For years it has been of great interest to social scientists to predict presidential election results. Throughout the course of these years many tools and techniques have been developed to aid in this endeavor. While some models do predict the winner within a certain reasonable margin of error, often times this margin of error negates the benefit of having such a model. In other words, the margin of error is so large that it provides little in the way of predictive power that a public opinion poll or simple guess would prove almost as accurate and fall within that same margin of error. Nonetheless, there are models that have been developed that are able to predict with relatively great accuracy the percentage of votes the incumbent party is expected to receive. One type of indicator that many of these models use to help predict the election results is economic. According to these models, if the economy is performing well then the incumbent party in the White House will be rewarded accordingly. However, if the economy is performing poorly then the incumbent party will be punished.

The exact definition of what it means for the economy to be performing well (or poorly) will vary by model and the economic indicators used to predict the results of the election. Nonetheless, if “inflation rate” as an independent variable is used to help predict the dependent variable “percentage incumbent party receives of popular vote” then we would expect a lower value for the dependent variable when the independent variable is higher, assuming a statistically significant negative linear relationship.

The use of economic indicators in predicting presidential election results is appealing for a simple reason—the economic data used in these models is generally accurate and not subject to measurement errors that are common in public opinion polls and other independent variables one may decide to use. For example, the Gross Domestic Product (GDP) is measured in a consistent manner that leaves minimal room for errors caused by the subjective disposition of the researcher and is henceforth able to remain a relatively objective data measurement. Previous models that have garnered attention have combined both these objective economic indicators with additional measures, often times including more subjective ones. The average absolute error of one model forwarded by Steven J. Rosenstone’s was only 2.9 percent (Rosenstone). In contrast, the average absolute error of the Thomas M. Holbrook’s work was 4.4 percent (Holbrook 105). However, one should be cautious in comparing the two models using numbers derived from simple statistics since the methods used to formulate the equations and what those equations are measuring are quite different from one another. For example, Holbrook is explaining the percentage of votes the incumbent party in the White House receives, while Rosenstone is measuring the percentage of votes the democratic candidate will receive. Additionally, some of the independent variables Rosenstone uses cannot be obtained until after the election, which significantly limits its value in predicting results. Nonetheless, Rosenstone’s model did correctly predict the winner 91 percent of the time between 1948 and 1972.

It has been shown that some economic indicators are effective in helping to predict presidential election results. Therefore, it may be beneficial to ascertain how they can be used most effectively. In this research I will attempt to determine how to use the economic indicators with predictive value in a manner that is most effective and accurate.

My goal is not to create a model with the greatest predictive value as these models generally become quite cumbersome and overly complex. For example, Holbrook created a model that helped explain presidential election results, but it also contained a total of 74 independent variables, 25 of which were political or economic in nature (93). Therefore, my approach will significantly simplify this by using far fewer variable(s), but variable(s) that can easily and consistently be measured in an objective manner. Additionally, the analysis of these variable(s) will be undertaken in a completely different manner. Instead of looking at static figures, I will calculate how certain economic indicators change in time and how these changes may help predict presidential election results. Thus, the new independent variable(s) will be “% change economic indicator A over time period Z”. This is a different approach that has not been done in exactly this manner previously and thus warrants attention.

Predictive Model for 2008
I have a more detailed paper in the final stages that includes my complete methodology, but I will leave that for another time when it is appropriate to share the complete analysis. At this time I simply want to share what my models predict for this election cycle prior to the election. There were two main approaches I used to predict the outcome of the presidential election: 1) models designed to predict the percentage of popular vote received by the incumbent party, and 2) models designed to predict the Electoral College vote received by the incumbent party. In reality, all that matters in an election is which party ends up with a majority of Electoral College votes. Nonetheless, percentage of popular votes has generally been a good indicator of this with some notable exceptions. Additionally, models designed to predict this have much smaller margins of error and thus lend themselves towards greater accuracy.

On the other hand, the threshold for whether a model is correct that is designed to predict the percentage of Electoral College votes received by the incumbent party should be nothing more than noting whether the incumbent party is expected to receive a majority of said votes or not. If the model predicts the incumbent party is to receive 10 percent of the Electoral College votes and in actuality it receives 45 percent then the model is still correct because in both cases the incumbent party did not receive the required threshold to be considered victorious.

I looked at the following eight economic indicators: 1) Personal Consumption Expenditures; 2) Unemployment Rate; 3) Personal Income and its Disposition; 4) GDP; 5) Consumer Price Index; 6) Inflation; 7) Dow Jones Industrial Average High, and 8) Disposable Personal Income, Per Capita (Constant 2000 Dollars). I calculated the percentage change in each of the economic indicators over various points in time during the current administration’s term to determine the greatest (if any) statistical significance for both percentage of popular vote and percentage of Electoral College votes. From this I created linear regression models designed to predict the outcomes of future elections. A breakdown for the results from percentage of popular vote is presented below:

Model Statistics for the Percentage of Popular Vote Received by the Incumbent Party

Indicator # of Qrts % ∆ R R2 S.E. of Est. a b S.E. t Sig. Avg. Abs. Err.
Unemp. 6 0.688 0.473 5.529 48.70 -0.360 0.110 -3.282 0.0066 3.91
GDP 7 0.737 0.543 5.147 38.72 1.750 0.463 3.779 0.0026 3.97
DJIA 8 0.538 0.290 6.419 42.96 0.293 0.132 2.213 0.0470 5.07
Disp. Inc. 10 0.785 0.616 4.719 39.60 1.904 0.434 4.389 0.0009 3.34

Given these models and current economic data the following predictions are made. Based on current unemployment trends over the last 18 months, Senator John McCain is predicted to receive 36.96% of the popular vote. GDP’s trending over the past 21 months leads to a prediction that McCain will receive 44.32% of the popular vote. According to the model above, the DJIA High’s trending over the last two years gives us a prediction that McCain will receive 43.85% of the popular vote. Finally, the model with the highest R2 predicts that McCain will receive 48.3% of the popular vote. The average of all of the models for percentage of popular vote is 43.36% for Senator John McCain.

To Summarize for Predicted Popular Vote for McCain:
Unemployment (18 mos.): 36.96%
GDP (21 mos.): 44.32%
DJIA High (24 mos.): 43.85%
Disposable Income (30 mos.): 48.3%
Average: 43.36%

It is interesting to point out that there is one important economic indicator showing Senator McCain very close to having 50 percent of the popular vote. The latest data for this indicator that was just released (10/30/08) shows a significant decline. However, keep in mind that this does not have a dramatic impact on this model because what the model is ‘interested in’ is the change in the economic indicator from one point in time to another. In this case, the percentage change 10 quarters prior to the last quarter for which data is available prior to the election, which happens to be the third quarter of the election year. All of these models use the latest data available. It is of course possible to use earlier data and try to predict further out, but given the propensity of many voters to make up their mind during the last month or two leading up to the election there is some question as to how useful this would be.

I don’t have the statistical information on the models for the Electoral College readily available, but do have the predictions so I will post them here.

To Summarize for Predicted Electoral College Vote for McCain:
Unemployment (15 mos.): 5.3%
GDP (21 mos.): 31.39%
DJIA High (24 mos.): 28.92%
CPI (24 mos.): 57.78%
Average: 30.24%

__________________
Marc
--------------------------------------------
For 5 little cents per day, you can view the forums w/o advertising, search the forums endlessly, feel good about keeping this place up and running, among the other benefits that are offered to contributing members.

Click HERE to subscribe now!!!

I originally began development on this type of model 8 years ago. I first used it to predict a presidential election in 2004. My boss asked me what the model said. I was quite busy but was determined to run the calculations to see what it would say. I used only two economic indicators and told him, "I don't believe the model. I think it is wrong and these types of models are based on a very small N. Don't believe what I am going to tell you." His response was simply, "Just tell me!". I gave him the average value of the two calculations, which were both within 1 percentage point of the other...and the actual result was less than .1 percent from the predicted result. Obviously this was not normal but yet very interesting. You can see from above that average absolute errors of the four models in the current analysis are all greater than this by quite a bit.

Do I think McCain will get approximately 43.46% of the popular vote if I use the average approach? No, I do not. Probably closer to 45-47 percent. However, I also thought Bush was going to get less. Also, interestingly McCain has consistently been polling around 43% nationally.

I know for a fact the Electoral College prediction will be way off. The margin of error on these models is quite high. Three of the four shows he will lose, while one does actually show he will win. The average of course shows he will lose this election.

All of this being said, the overall analysis is unidimensional...it is about economics. I had an idea after the 2004 election about writing an article titled, "It's Still the Economy, Stupid!" I didn't...but my model at least showed that voters in part were voting with the way the economy was trending. Of course looking at the R2 values, we know there are other issues...national security and what not. My models intentionally do not look at these things...the strength and weakness of them.

__________________
Marc
--------------------------------------------
For 5 little cents per day, you can view the forums w/o advertising, search the forums endlessly, feel good about keeping this place up and running, among the other benefits that are offered to contributing members.

Click HERE to subscribe now!!!

Interesting project. What motivated you to develop the model?

It looks as if your model was pretty good........

The final electoral tally will probably be (assuming MO for McCain and NC for Obama):

Obama = 364
McCain = 174

Your model predicted (using the 30.24% of 538 calculation)
McCain 162, Obama 376


So the accuracy here is incredible. Nice job. I don't follow the whole statistical analysis, but I can say that coming up with such an accurate prediction is impressive.

__________________

The world could use more Stan Ovshinsky's

yea and the pp. vote was very very close too, well done.

__________________

"The captain has turned off the `No Dubbing' sign. You are free to speak any language you choose."

Thanks.

As for my motivation...it all began during my graduate studies in political science. We read about other models that were developed and many of them seemed overly complex. I wanted something simple, but also unique. I believed that people rewarded the incumbent party not only for how good the economy was currently...or has been...but if it was getting better or worse. Thus, my use of percentage change in the economic indicators. The result has been quite good. While my four models here for popular vote averaged together showed McCain getting 43.36...I also thought there would be some "Bradley Effect" at play...hence my adding of a few points to this model and my final prediction for McCain of 45-47 percent given in my 2nd post. I think he got 46.3 percent or something.

As for the Electoral College prediction...I am pleasantly surprised that it worked out to be so close to my average of the four models. And the average of my four models actually predicts McCain will get 163 compared to 375 for Obama. The individual models themselves have high errors and you can see the large differences between the models in what the predict. The current models available in the PS: Political Science and Politics only predict the popular votes of either Obama or McCain with one exception. Klarner predicted Obama would receive 346 Electoral College votes...also not far off. Go here to see all of the latest models.

__________________
Marc
--------------------------------------------
For 5 little cents per day, you can view the forums w/o advertising, search the forums endlessly, feel good about keeping this place up and running, among the other benefits that are offered to contributing members.

Click HERE to subscribe now!!!

I'm impressed with that. Well done.