Recently I received a message from Professor Stephen Marrin (of Mercyhurst College) via the IAFIE (International Association for Intelligence Education) listserv talking about a U.S. presidential election prediction model developed by Professor Allan J. Lichtman of American University.  What was cited by Professor Marrin was the 08-25-08 Washington Post article entitled “In the Quake Model, Rumblings Favor Obama.”  Following the chain, the Washington Post article was really just a summary of Professor Lichtman’s peer-reviewed article entitled “The keys to the white house: An index forecast for 2008″ published in the International Journal of Forecasting, Vol. 24, No. 2, pp. 301-309 (doi: 10.1016/j.ijforecast.2008.02.004).  The abstract for this article is as follows:

The Keys to the White House are an index-based prediction system that retrospectively accounts for the popular-vote winners of every American presidential election from 1860 to 1980, and prospectively forecast the winners of every presidential election from 1984 through 2004 well ahead of time. The Keys give specificity to the theory that presidential election results turn primarily on the performance of the party controlling the White House. The Keys include no polling data and consider a much wider range of performance indicators than economic concerns. Already, the Keys are lining up for 2008, showing how changes in the structure of politics will produce a Democratic victory, in a dramatic reversal from 2004. The Keys also suggest that candidates need not follow the empty scripted campaigns of the recent past, but should instead be liberated to offer forthright discussions of the issues and ideas that will shape America’s future.

While I haven’t yet read this paper in great detail, I am curious as to how it works, and what amount of analytic confidence can be afforded to the model’s output.  What I am particularly curious about is the form of the output – is it an absolute answer, or does the model provide a probability distribution over the space of party candidates?  And what about third parties (or are they just in the noise)?

But what I can say is that, in principle, prediction technologies such as the “keys to the White House” model can be used by any individual or group of individuals (e.g., foreign countries) with a vested interest in the outcome of this or any other US presidential election to assess their environmental (e.g., political) risks, thereby providing a frame with which to evaluate their future decision options.  But do you think the model is robust enough to include the effects of supposed “game changers” (e.g., Ms. Palin’s recent entry into the GOP ticket)?  Keep in mind that if the “keys to the White House” model is, in fact, a credible model, it could also be used, at least to some degree, as a tool to help the otherwise losing party reshape their campaign toward winning, whether by pursuing a target PR campaign, changing policy positions, or whatever else that can influence the underlying “keys” (or rather, indicators) that enable prediction.  So perhaps the choice of Ms. Palin to run as the GOP vice president was targeting one of these keys…  After all, according to Professor Lichtman,

“Not only elections, but also election forecasts have consequences for politics.  The models that we use to predict and understand presidential elections shape the conduct of campaigns, the relationships between candidates and the American people, and ultimately the policies of government.”

Send article as PDF to

At 5PM EST, Saturday, 08/30/08, the National Weather Service, National Hurricane Center released their estimate of probability of storm surge afflicting New Orleans sometime over the next 72 hours (see below):

Mayor Ray Nagin of New Orleans is labeling Gustav as the “mother of all storms.”  Apparently, New Orleans (the brownish colored area amidst the probability contour) has between a 40-60% probability of storm surge (it is hard to tell whether New Orleans is in the yellow, brown, or partial to both).  What was my estimate from the 2PM EST data?  My estimate was 50% that Gustav would result in an appreciable storm surge.  So, I suppose this is confirmation of my initial estimate (perhaps not, but it is nice to think so).  The next question is whether Gustav will result in storm surges as bad or worse than Katrina.  Based on what I am reading, I suspect so, but we shall see.

(note: this post references my previous post from just a few minutes ago)

Send article as PDF to

In light of the temporal intersection of two very significant events – Hurricane Gustav and the Republican National Convention – I can’t help but perform a risk assessment of Gustav from the point of view of the Republican presidential campaign.  While one can speculate on a variety of bad things that could come out of Gustav that could affect Republicans or Democrats, I will focus my attention on the simple questions “will Gustav afflict New Orleans?” and “what if it does?”

What Can Happen?: The Republican Point of View

Let’s begin this analysis by developing a full set of hypotheses and sub-hypotheses for this study that are of interest to the Republican presidential campaign.  The events of concern for this analysis center on Hurricane Gustav and New Orleans.  The outcomes of concern are assessed in terms of “delta-support” for the Republican presidential campaign, that is, whether in the end the situation increases support, decreases support, or does not effect support for the presidential candidate in the time frame immediately following Gustav.  This analysis does not examine whether the “delta-support” can be sustained for the remainder of the election season.

The first question is whether Gustav will afflict New Orleans?  More specifically, will Gustav afflict New Orleans in such a way that causes an appreciable storm surge?  There are two possible answers to this question:

• A1: Gustav will afflict New Orleans
• A2: Gustav will not afflict New Orleans

Since the scope of this risk assessment is on the risks to the Republican party due to Gustav afflicting New Orleans, we will restrict most of our attention to A1.  This hypotheses can be further divided into several sub-hypotheses as follows:

• B1: Gustav will be as bad as or worse than Katrina in terms of intensity
• B2: Gustav will not be as bad as Katrina in terms of intensity

I partitioned A1 into B1 and B2 above for a simple reason: it is easier to make comparisons between Gustav and Katrina if they are similar in intensity than if Gustav is weaker than Katrina.

The next issue to consider is the levees – given that Gustav afflicts New Orleans, we have the following set of hypotheses:

• C1: The hurricane protection system will fail to protect New Orleans
• C2: The hurricane protection system will successfully protect New Orleans

If C1 occurs, then all is well for New Orleans (and as we will see later, potentially very good for the Republicans).  However, if C2 occurs, then we have two more hypotheses to consider:

• D1: The consequences of breach are disastrous
• D2: The consequences of breach are not disastrous

For the purposes of this analysis we need not qualify consequence in terms other than “disastrous” or “not disastrous” as it would not offer any additional insight in light of the cost of a more refined analysis.

Given Ai, Bj, Ck, and Dm, we can construct a possibility tree for this situation as shown below, where there are 7 possible end points, some good, some bad, and some neutral from the Replublican campaign point of view.  This possibility tree defines a complete (i.e., exhaustive) set of distinct (i.e., mutually exclusive) hypotheses relevant to this assessment.

How Likely Is It To Happen? (Likeliness Assessment)

This next phase of the analysis assesses the likeliness of events A1, A2, B1, B2, C1, C2, D1, and D2.  Let’s begin with A1 and A2.  According to the National Weather Service in Advisory 24 released at 2:00pm EST on Saturday 8/30/08, New Orleans falls within the 67% probability envelope of potential hurricane paths (see below).  This means that NWS assigns a 2:1 odds that Gustav’s path will pass within a region shown in white.  The residual 33% of total probability is reserved for the region outside of the white envelope, where there exists hurricane paths also with the potential to afflict New Orleans.

So, what is the probability that Gustav will cause a storm surge to occur that challenges the New Orleans hurricane protection system? If you examine the potential storm track of Gustav, the region to the right of the envelope centerline corresponds to a family of storm tracks that seem to push water toward New Orleans.  The family of storm tracks to the left of the centerline seems to push water toward the coast of western Louisiana and the eastern shore of Texas.  Treating the centerline as the median line, and assuming any storm track to the left of the median will not cause an appreciable storm surge and everything to the right of the centerline will, I assess it to be about even odds (50% chance) that Gustav will afflict New Orleans in an appreciable way.  Thus:

• Pr(A1) = 0.5
• Pr(A2) = 0.5

From the historical record maintained by the NOAA Coastal Services Center Historical Hurricane Tracks database, note that Katrina’s path (shown below), too, seemed to push water toward New Orleans.

To answer the question whether Gustav is poised to be as bad or worse than Katrina given that Gustav causes an appreciable storm surge, let’s compare the projected storm track of Gustav with that of Katrina.

What does this tell us?  First, note that starting at some point in time when Katrina was north of the western tip of Cuba, Katrina transitioned into a category 4 storm.  Similarly, Gustav as of the time of this image (at the Orange circle), too was a category 4 storm.  For both Katrina and Gustav, the paths between the point of transition of the estimated point of landfall more or less allowed for free movement over a body of water more than willing to fuel the storm’s intensity.   Much like Katrina, I suspect that Gustav will trudge along the Gulf of Mexico toward land, picking up speed along the way.

Also from this figure, the worst part of Gustav (in terms of wind speed) will happen with chances 67% or greater sometime between 8AM EST Monday 9/2/08 and 8AM EST Tuesday 9/3/08.  To see this, just see where the coastline intersects the white uncertainty envelope.  According to National Weather Service Advisory 24, the probability that the storm is a category 3 in or higher storm is 25% or greater.  However, I would argue that were Gustav to follow a trajectory to the left of the storm track centerline, the probability of occurrence would be much higher for category 3 and higher.  But since I lack this information at present, I will settle for the following probability judgement expressed in terms of probability intervals:

• Pr(B1|A1) = [0.25 1.00]
• Pr(B2|A1) = [0.00 0.75]

Now on to the integrity of the New Orleans Hurricane Protection System (NOHPS).  This question is the true object of contention with respect to the purpose of this risk assessment.  Given that Gustav strikes New Orleans, will the NOHPS hold up?  The answer to this question depends on the two factors: the intensity of the storm surge resulting from Gustav, and the structural resistance of the NOHPS with respect to this storm surge.  According to an article in the Seattle Times summarizing US Army Corps of Engineers progress toward upgrading the NOHPS, the $15-billion NOHPS upgrade scheduled for completion in 2011 is only 20% complete.  Of course, one doesn’t improve a system as immense as the NOHPS uniformly, but rather in discrete chunks.  So it is reasonable to assume that 20% project completion means 80% of the length of levees and flood walls in New Orleans is left “as good as old” where “old” = “Pre-Katrina.”  But because the USACE has a lot of smart people engaged in this project, we should assume that the project scheduled fortification of the most vulnerable sections first, leaving fortification of the less vulnerable lengths for later on in the project.  So, this 20% complete might very well mean that the weakest sections have been fixed, with less weak sections on the schedule for future improvement.

That said, I am going to go with the following (admittedly waggish) probability estimates:

• Pr(C1|A1,B1) = 0.20 (20% chance the NOHPS will fail if the storm surge is as bad or worse than Katrina)
• Pr(C2|A1,B1) = 0.80 (80% chance the NOHPS will not fail if the storm surge is bad or worst than Katrina)
• Pr(C1|A1,B2) = 0.05 (small chance that the NOHPS will fail if the storm surge is not as bas as Katrina)
• Pr(C2|A1,B2) = 0.95 (very high probability that the NOHPS will not fail if the storm surge is not as bad as Katrina)

What remains is to assess the likeliness of D1 and D2.  If the NOHPS fails under conditions as bad or worse than Katrina, it is reasonable to assume that the impacts to New Orleans would be perceived as disastrous (though perhaps not as disastrous as Katrina given Gov. Jindal’s preparedness efforts).  Now, if the NOHPS fails under storm surge conditions not as bad as Katrina, I admit a small chance to the whole thing not turning out to be a disaster, say 10%.  Thus we have:

• Pr(D1|A1,B1,C1) = 1.00 (certain that it will be a disaster if the NOHPS fails and the effects of Gustav are as bad or worse than Katrina)
• Pr(D2|A1,B1,C1) = 0.00 (ditto above)
• Pr(D1|A1,B2,C1) = 0.90 (see comment for the very next probability)
• Pr(D2|A1,B2,C1) = 0.10 (my admission of a small probability that failure of the NOHPS under conditions not as bad as Katrina will not yield a “disaster”)

At this point I have all the probabilities I need and can now calculate the probability for each scenario E1 through E7 as follows (expressed as percentages).  Please ignore the decimal precision as it is a by product of multiplying otherwise imprecise, representative numbers.

What Are the Political Outcomes?

This final step of the risk assessment examines each of the 7 end scenarios and attempts to assess whether each scenario will increase support for the Republican party, decrease support, or have no effect.  A five-level ordinal consequence scale is constructed for expressing outcomes as shown in the table below.

Now let’s rate even scenario E1 through E7 according to how it will impact near term public support for the Republican candidate.

• E1: No one can deny that if Gustav turns out to be Katrina-like or worse AND the NOHPS fails, then the Democrats are in a strong position to argue that the Republican party once again failed to protect New Orleans even when it had 3 years to do so since Katrina.  My bet is that Obama will argue that McCain will just offer “more of the same.”  Thus, I assess the political outcome from this scenario will be “- -” (two minuses, or a score of -2).
• E2: However unlikely this scenario might be, if Gustav turns out to be as bad or worse than Katrina, and the NOHPS fails to protect New Orleans, the immediate political outcome is unfavorable to the Republicans even if the event does not turn into a disaster.  But because Gustav did not result in a disaster, technically the situation is not as bad as E1.  Thus, I assess the political outcome from this scenario to be “-” (one minus, or a score of -1).
• E3: If Gustav turns out to produce a storm surge that is as bad or worse than Katrina, the fact that the NOHPS system protects New Orleans against a disaster stemming from flooding can be spun to mean that the Bush administration took the necessary steps to protect against hurricanes in the wake of Katrina.  The Republicans can (and will) play this up as a policy success.  Thus, I assess the political outcome from this scenario will be “++” (two pluses, or a score of +2).
• E4: If Gustav turns out to not be as bad as Katrina, yet the NOHPS fails to protect New Orleans against disaster, the Republicans should run and hide.  This situation can be spun as if the NOHPS, after several billion-dollars investment toward upgrades by the Bush Administration, is actually worse now than it was before Katrina (an arguable position, true, but one that seems likely to come from election opposition).  Thus, I assess the political outcome from this scenario will be “- -” (two minuses, or a score of -2).
• E5: If Gustav turns out to not be as bad as Katrina, yet the NOHPS fails to protect New Orleans, it really doesn’t matter what the outcome is (disaster or not).  This situation can be spun as if the NOHPS, after several billion-dollars investment toward upgrades by the Bush Administration, is actually worse now than it was before Katrina (an arguable position, true, but one that seems likely to come from election opposition).  But the discussion of whether New Orleans should be inhabited at all will likely rise to the forefront, and it is for me to credibly speculate on which party (or candidate) will offer the best (or most palatable) advice.  Thus, I assess the net political outcome from this scenario will not favor the Republicans, so I give it a “-” (one minus, or a score of -1).
• E6: If Gustav turns out to not be as bad as Katrina, and the NOHPS does its job successfully, then the Republicans can tout that they are doing a good job making progress toward a safer New Orleans.  While not nearly as strong as E3, the Republicans can, and probably will, squeeze out some brownie points for their effort.  Thus, I assess the political outcome from this scenario will be “+” (one plus, or a score of +1).
• E7: If Gustav does not appreciably affect New Orleans, then really there is nothing to worry about.  This does not mean that the topic of domestic preparedness will not come up during the remainder of the election season; rather, all I am claiming is that neither party can leverage Gustav to their benefit in any substantial way.  Thus, I assess the political outcome from this scenario to be “N” (neutral, or a score of 0).

With this consequence assessment in hand, the probabilities from the previous section can be used to construct a discrete probability distribution over the space of five political outcomes as shown below.  This figure communicates political risk, or rather, the uncertainty surrounding future political outcomes.  Again, ignore the apparent precision in the probability numbers – the interval actually makes them imprecise, but even so the digits past the decimal points are only a biproduct of arithmetic.  In the next section, you will see that the digits really don’t matter.

Summary of Results

Based on this assessment and without getting too detailed, we can arrange the three major outcomes in order of decreasing likeliness as follows:

• Neutral, or no affect [50%]
• Favorable political outcomes for the Republicans [~38%, 47.5%]
• Unfavorable political outcomes for the Republicans [2.5%, ~12%]

I think if I were the Republican campaign manager and I trusted this analysis, I would be excited and would be preparing myself with press statements and political advertisements aimed at bolstering the Republican candidate.  In contrast, if I were the Democratic campaign manager, I would start investing resources now in an effective press release aimed at mitigating the extent of any Republican political advantage following this event.

IMPORTANT CAVEAT: All things considered, and admitting that I am not learned in storm surge hydrodynamics nor do I fully appreciate the complexities of politics and political spin, I assign an overall analytic confidence of LOW (25% confidence, or a 67% chance that my analysis is valid).  Also, I am not saying what will happen one way or the other; rather my analysis just produced a subjective probability distribution over the space of potential political outcomes.  So, if a disaster does occur, my analysis is still not wrong.  After all, I never said disaster was impossible…

Afterthought

I could also do such an assessment from the Democrat point of view, but in this case the Republican’s loss is the Democrat’s gain (though admittedly the third parties may also absorb some of the benefit, so it is not quite a conceptual zero-sum game).  The conclusions would remain unchanged, however.

Send article as PDF to