The most convincing piece of data that things are looking very bad for Republicans in November

Fortunately for the political junkies who have a distrust in the polls after the 2016 election, there is another way to forecast the midterms — using actual voting data. We can look at all the special state and federal elections. Congressional contests, like Doug Jones' and Conor Lamb's big wins in Alabama and Pennslyvania, are in purple. State-level contests are in yellow.
By comparing the parties' performance against past elections, we can construct a baseline against which to gauge performance. Arizona's 8th district, for example, has voted for Republicans by an average of 25 pts more than the country in recent presidential elections. Republicans won the special election there by only 5pts, resulting in a 20pt Democrat overperformance.
When we look at every election in terms of performance relative to partisan baseline, we find that Democrats overperformed past elections in 83 of 114 elections. Democrats have overperformed by an average of 13 points in these elections.
Democrats are overperforming in districts, regardless of which way the district usually leans. But Democrats overperformance from the baseline has actually been greater in districts with a Republican lean.

These results suggest that the Republican Party is in trouble heading into the midterm elections. If past trends hold, it is possible Democrats could see a double digit swing in the average House district in 2018 compared with past elections.

The reason is fairly clear: The Democrats are outperforming their baseline far more often than they are underperforming it. The average swing across all elections has been +13 Democratic, signaling a national political environment is 13 points in the Democrats' direction.

In Arizona’s 8th congressional district special election on Tuesday night, Republican Debbie Lesko won by 5 points. That’s considerably less than the 25 percentage point partisan baseline and is another bad sign for Republicans heading into the midterms.

A baseline for partisanship is constructed by looking at the 2016 and 2012 presidential results. This lets us read which party elections are swinging towards.

The last time the average shift from the partisan baseline in federal elections looked anything like this cycle was in 2006. That year Democrats won the national House vote by 8 percentage points and had net gain of 30 House seats. Democrats need a net gain of 23 seats to take back the House this November. Like 2006, we shouldn’t expect Democrats to win the national House vote by nearly as much as they overperformed in the special elections.

Past cycles instead suggest that given the Democratic performance in special elections that they should win the national House vote by around 11 percentage points. That estimate comes with a 95% confidence interval such that a result of Democrats winning the national House vote between 3 points and 21 points is plausible. The bottom of this range would mean that Republicans most likely would maintain control of the House, while a 11 point win would easily give Democrats the majority.

It’s now about six months until the November election, which is still a lot of time in an election year, but there also doesn't seem to be much of a time trend in the Democrats’ performance in special elections.

While Democrats have done better in some months than others, it seems to be statistical noise more than anything else. Democrats outperformed their 13-point overperformance in March 2018 by 4 points and have been underperforming it by 6 points in April 2018. This Democratic lead on the generic congressional ballot in was similar in both months, after declining from the end of last year.