Hi Roger,
Here are some plots of the network variance and the presidential poll ratings.
The gif plot "xxxxsmoo" has a 3 point moving average smooth of the poll results. It is a bit easier to read.
The horizontal axis is time and the vertical is the percent of positive approval for the US president as given by various public opinion polls.
First the polls.
I choose 5 or 6 or national polls that addressed the question:
"Do you approve or disapprove of the way [President X] is handling his job"
I got the poll numbers at the site pollingreport.com
I plot the percentage of positive approval given by these polls for the day the poll closed (polls take several days to complete). I averaged results for days that had more than 1 poll result. Thus I have 506 days with poll results over the period spanned from Aug. 9 1998 to Dec 15 2004 (span of 2326 days). The
As an aside, it's interesting to look at this curve as a civics lesson. We know attack and war (far more than natural disasters) effect the public's opinion of their government. It's a bit chilling to see how strong this really is as is shown by this plot. Note that in this long concatenation of poll results there is the Clinton affair, hurricane disasters, electoral turbulence. These have small effects in the plot. But being attacked or attacking others. Ooof. Off the charts.
Back to the plots.
The network variance plot that I show as a comparison... and to show that there appears to be a correlation with the poll results... is the variance of the Stouffer Z per day expressed as the cumulative deviation .Remember that this is vary close to the cumdev of the Stouffer Z-sqrd that we use for the standard analysis. So this is our main statistic. I scale the cumdev for the plot. The only criteria for the scaling is to make it visually comparable to the poll results. The verical axis corresponds to the pool results and not to this scaled cumdev.
Markers.
I placed some vertical grey bars to mark significant events.
These are:
Bush inauguartion
9/11
Start of Iraq war
Capture of Saddam Hussein
Re-election of Bush
Correlations.
There seem to be several qualitative correlations.
1. Jumps at 9/11, Iraq war and (perhaps) Saddam capture.
Note that the poll jumps are nearly immediate and the netvar cumdev is more gradual, somewhat as responding with some inertia to an transient impulse.
2. General trend after 9/11 is strongly down (aside from transient jumps) up until about mid-2004
3. Clinton period is "calm" for both plots.
4. Quantitative:
I haven't done careful calcs yet, but the Kendall Rank Correlation comes in huge(t score of 16). But, I will need to the correlation calcs using values of the netvar that are not cumulative sums, I think. More on this later.
The more I stare at these plots, the more I find them amazing.
It's clear that the Presidential approval rating is a very general barometer of how people feel. This is really interesting.
We might also want to be a bit prudent with this since it is the kind of story the press could possibly really jump on should they get wind of it. You know, politics and weird science is a good hook for a story.
That's it. I'm off to celebrate the New Year.
A very Joyful one to you and yours.
cheers,
Peter