Literature and Theory

The Geography of Television Media Markets

A major impediment to making inferences about the media’s effects on electoral politics and voting behavior is that different types of voters choose to consume different sources of media. As a result, any observed differences in voting behavior could be the result of underlying differences in voter characteristics rather than the media sources that voters opt to consume.Footnote ⁸ To remedy this selection problem, scholars have a long history of utilizing clever natural experiments that exploit quasi-random variation in exposure or access to the medium of interest.Footnote ⁹ For instance, Simon and Stern (Reference Simon and Stern1955) exploit a Federal Communications Commission (FCC) freeze on the issuance of television licenses to examine the effect of television on turnout and vote share in the 1952 presidential election in Iowa, where television was expanding at the time. Since this seminal work, scholars have continued to use natural experiments to study media effects, including an eight-month driver strike that disrupted newspaper delivery in Pittsburgh prior to the 1992 election (Mondak Reference Mondak1995); the roll-out of cable news channels and local Spanish-language news (DellaVigna and Kaplan Reference DellaVigna and Kaplan2007; Hopkins and Ladd Reference Hopkins and Ladd2014; Oberholzer-Gee and Waldfogel Reference Oberholzer-Gee and Waldfogel2009); the ordering of channels within a channel lineup (Martin and Yurukoglu Reference Martin and Yurukoglu2017); and the geographic structure of media markets for radio, newspapers, and television (Althaus and Trautman Reference Althaus and Trautman2008; Ansolabehere, Snowberg, and Snyder Reference Ansolabehere, Snowberg and Snyder2006; Hopkins Reference Hopkins2018; Levy and Squire Reference Levy and Squire2000; Schaffner Reference Schaffner2006; Snyder and Strömberg Reference Jr, James and Strömberg2010; Stewart and Reynolds Reference Stewart and Reynolds1990).

The geography of television media markets conveniently allows for the comparison of voters participating in the same state-level electoral context with access to entirely different television stations and, thus, different local news content. The present-day media market boundaries defined by the Nielsen Company (known as designated market areas—DMAs) approximate the geographic reach of television stations’ signals in the 1950s and 1960s (Gentzkow Reference Gentzkow2006). Consequently, these boundaries are largely idiosyncratic, not subject to contemporary political considerations, and often span multiple states. Indeed, as discussed in greater detail below, the idiosyncratic nature of these boundaries results in remarkable covariate balance across the in-state share of DMA measure within states. In other words, counties located in in-state media markets are very similar to counties located in out-of-state media markets along a host of observable characteristics.

Media market boundaries generally determine the television stations to which a viewer has access. With very few exceptions, multichannel video programming distributors (MVPDs), such as cable and satellite providers, are restricted from including out-of-market television stations in their channel lineups.Footnote ^{10 ,} Footnote ¹¹ While residents in certain areas may be able to receive over-the-air signals from out-of-market stations, the vast majority of television households subscribe to a cable or satellite provider.Footnote ¹² Prior to the rise of MVPD subscription rates (see Appendix Figure A1), media market boundaries were likely far more porous due to over-the-air signals penetrating across media market boundaries. In sum, in the present context, viewers residing within a given media market generally only have access to the television stations within that market.

Because media markets can span multiple states, a person may reside in a market that is predominantly comprised of residents from another state. The degree to which a person’s media market is comprised of residents from that person’s state is the key geographic feature of media markets that induces variation in exposure to local news content. A person lives in an out-of-state market if most of the residents of that media market are from another state, while a person lives in an in-state market if most of the residents of that market are from the same state. For example, Van Wert and Mercer Counties in Ohio are adjacent to one another (see Figure 1). Van Wert County is located in the Ft. Wayne DMA (7% of this market resides in Ohio and 93% in Indiana), while Mercer County is located in the Dayton DMA (100% of this market resides in Ohio). The in-state share of DMA measure that I employ in analyses below is coded as 0.07 for Van Wert County and 1 for Mercer County. As I demonstrate empirically below, a station’s news coverage of a given state’s governor (senators) increases with the proportion of residents in the media market from the governor’s (senators’) state. Thus, residents of Mercer County (in the Dayton, OH media market with an in-state share of DMA of 1) are expected to receive substantially more coverage of Ohio officeholders than residents of Van Wert County (in the Fort Wayne, IN market with an in-state share of DMA of 0.07). For more information on media markets and how the in-state share of DMA measure is constructed, see Appendix Section A.2.

Figure 1.

Media Market Boundaries and In-State Share of DMA: An Example

Note: Mercer County, which is shaded in dark green on the map, is located in the Dayton DMA. The boundaries of the Dayton DMA include the counties shaded in green: 100% of the Dayton DMA’s residents live in Ohio, as the market does not span across the state boundary. On the other hand, Van Wert County, which is shaded in dark blue on the map, is located in the Fort Wayne DMA. The boundaries of the Fort Wayne DMA include the counties shaded in blue: 7% of the Fort Wayne DMA’s residents live in Ohio and 93% live in Indiana.

There is broad variation across the country in the degree to which counties are located in in-state vs. out-of-state markets. Overall, about 19% of counties are located in out-of-state markets in which less than 50% of the market’s residents are in-state, and 43 states have at least one county that is located in an out-of-state market based on this threshold.Footnote ¹³ See Figure 2 for a map of the lower 48 states and variation within each state in terms of counties in relatively in-state (lighter shades of blue) vs. relatively out-of-state markets (darker shades of blue).

Figure 2.

Geographic Distribution of Counties in Out-of-State Media Markets

Note: Darker shades indicate that counties are located in predominantly out-of-state media markets, while lighter shades indicate in-state media markets. Counties in out-of-state markets have lower values for the in-state share of DMA measure and are located in media markets in which most of the market’s population resides in a different state.

The validity of this empirical strategy rests on the assumption that these media market boundaries are haphazard such that residents of in-state media markets are comparable to individuals from the same state who reside in out-of-state media markets.Footnote ^{14 ,} Footnote ¹⁵ I assess the plausibility of this assumption in Appendix Section A.3; Appendix Figure A3 indicates remarkable covariate balance with respect to race/ethnicity, educational attainment, age, residential mobility, median income, poverty, party vote share and turnout rate in the 2016 presidential election, and several other characteristics. The impressive covariate balance weighs against the possibility that people’s political preferences, attitudes, or propensity for certain behaviors are correlated with their in-state media exposure such that differences in voter knowledge or voting patterns would be attributable to an unobserved/unknown confounder. If I had observed serious differences in the populations residing in counties located in-state vs. out-of-state media markets, that imbalance would have called into question the validity of the research design.

Of note, counties located in relatively out-of-state media markets do tend to be more rural, less populated, have a daily newspaper, and, unsurprisingly, these counties tend to be geographically located relatively near to state boundaries (based on a visual inspection of Figure 2). To alleviate concerns that these differences may jeopardize the comparability of individuals residing in in-state vs. out-of-state markets, I present additional specifications in which I subset the sample to different subsamples of geographically proximate residents (e.g., residents of counties that border a neighboring state). In the subsamples of geographically proximate residents, the covariate balance improves markedly in terms of these characteristics.Footnote ¹⁶ These specifications are described in greater detail subsequently.

How Media Markets Structure Local News Coverage

The biggest difficulty in studying patterns in local television news coverage is data availability. It is rare for stations to make transcripts of their newscasts publicly available, and few research repositories exist that collect local television newscast transcripts.Footnote ¹⁷ To overcome this obstacle, I rely on publicly available closed-captioning data from the Internet Archive’s TV News Archive.Footnote ¹⁸ In total, the TV News Archive contains about two million hours of coverage, equating to 5.7 billion words, from over 150 stations (Leetaru Reference Leetaru2016). Many of the stations in the archive are national or international networks such as CNN, MSNBC, Fox News, and BBC News London.

I include in the sample all local stations affiliated with one of the four major networks (ABC, CBS, NBC, or Fox) and analyze coverage during 2016.Footnote ¹⁹ The sample includes 99 stations from 25 media markets and, in total, about 24,000 hours of nonentertainment coverage. I retrieve data through an API to count the number of 15-second clips that mention a specific governor or senator from all states that overlap with the station’s media market (see Appendix Section A.4 for more technical details). For example, the Boston DMA overlaps with three states—Massachusetts, New Hampshire, and Vermont—so I count clips that mention senators and governors from those three states for all stations in the Boston DMA.

Figure 3 demonstrates this first-stage relationship between a state’s population share of a DMA and the level of news coverage that the governor and senators from that state receive from stations in that DMA. As is clear from the figure, television stations provide greater coverage of an officeholder as the officeholder’s state makes up a larger share of the population in the DMA.

Figure 3.

Local TV News Coverage of Senators and Governors by State’s Share of DMA

Note: A station devotes more coverage to senators (left) and governors (right) from states within their media market as the share of residents from that state comprise a larger proportion of the market. The plot is residualized to account for senator/governor fixed effects. Each point corresponds to a local mean, and the regression line is estimated using OLS. Officeholders who were presidential or vice presidential candidates in 2016 are excluded from the analysis.

Consequently, a resident of an entirely in-state market (i.e., a media market that is entirely contained within the resident’s state) is expected to receive about 1.5 additional mentions of her governor per hour of coverage relative to a resident of an out-of-state market in which only a minuscule share of the media market’s population lives in the same state as the resident.Footnote ²⁰ The same estimated quantity for a senator is about 0.5 additional mentions per hour hour of coverage.Footnote ²¹ While the magnitude is substantially larger for governors, both quantities represent substantively meaningful increases in coverage.Footnote ²² In sum, the relationship between a state’s population share of the media market and local news coverage for that state’s governor and senators is positive and strong. The full results from the local news coverage analysis along with additional technical information are available in Appendix Section A.4.

These patterns provide strong evidence that residents of in-state media markets have access to far more news coverage of their senators and governor than residents of out-of-state markets. The relationship is approximately linear, which provides an empirical basis of support for using a continuous measure of the in-state share of DMA.

Local News Coverage Informs Voters

As elaborated earlier, the theoretical logic linking the nationalization of news to nationalized voting can be summarized as follows. Higher exposure to relevant local news provides voters with more information about down-ballot officeholders and candidates, which better enables voters to evaluate candidates for down-ballot offices separately from their national assessment at the top of the ticket for president. A voter with less exposure to relevant news coverage is more likely to apply her assessment from the top of ticket to down-ballot offices and engage in nationalized voting (i.e., straight-ticket voting). This logic yields two testable implications. First, voters with greater exposure to relevant local news coverage should have more knowledge of their down-ballot officeholders (e.g., senators, governor). And, second, voters with greater exposure to relevant local news coverage should have a higher probability of casting a split-ticket ballot. In this section, I present evidence on the intermediate outcome: voter knowledge of down-ballot officeholders and candidates. In the subsequent section, I examine the effect of exposure to relevant local news coverage on split-ticket voting.

Remarkably, Representative Andrew Maguire of New Jersey’s 7th district pointed to the idea that residents who lack access to in-state television stations are unable to stay informed: “The citizens in the northern part of the State are better able to recognize Mayor Koch of New York City and the political issues facing local legislators in Albany then [sic] they are able to identify their own local political leaders and State officials. They are deprived, for the most part, of local advertising and public affairs programming that is targeted to their own needs. The citizens in southern New Jersey suffer the same fate at the hands of the Philadelphia television programmers.”Footnote ²³ Does exposure to relevant local news coverage about governors and senators inform voters?

For this analysis, I use data from the 2012 and 2016 Cooperative Congressional Election Study (CCES). The large sample size of the CCES provides substantial geographic coverage of respondents in counties located in both in-state and out-of-state markets. Conveniently, the CCES asks respondents several questions that gauge knowledge of their elected officials. The empirical setup is as follows:

$$ \unicode{x1D53C}\left[{knowledge}_{ist}\right]={\lambda}_{st}+\phi \cdot {p}_{ist}+{X}_{ist}\cdot \beta, $$

where knowledge_ist is the knowledge of respondent i in state s in year t (measures of knowledge are discussed below), $ {\lambda}_{st} $ is a state-year fixed effect, p_ist is the in-state population share of the DMA in which respondent i lives, and X_ist is a vector of individual-level control variables. The inclusion of state-year fixed effects ensures that comparisons are of individuals within the same statewide electoral context. Given that this is a study of statewide offices (i.e., governor, senator), the state-year fixed effects hold constant the officeholders and candidates and all other features of the statewide setting in an election year. The coefficient of interest is $ \phi $ . All models are estimated using ordinary least squares.

Table 1 displays the relationship between the in-state share of DMA (i.e., the proportion of the DMA’s population from the respondent’s state) and three measures of knowledge. The party recall measure indicates whether the respondent can correctly recall their senators’ or governor’s party (see specifications 1 and 4, labeled: “Recall”), the ability to evaluate measure indicates whether the respondent is able to evaluate (i.e., approve or disapprove) their senators or governor (see specifications 2 and 5, labeled: “Eval.”), and the ability to place on an ideological scale measure indicates whether the respondent is able to place their senators or governor on an ideological scale (see specifications 3 and 6, labeled: “Ideo.”).Footnote ²⁴ All specifications reported in Table 1 include demographic control variables: family income, gender, race, education, age, and marital status.Footnote ²⁵ Alternative specifications are reported in Appendix Tables A3–A5. Estimates are stable across all specifications.

Table 1.

Voter Knowledge of Senators and Governors

Note: Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

The dependent variable is whether the individual can recall the party of the senator/governor in specifications (1) and (4), whether the individual is able to evaluate the senator/governor in specifications (2) and (5), and whether the individual is able to place the senator/governor on an ideological scale in specifications (3) and (6). Basic control variables include family income, gender, race, education, age, and marital status.

Across the three knowledge measures for both senators and governors, the estimated coefficient for in-state share of DMA is substantively large. Importantly, the in-state share of DMA variable ranges from < 0.01 to $ 1 $ , so it is approximately correct to interpret the estimated coefficient as the expected increase in voter knowledge moving from an almost entirely out-of-state market to an entirely in-state market.Footnote ²⁶ Thus, a person living in an entirely in-state market is about 9 percentage points more likely to recall their senator’s party correctly and 11 percentage points more likely to recall their governor’s party correctly than a person living in the same statewide electoral setting in an entirely out-of-state market. Similarly, relative to residents of entirely out-of-state markets, residents of entirely in-state markets are about 9 percentage points more likely to be able to evaluate their senator, 4 percentage points more likely to be able to evaluate their governor, 8 percentage points more likely to be able to place their senator on an ideological scale, and 7 percentage points more likely to be able to place their governor on an ideological scale.Footnote ²⁷

Exposure to local television news not only increases voter knowledge about officeholders but also knowledge about nonincumbent candidates. The CCES asks respondents to place Senate candidates, including those candidates challenging an incumbent and those running in an open-seat race (unfortunately, CCES respondents are not asked about nonincumbent gubernatorial candidates). The results are displayed in Table 2. For all three types of candidates—challenger, Democratic open-seat, and Republican open-seat—the relationship between in-state share of DMA and ability to place the candidate on an ideological scale is substantively large. Residents of entirely in-state markets are 10–13 percentage points more likely to be able to place a candidate on an ideological scale relative to residents of entirely out-of-state markets.

Table 2.

Voter Knowledge of Nonincumbent Senate Candidates

Note: Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

The dependent variable is coded = 1 if the individual is able to place the candidate on an ideological scale and = 0 if the individual is unable to make an ideological placement. Basic control variables include family income, gender, race, education, age, and marital status.

One especially important concern—and serious threat to inference—is that the residents of in-state markets are more politically knowledgeable than residents of out-of-state markets due to differences between the two groups of residents. While the strong covariate balance described above allays this concern, the possibility remains that an unobserved difference accounts for these gaps in political knowledge. If this were the case, then residents of in-state markets should demonstrate greater general political knowledge as well. Table 3 displays results from six placebo tests that all capture general knowledge of national politics: knowledge of which party controls the House (1) and Senate (2), ability to place the Democratic Party (3) and Republican Party (4) on an ideological scale, correctly placing the Democratic Party to the left of the Republican Party on an ideological scale (5), and ability to place the Supreme Court on an ideological scale. Across all six placebo tests, the coefficient on the in-state share of the market is estimated precisely and near to zero. These results provide strong evidence that residents of in-state markets are more knowledgeable about their governor and senators due to exposure to in-state television rather than bias from confounding variables.

Table 3.

Voter Knowledge Placebo Tests

Note: Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

The dependent variable for models (1) and (2) is whether the individual knows which party has a majority of seats in the House (1) and Senate (2); the dependent variable for models (3), (4), and (5) is whether the individual can place the Democratic Party (3) and Republican Party (4) on an ideological scale and whether the individual correctly places the Democratic Party to the left of the Republican Party (5); and the dependent variable for model (6) is whether the individual is able to place the Supreme Court on an ideological scale. Basic control variables include family income, gender, race, education, age, and marital status.

Local News Coverage and Split-Ticket Voting

As discussed above, greater exposure to local news coverage may allow voters to assess down-ballot candidates separately from their national, partisan judgment for president. Voters with less exposure to local news coverage may be more likely to apply their national, partisan judgment to down-ballot races, resulting in a straight-party vote. Again, using a state’s population share of the media market as the key source of variation in the level of local news coverage, I examine the degree to which exposure to local news coverage increases the probability of a voter engaging in split-ticket voting for governor and senators. The empirical strategy remains the same as in the previous section:

$$ \unicode{x1D53C}\left[{split}_{ist}\right]={\lambda}_{st}+\phi \cdot {p}_{ist}+{X}_{ist}\cdot \beta, $$

where split_ist is coded = 1 if respondent i in state s in year t casts a split-ticket ballot (i.e., votes for candidates from different major parties for president and senator/governor) and is coded = 0 if respondent i casts a straight-ticket ballot (i.e., votes for candidates from the same major party for president and senator/governor), $ {\lambda}_{st} $ is a state-year fixed effect, p_ist is the in-state population share of the DMA in which respondent i lives, and X_ist is a vector of individual-level control variables.Footnote ²⁸ Again, the state-year fixed effects allow for the comparison of voters within the same statewide electoral context across different media markets. This strategy exploits quasi-random variation in the level of relevant local news coverage but ensures that these voters experience identical statewide electoral conditions (e.g., candidates, incumbency, closeness of the race). And, the coefficient of interest is $ \phi $ .

Table 4 displays the main results. For Senate elections, voters in an entirely in-state market are about 2–3 percentage points more likely to cast a split ticket than voters in an entirely out-of-state market within the same statewide electoral context. For gubernatorial elections, the estimated coefficient implies a 4–5 percentage point increase in ticket splitting for voters residing in entirely in-state markets relative to voters living in entirely out-of-state markets.Footnote ²⁹ Of note, the larger magnitude of the estimated effect of local news coverage for the governor specifications is consistent with the higher level of local TV news coverage that governors garner. The estimated effects for both senators and governors are substantively large: based on estimates from the CCES, in total, about 8% of voters in 2012 and 2016 cast a split president-senator ticket and 9% of voters cast a split president-governor ticket in 2012 and 2016. Thus, media markets and local news coverage seem to play a meaningful role in ticket-splitting behavior in contemporary U.S. elections.

Table 4.

Split-Ticket Voting: President-Senator and President-Governor

Note: Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

Dependent variable is coded = 1 if an individual votes for the Democrat (Republican) for president and Republican (Democrat) for senator/governor. Basic control variables include family income, gender, race, education, age, and marital status. Partisan/ideological controls include opposite-party incumbent, strength of partisanship, and strength of ideology.

Robustness Checks and Alternative Explanations

Robustness Checks

The strong covariate balance and placebo tests related to general political knowledge strongly suggest that voters living in out-of-state markets are a valid comparison group for voters living in in-state markets. Nevertheless, to provide further evidence that the observed relationship between in-state share of DMA and split-ticket voting is indeed causal, I subset the sample to geographically proximate residents. In the first subsample, I restrict the sample to residents of counties that are adjacent to counties in another state (i.e., residents of border counties). And, in the second subsample, I restrict the sample to residents of counties for which the in-state share of DMA is less than 0.5 as well as residents of counties adjacent to such counties (i.e., residents of counties in out-of-state markets and adjacent counties). The two restricted samples have strong covariate balance across a host of characteristics, including how rural and how populated the counties are (see Appendix Figures A4–A5). The results based on the geographically proximate subsamples are displayed in Table 5; see specifications (1) and (4) for results from the border counties subsample and specifications (2) and (5) for results from the adjacent counties subsample.Footnote ³⁰ While the precision of these estimates is unsurprisingly diminished due to the substantial reduction in sample size and the number of clusters, the estimates—for both geographically proximate subsamples for senators and governors—are of a reasonably similar size to the estimates reported in Table 4. These results from narrower samples of voters who are geographically closer to one another and reside in similar locales provide additional evidence that the relationship is causal.

Table 5.

Alternative Specifications: President-Senator and President-Governor Split-Ticket Voting

Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

Dependent variable is coded = 1 if an individual votes for the Democrat (Republican) for president and Republican (Democrat) for senator/governor. Basic control variables include family income, gender, race, education, age, and marital status. Models (1) and (4) restrict the sample to residents of counties adjacent to another state, models (2) and (5) restrict the sample to residents of counties in which in-state share is less than 0.5 or residents of counties adjacent to such a county, and models (3) and (6) use a dichotomized version of in-state share of DMA (coded = 0 if in-state share of DMA < 0.5 and coded = 1 if in-state share of DMA ≥ 0.5).

Table 5 also contains results based on a dichotomous measure of an in-state market (see specifications 3 and 6).Footnote ³¹ A voter is considered to live in an in-state DMA if at least half the market’s population resides in the same state (i.e., the in-state share of DMA ≥ 0.5). Again, the magnitude of the estimates based on the dichotomous in-state DMA measure are very similar to those based on the continuous measure. These results are important for a few reasons. First, the main findings in the paper are robust to a nonlinear characterization of the explanatory variable of interest. Second, these results imply that variation in a narrow part of the distribution of the in-state share of DMA measure is not driving the observed relationship. That is, it is not simply the case that the analysis is leveraging variation from residents of markets with an in-state share between, say, 0.9 and 1. Finally, the similarity in the magnitude of the estimated coefficients for the specifications with dichotomous and continuous measures bolsters the claim that exposure to in-state television news increases the probability of casting a split president–senator ballot by about 2–3 percentage points and a split president–governor ballot by about 4–5 percentage points.

To provide further evidence of the effect of local news coverage on ticket splitting, I also examine the relationship between the in-state population share of media markets and aggregate election returns at the precinct level. The full technical details of the analysis and results are displayed in Appendix Section A.6. The aggregate approach guards against potential survey nonresponse bias and measurement error in recalling vote choice across offices. This approach has a disadvantage in that it requires an aggregate proxy measure of split-ticket voting, which is an individual-level phenomenon. Thus, I first demonstrate the validity of the precinct-level proxy using unique ballot-level data from the South Carolina Election Audit. Then, using precinct-level data from the 2012 Harvard Election Data Archive, I examine the relationship between the in-state share of DMA and the precinct-level measure of split-ticket voting. The positive, significant estimates from the precinct-level analysis provide strong corroborative evidence that local news coverage increases split-ticket voting (see Appendix Table A16). And, as in the individual-level specifications, the relative magnitude of the estimated coefficient is higher in the gubernatorial specifications than the senatorial specifications, consistent with the higher level of local television news coverage that governors receive relative to senators. This analysis serves to demonstrate that the observed relationship between the in-state population share of the media market and split-ticket voting is not merely an artifact of survey measurement.

Is It Local News Coverage or Campaign Advertisements?

Thus far, I have demonstrated that voters residing in in-state markets have greater exposure to news coverage about their governor and senators, have more knowledge about these officeholders, and are more likely to engage in split-ticket voting. While my theoretical claim is that exposure to local news coverage accounts for the knowledge advantages and the increased ticket splitting, a plausible alternative mechanism, however, is exposure to campaign advertising on television. In this scenario, voters are better able to assess down-ballot candidates independently of their national partisan assessment due to higher levels of television advertising in in-state markets. Campaigns are strategic and tend to air more ads in in-state markets than out-of-state markets (see Appendix Section A.7 for an analysis demonstrating this relationship). This pattern is unsurprising: it is a far more efficient use of limited campaign resources to air advertisements in in-state markets in which most viewers are eligible to participate in the relevant electoral contest.

Voters are typically exposed to advertisements only during the period in which an officeholder is seeking reelection. Thus, if voters are primarily acquiring information from advertisements (rather than local news coverage), the relationship between in-state share of the DMA and voter knowledge should be driven by the officeholders running for reelection contemporaneously to the administration of the survey. In Table 6, I present specifications that test whether the magnitude of the in-state effect is contingent on whether the officeholder is running for reelection. The coefficient of interest from this table is the interaction between running (for reelection) and the in-state share of DMA. This coefficient indicates the extent to which the effect of in-state television is larger for officeholders running for reelection.

Table 6.

Voter Knowledge: Running for Reelection vs. Not Running

Note: Robust standard errors clustered by state-DMA in parentheses. *p $ < $ 0.05.

The dependent variable is whether the individual can recall the party of the senator/governor in specifications (1) and (4), whether the individual is able to evaluate the senator/governor in specifications (2) and (5), and whether the individual is able to place the senator/governor on an ideological scale in specifications (3) and (6). Basic control variables include family income, gender, race, education, age, and marital status.

The estimated coefficient for the interaction term for all three measures of knowledge across both offices is of modest size and not significant. While not statistically significant, the interaction term is positive in 5 of the 6 specifications, which is at least suggestive that voters absorb some informational content from ads.Footnote ³² Nevertheless, the coefficient for the main effect of in-state share of DMA demonstrates that voters residing in in-state markets have substantially greater knowledge of officeholders not running for reelection compared with residents of out-of-state markets.

In addition to campaign ads, I consider two other alternative informational mechanisms that could explain the observed relationship: exposure to national cable news and exposure to local newspapers. In Appendix Section A.8, I demonstrate that residents of in-state markets and residents of out-of-state markets report almost identical levels of cable news viewership. Furthermore, in Appendix Section A.5.5, I control for the presence of a local newspaper in the voter’s county, and estimates are insensitive to the inclusion of this control. Overall, these results provide strong evidence that exposure to local TV news primarily accounts for the increased voter knowledge and ticket splitting rather than exposure to campaign ads, cable news, or local newspapers.

Discussion and Conclusion

This paper demonstrates the important effects that exposure to local television news coverage has on voter knowledge and split-ticket voting. The geography of television media markets structures the level of relevant local news coverage to which voters are exposed. Voters who live in in-state media markets, which are comprised mostly of residents from their own state, are exposed to higher levels of relevant local news coverage about their governor and senators than voters who live in out-of-state media markets, which are primarily comprised of residents from a neighboring state. Exposure to relevant local news coverage has considerable effects on voter knowledge of both officeholders and nonincumbent candidates. For instance, residents of in-state markets with access to relevant local television news coverage are 9 percentage points more likely to recall the party of their senators and 11 percentage points more likely to recall the party of their governor than residents of out-of-state markets with little access to relevant local TV news. Moreover, voters residing in in-state markets are about 2–3 percentage points more likely to cast a split president–senator ticket and 4–5 percentage points more likely to cast a split president–governor ticket. Given that the overall rate of split-ticket voting is about 8–9% for these offices in these election years, the estimated effects of exposure to local television news on ticket splitting are considerable. Importantly, these findings are robust to the inclusion (or exclusion) of various control variables, alternative characterizations of the explanatory variable of interest, restricting the sample to subsets of geographically proximate residents, and different sources of data (individual-level survey data and precinct-level election data). Auxiliary analyses, including placebo tests, all point to local television news coverage as a causal factor rather than campaign television ads or unobserved confounders.

The results accord with the theoretical expectation that greater exposure to local news coverage provides voters with relevant information about officeholders and candidates, which allows voters to make assessments for down-ballot elections separately from their national, partisan judgment in the presidential race. In the absence of exposure to relevant local news coverage, voters are more likely to apply their national, partisan judgment to down-ballot races. The results suggest that the decline of local news likely contributes—to an important degree—to the nationalization of U.S. elections. Unfortunately, given the limited period under examination, it is difficult to assess the degree to which declining local television news audiences account for the decrease in split-ticket voting observed over the past couple of decades. It is possible that local television news had even larger effects on ticket splitting in the recent past, prior to the downturn in local newscast audiences. Moreover, to the extent that these results are indicative of the effect of local news coverage beyond television, the broader decline in traditional sources of local news likely contributed substantially to the nationalization of elections. Local newspapers have experienced steep declines in their circulation and revenues, resulting in considerable cuts to staffing and other resources required to provide extensive local coverage. Less speculatively, the findings suggest that local television news has mitigated nationalizing electoral forces in Senate and gubernatorial elections even in the contemporary polarized context.

The nationalization of elections has serious implications for the functioning of representative democracy. On the one hand, to the extent that party labels are more clearly defined across offices and levels of government (i.e., if a party label means the same thing or at least similar things), electoral nationalization could simplify the task of voters and improve ideological or policy-based representation. On the other hand, nationally meaningful party labels could crowd out important state and local issues in the calculus of voters. Moreover, with highly nationalized elections, the electoral prospects of candidates and officeholders are based less on their (prospective) performance in office (e.g., competence, service delivery, economic performance, etc.) and more on national forces such as the popularity of the president. Unless the performance of officeholders is correlated within party and across levels of government, and these correlations in performance correspond correctly with national electoral forces (e.g., the president’s popularity), it is difficult to hold elected officials accountable based on their performance. In the context of U.S. federalism, the actions and responsibilities of state and local elected officials are mostly distinct from those of federal officeholders. If voters are unable to assess down-ballot officeholders separately from their national judgment at the top of the ticket, these officeholders will have little incentive to defect from their parties. These forces are likely to exacerbate elite polarization. After all, if voters have limited ability to discern differences among officials within the same party, these officials have little reason to separate themselves from their party in service to their local constituents if there is no electoral reward for doing so.

With that said, scholars in the past have bemoaned indistinct parties and incumbency advantage as problematic for accountability and representative democracy. The parties are now more distinct and the incumbency advantage has declined, which have yielded the present accountability concerns due to increasingly nationalized elections. Thus, it is easy to point to present ills and far more difficult to diagnose potential future ills.Footnote ³³ More optimistically, U.S. primary elections help to filter out less qualified candidates as well as officeholders involved in scandals even in safe districts in which general elections fail to act as an effective filter (Hirano and Snyder Reference Hirano and Snyder2019). Of course, the first-order concern with the decline of local news should be its direct impact on being able to hold politicians accountable (e.g., Arnold Reference Arnold2004; Snyder and Strömberg Reference Jr, James and Strömberg2010). As local news sources disappear and their resources decrease, scandals, corruption, and other bad behavior may be more likely to go unnoticed. Politicians may also be less deterred with fewer watchdogs, increasing the incidence of such behavior. If the wrong-doing goes undetected, it is impossible for voters to hold the officeholder accountable in a primary or general election. Moreover, as citizens’ reliance on national news presumably increases with the decline of local sources of news, the character of the news coverage of political elites is likely to differ in ways that could have important consequences for mass polarization. For instance, research indicates that national sources of news, such as the New York Times, broadcast television news, and cable television news, provide greater coverage of more ideologically extreme elected officials relative to more moderate officials (Padgett, Dunaway, and Darr Reference Padgett, Dunaway and Darr2019; Wagner and Gruszczynski Reference Wagner and Gruszczynski2018).

A series of FCC rule changes appears to erode important protections for localism in broadcast news. First, the FCC eliminated the main studio rule, which required stations to have a physical studio in or near the community of license (Federal Communications Commission 2017c). FCC Commissioner Mignon L. Clyburn, who voted against the rule, issued a strong warning when the rule was proposed: “By tentatively proposing to eliminate the Commission’s main studio rule however, it seems to me that we are embracing a world in which automated national programming is the new normal” (Federal Communications Commission 2017e, 1). The Commission also has loosened restrictions on media consolidation within the same market area (Federal Communications Commission 2017d), and it has taken steps that effectively raise the cap on the national audience reach of a single owner (Federal Communications Commission 2017a).Footnote ³⁴ The possible consequence of these rule changes is that a single entity can own television stations with a vast reach throughout the country and pipe into those stations nationally oriented news programming with little local content. Sinclair Media Group is one such company that has acquired several stations recently and thereby expanded its reach.Footnote ³⁵ Alarmingly, Martin and McCrain (Reference Martin and McCrain2019) find that stations acquired by Sinclair increased their coverage of national politics and reduced their coverage of local politics. The direction of these new FCC rules is likely only to accelerate the nationalization of elections in the near future.