Jump to content
270soft Forum
vcczar

Why is this happening?

Recommended Posts

@admin_270 I just spacebarred through the General Election (no primary) on my 2020 scenario as the Simulation Party (to watch). The results didn't make any sense. Warren won UT by a large margin. The Green Party won ME, despite scoring only 3% nationwide in the PV. A few other states went odd directions. I had hoped to combat this by making the voters more certain of who they would vote for. In fact, I have no uncertain voters for any party. They're all set to vote certainly for one party, or whatever the middle option is. I would assume this would prevent such strange election results, but it didn't. 

Here is the election result file. 

I'll also attach the scenario. 

Can you explain this and hopefully prevent it? 

 

WarrenTrumpresults.csv

United States - 2020 (8-16-18).zip

Share this post


Link to post
Share on other sites
1 hour ago, vcczar said:

@admin_270 I just spacebarred through the General Election (no primary) on my 2020 scenario as the Simulation Party (to watch). The results didn't make any sense. Warren won UT by a large margin. The Green Party won ME, despite scoring only 3% nationwide in the PV. A few other states went odd directions. I had hoped to combat this by making the voters more certain of who they would vote for. In fact, I have no uncertain voters for any party. They're all set to vote certainly for one party, or whatever the middle option is. I would assume this would prevent such strange election results, but it didn't. 

Here is the election result file. 

I'll also attach the scenario. 

Can you explain this and hopefully prevent it? 

 

WarrenTrumpresults.csv

United States - 2020 (8-16-18).zip

What happened in that game? Those spreadsheet results are laughably unrealistic.

Share this post


Link to post
Share on other sites
9 minutes ago, NYrepublican said:

What happened in that game? Those spreadsheet results are laughably unrealistic.

I just space-barred as the simulation party. It was Trump/Pence vs. Warren/Castro. The map looked believable until about a week before the election. Green didn't show up as having Maine until the results came in. I think the same with Utah. I'm just out of ideas on how to curb this behavior. I thought making the voters more set in their ways would prevent this.

Share this post


Link to post
Share on other sites
10 hours ago, vcczar said:

@admin_270 I just spacebarred through the General Election (no primary) on my 2020 scenario as the Simulation Party (to watch). The results didn't make any sense. Warren won UT by a large margin. The Green Party won ME, despite scoring only 3% nationwide in the PV. A few other states went odd directions. I had hoped to combat this by making the voters more certain of who they would vote for. In fact, I have no uncertain voters for any party. They're all set to vote certainly for one party, or whatever the middle option is. I would assume this would prevent such strange election results, but it didn't. 

Here is the election result file. 

I'll also attach the scenario. 

Can you explain this and hopefully prevent it? 

 

WarrenTrumpresults.csv

United States - 2020 (8-16-18).zip

Oof. I haven't designed scenarios before and I know Anthony said it might be an AI behavior thing that could be fixed, but a few questions:

(1) Do you get the same unrealistic results if you turn third parties off (other than the simulation party)? Might be an issue with how they're implemented or the goals or something.

(2) Did you have primaries on? Maybe there's an issue with candidates building up infrastructure and foot soldiers in the primaries in certain states in one party, while in the other party those states don't matter.

(3) Was there any indication this would happen from the state polling (also does the observer have the max polling attribute) leading up to election day?

(4) Is the unrealistic state voting thing just something from the latest version? Or had you been seeing it for a while?

Share this post


Link to post
Share on other sites
1 hour ago, thr33 said:

(1) Do you get the same unrealistic results if you turn third parties off (other than the simulation party)? Might be an issue with how they're implemented or the goals or something.

(2) Did you have primaries on? Maybe there's an issue with candidates building up infrastructure and foot soldiers in the primaries in certain states in one party, while in the other party those states don't matter.

(3) Was there any indication this would happen from the state polling (also does the observer have the max polling attribute) leading up to election day?

(4) Is the unrealistic state voting thing just something from the latest version? Or had you been seeing it for a while?

1. You still get unrealistic results without the 3rd parties. For instance, Biden might win Oklahoma with 60% of the vote because Biden's campaign will obsess about Oklahoma, and Trump won't do anything to defend against it. That's just an example. It will happen in other random stats. Generally, results are only unrealistic in two or three states, but that's still too many. Results should pretty much always be within the realm of believability when using the simulation party. I don't think the CPU has a built in strategy that is used rationally. 

2. I usually play from the primaries, but in this playthrough I used the general election only. 

3. I didn't look at the specific polls, but the state colors seemed reasonable until the final week or two. As I said, the Green Party didn't show up as leading Maine until the results came in. 

4. Election results, in my opinion, have been getting worse with each update it seems. It has to be about two years since election results were more often believable than not. Now it seems as if 9 of 10 results are unrealistic. Unrealistic would include someone winning a competitive state with 80% of the vote. Really California shouldn't even go 80% for Democrats, which often happens. 

I've suggested this numerous times, but I think the editor needs to have a high/low number for each party in each state that the party can't go above or sink below. This will help. For instance, California could be set for Dems as High: 72 and Low: 45. Florida, which is always really close, could be set to Dems and Reps as high: 58 and low: 42. To prevent Green or LIbertarian from randomly winning a state, even when they have 3% nationally, you could set the high for Libertarians in even their best states to: 11%. And for Green: 5%. If people don't like this, then they can select to ignore the high/low restrictions. Ultimately, this will lead to very believable results. It's especially useful for historical elections. 

Share this post


Link to post
Share on other sites

My best guess is that, because the AI has gotten better, they are employing unconventional strategies that work within the rules of the game but produce unrealistic results. Again, I'll test for this to see if this is what's happening.

Share this post


Link to post
Share on other sites
1 hour ago, admin_270 said:

My best guess is that, because the AI has gotten better, they are employing unconventional strategies that work within the rules of the game but produce unrealistic results. Again, I'll test for this to see if this is what's happening.

That's partly what I figured. The AI has gotten really good in primaries, and reacts very well to user state infrastructure and foot soldiers, makes playing as an underdog really difficult.

I know that voter blocs/demographics are possibly planned for way down the line, but I wonder how much that could help? I guess maybe not all that much, since it still boils down to each state having definite and persuadable voters.

Since a lot of the AI models real life I'd say, I guess the question is, what happens if a candidate puts a ton of money into a state that can't be won? I mean you don't want to go *too far* (recall several of the states Trump targeted were considered out of reach, but in reality the Midwest was ripe to flip). On the other hand, you have examples like Bush putting $10+ million or so into California in 2000, and he only got 41.7% of the vote there.

I mean in real life even if you campaign in an out-of-reach state is there an absolute ceiling on momentum? What would be the response (maybe the Super PACs would target it hard)?

Just thinking aloud, but I wonder what makes sense here.

Share this post


Link to post
Share on other sites

I added a new "Partisanship" Issue in my campaign and assigned values for each state based on Dem/Republican voter registration and voting history (as well as number of independents, so if a state is a safe dem but has high independents I made it slight dem instead, if a state has no lean but high independents, I made it centrist, however if a state has high "independents" but never voted for the other party recently, I made it safe, used 4 sources to get the data). This issue represents the importance of partisan politics in a given state (the "vote for our team" dynamic, or how open individuals in the area are to messages when coming from a candidate of the other stripe). I then tested different issue importance values. Having the importance be High or Very High makes it MUCH less likely for there to be unrealistic results. If you make the issue a scandal + VH start, basically no ahistoric results will happen. This is because the AI would focus nothing but ads on Partisanship (as it tends to be very effective). This is probably not the intended behavior at campaign start.

Instead, I start the issue on Low/Medium and have scripted events as the game approaches the GE which bumps up the issue importance. Low in Primaries, Medium when the first Conventions happen, High once GE starts, and Very High once in October. It had very good results.

This issue works really in the primaries too for '16 since both Bernie and Trump are more independent on Partisanship than their other primary candidates in '16, so Trump naturally gets an advantage in liberal states (like NY and CA) while getting a disadvantage in conservative states. It also works well in the general election as a natural backstop in momentum, simulating "out of reach" states and parties rallying around their candidates once the primary is over.
____
In order to squash third party votes, I added negative momentum events as GE approaches for third parties. They are recoverable if a player is in a good situation as a third party, but it helps to prevent really large voteshares by the AI...most of the time. Other times I've found if a mainline candidate gets a bunch of scandals they can't recover (their CP drops to 1-3) and a third party may win some delegates.

Share this post


Link to post
Share on other sites

@thr33

The general, national approach of the Libertarian, Green, and so on parties, in the U.S., seems foolish. Who was the 3rd party candidate who almost won a state in 2016? Evan McMullin, who focused heavily on Utah, and at one point was neck-and-neck in polling, in the end getting >20% of the vote in that state. Similarly, in Canada the Bloc Quebecois has had significant successes by focusing exclusively on the province of Quebec. So, if 3rd parties were smart, they would figure out how to appeal to specific states in order to break through the winner-take-all nature of the electoral college.

Share this post


Link to post
Share on other sites
20 hours ago, admin_270 said:

My best guess is that, because the AI has gotten better, they are employing unconventional strategies that work within the rules of the game but produce unrealistic results. Again, I'll test for this to see if this is what's happening.

I think that might be the case. Although, the AI might not have gotten better about defending against the AI. I have the hunch that the AI is more offensive than defensive. For instance, Green zeroing in on Maine, Democrats zeroing in on Oklahoma rather than defending Maine, Republicans zeroing in on Oregon rather than defending Maine or Oklahoma. 

I've just made UK 1874 Election, and I noticed that I haven't seen similar issues there. Part of that might be because the election is only 24 days long. 

Share this post


Link to post
Share on other sites

@admin_270

What if you implemented a feature which could set importance for each state (so Ohio and Florida would be 5's in today's environment)?  That way the CPU would target the states that are important for victory.

Share this post


Link to post
Share on other sites
17 minutes ago, jvikings1 said:

@admin_270

What if you implemented a feature which could set importance for each state (so Ohio and Florida would be 5's in today's environment)?  That way the CPU would target the states that are important for victory.

Brilliant idea. I think this with an optional max/min % in the editor could fix the issue. 

Share this post


Link to post
Share on other sites
22 minutes ago, jvikings1 said:

@admin_270

What if you implemented a feature which could set importance for each state (so Ohio and Florida would be 5's in today's environment)?  That way the CPU would target the states that are important for victory.

For 3rd parties (such as McMullin), targeting Utah is probably the best strategy, not Florida or Ohio.

Share this post


Link to post
Share on other sites
2 minutes ago, admin_270 said:

For 3rd parties (such as McMullin), targeting Utah is probably the best strategy, not Florida or Ohio.

True.  Although you could do it by a party basis (which would allow different parties/candidates to target different states.  Although, I have no knowledge on how easily (or hard) that could be coded.

Share this post


Link to post
Share on other sites

How realistic is it that the Republican nominee in 2016 would lose Utah? Yet, that almost looked like it was going to happen at one point. Our intuitions about these things can trick us into thinking something isn't likely or not. A better approach re AI, I think, is a setting for that party or candidate's strategy (breakthrough or increase national %s, say).

Share this post


Link to post
Share on other sites

So, McMullin would be set to try to maximize electoral votes, the Libertarian and Green parties to maximize %s. That would prevent weird Green party breakthroughs. The other problem, it seems, is that computer players aren't responding quickly enough to gains in momentum and %s from 3rd parties.

Share this post


Link to post
Share on other sites
1 minute ago, admin_270 said:

So, McMullin would be set to try to maximize electoral votes, the Libertarian and Green parties to maximize %s. That would prevent weird Green party breakthroughs. The other problem, it seems, is that computer players aren't responding quickly enough to gains in momentum and %s from 3rd parties.

Would it be possible to add a setting for state party infrastructure attribute (maybe an irl representation of how strong parties are at the state level) alongside campaign infrastructure? Third party candidates who choose to specialize could lower a state's party infrastructure in their pet state(s) (i.e. McMullin, or a regional candidate like Thurmond or Wallace). Either visible or hidden.

Not sure if this exists already under the hood, or if it doesn't, how hard it would be to implement (especially given how org. strength and foot soldiers interact with momentum; in reality a candidate could have a high floor performance but low momentum in a state).

Share this post


Link to post
Share on other sites
4 minutes ago, admin_270 said:

I think this could be modeled by giving a candidate a higher starting regional Org. Strength value ..

Makes sense. Would you move it to a larger scale (I think org. strength currently maxes at 5 points, so maybe 10)? I ask this because you'd probably want some separation between the majority party, the minority party, and third parties.

Though maybe it's unnecessary, and with a quick tweak of org strength value the solid/leaning/undecided tier system would work fine.

Share this post


Link to post
Share on other sites

Usually, you don't want to use a sledge hammer to fix something where a thumb tack will do. I'll look at the AI option first, and then, if that doesn't solve the problem, consider other options such as this.

Share this post


Link to post
Share on other sites
On 8/22/2018 at 1:28 PM, admin_270 said:

For 3rd parties (such as McMullin), targeting Utah is probably the best strategy, not Florida or Ohio.

You might also want to make losing voters harder as the vote share decreases (cutting closer to the base).

Also, with the Favorability update are you just adding voter preference numbers, or will voter issue centers be a part?

crazyvirginiaresults.PNG

Share this post


Link to post
Share on other sites

In order for Favorability to work, voters will have to have implicit platforms or issue centers. Is this what you mean?

Share this post


Link to post
Share on other sites
On 8/24/2018 at 4:13 PM, admin_270 said:

In order for Favorability to work, voters will have to have implicit platforms or issue centers. Is this what you mean?

I would love for us to be able to shift platform/issue centers over time in the primaries/general election in the US naturally (rather than current method via event). Eg voters who shift over to support one candidate would slowly adopt that candidate's platform.

Share this post


Link to post
Share on other sites
On 8/26/2018 at 3:36 AM, IonicAmalgam said:

I would love for us to be able to shift platform/issue centers over time in the primaries/general election in the US naturally (rather than current method via event). Eg voters who shift over to support one candidate would slowly adopt that candidate's platform.

and @admin_270 Yeah, I wish the CPU would shift their platform. At least some of them. Kerry and Romney were accused of flip-flopping. Hillary Clinton veered Left. 

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×