CLEVELAND, OH – JULY 11: Windows at the team shop for the Cleveland Cavaliers are decorated for the Republican National Convention on July 11, 2016 in Cleveland, Ohio. The convention will be held at Quicken Loans Arena July 18-21, 2016. (Photo by Angelo Merendino/Getty Images)

The 2016 Republican National Convention begins next week, with the Democratic National Convention in Philadelphia following the week after. The Republican National Committee and Democratic National Committee chose these cities among several competing bids in 2014 and 2015, respectively, long before the presidential campaign began in earnest.

One reason parties choose a city is logistics the host city has to have hotels, a major airport, a large stadium, security, financial resources and more. Electoral considerations also play a role, as parties often place their convention in a battleground state — places such as Ohio and Pennsylvania.

Does the convention actually deliver votes in these states? It’s well known that conventions provide bumps in national polls, but until recently, the research had found little effect at the state level. However, our new research finds that conventions do win votes, just in a more localized fashion. Here’s how.

We make two key arguments. First, states are the wrong place to look for the local impact of conventions. The hoopla from a convention will not travel evenly through a state with multiple media markets and political ecosystems. For example, we would expect that any impact of a convention in Cleveland will diminish for voters who live hours away. The “information flow” that a convention generates — the factor that usually produces the bump — won’t be as intensive in, say, Cincinnati as in Cleveland. Similarly, a convention held in Philadelphia should actually affect voters in nearby New Jersey more strongly than voters further away in Pittsburgh.

Second, we argue that — even though parties strive to make their convention one big infomercial — the impact won’t always be positive. It depends on whether an area is predisposed to support the party in the first place. Conventions should rally local voters who are predisposed to support the party and perhaps even win over some voters who would have otherwise voted for the other party. But the convention could also create a backlash by mobilizing opposing partisans or turning off voters who realize they are out of sync with that party’s candidate.

Our evidence is based on county-level election returns for every presidential election from 1972 to 2012. We find that having a convention in a particular media market can lead to a greater vote share for that party’s candidate in the counties in that market, compared with the previous election. But this depends on whether a county was predisposed to favor that party. Here’s our key graph:

The left-hand side shows how the apparent impact of the Democratic convention depends on how Democratic a county is. Within the media market where the convention occurs, the Democratic candidate tends to gain vote share in counties that are Democratic or slightly Republican — suggesting that the convention mobilized Democratic voters and may have persuaded a few non-Democrats, too. But the Democratic candidate loses vote share in counties that are largely Republican.

The right-hand side shows the similar, but smaller, impact of the Republican convention. For counties that strongly favor the GOP, exposure to the Republican convention increases the Republican candidate’s vote percentage by about one percentage point, whereas the opposite happens in counties that strongly favor the Democrats. There is not much of an effect for the Republican convention in toss-up counties.

Consequently, Democrats would appear to do best by siting their national convention not just in “swing states” for electoral college advantage, but in media markets with toss-up counties, and/or heavily Democratic counties. Republicans are best off siting their convention in places that lean modestly toward the GOP.

Consequently, Democrats would appear to do best by siting their national convention not just in “swing states” for electoral college advantage, but in media markets with (a) toss-up counties, where persuasion could turn GOP voters in their favor; and/or (b) heavily Democratic counties (Democratic votes above 78 percent). The RNC is generally best off siting their convention in places that do not lean too heavily in either direction, but which lean modestly towards the Republican Party.

Do any local benefits from the convention actually affect whether candidates win a state, or even the election itself? Presidential elections results are so rarely razor-thin that several percentage points in one media market will tilt the outcome, but our analysis suggests that that in 2000, Al Gore would have won the election had either the Democrats or Republicans held their convention in  Miami or Tampa.

In other election years, parties could have “flipped” certain states, but not the outcome of the entire election. For example, Democrats could have won Massachusetts (by siting the convention in the Boston-Manchester media market) or Delaware (Philadelphia media market) in 1980 and Missouri (Kansas City media market) in 2008. The GOP could have won Minnesota by siting in Minneapolis-St. Paul in 1984, and in 1992 they could have won Georgia by siting in Atlanta.

So what does our analysis suggest about the 2016 convention sites? We are reluctant to make confident predictions given the unusual features of 2016 — the potential for protests at both conventions, Trump’s unusual candidacy, and the historic unpopularity of the nominees.

But if the past is any guide, the easiest prediction is that the Democrats will see little Electoral College impact from the parts of the Philadelphia media market in safely blue Delaware, Maryland and New Jersey. In the swing state of Pennsylvania, the Philadelphia media market is the kind of place our analysis suggests the Democrats should pick: heavily Democratic counties in 2008 and 2012 where the convention can mobilize and narrowly Republican counties where the convention may help persuade voters.

By contrast, the RNC’s selection of Cleveland may not help the GOP much. The media market is largely composed of Democratic counties, where we would expect a backlash to a typical Republican convention, and only a few moderate Republican counties where the convention might help the party. If the RNC’s goal was to influence the results in Ohio, the choice of Cleveland provides more risk and less reward than Columbus or Cincinnati.

Christopher B. Mann is assistant professor of political science at Skidmore College, where he focuses on elections and political participation. Joseph E. Uscinski is associate professor of political science at the University of Miami and the author of American Conspiracy Theories.