thesis - us voter knowledge

45
VOTER INFORMATION IN THE UNITED STATES PRESIDENTIAL ELECTIONS A Senior Integration Project Submitted to the Economics Department of Covenant College in Partial Fulfilment of the Requirements for the Degree of Bachelor of Arts in Economics by Kevin Lambert ______________________________________ Dr. Lance Wescher, PhD| Chair Dept. of Economics and Community Development Covenant College Lookout Mountain, GA Spring, 2016

Upload: kevin-lambert

Post on 22-Jan-2018

107 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: THESIS - US VOTER KNOWLEDGE

VOTER INFORMATION IN THE UNITED STATES PRESIDENTIAL ELECTIONS

A Senior Integration Project

Submitted to the Economics Department

of Covenant College

in Partial Fulfilment of the Requirements

for the Degree of

Bachelor of Arts in Economics

by

Kevin Lambert

______________________________________

Dr. Lance Wescher, PhD| Chair – Dept. of Economics and Community Development

Covenant College

Lookout Mountain, GA

Spring, 2016

Page 2: THESIS - US VOTER KNOWLEDGE

i

© Copyright 2016

Kevin Lambert

Page 3: THESIS - US VOTER KNOWLEDGE

ii

ACKNOWLEDGEMENTS

I would like to first thank my parents for their overwhelming support throughout my time

at Covenant as well as the many years they have selflessly provided for me by seeking Christ

first and making His glory preeminent in their parenting. Thanks to my beautiful fiancé Kendi

who has supported me through this whole process, giving me the drive to produce my best work

and finish to the end. Special thanks to Dr. Lance Wescher who has very patiently worked with

me during my time at Covenant, sharing his expertise and passion for economics through an

applicable biblical framework. I would also like to thank Anna Rannou for her constant support

and direction throughout the SIP process. Without her help and encouragement, I would not have

been able to complete a paper of this caliber. Lastly, I would like to thank my other professors

and classmates who have all played a part in giving me an enjoyable and fruitful college

experience.

Page 4: THESIS - US VOTER KNOWLEDGE

iii

VOTER INFORMATION IN THE UNITED STATES PRESIDENTIAL ELECTIONS

Abstract

by

Kevin Lambert

Voter information is an increasingly relevant topic in the United States today. Significant

academic research has been performed assessing the general levels of political knowledge and

the specific factors which aid in explaining the prevalence of political knowledge in individuals.

This paper engages the political knowledge discussion in order to review and further examine the

factors which are thought to influence differences in levels of voter information during the

electoral period. My central argument hinges on three primary determinants of voter information:

duration of US presidential election process, the level of partisanship within the electorate, and

media presence. By examining the more recent election cycles of 1992 through 2008, I will

construct a unique explanation of voter information by addressing the following questions: What

factors explain the differences in the levels of voter information and have these correlates

changed over time? By implementing multiple testing procedures, results seem to confirm past

research attributing education as the primary political knowledge indicator. Increased

partisanship, longer election duration, and the presence of written media sources also were found

to be positively related with political knowledge.

Page 5: THESIS - US VOTER KNOWLEDGE

iv

TABLE OF CONTENTS

I. Introduction……………………………………………………………………………………..1

II. Literiture Review……………………………………………………………………………….3

III. Analysis and Theoretical Framework

A. Data ................................................................................................................................ 6 B. Dependent Variable ........................................................................................................ 6 C. Independent Variable ..................................................................................................... 9

D. Empirical Methods…………………………………………………………………. . 13

IV. Results and Conclusions

A. Results………………………………………………………………………………..16 B. Conclusion .................................................................................................................. 21 C. Research Limitations and Application for Further Study ........................................... 22

V. Reformed Christian Response.................................................................................................. 24

VI. Tables and Graphs

Table 1: Variable description ........................................................................................................ 28 Table 2: Summary Statistics ......................................................................................................... 29 Table 3: Calculated Election Duration.......................................................................................... 29

Table 4: Calculated Incumbent Re-competing statistic ................................................................ 29 Table 5: Average Annual Unemployment Rate………………………………………………….30

Table 6: Knowledge Variance Across Election Years…………………………………………...30 Graph 1: Knowledge Levels Across Election Years……………………………………………..31 Graph 2: Agregated Knowledge Levels Across Election Years…………………………………31

Table 7: Ordinary Least Squared Regression……………………………………………………32 Table 8: Ordinal Logistic Regression……………………………………………………………33

Table 9: Odds Ratio Test………………………………………………………………………...34

VII. Additional Notes.................................................................................................................... 35

VIII. Bibliography………………………………………………………………………………..37

Page 6: THESIS - US VOTER KNOWLEDGE

1

I. Introduction:

Political knowledge and engagement of the American electorate has been a vital concern

since the conception of this country. In a country whose leadership is chosen from a democratic

system, one could expect a potential flaw to be an uninformed populace. Due to the relatively

inadequate weight of a single vote, it is not a surprise that many citizens lack the desire to be

involved in the political process at all.1 Since there are no checks on the level of political

knowledge one possesses when going to the polls, uninformed voter decisions are inevitable.

This issue was recognized during the initial steps of our country’s establishment. One of our

founding fathers, James Madison, said in a letter to W.T. Barry in 1822, “A popular government

without popular information, or the means of acquiring it, is but a Prologue to a Farce or a

Tragedy; or perhaps both. Knowledge will forever govern ignorance. And a people who mean to

be their own Governors must arm themselves with the Power that knowledge gives.”2 Our

founders expressly stated the importance of an informed populace and the inherent duty that we

possess as American citizens. John Adams wrote to the Boston Gazette on August 9th, 1863

reinstating the need for the voting populace to be sufficiently informed in order to understand the

policies of our governing authority:

“We electors have an important constitutional power placed in our hands; we have a

check upon two branches of the legislature . . . the power I mean of electing at stated periods [each] branch. . . . It becomes necessary to every [citizen] then, to be in some degree a statesman, and to examine and judge for himself of the tendency of political

principles and measures. Let us examine, then, with a sober, a manly . . . and a Christian spirit; let us neglect all party [loyalty] and advert to facts; let us believe no man to be

infallible or impeccable in government any more than in religion; take no man’s word against evidence, nor implicitly adopt the sentiments of others who may be deceived themselves, or may be interested in deceiving us.”3

1 (Barzel & Silberberg, 1973) 2 (Madison, 1822): The Writing of James Madison, vol. 9. 3 (Adams, 1763): Papers of John Adams, Volume I.

Page 7: THESIS - US VOTER KNOWLEDGE

2

If a democratic society can function sufficiently without an informed populace, than the

issue of voter knowledge would be irrelevant. Extensive research has been performed however,

testing the hypothesis that the American electorate is not informed, and that this problem has a

negative effect on the election process. While there have been theories suggesting the possibility

of an uninformed populace still having their ideas conveyed through a large sampled population,

Larry Bartels (1996) seemed to contradict this intuition through his extensive research.4 By

defining and outlining the various levels to which the American populace are uninformed, he

goes on to empirically test how this lack of information affects the voting process. His research

shows us that political ignorance has systematic and significant political consequences, as

democrats do nearly 2% better and incumbents nearly 5% better than they would have if the

electorate had been fully informed. Adding to the argument are researchers Delli and Keeter

(1997) who conclude in their book on why political knowledge is important, “…the more

knowledgeable are more likely to participate in politics, more likely to have meaningful, stable

attitudes on issues, better able to link their interests with their attitudes, more likely to choose

candidates who are consistent with their own attitudes, and more likely to support democratic

norms.”5 Knowledge does matter, and since the lack of an informed populace will directly affect

the policies that end up shaping the course of our nation, research such as this can help us pin

point the areas in which we can best inform the American people, strengthening our democratic

system.

4 (Bartels, 1996); Page 195 reflects upon past studies on political knowledge. 5 (Delli Carpini & Keeter, What Americans Know About Politics and Why It Matters, 1997): 272.

Page 8: THESIS - US VOTER KNOWLEDGE

3

II. Literature Review:

As an integral part of our American society, the election process serves not only to decide

who the next presidential candidate will be, but also to provide the American public with

information that can allow them to connect desired ideologies with potential candidates. With

this in mind, economists and political scientists continue to dedicate immense research toward

analyzing the changing variation in voter information among the electorate. These scholars

employ a range of methodologies to explore the factors which either enhance or detract from an

individual’s development of a constructive, politically- informed toolbox.

The majority of the literature on voter information and political knowledge in the US falls

into two major categories: (1) Analysis of the specific factors which diminish or enhance

political knowledge OR (2) Demographic research studying the variation of knowledge among

people groups.6 In comparison, tracking the general levels of political knowledge over time has

received relatively small amounts of scholastic attention. The studies referenced throughout this

paper, analyze a specific sample at a particular instance in time. While a couple studies do draw

from panel data, the time frame analyzed is most often no more than 2 years in length. Stephen

Bennet from the University of Cincinnati addresses this area through his research. Bennet (1989)

uses the National Opinion Research Center surveys as well as the Center for Political Studies

data from 1960 to 1986 to show that despite rising levels of public education, general levels of

political knowledge have remained relatively stagnant over time. Bennet identified two primary

causal mechanisms: lessened political interest among the electorate and decreased reliance on

newspapers within the US.

6 See cited evidence presented throughout my variable selection and explanation.

Page 9: THESIS - US VOTER KNOWLEDGE

4

Due to the sheer size of the United States, the level of political knowledge varies greatly

throughout all demographic sectors of the population. Depending on who you ask, there are

mixed opinions on the level of political knowledge individuals possess. Furthering his past

research, Bennet (2003) uses the National Election Studies of 1988 and 1992 to find that,

“Although most people know something about politics, the typical citizen is poorly informed,

and only a small group is very knowledgeable about politics.”7 As is seen throughout political

knowledge scholarship, general levels of political knowledge are arguably insufficient and the

number of the population that is highly informed is very small. Another author, Larry Bartels

(1996) begins his paper by stating America’s political ignorance is one of the best documented

features of contemporary politics. However, he goes on to cite several contributing authors who,

rather than focusing on quantitative studies on the population’s levels of political knowledge,

have written on the theoretical pitfalls of the democratic voting system.8 Ahn et al (2010) makes

reference to Bartels’ statement and noticed that many past studies have simply assumed general

public ignorance rather than continuing to test whether or not this observation has changed over

time. While a democratic voting system may clearly lead to inevitable problems, it is helpful for

researchers to point out these problems by supplementing theory with empirical supporting

evidence.

One of the most prominent pieces of literature addressing this topic is a book written by

Delli and Keeter (1996) called, What Americans know about Politics and Why it Matters.9 By

drawing from several data sources including the NES, the Roper Center for Public Opinion

7 Bennet (2003): 307. 8 Works include: The American Commonwealth (James Bryce, 1893), Democracy and its critics

(Robert Dahl, 1989), Public Opinion (Walter Lippmann, 1922), and Capitalism, Socialism, and Democracy (Joseph Schumpeter, 1950). 9 These authors are repeatedly referenced throughout many subsequent research studies in the political knowledge arena.

Page 10: THESIS - US VOTER KNOWLEDGE

5

Research, and a series of surveys that the authors conducted themselves, these gentlemen greatly

strengthen the literature on political knowledge. To name a few conclusions their work produced,

Delli and Keeter find that knowledge levels are extremely diversified throughout all social,

economic, and geographic spectra.10 They also determine education levels to be most indicative

of higher political knowledge. While the knowledge range is diverse among the population, they

discover low levels of knowledge to have persisted over time despite increasing education and

availability of political information. Expanding on this finding, these researchers conclude by

realizing that overall levels of knowledge have hardly moved throughout the past century.

This survey of literature reveals several important considerations including the

assumption of persistently low political knowledge among the electorate and the weak aggregate

impact of education on voter information. These works have, however, tended to analyze

particular factors and their relevance at various cross-sections of time rather than how factors

have changed over the course of sequential election years. Using data on past elections from the

American National Election Studies (ANES), I will show the impact that various correlates have

on political knowledge levels. The election period I have chosen to examine, 1992 to 2008, is a

very rich political atmosphere characterized by evolving political parties, tight presidential

elections, and increased media coverage. Because these elections are relatively recent, this paper

aims to foster an understanding of the variables that are most relevant in an explanation of

political knowledge within a modern context.

10 This was accomplished through isolation techniques focusing on several demographic characteristics.

Page 11: THESIS - US VOTER KNOWLEDGE

6

III. Analysis and Theoretical Framework:

A. Data

The primary data set used in my analysis is the ANES. The survey was administered by

the University of Michigan’s Survey Research Center which began its observations in 1948. The

data set was constructed as a scientific study in order to better understand public opinion and

political behavior with focus geared toward voter turnout and vote choice.11

The majority of the surveys were completed as face to face interviews, including both pre

and post-election questions, with the remaining interviews completed over the phone. In

addition, each member in the specified target population has a nonzero probability of being

selected arrived upon by probability samples. While the data set has both panel and cross

sectional attributes, the time series data consists of multiple cross-sectional surveys rather than a

single time series cohort. In order to construct a non-biased sample, the ANES used a sample

selection system known as “complex sampling” which employs methods such as oversampling,

stratified cluster sampling, and within-household sampling. In essence, this system enables

researchers to gather as wide and diverse a sample as possible in order to more accurately

measure traits of the entire US population.

B. Dependent Variable

There have been many attempts to operationalize the relatively intangible concept of

political knowledge. The primary difficulty in assessing such a variable stems from the fact that

political awareness is fundamentally multi-faceted, including aspects such as institutions and

processes opposed to political figures and current issues.12 Perhaps the most valid approximation

11 (Debell, 2010) 12 (Delli Carpini & Keeter, An Analysis of Information Items on the 1990 and 1991 NES Surveys, 1992): 1188.

Page 12: THESIS - US VOTER KNOWLEDGE

7

of such a concept is through the creation of an index of indicators that attempt to proxy a more

comprehensive measure of political knowledge. A prominent example is found in a paper

published by Carpini and Keeter (1993) in which they developed a 5 item index with questions

including the party control of the house, veto override percent, party ideological location, judicial

review, and vice president identification. The authors of this study employ item Response

Theory (IRT) in order to test the relevance of political knowledge questions. “If appropriate

assumptions are met, techniques utilizing item response theory can yield estimates of both the

discriminating power and the difficulty of a test item-independent of the particular sample on

which they are tested.”13 An important contribution produced by these researchers has been

specifying the need for an index to be a case by case measure, with respect to the data set being

used. While their results are based off a National Election Study pilot group, the index that they

suggest may not be an optimal index for a differing data set.

Alternatives to the indexing method have also been suggested, as many political surveys

include an interviewer assessed rating of political knowledge. The primary issue in using an

interviewer method is the inherent problem of heterogeneity, as discussed by Levendusky and

Jackman (2003). This refers to the lack of objectivity that exists across interviewers. In addition,

simple factors such as the mood of an interviewer can also convey bias to the interviewer rating

depending on the day the interview was performed. Levendusky and Jackman conclude that

while the interviewer method is not completely irrelevant, it does appear to be an overly

simplistic measure with a high likelihood for measurement error. While there are potential

problems with using such a method, it has also been described as the single most accurate

13 (Delli Carpini & Keeter, Measuring Political Knowledge: Putting First Things First, 1993)

Page 13: THESIS - US VOTER KNOWLEDGE

8

measure of political knowledge.14 Sekhon (2004) makes reference to 5 commonly used

techniques for observing political knowledge in respect to their relative effectiveness. Drawing

from past research of Zaller (1986) and Bartels (1996), she points out that Zaller used a 27 item

index, which had an estimated reliability of .89, and Bartels’ more reasonable 7 item index, with

an estimated reliability of .4 to .6.

In comparison, Sekhon mentions the single interviewer measure to have a sole reliability

of .78.15 Established American Political Scientist Larry Bartels discusses the interviewer rating in

his detailed work on political knowledge throughout the 1972 to 1992 presidential elections:

“Even rather elaborate information scales based on these sorts of items turn out to be only

slightly more reliable than the interviewer ratings, however; scales based based on as many as 15 separate “test” items have estimated reliabilities between .80 and .85, as

compared with about .78 for the interviewer rating. Interviewer ratings also turn out to be no less (and sometimes more) strongly related than factual information scales are to relevant criterion variables such as political interest, education, registration, and turnout.

Given the added difficulty of making comparisons form election year to another using scales based on rather different sets of available information items of variable quality, the

simpler interviewer ratings seem preferable for me purposes here.”16

This evidence suggests that while not perfect, the simplified interviewer attributed metric seems

to be sufficiently accurate for a basic study of political knowledge. Given the steep learning

curve in creating a comprehensive “knowledge index” as well as the inherent difficulty in

understanding the produced results from using such a method, the interviewer ascribed rating

becomes more and more reasonable. Additionally, the interviewer method will detour the

problem of question guessing by the interviewee which produces bias that is extremely difficult

to navigate as Carpini and Keeter (1997) discuss briefly in their book.

14 (Zaller, Proposal for the Measurement of Political Information, 1985); (Zaller, Analysis of Information Items in the 1985 Pilot Study, 1986) 15 (Sekhon, 2004): 7. 16 (Bartels, 1996): 203.

Page 14: THESIS - US VOTER KNOWLEDGE

9

The ANES offers an interviewer assessed rating which evaluates the level of political

information the respondent has before and after the interview. I will be using the post-interview

observation as a proxy for political knowledge for the purposes of this study.

C. Independent Variables

The independent variables chosen for this study have all been shown to contribute to the

explanation of voter knowledge. All correlates were initially chosen from a theoretical standpoint

and then confirmed on the basis of relevance from past field research. By using the ANES and

other supplemental variables, I will be able to test the presence that these measures have in

explaining knowledge in these samples over time. My independent variable list includes:

partisanship of the electorate, duration of the election process, media presence, incumbent re-

competing, gender, age, education, the unemployment rate, and whether or not an individual

voted in the given election year.

The first theorized variable I will be testing is the partisanship of the electorate. This

variable is drawn from the ANES survey questions asking respondents to subjectively rate from

one to seven their interpretation of partisanship.17 Due to subjectivity, natural problems may

persist such as the perceived sense of democrat-ness rather than the actual level which would be

extremely difficult to observe. Nonetheless, this variable is sufficient for my purposes as a broad

measure for partisanship. In a cross-national comparative study, Gronlund and Milner (2006)

examine determinates for political knowledge and find that party identification increases the

likelihood of a person having voter information, suggesting a positive correlation. In a more

recent study on partisan bias in the 2008 election however, Jesse (2009) finds that people with

17 A rating of ‘1’ refers to one thinking of themselves as a strong democrat while a ‘7’ is a strong republican. In my actual analysis, I plan to reorient these data by assigning answers 1&7 a measure of ‘3’, answers 2&6 a measure of ‘2’, and an answer of 4 with a measure of ‘1’. This will allow me to order all respondents from low to high levels of partisanship.

Page 15: THESIS - US VOTER KNOWLEDGE

10

lower levels of political knowledge will gravitate toward the views of their identified party rather

than assessing the actual policy views of candidates, suggesting a negative inferential

connection.

The second variable is the duration of the election process. In the past decade, the entire

election process has continued to elongate in a comprehensive sense and has become nearly

inescapable.18 It would make sense that as people are being exposed to longer election cycles,

and since these cycles immerse themselves into our daily lives, we would expect that greater

exposure to such would increase political learning. Election duration can be derived either by

measuring the number of primary debates leading up to an election, or by looking at the specific

time involved between the two. For the purpose of this study, I will use the time in days between

the first televised primary debate, and the actual day of the presidential election of the

corresponding election year to avoid having to compensate for elections in which the incumbent

is running.19 Evidence for this variable is found in a Danish study in which positive campaign

affects were found to be universal across genders, generations, and educational groups.20 Hirano

et al (2014) conclude their research by stating that significant political learning is transmitted

throughout the entire primary election process.

Thirdly, I will analyze the effect of media presence. As election duration has continued to

increase, the presence of the media following the entire process has also increased. Since mere

media presence may be serially correlated with duration, I will assess the type of media the

electorate is exposed to, essentially testing quality over quantity. Due to its prevalence in our

society, the effectiveness of news media has been thoroughly observed and in some cases

18 (Kondik & Skelley, 2015) 19 Election Day was chosen by congress in 1845 to be the Tuesday following the first Monday in November in years divisible by four. (Leip, 2008) 20 (Pedersen & Hansen, 2014)

Page 16: THESIS - US VOTER KNOWLEDGE

11

scrutinized. In a study by Chaffee and Frank (1996), the effects of media are assessed in order to

differentiate between strong and weak effects by media type. Drawing from panel data, Chaffee

and Frank found that televised media reaches the groups that lack political information, while

people who are actively seeking political information will tend to draw from print media such as

news magazines or newspapers. Guo and Moy (1998) contributed to the discussion by finding

television to be effective for developing political interest while newspapers provide a more

conducive learning medium. To broaden the potential mediums used, Drew and Weaver (2004)

and Kenski and Stroud (2006) both find that internet news and even simply having internet

access will increase one’s likelihood of possessing political knowledge. Both studies performed

draw their conclusions by retrospectively observing presidential elections. In order to take these

findings into account, I will use whether or not an individual used the newspaper as a resource

for gaining information as an independent variable.

While not my primary focus of this study, I have also included a number of control

variables which have been shown to contribute to political knowledge: incumbent re-competing,

gender, age, education, the unemployment rate, and whether or not an individual voted in the

given election year. Aside from unemployment rate, all other variables are quantified under the

ANES. Gronlund and Milner (2006), in an international comparative study, have shown that

people who vote are more likely to possess actual information. In a more recent study attempting

to measure how the economy affects levels of political information, Burden and Wichowsky

(2012) show that high unemployment rates increase the level of voter participation. Assuming

the validity of an individual’s vote as a legitimate predictor of knowledge, we can also see the

affect that the unemployment rate is likely to be a factor at play. My study corresponds the

election year with the unemployment rate from two years prior to allow individual adjustments to

Page 17: THESIS - US VOTER KNOWLEDGE

12

take place for the coming election. There has been great discussion on whether or not gender

plays a part in explaining knowledge. Mondak and Anderson (2004) demonstrate in their

analysis, by using the 1998 NES Pilot Study, that approximately 50% of the gender gap is

illusory; this reflects response patterns that work to the collective advantage of male respondents,

therefore insinuating being a male may have legitimacy as a knowledge indicator.

Many people assert that lack of education is strongly related to voter ignorance. Krause

(1997) by analyzing aggregate economic expectations found that as education declines, there is

an increase in retrospective voting criteria. Since retrospective voting is based upon policy

outcomes rather than policy implications, it seems likely that Krause’s research would lead us to

see a relationship between education and political knowledge. As previously mentioned,

Gronlund and Milner’s cross cultural study was performed to find valid knowledge correlates.

Their results were as follows, “Education remains the most powerful predictor of knowledge in

the MLS regression, followed by age. Overall, older, better educated, employed males are the

most likely to be politically knowledgeable.”21 Using the information from their research, I will

include both education and age in my analysis.22

Statistics are drawn from the ANES from years 1992-2008. The ANES offers data in 2

year increments so I will only be observing the election years themselves for a total of 5 separate

samples.23 Since some of my variable are dummy variables whose values depend on the election

year itself, I will run a single regression of all the years combined. In cleaning the data, I had to

delete observations so that my data set would only include election years 1992-2008. I then had

to drop Know observations where no assessment was given and where the interviewer did not

21 (Gronlund & Milner, 2006): 393 22 See Table 1 for variable list and their corresponding descriptions. 23 For Additional data cleaning information, see Section VII Additional Notes

Page 18: THESIS - US VOTER KNOWLEDGE

13

know how to answer. Additionally, because the interviewer rating for knowledge was only

administered during face to face interviews, I will have to drop all telephoned observations. I

also dropped Age observations where no age was specified. I dropped Partis observations when

respondents refused to answer. Lastly, I dropped MedNews when the respondent did not know

how to answer.24 Variables DurPrim and Incum were manually constructed, while Rate was

found from the U.S. Bureau of Labor Statistics.2526

D. Empirical Methods

Given the ordinal nature of the knowledge variable, it is important to find an estimation

technique that will provide the most accurate interpretation of the causal predictors. While some

of the more complex methods will yield more substantial and even more accurate results, we

must realize that this will be at the expense of complexity and therefore interpretability. In order

to analyze the data sufficiently, I have analyzed a few potential options; ordinary least squared

regression (OLS), multinomial logistic regression (Mlog) and ordinal logistic regression (Olog).

Aside from simplicity, OLS allows us to minimize the sum of the squared residuals while

showing the change in my dependent variable associated with a one unit increase in the

independent variable. For example, if a person’s age goes up by 1 year, the coefficient will tell us

the expected increase that will occur in one’s level of political knowledge. Dimitrova et al (2011)

used OLS regression in order to assess the affects that digital media has on political knowledge

and participation by drawing from two representative panel surveys. Providing an easy means of

interpretability, this model also has its problems. A key consideration is that OLS assumes that

the dependent variable, in this case Know is continuous. As was discussed, we are rather dealing

24 See table for mean variable differences before and after I dropped variables. 25 See Tables 3-5 for variables DurPrim, Incum, and Rate. 26 For Additional data cleaning information, see Section VII Additional Notes

Page 19: THESIS - US VOTER KNOWLEDGE

14

with a variable that takes on a discrete value from 1 to 5. Running OLS with a categorical

dependent variable must be interpreted with caution as it produces bias, therefore no longer

acting as our best linear unbiased estimator. While the results produced by OLS will not be

irrelevant, it is necessary to understand the limitations this model presents and therefore

implement another model in order to verify the results that OLS produces. The OLS model will

take the following functional form:

Know = β1Partisi + β2MedNewsi + β3DurPrimi + β4Incumi + β5MFi + β6Agei + β7Edui +

β8Ratei + β9Votei + ϵi

In order to avoid some of the problems of our OLS model, we will use the ordered

logistic model which we can then compare to the output from OLS. An ordered logit model will

produce much better results as it examines Know at each distinct sequential level.27 After running

the regression, results will tell us the log odds of occurrence of a particular outcome. This model

is similar to a multinomial regression in that it views each observation as independent, allowing

us to view the effect that various levels of a categorical variable are likely to have independently

on various levels of political knowledge. For example, if Ologit produces a coefficient of .5 for

respondents who have received advanced degrees, or an education level of 7, this tells us that the

log odds of someone possessing higher levels of political knowledge are .5 times higher than

they are for people without advanced degrees, holding other variables constant. For this analysis,

I will be implementing the Ologit model, as it takes into account the meaningful sequential order

of my dependent variable, ranked 1-5. The ordered logistic model has the following functional

form:

27 Torres-Reyna (post 2008)

Page 20: THESIS - US VOTER KNOWLEDGE

15

ln[prob(𝐾𝑛𝑜𝑤)

1−prob(𝐾𝑛𝑜𝑤)] = β0 + β1Partisi + β2MedNewsi + β3DurPrimi + β4Incumi + β5MFi +

β6Agei + β7Edui + β8Ratei + β9Votei

Page 21: THESIS - US VOTER KNOWLEDGE

16

IV. Results and Conclusions

A. Results

Prior to running the regression models, it is helpful to examine potentials trends in the

data itself. While regression analysis serves to show us the effect that particular variables will

have on various levels of political knowledge, in order to see how aggregate knowledge levels

have changed over time we can compare the ratios of various knowledge levels for a given

election year. Using the calculated percentages in Table 6, we can compare knowledge levels

relative to the sample size for a given election year. When comparing ‘very low’ to ‘very high’

levels of political knowledge we can see that there is not a drastic change other than in 2004.

Drawing form our sample data, 2004 has a higher proportion of people who possess very high

levels of knowledge opposed to lower knowledge levels. Even when combining very high and

fairly high knowledge levels, 2004 shows a significantly higher proportion of high political

knowledge. Graphs 1 and 2 examine the knowledge levels over time. As can be seen from the

trend line, knowledge levels appear to be at a slight increase over these 5 election periods.

Interestingly, over all 5 election cycles, the aggregate levels of knowledge appear to be

consistently high. The sample drawn from the election year of 2000 had the highest proportion of

‘very low’ levels of political knowledge at 5.67% with corresponding ‘very high’ levels at a

significantly higher level of 13.74%. If one were to draw conclusions based off of this data

alone, it would seem as though the American populace is aggregately well-informed, therefore

contradicting the volumes of research previously discussed (Bennet 1989, 1993, Bartels 1996,

Delli & Keeter 1996 etc.). This finding is significant and should be kept in mind as we interpret

the results of our regression analysis. As a caveat however, these “high levels” can most likely be

explained by the nature of the dependent variable as an interviewer assessed rating.

Page 22: THESIS - US VOTER KNOWLEDGE

17

OLS regression coefficients are simple to interpret as they represent how much the

dependent variable will change if the independent changes by one unit. OLS results can be found

in Table 7. Upon first observation, we see that variables Age, MF, Edu, Partis, Vote, Mednews,

and Durprim are statistically significant at the 95% confidence interval. This tells us that the

chosen independent variables are explaining the changes in Know levels. Let us first examine

partisanship. When the value of partis moves from very low to low, we see an increase in

knowledge of .25. Interestingly, moving from low to high partisanship produces the smallest

increase, while still positive, than other movements. We also see that people with very high

partisanship possess the highest level of political knowledge with a value of .34 as we move

from high to very high knowledge levels. While not seeming to be a strong connection,

intuitively speaking, these results seem to support Gronlund and Milner’s research performed in

2008 thereby opposing Jesse’s in 2009. Despite a low correlate, results are statistically

significant.

The duration between the first primary debate and the actual presidential election has

continued to increase over time as is shown in Table 3. The increase in duration does not seem to

translate to increases in aggregate knowledge levels as a one day increase between debates and

election leads to a .00088 increase in knowledge. However, we must remember that this result is

statistically significant and the degree to which it is significant has no relation to coefficient

itself. In order to assess this result further, we must seek the ordinal regression model.

Media presence is an intuitively weaker variable as the time this study was performed

was a transitional period away from traditional media sources toward greater use of electronic

news. Given this information we still see a positive correlation in the data wherein people who

seek newspapers for information will tend to have knowledge levels .29 higher than those who

Page 23: THESIS - US VOTER KNOWLEDGE

18

do not. This statistically significant result supports Chaffee and Frank’s (1996) research

revealing the fact that people who are truly politically informed, are so because they made an

intentional effort to obtain political knowledge. MedNews tries to capture this notion as it is a

source of media requiring a deeper level of effort opposed to radio or television sources.

The presence of an incumbent voter seems to have little effect on knowledge levels, seen

through the lack of statistical significance. This does not come as a large surprise considering

there has not been a great deal of past research substantially supporting incumbency as a

correlate. The unemployment rate also fails to be statistically significant but suggests a negative

correlation with knowledge levels. This evidence goes against the theory purported by Burden

and Wichowsky’s 2012 study.

There has been conflicting evidence as to whether age has an effect on knowledge levels,

despite the study by Gronlund and Milner. OLS shows gender to be statistically significant with

a coefficient of .0055, which supports their research of greater age being a factor of higher

political knowledge. It should be noted that the variable is in fact positive, suggesting that

knowledge levels clearly do not fall with age. Gender is a much more contested variable as the

evidence can easily increase gender based tension that is increasingly pervasive in America

today. This model explains that by being a male, the knowledge level will be .28 higher than it

would be for a woman.

Finally, as we look at the coefficients for education levels we can see the continual

increases in knowledge levels as people attain more education. The greater degree of education

someone attains, the more likely they are to also possess political knowledge. This clear finding

is consistent with Gronlund and Milner’s (2006) cross cultural study which also showed

education to be the clear winner in explaining higher levels of political knowledge.

Page 24: THESIS - US VOTER KNOWLEDGE

19

Interpreting the Olog coefficients requires more attention to detail than the OLS model.28

Before we interpret the results it is important to understand the output that will be given. The

coefficients produced from running the Olog model explain how much the logit changes based

on the values of the independent variables.29 Mentioned previously, the logit is the log odds of a

particular observed event occurring. The observed ordinal variable used in this model, Know, is a

function of the variable Know* which is an unmeasured latent variable. This latent variable is

continuous and is based on a particular threshold which determines the categorical values of the

observed values, Know. Using this information, we are able to make sense of the ancillary

parameters, seen in Table 7, which define the threshold that is given by Know*. To aid in the

coefficients interpretability, we will consult the results of an ordinal logistic odds ratio test.

Ologit does not tell us the significance that a particular variable has on different levels of

knowledge, but simply tells us if the relationship exists and whether it is positive or negative.

With the odds ratio results, we will be able to compare the impact that an independent variable is

likely to have on different levels of knowledge.

The Ologit model results appear to confirm those found with OLS referring to variable

significance and polarity as seen in Table8. To give an example of Ologit interpretability, let’s

begin by examining education levels since they have been proven to have the greatest and most

consistent impact on political knowledge. As we move to higher levels of education the

coefficients increase in value. As we move from individuals who possess a high school diploma

to those with some college exposure, the logit increases by 1.02. That is to say, the log odds of

possessing high levels of political knowledge increase by 1.02 as people move from high school

28 In order to understand Ordered Logistic Regression, I used the following resources from Princeton, Notre Dame, and the University of Michigan: Torres-Reyna (post 2008), Williams (2015), and Norusis (2008) 29 Norusis (2008)

Page 25: THESIS - US VOTER KNOWLEDGE

20

diplomas to having some college experience. Because log odds of occurrence do not make

intuitive sense, we apply the odds ratio results found in Table 9.30 The results produced from the

odds ratio test affirm the findings found in both Ologit and OLS. There is a noticeable increase in

knowledge levels as people gain higher levels of education. From table 9, for people with

advanced degrees, the odds of someone having very high levels of political knowledge is 19.11

times higher when comparing to all lower levels of knowledge, holding all other variables

constant. If increased levels of education have a substantial positive effect on someone’s level of

political sophistication, this is the result we would expect to see. By examining the incremental

increases between correlates, we can also see that the jump from some college to an actual

degree is much larger than the difference between some high school and a high school diploma.

Let us now move to the three primary variables of interest. Levels of partisanship do not

give a clear positive correlation as the odds of very high knowledge are 2.16 times higher for

people with very high partisanship than those without. Interestingly, these results also tell us that

as we move from very low to low levels of partisanship the odds of someone possessing very

political knowledge increases 1.77 times while an increase from low to high levels of

partisanship increases the odds by only 1.54. This is not an expected result as intuition would

lead us to believe that, assuming a positive relationship, the increase in knowledge, as we move

to higher partisanship levels, should be linear. This discrepancy is most likely explained by the

fact that partisanship is a self-attributed measure and it is likely that respondents fail to

consistently assess their own partisanship across the board. Primary duration also proves

statistical significance, as a one day increase in the duration between primary and Election Day

30 We will consult the odds ratio results for the remainder of this paper. Keep in mind that the odds ratio test is applied to the ordered logistic regression, meaning that the odds ratio results are directly drawn from Ologit.

Page 26: THESIS - US VOTER KNOWLEDGE

21

increases the odds of someone having very high knowledge by 1.002. Lastly, the proportional

odds ratio for someone with a very high knowledge rating is 1.89 times higher for people who

consult written media as an information source. In summary, the results produced by the Ologit

model confirm OLS results and support past research findings on political knowledge indicators.

B. Conclusion

My goal for this study is to contribute to the ongoing research of political knowledge in

the United States. By drawing from the fields of both political science and economics, my

research intends to observe various factors of political knowledge to assess their usefulness in the

political knowledge debate. By revealing which factors seem to be the most relevant in a modern

context, the goal of this research is to contribute to the ongoing discussion of voter knowledge

which can help shape the country’s attempt for creating an informed populace. These results

support my initial hypothesis in Table 1 that the increased duration between primaries and the

actual Election Day as well as people seeking written media are correlated with higher levels of

political knowledge. Since the causal relationship for increased partisanship was not intuitively

obvious, it was helpful to find results that were in support of Gronlund and Milner (2006),

suggesting a positive relationship. Given the fact that these three variables have received

relatively little discussion in the political knowledge debate, it is important to discover that all

three of these variables play a part in potentially predicting higher knowledge levels. Further

research is necessary to test the pervasiveness and degree of accuracy these correlates have on

the current electorate. This research reaffirms the presence of some attributes that exists among

voters and is designed to open the door to further study.

This project was designed to contribute to the greater body of political studies by

implementing an economic mindset through econometric analysis. In a season of political

Page 27: THESIS - US VOTER KNOWLEDGE

22

bombardment, research such as this is geared toward an efficient dissemination of information in

attempts to create a more just and well-functioning democratic system. By recognizing the

factors that contribute to a more well-informed populace, the spread and retention of viable

political knowledge should continue to increase throughout the country.

C. Research Limitations and Applications for Further Study

Given the limited scope of this research project, there are many inherent limitations that

exist with the data set, the empirical model, and the measure of knowledge itself. One of the

main difficulties in econometric research is choosing the correct data set to ensure that the

research will produce the most applicable results. In this case, the ANES provided a close to

ideal data set given the obvious limitations of using a stratified sample. Sampling methods have

remained relatively constant over time and the ANES takes special care in ensuring a well-

diversified representation of the American population. While this gives a more accurate

representation than other samples may, it is still an imperfect representation of the entire

American population and the results must be interpreted with this fact in mind. The results are

also dependent on the accuracy of this study’s data collection as the potential for human error is

likely to persist in any type of data gathering at this scale.

Political research generally tests one specific variable with a host of control variables,

while this research has tested three variables in particular. The specific method of my research is

unique, and therefore requires additional attention to verify the validity of my findings. While it

is very helpful to discover these statistical trends, the application of this research is limited as it

is drawn from a small sample from a specific point in time. If we were to apply these findings

across demographics and across time, we would need further a great deal of more evidence.

Page 28: THESIS - US VOTER KNOWLEDGE

23

Measuring political knowledge by using an interviewer assessed rating takes on a host of

assumptions. We are assuming a large degree of objectivity, we assume that the interviewer is

able to adequately assess individuals after speaking with them briefly, and we also assume that

the interviewer can differentiate between the various knowledge levels accurately and

consistently. Although imbued with problems, this method has proven sufficient for analyses at

this level. A method where knowledge is alternatively assessed through a host of proxy variables

and then fitted to an index has proven to yield slightly more accurate results. In this case, it came

down to a simple cost benefit judgment. Indexing knowledge levels requires an extensive amount

of work with an exponential learning curve. In the future, a further and more detailed study of

knowledge correlates would be aided by using a more accurate and detailed measure of

knowledge such as the index proposed by Bartels, Zaller, or Delli and Keeter.

Page 29: THESIS - US VOTER KNOWLEDGE

24

V. Reformed Christian Response

While the discussion of politics and religion has historically been taboo, I will argue that

the involvement in such discussions is an important aspect of living as Christians in a fallen

world. It has been interesting to see how the presence of Christian involvement in politics has

played a role in the past and present elections. While a moral compass and source of life for

some politicians, Christianity has also proven to be a political tool that many have used to swing

the evangelical vote toward their cause. This being said, many Christians have been turned off

completely to politics in general due to the great temptation politicians have to use Christianity

as a platform for their agenda. Paradoxically rather than evidence for deterrence, the level of

corruption should give Christians a greater reason to become involved in the political process.

This is first accomplished by increasing individual levels of political knowledge by performing

objective candidate research, and then using this knowledge to determine which candidate aligns

closest the values and principles held in scripture. Once a choice has been made, the most

important step is to take action and by placing a vote.

Christians have a direct responsibility to be actively involved in this nation’s politics. I

believe this principle is rooted in the parable of the talents in Mathew 25:14-30. God has given

us the gift to live in a free country where we face minimal persecution for worshiping God and

spreading His name. I naturally follows that we should have an interest in the people who are

directing the course of this nation, and therefore taking direct influence over our individual

liberties. As a young adult I am shocked with the level of political apathy that exists within my

generation, especially within the Christian community. In my lifetime alone I have seen the

government stray further and further from what the founding fathers designed for this country in

the founding documents. Whether it be issues from gay marriage, to abortion, to oppressive and

Page 30: THESIS - US VOTER KNOWLEDGE

25

self-interested foreign policy, numerous forms of US policy fail to reflect the biblical heritage

this nation was built on. How much longer are Christians going to sit back and watch? As

government continues to have a stronger hand on our daily lives, the time for action is passing us

by. A vital step needed to change this pathway is to increase our own levels of political

knowledge by further educating ourselves in a relevant and applicable context (Proverbs 14:18).

Many have twisted scriptures such as Mark 12:13-17 and Romans 12:1-2 to make the

case against having any involvement in politics claiming it is corrupt and inherently worldly. II

Corinthians 4:18 tells us to focus on what is unseen, the heavenly kingdom, rather than what is

seen, the earthly kingdom. Rather than seeing this as our ticket out of politics, this verse should

motivate us to become more involved in what is in front of us. If you were told by your boss that

you had to spend a month in a foreign country as part of a marketing strategy for the business to

advertise and ultimately negotiate business deal with them, would you not immediately start

preparing for that endeavor by educating yourself in the nation’s language, culture, geography,

and current events? If you want to make the best possible representation for your business and

perform successfully, this would be the natural step. Why is it that we do not feel this way about

learning about our own government? This leads me to another point of reason; without

understanding the foreign language and culture, the chances of being taken advantage of can

only increase. Whether or not this can be proven in the United States, people in power have

historically fed on mass public ignorance. Political knowledge as an argument against ignorance

is not only rational, but inherently biblical. While the majority of scripture warns of the

ignorance that persists by not heeding the word of God, passages such as Proverbs 4, 8:11, 9:6,

23:23, 24:5-6, James 1:5, Matthew 10:16 speak more generally of the importance of gaining

knowledge. Romans 1:28 speak against ignorance saying, “Furthermore, just as they did not

Page 31: THESIS - US VOTER KNOWLEDGE

26

think it worthwhile to retain the knowledge of God, so God gave them over to a depraved mind,

so that they do what ought not to be done.” As Christians we must be aware of what is going on

around us; if our government is acting unjustly, while we may not change it necessarily, we can

have an impact. If we are not first aware of the evils that are taking place, we will have no

chance to ever change the problems at hand.

If it is indeed biblical to be involved in the political process in this country, empirical

study such as this is can benefit the Christian community in several different ways. First, it will

provide us with political ammunition showing us the most receptive channels to send and

produce an educated populace. One must recognize the fact that it is not our duty as Christians to

spread propaganda, but to disseminate factual information that will allow the democratic election

process to function more efficiently. Second, we need real Christians in economic and political

academia. It is one thing to read and understand some of the great minds who have come into

power through these pathways, it is another to integrate ourselves amongst them. For those who

believe Christians have no business in ‘secular’ academic pursuit, I would direct them to any

scriptural example where Christ himself immersed himself among sinners with the sole purpose

of showing them the love of the Father. An example can be found in Jesus’ meal with tax

collectors in Mark 2:13-17.31 Third, in a broader sense, this research helps us understand the

world in which we live to a greater extent. There are numerous verses in the Bible that speak of

loving the world (people) yet hating the dark, such as John 2:15-17. If we are going to make an

impact for Christ on the people around us, understanding how they think is an important aspect

of apologetics.

31 PCA Pastor Kevin DeYoung wrote a great article called Jesus, friend of sinner: but how? in which he gives scriptural examples emphasizing the fact that He came to the earth FOR sinners; to save them from their sins.

Page 32: THESIS - US VOTER KNOWLEDGE

27

While politics in general may seem grueling and even distracting form our earthly and

heavenly focus, we must realize that the decisions we make today will continue to affect the

people around us for generations to come. Not to be misunderstood, we do live in a fallen world

and we cannot let ourselves become overwhelmed by constantly trying to fix everything around

us (Romans 3:23; Matthew 6:25-34; Philippians 4:6-7). Nevertheless, we will be held

accountable for our actions, and lack thereof. I strongly believe that by ignoring the opportunity

our government gives us to participate in elections and on the immediate world around us, we are

being selfish and irresponsible (Matthew 25:14-46). In summary, it is our duty as Christians to

be aware of our surroundings and to engage the broader body of believers through our power of

suffrage. In the world we live in, the political process is another way we can serve Christ, by

fighting for our constitutional and God-given rights as American citizens.

Page 33: THESIS - US VOTER KNOWLEDGE

28

VI. Tables and Graphs

Table 1: Variable description

Variable (hypothesized coefficient) Variable

Abbreviation

Definition

Political Knowledge (Dependent) Know An interviewer assessed rating on a scale of 1-5

Partisanship of electorate (+) Partis Respondents rated their personal level of affiliation with

either the republican or democrat party on a scale from

1-7, but in this study it has been collapsed to a 1-4 scale (see independent variable

explanation and footnote)

Media Presence (+) MedNews Whether respondents sought newspaper articles for

campaign information (0 = respondent did not use newspaper as an information

source)

Duration of Election Process (+) DurPrim The duration in days between the first primary debate and the

election day of the corresponding year

Incumbent re-competing (-) Incum Whether the incumbent is re-

competing (0 = incumbent is not re-competing)

Gender (+) MF Male or Female (0 = female)

Age (+) Age The age of the respondent in

years

Education (+) Edu Education level on an increasing 7 point scale (1

being 8 grades or less; 7 meaning advanced degrees)

Unemployment Rate (+) Rate Unemployment rate taken as an

annual average of two years prior to the actual election observed (Taken from Bureau

of Labor Statistics)

Voted in election (+) Vote Whether or not respondent voted in the presidential

Page 34: THESIS - US VOTER KNOWLEDGE

29

election (0 = respondent did not vote)

Table 2: Summary Statistics

Table 3: Calculated Election Duration32

Table 4: Calculated Incumbent Re-competing

32 I used a date calculator by comparing the first candidate primary debate of the election cycle with the actual date of that year’s election (Kondik & Skelley, 2015).

Variable Obs Mean Std. Dev. Min Max

know 4,649 3.405679 1.022855 1 5

age 4,649 47.83889 17.0884 17 93

mf 4,649 0.470424 0.499178 0 1

edu 4,649 4.523123 1.637144 1 7

partis 4,649 2.875887 0.984976 1 4

vote 4,649 0.804044 0.396978 0 1

mednews 4,649 0.640783 0.479823 0 1

incum 4,649 0.608303 0.488182 0 1

durprim 4,649 427.026 89.44792 325 559

rate 4,649 5.337294 0.659911 4.5 6.1

Election Year 1st debate (D) 1st debate (R) Election Day Duration in days

1992 - Clinton, H Bush, Perot 15-Dec-91 3-Nov 325

1996 - Clinton, Dole, Perot 11-Oct-95 5-Nov 392

2000 - Bush, Gore, Nader 27-Oct-99 22-Oct-99 7-Nov 383

2004 - Bush, Kerry 3-May-03 2-Nov 550

2008 - Obama, McCain 26-Apr-07 3-May-07 4-Nov 559

Election Year Incumbent Competing

1988 0

1992 1

1996 1

2000 0

2004 1

2008 0

Page 35: THESIS - US VOTER KNOWLEDGE

30

Table 5: Average Annual Unemployment Rate (US Bureau of Labor Statistics)

Table 6: Knowledge Variance across election years

Year Observations know-Low know-FL

know-H,FH know-H

1992 957 30 135 274 122

1996 1095 44 149 355 158

2000 1135 62 182 329 156

2004 776 16 84 258 177

2008 686 19 91 219 104

ratio-L,FL ratio-FH ratio-L ratio-H

17.24% 41.38% 3.13% 12.75%

17.63% 46.85% 4.02% 14.43%

21.50% 42.73% 5.46% 13.74%

12.89% 56.06% 2.06% 22.81%

16.03% 47.08% 2.77% 15.16%

Year Annual Average

1990 5.6

1994 6.1

1998 4.5

2002 5.8

2006 4.6

Page 36: THESIS - US VOTER KNOWLEDGE

31

Graph 1: Knowledge levels across election years

Notes: This graph was constructed using the data in Table 6. Using the sample data, this graph compares the percentage of ‘very high’ and ‘very low’ levels of political knowledge during over

the course of the election years.

Graph 2: Knowledge levels across election years

Notes: Also using data from Table 6, this graph combines ‘very high’ and ‘fairly high’ levels of

political knowledge to compare them to ‘very low’ and ‘fairly low’ percentages from the same election years.

0.00%

5.00%

10.00%

15.00%

20.00%

25.00%

1992 1996 2000 2004 2008

Re

lati

ve

Kn

ow

led

ge

Ra

tes

Election Years

Knowledge Variance

Low

High

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

1992 1996 2000 2004 2008

Re

lati

ve

Kn

ow

led

ge

Ra

tes

Election Years

Knowledge VarianceLowandFairlyLow

HighandFairlyHigh

Page 37: THESIS - US VOTER KNOWLEDGE

32

Table 7: OLS Regression

Dependent Variable: Knowledge level

Coefficient Standard Error

Age 0.0055874* 0.0007768

Gender (1 = male) 0.2815166* 0.0248683

Education: 2. 9-12 grades ('high school'), no diploma/equivalency 0.1939681 0.0862962

3. 12 grades, diploma or equivalency 0.4636092* 0.0761945 5. Some college, no degree; junior/community college 0.7993175* 0.0775453 6. BA level degrees 1.110194* 0.0796698 7. Advanced degrees incl. LLB 1.310726* 0.0827243

Partisanship Low 0.2545271* 0.0472396

High 0.1964986* 0.0470996 Very High 0.342673* 0.0473319

Vote (1 = vote) 0.5343227* 0.0341128

Media Use (1 = used media) 0.2896763* 0.0274414

Incumbent Re-competing (1 = re-competing) 0.156335 0.1041106

Primary duration 0.0008839* 0.000144

Unemployment Rate -0.0200609 0.0761799

Constant 1.023964* 0.3453342 Observations

4,649

R-Squared

0.3336

Mean Squared Error

0.83634

Notes: Regression is for all election years 1992-2008. Value adjacent to variable label is the regression coefficient, with the corresponding standard deviation beneath. * denotes statistical

significance at 5% level for a two-tailed t-test.

Page 38: THESIS - US VOTER KNOWLEDGE

33

Table 8: Ordinal Logistic Regression

Dependent Variable: Knowledge level

Coefficient

Standard Error

Age 0.0123852* 0.0017473

Gender (1 = male) 0.6319139* 0.0561003

Education: 2. 9-12 grades ('high school'), no diploma/equivalency 0.4761058 0.1980144

3. 12 grades, diploma or equivalency 1.019301* 0.175275 5. Some college, no degree; junior/community college 1.784319* 0.1795279 6. BA level degrees 2.485346* 0.185506 7. Advanced degrees incl. LLB 2.950428* 0.1933412

Partisanship Low 0.570537* 0.1073609

High 0.4294315* 0.1071373 Very High 0.7682538* 0.1079617

Vote (1 = vote) 1.20569* 0.0797642

Media Use (1 = used media) 0.6363105* 0.0619681

Incumbent Re-competing (1 = re-competing) 0.3660689 0.2318067

Primary duration 0.0020068* 0.0003246

Unemployment Rate -0.0440009 0.1695784

Threshold 1 1.258886 0.768936 Threshold 2 3.337927 0.7689303 Threshold 3 5.610565 0.7724573 Threshold 4 7.555189 0.775003

Observations

4,649

LR Chi-squared

1860.86

Notes: Regression is for all election years 1992-2008. Value adjacent to variable label is the regression coefficient, with the corresponding standard deviation beneath. * denotes statistical

significance at 5% level for a two-tailed t-test

Page 39: THESIS - US VOTER KNOWLEDGE

34

Table 9: Odds Ratio Test

Dependent Variable: Knowledge level

Odds Ratio Coefficient

Standard Error

Age 1.012462 0.0017691

Gender (1 = male) 1.881208 0.1055364

Education 2. 9-12 grades ('high school'), no diploma/equivalency 1.609793 0.3187622

3. 12 grades, diploma or equivalency 2.771256 0.4857319 5. Some college, no degree; junior/community college 5.955523 1.069183 6. BA level degrees 12.00527 2.22705 7. Advanced degrees incl. LLB 19.11413 3.695549

Partisanship Low 1.769217 0.1899447

High 1.536384 0.164604 Very High 2.155998 0.2327652

Vote (1 = vote) 3.339062 0.2663377

Media News (1 = used media) 1.889497 0.1170886

Incumbent Re-competing (1 = re-competing) 1.442055 0.3342779

Primary Duration 1.002009 0.0003253

Unemployment Rate 0.9569531 0.1622786

Threshold 1 1.258886 0.768936 Threshold 2 3.337927 0.7689303 Threshold 3 5.610565 0.7724573 Threshold 4 7.555189 0.775003

Observations

4,649

LR Chi-squared

1860.86

Probability > Chi-squared

0

Page 40: THESIS - US VOTER KNOWLEDGE

35

VII. Additional Notes

Edit Procedure for ANES

I first deleted all years not being observed. I then deleted all variables not being observed. (9,540 observations)

Know: I dropped observations ‘9’ where the interviewer rating was unavailable (901 dropped). I also dropped observations ‘0’ where there was no post interviewer rating (1,028 dropped). Both of these were described as ‘missing codes’ in the codebook.

Age: I dropped all missing codes coded as ‘00’. (47 dropped)

MF: Dropped missing codes coded as ‘0’. (0 dropped)

Edu: Dropped variables ‘8’ where respondent didn’t know what their education was (5

dropped). Also dropped all missing codes coded as ‘9’ (47 dropped).

Partis: Dropped all missing codes where responded either refused to answer or didn’t know coded as ‘0’ (68 dropped).

Vote: Dropped missing codes if respondent refused to answer coded as ‘0’ (82 dropped).

MedNews: Dropped all missing codes if the respondent didn’t know coded as ‘0’ (2,710

dropped)

After making the necessary changes, the data set is now down to 4,652 observations.

Edit Procedure for selected Variables

By using the replace function in STATA, I changed the values for Female (under variable

‘gen’) from ‘2’ to ‘0’ to be used in my analysis.

I applied a similar method to my remaining categorical and binary variables (see variable

description for more information).

Vote: An observation coded as ‘7’ refers to a non-vote. I used the replace function to

change all ‘7’s to ‘0’. The remaining observations were coded to signify a vote for either a republican, democrat, 3rd party candidate, or other minor candidates. I then condensed the remaining variables into ‘1’ in order to simplify the observations into vote and non-

vote. Using the replace function, I changed values 2, 3, and 4 to be equal to 1. This provided me with only 0’s and 1’s signifying a vote and non-vote.

Summary Statistics Comparison

Prior to dropping any variables, my summary statistics were as follows:

Variable Obs Mean Std. Dev. Min Max

know 14,127 2.90911 2.033579 0 9

age 14,127 46.78325 17.6348 0 99

mf 14,127 1.550081 0.4975031 1 2

edu 14,127 4.335386 1.73656 1 9

partis 14,127 3.658314 2.096593 0 7

vote 9,540 2.542138 2.449833 0 7

Page 41: THESIS - US VOTER KNOWLEDGE

36

mednews 9,540 1.039623 0.8743843 0 2

After deleting necessary variables, my summary statistics are as follows:

Variable Obs Mean Std. Dev. Min Max

know 4,649 2.595612 1.023787 1 5

age 4,649 47.83932 17.09632 17 93

mf 4,649 1.529791 0.4991654 1 2

edu 4,649 4.519682 1.63824 1 7

partis 4,649 2.876317 0.9839375 1 4

vote 4,649 2.632394 2.241582 1 7

mednews 4,649 1.640998 0.4797594 1 2

Speaking in terms of dropping observations, I will summarize the comparative data

summaries and their implications for my analysis:33

- Know – the mean has decreased telling us that more people possessing low levels of knowledge were dropped in proportion to those possessing high levels.

- Age – the age variable remained relatively constant - Mf – with a slight mean decrease, our proportion of males to females in our sample

has slightly increased (we dropped more females than males)

- Edu – the increase in mean Edu implies a slight increase in the proportion the education levels of our sample (we dropped more people with low levels of acquired

education) - Partis – the increase indicates we our sample on average has higher partisanship (we

dropped more observations of people with low partisanship)

- Vote – cannot be accurately interpreted because 1-3 signify a vote while a 7 indicates a non-vote

- Mednews – the increase means our sample now has more people who read news articles about the campaign (we dropped a greater number of people who did not read news articles)

33 Keep in mind that I have not yet changed the form of the variables, so interpretation will be different than it is my regression analysis. See Data Editing to understand how to interpret.

Page 42: THESIS - US VOTER KNOWLEDGE

37

Bibliography

Adams, J. (1763, August 29). Papers of John Adams, Volume I. The Boston Gazette. Retrieved

October 2015, from https://www.masshist.org/publications/apde2/view?id=ADMS-06-01-02-0045-0007

Ahn, T. K., Huckfeldt, R., Mayer, A., & Ryan, J. (2010). Politics, Expertise, and

interdependence within electorates. In J. Leighley, Oxford Handbook of American Elections and Political Behavior (pp. 278-281). Oxford: Oxford University Press.

Bartels, L. M. (1996, February). Uninformed Votes: Information Effects in Presidential

Elections. American Journal of Political Science, 40(1), 194-230. Retrieved October

2015, from http://www.jstor.org/stable/pdf/2111700.pdf?acceptTC=true

Barzel, Y., & Silberberg, E. (1973, Fall). Is the Act of Voting Rational? Public Choice, 51-58.

Retrieved October 2015, from http://www.jstor.org/stable/pdf/30022735.pdf?acceptTC=true

Bennet, S. E. (1989, October). Trends in American's Political Information, 1967-1987. American

Politics Research, 17(4), 422-435. Retrieved from http://apr.sagepub.com/content/17/4/422.short

Bennet, S. E. (2003). Is the public's ignorance of politics trivial? A Journal of Politics and

Society, 15(3-4), 307-337. Retrieved from http://www.tandfonline.com/doi/abs/10.1080/08913810308443585

Benoit, W. L., & Hansen, G. J. (2006). Presidential Debate Watching, Issue Knowledge,

Character Evaluation, and Vote Choice. Human Communication Research, University of

Missouri. Retrieved November 2015, from https://www.researchgate.net/publication/229598695_Presidential_Debate_Watching_Issue_Knowledge_Character_Evaluation_and_Vote_Choice

Brussino, S., Medrano, L., Sorribas, P., & Rabbia, H. (2011, May). Young adults' knowledge of politics: evaluating the role of socio-cognitive variables using structural equations.

Spanish Journal of Psycology, 183-194. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/21568176

Burden, B. C., & Wichowsky, A. (2012). Unemployment and Voter Turnout. Academic Study,

University of Wisconsin, Marquette University, Political Science. Retrieved November

2015, from http://users.polisci.wisc.edu/behavior/Papers/Burden%26Wichowsky2012.pdf

Chaffee, S., & Frank, S. (1996, July). How Americans Get Political Information: Print versus

Broadcast News. The Annals of the American Academy of Political and Social Science, 546(The Media and Politics), 48-58. Retrieved November 2015, from

http://www.jstor.org/stable/1048169?seq=1#page_scan_tab_contents

Page 43: THESIS - US VOTER KNOWLEDGE

38

Clarke, P., & Fredin, E. (1978). Newspapers, Television and Poltical Reasoning. Public Opinion Quarterly, 42(2), 143-160. Retrieved October 2015, from

http://poq.oxfordjournals.org/content/42/2/143.short

Debell, M. (2010). How to Analyze ANES Survey Data. ANES Technical Support Series,

University of Michegan, Standford University, Palo Alto, CA; Ann Arbor, Michegan.

Retrieved November 2015, from http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.178.1160&rep=rep1&type=pdf

Delli Carpini, M. X. (1999). In Search of the Information Citizen: What Americans Know About

Politics and Why it Matters. University of Pennsylvania. University of Pennsylvania

Scholarly Commons. Retrieved from http://repository.upenn.edu/cgi/viewcontent.cgi?article=1006&context=asc_papers

Delli Carpini, M. X., & Keeter, S. (1992). An Analysis of Information Items on the 1990 and

1991 NES Surveys. University of Pennsylvania. Annenberg School for Communication.

Retrieved 2015, from http://repository.upenn.edu/cgi/viewcontent.cgi?article=1016&context=asc_papers

Delli Carpini, M. X., & Keeter, S. (1993). Measuring Political Knowledge: Putting First Things

First. University of Pennsylvania. Annenberg School for Communication. Retrieved 2015, from http://repository.upenn.edu/cgi/viewcontent.cgi?article=1023&context=asc_papers

Delli Carpini, M. X., & Keeter, S. (1997). What Americans Know About Politics and Why It

Matters. Yale University Press.

Dimitrova, D. V., Shehata, A., Stromback, J., & Nord, L. W. (2011, November). The effects of

Digital media on political knowledge and participation in election campaigns: evidence from panel data. Communication Research, 41(1), 95-118. Retrieved from

https://www.researchgate.net/profile/Jesper_Stroembaeck/publication/223956376_The_Effects_of_Digital_Media_on_Political_Knowledge_and_Participation_in_Election_Camp

aigns_Evidence_From_Panel_Data/links/542cf7940cf29bbc126cf420.pdf

Drew, D., & Weaver, D. (2006, March). Voter Learning in the 2004 Presidential Election: Did

the Media Matter? Journalism & Mass Communication Quarterly, 83, 25-42. Retrieved November 2015, from http://jmq.sagepub.com/content/83/1/25.short

Econ, D. (1999, November). What is the single most important economic indicator for

policymakers? Retrieved March 24, 2014, from Federal Reserve Bank of San Francisco: http://www.frbsf.org/education/publications/doctor-econ/1999/november/economic-indicators-policy

Gronlund, K., & Milner, H. (2006, December). The Determinants of Political Knowledge in

Comparative Perspective. Scandinavian Political Studies, 29(4), 386-406. Retrieved

Page 44: THESIS - US VOTER KNOWLEDGE

39

November 2015, from http://onlinelibrary.wiley.com/doi/10.1111/j.1467-9477.2006.00157.x/abstract

Guo, Z., & Moy, P. (1998). Medium or Message? Predicting Dimensions of Political

Sophistication. Internation Opinion Research, 10(1), 25-50. Retrieved October 2015, from http://ijpor.oxfordjournals.org/content/10/1/25.short

Hirano, S., Lenz, G., Pinkovskiy, M., & Snyder, J. (2014, January). Voter Learning in State

Primary Elections. American Journal for Political Science, 59(1), 91-108. Retrieved from

http://onlinelibrary.wiley.com/doi/10.1111/ajps.12093/abstract

Iyengar , S., & Simon, A. F. (2000). New Perspectives and Evidence on Political

Communication and Campaign Effects. Annual Review of Psycology, 149-169. Retrieved

November 2015, from http://www.uky.edu/AS/PoliSci/Peffley/pdf/Iyengar%20%26%20Simon%202000%20Political%20Communication%20AnuRevPsych.pdf

Jessee, S. A. (2009, July 21). Partisan Bias, Political Information and Spatial Voting in the 2008

Presidential Election. The Journal of Politics, 72(02). Retrieved November 2015, from http://journals.cambridge.org/action/displayAbstract?fromPage=online&aid=7417980&fi

leId=S0022381609990764

Kenski, K., & Stroud, N. (2006). Connections between Internet use andpolitical efficacy,

knowledge, and participation. Journal of Broadcasting & Electronic Media, 50, 173-192. Retrieved November 2015, from

http://sspa.boisestate.edu/communication/files/2010/05/Kenski-and-Stroud-Connections-Between-Internet-Use.pdf

Kondik, K., & Skelley, G. (2015, July 30). Eight Decades of Debate: A brief history of

presidential primary clashes. (C. f. Poltics, Producer, & University of Virginia) Retrieved

November 2015, from Sabato's Crystal Ball: http://www.centerforpolitics.org/crystalball/articles/eight-decades-of-debate/

Krause, G. (1997). Voters, Information Heterogeneity, and the Dynamics of Aggregate

Economic Expectations. American Journal for Political Science, 1170-1200.

Leip, D. (2008). Presidential Election Dates. Retrieved November 2015, from US Election

Atlas: http://uselectionatlas.org/INFORMATION/INFORMATION/dates.php

Levendusky, M. S., & Jackman, S. D. (2003). Reconsidering the measurement of political

knowledge. Stanford University. Retrieved from

http://web.stanford.edu/class/polisci353/2004spring/reading/levendusky_final.pdf

Lipsitz, K. (2011). Competitive Elections and the American Voter. Pennsylvania, Philadelphia:

University of Pennsylvania Press. Retrieved November 2015

Page 45: THESIS - US VOTER KNOWLEDGE

40

Madison, J. (1822, August 4). The Writting of James Madision, vol. 9. Retrieved October 2015, from Online Library of Liberty: http://oll.libertyfund.org/titles/1940

Mondak , J. J. (1999). Reconsidering the Measurement of Political Knowledge. Tallahassee:

Florida State University. Retrieved from http://www.csus.edu/programassessment/Docs/theoreticalcontext/Assessing%20Political

%20knowledge.pdf

Mondak, J. J., & Anderson, M. R. (2004, March 31). The Knowledge Gap: A Reexamination of

Gender-Based Differences in Political Knowledge. The Journal of Politics, 66(2), 492-512. Retrieved November 2015, from http://onlinelibrary.wiley.com/doi/10.1111/j.1468-

2508.2004.00161.x/abstract?userIsAuthenticated=false&deniedAccessCustomisedMessage

Norusis, M. (2008). SPSS 16.0 Guide to Data Analysis. Saddle River, New Jersey: Prentice

Printing Press. Retrieved from http://www.norusis.com/pdf/ASPC_v13.pdf

Pedersen, R. T., & Hansen, K. M. (2014, April). Campaigns Matter: How Voters Become

Knowledgeable and Efficacious During Election Campaigns. Political Communications, 31, 303-324. Retrieved November 2015, from

http://kaspermhansen.eu/Work/Hansen_Pedersen_2014.pdf

Sekhon, J. S. (2004, July 26). The Varying Role of Voter Information across Democratic

Societies. Retrieved November 2015, from Research Gate: https://www.researchgate.net/profile/Walter_Mebane/publication/228934503_The_varyin

g_role_of_voter_information_across_democratic_societies/links/53cf3d1a0cf2fd75bc59b504.pdf

Torres-Reyna, O. (post 2008). Getting Started in Logit and Ordered Logit Regression. Retrieved

from Princeton University: https://www.princeton.edu/~otorres/Logit.pdf

White, J. (1997, August). Book Reviews. The Journal of Politics, 59(3), 925-928. Retrieved

from http://www.jstor.org/stable/2998645

Williams, R. (2015). Ordered Logit Models - Overview. University of Notre Dame. Retrieved

from https://www3.nd.edu/~rwilliam/stats3/L11.pdf

Zaller, J. R. (1985). Proposal for the Measurement of Political Information. Report to the NES

Board of Overseers, University of Michegan, Center for Political Studies.

Zaller, J. R. (1986). Analysis of Information Items in the 1985 Pilot Study. NES Pilot Study

Report, University of Michegan, Center for Political Science. Retrieved from http://www.electionstudies.org/resources/papers/documents/nes002261.pdf