Identifying Friend or Foe

Given how often what’s said on the campaign trail bears so little resemblance to votes cast on the House or Senate floor, it’s become a full-time job holding our elected officials accountable. Stakeholder groups will spend hundreds of hours tracking bills and legislative voting records to generate candidate ratings every year, and some even pay professional lobbyists to provide those services. To what end? Simply to help identify friend or foe.

Voter Science has provided tools in the past to track bills of interest through the legislature, but we’ve found that the data entry required to tag bills that an organization supports or opposes can be a significant barrier to entry. Folks simply don’t have the time to maintain these lists because they’re too busy trying to work the halls of the state capitol and advocate for their positions with members one-on-one and in committee hearings. Fortunately, since 2014 we’ve had an online committee sign-in system for public hearings that’s a public record of positions that individuals and organizations have taken in support or opposition to every bill that’s been granted a hearing. In fact, at the beginning of the 2022 legislative session there were exactly 322,706 records from such public testimony. With this public data, we already know which bills that stakeholder groups have decided to support or oppose, so there’s really no need for any tedious data entry. Moreover, our state’s Legislative Web Services provide the public easy access to member voting records, so we can now automate the entire process of determining how those voting records correlate to each organization’s public policy position.

Today I’d like to introduce you to my latest pet project, codenamed Identify Friend or Foe (IFF) after the transponder system used by our military to identify combatants on the battlefield. You can access it from my WhipStat prototyping site here:

http://whipstat.com/Projects/Advocacy

The user interface is quite similar to my Partisan Leaderboard page, where I display a stack chart for all members by chamber and date range. However, here the main Organization drop-down lists over 1,500 lobbyist employers registered by the PDC that were referenced from hearing testimony records. When you select an organization, an aggregated list of “bills of interest” will be displayed beneath the chart. This table includes the bill number, title, total number of references from the selected organization, number willing to testify, and the percentage supporting the bill, with “Pro” counting as 1, “Con” as -1, and “Other as 0. Note that I’m using a fuzzy matching algorithm to match the organization name entered in committee sign-in to the official PDC records, so they may not be perfect…but we’re getting better every day. Use this list as a quick sanity check to ensure that your organization’s testimony records have been aggregated accurately.

The horizontal axis of the stack chart show shows the correlation between member voting records and organization position for each bill. Members who always vote the organization’s position will have 100% correlation and those who always take the opposite position will have -100% correlation. The dots for each member are color coded by party and if you hover over each you’ll see a tooltip with each members information and their actual correlation coefficient. To save a tab-delimited “leader list” of the member scores that can be imported into Excel, you can simply press the Download button.

Note that I’m currently collecting additional information that could potentially be used to further weight these scores (but that would make them less than a true Pearson correlation). Here are some examples:

  • A stakeholder’s willingness to testify or whether they’ve travelled from out of town
  • Committee votes made by members when advancing the bill to the floor
  • Committee leadership that could be positioned to advance or kill the bill

IFF is obviously a work in progress and we would welcome any feedback you have on our current user interface or algorithms. Since committee sign-in data now must be obtained by formal public records request, our plan is to update this tool at the end of every session, but if more frequent updates would be valuable to legislative advocacy groups we should encourage the Legislative Service Center (a.k.a. LegTech team) to incorporate the sign-in data into the Legislative Web Services, where it probably belongs.

We at Voter Science hope that IFF can usher in a new era of transparency for state government, freeing stakeholders and lobbyists from the tedious process of generating their own candidate rating systems and holding elected officials more accountable for their actual voting records when they inevitably come asking for campaign donations. It may seem obvious, but up until now it’s been surprisingly difficult to know who your friends in Olympia really are.

Case Study: Handling the MN GOP Convention

This article is a technical case-study for how Voter-Science’s services handled a high-traffic event: the Minnesota GOP State Convention on May 30th.  The MNGOP was clear that it was absolutely critical for the site to keep up with the surge in traffic and stay fully responsive during their event. The event was successful. There were surge periods hitting over 1000 requests/second to our servers, and the servers averaged responses in under 100ms.

Here were the engineering steps we took to provide the MNGOP that guarantee…

Continue reading “Case Study: Handling the MN GOP Convention”

A candidate’s first task: creating an online petition

As a new candidate, a great first task is to create an free online petition at https://PetitionBuilder.org and share it out.

An online petition lets you pick a topic and people can sign up with their name, zip code and email address. They can also leave comments and upvote on other comments – which is empowering to the signers.

An online petition is an opportunity to test the waters in March, not at the August primary.  Specifically:

  1. Pick a meaningful topic – Avoid frustrated partisan rhetoric that only appeals to the base. Choose something that resonate with their community and motivates voters.
  2. Get community feedback – If nobody signs your petition, it gives you a pulse that perhaps the topic is not broadly important and you should focus elsewhere. Signers can also leave comments and upvote on a petition, so that’s another signal you can use.
  3. Exercise your influencer network –To really get traction, you’re going to have to do more than just share it once on Facebook. Roll up your sleeves and go to community meetings, meet with other people, and be seen as a leader on the topic in the community. This is hard work, but all essential skills you will need on the campaign to get votes.

The bottom line is if you can’t even get 100 signatures on a petition, you certainly won’t get 10,000 votes in August!  For many, running an online petition is a great wakeup call – but early enough that they can do something about it.

 

Now what?

As you get your signatures, you can monitor the statistics page to see things like view rates, signup rates, share rates. You can even see a heat map of where the signups are coming from.

petition-stats

Some practical next steps after you get signatures:

  • Use screen shots from the stats pages to make followup posts promoting the petition.
  • Update your petition’s description with new information.
  • Use the stats to identify the biggest influences
  • Contact petition signers with followup messages and action items. You can import the signers into your own mail list or contact them via PetitionBuilder.
  • Match your signers back to the voter-database to determine other attributes such as legislative district, party score, voting history, or other demographics. Voter-Science can help with this.

 

Easy integration between your CRM and VS Canvasser

Voter-Science provides a free door-to-door canvassing app, and you can bring your own data and get started immediately at https://Start.Voter-Science.com

But for Developers, there’s also VoterScience API access that lets you can quickly add canvassing support to your existing app.

This is ideal for apps, such as CRMs or outreach platforms, that have a list of names. You can call an API to create a new canvassing sheet with those names, and then receive a webhook as the canvassing results are filled out. The general flow here would be:

  1. In your CRM app, add a button like “Export to Walklist” which takes a list from your app and passes it to the VS API. You’ll also specify a webhook to receive results and which users are allowed to access this sheet. Your app is then in full control of list management.
  2. Users can then open the walklist on the VS Canvasser app. They will log in via their email and are matched against permissions  you provided in the first step.
  3. As users fill in canvassing results, VS will fire the webhook you provided in the first API call.
  4. Your app listens on a webhook and fills in results in your system. This could be adding tags, filling in fields, etc.

See https://github.com/Voter-Science/TrcLibNpm/wiki/Create-New-Sheets  for API usage.

A few additional notes:

  • This can also be used to integrate with an existing CRM. For example, we use this APIs to integrate between VS Canvasser and NationBuilder.
  • Users for the canvassing can be separate from your CRM users. For example, you may have a few staff members that can access your CRM, but a totally separate field team for running canvassing.
  • The VoterScience system also has a powerful data mashup engine that can merge in additional data sets or even provide geocoding.

So stop writing your own canvass apps and focus on more interesting problems!

3 takeaways from WA Presidential Primary

Here are some key takeaways from the Washington State 2020 presidential primary yesterday.

Background

Voters were required to mark a party on their ballot and then Democrats could vote for the Democrat nominee (a race down to Biden vs. Bernie) while Republicans could vote for the Republican Nominee (Trump).

While everyone’s specific vote (ie, Biden vs. Bernie) is private, the list of who voted and their party preference on the ballot is public (Democrat vs. Republican) and maintained by the Secretary of State.

Results

As our snapshot last night (midnight at Mar 10th) , there were 1.8 million ballots received (about 37% of the total voters) with the following split:

WaPresPrimaryResults

[Source: Secretary of State March 10th Election Results.]

We expect the absolute numbers to change as more ballots are received in the mail; but the percentages and trends will likely stay similar.

96% of voters successfully marked a party preference. Leading up to Tuesday, there was some controversy about the need to mark a party preference, but in practice, the overwhelming majority complied.

Leveraging a party score database

Voter-Science maintains a Party Identification database that associates each voter with a Party ID score.  This database is used by hundreds of candidates across the state and has frequently predicted elections to 99%+ accuracy. (contact info@voter-science.com to learn more about our database).

We can then join the ballot results with the party scores to gain additional insights. Here’s the pivot showing both party score (rows) and ballot marking (columns).

WaPresPrimaryResultsByPartyScore

Voter-Science has a party score for over 90% of the voters.

  • A “hard” voter is that party’s base and likely to vote straight party line.
  • A “soft” voter likely identifies with a party but is still considered persuadable.
  • The “Unknown” row is people that VS doesn’t yet have a party score for.

For example, this reads that 1.1 million ballots were marked Democrats, and of that 544k of those voters have voter-science party score of “soft democrat”.  The boxes inline show the cross over votes.

Independents went 67.3% : 32.7%  for a Democrat ballot over a Republican one.  That could spell trouble for Republicans in November, or it may be because the Democrats still had an interesting choice on their ballot whereas Republicans just could vote for Trump.

 

What about cross-over voting?

Dedicated party voters stuck with their party ballot. Only 27k GOP and 10k democrats did cross over and vote on the other ballot.  The 10k democrat voters may seem significant, but that’s only 0.58% of the total votes – a small enough number to be attribute to voter error in filling out their ballot. This won’t be an issue in November once there’s just a single general ballot.

But, there’s interesting cross-over from Soft Dem/GOP:

76k soft democrats (8.3% of total Dems) voted on an uncontested GOP ballot to support Trump. That’s 5% of the total vote, which could be an interesting sector if Republicans can identify and leverage them in November.

20.3% of total GOP voters crossed over to vote on the democrat ballot. That could be because the GOP ballot has just Trump, so these GOP may have weighed in on the more interesting Bernie/Biden debate.

 

Summary

  • 96% of voters successfully marked a party preference
  • Independents went 67.3% : 32.7%  for a marked a Democrat ballot over a Republican one
  • 20.3% of total soft GOP voters crossed over to vote on the democrat ballot. Only 8% of total soft Democrats

Canvassing With Gestures

While the use of obscene gestures as part of any campaign communications strategy is to be discouraged, some gestures can be an intuitive and efficient means of data entry on mobile devices.  So as we knock on doors, anything that helps us shift focus from our phones to our neighbors not only saves time and effort, but also promotes a more positive image in our communities.

Canvasser has always supported the swipe gesture to proceed to the next household or household member, but with the release last week of v1.6 for both Android and iOS, we’ve introduced support for a powerful new gesture: Shake.

To enable this feature, simply open up Settings from the main menu.  The new Gestures section adds two new settings:

Screenshot_1563306118

The first setting assigns an action to be initiated when a shake gesture is detected.  This is disabled by default, but by tapping the control you can pick the option you’d like automatically entered into the Result field when your phone is shaken.  For canvassing, that’s usually “No contact” or “Left literature”, but “No answer” might be more appropriate if you’re on the phone working through a call list to remind people to vote.

The second setting is to provide audible feedback when an action is triggered by gesture, which is done using the built-in text-to-speech capabilities of your phone.  By default, this is enabled so that when you shake your phone you’ll immediately hear spoken feedback (e.g. “Left literature”) to indicate that the gesture was detected and Result field automatically filled.  You can then simply swipe to move on to the next household.

Note that the volume of the audible feedback will be subject to both the global and app-specific volume settings on your phone.  For more details on how to set these, click here for Android and here for iOS.

Shake gesture support will primarily be used from the household detail page, but with the v1.6.1 release we’ve added support for the voter detail page as well.

Of course, we’re just getting started with introducing more intuitive gestures to use with Canvasser that will help speed data entry when you’re out knocking on doors.  Personally, I’ve knocked on over 30,000 doors during my two last campaigns and so I have some opinions on what improves my efficiency, but I’m always anxious to hear more suggestions from the field.  If you have an idea for the next kick-ass new feature for Canvasser, please let me know at chad@voter-science.com.

Getting the exact right voters in a District

Information specific to the WA state VRDB

In the WA VRDB, the City column may give the postal office associated with that voter; but if it’s an unincorporated precinct, they may not be able to actually vote in that city’s election. In other words, “City=Renton” gives 79k people, but nearly 25k are unincorporated and can’t actually vote.

In addition, individuals may provide the wrong value for City, which could be wrong.

Fortunately, the VRDB has a separate table, DistrictLookup, that specifies exactly the precincts for a given (DistrictType,DistrictName) pair.

So (DistrictType=”City/Town” and DistrictName=”City of Renton”) would give you the actual precincts for Renton City Council, which you can then join and compare with the VRDB to get the list of voters … which is around 50k voters. However, that query is much cumbersome to provide!

When using the query portal, Use the “DistrictTypeCity” field to select just voters in a city.

(DistrictTypeCity == XXX) corresponds to a join with the DistrictLookup table and (DistrictType== “City/Town” and DistrictName ==XXX)

There are similar new fields for supporting other DistrictTypes, such as School, Fire, Port, Other, etc.

 

Some Examples

1 City of Renton

When queried, this gives 79k voters as a result, which includes unincorporated areas that can’t vote in a Renton City Council race, but are still categorized under Renton for mailing purposes.

 

2 DistrictType Renton

This query will give just the Renton voters, and results in a list of 53k voters.

 

3 Renton Mailing Area

This  query shows you the voters in the Renton mailing area, but can’t vote in a Renton City Council race (which results in a list of roughly 26k voters).

 

Other trivia:

  1. Some voters may match multiple “DistrictType=City/Town”. For example, Tacoma is split into multiple districts, so a voter in Tacoma district 2 would match “CITY OF TACOMA” and “CITY TAC-2”.
  2. Some cities put their districts in “DistrictType=Other” instead of City/Town. It’s arbitrary.
  3. DistrictNames in the VRDB are not normalized. This isn’t really an issue for cities, but can be an issue for other districts. Each county can report it their own way. Sometimes that just means a capitalization difference, but here the different ways that CD 5 shows up:
    • Congressional District 5
    • CONG 05
    • Congressional District – 05
    • CONGRESSIONAL DISTRICT – 005
    • CONGRESSIONAL DISTRICT – 5
    • CONGRESSIONAL 5

Partisan divide widening dramatically in Olympia

For the past few years I’ve posted on our interactive online tool that analyzes the partisan distribution of our state legislature.  The goal is to call out members with the courage to vote independently of their caucus.  Often candidates will run as moderate or independent during the campaign, but we find that their floor votes in Olympia are right down party lines.  This tool provides some transparency into their actions, versus their intent.

http://www.whipstat.com/Projects/Records

If you recall, the methodology is simple:  The partisanship score for each floor vote is calculated as the percentage of Republican supporters minus the percentage of Democrat supporters, giving each a range from 100 (exclusively Republican) to -100 (exclusively Democrat) with unanimous votes scoring zero.  The member’s aggregate score is just the average of all the scores of floor votes they supported, minus the scores from those they opposed.

Looking back to 2003, we see a relatively normal distribution curve for both parties.  And while there isn’t as much overlap in the middle as there’s been in generations past, we do see that there are moderates on both sides of the aisle and even some true independents that have represented us in Olympia.

Partisan Leaderboard - All Policy Areas, Both Chambers (2013-2020)

It is interesting to note that during this time period the most independent members have run as Republicans and that Democrats are generally much less likely to vote against their party.  You can also easily identify the three members who have switched caucuses.

So now, consider the 2019 legislative session results:

Partisan Leaderboard - All Policy Areas, Both Chambers (2019-2020)

Notice a problem?

Both parties are now considerably more partisan and there are no independents (or arguably even moderates) left in the state legislature.

So how has this changed over time?  Let’s take a look…

Partisan Distribution by Party (2003-19)

In the above graph, range lines show a standard deviation above and below the mean.  Markers represent the median value.  The bars in the center represent the party balance, which has almost always favored Democrats.

What can we conclude?

  • Both parties have trended more partisan during this time period.
  • The median tends to consistently fall the left of the caucus mean with Democrats.
  • Democrats are now over twice as partisan as they were in 2003-04 under Gov. Gary Locke.
  • The partisan divide is almost twice as wide now as it was in 2007-08, when the Democrats had a 42 seat advantage.

Wrap-Up

This “death of the middle” I see as a unhealthy development for our state (and not just because I was one of the moderates unseated with this wave of political polarization).  Compromise is a necessary part of the political process, and we need moderates on both sides of the aisle willing to bridge the divide to find common ground.  During the 5 years there was divided control of the legislature, it admittedly took much longer to hammer out bipartisan agreements…but the resulting work product was worth it.  Our bipartisan budgets typically passed with 90% support, while only 57% voted for this last biennial budget (including no Republicans and not even every Democrat).

Clearly, changes are needed.