I noted earlier that the Democratic losses in the House were less than expected given what usually happens during the midterms. It is harder to make such a statement with the Senate because of the lower numbers, with fewer than a tenth of the total number of elections at stake when compared to the house. But, there is a pattern that makes the loss of a few seats in the Senate not unexpected. As is the case with the House race, the null model — what is expected despite any other political factors — is that this particular year for Senate races would favor Republicans when a Democrat is in office.
President Obama mentioned this just before the election, as I recall, but the press ignored it, possibly because it is a little hard to explain.
There are 100 Senators, each elected for a 6 year term, and some are up for election every two years. The Senate is divided into three classes, about one third of the Senate in each, distributing them evenly across these two year intervals. This year, Class 2 was up for election.
Apparently a sample of 100 Senators divided into three parts does not produce even and identical results. The following table indicates the average percent of the vote among Democratic and Republican voters that went to Obama-Biden for the states represented by each Senate Class. The values are similar, but not identical. The mean for Class 2 is slightly lower, as is the maximum.
This graph shows the data in more detail. Class 2 is not an Obama-Biden friendly set of states, as a sample, because it lacks a peak representing a small number of strongly supportive states. Class 3 has other problems; it has a few very unsupportive states. Clearly, when it comes down to just a few races (which is what happens when there is a close Senate) Class 1 is likely to be friendliest to a Democratic executive (both high maximum and lack of low support) while Class 2 and 3 are less friendly, with Class 2 possibly being the least friendly (we can assume the low value states would be Republican anyway; it is probably the upper part of the curve that matters more).
The total number of Senate seats lost was low, and the chance of losing some were relatively high. So, as is the case with the House losses, what happened this year was more or less expected, and not exceptional. This was not an historic loss.