National Average Trending: An Incredible Waste of My Life

[Alcon]

I have set out to make trend maps for every state in the U.S. by taking the margin for each county in 2004, subtracting the national margin, doing the same for 2000, and comparing the numbers.

Say a county was tied in 2004 and 2000.  2004 was Bush +2.47, and 2000 was Gore +0.51.  Thus, the county would be 2.47 more Democratic than the national average in 2004, and 0.51 more Republican than the national average in 2000.  Thus, a swing of 2.98 to the Democrats.

[Alcon]

I'm willing to make a small wager that you aren't anywhere close to completion by the end of the 2005 calendar year.

I'm willing to bet I will die without finishing this project.  But anyway...the colour scale:

>10% = <4%
>30% = 4%-8%
>50% = 8%-12%
>70% = 12%-16%
>90% = >16%

This actually takes less time than you'd think.  I just have to copy it from the Atlas table, do an Excel calculation, etc.  The trouble is making the map for states that I haven't memorised the counties for...which I am now working on.

By the way, I consider +4% a statistically significant trend.  This is entirely arbitrary.

[Alcon]

First, of course, is always Washington, because I'm a self-centred jerk.

There are 39 counties in Washington State.  Of them, 35 counties trended Democratic and 4 trended Republican.



Of these, 20 of the Democratic counties trended in a statistically significant way:  Asotin, Chelan, Clallam, Columbia, Ferry, Garfield, Jefferson, King, Kittitas, Lincoln, Okanogan, San Juan, Skagit, Skamania, Stevens, Thurston, Wahkiakum, Walla Walla, Whatcom, and Whitman.

No counties trended Republican in a statistically significant way.

Nine counties trended Democratic by ten points or more:  San Juan (Friday Harbor, 18.76);  Okanogan (Omak, 16.74);  Jefferson (Port Townsend, 15.77);  Columbia (Dayton, 12.77);  Whitman (Pullman, 12.18);  Whatcom (Bellingham, 12.14);  Garfield (Pomeroy, 11.12);  Klickitat (Goldendale, 10.66);  and Ferry (Republic, 10.25).

The Republican-trending counties were an interesting mix.  They were:

• Pierce, a Democratic-leaning county which contains Tacoma
• Adams, a small farming, extremely Republican county on the verge of becoming Hispanic majority
• Franklin, heavily Republican and Hispanic, containing the tri city of Pasco with a mix of agricultural and manufacturing
• Spokane, a solidly Republican county containing the Dem-leaning city of Spokane

[Alcon]

I have set out to make trend maps for every state in the U.S. by taking the margin for each county in 2004, subtracting the national margin, doing the same for 2000, and comparing the numbers.

Say a county was tied in 2004 and 2000.  2004 was Bush +2.47, and 2000 was Gore +0.51.  Thus, the county would be 2.47 more Democratic than the national average in 2004, and 0.51 more Republican than the national average in 2000.  Thus, a swing of 2.98 to the Democrats.

But to find a swing between the elections, shouldn't you subtract 0.51 from 2.47 instead of adding them together?

This results in the same numbers, and would be easier.  Thank you.  That's one of those things I would have realised tomorrow and yelled "stupid, stupid" at myself for.

I'm taking requests.  I should have some decent free time to finish it, and - Tweed - it's much faster than I imagined.  I'm doing states I know the counties of first, though.  Georgia and Texas will be a pain!

Queue
California
Delaware
Rhode Island
Connecticut
Massachusetts
Maine
Michigan
New Jersey
South Dakota

[Alcon]

There are 36 counties in Oregon.  Of them, 32 counties trended Democratic and 4 trended Republican.



Of these, 21 of the Democratic counties trended in a statistically significant way.

No counties trended Republican in a statistically significant way.

Four counties trended Democratic by ten points or more:  Benton (Corvallis, 11.12);  Multnomah (Portland, 12.10);  Hood River (Hood River, 13.71);  and Wallowa (Enterprise, 18.66).

The four Republican counties were all heavily Republican agricultural areas with ranching and timber industries.  Interestingly, so was Wallowa County, the biggest Democratic swing.

[Alcon]

I have set out to make trend maps for every state in the U.S. by taking the margin for each county in 2004, subtracting the national margin, doing the same for 2000, and comparing the numbers.

Say a county was tied in 2004 and 2000.  2004 was Bush +2.47, and 2000 was Gore +0.51.  Thus, the county would be 2.47 more Democratic than the national average in 2004, and 0.51 more Republican than the national average in 2000.  Thus, a swing of 2.98 to the Democrats.

But to find a swing between the elections, shouldn't you subtract 0.51 from 2.47 instead of adding them together?

This results in the same numbers, and would be easier.  Thank you.  That's one of those things I would have realised tomorrow and yelled "stupid, stupid" at myself for.
South Dakota

But the original numbers are -0.51 and 2.47, if we count a Democratic lean to be a positive number.  So, you should subtract -0.51 from 2.47, which is adding 0.51 to 2.47, so what you were doing initially was correct.

Democratic lean is negative.  Bush won by 2.47, Gore won by -0.51.  Thus, the overall lean adjustment should be -2.98, yes?  That is what I used for Oregon and Washington.

[Alcon]

There are 16 counties and one independent city in Nevada.  Of them, 11 counties and the independent city trended Democratic and 5 trended Republican.



Of these, five of the Democratic counties trended in a statistically significant way, and three Republican counties trended in a statistically significant way.

No counties trended by ten points or more.

More tomorrow.

[Alcon]

Alcon,
how did you calculate your percentages?  If I look at Snohomish County, I have a margin in 2004 of 7.50 in 2004 (+ = Dem) and 7.96 in 2000 for a net shift to Republican of 0.46, not a shift to the Democrats.

Here is my auto-generated map
<5, < 10, < 15, < 20, > 20 as my color coding.

I compared it to the national average, so an additional movement of 2.98 to the Democrats occurred.  Maps of just the swing are pretty readily available, so I thought this would be more interesting.

Would you prefer I used a different colour scale?

[Alcon]

Well, it would automatic with Dave's script, and would only count the R and D votes each year.

That would probably be best...in 2004 and 2000, I only did overall margin.  It would probably be better to do D vs. R, although it barely matters in 2004/2000.

It still ended up with some pretty crazy results, I recall, when Perot was considered in otherwise ultra-Republican counties.  Then again, now that look at the numbers again, it might be interesting.  It would sure show how Gore bled in rural areas.

Trinity County, California, in 1992 would have to be accounted for, I suppose, but other than that, why not?

[Alcon]

Yes, this does tend to overestimate the trend.  This should not be regarded as a liberal/conservative trend.  At best it is Democratic/Republican trend.  This is very true in areas where Nader did well - a lot of them were just Nader voters moving to Kerry.

I suppose Democratic vs. Republican would probably be best.  Dave might reconsider, if it is not too hard, doing it with the two-party comparison instead.

[Alcon]

Yes, this does tend to overestimate the trend.  This should not be regarded as a liberal/conservative trend.  At best it is Democratic/Republican trend.  This is very true in areas where Nader did well - a lot of them were just Nader voters moving to Kerry.

I suppose Democratic vs. Republican would probably be best.  Dave might reconsider, if it is not too hard, doing it with the two-party comparison instead.

Well I suppose Dave can do it in a lot of different ways without too much difficulty, if he wants to. I was thinking more along the lines of just looking at the percentage Bush vote, forget about the Democratic vote altogether. Since the libertarian party has been pretty marginal in both 2000 and 2004, I think that would give a more accurate picture of things.

I suppose, although Nader would still make things bothersome, wouldn't he?

I'm tired, so forgive me if I don't make sense. 

[Alcon]

I figure we'll bounce this around a bit before I mass-produce maps and data for previous years.  The current algorithm is
swing: (voteD1 - voteR1)/TotalVote1 - (voteD2 = voteR2)/TotalVote2. 
trend: (voteD1 - voteR1)/TotalVote1 - (voteD2 = voteR2)/TotalVote2) - ((voteDnational1 - voteRnational1)/nationaltotalvote1 - (voteDnational2 = voteRnational2)/nationaltotalvote2)

Minor parties will influence these meaning of the trend/swing figures - although even in the Case of Ralph Nader, its not obvious that all the votes would have gone Democratic -> how many would not have bothered to vote at all?  How many would have voted for Bush? How many for Badnarik?  (I've seen data in NH that more Bush voters cast ballots for Nader than Kerry voters).   

I think that "third party" votes are another level of analysis that can be performed when considering vote swings ant trends (there are maps for third-party strength already).
Enjoy,
Dave

That's a good point.  The numbers I've seen say nationally something like 50% Gore, 30% non-vote, 20% Bush.  Of course, there's the question of how many of those non-votes were just saying that on principle.

I'm pretty sure that, contrary to what A18 said, the way you currently have it would have third parties affect the result because they affect the margin (total vote).  Am I understanding this correctly?  It might be worth trying it with just D vs. R, since minor parties will introduce a variable over time that might vary significantly.

[Alcon]

Good idea, but has some problems.

For example areas in the northeast & mid atlantic that had a 9/11 shift (Long Island especially) will show a pretty solid rightward trend when in fact their is none (recent polls show Bush's approval hovering around 30% here, with most of those polls being when he was in the low 40's nationally) & the 9/11 impact that would show as a Republican trend is completley gone

While accounting for Nader involves throwing potentially unscientific numbers in, this is downright impossible.

Besides, I have certain troubles buying the 9/11 theory.  Just look at upstate New York.