I want to thank everyone for voting on my January 25th diary. I’ve received 28 votes (and counting), and without your help my current prediction model couldn’t have been created. So thank you all for voting!
Results from my January 25th Diary as of January 28th, 2:15 pm EST:
Which past mid-term election is most similar to the 2010 mid-term election?
10 votes for 1982
9 votes for 1978
4 votes for 1966
2 votes for 1938
2 votes for 1994
1 vote for 1946
Interestingly, 1982 was the only mid-term election above where the sitting President was a Republican! Not sure if this is significant, but would love to hear other opinions.
Creation of Tarheeman1993’s 2010 mid-term election model
I decided to base my formula on the results received from this survey. Each vote gets treated in the same manner as the results for the respective past mid-term election. At first glance, the model would look like:
2010 mid-term model predictor= (10/28)*(1982 mid-term election results) + (9/28)*(1978 mid-term election results) + (4/25)*(1966 mid-term election results) + (2/25)*(1938 mid-term election results) + (2/25)*(1994 mid-term election results) + (1/25)*1946 mid-term election results).
However, this in itself could be unreliable. I first had to see what the weighted average of seats up for re-election that were held by the current President’s party before the mid-terms took place in each given year. Note: except for 1982, the President in office for the respective year was a Democrat:
Seats up for re-election in past mid-term elections held by that President’s party on the given mid-term election:
Year House Senate
Seats Seats
1982 192 13
1978 292 18
1966 295 20
1938 334 30
1994 258 22
1946 242 24
Using my model, the average # of seats can be computed as follows:
House avg=(10/28)*192 + (9/28)*292 +(4/28)*295+(2/28)*334 + (2/28)*258 + (1/28)*242=255.50 seats. Currently the Dems will have 257 seats up for re-election, a difference of 1 1/2 seats!
Senate avg=(10/28)*13 + (9/28)*18 +(4/28)*20+(2/28)*30 + (2/28)*22 + (1/28)*24=17.86 seats. Currently the Dems will have 18 seats up for re-election in the senate, a difference of .14 seats.
Since I was satisfied with the above results, I decided to look at the seats that were lost by the President’s party in each given mid-term election. Here are the actual numbers of seats lost:
Year House Senate
Seats Seats
1982 27 0
1978 15 3
1966 48 3
1938 72 7
1994 54 8
1946 55 12
Using my model, the average # of seats lost by the President’s party in the mid-term election (net) can be computed as follows:
House seats lost prediction=(10/28)*27 + (9/28)*15 +(4/28)*48 + (2/28)*72 + (2/28)*54 + (1/28)*55= 32.29 seats.
Senate seats lost prediction=(10/28)*0 + (9/28)*3 +(4/28)*3 + (2/28)*7 + (2/28)*8 + (1/28)*12=2.89 seats.
Obviously this model has flaws (specifically that I’m creating it and that I’m only asking fellow SSP members their opinion). However, since the 2010 election is complex, I’m asking whether the end result, 32 seats lost in the House and 3 seats in the Senate is accurate.
Please vote on this survey and give me your feedback. I’d appreciate it!
Try multiplying both of those numbers by 2.
– generally accurate for House, but in Senate i expect greater Republican gains – really 5 or 6
Wow, you’re Mr. Optimism, given that Republicans are polling ahead of the margin of error in Arkansas, Colorado, Delaware, Florida, Kentucky, Missouri, Nevada, New Hampshire, North Dakota, Ohio, and Pennsylvania.
but the ratio of Senate seats lost to house seats lost isn’t all that consistent, and considering the 1982 figures are an outlier with 0 senate seats lost it was bound to skew your results. Factor that with 1982 getting the most votes and your Senate projection was bound to be “optimistic”.
Without a larger sample I don’t think there’s any correlation analysis or removing of outliers you could do to make the model work any better.
Maybe you could do something to take a look at all of the years and see the average number of senate seats lost per house seat, and then have your model work the way it does now for house seats but have senate losses as a function of house losses. Of course this is only done because the house figures appear more consistent and I cant say as though it would really improve the model, just make it more towards what my mind tells me it will be (more negative)
Or maybe you could look at percentage of incumbent senate seats lost and use that to get a vote-weight-averaged percent and apply that % to 2010’s count of 18 Senators up for re-lect. Still, because the highest vote getter (1982) had 0 incumbents losing, your model will be optimistic no matter what.
Still fun to look at and think about, reminded me of some old stats classes (which, to be clear, weren’t fun).