A new American Research Group poll, taken 6/18-21/97, finds approval at 27%, disapproval at 67%. This puts the approval trend estimate at 29.1%, continuing the decline that began in late April.
This is in line with Newsweek's 26% and NBC/WSJ's 29%, and is within range of Gallup's 32%, the three most recent polls. My post on the Newsweek result yesterday is unchanged by this result. None of the recent polls is an outlier, all falling within the +/-5% confidence interval around the current 29.1% estimate.
By popular request, I've added a plot to my standard approval post contents: The fourth graph contains both the standard "old blue" trend estimate, which is more likely to be correct in the long run but which is slower to detect change, and the "ready red" estimator which catches change quickly but is easily mislead by random noise that isn't actually a change in trend. Now you can see both as part of these posts. The blue line is plotted second, so when the red agrees closely with the blue estimator, the red line is covered up by the blue line. So where you can see red at all is where the estimators disagree. Otherwise red and blue are tracking together.
4 comments:
The red trend estimator seems like a fishy concept to me.
And, is it just me, or do the polls look like they are getting more dispersion as they go down. Shouldn't the variance across polls be greatest when true approval is 50%, or do I have this backwards?
jeremy,
Could you say why red is fishy?
The sampling variance at 30-70 is .21/n vs .25/n at 50-50, so not a whole lot of change in that yet.
I checked the variance. It declines from 2002 to the present (I attribute that to the war related spikes) but since 2005 the trend is not statistically significant.
Dependent variable is squared residual:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -22.803437 19.484106 -1.170 0.242
enddate 0.002196 0.001470 1.494 0.136
Ugh. Well, the slope is +.0022 per day, but with a se of .0015 for a t-value of 1.494. So your eyes may not deceive you, but the data are not yet clear enough to convince the t-distribution.
The red line seems kind of like the trend estimator for those who appreciate all the rigor that goes into your data and analysis, but still also like the little added thrill that comes from overreacting to the latest poll.
Post a Comment