Conor Lamb, a Democrat who won a congressional seat in western Pennsylvania last month, mingles with supporters after giving his victory speech. (Salwan Georges/The Washington Post)

Over the course of 93 special elections since the 2016 presidential contest was settled, Democrats and Republicans have split the results, according to data compiled by Daily Kos. Democrats have won 47 races; Republicans, 46. But that doesn’t tell the whole story. Going into those contests, Republicans held 65 of the 93 seats. So the Democrats have picked up 19 seats in those special elections.

That includes races at both the state and the federal levels. Many of those wins for the Democrats were a function of off-year turnout dropping less for them than it did for the Republicans. In other words, while both parties saw fewer people come out and vote in the special elections, relative to 2016, the drop-off for the Democrats was less. But that isn’t always the case. The special election in Pennsylvania’s 18th District had fairly high turnout on both sides — but the Democrats won anyway.

The question, of course, is what these particular tea leaves tell us about the midterm elections in November. It would be natural to assume, given Republicans losing 30 percent of seats they held and Democrats losing zero percent of theirs, that the House might lose 30 percent of their existing seats, easily handing the chamber to the opposition. But does a state legislative race in New Hampshire actually tell us who will win a House race in Mississippi?

Here’s what we did instead. Looking only at House special elections, we took two different measures and applied them to each of the 435 contests that will be determined in November. Then we took it a step further.

Shift relative to 2016 presidential margins

One of the ways in which the special-election results have been measured is by comparing the results in each contest to how Donald Trump and Hillary Clinton fared in the same area. Daily Kos has good data on the presidential vote in House districts that allows us to see the shift between 2016 and the special election.

For example, Georgia’s 6th Congressional District voted 2.1 points more Republican in last year’s special election than it did in the 2016 presidential race. Montana voted 15 points more Democratic in its special House election than in 2016. If we take each of those shifts from the seven House special elections (in which Republicans held five seats and lost one and the Democrats held one), we can evaluate how a similar shift in each House district would affect November’s races.

There’s a wide spread.

The Democrats win four of the seven scenarios. (This analysis uses Pennsylvania’s old district boundaries, for what it’s worth.)

If the shift in each House district is equivalent to Georgia’s 6th District, the Republicans come out with a 39-seat lead. If it’s equivalent to Montana, the Democrats have a 135-seat lead. And those aren’t even the extremes. Applying the shift in Kansas’s 4th District nationally gives the Democrats a 205-seat lead. Applying the shift in California’s 34th District, the Republicans end up with a 113-seat advantage.

Applying different shifts leads to two very different maps.

Let’s set this aside for a second and look at the other metric.

Shift relative to 2016 turnout

The shift relative to the 2016 presidential vote is fairly rough. The reason the best map for the Democrats was the Kansas district is because that district voted heavily Republican in the presidential race, so it was easier for a Democratic House candidate who wasn’t Hillary Clinton to outperform her. The opposite is true in that California district.

So we applied another metric. What if we looked at the change in turnout relative to 2016 in each House district and applied it to the number of votes for each candidate in the House race in the district in 2014?

That measure results in a lot less volatility. (The districts in Pennsylvania and California drop out of the equation because the winning candidates in those districts in 2014 essentially ran unopposed.)

Here, the Democrats win three of the five scenarios.

Instead of looking at maps, let’s cut to the chase: What if we average out the results from the House special elections and apply the same math to every House district?

Using the average shift relative to the 2016 presidential vote, the Democrats would gain a 35-seat majority. Applying the shift in turnout (or, really, votes relative to the 2016 total), the Democrats come up with an 18-seat majority.

This should not be considered a prediction as such, so much as it is an attempt to get a sense of the direction the special elections are pointing.

President Trump likes to tout that his party went 5-and-0 in special House elections before the Pennsylvania race last month. (It was actually 5-and-1, including the California race that he likes to omit.) But just as the split between the Democrats and the Republicans in those 93 total special elections is misleading, so is Trump pretending that the special House races show a positive result for his party.

Applying the shifts in those races nationally means his party ends election night in November having lost control of the lower chamber of Congress.