Why did we build a model when other people do it professionally?

Do you remember the first thing you learned how to cook by yourself? For me, it was scrambled eggs with a little cream cheese and some green onions. When I finally finished up that slightly overcooked and dry egg mess and plated it - I was so proud of myself. I had done it! I had cooked breakfast! From there, things only got more interesting for me in the kitchen.

And the same goes for Tiny Model. When we set out to build an election model we wanted to start simply, because we believe that the best way to learn something is simply by doing it. Even if you suck at it (which I do, Alex does not). By starting simply and building to the more complex, we aim to gain a deeper understanding of how polling models work, why they are so contentious, and how to read them to best understand and navigate the news that inevitably comes out about a wave election year.

To put it in slightly different terms, if you want a highball cocktail, typically composed of a spirit + a non-alcoholic mixer (your gin and tonics, rum and cokes, screwdrivers, etc) all you need are two simple ingredients. This makes a simple, maybe even a naive cocktail - much like Tiny Model! But if you wanted something more complex, like a tiki drink, you’re going to need to mix multiple types of rum, specialty syrups (orgeat anyone?), intricate garnishes, and that all important pebble ice. Polling models work similarly.

Today, we want to walk you through some of the basic mechanics that talk about how polling logic functions. We’ll start with the highball (beginning) and work our way up to the tiki (more complex).

Soon, Tiny Model will be more complex. But today, it’s very simple. Tiny Model’s “generic congressional ballot” model. This kind of model basically asks: “if the election were held today, what would the D/R margin be? Right now, Democrats are probably ahead by +6 to +8 points, or D+6 to D+8.

This model is pretty simple. It’s only slightly more complicated than taking a plain-old average of the margins of polls: during the average, we give each poll a “weight” based on stuff like the quality of the pollster and how recent the poll was. But that’s still pretty simple in terms of the statistical mathiness we could be applying to our model.

The gold-standard generic congressional ballot poll is this one published by Nate Silver’s indie publication, The Silver Bulletin. So how does Tiny Model do against it? Look at the graph on Tiny Model’s home page and you’ll see: it seems reasonably close!

Generic congressional ballot

Democrats lead by 6.9 points

D

48.4%

R

41.6%

As of Oct 5, 2026

Tiny Model

Silver Bulletin

D+10.0D+5.0EVENR+5.0R+10.0

Dec 2024May 2025Nov 2025Apr 2026Oct 2026

The shaded band is Tiny Model's rough uncertainty range. In our backtests, actual results often fell outside it. This is a polling average, not an election forecast.

And it is! But it's not exactly the same. Why?

One of the big reasons is: they just have much better data. Premium election data sources cost $15,000. One of the most important things those sources provide is population data: for historical polls stretching back to the 90s, they have information like how many likely voters (LV) and registered voters (RV) reported. Essentially all the “big” models will have this data, too: NYT, The Economist, and so on. Tiny Model doesn’t have this data, so it can’t really compete. You're truly only as good as your data.

Bringing this all back to the kitchen counter - they have a house spice blend that they carefully prefabricate over time and taste preference and then sprinkle onto their eggs liberally. They have access to all kinds of professional machinery, food, and help.

Tiny Model, in contrast, is more like a nice home kitchen. We have a pretty nice food processor to help.

After all of this, you may be asking yourself “Why did a Utah focused publication decide to write their own polling model when New York Times/Siena and Silver Bulletin exist and are professionals at this?” Well friend, that’s a great question. And the answer is simple: we are curious people and Molly has always wanted to understand polling better.

Sure, those other models are smart, less naive, and have full staff dedicated to them nearly year round. But in creating our own model, we are experiencing the true joy of figuring out what our special spice blend is and then adding it to our own eggs, with cream cheese!

As of this writing, our model shows Democrats leading Republicans by 6 (D+6) points and that likely means that the Democrats will have a good year at the polls in November. According to the NYTimes Chief Political Analyst, Nate Cohn, “That’s “blue wave” territory. At least in terms of the national vote, the Democratic margin in the [NYTimes/Siena] poll — 8.5 points — would almost exactly match the party’s 8.6-point margin in 2018. And it would exceed the winning margins for the party out of power in other recent wave elections, like in 1994, 2006 and 2010.”

Traditionally, polling models anchor their midterm expectations based on a president’s approval rating - which makes sense because a popular president with a popular congress isn’t likely to cause people to march out of their houses during election season with a bone to pick. Whereas, an unpopular president with an unpopular congress could galvanize voters strongly enough that they not only march out of their houses to vote, but they vote against the very people they elected two years prior! The traditional anchor for this metric is roughly a president (see Gallup article) underwater (meaning they have more negative than positive approval ratings) by 15 or more points. As of September 19, Trump’s approval rating on Silver Bulletin is -19.5. Well within the territory of a “Blue Wave”.

Looking at the Economist, they have his net approval rating even lower with an additional ~6 points at -26! Although we will note that his trend line for approval has mostly tracked evenly with his first term and his first year back in office. Starting in early 2026 his approval splits downward from his first term by about 15-20 points more negative. That roughly coincides with the extraordinary military intervention in Venezuela resulting in the capture of President Nicolas Maduro and extradition to the US, where he remains.

Shortly following that, the US and Israel began joint maneuvers in Iran which ultimately resulted in the ongoing war/not war. The full results of that decision are likely to be felt at the ballot box (we speculate!) largely due to the dramatic increase in gas and diesel prices. But we digress.

Targeting an issue Trump has long been particularly strong on: Immigration, let’s look at how people feel about his work on this issue since he was elected. Looking at the Silver Bulletin Trump approval rating, Trump was sitting at a pretty 53% approval rating in March 2026, but by mid April he had fallen below 50% approval, spiking above water only briefly in early June, before quickly retreating to the low 40’s where he remains.

His other major issue where he was particularly strong has historically been the economy. Here we see an even starker picture of a president who has likely set his party on a path to loss at the ballot box.

According to Silver Bulletin, Trump has never seen a positive approval rating on the economy during his second term. His disapproval rating stays below 50% in January and February before spiking upwards almost continually from early March to today. As of this writing, his disapproval rating on the economy is 63% - that’s a staggering -31 point net approval rating. Keeping in mind that ‘wave’ territory starts at ~15 points, double that likely portends tough days ahead for Republicans at the ballot box.

Bringing this back home to Utah, land of the industrious, we wanted to understand how our very exciting congressional district 1 race would shake out in our polling model.

In this case, it's very difficult for our model to understand how to rate and place a race like Ben McAdams vs. Riley Owen. The vibes on the ground tell us this is McAdams' race to lose. He’s got all the fundamentals down - great name recognition for having been a perennially elected individual (State Senator, County Mayor, House Representative), fundraising through the roof ($2+ million raised to date), and the economy is not doing as well as folks had hoped it would do under Trump 2.0 likely resulting in more Democratic voters.

Unfortunately, we’re sad to inform you that this race did not even merit a single professional poll that we deem acceptable and therefore we have no real idea on the state of the race.

Well.

So, what’s a girl to do when she lives in a polling desert but loves the horse race of political ballot box races? Build her own model, feel sad about lack of representation for Utah, daydream about starting her own polling consortium, and then build a model to talk about the other states in the nation that are more heavily polled. Obviously.

All quips aside, what this showcases is that a Tiny Model that is naive and simple is actually one of the best ways to view a race. Since we don’t have any polls that meet our standards, we simply cannot predict the election outcome.

It’s a lot like you wanted to make a quiche with those eggs instead of scrambling them, only to realize you don’t have a crust, you’re out of a savory meat, and you only have moldy lettuce in your crisper drawer. You know what a good quiche tastes like, but you don’t have the ingredients to actually make the quiche. Luckily for you, Harmons likely has a good version in their pre-made section. Or, in the case of polling, New York Times/Siena and Silver Bulletin have your back.

Next week we’ll dive into the raw signal and the generic ballot more. We have a lot of questions about what a poll actually measures, what the Generic Congressional Ballot actually tells us, and why a single poll with a +/- 3% margin of error is effectively a coin flip.