last result implies that Latin American citizens, in their democratic settings, respond to electoral cues roughly the same way as North American citizens do. Still, the suggestion does not speak to how Latin Americans behave politically when they are actually in North America, either as naturalized U.S. citizens or as immigrants as yet without citizenship. This, of course, is the burning concern of this paper. How do these immigrants from Latin America comport themselves in the U.S. national political environment? Can the political preferences of these Hispanic immigrants be explained in much the same way as the political attitudes of native-born Americans? In other words, does the funnel of causality apply to them?

Our hypothesis is that it does, because they are subject to the same forces as other democratic actors in the American system. Many immigrants came to the United States as adults, missing out on the traditional political socialization during the teenage years in the United States. But, like all people, they absorbed perspectives and attitudes in their countries of origin that likely provided guidance as they acculturated to the new political environment. Further, adjusting to life in a new country involves much change and flexibility. During such a transition, immigrants learn from their new friends, coworkers, leaders, and experiences. They are affected by the same political climate as all Americans, and therefore we anticipate that this will lead them to develop party identification, views on political issues, evaluations of the economy, and opinions about candidates.

In our analysis, we test whether the same structural model of political preference—essentially the Michigan model—works for Hispanic immigrants and for native-born Americans, with regard to the 2012 U.S. presidential election. We find that it does. We come to this conclusion after comparative analysis of the Latino Immigrant National Election Study (LINES) and the American National Election Studies (ANES), gathered in 2012 preelection surveys conducted either by phone (LINES) or face-to-face (ANES).¹ We first explicate the dependent variable. Then, we lay out the estimation strategy, a block-recursive one, successfully employed elsewhere. Next, we consider the independent variables as sets, in order of their appearance in the funnel: socio-demographics, partisanship, issues, candidates. As we shall see, the estimated models of presidential support, LINES versus ANES, appear similar. However, that does not mean that their political choice processes are identical. On the contrary, some differences are intriguing but nevertheless seem contextually understandable.

THE DEPENDENT VARIABLE: VOTES, PREFERENCES, APPROVAL

Initially, one might suppose the preferred dependent variable would be vote intention, measured equally in both these preelection surveys, with an item something like, “If the presidential election were held tomorrow, who would you vote for?” However, upon inspection, several obstacles to such a comparison appear. First, a question of that type was not asked in the LINES survey. Instead, the closest thing to a vote question (translated into English) reads as follows (see PREVOTE_PREF-PRWHO variable, LINES preelection): “Talking about the elections for president in the United States, do you have a preference for one of the presidential candidates? If ‘Yes,’ which candidate do you prefer?”

The obvious reason for this nonstandard formulation is that the majority of respondents in the survey are not U.S. citizens and so are not eligible to vote (as assumed by the standard formulation). Nevertheless, this preference question could perhaps be used as a vote intention proxy, assuming that an Obama (or Romney) response would reflect their vote intention (if they could vote). We did this, and uncovered the following distribution of among those who expressed a preference: Obama =90.3 percent, Romney = 9.4 percent, Other candidate = 0.3 percent. This is a useful result, suggesting an overwhelming desire to vote for Obama (even if they could not). However, it is not very helpful beyond that, since it is a blunt instrument. Because almost everyone selects one candidate (Obama), little variance is left to explain.

Put another way, if this simple dichotomy, Obama or Romney, were used as a dependent variable in a logistic regression analysis of these data, few independent variables would be found statistically significant. We would learn very little about what caused presidential support to vary in this particular population. Thus, we need another, more finely calibrated, measure of presidential support. Also, we need a measure of support that is theoretically more universal (stretching across this immigrant population of citizens and noncitizens, and across the U.S. population of mostly native-born citizens). The ideal measure would be some continuous assessment of presidential satisfaction, or approval, which fortunately we do have. It reads (translated into English) as follows (see PRESAPP_APPPRES, LINES pre-election): “Do you approve or disapprove of the way Barack Obama is handling his job as president? (Do you approve strongly or not strongly? Do you disapprove strongly or not strongly?)”

On the basic approval question, 90 percent of those surveyed provided an assessment. When constructing the approval scale, we faced the problem that respondents were not given an explicit neutral category. As Herbert Weisberg remarks in his book on survey error, “some people will not be able to translate their opinion into one of the available response alternatives, such as when they have a neutral opinion but no neutral alternative is offered” (2009, 132). Therefore, we coded responses of don’t know as the neutral category of 3, which does not seem to create theoretical or empirical difficulties.² On this 5-point scale, then, respondents spread themselves out well along the distribution (from 1 = strongly disapprove, to 5 = strongly approve). Fewer than half (49.2 percent) chose the 5 option, meaning that most of these respondents had less than complete satisfaction with President Obama’s performance. In other words, although many were fully behind him, others were less so, and to differing degrees. The item, in sum, allows the respondents a finer calibration of their attitude, making this a patently attractive indicator of support, and one that could also be followed in the ANES preelection survey as well. Admittedly, it is not a vote variable. But, we must remember that such approval measures are routinely used in comparative electoral behavior studies to construct a “popularity function” (setting the dependent variable as some sort of approval measure) when a “vote function” (setting the dependent variable as some sort of vote measure) is not available (Nannestad and Paldam 1994; Lewis-Beck and Stegmaier 2013).

THE ESTIMATION STRATEGY

The funnel of causality theory posits a causal flow from one set of variables to another. The more remote variables influence the less remote variables, in turn, finally arriving at the tip of the funnel, where the individual political actor responds. For example, religion shapes party identification, party identification shapes gay marriage attitudes, gay marriage attitudes shape candidate feelings, and candidate feelings trigger presidential support (or its withdrawal). Movement through this causal chain proceeds from long-term forces, such as religion and party identification, to short-term forces, such as gay marriage attitudes and candidate feelings. The long-term forces tend to be enduring, stable, lasting—in a word, exogenous. The short-term forces tend to be fickle, unstable, of the moment—in a word, endogenous. The generative process of political behavior takes place over time, among many variables, and implies the operation of a multiequation system rather than a single-equation one. Precise specification, and estimation, of such a multiequation system can be daunting.

A practical solution is block-recursive modeling, first extensively employed by Warren Miller and Merrill Shanks (1996). A sequence of equations is specified, then estimated one after the other, each containing the prior causal variables. The final equation contains all the relevant causal variables and estimates their direct effects on the political behavior under study. Their indirect effects, of course, are transmitted through the coefficients of the more remote equations. Econometrically, a core assumption is that the independent variables in a prior block are exogenous to the variables in a later block. An additional assumption is that the estimation error in a prior block lacks correlation with the estimation error in a subsequent block (Kmenta 1997). Under these assumptions, estimating the series of equations with ordinary least squares (OLS) yields consistent coefficients, given the continuous measure we have of the approval variable. We estimate (OLS) a four-equation block-recursive causal system of presidential support in the 2012 election, beginning with the first block, and with the LINES data. We then go on to similar estimation for the ANES data, ultimately comparing the two sets of results.

SOCIO-DEMOGRAPHICS (LINES)

An adult’s socio-demographic characteristics heavily determine his or her place in the hierarchy of society. In the Hispanic community we are studying, for example, a churchgoing, college-educated man who lives with his family in a nice house and earns a respectable salary conveys a certain image of himself in the political world. We might expect such characteristics to influence the kinds of candidates he prefers. These socio-demographic variables are measured in the LINES survey, as follows, in a series of dummy variables (1,0): Gender (1 = female), Church Attendance (1 = nearly weekly or more), Marital Status (1 = married living with spouse), Education (1 = high school or more), Class (1 = middle or upper), Homeowner (1 = owner), Income (1 = $20K or more), Age (1 = forty-five or older), Health Insurance (1 = possess). To these standard socio-demographic variables, we add additional variables that uniquely characterize the socio-demographic status of immigrants: the dummy variable of Citizenship (1 = citizen), Year of Arrival in the United States (actual year), Language Spoken at Home (1= only English to 5 = only Spanish), and country of origin. Approximately 70 percent of survey respondents were born in Mexico, about 5 percent in Cuba, and 4 percent in the Dominican Republic. We create separate dummy variables for each of these countries. We also create dummies for the regions of Central America (15 percent of the respondents) and South America (4 percent). In our multivariate analysis, we include dummies for these three countries and Central America, with South America serving as the omitted category.

Before looking at these multivariate results, we ask whether these variables relate to presidential approval for Obama in a simple bivariate way. Citizenship and year of arrival achieve the standard level of statistical significance (0.05, one-tail). Their correlations (Pearson’s r) with the 5-point approval scale (1 = strong disapproval and 5 = strong approval) are -0.058 and 0.091 respectively. These relationships mean that citizens were more likely than non-citizens to disapprove of Obama, and those who arrived in the United States more recently were more likely to approve of him than those who arrived long ago. Further, among the countries and regions of origin, dummy variables—Cuba, Central America, and South America—are statistically significant. The relationship is strongest for immigrants from Cuba, showing a correlation of -0.193. Those from Central America were more likely to approve of Obama (0.077) and those from South America less likely to do so (-0.084), though notably the strength of these relationships is much weaker than for Cuban Americans. In the bivariate correlations, Mexican origin just misses significance (a 0.055 correlation), and Dominican Republic origin is far from it.

A few other socio-demographic variables came close to conventional statistical significance. Specifically, women and homeowners were more approving of Obama’s performance, correlations of 0.055, and 0.053, respectively. Also, respondents who identified as middle or upper class and those who spoke more Spanish at home held more negative assessments, both showing correlations of -0.047. Although thin, these findings merit pursuit in a multivariate context.

Our first equation, for the socio-demographic block, can be read as follows:

Approval = f (Socio-demographics)     Model 1

where the socio-demographic variables are gender, church, marriage, education, class, home, income, age, health insurance, citizenship, year of arrival, language spoken at home, and country or region of origin. In other words, the model bases itself on our sixteen available variables, which we regress (OLS) on the 5-point presidential approval scale. This estimation appears in table 1 (column 1).

Do these multivariate estimates change the picture much from the bivariate correlations? Only a little. We see that those who arrived in the United States more recently are more likely to support Obama, as are respondents who were born in Mexico, the Dominican Republic, and Central America; respondents of Cuban origin were less likely to approve of him. Citizenship and language spoken at home fail to reach statistical significance. Looking at the other variables in this model, we observe that women are more likely to support Obama, as are homeowners. The impact of homeownership

Table 1.

LINES 2012 Regression Analysis of Presidential Approval

Source: Authors’ compilation based on McCann and Jones-Correa 2012.

Note: Unstandardized regression coefficients are reported, with the standard errors in parentheses. The dependent variable, Presidential Approval ranges from 1 = strongly disapprove to 5 = strongly approve. Don’t knows are coded in the middle category of 3. Those who refused are excluded from the analysis. Independent variables: Year of Arrival is the actual year reported by the respondent; Language spoken at Home ranges from 1= only English to 5= only Spanish. Don’t knows and refusals on these questions were coded as missing. Refused and don’t knows are included in the 0 category for the following dichotomous variables: all other socio-demographic variables including country of origin (coded into a set of dummy variables: Mexico, Cuba, Dominican Republic, Central America, and the excluded variable South America), party identification and ideology, and all social issue variables. The economic variables are coded as 1 = much worse to 5= much better, with don’t knows included in the middle “the same” category, and refusals are excluded. The candidate feeling thermometers range from 0 to 100, with don’t knows, refused, and don’t recognize coded as the neutral 50. We code the variables this way to avoid drastically reducing the N. Our experiments suggest that this coding decision does not seriously influence the magnitude of the coefficients reported.

Standard errors in parentheses.

*p < .05 (one-tailed test)

suggests that “having a stake” in the economics of the community can engender more broad system support for government officials, such as the president. We return to other, short-term economic effects later.

Overall, how do socio-demographics account for differences in presidential approval? Not well. The correlation (R) between all sixteen variables and approval reaches 0.280. However, even that modest magnitude diminishes when the R-squared (0.078) and the adjusted R-squared (0.057) are examined. We must conclude that, despite applying a rather full and more or less standard battery of socio-demographic measures, they make just a small dent in our understanding of presidential support among the Hispanic immigrant community. For that understanding, we need to move farther down the funnel.

PARTISANSHIP: PARTY AND IDEOLOGY (LINES)

Given that we are working with a sample of immigrants to the United States, rather than a sample of the native born, it might be supposed that the question of party identification would be met with an overwhelming response of “don’t know.” Zoltan Hajnal and Taeku Lee argue that immigrants “will be motivated to remain neutral—at least in their self-identification” (2011, 277). Theoretically, they draw on both the Michigan model, which posits party identification is acquired initially through socialization (Campbell et al. 1960), as well as the rational choice approach, which emphasizes the role of information and updating (Fiorina 1981). Hajnal and Lee contend that because many immigrants spent their formative years in their country of birth, they missed out on U.S. political socialization and might not have enough information about the parties and their positions to choose which party to align with.

However, elsewhere in this issue, David Sears, Felix Danbold, and Vanessa Zavala find that many Latino immigrants have indeed acquired a party identification. Using the 2012 LINES and ANES surveys, they examine party identification among naturalized and native-born citizens, and find that Latinos are right on par with white Americans (Sears, Danbold, and Zavala, this volume, table 2). Nonincorporation is more prevalent among non-naturalized Latinos, but they find signs of partisan crystallization in this group. Ultimately, Sears and his colleagues conclude that Hispanic immigrant incorporation into the party system has come about “following a path shaped by a combination of exogenous and endogenous forces … that aligns with the socialization approach put forth in The American Voter.”

Just over half of LINES respondents identify with a party, according to responses from the standard party identification question, which reads as follows (PTYID_RPTYID): “Generally speaking, do you usually think of yourself as a Democrat, a Republican, an Independent, or what?”

The relative frequency distribution on the variable is Democrat = 43.0 percent, Republican = 8.1 percent, Independent = 31.9 percent, don’t know/other party/no preference = 17.0 percent. Clearly, these respondents show knowledge of the major political parties in the country, and are able to evaluate their relationship to these parties. Of those who selected one of the two major parties, the preference slanted heavily in favor of the Democrats (84.1 percent) over the Republicans (15.9 percent).

We would expect Democratic identifiers to be much more approving of Obama, and they are, as we shall see. However, party identification does not exhaust the possible sources of partisan attachment. In particular, liberal-conservative ideology has been put forward as an additional long-term social-psychological anchor in the electorate. The extent to which Americans have a meaningful ideology remains a source of controversy. Campbell and his colleagues (1960) were actually skeptical that voters had coherent patterns of political thought that could be labeled liberal or conservative. Lewis-Beck and his colleagues echo this skepticism in their update of Campbell, reporting that “only a bare majority of our respondents (51.4 percent) consider themselves to be either liberal or conservative” (2008, 223). Further, too, is the question of what the terms liberal and conservative substantively mean, even for those who so place themselves (Lewis-Beck et al. 2008). Other work argues that these ideological orientations do have politically meaningful content (Jacoby 2002; Sears 2001). In any case, it seems worth exploring their impact in these LINES data.

In the LINES survey, an ideology self-placement item was posed, of the following form: “We hear a lot of talk these days about liberals and conservatives. How would you describe yourself? Extremely liberal, liberal, moderate or middle of the road, slightly conservative, conservative, or extremely conservative, or haven’t you thought much about this?”

The distribution on this item does show a fairly large group, 24.7 percent, who simply have not thought much about it. Also, a sizable percentage, 18.9 percent, located themselves in the middle of the road. Still, a fair share were willing to call themselves liberal, 21.6 percent in all. This group is of special interest to us because President Obama has frequently been tagged with the liberal label. We reasoned that, among immigrants who explicitly saw themselves as liberal, they would be more supportive of Obama independently of whether they identified with the Democratic Party. As we shall see, that is the case.

Incorporating this second block of variables, the model now reads

Approval = f (Socio-demographics, Partisanship)     Model 2,

where partisanship includes party identification (Democrat or not) and ideological identification (Liberal or not). The estimates for this model appear in table 1 (column 2).

Clearly, these social-psychological anchors are working as expected. The coefficients of both variables are in the right direction and highly significant (p > 0.001). Further, party identification has special weight, as the coefficient of 0.760 shows (standardized coefficient = 0.311). Democratic identifiers, not surprisingly, are much more likely to approve of President Obama’s job performance. But, beyond that, liberal identifiers (not all of whom are Democrats) also back Obama. Because these two variables—party and ideology—have the same (1, 0) metric, their relative strength can be compared in an examination of the unstandardized coefficients and are, respectively, 0.760 and 0.474. We observe that party effects are about double ideology effects, again in line with expectations.

How much does the addition of these long-term factors help the model, in terms of its explanatory power? A good deal. The multiple correlation (R) has climbed to 0.449, and the adjusted R-squared to 0.181. The model appears to be gaining traction, as the forces of electoral politics fill the funnel.

ISSUES (LINES)

We now enter into the realm of short-term forces, namely issues. Campbell and his colleagues (1960) saw that the issues of the day would influence preferences. However, they felt that that influence would be limited, because of the heavy requirements an issue must meet, to change political behavior. These criteria are three and may be summarized as follows: the voter has to see the issue, the voter must have a preference on the issue, and the voter must believe that one candidate is closer. Because these criteria are demanding, we should not expect issues to overwhelm the model. But we should imagine that some issues will make a difference.

Preferences on a number of issues were measured in the survey. First, we constructed a series of dummy (1, 0) variables on five social issues: abortion (1 = women should have full access), citizenship for illegals (1 = favor), government services (1 = should provide more), death penalty (1 = favor), gay marriage (1 = allow). Responses of don’t know or those who refused were coded as 0 to avoid losing too many observations. The univariate distributions of these variables merit attention. For one, fewer than half (44.8 percent) felt the government should clearly provide more services, a finding that flies in the face of the stereotype that “the immigrant is looking for a handout.” For another, these respondents overwhelmingly favor (at 84.1 percent) citizenship for illegal visitors, which speaks to a sense of grievance they feel over current policies. With respect to cultural issues, the findings are not surprising: only 29.1 percent favor gay marriage, only 18.0 percent favor full access for women to abortion care, and a large minority (40.8 percent) favor the death penalty. Of these issues, three reach conventional bivariate statistical significance in the expected direction, and their correlations with presidential approval are reported in parentheses: citizenship for illegals (r =0.079); gay marriage (r = 0.093), and increasing government services (r = 0.099).

As well, a series of 5-point economics questions was asked (where 1 = much worse and 5 = much better): past personal finances, future personal finances, past national economy, future national economy. These economic evaluations tap the classic dimensions of time (retrospective versus prospective) and target (personal versus national) that are employed in the economic voting literature, as applied to election surveys (on the evolution of economic voting research from TAV to the present, see Lewis-Beck and Stegmaier 2009; for classics on economic voting in American national election surveys, see Fiorina 1981; Kiewiet 1983). Here, for example, is the wording for the past national economy item, which follows standard phrasing: “Now thinking about the economy in the country as a whole, would you say that over the past year the nation’s economy has gotten better, stayed about the same, or gotten worse?”

This evaluation, a retrospective on the national economy, has been labeled sociotropic. The evaluation of past personal finances has been labeled pocketbook (see Kinder and Kiewiet 1981). The distribution on pocketbook evaluations shows more people think they became better off (38.3 percent) than worse off (31.0 percent) over the past year. The distribution on sociotropic evaluations shows the same pattern, though more lopsided: 36.5 percent saw a better national economy, but only 16.8 percent saw a worse one. Interestingly, this squares with the balance of consumer confidence in the 2012 American electorate as a whole (Lewis-Beck and Tien 2014).

Which of these issue variables sustain their effect once they are included in a multivariate model? Here is the equation to be estimated, showing the third block of variables entered:

Approval = f(Socio-demographics, Partisanship, Issues)     Model 3

where the issues are those noted and have the same coding. The estimates appear in table 1 (column 3).

As can be seen, the only social issue to survive the controls in terms of statistical significance is gay marriage. Although its coefficient does not suggest a large effect (that is, the standardized value = 0.070, which is much less than the Democratic and Liberal standardized coefficients in this model, 0.246 and 0.122 respectively), it remains noteworthy as a lavender button issue that, when pushed, tends to distance Hispanic immigrants from Obama. But the real action rests with the economy. Three of the four economic issue variables easily attain statistical significance. Moreover, the expected pattern from the literature is reproduced, with sociotropic evaluation clearly mattering more than pocketbook (Stegmaier and Lewis-Beck 2013). Indeed, the retrospective national economic evaluations coefficient, with a standardized value of 0.225, implies that evaluations of the national economy weigh heavily in the minds of these immigrants. For them, perhaps more than any other issue, and certainly more than any issue we have in our analysis, the national economy is what counts. Given the realities of their circumstances, and their aspirations for their families, this seems unsurprising.

Overall, does consideration of the role of issues increase much our understanding of presidential support within this community? Yes. The addition of this issues battery of nine items to the explanatory mix boosts substantially our ability to account for Obama support. The multiple R now stands at 0.574. Most impressively, the adjusted R-squared has risen from 0.181 in model 2 to 0.302 in model 3.

CANDIDATES (LINES)

We now arrive at that most immediate of voter concerns, the qualities of the candidates themselves. Their traits are very much at the fore-front of a presidential campaign, and the Obama-Romney contest was no exception. What is it about Obama’s leadership style that increases his support? Decreases it? A considerable literature has examined this general question, beginning with Campbell and his colleagues (1960). At least three qualities seem consistently important across studies in the United States and Great Britain: competence, honesty, and empathy (Clarke et al. 2004; Kessel 2004; Kinder 1986). Two basic methods have been used to measure these leader images among electorates: one direct (specific items, such as “Is the candidate honest?”) and the other indirect (general feeling thermometers, such as “rate how warm you feel toward candidate X”). The LINES survey employs the latter approach. The wording (translated into English) of the lead question (THERMPRE_THIN-TRO) is this: “I’d like to get your feeling toward some of our political leaders and other people who are in the news these days. I’ll read the name of a person and I’d like you to rate that person using something we call the feeling thermometer. Ratings between 50 degrees and 100 degrees mean that you feel favorable.”

These feeling thermometers, when applied here to Obama and Romney, are telling. First, most respondents were able to use the scale (don’t knows and refusals for Obama = 3 percent, for Romney = 13.3 percent). This suggests that they are well aware of these political figures, and have opinions about them. The modal feeling toward Obama is 100, more than one in four respondents (26.7 percent) giving him this score. Obviously, he is a major object of attention in their American political universe. The median responses reveal a good deal about relative candidate evaluations: Obama median = 80, Romney median = 45. The typical Hispanic immigrant, then, likes Obama almost twice as much as Romney. One would expect this image advantage would convert to job support for Obama. We incorporate these feeling thermometers into our model, as a fourth block:

Approval = f (Socio-demographics, Partisanship, Issues, Candidates)     Model 4

where candidates are measured by a feeling thermometer variables for Obama (0–100) and for Romney (0–100). So as to not lose observations, respondents who did not express a thermometer score on the candidates were recoded at the neutral position of 50 on the scale. The estimates for this model appear in table 1 (column 4).

An examination of these feeling thermometer coefficients demonstrates that affect toward both candidates, positive or negative, contributes to the Obama approval score. But the respondent’s feeling toward Obama counts for about twice as much as that toward Romney (from table 1, the Obama coefficient = 0.018, the Romney coefficient = -0.009). Moreover, Obama’s likeability has a greater impact on his job approval than any of the other independent variables. For example, the effect of the Obama thermometer even surpasses the effect of Democratic Party identification, when we compare the standardized coefficients from model 4, 0.385 and 0.142 respectively. In other words, the most important long-term force in the causal explanation, party identification, is trumped by this even more important short-term force of incumbent candidate appeal.

What of the other direct effects captured in this final, more fully specified, model? With respect to the long-term forces, the only influences that remain are from income, immigrants who came from Mexico and Central America, party identification, and ideology. Notably, the other immigrant-specific characteristics (citizenship, year of arrival, and language spoken at home) do not improve our understanding of presidential approval in our complete Michigan model, at least in their capacity as direct effects. With respect to short-term forces, such as social issues, attitudes on gay marriage still manage to affect Obama support. Turning to economic issues, we find sharp, and multiple, effects. Both retrospective and prospective sociotropic evaluations easily achieve statistical significance, as do past pocketbook evaluations. In terms of relative magnitude, the sociotropic evaluation dominates the pocketbook, suggesting that Hispanic immigrants concern themselves, first, with the larger national picture (Kiewiet and Lewis-Beck 2011).

The fit statistics have now made it into the familiar territory of well-specified models of U.S. presidential survey data: R = 0.688, adjusted R-squared = 0.450. The story that they tell about the political behavior of contemporary Hispanic immigrants to the United States sheds considerable light. On the whole, the story affirms our hypothesis: their presidential support can be explained, to a considerable extent, by the classic Michigan model. Citizen status, year of arrival, and language spoken at home undoubtedly affect one’s status in a community, but they do not enhance our immediate understanding of voting and presidential support among Hispanic immigrants. Next, we move on to compare the Michigan model’s performance with the same model applied to the 2012 ANES national sample.

THE BLOCK-RECURSIVE MODEL: ESTIMATES FROM THE ANES

The application of the funnel of causality, via estimation of the block-recursive model with the LINES data, shows promise. The time has come to apply the same theory and estimation strategy to the parallel ANES survey (pre-2012 election, face-to-face). But first we need to examine the sample and measures of the two in order to establish similarities and differences. The dependent variable, presidential approval, is worded and calibrated the same way, though the distribution of responses clearly favors Obama less in this sample. Further, all of the independent variables available in both the LINES survey and the ANES survey are coded the same way. Finally, although race is effectively a constant in the sample of Hispanic immigrants, we do use race variables with the ANES data. In sum, the two variable sets match almost exactly. The parallel estimates (OLS), on the ANES survey appear in full, from block one to block four (see table 2). We introduce the main results from each block, beginning with socio-demographics, and compare them with the LINES findings before fully and systemically evaluating them.

In the initial block, we again relate the socio-demographic variables to presidential approval. We note right away a different pattern from the LINES data. As preface, with respect to the bivariate correlations, almost all are statistically significant, and in the expected direction. Several of these correlations exceed 0.10 in magnitude, for example, marital status, homeownership, religious attendance, social class, age, and income. Further, the correlations of race-ethnicity with Obama support stand out: r = 0.33 (black versus others); r = 0.14 (Hispanic versus others).³ These variables, taken together in the multiple regression equation of block 1, appear in table 2 (column 1) with their respective unstandardized coefficients. Overall, they definitely yield a higher fit (an R = 0.445, an adjusted R-squared = 0.193) when compared with the LINES results. Clearly, socio-demographics matter.

What about partisanship, whose effects are estimated in table 2 (column 2)? When party and ideology are added to the model, they are again significant. Further, they again contribute substantially to an increase in model fit. Where we see the differences are in the magnitudes of the coefficients. For LINES (table 1, column 2), the party identification coefficient is 0.760; for ANES, it is 1.403. Similarly, for LINES, the ideology coefficient is 0.474; for ANES, it is 0.810. A comparison of these un-standardized coefficients suggests that partisanship, by either measure, exercises about twice the pull on the national sample, as opposed to the immigrant sample.

In the third block, the issues variables are included, table 2 (column 3). Issues appear important for both the LINES and the ANES samples. The influence of two issues stands out: gay marriage and the economy. For both, gay marriage is the most important social issue: as judged by the unstandardized coefficients, for LINES it is 0.180 and for ANES it is 0.299. Note, further, that the issue of gay marriage has a bigger impact within the national sample than the immigrant sample.

Of all the issues under study, social or not, the past performance of the economy manages the biggest effect, in both samples. Interestingly, the coefficients of the sociotropic retrospective variable come close to each other: for LINES, 0.280, and for ANES, 0.439. For both samples, then, the national economy is highly important. Also the pocketbook variable, as measured by past personal finances, achieves statistical significance across the samples. However, interestingly, its coefficient attains nearly twice the magnitude in the Latino sample, suggesting that for hard-pressed groups, such as immigrants, personal financial well-being has more relevance.

The last set of variables to be entered is the feeling thermometers for the candidates, table 2 (column 4). For both samples, opinions about the candidates add greatly to understanding their presidential support. Moreover, of all the increments to variance explained, in moving from block to block, this makes the largest relative contribution for LINES. Candidate attributes, more than any other set of factors, distinguish their choices. This contrasts, for example, with the powerful role of partisanship

Table 2.

ANES 2012 Regression Analysis of Presidential Approval

Source: Authors’ compilation based on ANES 2012.

Note: Unstandardized regression coefficients are reported, with the standard errors in parentheses. The dependent variable, Presidential Approval ranges from 1 = strongly disapprove to 5 = strongly approve. Don’t knows were coded in the middle category of 3. Those who refused are excluded from the analysis. Independent variables: Refused and don’t knows were included in the 0 category for the following dichotomous variables: all socio-demographic variables, party identification and ideology, and all social issue variables. The economic variables are coded as 1 = much worse to 5 = much better, with don’t knows included in the middle of the same category and refusals excluded. The candidate feeling thermometers range from 0 to 100, with don’t knows, refused, and don’t recognize coded as the neutral 50. We code the variables this way to avoid drastically reducing the N. Our experiments suggest that this coding decision does not seriously influence the magnitude of the coefficients reported.

Standard errors in parentheses

*p < 0.05 (one-tailed test)

for ANES. In sum, candidate appeal—personality if you will—helps structure immigrant preferences to a high degree, especially when understood comparatively.

COMPARING MODEL PERFORMANCE ACROSS THE TWO SAMPLES

Thus far, we have examined each survey, and each block, separately. Now we evaluate the Michigan model across samples and blocks. In table 3, a comparison is made across three criteria: contribution, structure, and strength. First, we observe that the funnel of causality provides a useful overall framework of explanation. However, at each stage, it does not always register the same value. The roles of partisanship and issues are rather similar across the two samples. That is, both sets of variables make a strong contribution to explaining presidential support. However, that does not mean that the structure and strength of the relationships within those two blocks is identical. Partisanship exercises stronger effects in the national sample than in the Hispanic immigrant sample. We expected that immigrants would acquire party identification as they assimilated into American society, but, as Sears, Danbold, and Zavala show in this volume, this is true for naturalized immigrants; among the non-naturalized, however, fewer self-identify with a party, though many show signs of nascent party identification. Because of this, and because immigrants have not spent their entire lives in the U.S. political system, the long-term anchor of partisanship has a weaker influence on vote choice. Finally, although both samples are responsive to issues, the set is not the same; the immigrants come close to being single-issue voters, in that it is mainly the economy (in its multiple dimensions) that moves them. In contrast, the national sample respondents have multiple concerns beyond the economy.

Table 3.

Comparison Across the Michigan Model: LINES and ANES, 2012

Source: Authors’ calculations based on tables 1 and 2.

Note: Contribution+ refers to the increment to the adjusted R-squared, once the variables are added to the block; Structure^ refers to the variables added to the block that have statistically significant regression slopes (in the expected direction); Strength# refers to the effects (slopes) of the variables added to the block that have statistically significant coefficients (in the expected direction).

Now consider the role of candidate appeal, which contributes more than any other block regardless of sample. This result suggests the vital role of candidate characteristics, such as leadership, honesty, and—more generally—personality traits. These things matter a great deal for both Hispanic immigrants and the national sample. Further, the structure of the response appears the same: liking Obama generates the expected presidential approval in both groups, as does disliking Romney. In other words, both these attitudes have an independent effect. Moreover, that effect operates asymmetrically, in that feeling toward Obama has at least double the impact of feeling toward Romney.

It is with respect to the first block, sociodemographics, that we observe—ostensibly—the biggest difference across samples. In the immigrant survey, it appears to add very little to the explanation, even with the inclusion of immigrant-specific traits and country of origin. In the national survey, by way of contrast, it provides variance, structure, and strength. However, looks can be deceptive. Most of this socio-demographic power comes from the necessary addition of the race-ethnicity variables to the ANES data (in table 2, column 1). These variables are not added into the LINES analyses in an obvious sense, because that sample is racially and ethnically essentially homogenous, that is, it composes in itself a Hispanic demographic. Statistically, that means that it is added, but it is added as a constant, into the intercept. Demographics, then, make an important contribution to the explanation of presidential support here; but to better capture that effect structurally one needs data over time rather than cross-sectionally. This same difficulty is occurring elsewhere as analysts try to capture the effects of this restricted variance problem with regard to the observation of economic crisis (see, for example, Lewis-Beck and Fraile 2014).

Overall, how well does the funnel of causality explanation account for Obama support? One answer comes from assessment of goodness-of-fit at the last stage, with model 4. At first glance, it looks as if the national sample receives better explanation than the immigrant sample: that is, the adjusted R-squareds, respectively, are 0.761 and 0.450. However, as we know, a comparison of the R-squared (adjusted or not) can be perilous across samples, particularly if we have variation differences in the X or Y variables (Lewis-Beck and Skalaban 1990). In such a situation, which clearly is the case here, a more secure measure of comparative model fit comes from the standard error of estimate, or SEE (Lewis-Beck and Kruger 2007). As the bottom of tables 1 and 2 (column 4) show, the SEE is 0.89 and 0.85, respectively. In other words, by this measure, the models yield about the same fit. Put another way, if we wish to predict presidential support, we are able to do about equally well for either the immigrant sample or the national sample. In sum, the funnel of causality argument, taken to its last stage, receives about the same level of empirical confirmation, on either dataset.

CONCLUSION

We began with the notion that the funnel of causality explanation for political behavior, as created by Campbell and his colleagues (1960) and developed by their adherents, could be successfully applied to yet another democratic setting. In particular, we pursued the idea that it could explain the political preferences of Hispanic immigrants in the context of the 2012 U.S. presidential election. According to our findings, the Michigan model shows itself capable of rendering a more than satisfactory account of Obama support in that contest, for both Hispanic immigrants and American nationals generally.

Our finding that the Michigan model holds for immigrants suggests that people are able to assimilate to new political environments. They have common experiences with other Americans, form attachments to parties, develop opinions on political issues, evaluate the economy, and assess the candidates. These factors, following the funnel of causality, predictably affect how immigrants judge the president’s performance in much the same way they affect how Americans, in general, judge the president.