The Effects of the State of Union Address of USA Presidents to Their Performance Ratings

Introduction

Text data is a major type of data in the modern society. According to Grimes (2011), over 80% of the data are unstructured, and most of them are text. Analysis of text data started in the 1980’s. Technology advancement in the recent decades has enabled the development of text mining techniques to analyze patterns in text data. Many software, both commercial (e.g., SAS/Enterprise Miner, IBM SPSS, StatSoft) and open sources (e.g., Rapid Miner, Apache UIMA, TESLA), are now available for researchers to analyze text information and further to combine the information retrieved from text data with quantitative data for advanced research Due to the broad range of applications of text mining technology, Feldman (2004) suggested the use of text analytics in place of text mining. Examples of some common applications include biomedical, linguistic, online text mining, marketing, sentiment, academic, security applications, and so on.

The USA Constitution mandates that the President shall deliver the annual State of the Union Address to the Congress. Today, the State of the Union Address is delivered during the month of January. The Congress Research Service Report for Congress (2006) of USA states

These annual addresses often have very significant impact to USA as well as to the world since they reflect the current situation of USA and set the direction and policies that the president recommends for the coming year.

An interesting question is “Does it matter what the President says in the State of the Union Address?’. In order to answer this question, we first conduct a text mining study to investigate if there are common patterns from these addresses, for example, ‘Are there commonality for the presidents from the same party?’; ‘Are there different patterns between Democrat and Republican presidents?’; and so on. Secondly, we integrate these text patterns with their presidential approval rating data to study if these text patterns have any effects on their performance ratings. Each year, there are a list of surveys regarding the president approval rates in a variety of issues, such as economic, social and health care, budget and debts, foreign affairs, and education issues, etc.. By integrating the patterns found from their state of the union addresses from previous year with the performance approval rating survey data, it allows us to study the effects of their addresses to their approval rating on different issues.

The annual State of the Union Addresses that we analyze in this study are from four USA presidents, George Bush (1989 – 1993), William J. Clinton (1993 – 2001), George W. Bush (2001 – 2009) and Barack Obama (2009 – 2011) for a total of twenty three State of the Union Addresses. The documents are downloaded from the American Presidency Project by Woolley and Peters (2011). The approval rating survey data are from the U.S. Officials' Job Approval Ratings (JARs) project funded in part by a National Science Foundation grant (Niemi, Beyle & Sigelman, 2010). The detailed results of the speech patterns of these 23 addresses have been reported in Lee and Crockett (2012). This article first reviews some of the results from the text pattern analysis and then presents the findings of the effects of these text patterns on the presidential approval ratings. The article is organized as follows. Section 2 briefly describes the speech data preparation and how we integrate the speech data and the approval rating survey data. Section 3 reviews some important patterns found from the text analysis given in Lee and Crockett (2012) that will be used in this effect study. Section 4 describes in detail the manipulation of the approval survey data. Section 5 discusses the methodology of analyzing the effects of the text patterns to the approval rating. Section 6 presents the results and discussions of the text effects.