1. Local elections as motivation for local news use

What can increase local news demand? We propose events which shift the public's attention to topics covered by local news sources might increase local media use. Following work demonstrating elections can reshape news audiences (Archer, Reference Archer2018; Kim and Kim, Reference Kim and Kim2021; Archer and Darr, Reference Archer and Darr2022), we consider whether local elections increase demand and consumption of local news.

Accounts of political information acquisition suggest several reasons the public may be more willing to consume coverage from local media when local elections are imminent, even for those who are otherwise uninterested in news coverage of local politics. In the lead up to these elections, voters must decide which candidates to support and information about local politics has more utility for the public's decision-making than at other times (Downs, Reference Downs1957). Elections also facilitate horse-race coverage of local politics that adjudicates who leads or trails in the campaign based on recent events, a style the public finds more entertaining than substantive policy news (Iyengar et al., Reference Iyengar, Norpoth and Hahn2004). Finally, the issues raised during local election campaigns may prime the importance of local government, which increases the appeal of local news in survey experiments (Hayes and Lawless, Reference Hayes and Lawless2021).

While these accounts tap different motivations for political news use, they all highlight an advantage local media outlets possess over national news organizations during local elections: the ability to provide timely, community-focused information about local politics (Mahone et al., Reference Mahone, Wang, Napoli, Weber and McCollough2019; Darr et al., Reference Darr, Hitt and Dunaway2021; Hayes and Lawless, Reference Hayes and Lawless2021). This leaves local media well positioned to attract anyone motivated to seek out information about local politics by elections happening in their community.

2. Data

Understanding of local news use has been limited by data constraints. Survey self-reports of media exposure contain measurement error, are oriented toward national news sources, and are collected too sporadically to assess how events like elections change local media consumption. Some studies consider more concrete aspects of local news choice, such as the public's interest in specific stories or news sources (Hayes and Lawless, Reference Hayes and Lawless2021; Peterson and Allamong, Reference Peterson and Allamong2022). However, it remains unclear whether such evidence generalizes to explaining habitual news consumption over a longer time period, a form of news demand with important economic implications for local media due to their need to appeal to advertisers (Hindman, Reference Hindman2018; Usher, Reference Usher2021).

Here we consider two behavioral measures of local news use: viewership of local television news and visits to local newspaper websites. Our data on television viewership come from TVEyes, which provides station-program-day level three week averages of Nielsen ratings (the share of a designated market area's audience watching local television news programs) during 2013–2019.Footnote ¹ Following previous work on local television, we focus on viewership of local news broadcasts during weekdays (Martin and McCrain, Reference Martin and McCrain2019). Our website traffic data comes from the Alliance for Audited media and measures monthly impressions on local newspaper websites from 2015 to 2021. As this measure is collected by several different web traffic providers depending on the newspaper, we demean and standardize it for each newspaper to understand fluctuations in online local news consumption. Using these indicators we consider how the monthly audience for these news sources changes over time. It is important to note that, beyond the general value of behavioral media use data, these measures matter because they are generated for advertisers and affect a news outlet's advertising revenue.

We combine these measures of local news use with data on the timing and outcomes of mayoral elections in the cities these news sources cover (Warshaw et al., Reference Warshaw, de Benedictis-Kessner and Velez2022). We see these elections as a likely case for local news consumption to change based on a regularly occurring local political event. This is because these elections focus on an executive office and their results have clear ramifications for the entire community. To ensure it is plausible that news audiences might be altered by the election, we focus on elections in the city a newspaper is located in or the largest community in a television station's coverage area. Elections in local news outlets with a strong overlap between their audience and relevant political constituencies are the most likely to engage news audiences and are also the types of media shown to affect local political outcomes (e.g., Hopkins and Pettingill, Reference Hopkins and Pettingill2018; Moskowitz, Reference Moskowitz2021). Later we also focus on close elections, those happening off the federal cycle, and partisan elections involving candidates from opposing parties to consider races that may be best suited to motivating interest in local news (e.g., Archer, Reference Archer2018).

Altogether, we consider a broad set of local news outlets and elections after combining these data sources into a panel. We analyze over 229,000 weeks of local television viewership from 633 television stations in 187 markets. For local newspapers, we consider 9,000 months of newspaper web traffic data from 169 local newspapers. All told, we observe 506 mayoral elections in the markets with news use data (Appendix A).

3. Estimation

Both the immediate and near-term impacts of mayoral elections on news use are important to consider, which requires us to estimate the dynamic effects of local elections on news use. This setting also includes a staggered treatment, as mayoral elections happen at different times, and their effects could differ based on timing (Anzia, Reference Anzia2013; de Bendictis-Kessner, Reference de Bendictis-Kessner2018).

Recent work establishes these features can lead two-way fixed effects models (TWFE), a common approach to analyzing panel data, to produce biased effect estimates by weighting units differently depending on when a “treatment”, in this case a mayoral election, occurs in them (Goodman-Bacon, Reference Goodman-Bacon2021). For this reason we follow best practices for studying staggered treatments with potentially time-varying, heterogeneous treatment effects (Callaway and Sant'Anna, Reference Callaway and Sant'Anna2021; Sun and Abraham, Reference Sun and Abraham2021). Specifically, we create a “stacked” estimation dataset, an approach that effectively estimates a standard dynamic TWFE model for each event window (i.e., each month in which a mayoral election occurs, see Baker et al., Reference Baker, Larcker and Wang2022). This approach addresses recent concerns raised in the literature as it permits dynamic treatment effects, allows for time-varying heterogeneous treatment effects and excludes already-treated units from the control group, generating a clean comparison group for evaluating the influence of local elections on news use. This method is also more statistically efficient than other approaches that address staggered treatment timing.

To implement the stacked approach, we construct treatment cohort groups based on the date on which mayoral elections occurred. For a given election date, all cities with a mayoral election (the treated units) are grouped together with cities that do not have a mayoral election in a given window surrounding that date (the control units). Relative time to treatment indicators are created for the treated units (e.g., one month pre-treatment, one month post-treatment, etc.)Footnote ² , while control units and the month of the election are set to zero. Finally, each treatment cohort receives a separate dummy variable.

We then estimate models of the following form:

(1)$$Y_{it} = \gamma_{id} + \lambda_{td} + \sum_{\tau = -l}^{-1}\gamma_{\tau}D_{i\tau} + \sum_{\tau = 1}^l\delta_{\tau}D_{i\tau} + \varepsilon_{idt}$$

Where Y _it, our measure of local news use in market/outlet i in month t, is regressed on unit (a newspaper or television station) fixed effects (γ _i), month fixed effects (λ _t), and treatment dummies for each pre- and post-treatment window from −l to l, excluding 0, which is when the mayoral election occurs. The parameters of interest are the pre-treatment lags γ and post-treatment leads δ of time to mayoral election. These estimate the dynamic treatment effects within each treatment cohort relative to untreated units. In the present application these terms reflect whether local media use in communities with an impending mayoral election changes before, during or after the election, relative to a reference group of outlets in communities without an election during that same time period. It is important to note that in this case, unlike many other event study analyses, we do not necessarily expect to see these pre-treatment dummies centered around zero. This is because the campaign that occurs in the lead-up to a mayoral election could increase engagement with local news before the election happens, a possibility our dynamic approach can capture.

4. Elevated attention to local politics during elections

Our argument that local elections could increase local news use is based on the assumption they are a time of elevated attention to local politics. Before considering the effects of local elections on media use, we provide evidence for this part of the research design.

To consider variation in the public's attention to local politics we follow previous work and use city-level Google Trends data on searches for the term “Mayor” in the same cities that later appear in our news use analysis (Sinclair and Wray, Reference Sinclair and Wray2015; Hayes and Lawless, Reference Hayes and Lawless2021). This measure captures relative monthly search for this term and in each city is standardized between 0 (lowest) and 100 (highest) interest for the term during the time period under consideration. To consider the media's attention to local politics, we use transcripts from a sample of the television stations in our news use analysis and count how often the term “Mayor” appears in news broadcasts each month. Our expectation is these indicators of public and media attention to local politics should increase during mayoral elections.

Figure 1 examines how attention to local politics varies in the months before, during and after mayoral elections. This uses the stacked regression approach described in the previous section. The coefficient estimates represent the difference between communities with a mayoral election, relative to those without an election, on the dependent variable. These estimated differences are all relative to the month in which the election occurred (t = 0), indicating whether this changes based on proximity to the election.

Figure 1.

Local Elections and Attention to Local Politics. (a) Searches for Mayor on Google, and (b) Mentions of Mayor on Local TV.

The results displayed in Figure 1 are consistent with our expectations. Compared to the months before and after the elections, there are substantial spikes in public (Fig. 1, Panel A) and media (Fig. 1, Panel B) attention to local politics the month the election occurs. This is indicated by the negative and statistically significant signs on the coefficients for the months before and after the election, a pattern that is similar for both outcomes. Compared to the month before the election, relative search for information about the mayor is 15 percentage points higher during the election month. Similarly, there are 4 more mentions of the mayor during newscasts the month of an election, relative to the month before it. Given there are on average 3.4 mentions of the mayor per month in this data, this is a large increase, with an election more than doubling the attention the mayor receives from local television stations compared to the typical month. In both cases the increased attention to local politics declines after the election. Overall, this is consistent with our claim that local elections increase attention to local politics, while also showing this method can successfully recover differences in attention brought about by elections.

5. Effects of local elections on local news use

Does the elevated attention to local politics that occurs during mayoral elections increase local news consumption? To assess this we compare monthly demand for local news at media outlets with an impending local election to media outlets without an election in their community. If the increased salience of local politics that accompanies elections shifts demand for local news, we expect to see higher levels of consumption for the news sources with a local election compared to the others.

Figure 2 shows the paired event studies for local television stations (Fig. 2, Panel A) and local newspapers (Fig. 2, Panel B), using the same approach as in the previous section. The coefficients reflect the difference in the size of the monthly audience of local news outlets with an impending mayoral election, relative to sources in communities without a mayoral election in the months before and after the election takes place. In Panel A, coefficients are interpreted relative to the three week period containing the election, such that a positive (negative) coefficient indicates higher (lower) viewership relative to when the election occurred. In Panel B, coefficients are interpreted relative to the month in which the election occurs.Footnote ³ Thus, if local news use gradually increased leading up to the election, and then dropped immediately after, the pre-election coefficients would be negative and approach zero, while the post-treatment coefficients would return to negative values.

Figure 2.

TV Viewership and Newspaper Traffic Event Studies. (a) Effect of Local Elections on TV Viewership, and (b) Effect of Local Elections on Newspaper Traffic.

We first discuss the estimates for local television viewership and newspaper website traffic in the months before the election. Here, there are not detectable differences in local news use between the election and non-election local news sources. The coefficients for local television are small and centered around zero, indicating viewership is stagnant in the run up to the election. Immediately following the election, there is a small short-lived increase in ratings that quickly dissipates in the second post-election three week period. In contrast, the point estimates for newspaper web traffic are negative, indicating lower levels of web traffic in the months before the election compared to the month of the election. However, in both cases these estimated differences are substantively small (roughly .04 ratings points for television and .1 standard deviations for newspapers) and do not reach statistical significance. Considering the dynamics of news use, the coefficients do not illustrate a change in the trajectory of news use leading up to the election. This indicates the presence of an upcoming mayoral election does not generate increased local news use in the lead-up to the election, a perhaps unsurprising pattern given our earlier evidence of a sudden shift in attention to local politics during the month of the election that is not preceded by a period of increase beforehand.

We now turn to the post-election time period. For local television, the post-election coefficients are again positive, indicating higher viewership during this period than when the election occurred. The magnitude of these estimates is largely similar to those from before the election, although in this case the positive estimate from two months after the election does reach statistical significance. In sum, the findings here indicate that the presence of a local election harms viewership of local television stations, showing they do not benefit, in terms of news use, from the increased attention to local politics that is present during this time period. Indeed, when jointly considering the monthly estimates from before and after the election, this indicates a statistically significant decline in local television viewership due to the election (−.04, 95% CI [−.08, − .01]). While this difference is not large–it represents only a .01 standard deviation shift in television news use–we conclusively rule out any election-related viewership benefits for local television stations.

For local newspapers, the post-election time period differs somewhat. Here the coefficient for the month after the election is negative and statistically significant, indicating a .1 standard deviation decline in website impressions for newspapers in communities with a mayoral election compared to those without one. This difference dissipates over time and the two groups of newspapers are no longer distinguishable after the first post-election month. This evidence does not show any sustained increase in local newspaper use due to a local election. In a more qualified sense there is evidence that local newspaper web traffic may increase during the month in which a mayoral election takes place, as estimates from both the month before and after the election are negative compared to this reference point. However, we note that the increase from the month before the election to the month of the election is not itself statistically significant. In any case, as with local television, there is not evidence that local newspapers receive either sizeable or long term benefits in the size of their audience in the wake of local elections in the communities they cover.

We can rule out large benefits for local media outlets from the increased attention to local politics that occurs during local elections, a finding that runs counter to our expectations that the presence of local elections would increase local news use. Local television viewership suffers during the election when compared to the time period before and after it. While the post-election decline in local newspaper web traffic is more in line with an account in which the presence of an election elevates local news use, we see no indication the election month itself is a time of elevated local news consumption compared to previous months and, ultimately, differences between the election and non-election communities dissipate by two months after it takes place. Here we note that this general pattern persists when we probe its robustness in several ways, including an alternative estimation approach (Appendix A).

6. Heterogeneity by election context

The preceding analysis shows limited evidence that local elections increase local news audiences. However, it includes different types of elections, some with features that could dampen any motivation for the public to consume local news. We now consider heterogeneity based around institutional and contextual variation in local elections.

One potential source of heterogeneity is whether the election occurs on or off the federal election cycle. Given the American public's greater interest in national politics, any role local elections have in motivating the local media use may be attenuated if they co-occur with more engaging elections for Congress or the Presidency (Hopkins, Reference Hopkins2018). For this reason we separate out mayoral elections that happen alongside federal elections (47% of the elections in our sample) from off-cycle elections at different times (53% of the sample) and which may be better positioned to shift local news demand.

Another feature we consider is whether or not the election is competitive. Many mayoral elections consist of a clearly favored candidate winning office in a relatively uncontested fashion, with the results largely determined at an earlier stage of the electoral process (e.g., Keele et al., Reference Keele, Shah, White and Kay2017). These races seem unlikely to stimulate demand for local news. For this reason we separately assess the effects of local elections on news use between cities with competitive mayoral elections (decided by 10 points or less, 27% of our sample) and uncompetitive elections decided by more than this this margin (73% of our sample).

We begin by considering this election-related heterogeneity for local television. The left panel of Figure 3 separates the event study for off-cycle elections, shown in blue, and on-cycle elections, with point estimates shown in red. Here differences are in the anticipated direction as television stations covering off-cycle elections experience an increase in viewership following the election of 0.05 rating points in month one. However, this effect size is substantively small at 0.06 within-channel standard deviations – effectively a precise difference only slightly larger than zero. Ratings quickly begin to decrease in the second period and become undifferentiated from zero. Elections that coincide with state-wide and federal elections see no increase in viewership post-election.

Figure 3.

TV Event Studies by Election Type. (a) Election Timing, and (b) Election Competitiveness.

The right panel of Figure 3 considers competitive elections, with point estimates in blue, and non-competitive elections, with point estimates in red. In close elections see some evidence of a small election run-up effect, with ratings peaking during the period of the election and the period immediately after. Relative to normal monthly variation in channel viewership these coefficients are small – roughly 10% of a standard deviation – and are better interpreted as precise zeros. For noncompetitive elections, viewership remains remarkably stable throughout the time period, indicating no election-related increase in viewership.

We now turn to examine the same sources of heterogeneity for local newspapers in Figure 4. Here the implications of the pooled results are not much changed when considering potential sources of heterogeneity. For both election timing and competitiveness, the one month post-election point estimates remain negative, rather than positive, indicating a roughly .1 standard deviation decline in web traffic following the mayoral election that is of similar magnitude across the different timing of elections or degrees of competitiveness. Like in the pooled newspaper results, the one month pre-election point estimates are also negative, though not statistically significant. This again offers suggestive evidence of an election-month boost in web traffic for local newspapers in communities with mayoral elections compared to those without one. In general, however, there is not any additional support for an increase in attention to local newspapers following an election when considering likely cases like off-cycle elections or those that are particularly close.

Figure 4.

Newspaper Event Studies by Election Type. (a) Election Timing, and (b) Election Competitiveness.

As a final source of heterogeneity we further subset the analysis to close elections that also involve candidates from opposing political parties, as past work finds partisanship to be a moderator how news audiences respond to elections (e.g., Archer, Reference Archer2018). Here the results resemble the results for close election as a whole and do not reveal any larger or more durable increase in local media use for these partisan, close elections (see Appendix Fig. A10). Altogether, even evaluating the subsets of elections with features that could be particularly engaging for audiences does not reveal sizeable shifts in local news consumption.

7. Implications for local news demand

We show mayoral elections are a time of elevated attention to local politics, but this increased attention is insufficient to elevate local television viewership or traffic to local newspaper websites. In contrast to our expectations, this suggests a stable audience for local political news, with few people motivated to opt in to local news use due to elections.Footnote ⁴ This evidence has several important implications for local news demand.

Efforts by elected officials, policy advocates, and journalists aimed at increasing attention to local politics by local media have many goals beyond changing the public's demand for local news. Our findings, however, indicate that some of the most important and salient local political events – mayoral elections – do not increase the local news audience's size. While it is certainly possible other political events may increase the local news audience, our results suggest a need for pessimism on this possibility. Even among the promising scenario of highly contested and salient elections, we find precise null effects on increasing audience sizes. These results align with recent field experiments which find that another promising shift for increasing news use, reducing the costs of local news subscriptions, also has limited effects on local media demand (Hopkins and Gorton, Reference Hopkins and GortonForthcoming; Trexler, Reference Trexler2023).

Given evidence on the limits of politics at increasing local news use, future efforts might apply a similar logic to the one used here, that local news outlets will be advantaged when public attention focuses on events that do not receive national media coverage, but instead consider non-political situations. Here we suggest local crises in which community information is particularly valuable (e.g., the COVID-19 pandemic, natural disasters) or non-political events (e.g., sporting events) as alternatives that could have a larger role than politics in shaping local media engagement and inform efforts to durably shift local news use habits.