North Carolina has been conducting a real-time experiment on the efficacy of voter suppression—and the results suggest both that it works, and that it’s far from the only factor at play in the state.

“Voter suppression” has become a watchword in the state after a multi-year fight against the 2013 HB 589 voter law, a law championed by Republicans that reduced the early-voting period for elections by seven days, ended same-day voter registration, and established strict voter ID requirements. Those provisions—which have been demonstrated to affect minority communities most—were struck down by the Fourth Circuit Court of Appeals this summer for naked discrimination. But data from North Carolina’s State Board of Elections on the early-voting period indicate that continued efforts by Republicans at the county levels are having real effects on the election.

The general strategy that many Republicans pursued to make party-line changes to the first week of early voting were outlined in correspondence from state GOP executive director Dallas Woodhouse. Several county boards of elections eliminated Sunday voting during that week and severely curtailed the number of polling places and their hours of operation. Three additional county boards of election were successfully sued by the state NAACP for discriminatory purges of black names from voter rolls. Both the Fourth Circuit’s rulings, and the pattern of decisions by counties in circumventing it, suggest that Republicans were trying to gain an electoral advantage by reducing the number of African American voters.

Whatever the intent of these changes, though, it is difficult to calculate the impact of laws that reduce turnout and make it harder to vote. It requires teasing apart the effects of any given new law from the latent effects of historical disenfranchisement, a low baseline level of minority turnout, and the motivations of voters in a particular election. Conversely, that’s also the difficulty with simply comparing  black turnout from 2008 and 2012 to turnout in 2016; it’s impossible to know if the historic turnout in those years was an irreplicable surge in response to a historic candidacy, or a new normal that suggests black people are having an easier time voting now than in the past.

Unwittingly, perhaps, Republicans in North Carolina provided a natural experiment in their response to Woodhouse’s request that sheds some light on these dynamics. In all, 17 county plans reduced polling places or hours during the first week of early voting; others remained unchanged. That makes it possible to see the actual effects of those reductions on turnout. Alongside those 17 counties with first-week restrictions, I included counties identified in this report as having more than three fewer polling places than in 2012 for the entire duration of voting, and the three counties—Beaufort, Cumberland, and Moore—sued by the NAACP for voter-roll purges.

Although 2012 is an imperfect baseline, it does allow for comparisons between racial groups. In the counties marked as “suppressed” on the map, both black and white turnout suffered over the first week of early voting. But it rebounded after the first seven days, when counties were mandated to provide a full complement of voting sites. Black voters in those counties suffered the worst, with day 7 cumulative turnout only reaching 60 percent of the cumulative turnout at the same point in 2012. But white voters in those counties suffered too—notably at a worse rate relative to their unsuppressed counterparts than black voters—though they still managed to increase their vote share over the course of the first week. Voters in other counties did not see such sharp reductions over the first week, which is a rather clear indication that reducing the availability of polling locations and shortening their hours reduces turnout. The Republican scramble to restrict access to the polls at the last minute appears to have hurt white voters and black voters alike.

But turnout numbers for both black and white voters in these “rogue counties” did bounce back after the first week of voting, and both eventually caught up to turnout levels in other counties. This could be attributed to the considerable organizing effort in the second week of early voting by several groups, including the state NAACP, and to a surprisingly strong Republican turnout machine. But although white voters generally soared above 2012 early-voting numbers, in all counties black turnout appeared to reach a ceiling of around 90 percent of 2012’s results.

Though county-level efforts to suppress votes appear to have been ameliorated by organizing, activism, and a 10-day recovery period, that ceiling is what should concern Democrats in the state. While the state GOP’s celebratory press release about early-voting numbers cited a “crumbling and tired” state coalition and flagging black enthusiasm as reasons for this ceiling, it’s simply impossible to separate lower black early-voting turnout from the GOP’s three-year campaign to reduce turnout. That campaign was ostensibly halted by the Fourth Circuit Court’s decision, but it left voting-rights activists with only months to mobilize an electorate already facing barriers and conflicting information. And then, massive flooding from Hurricane Matthew set back those efforts.

These maps show turnout effects in the rogue counties by race. The eastern third of the state illustrates how barriers to voting come naturally along racial lines—and how the same events can have disparate impacts. The lowest-lying portions of these counties are the most prone to flooding, generally hold the lowest-income communities, and are disproportionately populated by voters of color. In the counties hit hardest by Hurricane Matthew, white early-voting turnout still managed to increase by large margins relative to the same period in 2012. But black voting in those counties tapered off. Notably, Robeson County—much of which was underwater well into October, and which had several polling places damaged and relocated—saw large increases in white early-voting turnout, but also the largest depression of black turnout in the state. By the last day of early voting on Saturday, black voters had only reached only 75 percent of 2012 turnout levels.

What do these results mean for Election Day? For one, the first-week returns provide a proof-of-concept for the efficacy of voter suppression efforts. Also, though early voting is not necessarily a predictor of how people vote on Election Day, these results should buoy Donald Trump’s chances of winning North Carolina. Regardless of the outcome, the 90 percent turnout ceiling should concern voting-rights activists for years to come, especially in a race in which Florida’s black early-voting numbers actually jumped past 2012 levels. What is it about North Carolina that seems to have especially depressed black turnout?