If, as we seem to have discovered, the U.S. electorate prefers to elect the taller candidate, and this has been going on for at least the 120 years for which we have data, then it seems wholly reasonable that this information is fully known to candidates. This observation leads us to expect that there should be an upwards trend in candidate height very much as there has been an upward trend in the height of professional basketball players.

Above we have plotted all the candidates heights and a best fit regression line. This line is the result of a panel data analysis with individual effects and has a gradient of 0.0022 ± 0.0064 feet/year. The regression has an F statistic of 7.4 with 2 and 56 degrees of freedom and a p-Value of 0.001. Thus we can reject the hypothesis that there is no trend with over 99.9% confidence — the trend is clear to see in the figure.

I imagine the sporting career of an NBA player is brief. This data suggests that they should retire to national politics!

 

I wrote up the Presidential Probit Model work as a working paper that is available from the Social Science Research Network.  

Our presidential probit model, which describes success based on the appeal of tall candidates and a desire for change after two full terms, has an in-sample forecasting skill of 54%. To get a grip on the error on that measurement, I ran 10,000 bootstrap simulations of the process. The chart below suggests that the skill uncertainty is ±19% so this significance is at the 2.8σ level.

 

Previously on this blog I analysed the New York Times' presidential height and weight data. We built a probit model and found that the height difference and change variable (meaning the opposite of the outcome of the election two cycles ago), were significant. Neither weight difference or incumbency variable (meaning the outcome of the election one cycle ago) were significant. The full model is:

Prob(Republican Win) = Φ(constant+dHeight+change).

Here Φ(·) is the generalized linear model transfer function, which is the cumulative normal distribution for a probit model.

Taking the candidates heights from the internet we find (Barack Obama is 6'1.5" tall):

Newt Gingrich is 6' tall, Prob(Gingrich|Obama) = 16%

Mitt Romney is 6'1" tall, Prob(Romney|Obama) =  29%

Rick Santorum is 6'4" tall, Prob(Santorum|Obama) = 76%

If this model is a good predictor, then the Republicans should run Rick Santorum against Barack Obama for President of the USA.

You can find my RATS code to evaluate the probit model at Google Docs. Bear in mind that the purpose of this model is fun!

 

The Tweet class described in the prior post has the defect of blocking on interaction across the wan with an internet service, twitter, that does not guarantee that it can process your message within a negligible, or even with in a predictable, time. Frequently I code systems that have to guarantee a predictable and small response time and we can achieve that with twitter by dividing our task between two threads:

1. the first thread stores your tweet in a queue and returns immediately;
2. the second thread processes the queue, forwarding the tweets to twitter.

For this to work the average tweet rate cannot exceed the average total round trip time to process the tweets — otherwise the queue will back up. However, providing we do satistfy that constraint on average we can allow the response time to vary without impacting the thread doing the tweeting.

The TweetQueue class provides this multithreaded facility, which is incredibly easy to implement in Java.

To use the object we:

1. Instantiate the object
   StatTrader.TweetQueue twitter = new StatTrader.TweetQueue(tokenKey,tokenSecret);
2. Create a thread for the object
   Thread worker = new Thread(twitter);
3. Start the thread
   worker.start();
4. And use it
   twitter.Update("status");

That's it. I don't claim that this is original, but I find it useful so I thought I'd share.

 

I did an earlier experiment with twitter, which you can read about at the SSRN.

That system used a very basic command line http client. For another reason I recently wrote a twitter client in Java that implements an asynchronous queue manager so that the tweeting process does not block on the wan traffic and authenticates properly using OAUTH. It uses the twitter4j library and wraps it even more to make just issuing tweets elementary.

You can access the first part, the Tweet class, here. This is what you have to do:

1. Go to twitter and register an application. This generates a consumer key and secret.
   
   
2. Generate a token and secret for your account by calling:
   twitter4j.auth.AccessToken at = StatTrader.Tweet.Generate(consumerKey,consumerSecret);
   Serialize(at);
   
   
3. Instantiate a twitter object:
   twitter4j.auth.AccessToken at = Unserialize();
   StatTrader.Tweet twitter = new StatTrader.Tweet(at.getToken(),at.getTokenSecret());
   
   
4. Tweet:
   twitter.Update("Status");