Leading into the 2018 US legislative elections, the Republican caucus settled on a unified strategy for keeping their House majority: talk about Nancy Pelosi.Footnote 1 While Pelosi's 25 per cent approval rating marked her as even less popular than President Trump, incumbents focusing on vilifying a national opponent instead of their personal records of success as legislators is out of step with much of the literature on legislative elections. For someone reading Fenno (Reference Fenno1978) – who defines legislative home-styles by the careful cultivation of a non-partisan ‘personal’ vote – contemporary legislative elections that feature substantially lower rates of both incumbency advantage and split-ticket voting would be hard to recognize.
I show that this change in politics, broadly defined as ‘nationalization’, is partly a function of a changing media environment. The politics of the past – where citizens would more regularly cast split-ticket ballots in order to vote for their incumbent member of Congress based on a record of good service – was made possible by a robust local media. The expansion of the communication environment has eroded these local sources of information and replaced them with a more national information environment. In the language of Zaller (Reference Zaller1992), voters receive a set of ‘considerations’ from this new communication environment that are more national than local. In line with other recent work that has examined the effect of a changing media environment on politics, I show that this new set of considerations has fundamentally altered the politics of congressional representation (Darr and Dunaway Reference Darr and Dunaway2017; Darr, Hitt and Dunaway Reference Darr, Hitt and Dunaway2018; Hayes and Lawless Reference Hayes and Lawless2018; Peterson Reference Peterson2017; Van Aelst et al. Reference Van Aelst2017). Using new data on the roll-out of broadband internet to all congressional districts over four elections to identify the effects of these changes, I leverage a within-district design to show that increasing broadband leads to less split-ticket voting. As a result, these districts have a lower incumbency advantage.
Split-ticket voting and the incumbency advantage are inherently linked. Measured in the traditional way (that is, Gelman and King Reference Gelman and King1990), the incumbency advantage measures the additional votes a legislator receives from running as an incumbent versus running in an open district above and beyond what is expected of that legislator based on the district's partisanship. As such, anything that causes individuals to cast straight tickets will also lead to a decline in the incumbency advantage. For example, the 1972 election is a high-water mark for the incumbency advantage as many districts voted overwhelmingly to re-elect Republican President Richard Nixon while also voting to re-elect their Democratic members of Congress – evidence that these voters had considerations beyond pure partisanship.
The incumbency advantage occurs as a consequence of voters having these additional local considerations in mind when entering the voting booth – considerations that are received through their information environment. When more people in an area get access to broadband internet, the types of considerations available change in fundamental ways. This happens through two processes. First, access to broadband has important ramifications for local media, which in the past has provided the voting considerations vital to the incumbency advantage. Internet use negatively impacts the provision of local information directly through declining subscriptions, but also indirectly via a weaker product that has fewer resources available for political coverage. Secondly, when individuals read news online they are far more likely to view news about national politics compared to local politics, both as a consequence of their own choices and the architecture of the internet. These two processes work in tandem to generate voters who are much more likely than in the past to have considerations about national, as opposed to local, politics. As a consequence, they are less likely to split their tickets, and therefore produce a smaller advantage for incumbents. In the first section of the article I discuss in more detail the accumulated evidence across disciplines for each of these steps in the causal chain.
To identify the effect of broadband on the nationalization of congressional elections, I merge Federal Communications Commission (FCC) data on broadband internet availability with election results in all congressional districts from the 2002–2008 period. I use within-congressional-district models to estimate the effect of the roll-out of broadband internet on both the incumbency advantage and partisan voting. I estimate that a 100 per cent increase in the number of broadband internet providers in a district decreases an incumbent's electoral advantage by 2.7 percentage points. This decline in the incumbency advantage seems to be due to an increase in the power of national partisan politics in House elections: the electoral margins of incumbents facing re-election are increasingly affected by ‘down-ballot’ effects from presidential elections in districts with higher levels of broadband internet connectivity. Citizens in districts with more access to broadband internet are also less likely to punish their incumbent for excessively partisan roll-call voting. Taken together, these results provide evidence that contemporary politics – in which both Republicans and Democrats approach re-election with a message focused on vilifying the most hated national figures in the opposite party – is, at least in part, a function of this changing media environment.
By focusing on changes to the aggregate information environment instead of individual self-reports, this article avoids the well-known pitfalls of self-reported exposure to the media, which often serves as little more than a proxy for political interest (Bartels Reference Bartels1993; Prior Reference Prior2009a; Prior Reference Prior2009b). Using unique data on the geographic roll-out of broadband internet, I am instead able to focus on the real-world effects of a change in communication environment on election results. The within-district design I employ rules out many potential confounders, such as broadband internet being endogenous to higher-income areas, as well as concerns that broadband internet was introduced at the same time as other contemporaneous media changes. Further, it is unlikely that these results are due to reverse causality, as this would involve broadband providers making distribution decisions based on changing levels of polarization in particular districts.Footnote 2
I do not claim that broadband is the sole cause of polarization and a decreased incumbency advantage in American politics. Politics has been becoming more nationalized in America for decades (Hopkins Reference Hopkins2018; Jacobson Reference Jacobson2015a), and the causes range from elite polarization (Hetherington Reference Hetherington2001) to geographic sorting (Sussell Reference Sussell2013). One major contributor to increasing nationalization, however, is thought to be a changing media system (Hopkins Reference Hopkins2018), which has also been becoming more nationalized since at least the 1980s (Clinton and Enamorado Reference Clinton and Enamorado2014; Prior Reference Prior2007). As I discuss below, a key component of this media nationalization is the expansion of broadband internet, which was perhaps the most significant and rapid change during this period. This rapid (and geographically uneven) change provides an opportunity to identify the relationship between these two time series that are changing in parallel. My identification strategy permits the observation of a plausibly exogenous change in the communication environment. Similar to a field experiment, the reaction to ‘treatment’ is important inasmuch as it helps us understand how similar changes – from the introduction of dial-up internet to the introduction of smartphones – have impacted, and will continue to impact, the nationalization of politics. Broadband is not the whole story, but understanding the effects of this colossal shift in how Americans learn about politics is vital to understanding trends in politics since that shift, and what is likely to happen as communication technology continues to evolve along the same track.
These data and results are important because they highlight a crucial shift in accountability and representation. The incumbency advantage is not just a statistical regularity, but a representation of the role of local information in legislative elections. The increase in straight-ticket voting, and the resulting decline in incumbency advantage, points to a shift in how representatives are held accountable; a shift from a model of legislators appealing directly to their constituencies to a version of accountability that is much more reliant on ‘responsible party government’ (Schattschneider Reference Schattschneider1942). While reducing all voting decisions to one plane of partisan conflict makes choices easier for voters, it also complicates the lines of accountability in a system of checks and balances. In Federalist 51, Madison states that essential to preserving liberty is a system in which ‘ambition [is made to] counteract ambition’ (Madison Reference Madison1982[Reference Madison1788]). The differing re-election constituencies and demands on members of Congress and the president are thought to generate two different sets of ambition that allow government to control itself. Evidence that an expanding communication environment leads to voters who are less able to separate their votes across their ballots generates concerns that House members and the president increasingly serve the same ambition.
Ultimately, the results I identify serve as a warning to those hoping these new communication media will democratize the flow of information, allowing thousands of small websites producing local information about politics to thrive (Hindman Reference Hindman2008, 2–3). It does not seem to be the case that increased access to broadband internet helps individuals ‘get information’ about local politics. Instead, the pattern of results presented here makes it clear that the move online simply makes it easier for citizens to ‘get in formation’.
The Role of Media in the (De)Nationalization of Voting
Whether political behavior is ‘nationalized’ or not can be simplified to a basic question: What considerations does an individual have in mind when making their voting decisions (Zaller Reference Zaller1992)? Do voters have a great deal of information about the local incumbent member of Congress, their policies and their performance? Or, do they instead have considerations about the policies of national parties, national political figures like the president and affective feelings of antipathy towards members of the other party? In this latter case, voting behavior will be more nationalized: political decisions will be increasingly made on a single plane of (partisan) conflict.
If the types of considerations held by voters is what is important in producing nationalized (or localized) voting outcomes, it is important to look at how a changing media environment may influence the considerations voters receive. Studies of media ‘priming’ have consistently shown that topics made salient by an information environment will be easily accessible at the ‘top of the head’ when voters evaluate candidates (Iyengar and Kinder Reference Iyengar and Kinder1987). For example, when foreign policy is more salient in the news media, individuals place greater weight on their foreign policy attitudes when evaluating politicians (Iyengar and Simon Reference Iyengar and Simon1993). This theory of political decision making is a natural outgrowth of psychological models of memory-based information processing, in which attitude formation stems from ‘the ease in which instances or associations [can] be brought to mind’ (Tversky and Kahneman Reference Tversky and Kahneman1973, 208; Scheufele and Tewksbury Reference Scheufele and Tewksbury2006). In a way similar to how the media drives the importance of foreign policy, a media environment that provides and prioritizes local information will produce voters who have local considerations at the ‘top of the head’ when making voting decisions. However, a media environment that provides and prioritizes information about national politics to voters – information which almost exclusively centers around conflict between the two major parties (Arceneaux and Johnson Reference Arceneaux, Johnson, Thurber and Yoshinaka2015) – will generate individuals who will instead base their voting decisions on national-partisan considerations (Clinton and Enamorado Reference Clinton and Enamorado2014).
I provide evidence in this section that broadband and similar technologies create an information environment that tilts the balance of information from local to national through two complimentary routes. First, broadband draws viewers away from local news sources, which has the subsequent effect of causing a decline in the quality of local journalism. Secondly, when individuals read news online, they are more likely to view information about national-partisan politics than local politics. The end result is citizens who go to the voting booth with fewer local considerations about Congress, and more national-partisan considerations.
The prominence of the incumbency advantage in the study of congressional elections is testament to the degree to which voting behavior in the past has been driven by considerations above and beyond simple national-partisan considerations. The traditionally measured incumbency advantage (see, for example, Gelman and King Reference Gelman and King1990) is defined as the additional votes an incumbent receives above and beyond what they would have received if they were running in an open seat, controlling for how their party performed in the election overall, and the two-party presidential vote in their district.Footnote 3 Figure 1 displays the incumbency advantage from 1960 to 2014 calculated using this equation. Incumbents gained 5–15 percentage points compared to those running in open seats during this period, though this advantage has clearly declined in all regions in recent decades.

Figure 1. Incumbency advantage, 1960–2014
The key variable in the equation that makes the incumbency advantage an indicator for national versus local voting is the inclusion of the District Presidential Vote. The incumbency advantage crucially does not include the advantage of being in a district that is filled with co-partisans. In order for incumbents to have an advantage over those running in open seats, they must regularly exceed the presidential vote for their party in their district. Put another way, if every individual in a district votes a straight ticket, then there could not, by definition, be an advantage to running as an incumbent (as any individual with the same party affiliation would receive the same percent of the vote regardless of whether they were an incumbent). This means that the existence of the incumbency advantage is necessarily evidence that local, non-partisan, considerations are driving the election of members.
What are these local considerations? Scholars believe there are both direct and indirect components (Cox and Katz Reference Cox and Katz1996; Levitt and Wolfram Reference Levitt and Wolfram1997). The direct component includes advantages like name recognition and the ability to use one's position in Congress to bring projects to the district;Footnote 4 the indirect component includes incumbents' tendency to be of higher quality on average, and their ability to scare off high-quality challengers. A third, more temporally specific, component of the incumbency advantage is the slow progress of the Southern re-alignment. In much of the late twentieth century, Southerners regularly voted for Republican presidential candidates while also supporting their segregationist Democratic members of the House of Representatives (Bartels Reference Bartels2000). Regardless of the reason, voters prioritizing any of these components is necessarily evidence of orthogonal, non-partisan, information driving voting decision making. This is true whether that information is received directly, such as voters rewarding an incumbent for good performance; indirectly, for instance if voters know and care about the quality differential between the incumbent and challenger; or due to Southern re-alignment, where state Democratic parties in the South were out of step with the national Democratic Party.
Voters receive these additional considerations largely from local news media. Empirical evidence to this effect is abundant: multiple studies show that local newspaper and television coverage predicts local political knowledge, support for incumbents and voters who rely less on partisan labels when making voting decisions (Arnold Reference Arnold2013; Peterson Reference Peterson2017; Prior Reference Prior2006; Schaffner Reference Schaffner2006; Snyder and Strömberg Reference Snyder and Strömberg2008). Snyder and Strömberg (Reference Snyder and Strömberg2008), for example, use nearly 20 years of newspaper data to show that having a quality newspaper in overlap with a congressional district strongly drives individuals’ ability to identify and evaluate their incumbent House member, and ultimately leads to less partisan voting outcomes. This effect is not limited to local newspapers: Prior (Reference Prior2006) found that local television coverage in the mid-century made an important contribution to the rise of the incumbency advantage via increased citizen knowledge of their representatives.
While voters could potentially receive information about their local candidates from national media, with 435 members of the House of Representatives, adequate coverage of all individual members is next to impossible. Indeed, Cook (Reference Cook1989) finds that in a typical year only 39 per cent of House members appeared on a network newscast. Legislators (and challengers) can, and do, communicate directly with their constituents, but the local media – which serves a larger audience than the politician can reach directly, and who additionally offer the reputational benefit of a neutral third-party observer – offer the most in terms of localizing political decision making. The advantages gained from local news media attention were made clear by at least one incumbent in Fenno's Homestyle, who remarked that the most ‘profitable thing politically’ of the day was not any personal interaction with constituents, but the presence of the local newspaper at a press conference (Fenno Reference Fenno1978, 205).
The incumbency advantage requires voters to possess and prioritize local information. The empirical record is clear that when individuals have greater access to robust local media, they gain independent, non-partisan, knowledge about their incumbents, and ultimately weigh that information against partisanship when making voting decisions. Examining the trends in Figure 1, the notion that the local media plays an important role in promoting the incumbency advantage seems plausible. The period when the incumbency advantage was the highest was also when the United States had a robust and independent local media infrastructure that could provide voters with the sort of non-partisan local considerations that supported this advantage. The incumbency advantage declined – both in the South and the rest of the country – as the media system became more nationalized (Prior Reference Prior2007). As the media environment became less conducive to the provision of local, non-partisan, information about legislators – including information about a record of good service, overall quality, or loyalty to the segregationist South – the incumbency advantage declined.
The introduction of broadband represents a particularly large disruption to this relationship between voters, the local media and their incumbents. This disruption presents an opportunity to illustrate the causality between more nationalized media and more nationalized voting. Broadband has disrupted this relationship in two ways. First, it has had significant effects on the local media itself. Secondly, the types of information voters view and have access to online is fundamentally different from offline news.
There is strong evidence that the use of the internet has had negative effects on local media use – and as a consequence – local media itself. The aggregate trends away from local news sources are clear: in the mid-1990s 71 per cent of Pew respondents reported regularly reading local daily newspapers and 72 per cent reported regularly viewing the local television news. In 2008 these figures dropped to 54 and 52 per cent, respectively (Kohut Reference Kohut2008). Using panel data, Hopkins (Reference Hopkins2018) shows that these broad trends in media use away from local sources towards national sources like cable television and the internet are not simply the consequence of generational replacement, but rather due to all Americans shifting away from traditional media. Other researchers have confirmed that internet use displaces traditional media use. Respondents report that accessing news online better satisfies their need for variety and convenience, leading them to reduce their newspaper readership (De Waal and Schoenbach Reference De Waal and Schoenbach2010; Gaskins and Jerit Reference Gaskins and Jerit2012; Ha and Fang Reference Ha and Fang2012). As such, the first step in the causal chain between increased access to broadband and a reduced incumbency advantage is that those who have access to broadband are simply less likely to access local media.
This decline in local media viewership due to broadband has the knock-on effect of eroding the profitability and quality of local media. The FCC's 2011 report on the information needs of local communities (Waldman Reference Waldman2011) lays out a litany of effects of how increased access to high-speed internet has decimated the local news economy. Newspaper revenue dropped 47 per cent from 2005 to 2009 and staffs have shrunk 25 per cent since 2006. While the 2005–2009 period did see online traffic for local newspapers balloon from 43 million to 70 million users a month, the $716 million in additional revenue generated online pales in comparison to the $22.6 billion in advertising losses to the print side (Waldman Reference Waldman2011, 17). Reporters are stretched thin, making it harder to report on complicated topics. As a result, topics like education, health care and government are reported on less, at the expense of topics like weather and crime (Waldman Reference Waldman2011, 13).Footnote 5
This erosion of local news readership and quality has appreciable effects on congressional elections. Just as there is a great deal of evidence that areas better served by a robust local news environment have citizens who are more knowledgeable about congressional elections and vote in a less partisan manner, districts with faster-eroding local news environments (for example, from newspapers closing, see Darr, Hitt and Dunaway Reference Darr, Hitt and Dunaway2018) have citizens who are less knowledgeable about their members of Congress (Hayes and Lawless Reference Hayes and Lawless2015; Hayes and Lawless Reference Hayes and Lawless2018), and are less likely to split their tickets across different levels of office (Darr, Hitt and Dunaway Reference Darr, Hitt and Dunaway2018). When local newspapers leave town, citizens lose exactly the sort of knowledge and behavior necessary to sustain an incumbency advantage.
The first important effect of broadband is its impacts on the local news environment. However, individuals are not simply reading less local news, but shifting their consumption to online sources. Therefore, it is necessary to know what sort of political news individuals view online. For example, if they access a great deal of local information online, this would mitigate the concern that the expansion of broadband will lead voters to place greater weight on national political attitudes. The evidence presented below, however, suggests that this is not the case.
The second disrupting effect of broadband is that using the internet to access news heavily skews the information citizens receive towards the national level. Hindman (Reference Hindman2011) and Tewksbury (Reference Tewksbury2003) have conducted large-scale web-tracking studies that look at the balance of national and local news that is consumed online. Using comScore data on millions of users, Hindman (Reference Hindman2011) estimates that of the time spent reading news online, only 15 per cent is spent on local sites, compared to 85 per cent on national sites. Tewksbury (Reference Tewksbury2003) similarly tracked the consumption patterns of online news readers from a representative sample provided by Nielsen/NetRatings, and found stories classified as ‘national’ in focus made up 10 per cent of all views, compared to 1.2 per cent for stories classified as ‘local’.Footnote 6
Attention to different types of websites is not a measure of what information individuals actually consume. However, a necessary condition for online news readership to provide the sort of information needed to sustain the incumbency advantage is that individuals at least visit websites covering a local angle of politics. Instead, the evidence suggests that when citizens go online their news consumption becomes more heavily skewed towards national content. This uneven distribution of national and local news online seems to be due to both demand and supply.
In terms of demand, individuals seem to gravitate to more national, and more partisan, news when given the choice. Hopkins (Reference Hopkins2018) uses an experiment to show voters are more likely to self-select stories about the president compared to those about governors or mayors. Lelkes, Sood and Iyengar (Reference Lelkes, Sood and Iyengar2015) find that individuals with broadband internet were more than twice as likely to visit national-partisan websites like the Drudge Report and Huffington Post compared to those with dial-up internet. Both of these tendencies suggest that a move online would produce both more national, and more partisan, considerations. This seems to clearly predict an increase in the propensity to read national-partisan websites, but even increased readership of non-partisan national news ought to increase partisan considerations, as US national news of all types is known to cover politics as a conflict between the two main parties (Arceneaux and Johnson Reference Arceneaux, Johnson, Thurber and Yoshinaka2015).
Another feature of online news that may cause a lower incumbency advantage is selective exposure to partisan-consistent information, often referred to as ‘echo chambers’ or ‘filter bubbles’ (Pariser Reference Pariser2011; Sunstein Reference Sunstein2001). This selective exposure may cause individuals to have stronger partisan attitudes, and therefore to vote in more partisan-consistent ways across different levels of government (Lelkes, Sood and Iyengar Reference Lelkes, Sood and Iyengar2015). More recent behavioral evidence, however, has cast doubt on whether concerns about selective exposure generalize beyond a set of highly motivated news consumers (Barberá et al. Reference Barberá2015; Gentzkow and Shapiro Reference Gentzkow and Shapiro2011; Guess Reference Guess2016). Regardless, the relationship between online use and nationalized politics does not require selective exposure to operate. Even a heterogeneous media diet that primarily focuses on national politics would cause individuals to access those national attitudes when making voting decisions, thus lowering the probability of split-ticket voting. If, instead, selective exposure is a widespread phenomenon and individuals become more polarized after exposure to online news, then the relationships described here will be heightened.
Even if individuals did not have differential preferences for national vs. local news, the online environment is still tilted towards national content due to supply. The architecture of the internet prioritizes large national content providers and lowers the probability of local news garnering attention. Hindman (Reference Hindman2008) uses web crawlers to determine the linking structure of websites. Websites with more in-links are more likely to be accessed, both directly through those links and, perhaps more importantly, due to Google using in-links as its primary method of ranking search results. Hindman finds that the distribution of in-links in all categories of websites approximates the ‘power law’, whereby the top websites on a given topic have exponentially more in-links than smaller sites. As such, naive searches for political information overwhelmingly lead consumers to national information – not to more specific local coverage of the same issues. This is true across political topics. One could imagine a world in which the distribution of news about Congress is more decentralized than news about the presidency – a constellation of high-quality local blogs and news sources providing in-depth coverage of local members – but this is not the case. Hindman shows that the link structures of websites covering Congress vs. the presidency are equally concentrated among large national producers.
The collective evidence suggests that when broadband internet is expanded, the balance of information consumed by individuals tilts towards the national and the partisan. Those with access to broadband are less likely to view local news media. This has negative effects for their reception of local considerations, but also harms the community at large by eroding the resources available to local journalism. On top of these changes, when news consumers move online, the content that they both choose and have access to is predominantly about national partisan politics.
Combined, these characteristics result in voters having – and giving the most weight to – national politics when deciding how to vote. In the remainder of the article I test three expectations arising from this literature.
The first-order question is whether the incumbency advantage rises or falls in areas exposed to broadband internet. I expect congressional districts with greater broadband availability to have a lower incumbency advantage than those with lower levels of broadband.
The second question is whether House incumbents are losing their advantage due to an increase in national forces in elections. I anticipate that incumbents in districts with greater broadband penetration will be more susceptible to national partisan forces. The availability of local information about politics has historically allowed voters to separate their votes for president from their votes for Congress. In an environment dominated by national information about politics, this separation is less likely. I therefore expect voting for members of Congress to be more strongly predicted by voting for president in areas with higher levels of broadband.
The third question is whether voters are decreasingly likely to punish incumbents for excessive partisanship. It has historically been assumed that constituents will not re-elect legislators who display excessive fealty to their parties. ‘There is no member of either house’, reports Mayhew (Reference Mayhew1974, 99), ‘who would not be politically injured…by being made to toe a party line on all policies’. Indeed, Carson et al. (Reference Carson2010) demonstrates a strong negative link between party unity voting and re-election margins. This relationship, however, similarly relies on voters having – and caring about – local information over national-partisan concerns. However, evidence that the expansion of broadband causes voters to be more affectively polarized (Lelkes, Sood and Iyengar Reference Lelkes, Sood and Iyengar2015) suggests that voters may increasingly reward members for being good partisan warriors. The expectation is therefore that voters with broadband access will be decreasingly likely to punish their incumbents for excessive partisanship; indeed, they may reward them.
Roll-Out of Broadband Internet
The main independent variable used in this article to determine the effect of an increasingly dense information environment on political nationalization is the roll-out of broadband internet providers to congressional districts. I show in Appendix Section 4 using individual-level data from the Current Population Survey Internet Supplement that the expansion of broadband providers strongly predicts the probability that individuals will have a home broadband subscription. More broadly, this measure has the advantage of frequent over-time measurement and being unrelated to districts’ time-varying political characteristics.Footnote 7
The measure of broadband internet providers comes from the FCC, which collects bi-annual data from all broadband providers operating in the United States on where they have deployed service. From 1999–2008 the FCC tabulated these results into data listing the number of broadband providersFootnote 8 in each zip code in the United States, which I transformed into congressional district-year observations.Footnote 9 Figure 2 displays the levels and changes in the number of broadband providers in congressional districts over this period, and Figure 3 displays the geographic density of broadband across the mainland United States in the year 2002.

Figure 2. Broadband in House districts, 2000–2008

Figure 3. Geographic density of broadband providers in congressional districts, 2002
The number of broadband internet providers is an indirect measure of the degree to which people in an area have increased access to information. If a district has five broadband providers operating within it, some individuals may have access to one provider, others may have access to three or four, while some may have access to none. At the same time, for a given district an increase in the number of broadband providers can only mean either increased competition or access for a new set of customers.
This article is not the first to use broadband providers as an indicator of greater access to information, and previous research has done much to validate the measure. The number of broadband providers has been found to have the expected effect on the cost and quality of broadband available to an area. Wallsten and Mallahan (Reference Wallsten and Mallahan2010) analyzed a similar dataset and found that areas with multiple broadband providers had, on average, lower costs and higher internet speeds. Lelkes, Sood and Iyengar (Reference Lelkes, Sood and Iyengar2015) show that the number of providers and the number of subscribers correlate quite strongly at the cross-sectional county level, a result which has been validated in multiple other articles at differing levels of aggregation (Hitt and Tambe Reference Hitt and Tambe2007; Kolko Reference Kolko2010). In Appendix Section 4, I add further validation by matching these broadband data to individual-level data from three waves of the Current Population Survey Internet Supplement. Leveraging within-countyFootnote 10 variation, I show that a 100 per cent change in the number of broadband providers in an area is associated with an 11 per cent increase in the probability an individual has a home broadband subscription.Footnote 11 We can be confident, in other words, that in areas with more broadband providers, more individuals have internet access, at a lower cost, and with higher speeds. As stated above, Lelkes, Sood and Iyengar (Reference Lelkes, Sood and Iyengar2015) found that the number of broadband providers has a significant impact on online consumption behavior.
Measuring and Identifying the Incumbency Advantage
I merge these data (from Jacobson Reference Jacobson2015b) on broadband internet availability with bi-annual House of Representative election outcomes for the four elections from 2002–2008.Footnote 12 The key election variables are: % Democratic Vote for House, the Democratic share of the two-party vote for the House of Representatives in a given district-year; and % Democratic Vote for President, the Democratic share of the two party vote for president in a given district-year.Footnote 13 For all of the below specifications I omit cases where members ran unopposed.Footnote 14
The first hypothesis is that the incumbency advantage is expected to decline in response to expansions in the number of broadband internet providers. To estimate the incumbency advantage I use the Gelman and King (Reference Gelman and King1990) method,Footnote 15 which estimates the incumbency advantage by regressing % Democratic Vote for House on: an indicator for Incumbency, whether there is an incumbent Democrat (1), incumbent Republican (−1) or no incumbent (0); Party, whether the seat is held (at the time of the election) by a Democrat (1), Republican (−1) or is a new seat held by neither party (0); and % Democratic Vote for President, the Democratic share of the two-party vote for president in a given district-year. Gelman and King (Reference Gelman and King1990) show that the coefficient on Incumbency in such a setup is a consistent and unbiased estimator of the additional votes an incumbent receives over and above what that candidate would receive in an open seat.
To see whether this relationship alters depending on the number of broadband providers, I interact Incumbency with ln(Broadband Providers). The full equation for estimating the incumbency bias conditional on the information environment for district k in time periods t is:

β 2 is the incumbency advantage when a district has one broadband provider, and β 5 is the change in the incumbency advantage for every additional (logged) broadband provider. The expectation is that β 5 will be negative and significant, that is, the incumbency advantage decreases as an area receives additional broadband providers.
Importantly, this specification includes fixed effects for districts and years. Simply pooling together all district-years into a single regression would lead to bias in two important ways. First, certain districts – those that are richer and more populated – are likely to have more broadband providers. In a pooled regression, this non-random distribution would possibly bias the effect of broadband on electoral outcomes. Secondly, pooling together all district-years would cause bias due to endogeneity based on time, since broadband access was increasing in all areas during these elections. If this time trend is not accounted for, then anything also systematically changing over the period would be correlated.
Having repeated observations of the same units, however, offers a powerful solution to both of these problems. By including fixed effects for districts (α k) I omit all non-time-varying features of districts. That is, any fixed feature of a district is removed, meaning that variation comes from within-district changes in the number of broadband providers. Bias can no longer come from any feature of districts – such as urban or rural status – that does not change over time. By including fixed effects for years (α t) I omit all non-district-varying features of years. Most importantly this controls for the time trend, but also for different political outcomes in each year. Together, these two sets of fixed effects remove many of the possible sources of bias.
Any remaining sources of bias must be due to systematic differences in within-district changes in broadband internet that are above and beyond what would be expected due to the national growth of access to high-speed internet. To deal with such sources of bias, I control for two time-varying features of districts that could affect the results: (logged) population and (logged) median income.Footnote 16 I include these two simple economic controls for parsimony, but also show in Appendix Section 6 that all relationships are robust to including time-varying controls for the median age of the district, the percent of the district living in poverty, the percent of the district identifying as white and black, as well as the percent of the district that have bachelor degrees. Appendix Section 7 further shows that the results are robust to including no demographic covariates. Another possible source of bias is if certain types of politicians are able to encourage broadband internet to expand to their districts. I investigate this in Appendix Section 3 and find no evidence of this possibility.Footnote 17
While time fixed effects provide leverage by detrending the data such that the models are not driven by the time trend, they do not alleviate all potential sources of confounding variation based on time. Two problems in particular may cause problems: violations of the parallel-trends assumption, and changes in the effects of variables over time. Appendix Section 5 discusses each of these problems and provides three tests. I first test the most likely violation of the parallel-trends assumption: that rural and urban districts have different rates of broadband adoption that are related to rates of nationalization. The results show that this is not the case. I then provide two more general tests showing that the main results are not due to spurious time trends. First I show simply that the same relationships are found when examining these data using year-by-year cross-sectional models with state fixed effects. Secondly, I test this more generally using a placebo test: merging the broadband data with election results from the 1990s with the expectation that future changes in broadband in an area should be unrelated to past changes in nationalization. The results are null, lending credibility that the main models in the article are not driven by some unknown source of over-time spuriousness.
Does Broadband Decrease the Incumbency Advantage?
Table 1 examines whether the incumbency advantage is conditional on the communication environment. The first column estimates the incumbency advantage without introducing the communication environment. The coefficient on Incumbency indicates – as expected – that incumbents receive an electoral advantage compared to what would happen if they were running in an open seat. In line with cross-sectional estimations of the incumbency advantage during this period (see, for example, Jacobson Reference Jacobson2015a), incumbents are expected to receive an additional 6.35 percentage points more support compared to a candidate from their party facing the same electoral environment in an open seat. The second column of Table 1 examines this relationship conditional on the information environment. The coefficient on Incumbency now represents the impact of being an incumbent in a district with a low rate of broadband connectivity (one broadband provider); it is quite high at around 12 per cent. The coefficient on the interaction term represents how the incumbency bias changes as the number of broadband providers increases. As expected, this coefficient is negative, indicating that a 100 per cent change in the number of broadband providers decreases the incumbency advantage by around 2.7 percentage points.Footnote 18
Table 1. Effect of communication environment on incumbency advantage

Note: cluster-robust standard errors. Relevant hypothesis test in bold. Observations are contested House districts 2002–2008. *p < 0.05
This estimation supports the hypothesis that the incumbency advantage is greatly diminished by exposure to broadband internet. Incumbents’ ability to win a share of the vote greater than expected based on the partisan conditions in their districts relies, in part, on voters holding considerations about local conditions, whether those considerations are about the quality of the incumbent, or what that incumbent has done for the district (Arnold Reference Arnold2013; Prior Reference Prior2006; Schaffner Reference Schaffner2006; Snyder and Strömberg Reference Snyder and Strömberg2008). As broadband has eroded the local sources that provide those considerations, citizens go to the ballot box without the information needed to support the incumbency advantage. Areas with higher levels of broadband internet do not see the incumbency advantage attenuate to zero, but the reduction is substantively significant. Whereas an incumbent in a district with one broadband provider is expected to gain a nearly 12-percentage-point boost in support compared to a non-incumbent running in the same race, at the median level of providers this advantage is halved to 5.9 percentage points.
This result alone tells us that voters exposed to broadband internet are increasingly using a decision metric orthogonal to incumbency to make their voting decisions, not what that decision metric is. The next section investigates whether this new metric is individuals' national-partisan attitudes.
Measuring the Impact of Partisan Voting on Incumbents
The finding that voters increasingly using partisan cues to make their voting decisions is a strong candidate for explaining why increasing broadband decreases the incumbency advantage. Recall that the incumbency advantage requires at least some voters to cast split tickets: voting for an incumbent even though he or she comes from their out-party. If individuals increasingly vote in line with their partisan identities – a clear prediction from previous literature on the effects of broadband internet – this will reduce the incumbency advantage.
To examine whether increasing partisan voting is behind the relationship between broadband internet and a decreasing incumbency advantage I test two hypotheses: whether an incumbent's re-election numbers are increasingly influenced by presidential politics in areas with high broadband connectivity; and whether an incumbent's re-election numbers are decreasingly influenced by their partisan behavior in areas with high broadband connectivity. To answer these questions I confine the data to incumbents facing re-election, and examine the conditional impact of two variables: Same Party Presidential Vote, the presidential vote in an incumbent's district recoded to be in the direction of the incumbent's party; and Party Unity, the incumbent's VoteView Party Unity score in Congress prior to the election. This latter score calculates the percentage of roll-call votes in which a legislator voted with their party on ‘party votes’, votes where the majority of one party votes against the majority of the other party. The median member votes with their party 91 per cent of the time on such votes, while the standard deviation is 10 per cent.
There are clear unconditional expectations for both of these variables. A positive coefficient on Same Party Presidential Vote represents down-ballot effects. If a Republican member is running in a district that surges in support for a Republican presidential candidate, then that member is expected to do better. The larger the coefficient on Same Party Presidential Vote, the more ‘nationalized’ elections are: the dynamics of the presidential race increasingly impact House races. For Party Unity, existing research suggests that frequently voting with one's party is an electoral liability (Carson et al. Reference Carson2010). The expectation, therefore, is that the unconditional coefficient on this variable will be negative.
The main hypothesis is the degree to which these relationships are conditioned by the information environment. To test this, I estimate the following two specifications for each incumbent j in time period t.


In each, a positive coefficient on the interaction term is expected. The impact of national electoral forces (represented by Same Party Presidential Vote) is expected to increase as a legislator's district receives additional broadband providers. Further, the impact of a legislator's personal partisan conduct is expected to cease to be a liability as his or her district receives additional broadband providers. These two equations are estimated with both legislator and Year × Party fixed effects. Legislator fixed effects remove all non-time-varying features of legislators that could bias the results, such as their personal ideology. Year × Party fixed effects remove all non-member-varying features of election years, specific to each party. This controls for the time trend, and removes any source of potential bias related to the electoral fortunes of a member's party in a given year.
Does Broadband Increase the Impact of Partisan Voting on Election Results?
Table 2 displays how incumbent re-election margins are affected by national forces and the partisan conduct of members, first unconditionally, and then conditional on the information environment. Columns 1 and 3 look at the unconditional effects of the presidential vote in the district and party unity scores. The coefficient on Same Party Presidential Vote, which is the presidential vote in an incumbent's district recoded in relation to their partisanship, is positive and significant. If a representative's district increasingly votes for the presidential candidate of the same party, then the candidate receives ‘down-ballot’ benefits. The coefficient on Party Unity in the third column is close to 0 and imprecisely estimated, indicating that a representative's party unity score has no bearing – positive or negative – on their re-election margin.
Table 2. Partisan effects conditional on information environment

Note: cluster-robust standard errors. Relevant hypothesis tests in bold. Observations are incumbents in contested elections, 2002–2008. * p < 0.05
Columns 2 and 4 allow these two main effects to vary by the level of broadband providers. The coefficient on Same Party Presidential Vote in Column 2 is now close to 0 and imprecisely estimated, indicating that at low levels of district broadband internet connectivity, there is no relationship between the presidential vote in a district and an incumbent's re-election margin. That is, a hallmark of nationalized politics – House incumbents whose fortunes are strongly tied to presidential politics – is not present in districts with a low number of broadband providers.Footnote 19 The coefficient on the interaction term is positive and has a standard error of sufficiently small size that I reject the null hypothesis of a coefficient equal to zero. This indicates that as the number of broadband providers increases, the impact of presidential vote on incumbent vote margins also increases. The left panel of Figure 4 displays this relationship visually. At low levels of providers, the marginal effect of the presidential vote on incumbent vote share is not distinguishable from zero. At the median level of broadband providers this marginal effect rises to 0.25, indicating that with every additional percentage point the presidential candidate of the same party as the incumbent receives in the election, the incumbent's vote margin rises by 0.25 percentage points.

Figure 4. Conditional effects
In Column 4, the coefficient on Party Unity is negative and significant, lending support – at least for districts with low broadband internet connectivity – to the hypothesis that incumbents pay an electoral cost for voting in lock-step with their parties in the legislature. The coefficient on the interaction term indicates that as the number of broadband providers increases, the effect of Party Unity attenuates to zero. This indicates that as a district adds broadband providers, a high degree of party unity voting becomes less and less likely to be an electoral liability. This relationship is displayed visually in the right panel of Figure 4. While the effect of Party Unity is imprecisely estimated at all levels, the trend makes it clear that incumbents running in districts with rising broadband connectivity rates are far less likely to face electoral sanctions for voting with their party. Indeed, there is an increasing probability that these legislators may be rewarded for their partisan behavior.Footnote 20
Taken together, these tests present strong evidence that an expansion of the information environment in congressional districts had a negative effect on incumbency through increasing the weight of national considerations in legislative elections. Areas with faster than expected rates of broadband internet growth experienced a greatly reduced incumbency advantage. Incumbents’ electoral margins in those areas were more affected by ‘down-ballot’ effects, which necessarily reduce an incumbent's personal vote. Further, the electoral sanction for excessive partisanship erodes in districts with high broadband connectivity. The roll-out of broadband internet quite clearly had a significant role in nationalizing elections.
Conclusion
The incumbency advantage is the natural outflow of the type of retail politics described so colorfully by Fenno's classic work. The personal vote historically enjoyed by incumbents was in part a consequence of an information system that has been significantly eroded. This article has shown that the expansion of broadband internet has a clear negative effect on the incumbency advantage. This decline is due to an increase in partisan voting. The impact of a district's presidential vote on incumbent margins has increased in districts with high broadband connectivity, which dole out a smaller electoral punishment to representatives more concerned with party unity than moderating to the median voter.
For representatives facing this new information environment, assuming that incumbency alone is enough to guarantee re-election seems an increasingly bad bet. As internet access becomes cheaper and easier to use, incumbents find it harder to break through the tide of information. As a consequence, the incumbency advantage declines and rates of partisan voting increase. This seems to require a change in ‘home-styles’ away from advertising constituency service and boosting name recognition, and towards promoting oneself as a partisan warrior. The continuing expansion of the information environment – and the concurrent erosion of local news resources – produces incentives for members of Congress to be less in tune with the valence needs of the median voter in their district, and increasingly concerned with getting in-formation with the national party.
New communication technologies are not necessarily a one-way street to more nationalization. Anything that allows individuals to make independent decisions across levels of their ballots will reduce nationalization. In the past, members of Congress fostered a personal connection to their constituency that was more powerful than partisanship. Candidates could use social media to communicate directly with constituents and thus counteract increasing nationalization. Several studies have highlighted the increasing use of platforms like Twitter by members of Congress to form the sort of personal connections highlighted in the classic work of Fenno (Reference Fenno1978) (Kreiss, Lawrence and McGregor Reference Kreiss, Lawrence and McGregor2018; Lassen and Brown Reference Lassen and Brown2011; Straus et al. Reference Straus2013). Out-party voters may remember instances of social-media-facilitated connections when going to the ballot box, which could increase instances of split-ticket voting. Interestingly, this development seems in line with the hopes of Progressive reformers of the early twentieth century that new communication technology would break party machines’ monopoly over the transmission of political information (Bimber Reference Bimber2003, 75–76). It is difficult, however, to determine the relative magnitude of such an effect in relation to the damage an increasingly online world has wreaked on the traditional venues of these local considerations. Politicians’ use of social media to highlight differences (or alignment) with their parties will be an interesting avenue of research to watch over the coming years.
The appendices extend the results found here and offer several robustness checks. Appendix 1 tests for these relationships at the individual level. Replicating the analysis completed by Snyder and Strömberg (Reference Snyder and Strömberg2008) on newspaper coverage, I merge the data on broadband availability to respondents in the American National Election Study to determine the plausibility of the causal path specified here. A relatively small number of observations make the results tentative, but the pattern is broadly consistent with the theory presented: individuals exposed to broadband have less knowledge about their incumbents and more knowledge about national conditions. Another possible causal mechanism, Prior's (Reference Prior2007) ‘polarization without persuasion’ hypothesis, is tested in Appendix Section 2. Time spent online is primarily not spent viewing news of any sort. This could lead to the selective drop-out of the electorate: less interested individuals, who are more politically moderate, may read less about politics and subsequently become less likely to vote. The resulting electorate will be more partisan, even though no individuals have changed their political opinions. I test this theory by adding turnout data to my analysis. While the expansion of broadband is associated with a reduction in the size of the electorate for House races, controlling for turnout does not affect the relationships found here. In other words, there is no evidence that the smaller electorate is comprised of more partisan individuals in a way that would explain these results.
Observational studies can never fully rule out the possibility that the results are due to spuriousness, but the analyses presented here do a great deal to account for potential confounding factors. Using fixed effects on repeated observations of the same units effectively deals with many potential sources of bias. The results cannot be biased by any fixed feature of districts or legislators, nor can they be contaminated by any variables related to the time trend. Inasmuch as other media variables are unrelated to where broadband providers decide to extend service, for example, there is no potential for these results to be explained by contemporaneous changes. Further, I have taken steps to rule out the impact of many potential time-varying confounders, including changes in the population and median income of the districts (and additionally in Appendix Section 6, changes in median age, poverty rates, race and education). Nor are these results a function of broadband roll-out being related to politician characteristics, as tested in Appendix Section 3, or due to a violation of the parallel-trends assumption, as tested in Appendix Section 5. Ultimately, these results do much to improve our understanding of how shifts in the aggregate media environment affect changes in national elections.
For many years, some scholars have found the incumbency advantage normatively troubling (Carson and Roberts Reference Carson, Roberts, Edwards, Lee and Schickler2011). The degree to which incumbents can take advantage of the good will afforded by their constituents to freely pursue their own personal agendas (or to get away with personal misconduct) is a potentially sub-optimal outcome that may be mitigated by nationalization. Further, if a more partisan electorate is receiving more partisan politicians, then this ought to be considered successful representation. However, there are (at least) three reasons why this change may be normatively troubling.
First, just because more partisan districts are getting more partisan members does not mean that the aggregate output of the legislature is maximized. A declining incumbency advantage means a less experienced legislature, which is a key component of legislative productivity (Cox and Terry Reference Cox and Terry2008). It is also likely to produce a less moderate chamber, which similarly is detrimental to policy making. Grimmer (Reference Grimmer2013) discusses how members of Congress from marginal districts have an incentive to moderate their policy positions in order to win over opposing partisans. But an increasing reliance on party labels as a voting metric makes it less likely that opposing partisan voters can ever be won over by moderate policies and appropriations. As such, an increasingly partisan electorate weakens this incentive for moderation. Districts may still get ‘representation’ through replacement, although replacing one ideologically extreme member with another is a sub-optimal outcome for the median member of the district. Perhaps more consequentially, a less moderate chamber is seen as a key component of policy gridlock (Binder Reference Binder1999). On a district-by-district level representation may be occurring, but this may come at the detriment of collective representation through efficient policy output.
Secondly, there is a difference between choosing to use different media to access news and choosing different sorts of political information directly. While the internet has bolstered individual proclivities for national information (Hindman Reference Hindman2011; Hopkins Reference Hopkins2018), other changes to the media system were not ‘chosen’. Voters have largely gravitated towards online news and away from local sources due to the convenience inherent in the former (De Waal and Schoenbach Reference De Waal and Schoenbach2010; Gaskins and Jerit Reference Gaskins and Jerit2012; Ha and Fang Reference Ha and Fang2012). The changes to the information they receive are a symptom of this more basic desire for convenience – both in their personal experience and through the weakening of journalistic resources through a collective decision to move away from traditional media sources. Part of the decline in local newspapers, for example, is due to a drop in revenue from classified ads as individuals increasingly utilize online peer-to-peer marketplaces (Seamans and Zhu Reference Seamans and Zhu2010). When citizens list their couches on Craigslist they are (inadvertently) changing the political information environment and their own voting behavior. But they are not doing so consciously.
Thirdly, as discussed in the introduction, voters increasingly voting on one plane of partisan conflict violates the assumption underlying a system of checks and balances: where politicians in different institutions will have competing interests such that ‘ambition counteracts ambition’ (Madison Reference Madison1982[Reference Madison1788]). The implications for accountability in a system in which politicians in all branches of the federal system face the same electoral incentives are not simple to calculate. However, it seems clear that in such a system political parties bear a great deal of responsibility for curbing excesses of power. In particular, if congressional politics is increasingly driven by national concerns rather than the characteristics of local candidates, it is crucial that parties take seriously their responsibility to nominate suitable candidates for office – and to censure incumbents who act irresponsibly.
One of the main political goals of progressive reformers in the early twentieth century was to reduce the ‘distortions’ caused by rampant partisanship (Schudson Reference Schudson2000, 190). Bimber (Reference Bimber2003, 75–76) discusses how part of the power of large party organizations in the nineteenth century was based on their control of the flow of information. New information technologies like radio and television meant both that grassroots groups of citizens could more effectively organize, and that candidates were no longer dependent on the central party structure to connect with voters. Reformers at the time saw information taking the place of parties: politics was moving from ‘outdoors to indoors and from the heart to the head’ (Schudson Reference Schudson2000, 193–196). It is ironic, then, that the return to nineteenth-century partisan politics comes to us through the ultimate freeing of information represented by the internet. While strict control over the press facilitated partisan politics, so does an erosion of local media meant to hold politicians to account on important local issues, and a public that is more interested in national partisan news than local concerns.
Supplementary material
Data replication sets can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/C4QGBT and online appendices at: https://doi.org/10.1017/S0007123419000577.