Illegal Immigration and Media Exposure: Evidence on Individual Attitudes

Illegal immigration has been the focus of much debate in receiving countries, but little is known about what drives individual attitudes towards illegal immigrants. To study this question, we use the CCES survey, which was carried out in 2006 in the United States. We find evidence that - in addition to standard labor market and welfare state considerations - media exposure is significantly correlated with public opinion on illegal immigration. Controlling for education, income and ideology, individuals watching Fox News are 9 percentage points more likely than CBS viewers to oppose the legalization of undocumented immigrants. We find an effect of the same size and direction for CNN viewers, whereas individuals watching PBS are instead more likely to support legalization. Ideological self-selection into different news programs plays an important role, but cannot entirely explain the correlation between media exposure and attitudes about illegal immigration.


Introduction
Immigration is one of the most salient policy issues in the United States. Gallup polls conducted in May 2006 andMay 2007 show that immigration was considered the second most important problem facing the country, with respectively 13 and 24 percent of respondents mentioning it. As the number of undocumented immigrants has increased (Passel 2005), much of the recent discussion has focused on illegal immigration. While a small literature is emerging which tries to measure the costs and benefits of illegal immigration (Hanson 2006), little is known on the factors that influence individual attitudes towards this facet of globalization. The purpose of this paper is to carry out what, to the best of our knowledge, is the first, national-level, systematic study of the economic and non economic determinants of public opinion on illegal immigration, and in particular on the legalization of undocumented aliens. 1 To undertake our analysis, we use the newly released Cooperative Congressional Election Study (CCES), 2 an individual-level survey conducted immediately before and after the 2006 midterm elections. At that time, two legislative initiatives on illegal immigration were being considered in the US Congress. The bill discussed in the House (H.R. 4437) focused on border enforcement and deportation of illegal immigrants. The Senate proposal (S. 2611) contained instead a more complex set of initiatives. Besides calling for increased security, it expanded the number of guest workers and, importantly, it introduced a path to legalization for undocumented immigrants.
The CCES contains two question which are particularly suited for the purpose of our study. The first one asks each respondent to state whether he/she would have voted for the Senate plan. The second one, which was asked only to a subsample of the population, directly compares the House and the Senate proposals. We use answers to these questions to assess the views of each respondent on illegal immigration. In analyzing the drivers of individual attitudes, we focus our attention on two main channels: the economic channel, which works through the labor market and the welfare state, and the non economic channel, with a particular emphasis on the role played by the media. The CCES is unique 1 Several papers have in fact focused on specific questions related to illegal immigration. Hood III and Morris (2000), Newton (2000) and Tam and Cain (2001) for instance have looked at the determinants of support for proposition 187 in California, which limited the access enjoyed by illegal immigrants to a series of welfare state benefits.
for our purpose as, besides providing a wealth of information on the socio-economic characteristics of the respondents, it also contains information on the TV evening news program they most frequently watch (ABC World News, CBS Evening News, CNN, Fox News, PBS The NewsHour, NBC Nightly News, MSNBC and others). To investigate the role played by the actual content of the newscast, we have supplemented this information with content coverage data obtained from the Dow Jones Factiva online archive.
To carry out our empirical analysis, we estimate a series of probit regressions. In our main specification, the dependent variable is a pro-Senate plan dummy, which is constructed from answers to the first policy question described above. Controlling for other economic and non-economic drivers, we find that media exposure plays an important role in shaping public opinion on illegal immigration. According to our estimates, respondents watching Fox News are 9 percentage points more likely to oppose the lenient Senate plan (relative to CBS viewers). This result is broadly consistent with the findings of recent studies on the political position and the persuasive role of Fox News (Groseclose andMilyo 2005, DellaVigna andKaplan 2007). Perhaps surprisingly, we find that CNN viewers entertain an attitude towards illegal immigration which is very similar to the one displayed by Fox News viewers. This result might be explained by a 'Lou Dobbs' effect. 3 Instead, the opposite effect holds for PBS viewers, who are 14 percentage points more likely, relative to CBS viewers, to support the Senate plan. It is worthwhile to notice that ideological self-selection into different news channels plays a relevant role, but does not completely explain the correlation between media exposure and policy attitudes on illegal immigration. In fact, when controlling for self-reported ideology and party identification, the estimated effect for Fox News and PBS is significantly smaller in absolute value than when not controlling for these variables, but still large and highly significant. In addition, and most importantly, the estimated effect for CNN is larger in absolute value when controlling for self-reported ideology and party identification. This means that, in the case of CNN, the impact of the news program works in the opposite direction with respect to self-selection and, thus, the coefficient on CNN is if anything biased towards zero due to self-selection. This result gives us confidence that our correlations are at least in part driven by the causal impact of media exposure. In addition, we also find that the count of stories covering migration in a given evening news program has a negative and significant impact on the propensity to support the Senate plan: This result is also unlikely to be driven by self-selection.
When forming their opinion about a given issue, individuals are also influenced by the real world events pertaining to that issue (Behr and Iyengar 1985). Our findings are consistent with this framework: controlling for media coverage and other confounding factors -e.g. the state-level fraction of legal immigrants -individuals living in states with a larger fraction of illegal immigrants are significantly more favorable to the Senate plan.
The estimated correlation is quantitatively significant: an additional one percentage point in the fraction of illegal immigrants is associated with a one percentage point increase in the propensity to support the Senate plan.
Turning to economic drivers, we find that individuals are more favorable to the Senate's 'more lenient' plan if they are more educated and richer. To better interpret these results, we also run the same regressions using -as the dependent variable -the dummy constructed from answers to the second policy question, which explicitly asks respondents to compare the Senate and the House plan. Our results in the two sets of specifications are very similar. This suggests that -when answering the question about the Senate plan -respondents have in mind the House Plan as the alternative.
Our finding that more educated individuals are more favorable to the Senate plan is in line with the existing literature on the drivers of attitudes towards overall migration.
Compared to the House plan, the Senate plan increases the labor market competition faced by unskilled workers, both by being more lenient on deportations as well as by legalizing illegal workers, thus broadening their employability across sectors. Since illegal immigrants are mostly unskilled, both these effects increase the relative supply of unskilled labor. 4 The positive impact of individual skill is then consistent with the existence of labor market complementarities, which have been shown to be important drivers of attitudes by Scheve and Slaughter (2001), Mayda (2006) and O' Rourke and Sinnott (2006). 5 We also find that the welfare state plays an important role in shaping attitudes towards legalization. In particular, richer individuals are more likely to support the Senate plan.
A plausible explanation of this result is that both poor and rich individuals feel penalized by the Senate plan through the welfare state channel -since they perceive legalization as worsening the position of public finances. However, the poor are more affected than the rich, because they expect benefits to fall, either in quantity or quality, and they use public services relatively more.
The remainder of the paper is structured as follows. In section 2 we review the related literature, while in section 3 we provide some background information about illegal immigration in the United States and the two legislative proposals discussed by the U.S. Congress during 2006. In section 4 we describe the data. Results of our statistical analysis are discussed in section 5, while section 6 is devoted to a series of robustness checks. Section 7 concludes.

Related literature
This paper is related to two strands of the literature. The first set of papers looks at the impact of the media on individual preferences. The second set considers instead the drivers of individual attitudes towards (overall ) immigration.
A vast body of literature, both in economics and political science, has analyzed the impact of the media on policy preferences, but not specifically on migration preferences.
In particular, several studies have focused on the persuasion role that the mass media can play through their "agenda-setting power", according to which the amount of coverage devoted to a particular issue can influence the importance readers and viewers attach to it (McCombs and Shaw 1972). In addition, as noted by McCombs (2002), the media not only can make an issue more salient by increasing the amount of coverage, but it can also emphasize particular attributes of the issue. The theory of issue priming describes how readers and viewers, when assessing a given situation or individual, are pushed towards giving a higher weight to the aspect emphasized by the mass media. 6 Recent analyses -taking as given and known the ideological position of a media outlet -have tried to quantify these effects by exploiting experimental or quasi-experimental settings on the degree to which individuals are exposed to a given outlet. 7 DellaVigna and Kaplan (2007) find that the gradual introduction of Fox News in cable markets has increased the Republican vote share in presidential elections between 1996 and 2000.
Using an experimental setting, Gerber et al. (2008)  extending the analysis to a multi-country framework, Mayda (2006) andO'Rourke andSinnott (2006) find that the sign of country-specific correlations -between individual skill and attitudes -depends on the relative skill composition of natives to immigrants in each country. In particular, consistent with the labor-market predictions of economic models, the data shows evidence of a positive correlation between individual skill and pro-migration attitudes in countries that receive unskilled migrants on average, and of a negative correlation for countries that receive skilled migrants on average. 8 A second important economic factor driving attitudes is the size and type of adjustment of the welfare state. In the presence of a redistributive welfare state, immigration can have a sizeable impact on the post-tax income of natives, since it affects the amount of benefits they receive and the tax bill they pay. The net effect for a given individual depends on the relative position of the individual in the ex ante income distribution, on the relative skill-mix of natives as compared to immigrants and on the mechanism through which the welfare state adjusts to the presence of immigrants. Empirically, Hanson et al. (2007), Facchini and Mayda (2009) and Dustmann and Preston (2007) all find strong evidence of the importance of this channel -for public opinion on overall immigration -using variation in the extent of redistribution carried out across US states as well as across advanced Western economies.
Immigration attitudes are also influenced by cultural factors. As pointed out by Dustmann and Preston (2007), 'differences' between the immigrant and the native populations might induce a more negative appraisal of the immigration phenomenon, to the extent that natives fear a dilution of nation-specific characteristics and/or have a preference for cultural (and ethnic) homogeneity. The results for Great Britain obtained by Dustmann and Preston (2007) suggest in fact that racist feelings have a particularly strong effect on attitudes towards overall migration.
Summing up, a substantial literature has investigated the drivers of public opinion on overall immigration. At the same time, little is known on what shapes attitudes towards illegal immigration, even though this phenomenon is at the forefront of the political 8 Using the 1992 and 1994 ANES surveys, Citrin et al. (1997) shows that the statistical significance of labor market factors in explaining individual migration attitudes is not robust to the inclusion of other attitudinal factors as regressors. However, these attitudinal regressors (e.g., the individual opinion about whether Hispanic and Asian immigrants might "take away jobs from people already here") can be considered as intermediate outcomes with respect to labor market treatment and, as a result, it is not surprising that the latter loses statistical significance when the former are included in the specification.
debate. The purpose of this paper is to fill this gap.

Illegal immigration in the United States
Illegal immigration refers to labor movements across national borders taking place in a way that violates the immigration laws of the destination country. There are many possible avenues through which an individual might become an illegal immigrant. Citizens from nations which do not have automatic visa waiver agreements, or who would not qualify otherwise for a visa, often enter a destination country by crossing the border without inspection (illegally). Individuals might also become illegal immigrants by simply overstaying the period of legal permanence in the country. Similarly, unauthorized immigrants who have applied for asylum or Temporary Protected Status are considered to be unauthorized residents. Finally, a third possible channel is represented by fraudulent marriages, through which a foreign citizen marries a national of the destination country, with the only purpose of obtaining a residence permit.
Assessing the number of illegal immigrants residing in the United States involves some educated guessing. It is well known that illegal aliens do respond to government surveys such as the Current Population Survey or the decennial Census. While the CPS and Census do not ask explicitly whether the foreign born is legally or illegally present in the country, a wide range of research institutes and the Office of Immigration Statistics (OIS) within the Department of Homeland Security have constructed estimates of the number of illegal immigrants, for example based on the socio-economic characteristics available in the CPS or Census data. The most common method to estimate the number of illegal immigrants is to take the difference between the measured immigrant population and the sum of past legal immigration flows. Estimates obtained using this methodology vary substantially 9 but, as of January 2006, the OIS reports that there were 11.6 million unauthorized immigrants in the United States. Of these, 4.2 millions had entered in 2000 or later, and close to sixty percent of the total number of illegal immigrants was from Mexico.
The number of illegals has been steadily increasing from the early nineties until today. Many estimates suggest that, between 1995 and 2005, the inflows of unauthorized migrants -at over seven hundred thousand per year -have actually been larger than those of legal arrivals (Passel 2005). Importantly, the distribution of illegals is highly concentrated. According to the OIS estimates, the ten states which were the largest receipients of undocumented immigrants accounted for approximately three quarters of the total, and California and Texas alone had more than four million illegal aliens in 2006. This is not surprising, as the two states share a border with Mexico and, at least since 1990, the vast majority of immigrants from that country are illegally living in the United States (Passel 2005).
To understand the impact of illegal immigrants on domestic residents, it is important to analyze the socio-economic characteristics of these individuals. Recent estimates by the Pew Hispanic Center (Passel 2005) suggest that the following stylized facts hold: 1) Unauthorized immigrants are younger than both natives and legal immigrants; 2) they are substantially less educated than both natives and legal immigrants; 3) they work in lower wage occupations, and they earn substantially less than natives and legal migrants in the same occupations; 4) they are concentrated in a relatively small number of industries; 10 5) Poverty rates among illegals are particularly high; 6) Well over fifty percent of illegal immigrants do not have health insurance.
These stylized facts suggest that attitudes towards illegals will be shaped by labor market considerations and also -even more so than for legals -by the working of the welfare state. Since illegals are often perceived as not paying their fair share into the government's coffers, their participation in the welfare state has been the subject of a heated debate in the public arena. In 1994, California voters supported by a wide margin the introduction of Proposition 187 -also known as the "Save our State initiative" -which was directed at excluding illegal immigrants from access to a large set of welfare programs, including public education for their children. The proposition was later overturned and rescinded by a federal court, but similar initiatives have been proposed in Illinois, Florida, New York and Texas. Illegal immigrants' access to the welfare state has also been reduced by the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996, in particular by excluding them from federal child care and federal low-income heating assistance (Zimmermann and Tumlin 1999). The question of illegal immigrants' access to the welfare state still continues to be at the center stage of the political debate.
There have been several calls for draconian interventions. As representative Ron Paul (R, Texas) has suggested in 2005: "We must end welfare state subsidies for illegal immigrants.
Some illegal immigrants -certainly not all -receive housing subsidies, food stamps, free medical care, and other forms of welfare. This alienates taxpayers and breeds suspicion of immigrants, even though the majority of them work very hard." Besides anecdotal evidence, very little systematic information exists on the actual contributions and benefits made and received, respectively, by illegal immigrants: In the few existing studies, it has been documented that a substantial share of illegals are employed "off the books" 11 -and 10 For instance, 26% of the workers in landscaping services and 20% of those in meat/poultry packing are unauthorized. According to Passel (2005), in ten industries illegal immigrants represent more than ten percent of the overall labor force.
11 See North and Houstoun (1976) and Passel and Clark (1997). In particular, in their study of New therefore are less likely to pay taxes -and that illegals are likely to be net beneficiaries of the welfare state. In addition, and most importantly for this paper, little is known on the fiscal effects -relative to the status quo -of the introduction of a legalization measure. 12 As a response to the rapid increase in the number of illegal immigrants in the last  Table 1.
While neither bill became law because they failed to pass the conference committee, they have been important catalysts for the immigration debate of this period. Thus, York state, Passel and Clark (1997) assume a 60% compliance rate but, as agriculture is much less important in New York than in the other major illegal migrants' destinations, this assumption is likely to be substantially upward biased. 12 For one of the few studies on this matter, see Camarota (2005). While there is little doubt that legalization will lead to an increase in tax payments by the migrants -relative to the status quo -an important question is whether legalization will lead to a big increase in welfare program participation rates and on which agency the burden of financing the additional cost will fall (i.e. the state or the federal government).
understanding the drivers of individual views on these measures will provide us with valuable information on how illegal immigration is perceived.

The data
The Cooperative Congressional Election Study (CCES) is a stratified cross-sectional survey, conducted around the 2006 midterm elections by Polimetrix. A crucial feature of the CCES project is that interviews are conducted on-line, rather than on the phone or live, and great attention has been paid to insure that the sample is representative. 13 The CCES data originates from the joint effort of researchers in thirty U.S. universities. As a cooperative project, it is characterized by a modular structure, i.e. some questions are asked to the entire set of 36, 421 respondents (the so called Common Content), while other questions (the Team Content) are specific to subsets of respondents and are "purchased" by individual research groups in the project. 14 The main dependent variable in our empirical analysis is the answer to a question about whether the respondent is in favor of the Senate plan, which offered a path to legalization and citizenship to illegal immigrants. The exact wording is as follows: "Another issue is illegal immigration. One plan considered by the Senate would offer illegal immigrants who already live in the U.S. more opportunities to become legal citizens.
Some politicians argue that people who have worked hard in jobs that the economy depends on should be offered the chance to live here legally. Other politicians argue that the plan is an amnesty that rewards people who have broken the law. What do you think?
If you were faced with this decision, would you vote for or against this proposal?" We create a pro lenient plan dummy variable which equals one if the respondent answers he would have voted for the Senate proposal, and zero if he/she would have turned it down.
We exclude from the analysis those who have answered 'Don't know'.
There is an additional question in the CCES regarding the respondent's likely vote on alternative illegal immigration proposals. In this case, the respondent is asked to give his comparative opinion about the Senate and the House plans. The exact wording of the question is as follows: "Congress has been debating different policies concerning immigration reform. The Senate proposal has a path to citizenship for illegal immigrants. The House proposal, on the other hand, contains stricter enforcement and deportation of undocumented aliens. Which of these two items of reform do you think is more important?" We create a pro lenient plan2 dummy variable which equals one if the respondent answers he would have voted for the Senate proposal, and zero if he/she would have voted for the House plan. We therefore exclude from the analysis those who have answered 'Don't know'. Since this question was asked to a subsample of 16, 231 respondents (i.e., not to the entire sample, but only to those who were interviewed in the pilot study, held in August 2006), we rely on this question solely for robustness checks of the main results.
Regarding media exposure, respondents are first asked to state the frequency with which they watched a national evening news program in the week before the interview.
For those who watched a news program at least once, there is an additional question, which asks them to mention the most frequently watched network for national evening news. 15 We create separate dummies for each most frequently watched network and an additional dummy for those who did not watch a national evening news program (the no evening news dummy).
The CCES survey also contains information on the age, gender, education, household income, employment status, immigration status and political views of respondents. Education is coded according to a 1 − 6 ordered scale, with a value of one for those who have not completed high school, and six for those with a post-graduate degree. Individuals are asked to classify their own income along a discrete scale of 14 income "brackets", with interval size increasing from $5, 000 to $30, 000. 16 Previous research Slaughter 2001, Mayda 2006) shows that the correlation between education and immigration attitudes is a function of labor force participation. Building on a question on employment status, we thus construct a dummy variable which equals one when the respondent belongs to the labor force, i.e. he/she is working full time or part time or is unemployed, and zero otherwise.
Another question in the CCES survey provides information on the immigration status of the respondent. We use it for two different purposes: first, we exclude from the statistical analysis those who declare to be immigrants; second, for those who are U.S. citizens and hence are included in our analysis, we use the question in order to extract some information on the family origin of the respondent. More specifically, we create a discrete variable (immigrant origin) on a 1-3 ordered scale, with a value of 1 for those whose parents and grandparents are U.S. citizens, a value of 2 for those whose parents are U.S. citizens, but at least one grandparent is an immigrant, and a value of 3 for those who declare that at least one of their parents is foreign-born.
Regarding political controls, the CCES survey has a question on self-reported ideology: individuals are asked to locate themselves on an ideological 0 − 100 scale, ranging from zero for extremely liberal views to 100 for extremely conservative ones. Moreover, there are two variables measuring the party identification of the respondent. The first variable is a 3-point scale party id variable, which equals minus one for self-identified Democrats, zero for Independents and one for Republicans. The second variable is a 7-point scale indicator, ranging from 1 for strong Democrats to 7 for strong Republicans.
One might consider religion as an organized system of beliefs which could be systematically correlated with attitudes regarding policy issues. Among others, the CCES survey contains a question about the religious preferences of the respondent, and one about the frequency of church attendance. We recode the church attendance variable as an ordinal one, which ranges from a value of zero for those who declare to go to church "never or almost never" to a value of 3 for those who (declare to) go at least once a week or more. Moreover, the survey also includes a question on the perceived importance of religion in everyday life. In section 6, we explore whether our baseline results are robust to controlling for the religious attitudes of respondents.
Whether the respondent lives in an urban, suburban or rural area might be correlated with the type and frequency of encounters with illegal immigrants and hence with his/her views on the issue. Also, it can be correlated with other unobserved features of his/her political views. Since respondents are not asked about their location in terms of urban, suburban or rural area but are asked about their county of residence, we match the CCES data with county level information, which are taken from the ICPSR County Characteristics File. 17 In particular, for each county we calculate a measure of population

Center tabulations of the Census Bureau's 2005 American Community Survey. 18
Summary statistics of these variables are shown in Table 2, separately for individual, county and state level variables. The large majority of respondents is against the Senate Plan. Regarding the most frequently watched national evening news program, Fox News obtains the highest ratings, with NBC and CNN ranking second and third, respectively.
Age, education, household income are positively skewed, and this is also the case for the conservative ideology score.
At the county level, population density, the unemployment and the crime rate are all positively skewed, and show a sizeable degree of cross-sectional variation. Regarding state-level variables, the estimated fraction of illegal immigrants displays some nonnegligibile variation as well, ranging from a quarter of a percentage point for West Virginia to around seven percentage points for California. This also applies to the fraction of legal immigrants, which ranges from about a fifth of a percentage point for Mississippi to almost 20 percentage points again for California.

Empirical Results
In this section, we empirically analyze the individual-level propensity to support the Senate plan on illegal immigration as a function of respondents' characteristics. We run a set of probit regressions with the pro lenient plan dummy as the dependent variable, excluding all individuals who describe themselves as immigrants. The regression output is displayed in Tables 3 and 5 Tables A1 and A2). We find that the estimates of the marginal effects of most variables in Tables A1 and A2 are very similar to the   corresponding estimates in Tables 3 and 5. 20 As a result, we conclude that the alternative respondents have in mind, as they evaluate the Senate plan, is the House plan. However, we will also discuss the interpretation of the results if, instead, respondents have in mind the status quo as the alternative.

The impact of media exposure on individual attitudes
In Table 3 we report our baseline results. We start with a specification where we omit ideological and political party controls (regression [1]). Next, in column [2], we introduce the conservative ideology score and the 3-point party id variable. In regression [3], we replace the 3-point party id variable with a 7-point one and we include the county-specific measure of population density as well as the state-specific measures of exposure to illegal and legal immigration. In columns [4] and [5] we introduce state and designated market area (DMA) fixed effects, respectively. 21 In the first five columns of Table 3 we use dummy variables for each media channel, with CBS viewers as the omitted category.
In order to analyze the role played by media exposure -in particular in relation to self-selection issues -it is crucial to compare the estimates with and without ideological controls, i.e. column 22 This result is in fact consistent with the Groseclose and Milyo (2005) analysis of think tank quotes, which places Fox News' Special Report with Brit Hume significantly to the right with respect to CBS Evening News. the extent that liberal individuals appreciate PBS and the more lenient Senate plan, the marginal effect of the PBS dummy is biased upwards when not controlling for ideology and party id.
On the other hand, those watching CNN are about 5 percentage points less likely to favor the Senate plan when not controlling for ideology and between 8 and 9 percentage points less likely when doing so. Prima facie, it is perhaps surprising that those watching CNN are systematically less likely to support the Senate plan than CBS viewers, but this result can be explained in light of what in the introduction we have dubbed the "Lou Dobbs effect". 23 It is the case that Lou Dobbs, the anchor and managing editor of CNN evening news, has always been very vocal regarding the costs imposed by illegal immigration on the American public, and has consistently opposed the Senate bill proposal. For example, his official website features a specific section on illegal immigration called "Broken Borders". 24 This is how he describes the issues at stake: "The single most critical issue to protect our nation is the securing of our borders and our ports. Every day, tens of thousands of containers enter our country from other nations and they are never inspected. At the same time, our government turns a blind eye to the thousands of people who illegally cross our borders. These scenarios exist because corporate America has convinced our leaders that this is one of the best ways to remain competitive." When the Senate passed the Comprehensive Immigration Reform Act, Lou Dobbs introduced the story with the following words: "Tonight, the Senate has just voted for a so-called comprehensive immigration reform bill. The vote, . The legislation gives amnesty to millions of illegal aliens and sharply escalates the war on our middle class and raises the cost of federal government substantially". 25 From the point of view of ideological self-selection, the bias towards zero of the CNN dummy variable in regression [1] -induced by omitting ideological controls -suggests that CNN viewers are more liberal than CBS viewers (the omitted category). In addition, and most importantly, the result for CNN -that the impact of the news program works in the opposite direction with respect to self-selection -gives us confidence that our correlations are at least in part driven by the causal impact of media exposure. In other words, if anything, the coefficient on CNN represents a lower bound of the causal impact of CNN on attitudes towards illegal immigrants.
It is also interesting to notice that the marginal effects of media exposure are remarkably stable once we start controlling for ideological and partisan preferences. As mentioned above, beginning with column [3] we replace the 3-point party id variable with the 23 See also Akdenizli, Dionne, Kaplan, Rosenstiel, and Suro (2008 Focusing on the other media channels, individuals watching NBC, MSNBC or other networks appear not to be significantly different from CBS viewers in their propensity to support the more lenient immigration plan, while viewers of ABC News are systematically less likely to favor the Senate plan than CBS viewers, with an estimated marginal effect between 3 and 4 percentage points. Finally, respondents declaring not to watch any national evening news program are significantly less likely than CBS viewers to be in favor of the Senate plan. Overall, the evidence we have gathered is consistent with the fact that self-selection explains a sizeable portion of the correlation between policy attitudes on illegal immigration and media exposure, but not the whole of it: persuasion, as best exemplified by the Lou Dobbs effect, appears to be a non-negligible factor. Since our estimates suggest that Corr(Viewer's Cons. Id., Attitudes) < 0, then if δ 1 > δ 5 it must be true that OV B > 0 and therefore that Corr (Viewer's Cons. Id., TV Channel) < 0. In other words, the TV channel reveals itself to be more liberal than the CBS benchmark, based on the preferences of viewers who choose it.
The outcome of this exercise is illustrated in Figure 1 where -for each TV channelwe plot the estimated marginal effect -on the propensity to support the Senate planagainst its overall ideological position. More precisely, the former is taken from column [5] in Table 3, while the latter is calculated as the difference in the estimated marginal effect of each channel dummy when moving from specification [1] to specification [5], again in Table 3.The relationship appears to be positive, since more liberal channels like PBS are associated with a more positive effect on the propensity to support the Senate plan, and vice versa for a conservative channel like Fox News. But it is especially interesting to focus on the relative position of CNN, which is close to CBS and NBC from the point of view of the overall ideological position (as revealed by its viewers), but has a negative effect on immigration attitudes, the size of which is comparable to that of Fox News.
In fact, when including CNN the relationship between the overall ideological position of each TV channel and its persuasive effect on immigration attitudes is positive but not statistically significant, while it is significant at the 10 percent level when excluding CNN.
One can also notice that the so called "Big Three", i.e. the oldest U.S. networks (ABC, CBS and NBC), are closely clustered in the scatter plot, while PBS, the publicly financed channel, displays the most liberal stance on both dimensions.
As mentioned in the introduction, the exposure to real world events is likely to affect individual attitudes, or at least to be a significant predictor thereof. At the county level, population density is positively and significantly correlated with the propensity to support the Senate plan: column [3] in Table 3 shows that an increase of density of 10, 000 individuals per square mile is associated with a 3 percentage points increase in the probability of being in favor of the more lenient immigration plan. This result is consistent with the idea that living in an urban area increases the frequency of encounters with illegal immigrants and hence positively affects the respondent's views on the issue.
Thus, this finding is consistent with the so called "intergroup contact theory" (Allport 1954). According to this theory, as developed within the social psychology field, the interaction itself between different racial groups can reduce intergroup prejudice and foster more friendly attitudes. Also, living in an urban area is likely to be correlated with other unobserved features of the respondent's political views.
In column [3] we also control for the state-specific fraction of illegal immigrants: we find that respondents living in states with a higher fraction of illegal immigrants are significantly more likely to support the more lenient Senate plan. The size of this correlation is quite large, as a one percentage point increase in the fraction of illegals is associated with a 0.7 percentage points increase in the propensity to support the Senate plan. This result is again consistent with "intergroup contact theory". One might be concerned that the partial correlation between attitudes and the state-specific fraction of illegal immigrants is picking up some omitted variable at the state level, e.g. exposure to legal immigrants. Thus, we also control for the state-specific fraction of legal immigrants and find that it is not significantly correlated with the dependent variable. Columns [4] -[5] replicate the specification featured in column [3] but with the addition of state and DMA fixed effects, respectively: the estimated marginal effects are remarkably robust to this change.
In the second part of Table 3 we further investigate the estimated correlations of immigration attitudes with the respondent's favorite evening news broadcast. In particular, we check whether the way each news broadcast covers illegal immigration helps explain the previously estimated marginal effects on the TV channel dummies. As discussed in section 2, both the overall amount and type of media coverage devoted to a given issue could matter. In our case we focus on overall coverage, since it can be easily measured in a replicable fashion. The Dow Jones Factiva online archive allows to search the transcripts of evening news programs on ABC, CBS, CNN, Fox News, PBS and NBC. We proxy the total amount of coverage to which respondents are exposed -before being interviewed during the 2006 midterm campaign -with the broadcast-specific number of stories featuring the words "immigration" or "immigrants" between January 2006 and October 2006. We do the same with the search terms "illegal immigration" or "illegal immigrants", and we compute the difference between those two counts, in order to obtain a measure of immigration coverage that does not mention illegal immigration. 27 Table 4 reports this media coverage data for each of the six TV channels. It is interesting to notice that Fox News and CNN devoted the largest amount of coverage to immigration overall, while PBS and CBS devoted the least. However, when looking at illegal immigration, PBS dedicated much less coverage than CBS, while CNN gave more coverage than Fox.
In the second part of Table 3, for comparison purposes, we start with a baseline specification (column [6]) which includes only the respondents mentioning the six channels as favorite ones -thus we exclude respondents who do not watch a national evening news program -but we do not control for media coverage variables. In column [7] we add as regressors the channel-specific count of stories mentioning illegal immigration and the channel-specific count of stories mentioning immigration but not the illegal aspect of it.
There is a very strong and negative correlation between those measures of media coverage and the propensity to support the Senate plan. In terms of magnitudes, ten additional illegal immigration stories during the January/October 2006 period are associated with a one percentage point decrease in the probability of supporting the more lenient plan 27 A word of caution regarding Fox News: Since the transcripts for its evening news broadcast are not available, we instead look at the O'Reilly Factor. on immigration. The size of the effect for stories mentioning immigration but not illegal one is the same. All other estimated marginal probabilities are practically unaffected by this cardinal specification of the media exposure variable.
To conclude, this last result -i.e., the sizeable and significant impact of the number of immigration-related stories on the propensity to support the Senate plan -is especially important because it provides additional evidence that the media effects we are estimating are not driven only by self-selection. As a matter of fact, while it is plausible that individuals self-select into evening news programs based on the broad ideological position of the channel, it is less likely that self-selection works on the basis of the count of stories covering immigration.

Labor-market and welfare state determinants of individual attitudes
Besides studying the role played by the media, the regressions reported in Table 3 allow us to analyze the labor-market and welfare-state determinants of attitudes in favor of legalization, accounting for a number of socio-demographic individual-level controls.
First, we find that the impact of age on the propensity to support the Senate plan is characterized by a U-shaped relationship, as shown by the negative marginal effect of age and the positive marginal effect of age squared. The estimated minimum in the propensity to support the Senate plan occurs at an age between 55 (column [2]) and 58 (column [1]). In other words, as they approach midlife young respondents become more opposed to the legalization plan, while older respondents become more in favor as they move towards retirement. A plausible interpretation of the positive marginal effect for elderly individuals is that illegal immigrants -who would be legalized under the Senate plan while they would be deported under the House plan -offer services which are mostly consumed by the old, for example elderly care and landscaping services.
Moreover, females appear to be significantly more supportive of the Senate plan only when not controlling for ideology (column [1]). In fact, once we control for self-reported ideology and party id in the following regressions, we do not find evidence of a gender effect on pro-legalization attitudes. This result can be easily explained by the fact that women are on average more liberal than men so that -when not controlling for ideology -the ideology effect on illegal immigration attitudes is absorbed by the female dummy.
A similar interpretation can be given to the marginal effect of the Black dummy variable, which is positive and significant in column [1] and becomes insignificant in the following regressions.
Not surprisingly, being a Latino has a positive and significant impact on pro-legalization attitudes, controlling or not for ideology and party id. Our estimates also show that individuals are more willing to support the Senate plan the more recent the immigration status of their family (as captured by immigrant origin). The size of this effect is relatively large: A one-point increase in the immigrant origin variable is associated with approximately a 2 percentage points increase in the propensity to support the lenient immigration plan.
Respondents are significantly more likely to support the Senate plan, the higher their Suppose instead that respondents are comparing the Senate plan to the status quo, rather than to the House plan. In this case, the legalization of illegal immigrants has two effects: it allows previously illegal immigrants to access occupations and industries which were until then precluded to them; and it eliminates the "unfair labor market competition" of illegal immigrants visà vis unskilled natives -in other words, once illegal immigrants enter the formal market, they need to be paid the minimum wage, enjoy higher employment protection, receive unemployment benefits, etc.. To the extent 28 In general, the labor market competition hypothesis predicts that the level of individual skill should be positively correlated with pro-immigration preferences in countries where immigrants increase the labor market competition for unskilled natives (for example, when immigrants are unskilled) and negatively correlated in countries where immigrants increase the labor market competition for skilled natives (for example, when immigrants are skilled). See Figure 2.
29 Remember that Tables A1 and A2 seem to suggest that, when answering the question, respondents are comparing the Senate plan to the House plan. that the first effect more than offsets the second one -which is plausible and consistent with the literature 30 -the Senate plan will still be perceived as increasing the labor market competition faced by domestic unskilled workers.
In Appendix Table A3, we split the sample according to whether the respondent belongs to the labor force or not. 31 We find that the estimate of the marginal effect of education is systematically larger for those who belong to the labor force than for those who do not. This evidence confirms the above labor-market interpretation of the marginal effect of education. In addition, the impact of education on legalization attitudes is positive and significant for individuals out of the labor force as well. Thus, our results are also consistent with a non economic interpretation of the effect of education, according to which the education variable is to some extent capturing the cultural openness of the respondent (see, for example, Hainmueller and Hiscox 2007). Thus, both non economic and labor-market considerations shape the impact of education on public opinion on the Senate plan.
The impact of household income gives us information about the welfare state determinants of pro-legalization attitudes. Theory highlights that a redistributive welfare system might adjust to immigration through changes in the tax burden or in the generosity of benefits (Facchini and Mayda 2009). In both frameworks what is crucial is whether migration improves or worsens the fiscal stance of the welfare state: unskilled migrants are expected to be a burden, while the opposite is true for skilled migrants. In addition, depending on which adjustment takes place, immigration will give rise to opposite income-distribution effects through the welfare-state channel. If the level of per capita benefits is kept constant and changes occur on the tax side of the budget (tax adjustment model ), unskilled immigration has a larger negative impact on individuals at the top of the income distribution. The reason is that increases in the tax rate mostly affect rich individuals, whose tax payments represent a larger fraction of their net income. Similarly, skilled immigration has a larger positive impact on rich individuals. 32 On the other hand, if tax rates remain constant and the adjustment takes place on the benefit side (benefit adjustment model ), the income-distribution effects of migration are completely reversed.
Low-income individuals are more negatively affected if migration is unskilled and more positively affected if migration is skilled. The reason is that per capita benefits represent a larger fraction of a poor individual's net income. 33 30 See Kossoudji and Cobb-Clark (2002) and Orrenius and Zavodny (2007). 31 It is worthwhile to notice that students are excluded from both subsamples. 32 Thus, in the tax adjustment model, individual income is negatively correlated with pro-immigration preferences in countries where immigrants represent a net burden (for example, when immigrants are unskilled) and positively correlated otherwise (for example, when immigrants are skilled). See Figure 3. 33 Thus, in the benefit adjustment model, individual income is positively correlated with pro-immigration preferences in countries where immigrants represent a net burden (for example, when immigrants are If respondents are comparing the Senate plan to the House plan -as Tables A1 and A2 seem to suggest -they will necessarily perceive legalization as worsening the fiscal position of the welfare state. The reason is that illegal immigrants are unskilled, therefore legalizing them as opposed to deporting them will worsen the government's budget. Therefore, the positive impact of household income in Table 3 is consistent with the benefit adjustment model : Both poor and rich individuals feel penalized by the Senate plan through the welfare state channel -since they perceive legalization as worsening the position of public finances -but the poor more so than the rich, because they use public services relatively more (see right panel of Figure 4). This result -in favor of the benefit adjustment model -is not consistent with the existing evidence on attitudes towards overall immigration (Hanson et al. 2007, Hanson 2005, Facchini and Mayda 2009). Thus, it appears that respondents have different perceptions of the adjustment of the welfare state in reaction to legal vs. illegal immigration. One possible explanation is that, in the case of legal migration, authorities do take into account the higher number of participants in the welfare state (i.e. migrants) and, therefore, adjust the welfare system accordingly. On the other hand, in the case of illegal migration, it is likely that there is no (quantitative) adjustment of the welfare state, i.e. the increase in the number of participants is simply ignored. Although more people now access the welfare system, tax rates do not change nor benefits -for example, schools, hospitals, etc. -vary in number. However, the quality of benefits changes, due to overcrowding brought about by the arrival of illegal immigrants. Since the poor are larger beneficiaries of welfare benefits, they are hurt the most by illegal immigrants through the welfare state channel.
On the other hand we could assume that, in assessing the Senate plan, respondents have in mind the status quo as the alternative. As pointed out above, the positive impact of household income is consistent with the tax (benefit) adjustment model as long as legalization is perceived to reduce (increase) the burden of migration on the welfare state (see left (right) panel of Figure 3 (Figure 4)). Most existing estimates suggest that only a fraction of illegal immigrants pay taxes (see Camarota 2005): this issue has been repeatedly highlighted in the public debate. Legalization will imply that tax compliance by legalized illegal immigrants will increase and this will substantially raise the government's tax revenues. 34 However, it is not clear ex-ante whether the increase in tax revenues -brought about by legalization -will more than offset the increase in public services used by legalized illegal immigrants. Therefore, on net, legalization might unskilled) and negatively correlated otherwise (for example, when immigrants are skilled). See Figure 4. 34 One must also take into account that, relative to the status quo, the Senate plan's proposed legalization is likely to shift the burden of financing some of the welfare programs from state to federal government, thus reducing the fiscal burden xxx felt by each respondent. either reduce or increase the burden of migration on public finances, relative to the status quo. This implies that, if respondents have in mind the status quo as the alternative, our estimates could be consistent with both the tax adjustment model and the benefit adjustment model.

Robustness checks
In this section we perform a series of robustness checks of our baseline results. The   Table 3 plus a control for whether the respondent owns a pick-up truck: as discussed in section 4, this variable might simultaneously capture whether the individual does not live in a city and something about his lifestyle. We find that owning a pick up truck is associated with a 5.5 percentage points reduction in the probability to support the Senate plan.
As thoroughly discussed by Guiso et al. (2003), individual attitudes might be systematically correlated with the type and intensity of religious beliefs being held. Regarding the specific topic of immigration, Guiso et al. (2003) show that individuals interviewed within the World Values Survey (WVS) are significantly more intolerant towards immigrants if they were raised religiously and they declare to be currently religious, while being actively religious is not significantly correlated with immigration attitudes. Moreover, Guiso et al. (2003) show that Catholics, Protestants and Muslims are more likely to be intolerant towards immigrants than agnostic individuals, while Buddhists display the opposite tendency. Table 5 we exploit the CCES questions tapping into the religious beliefs and habits of respondents. More specifically, we include dummy variables for the respondent declaring to be Protestant, Catholic, Jewish, Muslim, of another religion or of another Christian religion, keeping as the omitted category those who declare not to have any religious preference. We also control for each individual's church attendance frequency and for the importance he/she attaches to religion in everyday life. Self-declared Protestants and Catholics are around 6 percentage points less likely to support the Senate plan on illegal immigration, as compared to individuals with no religious preference. The correlation is of the same sign and larger in magnitude (around 8 percentage points) for those who declare to belong to another Christian religion. The level of significance is one percent for all three marginal effects. On the other hand, the propensity to favor the more lenient plan on illegal immigration is significantly and positively correlated with church attendance. 35 Finally, the religion importance dummy is not significantly correlated with the respondent's opinion on the illegal immigration plan.

In columns [2] and [3] of
From this point of view, our results are reasonably in line with those obtained by Guiso et al.: we similarly find that immigration attitudes are significantly more negative for Catholics and Protestants, however we find no significant effect for Muslims. This could be due to the lower number of Muslims in the U.S., as compared to the cross-country WVS sample. Moreover, differently from Guiso et al., we find that church attendance has a mitigating effect on negative immigration attitudes. Of course one should handle this comparison with some further caution, since (i) as mentioned above, our sample is a U.S. one, (ii) we are concerned with illegal immigration, rather than with the immigration phenomenon as a whole, and (iii) we use as the dependent variable a policy-related question, instead of a direct question about attitudes.
An additional concern regarding our results is the possibility that they are partially driven by omitted variables at the state or county level. In Table 3 we have already shown that our findings on individual-level variables are robust to the inclusion of state and DMA fixed effects as well as to controlling for population density at the county level.
In column [3] of Table 5 we further control for the county-specific 2004 crime rate and 2005 unemployment rate: the rationale for this is that individuals might be particularly sensitive to their economic and social environment when they are asked to think about a policy proposal regarding a potentially very contentious issue.
We find a positive, statistically significant and quantitatively non negligible correlation between the county-level crime rate and the propensity to support the Senate plan: a one percentage point increase in the crime rate (i.e., an additional crime event for every 100 inhabitants) is associated with about a one percentage point increase in the probability of favoring the more lenient plan on immigration. Conditionally on all the caveats regarding the distinction between causation and correlation, this result is consistent with the hypothesis that individuals on average believe that the illegal status of immigrants is conducive to a higher propensity to criminal behavior, and that legalization might help break this vicious link. On the other hand, we find no significant impact of the unemployment rate. It is also worthwhile to notice that, in columns [1] and [2], there is a mildly significant and negative relationship between the propensity to support the Senate plan 35 When distinguishing according to the religious faith of the respondent, further results -available upon request -show that the church attendance variable is statistically significant only for Protestants, Catholics and individuals belonging to another Christian obedience. and the state-specific fraction of legal immigrants. 36 It is still the case that the fraction of illegal immigrants is positively and significantly correlated with the pro lenient plan dummy, mildly so in column [3].
In the second part of  Table 3). When doing so, we do not find significant departures from the results in Table 3 and -regarding the robustness checks themselves -from the those displayed in the first part of Table 5.

Discussion and conclusions
Illegal immigration is at the center of the national debate in the majority of destination countries of immigrant flows. Several factors influence voters' views regarding this phenomenon, ranging from economic to cultural ones. To the best of our knowledge, this paper represents the first systematic empirical analysis of the determinants of public opinion on illegal immigration, and in particular on the legalization of undocumented aliens. Our main results suggest that respondents are more favorable to the more-lenient Senate plan on illegal immigration if they are more educated, more liberal and richer. Individuals watching Fox News or CNN are much more likely to be against the Senate plan than those who watch CBS, while PBS viewers are characterized by the opposite effect. Finally, individuals living in states with larger fractions of illegal immigrants are more favorable to the legalization of illegal immigrants.
Concerning the role of economic drivers of immigration attitudes, our analysis suggests that both labor market and welfare state drivers play an important role. We find that more skilled individuals are more likely to support the Senate plan, a result which is in line with the labor market complementarity hypothesis. Furthermore, we show that richer individuals are more likely to support the Senate plan. Under the reasonable hypothesis that respondents have in mind the House plan as the alternative to the Senate plan, our results are consistent with the benefit adjustment model. As discussed in section 5.2, a likely explanation for this result is that, in the case of illegal migration, there is little or no quantitative adjustment of the welfare state. As a consequence of this, the quality of benefits drops because of overcrowding effects, which particularly hurt the poor, since they are larger beneficiaries of such services. Future research should shed light on which adjustment takes place in reality, beyond individuals' perceptions, both in the case of legal and illegal immigration.
We conclude the paper by re-emphasizing a crucial issue in the empirical analysis. 36 The relationship is no longer significant in column [3].
While we have controlled for a number of potential omitted variable biases, we cannot claim we estimated a causal effect of media exposure on public opinion on illegal immigrants. In other words, to the extent that we do not have a clean natural experiment, although very suggestive, our estimates must be interpreted with due caution. As repeatedly pointed out, the crucial confounding factor is self-selection: individuals likely choose to expose themselves to the media outlet whose ideological position is closest to theirs.
In fact, when not controlling for self-reported ideology and party id, the estimated partial correlation between illegal immigration attitudes and the Fox News dummy is much larger in absolute value than when we add those controls. The same is true for the PBS dummy, albeit with the opposite sign. On the other hand, consistently with the fact that Lou Dobbs entertains a very negative (and influential) position about illegal immigration on a TV channel which is overall left leaning, the estimated marginal effect of CNN viewership is negative but smaller in absolute value when not controlling for ideological factors at the individual level. This last result gives us confidence that our findings are not driven uniquely by self-selection, but rather that our correlations are at least in part driven by the causal impact of media exposure. In addition, we also find that the count of stories covering migration in a given evening news program has a negative and significant impact on the propensity to support the Senate plan: This result is unlikely to be driven by self-selection. The exact reason why readers and viewers might self-select themselves to ideologically consonant media outlets in turn depends on whether they consume news in order to acquire information (Calvert 1985 andShapiro 2006), or to confirm their prior beliefs (Mullainathan and Shleifer 2005). One important direction of future research is to look for sources of exogenous variation in respondents' viewership of different channels to better disentangle and quantify those two competing hypotheses, i.e. the influence hypothesis, according to which the media has a causal impact on voters' attitudes, and the self-selection one, according to which individuals choose specific newspapers and media channels as a function of their ex-ante preferences.

Legalization of Undocumented Immigrants
Would give illegal immigrants who have lived in the United States for two years or more a path to eventual citizenship. Illegal immigrants who have been here less than two years would be required to leave the country altogether. They could apply for the guest worker program, but they would not be guaranteed acceptance in it.
No provisions for legalization, although a conservative leader in the House, Representative Mike Pence, proposed a separate bill that would allow illegal immigrants to become guest workers, but not permanent residents or citizens.

Temporary Worker Program
Creates a guest worker program with a path to legal permanent residence.
In December, the House defied President Bush's call for a guest worker program although the separate bill recently introduced by Mike Pence, the leader of the conservative caucus in the House, would allow illegal immigrants to become guest workers.

Number of Guest Workers to be Admitted Annually
Negotiations in the Senate bill have reduced the number of foreign guest workers to be admitted annually to 200,000 a year from 320,000.
No such provisions in the House bill.

Worksite Enforcement
The legislation would require employers to use an electronic employment verification system that would distinguish between legal and illegal workers.
Requires employers to participate in an electronic employment eligibility verification system within three to six years.

Criminal Penalties for Existing Illegal Immigrants
Mandates penalties for smuggling aliens, but offers exceptions for those who provide "humanitarian" assistance to immigrants, including medical care and housing. Also, illegal immigrants convicted of a felony or three misdemeanors would be deported.
Makes it a federal crime to live in the United States illegally. Individuals who help illegal immigrants to enter or stay in the country would also face criminal penalties.

Table 1-Details about the bills passed by the Senate and the House (source: New York Times) Border Security: Fencing
The bill initially called for limited "double-or triple-layered fencing" but as the debate progressed, the Senate added provisions for 350 miles of border fencing and 500 miles of vehicle barriers between the United States and Mexico.
Requires the construction of "at least two layers of reinforced fencing" as well as "physical barriers, roads, lighting, cameras and sensors" along approximately 700 miles of the U.S.-Mexico border.

Border Security: Personnel
Increases the number of Border Patrol agents by 2,400 each year through 2011 to the current force of 11,300 agents.
Hires more Border Patrol agents "as expeditiously as possible." Nearly 12,000 Border Patrol agents currently stand guard. Hires at least 250 active duty port of entry inspectors for each of the next three years.

Fraction of legal immigrants (state level)
Notes: The sample excludes all individuals who describe themselves as immigrants. The dependent variable is a dummy which takes on the value of one if the respondent would have voted in favor of the Senate plan on illegal immigration, and zero otherwise. The panel contains the estimated marginal effect on the probability of being in favor of the Senate plan, given an increase in the value of the relevant regressor, holding all other regressors at their mean values. * Significant at 1%; ** significant at 5%; *** significant at 1%. Income is coded according to a 1-14 ordered scale of income brackets. Education is coded according to a 1-6 ordered scale, ranging from a value of 1 from high school dropouts to 6 for those holding a post-graduate degree. The immigrant origin variable is coded as follows: 1, parents and grandparents were born in the US; 2, parents were born in the US but at least one grandparent is an immigrant; 3, at least one parent is an immigrant.The conservative ideology variable is coded on a 0-100 scale, with 0 for extremely liberal views and 100 for extremely conservative ones. The 3-point scale variable for party identificationtakes on a value of -1 for a Democrat, 0 for an Independent and 1 for a Republican. The 7-point scale party id variable takes on values on the 1-7 range, with 1 for a strong Democrat to 7 for a strong Republican. Each of the network dummy equals one when the respondent declares that he most frequently watches the evening news broadcast on network x, and zero otherwise. CBS is the excluded network. The "no evening news" dummy equals one when the respodent declares that in the past week he never watched a national evening news program, and zero otherwise. The count of stories on illegal immigration on the favorite evening news broadcast is the number of stories in 2005 where the expressions "illegal immigration" or "illegal immigrants" appear.      Standard errors, adjusted for clustering at the state level, are shown under each coefficient. * Significant at 1%; ** significant at 5%; *** significant at 1%. See Table 4 for the definition of the other variables.  Overall ideological position Fig. 1: Estimated ideological position of each TV channel and effect on illegal immigration attitudes Notes: the graph displays the estimated marginal effect on the propensity to support the Senate plan for each TV channel against its overall ideological position. The former is taken from column [5] in Table 3. The latter is a measure of liberal ideology and is calculated as the difference in the estimated marginal effect of each channel dummy when moving from specification [1] to specification [5], i.e. when including ideological and party id controls. See section 5.1 for additional details.