# Tracking Congress in the Age of TrumpFebruary 14, 2017 1:48 PM   Subscribe

An updating tally of how often every member of the House and the Senate votes with or against the president. The "votes" tab, organized by bill/nomination, is particularly useful.
posted by materialgirl (35 comments total) 55 users marked this as a favorite

Good for you, Tammy Duckworth.
posted by daisystomper at 1:58 PM on February 14 [4 favorites]

Oh, this is quality. I was able to call up Senator Cardin's office just now and say, "I see that Senator Cardin has voted in support of Donald Trump's positions 43.8% of the time since the inauguration. By contrast, Senator Van Hollen has only voted in support of Trump's positions 25% of the time. Senator Cardin needs to get those numbers down, or we'll have to find a more reliable progressive come the elections."

Live by the metric, die by the metric.
posted by Faint of Butt at 2:07 PM on February 14 [21 favorites]

This metric seems overly simple: “because Trump got X% of the vote in your district, we expect you to adopt Trump’s position X% of the time.” But there are all sorts of weightings that matter here, since types of bills and of how much interest they are to the populations that elected you aren’t uniformly distributed. (I assume.)
posted by Going To Maine at 2:27 PM on February 14 [1 favorite]

Going to Maine, I absolutely agree that that would be overly simplistic. I am not the person to ask re: the soundness of this model (which is partly why I wanted to share this here, so some savvier people could weigh in and assess), but I think the plus-minus metric is meant to roughly approximate the extent to which congresspeople deviate from their constituencies:
We’re also calculating a metric that we’re calling plus-minus. Plus-minus measures how frequently a member agrees with Trump compared with how frequently we would expect the member to, based on Trump’s 2016 vote margin in the member’s state or district. (The “predicted score” is calculated based on probit regression.) Put simply, we would expect a member in a district where Trump did well to be more in sync with him than a member in a district where Trump did poorly. As members vote on more bills, their predicted agreement score will change.
posted by materialgirl at 2:47 PM on February 14

I don't understand the last two columns, the likelihood and the plus-minus ones, for individual members. If these are all past votes, and they appear to be, what's this "likelihood" for? Of course Gillibrand is likely to vote 100% in agreement with Trump about Shulkin; she already voted in agreement. Isn't that like saying there's a 100% likelihood of me having woken up this morning?

I am not at all good at statistics. :( If there's somewhere where this is explained in smaller words, I'd be very grateful.

This is definitely useful at seeing all the votes in one place, though.
posted by XtinaS at 2:53 PM on February 14 [1 favorite]

Thank you Congressman Al Green.
posted by neutralmojo at 3:07 PM on February 14

I don't understand the last two columns, the likelihood and the plus-minus ones, for individual members. If these are all past votes, and they appear to be, what's this "likelihood" for?

The "likelihood" column is basically stating a hypothesis that reps from districts that swung for Trump are more likely to support him with their votes than reps from districts that swung for Clinton. It's a really naive hypothesis that really didn't need testing, but that's the basis of the "likelihood" scores.

But I can see why they'd do it, basically they're using the results from the 2016 presidential election as a proxy for how "safe" each seat is. They aren't using congressional election results to measure "safety" directly because not all Senators are up for reelection at the same time.
posted by tobascodagama at 3:26 PM on February 14

Boozman and Cotton, 100%.

I already knew they were heartless and bought out, but I do not enjoy seeing it quantified.
posted by middleclasstool at 3:40 PM on February 14

But there are all sorts of weightings that matter here, since types of bills and of how much interest they are to the populations that elected you aren’t uniformly distributed. (I assume.)

I think there's a clear advantage to the simplicity of the model, which is that it's not dependent on any outside assumptions about the individual votes. In order to make the sort of judgment you're suggesting, you'd either need extremely granular data (which doesn't really exist, because polling by congressional district is complicated and expensive) or you'd basically have some expert making ad hoc judgments of the form "Well, rural districts care more about passing the agriculture bill, so we'll bump the likelihood of support for those districts by 10 points." At a certain point you add up enough fudge factors and the model is more reflective of the biases and assumptions of the person who makes it than it is of the underlying data. Ultimately with a simple model, once you get a big enough database of votes, all of the little eccentricities should come out in the wash.

But I can see why they'd do it, basically they're using the results from the 2016 presidential election as a proxy for how "safe" each seat is. They aren't using congressional election results to measure "safety" directly because not all Senators are up for reelection at the same time.

The other big problem is that local effects can screw with your data. For example, Tom Price's district (GA-6) only went for Trump by a point and a half, but Price won by more than 20 points against a Democratic opponent who didn't campaign at all (literally the only things you can find about him online are that his name was on the ballot, and he didn't spend a single penny from his campaign account). Presidential vote margins are a nice proxy because people usually pay attention to the Presidential race, and we know that both sides run vigorous campaigns. Of course 2016 was a weird election in a bunch of ways, and you might want to use a broader range of presidential elections in calculating these. (That's what Charlie Cook does when he calculates PVI). But I could easily argue both sides of that particular question. For the House races, where everybody will be up for re-election in 2018 and 2020, it seems like a decent assumption that the dynamics of the 2016 race will still be in play, since Trump will presumably still be in the White House.
posted by firechicago at 3:58 PM on February 14 [2 favorites]

I'm onboard with the idea that this is too simple to be of real value. The measure of a representitive is definitely not how often they vote in opposition to the other party.
posted by grumpybear69 at 4:26 PM on February 14 [2 favorites]

The measure of a representitive is definitely not how often they vote in opposition to the other party.

Under normal circumstances, no. But these are not normal circumstances.
posted by Faint of Butt at 4:34 PM on February 14 [7 favorites]

So, the "Trump Score" is fairly straightforward, and its soundness as a metric pretty much depends on how much you think "voting with/against Trump's stated position" is a useful way to evaluate members of Congress and/or individual bills.

The "predicted score" is much more difficult to interpret, as FiveThirtyEight doesn't seem to have given sufficient information to understand how it was calculated ("based on a probit regression" doesn't tell us what factors are going into that regression). Since two individuals with the same "Trump Margin" can have different predicted scores (compare Diane Feinstein and Kamala Harris: both share a Trump Margin of -30.1, but have predicted scores of 31.3% and 29.5%, respectively), the Trump Margin is clearly not the only factor informing the predicted score, but they don't say what the other factors are. Thus, the "predicted score" statistic is something of a black box; they have some sort of model for calculating a predicted value for each individual's Trump Score, but they aren't telling us what it is. It's therefore impossible to evaluate the soundness of their model a priori. Once we have a lot of votes in, we can evaluate their model empirically to some extent, but quite frankly the difference between "predicted scores" of 31.3% and 29.5% is almost certainly meaningless. I would be surprised if their model really captures any resolution finer than about top-third, middle-third, and bottom-third.

Since the predicted score is impossible to evaluate, the "Trump Plus-Minus" derived from it is also impossible to evaluate. Basically, it works out to "this is how bad our model was at predicting this person's votes." There's probably an automatic tendency (for those of us who oppose the Trump administration, anyway) to view someone with a particularly low Trump Plus-Minus score as "doing better," "working harder," or "taking more risks" to oppose Trump, but this isn't really what it means at all. Overall I don't think it adds any more information that what we already qualitatively know. Personally I would treat it at best as an amusing diversion, and at worst as an irritating distraction.
posted by biogeo at 5:13 PM on February 14 [3 favorites]

What, exactly, explains Dianne Feinstein's voting record?
posted by jeremias at 5:16 PM on February 14 [3 favorites]

The list of who voted for what is better than I've seen anywhere else. With some digging you can usually find out who voted on a specific bill but I haven't seen them presented all together like this. It's pretty great.

It's too bad they tacked on the silly prediction model. It's very flimsy reasoning that's supposed to look reliable because they throw lots of numbers at you. It strikes me as the sort of thing FiveThirtyEight would make fun of in the "good use of polling" segment of their podcast. (Which I haven't listened to since the election... Maybe they covered this?)
posted by Gary at 5:21 PM on February 14 [2 favorites]

John McCain talks a good talk of standing up to Trump, and gets a lot of air time as a mavericky maverick. But when it comes to votes, he's with Trump 100%.
posted by jetsetsc at 5:23 PM on February 14 [13 favorites]

> What, exactly, explains Dianne Feinstein's voting record?

She's always been on the conservative end of the D scale. Her record in this case isn't surprising.
posted by rtha at 5:59 PM on February 14 [1 favorite]

Thus, the "predicted score" statistic is something of a black box; they have some sort of model for calculating a predicted value for each individual's Trump Score

I'm pretty sure they're not predicting Trump Scores, just estimating Pr(vote with Trump), which in the frequentist world would turn pretty directly into a proportion of votes with Trump.

I agree that the predicted scores aren't very useful, especially as they're protecting whatever secret sauce they added. The very simple model Going to Maine would probably be a better, or at least more straightforward, measure for how much more Pro-Trump than your state or district you're voting.
posted by ROU_Xenophobe at 6:14 PM on February 14 [1 favorite]

Yeah, not that I was surprised, but... based on Kaine and Warner's respective track records so far, I will be making a few extra calls to their offices this week. No more of this 50% tomfoolery.

And next week when Congress is in recess, those two better host town halls. I don't currently see any scheduled for either of them. Not sure where the idea originated but the discussion on the Pod Save America episode yesterday was that if your representatives aren't hosting town halls next week, organize one and invite them to it, and take a bunch of photos of the crowd to post on social media!
posted by nightrecordings at 6:33 PM on February 14 [4 favorites]

John McCain talks a good talk of standing up to Trump, and gets a lot of air time as a mavericky maverick. But when it comes to votes, he's with Trump 100%.

only took you 33 years to figure him out too.
posted by any major dude at 9:18 PM on February 14

What, exactly, explains Dianne Feinstein’s voting record?

She’s always been on the conservative end of the D scale. Her record in this case isn't surprising.

The Feinstein votes that have jumped her up the scale are her support for the cabinet positions held by Mattis (two votes, the waiver and the nomination), Pompeo, Kelly, Haley, Chao, Shulkin, and McMahon. Feinstein, in fact, is at least part of the reason why I consider this metric kind of garbage. Maybe, once all of the the cabinet votes are taken care of, it will be less garbage, but I kind of assume not.

So, in my ignorant opinion, this metric seems great for someone who, say, is trying to identify a flippable district - if their trump score doesn’t match their record, maybe they can be kicked out because there’s some kind of mismatch with the general consensus. A blue dog / RINO who is engaged in horse-trading on votes with the opposition party (which, realistically, is what the President’s position will always be) should balance out nicely if they are doing a good job of trading positions, and should be decidedly lopsided if they’re going too much one way or the other. In aggregate - after many more votes than are currently in the system, it may give some nice results.

That said, to me as an individual voter this is unhelpful. The cabinet votes highlight this problem: votes in favor of Mattis, DeVos, and Haley all qualify as votes “for” the President. But Mattis was eminently qualified for his position, while DeVos and Haley were not. As a decidely anti-Trump Californian, I’m entirely content with Feinstein supporting Mattis, and a token vote against him would seem to be a waste of time - the sort of idiocy that gets me riled up about the Freedom Caucus. In contrast, a vote for Haley cheeses me off because she’s markedly inexperienced. She got nearly 100% of support from all of the Democrats, so perhaps there’s political thought going on here that I’m unaware of, but on the surface it seems bad. The same with DeVos, but here we can say that the party did the right thing in opposing her strongly. But of course, education secretary is generally considered a “meh” position - DeVos got hit hard because she was so bad, but the appointment is waaay less important in the big picture than Mattis. So there’s an argument that you should care a lot more about the Mattis vote (for or against) than the DeVos position. On the other hand, there was a lot of public pressure to “do the right thing” on DeVos. So perhaps a better metric would be how well the Senator seemed to respond to the DeVos position. (Also, to point to another imbalance in the metric: Lisa Murkowski voted for DeVos to get out of committee and onto the Senate floor for the nomination vote. Then she deliberately reversed herself and voted against the nomination. That first committee vote isn’t included in this metric, but that vote was arguably more important than the second. Certainly it’s an important signal to Murkowski’s voters about the impact of yelling at her.)

In threads we’ve had here on the Supreme Court, I’ve often taken the position that hey, the justices really agree a lot more than they disagree. This gets a lot of (reasonable) harrumphing from others and the counterpoint that it’s the big controversial cases that matter. In the case of this metric, I feel like I’m the one doing the harrumphing. Most of these votes aren’t particularly important and don’t show very much.

John McCain talks a good talk of standing up to Trump, and gets a lot of air time as a mavericky maverick. But when it comes to votes, he's with Trump 100%.

The problem for Republicans, meanwhile, is that they can’t really do a lot to distinguish themselves for Trump. The President’s position is going to almost always be the standard Republican position. Republicans who sincerely hold their party’s positions are going to probably take them most all of the time. So unless Trump starts to really flip himself on crazy axes (which would make trouble for the Democrats) the Republicans are going to have a high level of agreement with Trump.

In the case of McCain and these cabinet votes, I -again- really don’t care about support for someone like Mattis (though others might). On the other hand, the idea that he wouldn’t flip on DeVos despite massive public outcry is a serious black mark.

One kind of fun thing about this is that it might tell us how gerrymandered particular congressional districts are. A good gerrymander get a solid 60-40ish split so that the other party’s votes are wasted. This will presumably show up in the Trump score, so congresspeople who are really over/under on their scores might be in gerrymandered districts.

Basically, I guess right now I’d rather see the Trump score correlated against DW-Nominate scores or some other metric - how partisan are you relative to how partisan your district is? Or at least wait to use this until we get a few further months into the presidency.
posted by Going To Maine at 10:19 PM on February 14 [3 favorites]

The elephant in the room – and it is an elephant, as opposed to a donkey – seems to be that he is doing a terrible job as president, with horrible poll ratings, yet the score so far is Trump Support 24, Trump Opposition 0.
posted by LeLiLo at 12:18 AM on February 15

The reason why a metric like this matters right now is because the Republicans have been at total war against bipartisan and democratic values for over a decade and the Democrats are still saying "peace in our time."
posted by Skwirl at 12:18 AM on February 15 [7 favorites]

DW-Nominate already provides information about the increasing partisanship in Congress by looking at how votes cluster. This seems to add nothing to that approach, and is perhaps dumber. It approaches the problem of evaluating how much your representative is compromising relative to your district's makeup, but doesn't seem like it's going to differ much from existing rankings of how often a politician votes with the opposition party.
posted by Going To Maine at 12:45 AM on February 15

Also - I'm not certain that angry progressives need to get more riled up. In many ways, I think the metric helps justify yelling at Republicans who come from anti-Trump districts and who should be deviating from the party line more often.
posted by Going To Maine at 12:49 AM on February 15

Michael E. Capuano, 7th MA, 0.0%

I knew I loved my rep for a reason.
posted by explosion at 7:56 AM on February 15

I'm a little sad not to be in Capuano's district any more, but Katherine Clark is also pretty awesome. So I'm not too sad.
posted by tobascodagama at 8:07 AM on February 15

I agree that the predicted scores aren't very useful, especially as they're protecting whatever secret sauce they added. The very simple model Going to Maine would probably be a better, or at least more straightforward, measure for how much more Pro-Trump than your state or district you're voting.

I don't think there is a secret sauce; the reason they didn't provide any more explanation is that they literally are using the very simple method mentioned by Going to Maine; the "predicted score" is based purely on the Trump margin of victory, using a probit regression. A probit regression is useful because it does a good job with binary choice data - these are "yea" or "nay" votes rather than a sliding scale - and because it produces an "s" shaped curve; this makes intuitive sense - the difference in behaviour we expect in going from NC (+4% Trump) to safe NE (+25%) should be greater than the difference from NE to ultra-safe WY (+46%). If NC senators are predicted to vote with Trump near the median of 70% and WY senators are predicted to 100% vote with Trump, then a linear model would suggest that NE would be halfway between, at 85%; instead the probit regression says they are 96% likely to vote with Trump.

This figure shows for senators (as of after the McMahon vote) the Trump margin versus the predicted score; for most states the two senators (coloured by party) are overlaid on top of each other - the few outliers appear to be associated with missed votes. For example, the two dots above the curve in the lower left are Feinstein and Markey (MA), who missed the controversial anti-ACA and energy company foreign bribery votes respectively - since they missed a close vote, but not any of the safer votes (Mattis or Haley for example) they are "predicted" to vote a little more with Trump than otherwise. The others are Coons (DE) at Trump -11% and McCaskill (D-MO) at Trump +18%, who missed the controversial Tillerson and Price nomination votes respectively.

I don't think it's as helpful to say that the model results are predictions, even though that is fairly standard statistical terminology (see also: "significant"). I don't think they are trying to accurately predict the results of votes so much as provide a benchmark - Feinstein is supporting Trump as much as Democrats from very competitive states even though she's from California; what's up with that? Tester is standing up against Trump much more than someone from red Montana would, yay! Look at all the Republicans backing Trump 100%, even those from pretty competitive states - party over country! and so on.
posted by Homeboy Trouble at 9:13 AM on February 15 [2 favorites]

This figure shows for senators (as of after the McMahon vote) the Trump margin versus the predicted score; for most states the two senators (coloured by party) are overlaid on top of each other

Huh, so they are. I'd assumed they'd at least put party in too.
posted by ROU_Xenophobe at 9:28 AM on February 15

Feinstein and Markey (MA), who missed the controversial anti-ACA and energy company foreign bribery votes respectively

Feinstein, incidentally, missed the ACA vote because of a medical emergency. That won't be the norm for abstentions, of course, and is another argument for waiting a few months before thinking about this.
posted by Going To Maine at 9:47 AM on February 15 [1 favorite]

Michael E. Capuano, 7th MA, 0.0%

I knew I loved my rep for a reason.

I clearly need to step away from this because I have a fixation, but to try to get closer to what I’m pointing to here: like the other representatives, the Capuano vote data is sourced from ProPublica’s Represent project and a few other spots. On Capuano’s Represent page you can see that he’s voted against his own party just 5.3% of the time (two votes), and why. (This include a vote on January 10th that isn’t logged on 538 - I’d need to dig further into their methods to see why.) Now, kudos to 538 for visualizing this in a pretty way and framing everything in terms of whether or not someone voted for the Republicans (as opposed to “voting against their party”, though that’s also a very useful metric.)

BUT I would suggest that if we like Capuano because he has a zero, and not because his actual score matches his predicted score (also zero), then we aren’t necessarily using this metric correctly. What, for instance, do we say about Richard M. Nolan (D-MN), who should really be supporting Trump’s positions much more heavily given how much his district went for the President? If we think that’s great (and I rather do), then I’d prefer a different metric. If, on the other hand, I’m a Trump voter in Nolan’s district, I should be giving him some really angry calls.
posted by Going To Maine at 11:34 AM on February 15

I think the key takeaway here is that if the Trump Plus-Minus score is positive, and they're a Democrat; PRIMARY THEIR ASS IN 2018.

Looking at you Feinstein (CA), Hirono (HI), Cardin (MD), Cantwell (WA), King (ME), Cuellar (TX), Costa (CA), Sinema (AZ), Correa (CA), Sires (NJ), Green (TX) ...
posted by Xyanthilous P. Harrierstick at 5:34 PM on February 15

Looking at you Feinstein (CA), Hirono (HI), Cardin (MD), Cantwell (WA), King (ME), Cuellar (TX), Costa (CA), Sinema (AZ), Correa (CA), Sires (NJ), Green (TX)

Just to build in some more historical data in order to better evaluate the metric itself:

According to Propublica, the average Democratic Senator votes against their party 4.1% of the time. (Not sure what the median is.)
• Feinstein has voted against the party 2.3% of the time
• Hirono has voted against the party 0.0% of the time.
• Cardin has voted against the party 0.0% of the time.
• Cantwell has voted against the party 1.5% of the time.
• King has voted against the party 13.5% of the time. (He’s also an independent, but what that means for Maine I have no idea.)
According to ProPublica, the average Democratic Representative votes against their party 5.1% of the time. (Not sure what the median is.)
• Cuellar has voted against the party 21.6% of the time
• Costa has voted against the party 15.4% of the time.
• Sinema has voted against the party 17.4% of the time.
• Correa has voted against the party 7.4% of the time.
• Sires has voted against the party 6.7% of the time.
• Green has voted against the party 6.4% of the time.
In other words, for the Senators this metric seems mismatched with the party votes. With the reps, it’s a bit closer Someone with more time should do a more thorough comparison. It’ll also be interesting to see what additional votes on confirming judgeships does to the Senate numbers. As I’ve speculated earlier, they’ll probably be more in line with the President because judgeships are less controversial.
posted by Going To Maine at 6:52 PM on February 15 [1 favorite]

(Looks like King was a Democrat until 1993, at which point he ran for Governor of Maine as an Independent. His Wikipedia page quotes him saying, "The Democratic Party as an institution has become too much the party that is looking for something from government". Despite that, it appears that he's always caucused with the Democrats while in the Senate.)
posted by tobascodagama at 7:05 PM on February 15

Bill Foster, IL-11, 0.0%. Glad I'm in his district!
posted by SisterHavana at 12:59 PM on February 16

Given my opinion that this metric should really be better suited for vetting Rs who should be challenged, here’s some numbers Rs who are voting too closely with Trump

According to ProPublica, Republican senators vote against their party 2.5% of the time
According to ProPublica, Republican representatives vote against their party 4.1% of the time
Again, it feels like the Trump margin is sort of insufficient here - or rather a broader critique of politics. All of these folks -even the Democrats- are voting much more as a block than one might expect given the lack of consensus around the presidential candidate. I can’t say that I know much about most of the Republican senators up there, but Rubio was at one point considered quite vulnerable, as was Toomey - the fact that the Trump score shows them as vulnerable seems kind of like a mark in its favor. It’s old news, but perhaps good as reinforcement for the general principal of the model. But - what if we just looked at the voting-lockstep difference with their margin of victory? That might give us something similar.

In terms of additional data that would be helpful, I wodner about adding the margin of victory for each of these candidates the last time they ran? It seems like the Trump vote is being used as a proxy for that - which is probably fine, but it’s being called out specifically as “Trump” score, not “republican lock-step” score.

Like, the heart of labeling it as a Trump-score seems like a way of calling out that the President might start passing unconventional laws. (As opposed to unconventional executive orders.) That difference - the idea that a Republican might go along with their party while at the same time opposing the President, feels like it’s absent.

But whatever - I’ve spent enough time writing about this thing that it clearly has some base magnetism, but mostly I’ll tie that to its tidy UI. Good job 538 on making the page, your metric still seems kind of bleh.
posted by Going To Maine at 6:02 PM on February 16

« Older South Korean and Japanese perspectives on North...   |   "All the German Shepherds are dead, but then again... Newer »