On average, from 1934 -2014, at midterm elections, the President's party has lost 27 House seats and 4 Senate seats.
Using the data we provide in our data archive "Seats in Congress Gaines/Lost by the President's Party in Mid-Term Elections" we estimate GOP seat losses of 33 in the House and 1 in the Senate. Of course, these data are aggregated and take no account of local political conditions, candidates, or issues.
Partisan seat swings in midterm elections are consistently correlated with two familiar factors. The first is the public's assessment of the incumbent President. The second is the number of seats in Congress controlled by the President's party.
In our table of midterm seat swings, starting with the 1934 election, we provide historical data on seat swings and seats defended, both in the House and the Senate. We also provide data on public approval (Gallup poll) at various points prior to the election.
Using four graphs (one displayed with this text), we show the relationship between seat swing and both approval and seats defended. In all cases the relationship is pretty clear. The more seats the President's party has to defend, the more they lose. The lower the President's public approval, the more seats his party loses.
What about the combined effect of approval and seats to be defended? They both matter.
For the House, most of the work is done by approval. The joint prediction of a GOP loss of 33 House seats is the same as for approval alone (but the "fit" is better). The range of the prediction in 2018 is for a GOP loss of between -15 and -51 seats. (In 2010, the House Democrats lost 63 seats with Obama approval and seats to defend similar to Trump's in 2018.)
For the Senate, approval and seats together predict a 2018 GOP loss of one Senate seat. The error range of that prediction is between -5 and +2.