Supplement to Dreaminonempty's Approval Diaries

From dKosopedia

Jump to: navigation, search

Work in progress. Coming soon: district and county level approval. The intent of this page is to put archives of old approval maps on here, as well as explanations of how estimates are done, and other things that can assist the occasional diary series.

Contents

Estimation Procedures

Regression used to estimate Bush approval in 26 states for September, 2007.
Regression used to estimate Bush approval in 26 states for September, 2007.
Summary: All states and other geographic areas tend to increase and decrease their approval of Bush in similar patterns. If we know the approval in some states, we can use some simple statistics to estimate the approval in others with a reasonable degree of accuracy. We can also estimate what the approval is in counties and congressional districts. When estimates are tested by comparing to real polls conducted at a slightly later date, we find that the estimates are close enough to the real values to be useful.

In order to estimate Bush’s approval in geographic areas where no polls are available, regressions are used. For example, for the maps of approval by state, Survey USA stopped polling every state in November 2006, and cut back to about 15 states. These polling results are combined and averaged with all other statewide polling data available on Bush’s approval rating, and plotted against the average Survey USA approval rating for the period August-October 2006. From this, a regression line and an equation that describes this line are generated. This equation can then be used to estimate the approval ratings in states where no polls were taken that month, typically with a 95% confidence interval of 5-6 percentage points. This means that if we estimate an approval of, say, 30% in Illinois, we can be pretty sure that the true approval lies between 25 and 35%. Note that the Survey USA polls themselves have a 95% confidence interval (or margin of error) of 4 percentage points.


Regressions of Bush average approval for Jan-Aug 2007 against percent vote in 2004.
Regressions of Bush average approval for Jan-Aug 2007 against percent vote in 2004.
Better estimates can be produced using more data, although the time resolution is sacrificed. When all the state-level approval data for the first eight months of 2007 are plotted against Bush’s percent vote in 2004, we can see the states falling into separate categories on very straight lines. Some states, then, moved further to the left than others, and a handful (not shown) shifted further right. For those states with little or no data available in 2007, their behavior in 2005 and 2006 was used to decide which category they belonged in.


Regression used to estimate Bush approval in Congressional Districts, Summer 2006, compared to actual poll results for Congressional Districts.
Regression used to estimate Bush approval in Congressional Districts, Summer 2006, compared to actual poll results for Congressional Districts.
The approval among Democrats and Republicans, as reported by Gallup, also falls along a similar straight line, showing that even at extremes of public opinion the same equations should provide decent estimates. This allows us to use the equations for counties and congressional districts as well. However, it is possible that congressional districts simply behave differently, as the regression was developed using state data. In late summer 2006, a regression using state and metropolitan approval ratings was developed and used to predict approval in congressional districts. These predictions were compared to polling data, and a generally good agreement was found. It is therefore reasonable to use the equations below for congressional districts, and likely counties as well.

The Use of Color

Ideally, maps of this sort should not use both red and blue. However, the convention of red states and blue states, with purple states in between, is well-established after the 2000 election. On either the red or blue side of the maps, darker colors correspond to more extreme values. In order to allow the viewer to differentiate colors, the difference from one level to another was so large that the far extremes of approval had to move into green (on the disapproval side) and purple (on the approval side).

One could argue that shades of red only should be used, as the maps show only approval of Bush, and disapproval of Bush is not equivalent to having a positive view towards Democrats. The use of blue and green shades on the map could therefore be considered misleading. On the other hand, disapproval of Bush is much more than simply the absence of approval. Also, in politics, the 50% level matters, and we like to be able to easily tell the difference between values greater or less than 50%. Neither way is perfect, but use of both blue and red allows for easier differentiation between levels of support. Finally, the 2006 election showed that Bush’s approval did indeed have a substantial relationship to the support for Democrats at the House level, which means the use of blue on the maps is not as misleading as one might otherwise think.

County and District Approval Archives

Coming soon.

State Approval Archives

Also available as an animation.

Personal tools