Literature Review

document2.pdf

Home >Social Science homework help >Literature Review

Why the Giant Sleeps So Deeply: Political Consequences of Individual-Level Latino Demographics∗

Rodolfo O. de la Garza, Columbia University

Seung-Jin Jang, Kookmin University, Seoul, Korea

Objectives. We seek to examine what common characteristics Latino voters share distinctively from nonvoting Latinos and how they differ from those of non-Latino voters. Methods. We use the method of classification tree to find what variables best describe the shared characteristics of Latino and non-Latino voters. Results. The results indicate that age, the strength of partisanship, and the level of education characterize both Latino and non-Latino voters. However, there is a sharp difference in how age interacts with other the two variables. In the Latino sample, the over- whelming majority of younger people do not turn out to vote, and it is among older Latinos that education and partisanship play an important role in distinguishing voters and nonvoters. By contrast, among non-Latinos, it is younger people whose participation is sensitive to the level of education or strength of partisanship, while the older are overwhelmingly voters regardless of other factors. Conclusions. At the individual level, Latinos in the United States still face substantial barriers in political incorporation: in contrast to non-Latinos, they do not seem to develop the habit of voting even after they have aged enough, unless they are helped by higher levels of education or stronger partisanship. At the aggregate level, the pool of Latino likely voters is relatively small, which in turn emphasizes the potential significance of the GOTV campaigns in increasing participation rates among Latinos.

Beginning in the 1980s, it has been a common feature of U.S. elections that politicians and pundits, at the beginning of the campaign season, assert that Latino votes will play a significant role in determining the election’s out- come, and at the conclusion of the election, lament their unrealized potential. Although, due to continuous flows of immigrants and high birth rates among the second generation, Latinos outnumbered African Americans in the 2000 Census as the nation’s largest minority, they nonetheless have failed to trans- form their increased number into higher levels of participation and thus to greater electoral significance (e.g., de la Garza and DeSipio, 1999, 2005).

∗Direct correspondence to Rodolfo O. de la Garza, Department of Political Science, 1432 IAB, Columbia University, New York, NY 10027 〈[email protected]〉. Seung-Jin Jang 〈[email protected]〉 will share all data and coding for replication purposes. The authors thank David Epstein, Robert Erikson, and Robert Shapiro for reading the article and providing helpful comments.

896 Social Science Quarterly

Scholarly efforts to explain this repeated failure of Latinos’ electoral power to materialize led to a vast literature around the question of why Latinos do not vote (for an extensive review of this literature, see de la Garza, 2004). Studies have found that demographic characteristics of Latinos—for instance, lower socioeconomic status, relatively younger age structure, and incomplete social assimilation—continue to create barriers to their full empowerment. Institutional constraints such as haphazard naturalization requirements or tricky registration and voting rules also disproportionately affect them (Jones- Corea, 2001). At the same time, contemporary Latino immigrants often find a void of mobilization drives by mainstream political institutions and major parties, which would help them overcome disadvantages in socioeconomic resources and political socialization (e.g., Leighley, 2005; Wong, 2006).

In this article, instead of once again asking why Latinos do not vote, we turn the question on its head and ask who Latino voters are. In other words, we examine what common characteristics Latino voters share compared to nonvoting Latinos, and how these differ from those of non-Latino voters. An- swers to this question have significant implications for understanding Latino electoral participation, since changes in an individual Latino’s demographic characteristics are not easily produced nor do they necessarily lead to changes in the likelihood of voting and, similarly, electoral involvement does not automatically endow Latino voters with political and socioeconomic charac- teristics equivalent to those of non-Latino voters. In addition, by carefully identifying what segment of Latinos are most likely to be voters and how large this segment is among the general population of Latinos, we can formulate a long-term prospect of by how much the pool of Latino likely voters is likely to increase and how it might be achieved.

Our main objective, therefore, is to efficiently classify voters and nonvoters among Latinos and to understand what, among the demographic and political variables that have traditionally been considered important in explaining voter turnout in general and among Latinos in particular, best describe the shared characteristics of Latino voters. In this sense, our inquiry is inductive and thus differs from traditional approaches that aim at finding a causal relationship through hypothesis testing. To this end, we employ a statistical method that has rarely been used in the study of political participation: the classification tree. The classification tree method, by searching for the optimal classification scheme among observations without the need to assume a particular para- metric model or interaction structures a priori, is particularly suitable for our purpose.

The article is organized as follows. In the next section, we provide a brief introduction to the classification tree method. The second section applies the method to a national sample of Latinos and non-Latinos and then discusses distinctive characteristics of Latino voters and how they differ from those of non-Latino voters. Based on these findings, the third section generalizes our discussion to the general population of Latinos in the United States using Census data. The last section concludes.

Political Consequences of Latino Demographics 897

Method: Classification Tree

Across various disciplines, researchers often create classification problems. For instance, when entomologists discover a new insect, they seek to classify it properly. Social network analysts investigate the interactions within a set of individuals and are interested in identifying individuals who have similar aims or attributes. With different objects of study and methods employed, the end result of classification studies is a partition of the set of objects or individuals into a set of disjointed classes such that members of each class possess some properties that distinguish them from members of other classes (Gordon, 1999:1–2).

Since a seminal study by Breiman et al. (1984), a wide range of statisti- cal techniques for classification, including cluster analysis, neural networks, and classification and regression trees, has been developed under the general rubric of “pattern recognition” (Ripley, 1996). These techniques, by search- ing inductively for patterns in the data rather than imposing a particular parametric model on the data, provide a much more flexible way of sum- marizing the data and presenting the relationships among variables than do standard regression approaches (Breiman, 2001). Political scientists, however, have paid little attention to them, though many hypotheses in the discipline are expected to be highly nonlinear, massively interactive, and heavily context dependent and thus cannot be appropriately captured by traditional regression approaches (Beck, King, and Zeng, 2000:22). Only recently has a small num- ber of judicial scholars begun to acknowledge advantages of these techniques in answering questions of interest. For instance, a group of researchers utilized classification trees to predict the outcome of Supreme Court cases (Ruger et al., 2004; for an abbreviated version, see Martin et al., 2004). Similarly, Kastellec (forthcoming) employed classification trees to show how the courts’ legal rules on search and seizure and confession cases have changed histori- cally. Eisenberg and Miller (2007) also adopted classification trees to explain the pattern of jury trial waivers in large corporate contracts. These examples show how the statistical classification method can be used to study important political phenomena in a way that traditional regression techniques cannot address.

The method of classification tree adopted in this article has a number of advantages in handling classification problems, compared to maximum likelihood estimation such as logit or probit.1 First, and most importantly, rather than assuming a particular parametric model that maps from predictor variables to the outcome variable, classification trees take a nonparametric ap- proach and directly partition the outcome space based on predictor variables

1A regression tree is constructed from data with a continuous or ordinal outcome variable in much the same way as a classification tree is constructed from data with a nominal outcome variable. In this article, we use rpart routine in R (Therneau and Atkinson, 1997) to construct classification trees.

898 Social Science Quarterly

selected for their efficacy in reducing classification errors.2 Second, classifica- tion trees do not require the analyst to specify interactions among predictor variables in advance, as the procedure of recursive partitioning inherently con- siders interactions among all predictor variables entered in a model. Third, they are better able to handle missing values in predictor variables, highly robust to outliers, and invariant to monotone transformations of predictor variables (Feldesman, 2002:258). Last, but not least, trees are easy to interpret: we can graphically represent complex data no matter how many dimensions are rele- vant, and easily classify a new observation based on the values of the variables used to construct the trees (Breiman, 2001:206).

Specifically, a classification tree is constructed as follows.3 First, a single variable is found that best (the meaning of “best” is explained below) splits the data into two groups (nodes).4 The data are separated, and then this process is applied separately to each node, and so on recursively until the number of cases at the node either reaches a (prespecified) minimum size or until no improvement in classification can be made (Therneau and Atkinson, 1997:4). More formally, let C = {c1, c2 , . . . , cJ} be the set of all possible classes, and χ be the space of all possible values of predictor variables x = (x1, x2 , . . . , xK). Then, a classification tree is a partition of χ into disjointed sets, B1, B2 , . . . , BJ, such that the predicted class is cj if x ∈ Bj.

At each stage in the recursive partitioning, all the allowable ways of splitting a subset of χ are considered and the one that minimizes the impurity function of the resulting nodes is chosen (Sutton, 2005:310). The impurity function is a function of the proportions of the data belonging to all possible classes of the outcome variable, denoted by p1, p2 , . . . , pJ. Typically, we define the impurity function using the Gini index of diversity:

g ( p1, p2, . . . , p J ) = 1 − J∑

j =1 p 2j .

Intuitively, this function assumes its maximum value when a node is composed of an equal number of cases from all possible classes, and its minimum value

2Breiman nicely summarizes the basic idea behind the methods of pattern recognition in general, and classification tree in particular: “The approach is that nature produces data in a black box whose insides are complex, mysterious, and, at least, partly unknowable. What is observed is a set of X ’s that go in and a subsequent set of Y ’s that come out. The problem is to find an algorithm f (X ) such that for future X in a test set, f (X ) will be a good predictor of Y ” (2001:205). This approach is sharply different from standard regression approaches, where we assume a stochastic data model for the inside of the black box.

3For more complete introduction, see Ripley (1996) and Hastie, Tibshirani, and Friedman (2001).

4Although one may want to consider splits involving a linear combination of two or more variables, or splits resulting in more than two nodes, rpart only allows single variable and binary splits. In fact, this restriction is beneficial, as it makes the resulting tree easier to interpret and decreases the computing time. In addition, if just single variable splits are used, the resulting tree is invariant with respect to monotone transformations of the variables (Sutton, 2005:319).

Political Consequences of Latino Demographics 899

when all cases in the node are of the same class. In practice, the value of the impurity function using the cases corresponding to the node being split is compared to the weighted average of the impurity for the two resulting nodes, with the weights proportional to the number of cases corresponding to each node (Sutton, 2005:311).

The trickiest part of constructing a classification tree is to determine how large the tree should be. If the tree becomes too large, it may overfit the data; if the tree is too small, then we do not exploit enough information present in the data and thus classification accuracy may suffer. To find a parsimonious tree that accurately describes the data, the established procedure is to let the tree grow very large and “prune” it back. Breiman et al. (1984), who introduced this approach, proposed a strategy that attempts to balance the complexity of the tree (number of terminal nodes) and the misclassification error rate: as trees grow, they become more complex, while misclassification error rates decline; the model is penalized each time another split occurs, and, if the additional split does not improve the fit enough to overcome the penalty (the tree “cost”), the smaller tree is selected (Feldesman, 2002:263). More formally, let |T| equal the size (number of terminal nodes) of a tree T, and R(T) be the estimate of the misclassification rate of T. Then, for α ≥ 0 , the cost-complexity measure is defined as:

Rα (T) = R (T) + α |T| , where α is the penalty for each additional terminal node. If α is 0, there is no penalty for additional nodes and the tree continues to grow until there is no case belonging to the same node with distinct values of predictor variables. If α is set to infinity (or to an arbitrarily large number), the tree with only one node is selected. Therefore, we indirectly control the size of a tree by specifying α, and the goal of the pruning procedure is to find the proper value of α that minimizes Rα (T).

We use the cross-validation procedure to determine the proper value of α.5

Simply put, the data are randomly divided into V subsets of roughly equal size.6 Each of the V subsets of the data is left out in turn as a test sample, and the remaining data are used to develop a sequence of trees with various values of α. Then, the test sample is run through the trees, and the classification errors are recorded, as well as the α that results in the lowest misclassification error. After this process is repeated V times, the results are averaged to arrive at a single error rate estimate corresponding to that value of α. This cross-validation estimate of the error rates serves as the estimate of the true misclassification

5The following rather complicated procedure is necessary because calculating the misclas- sification rate with the data used to create the tree is often biased. The estimate calculated as such, resubstitution error rates, can be a very poor estimate of the tree’s misclassification rate for future observations, since it can decrease as more nodes are created, even if the selection of splits is just responding to “noise” in the data, and not real structure (Sutton, 2005:311).

6The default in rpart is V = 10.

900 Social Science Quarterly

rate for the tree created using all the data and pruned using this value of α (Sutton, 2005:314).

In actual practice, it is advised that the analyst follows the “1-SE rule.” If we plot α against the cross-validation error rates, it is often the case that an initial sharp drop is followed by a relatively flat plateau and then a slow rise. Therefore, simply choosing α with the minimum cross-validation error may not be reliable, especially if it is on the flat plateau. To avoid this, each estimate of the cross-validation error rate is computed with its standard error, and any rate within one standard error of the achieved minimum is considered as equivalent to the minimum. Then, the simplest tree, among all those “tied,” is chosen (Therneau and Atkinson, 1997:13).

Analyzing the Classification Trees

In this section, we apply the method of classification tree to a national sample of Latinos. The data are drawn from the national survey of Latinos conducted by Harvard University, Kaiser Family Foundation, and the Wash- ington Post in 1999. The original survey was conducted by telephone among a nationally representative sample of 4,614 adults 18 years and older living in the United States, including 2,417 Latinos and 2,197 non-Latinos. Therefore, one immediate advantage of the Harvard/Kaiser/Washington Post data, com- pared to other survey data on Latinos, is that they allow us to compare our findings regarding Latino respondents to those of the non-Latino population. As our primary focus is on the question of turnout in the national elections, we limit our samples to U.S. citizens only, which results in the final sample of 1,580 Latinos and 2,148 non-Latinos (1,779 non-Hispanic whites and 369 blacks).

Before proceeding, it is important to reiterate what our primary objective is. We do not aim at discovering causal relationships; our method precludes that. For instance, even if we find that a certain variable X is important in distinguishing Latino voters and nonvoters, this does not necessarily imply that X causes Latinos to vote; instead, we can say that Latino voters are disproportionately concentrated in the group of people who commonly have the attribute X, which helps us predict that Latinos with X are more likely to be voters than nonvoters. In other words, the classification tree shows us what the attributes are that most parsimoniously characterize the difference between voters and nonvoters among the Latino population. By the same reason, one should also note that the omission of a variable in the final classification tree cannot be taken as evidence that the variable is not strongly related to voting turnout of Latinos. The tree only selects enough variables (and the interaction among them) up to a point where the tree does not overfit the data but maximizes prediction accuracy. Therefore, the classification tree method should be considered as a complement to, rather than a replacement of, traditional maximum likelihood estimations.

Political Consequences of Latino Demographics 901

FIGURE 1

Latino Voters and Nonvoters in the 1998 Election (N = 1,580)

25- -years old or older?

294/234

Is strong partisan?

Voters

361/276

No Yes

Yes

More than high school education?

Yes

Voters

530/324395/243

Voters

1286/752

Voters

925/476

Total Sample

812 Voters / 768 Nonvoters

Nonvoters

Latino Voters and Nonvoters in the 1998 Election

The classification tree between Latino voters and nonvoters in the 1998 election is depicted in Figure 1. In constructing the figures that follow, we include various political and demographic variables that the literature has found important in determining voting turnout of Latinos and the general population. Table 1 shows the list of variables included to construct the trees and how they are coded. One should note that, among the predictor variables considered, only three appear in the classification tree presented in Figure 1. This is because, though other predictors may also exert some influence on voting turnout among Latinos, only three attributes are enough to parsimoniously classify voters and nonvoters among Latinos.

The way to read the classification tree is as follows: starting with the root node, each branch describes a rule for splitting the data. Therefore, among all eligible Latino voters, those who are younger than 25 years old are predicted to proceed down the left branch to the next node, while those who are 25 years old or older are predicted to proceed down the right branch. Nodes surrounded by squares with solid line are terminal nodes: thus, Latinos who are 24 years old or younger are classified as nonvoters. Nodes surrounded by squares with dashed line are nonterminal nodes, which will be further split based on the next variable. The numbers in each node indicate how many

902 Social Science Quarterly

TABLE 1

List of Variables for Classification Trees

Variables Coding Scheme Min. Max.

Partisan strength

The extent to which a respondent identifies with a political party

1 4

Political trust The extent to which a respondent trusts the federal government

1 4

Nativity Whether a respondent was born in the U.S. or not

0 1

Proportion of life in U.S.

The number of years a respondent has lived in the U.S. divided by his or her physical age. For the native born, the value is 1

0.16 1

Pan-ethnic perception

The extent to which a respondent thinks that Latinos in the U.S. share common political interests and goals

1 4

Discrimination Whether a respondent, including a family member or a close friend, experienced racial discrimination during the last 5 years

0 1

English fluency

The extent to which a respondent uses English, relative to Spanish, at home

1 5

Ethnic activities

The number of activities that a respondent participated in: working for a Latino political candidate, attending a public meeting or demonstration regarding Latino concerns, and contributing money to a Latino candidate or Latino political organization

0 3

Education Level of education 1 5 Age Physical age 18 97 Religiosity Subjective importance of religion

in a respondent’s everyday life 1 4

Catholic A respondent’s religious preference 0 1 Employment Whether a respondent is currently

working full time (including self-employed) or not

0 1

Gender A respondent’s gender 0 1 Marital status Whether a respondent is currently

married or not 0 1

NOTE: Among these variables, pan-ethnic perception, discrimination, English fluency, and ethnic activities are operationalized only among Latinos, and thus are excluded in the analysis of non-Latinos.

Political Consequences of Latino Demographics 903

cases reach the node, followed by how many are correctly classified among the cases that reach the node. For example, at the left-most terminal node, 294 Latinos are classified as nonvoters based on their age, among whom 234 (79.59 percent) are correctly predicted (i.e., actually did not vote). On the other hand, those Latinos who are 25 years old or older proceed rightward from the root node and further split based on their strength of partisanship: if they are strong partisans, they become classified as voters; if not, the tree continues and splits them again based on their education levels: those Latinos with a high school diploma or less are classified as nonvoters and those with higher education as voters.

From the classification tree, we can see that the single most important variable that distinguishes between voters and nonvoters among Latinos is an individual’s age: almost 80 percent of Latinos who are younger than 25 are correctly predicted not to vote in the 1998 election. On the other hand, Latinos who are over 25 are predicted to vote and 58.48 percent of them actually did. Based solely on an individual’s age, we are able to correctly classify 986 (234 + 752 = 986) Latino voters and nonvoters out of a total sample of 1,580 (62.41 percent).

However, Figure 1 also shows that age is not enough to characterize the Latino voting population. Among 1,286 Latinos who are 25 years old or older and thus predicted to be voters, only 752 (58.48 percent) reported that they actually turn out to vote. To further classify Latino voters and nonvoters within this age group, the tree suggests that two additional variables are influential: strength of partisanship and level of education. One should note that partisanship alone does not improve our classification in that it classifies Latinos who are 25 years old or older as voters irrespective of their strength of partisanship. The strength of partisanship distinguishes voters and nonvoters only when it is interacted with levels of education.7 Specifically, strong partisans are overwhelmingly likely to vote regardless of their education: 361 Latinos fall into this class and 276 (76.45 percent) of them actually voted. On the other hand, Latinos with only weak partisanship need an additional boost from higher education to turn out: 530 Latinos who are not strong partisans and have more than high school education are classified as voters, and 324 (61.13 percent) of them actually did vote. Conversely, the tree predicts 395 Latinos who are not strong partisans and have only up to high school education to be nonvoters, and 243 (61.52 percent) of them actually did not vote.

Based on the classification tree, the basic profile of Latino nonvoters is that they are young or have a relatively low level of education in conjunction with weak partisanship: among 294 Latinos who are younger than 25, only 20.41 percent voted; among 395 Latinos who are not strong partisans and have only up to a high school level of education, only 38.48 percent voted. Similarly,

7In fact, uncovering this kind of nonlinear interaction is one of the advantages of the classification tree method with its nonparametric nature.

904 Social Science Quarterly

we can compare the turnout rates among different Latino subgroups. For instance, the turnout rates increase from 51.39 percent in the total sample of Latinos to 76.45 percent among Latinos with strong partisanship, or to 61.31 percent among Latinos with weak partisanship but more than a high school education. Overall, based on only three attributes, age, the strength of partisanship, and levels of education, we are able to correctly classify 600 voters (74 percent) and 477 nonvoters (62 percent) from a total sample of 812 voters and 768 nonvoters, which results in the proportional reduction in error of 34.5 percent.8

One important disadvantage of the classification tree method is that a proper measure of uncertainty is not available. In typical regression or maximum likelihood approaches, each coefficient estimate has its standard error, which reflects the amount of uncertainty associated with the estimation and thus is used for the purpose of hypothesis testing. In classification trees, on the other hand, we instead use cross-validation classification rates to measure the reliability of the tree we obtain. As explained in the previous section, the cross- validation classification rate is obtained by randomly dividing the sample into 10 subsets of roughly equal size and then using each of the 10 subsets as a test sample against the tree constructed from the remaining observations. The cross-validation estimate produced by this procedure is an unbiased and independent estimate of the true classification rate (Feldesman, 2002:264– 65). However, as this estimated rate depends on a particular random division of observations, we simulate the cross-validation procedure 1,000 times and calculate the mean cross-validation classification rate. In Figure 1, the mean cross-validation classification rate is 67.55 percent, with the mean proportional reduction in error of 33.25 percent.

Latino Voters and Nonvoters in the 1996 Election

Figure 2 displays a similar result of classification tree, but with respect to the 1996 election. In 1996, an individual’s age is still the single most important variable that distinguishes Latino voters and nonvoters. Among Latinos who are younger than 25, almost three-quarters abstain from voting. If an individual is over 25, the tree classifies him or her as a voter and over 65 percent of these individuals actually did vote. Based solely on an individual’s age, we can correctly classify 1,059 Latino voters and nonvoters (67 percent).

Among Latinos who are 25 years old or older, the next important variable that distinguishes voters and nonvoters in the 1996 election is how long they have been in the United States. If an individual is over 25 years old and

8As 51.4 percent of eligible Latino voters reported voting in the 1998 election, we can be at least 51.4 percent correct in predicting Latino voters by blindly classifying everybody as voters. As the correct classification rate from the classification tree is 68.2 percent, we improve the classification rate (i.e., reduce the classification error) by 100 × 68.2 − 51.4

100 − 51.4 = 34.5 .

Political Consequences of Latino Demographics 905

FIGURE 2

Latino Voters and Nonvoters in the 1996 Election (N = 1,580)

old or older?

294/220

Is strong partisan?

Voters

627/513

No Yes

Yes

More than high school education?

Yes

Voters

70/40220/150

Voters

1286/839

290/180

Total Sample

913 Voters / 667

Proportion of life in U.S. > 0.54?

Voters

996/729

YesNo

Voters

369/216

Is partisan?

YesNo

115/66

Voters

254/167

Nonvoters

25- -years

has spent over half his or her life in the United States, the classification tree predicts that the individual will be a voter, which is correct for 73.19 percent of the cases. On the other hand, 290 Latinos who are over 25 but have spent less than half their lives in the United States are classified as nonvoters, among whom 180 actually abstain from voting. By taking one’s proportion of life in the United States into consideration in conjunction with age, the correct classification rate improves to 71.46 percent.

Similarly to the 1998 election, one’s strength of partisanship and level of education come as the next important variables in classifying Latino voters and nonvoters. First, among Latinos who are classified as nonvoters based on their age and proportion of life in the United States, strong partisans now are predicted to be voters: in the sample, 70 Latinos fall into this class, and 40 actually voted. If they are not strong partisans, then the tree classifies them as nonvoters, which is correct for 68.18 percent of the cases. Second, among Latinos who are classified as voters based on their age and proportion of life in the United States we find an interaction between the strength of partisanship and level of education. If they have more than high school education, then we

906 Social Science Quarterly

classify them as voters: 627 Latinos fall into this class, and the classification is correct for 81.82 percent of the cases. Among those Latinos with only up to a high school education, only partisans (strong or weak) are predicted to be voters: 254 Latinos fall into this class, among whom 65.75 percent actually turn out to vote. On the other hand, if an individual is an independent, we predict him or her to be a nonvoter. Overall, with four attributes, our method is able to correctly classify 720 voters and 436 nonvoters among a total sample of 913 voters and 667 nonvoters (73.2 percent) in the 1996 election with the proportional reduction in error of 36.4 percent. The mean cross-validation classification rate from 1,000 simulations is 70.5 percent, with the mean proportional reduction in error of 30.1 percent.

Overall, Figure 2 shows that the basic profile of Latino voting population found in the 1998 election is still intact with respect to the 1996 election. Again, Latinos who are young or have a relatively low level of education in conjunction with weak partisanship are overwhelming failing to turn out. One difference across the two election years is that one additional variable, proportion of one’s life in the United States, plays an important role in the 1996 election. To the extent that the length of residence reflects the degree of an individual’s acculturation (being knowledgeable about and socialized into U.S. political life), Figure 2 suggests the roles of education and partisanship may differ across different levels of acculturation. Specifically, for those Latinos who are less acculturated, it takes much stronger partisanship to be motivated to vote. On the other hand, if they are highly acculturated, that is, have spent more than half their life in the United States, then relatively weaker partisanship is enough, especially combined with more than a high school education, to mobilize them to participate.

Classification Trees for Non-Latinos

For the purpose of comparison, Figures 3 and 4 present classification trees for the non-Latino population with respect to voting turnout in the 1998 and 1996 elections, respectively. The first thing to note is that classification trees for the non-Latino population are also composed of a similar set of variables as in the case of Latinos. As with Latinos, an individual’s age is the single most im- portant variable that distinguishes voters and nonvoters among non-Latinos. Then, each age group is further split by such variables as strength of partisan- ship and level of education. On the other hand, a caveat in reading Figures 3 and 4 is their relative low predictive power: the proportional reduction in error is only 19.3 percent (10.3 percent from 1,000 simulated cross-validations) in Figure 3 and 21.15 percent (14.2 percent from 1,000 simulated cross- validations) in Figure 4. This is mainly due to much higher reported turnout rates among non-Latinos in both elections, which is also reflected in the particularly simple classification tree in Figure 4.

Political Consequences of Latino Demographics 907

FIGURE 3

Non-Latino Voters and Nonvoters in the 1998 Election (N = 2,148)

Voters

773/388

Yes

Voters

1375/1090

Total Sample

1478 Voters / 670

251/168

Yes

Voters

522/305

135/81

Yes

Voters

387/251

Is partisan?

More than high school education?

No Yes

60/46

Voters

75/40

No Yes

22/17

Voters

53/35

Nonvoters

39-years-old or older?

27-years-old or older?

31-years-old or older?

One key difference between Latinos and non-Latinos in their classification trees concerns the role of an individual’s age. Obviously, for both Latinos and non-Latinos, an individual’s age is the single most important variable that distinguishes voters and nonvoters. However, once the classification tree initially splits the sample based on age, there is a sharp difference in the way education and partisanship come into play for each age subgroup of Latinos and non-Latinos. For instance, in the Latino sample, the overwhelming ma- jority of younger people does not turn out to vote and levels of education or partisanship do not improve their turnout rates. On the other hand, it is older Latinos for whom education and partisanship play an important role in distinguishing voters and nonvoters. Classification trees for the non-Latino

908 Social Science Quarterly

FIGURE 4

Non-Latino Voters and Nonvoters in the 1996 Election (N = 2,148)

80/75

Voters

320/241

No Yes

Yes

More than high school education?

Yes

Voters

1637/1418

Voters

431/285

Voters

2068/1703

111/67

Total Sample

1708 Voters / 440 Nonvoters

Nonvoters

33-years-old or older?

21-years-old or older?

sample present an opposite pattern: while the older are overwhelmingly voters, it is younger people whose turnout is sensitive to the level of education or strength of partisanship. In addition, this difference between the Latino and non-Latino populations does not disappear even if we increase the tree sizes in Figures 1–4 by intentionally overfitting the data.

The difference between Latino and non-Latino voters as to the role of age, and its interaction with education and partisanship, suggests an important implication with respect to the political incorporation of the Latino popu- lation. Traditionally, an individual’s age has been considered as one of the key components of his or her socioeconomic characteristics, which, in turn, significantly influence one’s extent of voting participation. In general, turnout rises gradually with increasing age until it reaches its peak and levels off in middle age, and gradually declines as maturity fades into old age (e.g., Nie, Verba, and Kim, 1974; Verba and Nie, 1972; Wolfinger and Rosenstone, 1980). This is probably because age represents the accumulation of political experience and information, along with changes in community attachment, political orientation, and civic competence (e.g., Strate et al., 1989). In this respect, declines in participation among the elderly imply the adjustment of political activities to more age-appropriate activities, rather than complete disengagement (Jennings and Markus, 1988). At the same time, age-related changes in participation are typically concentrated in early adulthood. For

Political Consequences of Latino Demographics 909

instance, Miller and Shanks (1996) find that participation rapidly increases over the first two or three elections after becoming eligible to vote, and the rate of increase diminishes after that. This reflects significant changes in one’s lifecycle in early adulthood, such as marriage (Stoker and Jennings, 1995) or parenthood (Jennings, 1979), that bring about changes in one’s social networks and general dislocations from previous modes and styles of living (but see Highton and Wolfinger, 2001). On the other hand, beyond the mid 30s, voting may become self-reinforcing or “habitual” (Gerber, Green, and Shachar, 2003; Plutzer, 2002), as represented by Figures 3 and 4.

However, the classification trees presented in Figures 1 and 2 suggest that this general pattern of political participation does not apply to the Latino population. Latinos are not likely to vote even when they have aged enough— in other words, when they are of sufficient age by which the habit of voting has developed after experiencing several elections in the general population— unless they are stimulated by high levels of education and strong partisanship. Note that the different role of age between Latinos and non-Latinos is not limited only to the foreign born: although the immigration generation status— whether one was born in the United States or not—is also considered in constructing all the trees, Figures 1 and 2 indicate that its inclusion does not help improve the classification of voters and nonvoters in each Latino age group.9

A number of factors may contribute to the distinctiveness of Latino partici- pation, but the most important one is the lack of political incorporation. The majority of the Latino population, especially those who are older, is immigrants and physical age does not necessarily reflect the extent of their experiences with and knowledge of the U.S. political system. For instance, Wong (2000) finds that the acquisition of party identification by Latino immigrants is positively associated with years of residence in the United States but is negatively as- sociated with age. This suggests that political incorporation among Latino immigrants is achieved primarily from experiences with the polity rather than from new social roles they take on as a function of aging. On the other hand, Figure 2 also suggests that the simple fact of longer residence in the United States does not necessarily imply a higher level of incorporation. Even among Latinos who have spent more than half their life in this country, including the native born, voters are still largely concentrated among those who have more than a high school education or explicitly identify with a political party. In other words, low levels of political incorporation of Latinos persist after a significant amount of time has passed in the United States, which lead them to remain “reticent”‘ and “reluctant” to participate (DeSipio, 1996) unless they are stimulated by high levels of education or partisan identification. Addition- ally, it may also be the case that even when Latinos have stayed in the United

9Nor is this due to the concentration of the native born in a particular age group. In the sample, although native-born Latinos are in fact significantly younger than foreign-born Latinos, the proportion of those who are 25 years old or younger accounts for only 30 percent of native-born Latinos.

910 Social Science Quarterly

TABLE 2

Distribution of Eligible Latino Voters in the Census 2000

Less than Half More than Half of Life in the U.S. of Life in the U.S.

Age < 25 19.84% Age ≥ 25 Up to high school education 11.45% 37.90%

More than high school education 4.69% 26.12%

NOTE: Entries are the percentage against the total number of Latinos who are eligible to vote. SOURCE: Integrated Public Use Microdata series (IPUMS), 5% sample data.

States long enough to become no less likely to endorse U.S. political values and attitudes than non-Latinos (e.g., de la Garza, Falcon, and Garcia, 1996), their full incorporation into the U.S. political system can easily be hindered by the persistence of informal mechanisms and practices that politically and socially discriminate against minorities (Hero, 1992).

Estimating Latino Voter Pool from the Census

Although the out-of-sample prediction poses problems in any statistical modeling, the cross-validation procedure used in the classification tree, as well as its interpretability, makes it particularly useful for the purpose of generaliz- ing our findings to the general population. While there are some complicated methods to improve the prediction accuracy of the classification trees, such as bagging or boosting (Sutton, 2005), they tend to make it very difficult to in- terpret the results (Breiman, 2001:208). Therefore, in this article, we adhere to the original analysis presented in the previous section and apply it to the gen- eral population of Latinos in the United States derived from the Census data.

The 2000 Census estimates that there are about 23 million Latinos who are 18 and over, 62.25 percent of whom hold U.S. citizenship. Among this total number of potential Latino voters, Table 2 presents the aggregate number of Latinos from the Census data who fit the profile of predicted voters. Recall that, in the classification trees of Figures 1 and 2, the basic profile of the Latino voting population is those who are 25 years old or older and have spent more than half their life in the United States, are strong partisans, and have more than a high school education.

From Table 2, the prospect of a large pool of Latino likely voters is rather discouraging. First, almost one-fifth of all eligible Latinos voters are between the ages of 18 and 24, most of whom are predicted to be nonvoters. This confirms the usual charges that the relative youth of the Latino population is one of the key factors contributing to their low levels of participation (e.g., DeSipio, 1996). On the other hand, the proportion of Latinos who

Political Consequences of Latino Demographics 911

are predicted to vote regardless of their strength of partisanship, that is, 25 years old or older with more than a high school education, consist of only 30.81 percent of all eligible Latinos. Furthermore, according to Figure 2, if we also take into account the length of stay in the United States, the pro- portion of predicted Latino voters diminishes to only 26.12 percent of all eligible Latinos. This small proportion of predicted voters among Latinos makes a sharp contrast to that of non-Latinos, that is, non-Hispanic whites and non-Hispanic blacks, reported in the Census 2000. For instance, among non-Hispanic whites and non-Hispanic blacks, the proportions of those who are predicted to vote irrespective of their strength of partisanship are 57.31 percent according to Figure 3,10 or 66.07 percent according to Figure 4.11

It is hard to estimate how large the pool of Latino likely voters is among the remaining half of Latino eligible voters, as their turnout is closely related to their strength of partisanship, which is not available from the Census data. On the other hand, a number of national surveys of Latinos (including the one used here) provide estimates for the number of Latinos who strongly identify with one of the two major parties. Typically, the proportion of strong parti- sans among Latinos ranges between 30 percent and 40 percent of respondents across different surveys. In addition, given strong relationships between par- tisan identification and levels of education, the proportion of strong partisans among Latinos is expected to decrease if we focus on those who do not have a college-level education.12 Therefore, strong partisanship only adds roughly 15–20 percent of Latinos to the pool of predicted voters.

Of course, turnout rates among Latinos can and do fluctuate depending on political contexts (e.g., Barreto and Woods, 2002; Pantoja, Ramirez, and Segura, 2001), as well as on mobilization efforts by political parties and nonpartisan organizations.13 However, it is still questionable whether these contextual catalysts would have persistent rather than occasional effects on levels of Latino turnout, as the expansion of political participation among Latinos is often interrupted by a number of factors, including social and res- idential segregation, lower levels of political efficacy and knowledge, and the

10Those who are 39 years old or older or who are between the age of 27 and 38 with more than a high school education.

11Those who are 33 years old or older or who are between the age of 21 and 32 with more than a high school education.

12For instance, according to the most recent national survey of the Latino population (Fraga et al., 2006), among Latino citizens who are 25 years old or older, 36.8 percent strongly identify with one of the two major political parties. If we focus on those who do not have a college-level education, the proportion of strong partisans reduces to 34.27 percent. Even when we further limit the sample to those who have spent more than half their life in the United States, the corresponding numbers increase only about 1.5 percentage points, respectively.

13In fact, the data used in this article do not provide the information regarding the exposure to mobilization efforts, and thus this important variable cannot be included in this article. Therefore, the analysis presented in this section should be considered as a baseline estimate of the Latino voter pool irrespective of the changes in the volume and intensity of electoral mobilization.

912 Social Science Quarterly

possible incompatibility between policy issues salient to Latinos and conven- tional partisan cleavages in U.S. politics (Hajnal and Lee, 2004).

Conclusion

Our answer to the question of who Latino voters are is that they are at least 25 years old or older and are strong partisans or have more than a high school education. Younger Latinos are in the great majority nonvoters regardless of their level of partisanship or education. Older Latinos may become more likely to vote, but only if they have high levels of education or strong partisanship; and this is true even when they have spent more than half their life in the United States. Our findings indicate that, unlike among non-Latinos, electoral participation of Latinos does not incrementally increase as they become older and obtain more experience with U.S. politics.

This, in turn, suggests an important implication with respect to whether Latino participation is shaped by the same factors that influence participa- tion in the general population. The literature typically finds that although standard demographic variables such as education and age and political vari- ables such as strength of partisanship have consistent effects on participation among Latinos as well as in the general population, the unique experiences of Latinos as an immigrant-majority and racial-minority group significantly influence their active participation in politics. Going one step forward, our analysis shows that even those seemingly consistent effects of standard demo- graphic and political variables may also be in need of reconsideration. It is true that among both Latinos and non-Latinos, the same set of demographic and political variables—age, education, and partisanship—is critical in differenti- ating between voters and nonvoters. We find, however, that there is a sharp difference in the way these variables characterize Latino and non-Latino vot- ers. Among non-Latinos, participation rates increase with age, which suggests that age represents the accumulation of political experience and information; and the role of partisanship or education in participation is evident primarily among those who are relatively too young to have accumulated other political resources. By contrast, among Latinos, physical age may not necessarily lead to the accumulation of political resources and thus to higher levels of partic- ipation unless they are further stimulated by higher levels of partisanship or education. By revealing a distinctive dynamic of how standard demographic variables are related to political participation among Latinos, which is not possible in a traditional maximum likelihood estimation procedure, our anal- ysis makes an important contribution for a nuanced understanding of Latino political participation.

The analysis presented in this article transcends pure scientific interests and carries an important practical insight with respect to various efforts to increase participation rates among Latinos. Taken together, our findings suggest that the current pool of Latino likely voters is relatively limited and this will not

Political Consequences of Latino Demographics 913

change in the near future as key variables that differentiate between voters and nonvoters among Latinos cannot be easily manipulated by the government or political parties. In this respect, “get out the vote” (GOTV) campaigns targeted to Latinos become even more important. Although the effect of GOTV campaigns is not limited to a particular racial group (e.g., Gerber and Green, 2000; Green, Gerber, and Nickerson, 2003), they have a particular importance for Latinos as they can mobilize those who are least likely to vote given their own demographic and political characteristics. In fact, a number of studies have found that mobilization efforts can significantly increase turnout rates among Latinos (e.g., de la Garza, Abrajano, and Cortina, 2002a; Shaw, de la Garza, and Lee, 2000), especially when targeting Latinos (DeFrancesco Soto and Merolla, 2006) or when being conducted by co-ethnics (de la Garza, Abrajano, and Cortina, 2002b; Michelson, 2003).

However, the effect of GOTV campaigns on Latinos in the long run may be limited. First, as political parties and candidates strategically target their scarce resources to likely voters (Rosenstone and Hansen, 1993), Latinos are less likely to be exposed to mobilization efforts precisely in places where a large number of Latinos is concentrated (Leighley, 2001). Second, young voters—those least likely to vote—comprise a very large proportion of the Latino potential electorate but it is much more difficult to mobilize younger voters than mature voters (Nickerson, 2006). Of course, this does not necessarily mean that the pool of Latino likely voters is fixed. Our analysis shows that in addition to education, the acquisition of partisanship is an important instrument to overcome a low level of political incorporation among Latinos. Research shows that Latino partisan attitudes are explicitly political, being influenced more strongly by political ideology and issue preferences than by economic factors (Alvarez and Garcı́a Bedolla, 2003). The way that Latinos acquire partisanship is closely related to whether they perceive a political party as more credible and supportive of their preferences and interests on issues that matter to Latinos (Nicholson and Segura, 2005; Uhlaner and Garcia, 2005). Therefore, in order to increase participation rates among Latinos, the government and political institutions should be more actively engaged in incorporating immigrants and their second-generation descendants to the political mainstream, and political parties and Latino civic associations in particular should take more explicit positions on policy issues salient to Latinos so as to accelerate the development of enduring partisan identification among them.

REFERENCES

Alvarez, R. Michael, and Lisa Garcı́a Bedolla. 2003. “The Foundations of Latino Voter Parti- sanship: Evidence from the 2000 Election.” Journal of Politics 65(1):31–49.

Barreto, Matt A., and Nathan D. Woods. 2002. “Latino Voting Behavior in an Anti-Latino Political Context.” In Gary M. Segura and Shaun Bowler, eds., Diversity in Democracy: Minority Representation in the United States. Charlottesville, VA: University of Virginia Press.

914 Social Science Quarterly

Beck, Nathaniel, Gary King, and Langche Zeng. 2000. “Improving Quantitative Studies of International Conflict: A Conjecture.” American Political Science Review 94(1):21–35.

Breiman, Leo. 2001. “Statistical Modeling: The Two Cultures.” Statistical Science 16(3):199– 231.

Breiman, Leo, Jerome H. Friedman, Richard A. Olshern, and Charles L. Stone. 1984. Classi- fication and Regression Trees. Belmont, CA: Wadsworth International Group.

de la Garza, Rodolfo O. 2004. “Latino Politics.” Annual Review of Political Science 7:91–123.

de la Garza, Rodolfo O., Marisa A. Abrajano, and Jeronimo Cortina. 2002a. Latino Voter Mobilization in 2000: Predictors of Latino Turnout. Report 3. Claremont, CA: Tomás Rivera Policy Institute.

———. 2002b. “Get Me to the Polls on Time: Coethnic Mobilization and Latino Turnout.” In Jane Junn and Kerry L. Haynie, eds., New Race Politics in America: Understanding Minority and Immigrant Politics. Cambridge, NY: Cambridge University Press.

de la Garza, Rodolfo O., and Louis DeSipio. 1999. Awash in the Mainstream: Latinos Politics in the 1996 Elections. Boulder, CO: Westview Press.

———. 2005. Muted Voices: Latinos and the 2000 Elections. Lanham, MD: Rowman & Littlefield.

de la Garza, Rodolfo O., Angelo Falcon, and F. Chris Garcia. 1996. “Will the Real Americans Please Stand Up: Anglo and Mexican-American Support of Core American Political Values.” American Journal of Political Science 40(2):335–51.

DeFrancesco Soto, Victoria M., and Jennifer L. Merolla. 2006. “Vota port u Futuro: Partisan Mobilization of Latino Voters in the 2000 Presidential Election.” Political Behavior 28(4):285– 304.

DeSipio, Louis. 1996. Counting on the Latino Vote: Latinos as a New Electorate. Charlottesville, VA: University of Virginia Press.

Eisenberg, Theodore, and Geoffrey P. Miller. 2007. “Do Juries Add Value? Evidence from an Empirical Study of Jury Trial Waiver Clauses in Large Corporate Contracts.” Journal of Empirical Legal Studies 4(3):539–88.

Feldesman, Marc R. 2002. “Classification Trees as an Alternative to Linear Discriminant Analysis.” American Journal of Physical Anthropology 119(3):257–75.

Fraga, Luis R., John A. Garcia, Rodney Hero, Michael Jones-Correa, Valerie Martinez-Ebers, and Gary M. Segura. 2006. Latino National Survey [computer file]. Ann Arbor, MI: Inter- University Consortium for Political and Social Research.

Gerber, Alan S., and Donald P. Green. 2000. “The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment.” American Political Science Review 94(3):653–63.

Gerber, Alan S., Donald P. Green, and Ron Shachar. 2003. “Voting May Be Habit-Forming: Ev- idence from a Randomized Field Experiment.” American Journal of Political Science 47(3):540– 50.

Gordon, A. D. 1999. Classification, 2nd ed. Boca Raton, FL: Chapman & Hall/CRC.

Green, Donald P., Alan S. Gerber, and David W. Nickerson. 2003. “Getting Out the Vote in Local Elections: Results from Six Door-to-Door Canvassing Experiments.” Journal of Politics 65(4):1083–96.

Hajnal, Zoltan, and Taeku Lee. 2004. “Latino Independents and Identity Formation Under Uncertainty.” Unpublished paper. San Diego, CA: University of California, San Diego.

Political Consequences of Latino Demographics 915

Hastie, Trevor, Robert Tibshirani, and Jerome Friedman. 2001. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. New York: Springer.

Hero, Rodney E. 1992. Latinos and the U.S. Political System: Two-Tiered Pluralism. Philadelphia, PA: Temple University Press.

Highton, Benjamin, and Raymond E. Wolfinger. 2001. “The First Seven Years of the Political Life Cycle.” American Journal of Political Science 45(1):202–09.

Jennings, M. Kent. 1979. “Another Look at the Life Cycle and Political Participation.” American Journal of Political Science 23(4):755–71.

Jennings, M. Kent, and Gregory B. Markus. 1988. “Political Involvement in the Later Years: A Longitudinal Survey.” American Journal of Political Science 32(2):302–16.

Jones-Corea, Michael. 2001. “Institutional and Contextual Factors in Immigrant Naturaliza- tion and Voting.” Citizenship Studies 5(1):41–56.

Kastellec, Jonathan P. Forthcoming. “The Statistical Analysis of Judicial Decisions and Legal Rules with Classification Trees.” Journal of Empirical Legal Studies.

Leighley, Jan E. 2001. Strength in Numbers? The Political Mobilization of Racial and Ethnic Minorities. Princeton, NJ: Princeton University Press.

———. 2005. “Race, Ethnicity, and Electoral Mobilization: Where’s the Party?” In Christina Wolbrecht and Rodney E. Hero, eds., The Politics of Democratic Inclusion. Philadelphia, PA: Temple University Press.

Martin, Andrew D., Kevin M. Quinn, Theodore W. Ruger, and Pauline T. Kim. 2004. “Com- peting Approaches to Predicting Supreme Court Decision Making.” Perspectives on Politics 2(4):761–67.

Michelson, Melissa R. 2003. “Getting Out the Latino Vote: How Door-to-Door Canvassing Influences Voter Turnout in Rural Central California.” Political Behavior 25(3):247–63.

Miller, Warren E., and J. Merrill Shanks. 1996. The New American Voter. Cambridge, MA: Harvard University Press.

Nicholson, Stephen P., and Gary M. Segura. 2005. “Issue Agendas and the Politics of Latino Partisan Identification.” In Gary M. Segura and Shaun Bowler, eds., Diversity in Democracy: Minority Representation in the United States. Charlottesville, VA: University of Virginia Press

Nickerson, David W. 2006. “Hunting the Elusive Young Voter.” Journal of Political Marketing 5(3):47–69.

Nie, Norman H., Sidney Verba, and Jae-on Kim. 1974. “Political Participation and the Life Cycle.” Comparative Politics 6(3):319–40.

Pantoja, Adrian D., Ricardo Ramirez, and Gary M. Segura. 2001. “Citizens by Choice, Voters by Necessity: Patterns in Political Mobilization by Naturalized Latinos.” Political Research Quarterly 54(4):729–50.

Plutzer, Eric. 2002. “Becoming a Habitual Voter: Inertia, Resources, and Growth in Young Adulthood.” American Political Science Review 96(1):41–56.

Ripley, Brian D. 1996. Pattern Recognition and Neural Networks. Cambridge, NY: Cambridge University Press.

Rosenstone, Steven J., and John Mark Hansen. 1993. Mobilization, Participation, and Democ- racy in America. New York: Macmillan.

Ruger, Theodore W., Pauline T. Kim, Andrew D. Martin, and Kevin M. Quinn. 2004. “The Supreme Court Forecasting Project: Legal and Political Science Approaches to Predicting Supreme Court Decisionmaking.” Columbia Law Review 104(4):1150–1209.

916 Social Science Quarterly

Shaw, Daron, Rodolfo O. de la Garza, and Jongho Lee. 2000. “Examining Latino Turnout in 1996: A Three-State, Validated Survey Approach.” American Journal of Political Science 44(2):338–46.

Stoker, Laura, and M. Kent Jennings. 1995. “Life-Cycle Transitions and Political Participation: The Case of Marriage.” American Political Science Review 89(2):421–33.

Strate, John M., Charles M. Parrish, Charles D. Elder, and Coit Ford III. 1989. “Life Span Civic Development and Voting Participation.” American Political Science Review 83(2):443–64.

Sutton, Clifton D. 2005. “Classification and Regression Trees, Bagging, and Boosting.” In C. R. Rao, E. J. Wegman, and J. L. Solka, eds., Handbook of Statistics 24: Data Mining and Data Visualization. Amsterdam: Elsevier B.V.

Therneau, Terry M., and Elizabeth J. Atkinson. 1997. An Introduction to Recursive Parti- tioning: Using the Rpart Routines. Mayo Foundation Technical Report. Rochester, MA: Mayo Foundation, Mayo Clinic.

Uhlaner, Carole J., and F. Chris Garcia. 2005. “Learning Which Party Fits: Experience, Ethnic Identity, and the Demographic Foundations of Latinos Party Identification.” In Gary M. Segura and Shaun Bowler, eds., Diversity in Democracy: Minority Representation in the United States. Charlottesville, VA: University of Virginia Press.

Verba, Sidney, and Norman H. Nie. 1972. Participation in America: Political Democracy and Social Equality. New York: Harper and Row.

Wolfinger, Raymond E., and Steven J. Rosenstone. 1980. Who Votes? New Haven, CT: Yale University Press.

Wong, Janelle S. 2000. “The Effects of Age and Political Exposure on the Development of Party Identification Among Asian American and Latino Immigrants in the United States.” Political Behavior 22(4):341–71.

———. 2006. Democracy’s Promise: Immigrants and American Civic Institutions. Ann Arbor, MI: University of Michigan Press.

its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's

express written permission. However, users may print, download, or email articles for individual use.