Decision Making In Complex Environments
BQOM 2521
Professor Thomas L. Saaty
A Study of the 2004 United States Presidential Election Using the Analytic Network Process (ANP)
By John J. Barsic
Introduction of the Problem The purpose of this article is to conduct an analysis of the 2004 United States Presidential Elections using the Analytic Network Process (ANP) developed by Professor Thomas L. Saaty of the University of Pittsburgh. Traditionally, this approach is used to structure highly complex decisions from the perspective of a single decision maker or a body of executive decision makers acting on behalf of a corporation or government entity. However, within the context of the election decision, the problem will be structured and developed from the perspective of the American public. Accordingly, the stated goal of the problem will be to select the best presidential candidate such that potential benefits and opportunities associated with the elected person’s leadership are maximized, and such that costs and risks are minimized. The specific components of the BOCR model will be duly outlined and discussed below. Necessary Assumptions Before proceeding, several reasonable assumptions will be necessary to clarify expectations about the context of the decision within this decision making framework. It is already widely acknowledged by the American public that the main two candidates for the presidential election are incumbent President George W. Bush, representing the Republican Party and Senator John Kerry, representing the Democratic Party. By way of completeness it is also necessary to acknowledge that there are various other candidates representing smaller political parties, special interests, and independent candidates (most notably Ralph Nader). Third party candidates may have had a significant impact on election results in past elections (for example, the influence of Ross Perot on the outcome of the 1992 presidential election). However, it is not anticipated as of yet that third party candidates will have a significant impact on the 2004 election, nor can that effect be measured or reliably estimated at this time. Therefore, for the purpose of this model, the election shall be regarded as a two-candidate race between George Bush and John Kerry. One other caveat necessary to mention is that forecasts of election results (and indeed the outcome of the election itself) are heavily impacted by public perception at the time of the forecast. The critical issues surrounding the election
-2-
will be highlighted in the next section. However, these issues are subject to change, and one or more of the issues which heavily influences public opinions at this time may be a non-factor on the day of the election. Conversely, critical issues with the potential to significantly change election results and margins of victory emerge quickly, and may impact the election at any time between Now and November 4, 2004. Indeed, even during the few course of the development of this model and the completion of this project, political news with the potential to influence election results emerged on several occasions (for example, the onset of increased violence and U.S. casualty rates in Iraq, the new book by Richard Clarke "Against All Enemies" and the flurry of other political books now in the spotlight, Condoleza Rice's 9/11 testimony, Spain's decision to withdrawal troops, and the list goes on). Similar events are prone to emerge at any time between now and the election. Therefore, it is necessary to realize that the results of the analysis reflect conditions and public sentiment as of April 2004 (the timeframe for the development of the model). To the extent that issues and conditions are similar during the week of the election, then the likelihood that the model forecasts an accurate result will be higher. Critical Election Issues One of the most important issues for voters in the upcoming election is the creation of more domestic jobs. As the United States economy continues to struggle during a period of recovery, this is a much more pressing concern for U.S. voters than it has been in the last few elections, which occurred during a very robust period for the economy. However, the overall increase in the number of jobs created is not necessarily meaningful to voters, as many jobs are outsourced to foreign countries. For example, it is estimated that IBM will create approximately 15,000 jobs over the course of the upcoming fiscal year. However, over 5,000 of these jobs are anticipated to be outsourced to India, Russia, and Europe where certain technical skills may be obtained at a lower salary cost. Clearly this impacts people who have been displaced from wellpaying jobs as well as college students vying for new positions. Therefore, U.S. voters regard the creation of domestic jobs that are not outsourced as an extremely important issue at this time. Another issue which holds the full attention of the American public, and one that will likely be pivotal during the election is the United States strategy for a successful exit from Iraq. There are at least two major dimensions to this issue. The first is to minimize further U.S. casualties by bringing troops back home as
-3-
quickly as possible. The second dimension is to establish a stable and selfsufficient government in Iraq before relinquishing military control. If power does not transition smoothly to a well-established Iraqi government entity, the country could succumb to chaos or civil war. In the event that this occurs, the primary mission of the United States in initiating military action will have failed, and there could be a tragic sense that U.S. soldiers have died in vain. Due to the extremely precarious circumstances surrounding the U.S. presence in Iraq, the effective resolution through a viable exit strategy will clearly be of utmost importance to voters. The third major issue likely to envelop the 2004 elections is the state of the Social Security system. Recently, predictions have been made which indicate that Social Security will encounter a funding deficit (the equivalent of corporate bankruptcy) earlier than anticipated, with some predictions indicating that this may occur by 2017 or even earlier. One major initiative is to prevent the bankruptcy of the Social Security system by ensuring proper stewardship of funding. This includes the prohibition of practices such as using social security surpluses to cover current deficits in other government programs. Only through proper financial stewardship can the Social Security system maintain its existing level of benefits. Another Social Security initiative is to provide prescription coverage to all beneficiaries. This has become an increasingly important issue, as the costs for prescription medication in recent years have outpaced inflation by many orders of magnitude. This increase in prescription costs have a particularly devastating effect on senior citizens who are likely to require more frequent and more expensive drug treatments. The fourth issue of significant importance for the election is tax policy. This issue frequently plays an important role in presidential elections, and the 2004 election is not an exception. There are several aspects of the tax policy which draw the attention of voters. The first is the introduction of a tax relief package that would involve tax cuts beneficial to all demographic groups rather than policies which only benefit wealthy individuals. Another issue involving corporate tax policy is to reduce or eliminate incentives which provide tax relief to corporations that engage in outsourcing of jobs. Clearly, this is a practice that most U.S. voters have a desire to eliminate. Development of the BOCR Model
-4-
In order to illustrate the development of the Benefits, Opportunities, Costs, and Risks (BOCR) model, the subnet under Benefits will be examined closely by drilling down into successive levels of detail until one of the decision subnets is reached. The outcome of the Benefits Subnet will be explained in terms of this progression. Since similar progressive logic can be applied to all subnets, it is unnecessary to reproduce the discussion for all aspects of the main model, as the reader can infer that the construction of the model follows approximately the same pattern as the one illustrated explicitly below. Therefore, the results of the Opportunities, Costs, and Risks modules will be summarized to eliminate the repetitive mechanical descriptions of model development in favor of analytical interpretation of the results. The following image illustrates the structure of the subnet under Benefits. Notice that the potential control criteria are grouped into four broad categories that closely reflect the issues highlighted in the discussion above. Each category features several nodes which are the potential control criteria.
-5-
Pairwise comparisons among all nodes and comparisons between clusters produce a prioritization which reveals the nodes which will be selected as control criteria. The following graphical illustration of the node priorities clearly illustrates that the creation of domestic jobs is the highest priority item within the Benefits subnet. As such, job creation is clearly a control criterion which merits the creation of a decision subnet under its node. Other nodes do not have levels of prioritization as decisive as the job creation node. However, there are several other nodes with prioritizations high enough to merit selection on a judgmental basis. In this case, the establishment of a stable government in Iraq and the provision of adequate social security funding were also selected as control critera, and decision subnets were created for these nodes as well.
In order to examine the very core of the model where true feedback occurs within a decision subnet, drilling down into the model one more level is required. Therefore, the decision subnet under job creation (the highest priority item under the benefits subnet) will be examined. An image of this subnet appears below.
-6-
Of course, the alternatives within the decision subnet are uniform across all decision subnets, and consist of two alternatives labeled Bush and Kerry. Within the job creation subnet, the factors affecting the selection between the alternatives were grouped into three clusters, including party platforms, demographic groups, and political action groups. Party platforms refer to the particular issues (or “planks” in a political platform) that are endorsed by each presidential candidate. Within the party platforms cluster, there are nodes corresponding to each “plank” or campaign promise, including more domestic jobs, tax policy, economic growth, trade policy, and international policy. Within the demographic groups cluster, there are nodes representing some of the largest groups of voters with similar characteristics and views, including white collar, blue collar, wealthy, and retired voters. The political action groups cluster contains nodes representing some of the most powerful and influential political action groups such as labor unions, conservatives, liberals, and businesses.
As in other subnets, pairwise comparisons were made among the nodes and also between clusters in order to determine which candidate had the highest
-7-
priority within the decision subnet. A snapshot of the synthesized outcome appears below, and shows that John Kerry has an overwhelming advantage of 64% vs. 36% for George Bush with respect to voter prioritization under job creation. This lopsided result can be explainable in terms of public perception, as John Kerry has made an attempt to position himself as a strong proponent of U.S. labor and keeping jobs in America while George Bush is perceived as an ally of big business and wealthy voters. In addition, George Bush has indicated that job outsourcing helps the U.S. economy, and this is a position which has drawn a great deal of public criticism by Americans concerned with retaining domestic jobs.
There were two other decision subnets under benefits: one for establishing a stable government in Iraq, and one for ensuring adequate Social Security funding. However, based on the prioritizations form above, these attributes did not have as great an influence on the outcome of the Benefits subnet as job creation. Accordingly, John Kerry also had a higher prioritization under benefits by a relatively wide margin of 61% to 39% for George Bush as illustrated by the synthesized results of the benefits subnet that appears below.
-8-
Summary of BOCR Results The following discussion will provide a summarization of the results of the benefits, opportunities, costs, and risks subnets in lieu of a repetitive mechanical analysis as presented for the benefits subnet above. Determination of the results for each of the B,O,C, and R subnets will be necessary for the development of the strategic (or external) model in the next section. As determined by the preceding discussion, John Kerry had a higher prioritization under benefits, due mainly to the public perception of his strong support for domestic job creation. Obviously, this is a tangible and highly desirable benefit for United States voters. For similar reasons, John Kerry also had a higher prioritization under opportunities, as economic growth and the prospect of lower unemployment also represent significant opportunities. In addition, there are also opportunities to improve relations with our allies and quickly evacuate U.S. troops form Baghdad. Relations with some U.S. allies
-9-
have been strained as a result of the U.S. invasion of Iraq. John Kerry’s stance is to garner as much international support from allies as possible by working with them instead of stubbornly imposing the will of the United States. This may enable a faster resolution of the problem in Iraq and enable more support from the United Nations. Ultimately, this could lead to an expedited withdrawal of U.S. troops from Iraq. Due to the perception that John Kerry is more willing to cooperate with allies, these opportunities have a higher chance of realization. This also made a significant contribution to Kerry’s higher prioritization under opportunities. Analysis of the costs and risks requires a reversal in logic due to the fact that the candidate with highest costs and risks are less desirable. Therefore, the candidate with the lower nominal prioritization in these subnets is the “winner”. This reversal of logic is necessary due to the additive (negative) formula utilized to synthesize the overall main model. Since costs and risks are subtracted from the optimization routine, a lower nominal prioritization for costs and risks indicates the favored candidate with respect to that subnet. In light of the above logical considerations, George Bush received a more favorable rating on both costs and risks. The primary reason for his advantage with respect to costs is that he has instituted tax cuts and has clearly indicated his commitment to keeping taxes as low as possible. On the other hand, John Kerry has criticized tax cuts, indicating that it could lead to high levels of national debt. His suspension of current tax relief initiatives backed by Bush could impose additional costs on individual and corporate taxpayers. Also, there is also an ancillary perception that Kerry, as a democrat, could introduce a “tax and spend” mentality leading to a larger and more costly federal government. With regard to risks, George Bush also received a more favorable rating due mainly to his tough stance on terrorism and experience in handling national crises such as the 9/11 terrorist attacks. Development of the Strategic Model The outcome of the benefits, opportunities, costs, and risks components within the main model enables the development of the strategic model. Completion of the analysis in the previous section reveals that John Kerry has a higher prioritization with respect to benefits and opportunities, and that George Bush has a more favorable prioritization with respect to costs and risks. Using this knowledge, the strategic criteria were rated using a scale of high, medium,
- 10 -
and low. The rating for each strategic criteria within each category (B, O, C, and R), requires consideration of which candidate was a superior alternative in each of these categories. This is the reason that the strategic model was complete after the determination of priorities within each B, O, C, and R subnet.
The actual structure of the strategic model is illustrated below. Rather than introducing unnecessary levels of complexity into the model, all strategic criteria were placed in one “supercluster” labeled as “Strategic Criteria”. The nodes within this cluster correspond to the strategic criteria, which include a high growth economy, good Social Security Benefits, good international relations, low unemployment, an efficient government with low levels of bureaucracy, and domestic safety from terrorism.
- 11 -
Synthesis of the above strategic model yields a prioritization for benefits, opportunities, costs, and risks. Benefits and costs received the highest prioritizations with 29.8% and 29.6%, respectively. Opportunities and risks received progressively lower prioritization at 25.3% and 15.2%, respectively. One intuitive explanation for this outcome is that benefits and costs are often more tangible and measurable than opportunities and risks (which are merely represent potential future outcomes that are less tangible at the present time). Since benefits and costs can be more readily quantified and assessed than opportunities and risks, presumably this is the reason that they received the highest prioritizations. Furthermore, since society tends to avoid the unpleasant prospect of assigning probability to large risks (such as terrorist attacks and war casualties), risks tend to receive the lowest prioritization. Risks are often regarded as a dynamic non-event until something drastic happens to capture public attention.
- 12 -
Final Synthesis of the Main Model The results of the BOCR prioritizations from the strategic model above were manually input into the main decision model to force the model to assign the priorities for benefits, opportunities, costs, and risks as determined in the strategic model above. Afterwards, a final synthesis of the top-level BOCR network reveals the final results for the prioritization between the two candidates. Based on the stated goal of the American public to select the best presidential candidate, we can interpret a higher prioritization as in indication of the winning candidate, with the percentage prioritization revealing an approximate margin of victory. The final synthesis of the main model reveals that John Kerry has a higher overall prioritization than George Bush, implying that the margin of victory will be 53% to 47% of the popular vote in favor of John Kerry. The following model excerpt displays the final synthesis which yields this result.
- 13 -
Sensitivity Analysis In the case of election results between the two candidates, the effective interpretation of sensitivity analysis for each of the four main subnets (benefits, opportunities, costs, and risks), is to illustrate the point at which election results would reverse. In this case, it can be interpreted as the minimum percent change in priority required within each of the BOCR subnets such that the election would change from George Bush to John Kerry. The model excerpt below illustrates the sensitivity analysis performed for the costs subnet. Since the lines cross at approximately .25, this indicates that a .25 change within the costs subnet would be enough to change the outcome of the election. Notice that the blue line representing John Kerry is the top line from 0 to .25, and the red line representing George Bush is the top line from .25 onward. By conducting a similar analysis for the other subnets (which is accomplished by changing the independent variables accordingly), we discover that the crossover point for benefits, opportunities, and risks are .35, .30, and .15, respectively.
- 14 -
Conclusion: Reconciliation of the ANP Model Outcome with Personal Intuition
One of the most valuable attributes of the analytic network process is the ability for the resulting model to provide a mathematical manifestation of the intuition which drives each component of the model and influences it’s pairwise comparisons. Therefore, as a natural conclusion of the modeling process, it is extremely important to reconcile the outcome with the intuition which helped to build the model. In the case of the 2004 United States Presidential Elections, the ANP Model forecasts that John Kerry will win over incumbent president George W. Bush. A close examination of some of the circumstances surrounding the upcoming election reveals some insight as to why this outcome may occur. It has already been determined that successful foreign policy (including a prompt and effective resolution to the crisis in Iraq) is a very important criterion for the selection of the next president. To the detriment of George Bush in this regard,
- 15 -
there is public criticism that he has isolated the United States from our allies by imposing the will of the United States as an aggressor in Iraq without the approval of the United Nations. There is also a certain degree of public mistrust that the Bush administration has lied about war issues such as the presence of weapons of mass destruction in Iraq. These perceptions, whether merited or unmerited, severely undermine President Bush’s credibility with respect to foreign policy. Another significant advantage for John Kerry in the upcoming election is that George Bush must actually prove that his exit strategy from Iraq will work by demonstrating tangible results (such as meeting the proposed transition of power to an established replacement government in Iraq by June 30, 2004). If there are significant delays or setbacks in the initiative to restore order and withdrawal troops from Iraq, these events could prove particularly damaging to the Bush campaign. On the other hand, John Kerry has the luxury of observing the outcome of events in Iraq without having responsibility for a potentially bad outcome. To this end, John Kerry essentially had the advantage afforded to an “armchair general” or “backseat driver” who can criticize bad results without the risk of the outcome affecting his own public perception. Kerry merely needs to formulate an exit strategy that is feasible, and claim that it would work better than the solution that Bush has implemented. He also has the benefit of hindsight in pointing out mistakes made by the Bush administration in the handling of the matter. One additional strength which may help to explain John Kerry’s advantage in the upcoming election is the public perception that he is more in favor of the working class than George Bush. Again, whether it is merited or unmerited, as a republican, George Bush has a reputation of favoring large corporations and wealthy individuals (for example through tax cuts that are more beneficial to the wealthy than to lower income classes and the average American worker). On the other hand, Kerry has strategically positioned himself as an advocate of creating and retaining new jobs in the United States, vehemently opposing the outsourcing of American labor. Since the creation of more domestic jobs and the continued strengthening of the U.S. economy are extremely important to voters, this is yet another reason why Kerry may have received a higher rating in the model.
- 16 -
Since public perception favors John Kerry on the factors which have the highest priorities within the BOCR model, naturally the overall outcome should reflect that he represents the best alternative, bearing the highest prioritization. This analysis shall conclude with one final reiteration that the model results are predicated upon issues that are important to the American public as of the April, 2004 timeframe. To the extent that new issues emerge between now and the election, the issue priorities may change in such a way that the final outcome is also affected. The only way to eliminate this uncertainty is to revise the model as new issues emerge.
- 17 -