Pollster.com

Articles and Analysis

 

Charles Franklin: Upturn in Bush Job Approval

Topics: George Bush

[Editor's note:   This post is the first on Pollster.com from our partner Professor Charles Franklin.  Charles also blogs at Political Arithmetik, but will also be a regular contributor here.  We will be changing the menus and titles next week to designate his contributions approrpriately, but the authors name always appears at the bottom of each post].

Approval of President Bush's handling of his job has taken a significant turn up since mid-August. In ten polls completed since September 1, approval has averaged 39.9%, while disapproval has averaged 55.7%.

Approval around 40% is not particularly good news for any President, but this represents a considerable improvement since earlier in the year.

The chart here plots each of 336 national polls taken since January 1, 2005 that asked a presidential job approval question. The dots represent the poll's approval rate. There is a good deal of scatter in these points, representing the variation due to sampling error, questionnaire effects, survey organization practices, interviewer variability and everything else that causes surveys to vary. While the individual polls vary quite a bit, the trend in approval over the last 21 months is clear. The blue line estimates the trend in approval, allowing each poll to have its say, but keeping to the middle of the range of polls at any point in time, so that the average deviation around the trend line is zero-- the polls above cancel the polls below the trend.

After starting the second term at 50.5% approval, the President's standing with the public declined steadily until November 2005, when it rallied briefly before again sinking to an all time low in May 2006. Following the President's television address on immigration on May 15, however, approval again rallied. The summer produced a steady plateau of approval, but starting around August 15, the trend has once more moved up. The total movement since May has erased most of the ground lost since January 2006, though approval still remains some 10 points below where it stood at the beginning of the second term.

As of polling completed September 14, the trend estimate of approval stands at 40.3%, slightly above the average of the most recent 10 polls. Most of the summer was spent around 37%, so the current standing is a gain of a little over 3 points in a month. The all time low came on May 12 when the approval trend hit 33.98%.

Notice that around that time, polls ranged from a low of 29% to a high of 38%, a good example of how much variation there is across surveys. If one wished to make the President's improvement seem large, one could compare the all time low of 29% with the current highest reading of 43% and conclude there has been a 14 point run-up in approval. If, conversely one wished to minimize the improvement, one might compare the high in mid-may of 38% with the lowest recent reading of 37% and conclude that in fact there has been a one point decline in approval. Both of those comparisons are misleading, of course. The virtue of focusing on the trend estimate is that it is not so sensitive to any single poll, nor to a choice of which poll to compare to which. Based on the trend, the President's approval has risen 6.3% since May 12. Or put in a different way, approval fell by 16.5 points from January 2005 through May 12, 2006. Since then, over six of those points have been regained. This leaves the President 10.2 points below where he started the second term, but with a strong upward trend over the past 4 months.

The interesting question is whether this trend will continue, accelerate, flatten or turn down over the next 53 days until the election, and how much the President's approval rating will affect the vote.

 

Comments
Ike:

Hey Charles,

It's nice to have you two together.

A simple question: can you put the month on your charts of approval --I think it helps all of us who speculate on the cause of movements (however futile that might be). For example, the round-up of the British terror suspects who planned (we think) to bomb airliners took place on August 10th...just before this upturn.

It would be nice to locate these points on the time axis.

An interesting comparison would be to overlay this apparent bounce with the increase in approval that began (right) just after the Republican convention in 2004.

Thanks again for these analyses. They are really invaluable.

Ike

____________________

Poll watchers, keep an eye on the details. Fox, AP Ipsos and WSJ/NBC are now using (with little fanfare) RV and LV samples which may skew ever so slightly Republican compared to "all adults" in previous poll samples. AP-Ipsos, for example, has Bush at 39 among all adults but 40 among LV.

I believe Charles' trend to be real, but the details are relevent for small changes.

I also suspect that movements below 43% reflect Rs who come and go (gas prices?). And the sustainability is, of course, key. I also supect the 9/11 anniversary represents a high point in the mini-blip.

But since the media reports each 2 point change with breathlessness (more so on the upswing than the downswing), the use of RV and LV should be clearly reported.

Don't hold your breath; look it up yourself.

____________________

Also, is Democracy Corps a 50 swing district and not a national survey?

Does the point get added to the aggregate anyway?

____________________

Ike-- thanks-- I'll see if I can't add the months lines. It is a tradeoff between too much in the graph and being able to find the time points you want. I'll fiddle and see if I can improve it.

DemFromCT: The Adults-RV-LV issue is an important one, and I'll see if I can give you a good answer on that soon. Looking within pollsters (those who gives us results for the different populations within a single poll) looks like about a 1-2 point difference due to the nature of the population. But they don't always go in the "obvious" direction. So I need to do some analysis, and then get back with a new post on this.

The Democracy Corps survey is a national survey, taken 9/6-10/06 among 1004 LVs. Democracy Corps also has a 9/12-14/06 survey for the competitive house districts. Interestingly, that poll ALSO finds a 43-53 approval-disapproval rate.

Charles

____________________

Rdandrea:

The price of gasoline is going down pre-election.

I'd like to see the Bush approval numbers superimposed on a plot of the average price of 87-octane regular.

____________________

Charles-

You said above that "The summer produced a steady plateau of approval, but starting around August 15, the trend has once more moved up."

Of all the specific elements of his job, Bush's strongest approval (appeal) is on handling terorism. See below. Do you see a causal relationship here?

Source: Britain's Financial Times: "Our plan swung into action on August 10 after 21 people were arrested in London, Wycombe and Birmingham for plotting to attack aircraft".

____________________

DemFromCT:

re gas prices and approval see prof pollkatz.

____________________

toad:

Does the approval rating of a president affect federal elections in his last term?

____________________

there's a historical correlate.

However, past performance is no guarantee of future results. ;-P

____________________


Just wanted to add more inormmation to my comment below - that the British terrorist capture (that Ike posted before I did) is the likely event that caused the upswing in Bush's approval, not gas prices.

Charles gave August 15 as the date when that sudden reversal in approval began, five days after the incident in Britain. Such a significant reversal suggests that a single event with universal instant awareness was the factor, especially since it plays to Bush's issue - terrorism.

As for gas prices, the Energy Information Administration link below of weekly regular gasoline prices shows major declines coming much later than August 15. Specifically, Aug. 7, $3.04; Aug. 14, $3.00, Aug. 21, $2.92; Aug. 28, $2.85; Sep. 4, $2.73; and Sep. 11, $2.62.

http://tonto.eia.doe.gov/dnav/pet/pet_pri_gnd_a_epmr_pte_cpgal_w.htm

Nick

Charles-
You said above that "The summer produced a steady plateau of approval, but starting around August 15, the trend has once more moved up."
Of all the specific elements of his job, Bush's strongest approval (appeal) is on handling terorism. See below. Do you see a causal relationship here?
Source: Britain's Financial Times: "Our plan swung into action on August 10 after 21 people were arrested in London, Wycombe and Birmingham for plotting to attack aircraft".
Posted by: Nick Panagakis | September 16, 2006 7:50 PM

____________________

toad:

Hmmm, I wonder what the beta is on presidential approval, that is how quickly and how much can it move up or down. The volitivity of it. My dimly remembered POL SCI 101 course had the "swing voters" in the middle of the ideological bell curve making up their minds just a few weeks before an election. Control of the House might go either way then in a time of war??

____________________

Here's further proof that ABC has been the primary Bush shill since 2001.

ABC's 57.95% average Bush monthly rating since 2001 is the highest among 10 pollsters.

The 10 pollster average is 54.75%.

AP (51.2%), Pew (52.1%) and Zogby (52%) have the lowest Bush ratings.

http://www.geocities.com/electionmodel/BushRating_29028_image001.png


Nwk Fox CNN Pew Harris CBS ABC Time NBC AP Zogby Avg
Avg 55.5 56.3 56.6 52.1 54.4 55.2 57.95 56.6 54.5 51.2 52 54.75

Median 52.5 52 53 50 51 53 55 53.5 50 49 49 52

Nwk Fox CNN Pew Harris CBS ABC Time NBC AP Zogby ARG Avg

Feb-01 52 55 57 53 na 53 55 52 na na 57 na 54.25
Mar-01 55 56 63 56 56 53 58 55 57 na 53 na 56.20
Apr-01 57 63 59 53 49 60 63 na 56 na 52 na 56.89
May-01 57 59 53 50 59 56 55 52 na na 52 na 54.78
Jun-01 na 59 55 51 50 57 na na 50 na 51 na 53.29
Jul-01 na 56 52 52 56 57 59 55 na na 47 na 54.25
Aug-01 na 55 51 50 52 53 55 na na na 50 na 52.29
Sep-01 86 81 87 80 na 84 86 84 82 na 82 na 83.56
Oct-01 84 80 na 84 88 89 92 89 na na 78 na 85.50
Nov-01 85 84 87 84 86 87 89 87 88 na na na 86.33
Dec-01 82 86 86 na 82 85 89 82 85 na 81 na 84.22

Jan-02 na 83 84 80 79 86 86 77 82 77 80 na 81.40
Feb-02 83 81 82 78 79 84 83 na na na 74 na 80.50
Mar-02 74 80 81 na 77 82 82 75 na 75 74 na 77.78
Apr-02 71 79 76 74 75 78 79 75 74 72 69 na 74.73
May-02 73 72 77 na 74 76 78 72 75 71 70 na 73.80
Jun-02 70 74 70 70 70 71 77 70 69 na 69 na 71.00
Jul-02 68 70 76 67 62 70 72 70 67 67 62 na 68.27
Aug-02 61 69 68 60 63 65 69 na na 65 na na 65.00
Sep-02 70 66 66 63 68 66 69 65 64 65 na na 66.20
Oct-02 61 65 67 na 64 66 67 61 63 63 63 na 64.00
Nov-02 60 68 63 61 65 63 67 64 na na na na 63.88
Dec-02 na 65 64 61 64 61 66 55 na 64 64 na 62.67

Jan-03 56 63 63 58 na 65 59 53 62 64 na na 60.33
Feb-03 61 57 61 54 52 59 64 62 54 58 63 na 58.64
Mar-03 53 60 57 55 na 56 62 54 61 57 62 na 57.70
Apr-03 71 71 70 67 70 68 71 62 62 51 54 na 65.18
May-03 65 65 69 65 na 67 71 63 71 64 61 na 66.10
Jun-03 61 66 64 60 61 67 68 na 62 na 58 na 63.00
Jul-03 55 60 62 58 na 66 59 55 na 60 53 na 58.67
Aug-03 53 59 60 56 57 60 59 na 56 58 52 na 57.00
Sep-03 52 58 52 55 na 55 56 52 na 52 45 na 53.00
Oct-03 51 52 55 50 59 52 54 na 49 51 49 na 52.20
Nov-03 52 52 54 50 na 54 56 52 51 52 48 na 52.10
Dec-03 54 52 63 57 50 60 59 54 58 59 53 na 56.27

Jan-04 50 58 60 56 na 50 58 54 54 56 49 na 54.50
Feb-04 48 48 51 48 51 50 50 54 na 47 na na 49.67
Mar-04 48 48 49 46 na 51 50 na 50 48 na na 48.75
Apr-04 49 50 52 48 48 46 51 49 na 48 47 na 48.80
May-04 42 48 47 44 na 41 47 46 47 48 42 na 45.20
Jun-04 na 48 49 48 50 42 47 na 45 48 46 na 47.00
Jul-04 48 47 47 46 na 45 50 50 48 50 49 na 48.00
Aug-04 45 51 51 46 48 46 50 51 47 49 44 na 48.00
Sep-04 48 50 52 46 45 48 50 53 47 54 47 na 49.09
Oct-04 46 49 46 44 na 49 53 53 49 47 49 na 48.50
Nov-04 46 49 53 46 na 51 50 na 49 51 51 na 49.56
Dec-04 49 48 53 48 50 49 48 na 49 51 50 na 49.50

Jan-05 na 52 52 50 na 49 52 49 50 49 48 na 50.11
Feb-05 50 51 57 46 48 49 50 53 50 45 49 na 49.82
Mar-05 45 52 52 45 na 49 50 na 50 48 46 na 48.56
Apr-05 na 49 48 44 44 43 47 53 48 44 46 na 46.60
May-05 na 47 50 43 na 44 48 na 47 47 46 na 46.50
Jun-05 na 48 47 42 45 46 48 46 na 43 46 na 45.67
Jul-05 na 47 49 47 na 42 48 na 46 42 43 na 45.50
Aug-05 42 45 45 43 40 45 45 na na 42 45 36 42.80
Sep-05 38 41 40 40 40 41 42 42 40 39 40 37 40.00
Oct-05 40 41 42 40 40 37 39 42 39 39 43 38 40.00
Nov-05 36 36 37 36 34 35 39 42 38 37 37 36 36.92
Dec-05 36 42 42 38 34 40 47 41 39 42 38 40 39.92

Jan-06 na 42 43 38 43 41 42 41 39 40 39 36 40.36
Feb-06 na 39 38 40 40 34 na 40 na 40 40 38 38.78
Mar-06 36 39 na 33 36 34 41 37 37 37 38 37 36.82
Apr-06 na 33 32 35 35 33 38 na 36 na na 34 34.50
May-06 35 35 36 33 29 35 33 na na 33 32 32 33.30
Jun-06 na 41 37 36 33 33 38 35 37 35 36 36 36.09
Jul-06 38 36 40 36 34 36 na na 39 36 36 35 36.60
Aug-06 36 38 42 37 34 36 40 38 na 33 35 36 36.82
Sep-06 41 42 37 40.00

Nwk Fox CNN Pew Harris CBS ABC Time NBC AP Zogby ARG Avg

____________________

LOTS of great comments-- I'll follow up with some new posts that address a number of these issues.

The gas price and approval relationship is a lot more complicated than any analysis I've seen recognizes. Both gas prices and Bush approval have strong trends, prices up and approval down, that GUARANTEES a strong correlation between the two even if there is absolutely no causal connection between the two. To further complicate things, international events may well cause spikes in both approval and in the price of oil, so both may respond to events making untangling the causal relationship still harder. Having said that, of course presidents are evaluated in part on economic conditions in the US and gas prices are part of that. So I'm prepared to believe there may be a some connection between gas price and approval, but I think the simple-minded plotting of the two series when we know both are trended for other reasons, is a dubious practice. Certainly focusing on the simple correlation between the two is deeply misleading.

On the timing of the British plane plot and approval, the immediate reaction in the polls did NOT seem to be at all dramatic. But now with many more polls to use, the timing of the upturn seems to be a little delayed but still shortly after the plot was uncovered. Cause and effect? I don't know. If we had seen an immediate effect of the plot that would be easier to credit.

On ABC and other house effects, I've done a lot of analysis of that. The latest update (from August) is at Political Arithmetik here

http://politicalarithmetik.blogspot.com/2006/04/house-effects-and-presidential.html

Various pollsters vary in their house effects, for a great many reasons. But the effects I find and analyze in the linked post are generally pretty modest and seem to me to be well within the range that is attribuable to the effects of standard operating practices and questionnaire design. I don't think any of these organizations stand inexplicably far from the mean. ABC/Washington Post is above the mean, but others are higher. Likewise, several are further below the mean than ABC/WP is above. So singling out this or other polls as "shills" seems to me not to be supported by the evidence. The politicization of polling and analysis of polling is quite remarkable. For every condemnation by a liberal I see a similar but opposite condemnation by a conservative. That is one of the reasons I've focused on house effects at Political Arithmetik. Polls DO vary, and we should be concerned with estimating that variation. The more we know about this the better we can understand differences across polls in their results. But these differences are not direct evidence of any partisan or ideological bias.

(Full disclosure: I work for the ABC News election night Decision Desk which analyzes exit polls and election returns. I have no connection with the ABC Polling Unit.)

Charles

____________________

Charles,

I'm sure that you'll take a serious look at the Adult/RV/LV sample universe issue, but my intuiution tells me it is probably worth 1-2 percentage points for Bush. We're seeing polls move to RV or LV samples, so some of the "significant upturn" you note since Sept. 1 is probably a consequence of changing sample methodology.

What will be the turnout rate in the 2006 election? We know that midterm turnout is related to competition in the top of the ticket races, so we can expect higher turnout where there are competitive Senate and Governor races. I am curious to know if the LV models used for national samples adjust for state effects, or if they assume a national turnout rate. I don't think that a typical national poll could reasonably correct for state level turnout effects because the sample size in any one state is too low. Given the variation in state level midterm turnout, I rather suspect that LV models based on a national turnout estimate introduce some unknown, but non-random error.

Do we know what the expected turnout rate that polling organizations are using? I suppose we might use 41.1% among the voting-eligible population in the 1994 election as a starting point. Again, national turnout is sensitive to competitive races in the big states of California, New York, Texas, etc. In 1994, California had a VEP turnout rate of 46.9% due to the competitive Senate race between Feinstein and Huffington. 8.7 million of the 75.1 million 1994 voters cast a ballot in California. Whether or not we break 41.1% will largely depend on the competitive nature of the California governor's race. I would be surprised if it topped 46.9%, especially considering that few (or none) of the congressional and state legislative races are competitive. So my best guess from the back of the envelope is a national turnout rate somewhere around 40%.

This says nothing about who will turn out to vote.

And doesn't it seem a little early to move to a LV model, which is what FOX/Opinion Dynamics is now reporting?

-Michael McDonald

____________________

rdandrea:

Charles--I'm smart enough not to confuse correlation with causation. I also know that the price of gasoline and presidential approval are both "dependent variables" that are caused by too many other factors to evaluate, most of which aren't domestic.

However, I think that when people are polled, they aren't worried about what's going on in far-off corners of the world. They're worried about bread-and-butter issues; the cost of going to work, going to the store, going on vacation, taking the kids to soccer practice among them.

So although the price of gasoline is certainly not something against which you'd regress presidential approval in any rigorous sense, it is an excellent predictor nonetheless.

I believe that manipulating the price of gasoline is an effective means by which to manipulate public sentiment. And to tell you the truth, there's an election just around the corner, all of a sudden crude and gasoline supplies are there, and nothing else has changed.

____________________

Drew Mackenzie:

Echoing rdandrea - people vote with their gut. Mostly, they vote on how best to fill it.

____________________

This small sample indicates that the difference beween the independent and media poll averages is significant, when you use Group MoE's.

A typical rating poll has a 1000 sample-size, a 3% MoE. Grouping polls increases sample-size and lowers the MoE.

4 independent pollsters have provided Sept. ratings. The average rating is 37.75%. The group MoE is 1.58%

3 corporate media pollsters have provided Sept. ratings. The average rating is 41.0% The group MoE is 1.83%

There is a 3.25% average difference in ratings between the two groups, nearly double the 2-group average 1.70% MoE.

Let's watch the trend as the election approaches and we have more ratings to analyze..

95% confidence Limits
Group Sample AvgApp MoE Low High
Indep 4 4000 37.75% 1.58% 36.2% 39.3%
Media 3 3000 41.00% 1.83% 39.2% 42.8%

Independent: Zogby, ARG, Harris, AP
Type NWK Fox CNN Pew Har CBS ABC Time NBC AP Zog ARG Avg
Indep na na na 37 38 na na na na 39 37 na 37.75
Media na 40 41 na na na 42 na na ba na na 41.00

____________________



Post a comment




Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.

MAP - US, AL, AK, AZ, AR, CA, CO, CT, DE, FL, GA, HI, ID, IL, IN, IA, KS, KY, LA, ME, MD, MA, MI, MN, MS, MO, MT, NE, NV, NH, NJ, NM, NY, NC, ND, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VT, VA, WA, WV, WI, WY, PR