Updated predictions- How many House seats do you think Dems will pick up?

[GeorgiaModerate]

Sticking with 42.

[GeorgiaModerate]

Sticking with 42.

Yeah, 42 is always the right number because Douglas Adams reasons. I agree with 42.

I will admit that the mere thought had begun to speculate about the merest possibility of crossing my mind.

[GeorgiaModerate]

I think Democrats will win some house seats, but not all of them.

Big if true.

Certainly a bold prediction.

[GeorgiaModerate]

So everyone is between 210ish and 250ish seats with a median  around 225-232?

Pretty much, with a few exceptions.  Ben Kenobi for one, who predicted the Democrats will lose 4 seats net.

[GeorgiaModerate]

You must be logged in to read this quote.

Not impossible, nor unlikely. Last time we were in a similar situation with Bush, the Rs picked up a few seats in the midterms. Right now RCP has the race as essentially tied.

Remember, this isn't a presidential election year and Rs tend to do better in the midterms.


No, the out party tends to do better in midterms.  The last two cycles that was the Republicans.  This time it's the Democrats.

If the Senate is going to have R pickups then we're likely to see a similar situation in the House.

This is a totally unjustified assumption.  They're not on equivalent playing fields.  The Senate battlegrounds are mostly in R-friendly states.  The House battleground districts are in much more D-friendly territory.

I've seen polls showing they are even in the generic ballot which means that Rs will net a bunch of seats.

You are choosing to look only at the extreme outliers in the range.  There have also been polls with D+15, D+16, etc., which are just as likely to be correct as the ones that are even -- i.e., not at all likely.  The truth will be somewhere near the average, or D+8 plus or minus a couple points.  IF the generic ballot was truly even, then yes the R's would likely gain.  But nobody really believes it's close to even; if you believe that, you are engaging in extremely wishful thinking.

(My inline comments in red.)

I don't mean this unkindly, but you should really stop posting analysis like this.  You're only embarrassing yourself.

[GeorgiaModerate]

The behavior I've seen here whenever a poll contradicts 'established knowledge' is to label the poll an 'extreme outlier'. That's not a particularly useful approach.  

You must be logged in to read this quote.

My first argument wasn't, "I believe the race to be X", but rather "I've seen some polls with the generic at even". I didn't even suggest that I thought the true value of the Generic to be even, all I said is that there's some evidence out there that the Rs are actually doing not too bad. R's don't need a generic even ballot to keep their seats, and win a few, D+3 would be sufficient.

And that is hardly something that is impossible to see.

But do go on.

I'm not saying the tied GCB polls are outliers because they contradict the general wisdom.  I say they are outliers because I looked at the actual data.  Looking at the GCB polls in the 538 database that include periods since August 1 (we could pick some other starting point and it wouldn't change the picture significantly), there are 97 polls.  They break down as:


D+14 4
D+13 5
D+12 3
D+11 7
D+10 5
D+9 9
D+8 14
D+7 12
D+6 9
D+5 10
D+4 7
D+3 4
D+2 3
Tie 4
R+3 1

With this data set, the four ties and the one R+3 are outliers BY DEFINITION, as are the ones at the other end of the range, as I noted in my previous response.  (FWIW, 3 of the ties and the R+3 were in August.)  Citing the outliers -- on either end -- is a very weak point in favor of an argument.

[GeorgiaModerate]

You must be logged in to read this quote.

Two things. One, not all the polls measure the same thing. You're looking at a dataset that has polls (by your own admission), that are spread out over 10 weeks.

This is why you're getting such high standard deviations in your sample, and why your sample isn't going to give you a good estimate of the current median.

If you wanted to do this properly, you'd confine it to 2 weeks, not 10 weeks and then try to estimate your outliers.

You need to be cautious with that. It's not normally distributed.

If we look at the last two weeks it does not significantly alter the situation, and actually includes only one tie and no R+ polls.  I deliberately went back far enough to include multiple ties, since you mentioned polls (plural) that showed an even ballot.

If you are confident that R's will gain seats, how about a friendly wager on it?  Specifically, if the D's end up with fewer than 195 total House seats, I'll change my avatar to match your current one until the end of November.  But if they end up with 195 or more, you change your avatar to I-GA for the same period.  Agreed?

[GeorgiaModerate]

You must be logged in to read this quote.

Yeah, just that going back that far ruins your sample.

You must be logged in to read this quote.

Nah. You take the atlas aggregate with the over and I'll take the under.

Nah, there are some predictions on the high end that are even more unlikely than yours (there's a D+153, IIRC).  How about which of our predictions is closest?  We're both on record in the thread, you at R+4 and me at D+42.  The average of those is D+19.  Want to go over/under on that?  (Exactly D+19 would be a push.)

[GeorgiaModerate]

You must be logged in to read this quote.

Which is sort of my point, no? Mine at least has polls that reflect the outcome.

You must be logged in to read this quote.

Why not straight up D takes the house you win. D falls short, I win. Simpler, and that's really the crux of the matter, no?

Fair enough.  It's a bet.