Q3 House Model Update Part 2: Navigating the Wild West of Campaign Finance
To see the first part of this with the full model update, click here: https://medium.com/me/stats/post/dd8757281af2
The interactive downloadable version of the model is here: https://1drv.ms/x/s!AvfJM5SfJyk7hVFaln5R43JkWE79?e=RI4zd0
Current spending for competitive House races has hit $1.3 billion, and will probably finish a bit under the 2018 total. When leadership and safe spammy races are removed, there are 65 where total spending has eclipsed $5 million and the full chart is below. Unlike 2018 though, Democrats do not have large spending advantages in every one of the top races. All data here comes from the FEC’s website or from the aggregator OpenSecrets.com
It’s part of why the model still thinks that there are a lot of competitive races where Democrats are outspending Republican candidates but outside groups are propping them up. The next chart shows the total spent but then what share of the spending from each party comes from outside groups. For instance, if the GOP candidate is spending $1 million and outside groups are spending $4 million, then the GOP Spending IE % would be 80%.
DMA Effect
I talked about adding in this effect back in my initial launch so I don’t want to spend too much time on it. It is one that I am constantly tinkering with because I believe that because this directly informs spending strategy it is crucial to get as right as possible. In 2018, I noticed an error in the direction of spending in both districts that had big markets or districts broken up between several markets. Because of this I now am putting a decent fraction of spending through a factor that takes into account both of these in order to reduce cash spent where I believe the power of the dollar doesn’t go as far. So if you’re adding cash to the model and find that the margin doesn’t move as much in a California seat compared to Arkansas’ 2nd, this is why.
Wasteful Spending
In 2018 some of the biggest misses due to cash were based off for the actual spending behaviors. While I don’t have all of the fixes I am implementing a new look at some of the practices throwing the model for a loop. To start, before I just used all of candidate spending and a few interesting candidates stuck out to me.
The first were the Team Players. This is mostly leadership but also a lot of people in safe seats raising cash or with higher ambitions. These people do spend but it’s mostly transfers to state or party committees, or their fellow candidates. For example, Northern Virginia’s Don Beyer is not in any trouble but he’s spent over $1.9 million this cycle. When you dig into his FEC file, you discover that nearly a million of that was in contributions. Obviously, the model can not tell the difference and assumes that goes to helping his campaign. In 2018, the model was about 4 points too Democratic in its estimate, which would have been cut by more than half with those filtered out. Starting now I have filtered out anything contributions or transfers from the total cash spent.
The second is that it could make for a decent minor scandal variable. If there’s an indictment, I include a 15-point penalty from the few of those we have in the past that come out to about there. More minor scandals however were a bit trickier. 538 had a measurement but it only introduced more error because the scandal and its impact were both subjective. I initially tried looking only at size or share of money raised that became a refund but it didn’t clear much up. Finally, it hit me. David Schweikert is the representative for Arizona’s 6th, who was under a congressional ethics investigation and this cycle was reprimanded and fined. While last time he faced a low spending opponent to win, the model gives him only a slight edge this time around. The model originally looked at his spending of $1.4 million and gave him a decent boost from that to shore him up. This version was not as forgiving. Below shows his top disbursements amount spent in each category.
By subtracting refunds and legal fees or legal consulting, it becomes possible to introduce a new scandal variable. One can also include that the bigger the scandal the more legal help will be needed so this fix can hopefully provide a good measure on the scale of the issue and penalize them more.
And last are the list churners which are the hardest to nail down on total cost and throw havoc in the model. These are the field reps that raise a lot of money but they do it by burning through lists and focusing on heavy direct mail costs. In the model you will notice that in TX-2, Dan Crenshaw is in a close seat up-ballot but expected to win by 20. This is because he is spending like crazy, with millions and millions. When you look at where it’s going though, it shouldn’t help him. The chart below shows how much goes to feeding that fundraising operation and postage alone for Crenshaw through the second quarter. Other representatives like Brian Mast do the same thing.
Unchallenged Underachievers
If there’s one thing people noticed, judging from the replies, there are a series of seats that are really Republican friendly despite the presence of an opposition campaign. These seats include but are not limited to: ME-2, MI-8, NY-19, WA-8, and maybe throw in MN-2 or NJ-3. Though they are a mix of districts, they all have something in common beyond being just flips with most in Trump friendly districts: all underperformed in the 2018 model. Now they all won, some by decent margins but the 2018 version of the model believes that having outspent in their races last time, they really should have won by more. Adjusting for spend in previous years, a practice that works excellently and I wrote up here, and so docks some points this time around if they don’t hit the same levels. While this works from high profile race to high profile race, it underestimates the incumbents when they should be facing a challenge but instead face a challenger with no cash and support. Where in 2018, someone could have won by a few points while outspending by a larger gap of $7 million to $4 million, if they are only spending half as much this time but their opponent has such a small number, the model still thinks they should do worse.
One solution I’ve been working up would be to look also at the ratio of cash spent against your opponent and testing that. The other is a ‘giving up,’ variable that only looks at that ratio or adds a fixed margin based on a predetermined set of criteria of how much cash you’re facing from the candidate and outside groups. Considering that these will all have some error, I could add this fix in now but it honestly it would feel like cheating and because we don’t know the outcome it may not even work. Instead if this group does have a systemic error, I will just take the hit on them and then happily add it to the next version when it’s less rushed and the rules have been properly verified in previous models as well.
Conclusion and Updates
There is a lot to sift through that skews any model that includes campaign finance. However, given past results I still believe that it would be negligent to ignore total spending completely and none of the fundraising shortcuts seem to cut it. This means going through those with legal fees, contributions to colleagues, or list chuggers. It also means constantly adapting. Money is, in my opinion, not only one of the most critical factors but also one of the only things to actually change a race. If I hope to fine tune this as a model that helps determine battleground states early or spending strategy for players such as the parties or IE funds then it is critical to always be exploring and trying new things.
