HOT and COLD: Quantifying the Variation of Sentiment in Supreme Court Confirmation Hearings

HOT AND COLD: Quantifying the Variation of Sentiment in Supreme Court Confirmation Hearings

Student Author the context of other judicial systems, such as that of Norway. He has published articles in journals such Noah Alderton is currently a as American Politics Research, Political Behavior, student at Purdue University Scandinavian Political Studies, and Social Science pursuing a BS in computer Quarterly, and he is the author or co-author of six science and data science and a books on judicial politics. His most recent book BA in political science. During (co-authored with Gunnar Grendstad, William the past three summers, he has Shaffer, and Jørn Øyrehagen Sunde), Proactive and interned as a software engineer Powerful: Law Clerks and the Institutionalization and plans on working full- of the Norwegian Supreme Court, is the first time in software development upon graduation. comprehensive study of law clerks in a European Additionally, Alderton has a passion for education, supreme court. and he has worked as a lecturer and curriculum creator for CS 193. He has been a teaching assistant Mintao Nie received his PhD for multiple different courses, including CS 240 and in political science at Purdue POL 300. Later in life he hopes to achieve a master’s University in 2020. His research degree in computer science. interests lie at the intersection of human rights, judicial Mentors politics, and global governance. He specifically studies how Eric N. Waltenburg received his contestation and collaboration PhD in political science from The between divergent actors determine human rights Ohio State University in 1994 policies in both domestic and international settings. in American politics. He joined His research is published in Du Bois Review: the Purdue faculty in 1994. His Social Science Research on Race and Justice research focuses on explaining System Journal. In his most recent research, the political role of high Dr. Nie has explored the politics of human rights courts in democratic systems. in intergovernmental organizations. In particular, Dr. Waltenburg investigates these issues primarily he studied human rights organizations’ advocacy in the context of the U.S. Supreme Court, but he has campaigns surrounding the Universal Periodic recently begun to investigate the same questions in Review at the United Nations.

8 Journal of Purdue Undergraduate Research: Volume 10, Fall 2020 http://dx.doi.org/10.7771/2158-4052.1484 Hot and Cold 9 LITERATURE REVIEW LITERATURE INTRODUCTION In 2012, Lori Young and Stuart Soroka published Young In 2012, Lori Lexicodera paper detailing their newly developed The purpose behind cre- Sentiment Dictionary (LSD). ating the LSD was to have a sentiment dictionary that caters better to analyzing political communications. The LSD measures sentiment using a dictionary-based word count algorithm that counts the number of words ed category in the dictionary. that match a specifi The LSD was created by combining three expansive lexical resources: General Inquirer (GI), Regressive Thesaurus. Imagery Dictionary (RID), and Roget’s the ectiveness of the LSD, In order to measure the eff authors compared it directly with six other frequently used sentiment lexicons on their ability to measure articles Times positive and negative tone in New York environment, across four topics: crime, economy, The benchmark to airs. and international aff determine the accuracy of a sentiment dictionary is a study and Soroka organized Young human coding. In 2018, Brett Kavanaugh was appointed by Kavanaugh was In 2018, Brett retiring to replace Trump Donald President Court. on the Supreme Anthony Kennedy Justice be highly proved to appointment Kavanaugh’s once multiple women controversial, especially allegations against him (Hauser, made sexual assault Judiciary ing the Senate dur 2018). Questioning for Kavanaugh was notably Committee hearing day of the rst on the fi contentious. For instance, Durbin of Illinois was quoted hearings, Senator Dick are the nominee of President Donald saying, “You President who has shown us This is a Trump. John is contemptuous of the rule of consistently that he are that President who has decided you . . . It’s law. his man” (Collinson, 2018). for The Senate Judiciary Committee hearings opportunity Supreme Court nominees provide a keen in play to observe the numerous political forces to the during the appointment of a new justice not yet been there has Supreme Court. However, explain substantial research that attempts to demeanor during the the variation in senators’ greater understanding of A nomination hearings. be obtained by the attitudes during the hearings can the general leveraging sentiment analysis to quantify The goal of this statements. attitude of the senators’ explain the project was to create a model that could attributes variance in senator sentiment using various the hearing, and the related to the individual senator, Senate body as a whole. , 8–17. , 8–17. Keywords Supreme Court, Senate Judiciary Committee, rmationhearings, sentiment analysis confi Abstract Sincethe appointment Harlan John M. II of every Supreme Court nomineein has 1955, ed Senatein a Judiciary Committee testifi hearing. These hearings a fertile provide ground partisanship, senator ideologies, for and to forces be political full on display. little research has systematicallyHowever, rmationhearings Supreme for analyzed confi Court nominees. In this quantitative paper, sentiment analysis is used the on transcripts rmationhearings Supreme Courtof confi leveraging By and 2018. between 1969 sentiment analysis, the attitudes each of the senatorsof during the hearings can be measured. Investigating the correlative impact that variables institution, the at senator, and sentiment on have nomination creates levels a better understanding the factors of that attitude uence during a senator’s infl may ect the hearings. correlative eff A positive sentimenton was with found an increase in the percentage the the nominating vote of during received president his recent most state. home A inelection the senator’s correlationpositive was also when found the nominating and president a senator members thewere same of party. political cant a statistically signifi Additionally, correlationnegative was measured when the departing justice was a swing and voter thewhen hearing was aired television. on This research for points to avenues new using textual data to study partisanship and ideological polarization. and cold: Hot Alderton, (2020). N. Quantifying the variation sentiment of in rmationhearings. Supreme Court confi Undergraduate Purdue of Journal Research, 10 where three human coders would assign each article Rinker and Spinu investigated the occurrence of either a positive, negative, or neutral tone. Based valence shifters in several textual datasets. During upon the results from these coders, the researchers the 2012 presidential debate, they found that negators assigned a classification for each of the articles on occurred in 23% of sentences with polarized words a 5-point sentiment scale. They found that the LSD and that amplifiers occurred in 18% of polarized more closely aligned with the human coding of the sentences. They also investigated the occurrence of articles compared to the other dictionaries. This valence shifters in the speeches of President Donald indicates that the LSD is the preferred sentiment Trump, finding that of polarized sentences, 14% had dictionary for this purpose. amplifiers and 10% had adversative conjunctions.

Young and Soroka’s paper was useful in deciding George Watson and John A. Stookey’s 1995 book, which sentiment dictionary is most appropriate Shaping America: The Politics of Supreme Court for this project. Several sentiment dictionaries are Appointments, provides detailed insight into nearly available, including Affective Norms for English every aspect of the Supreme Court appointment Words (ANEW), WordNet-Affect (WNA), and process. The book’s most relevant component for Whissell’s Dictionary of Affect in Language (DAL). this project is the discussion of nomination setting. It can be difficult to determine which lexicon The nomination setting is defined as the variety of would be most appropriate for a respective research circumstances that surround that appointment of a topic. However, Young and Soroka answered the new member of the Supreme Court that may affect question for this project by finding that the LSD is the amount of controversy the appointment creates. the most appropriate for political communications. Watson and Stookey recognize four primary factors Because the Senate Judiciary Committee hearings that influence the nomination settings: political for Supreme Court nominees are highly political, it composition of the Senate, the level of support in the appears that the LSD would be the most appropriate Senate for the president’s programs, public opinion dictionary for this project. regarding the president, and attributes of the vacancy itself. Sentimentr is an R package that measures sentiment in text (Rinker & Spinu, 2016). It was designed to With the political composition of the Senate, there address the needs of its authors, Tyler Rinker and is expected to be less controversy for the president’s Vitalie Spinu. Before the creation of Sentimentr, nomination if the president’s party holds a majority Rinker and Spinu found that other R sentiment librar- in the Senate. Additionally, it is not just a matter of ies were either too slow or too inaccurate for their pur- which party holds a Senate majority, the size of that poses and that packages that were quick enough did majority matters as well. If the party opposing the not do a sufficient job of considering valence shifters. president holds a large majority over the president’s party in the Senate, there is greater potential for an Valence shifters are words that impact polarized effective opposition campaign against the nominee to words. Polarized words are simply words that have a be organized. positive or negative meaning, which is detected by the sentiment package. There exist several different types Watson and Stookey measure Senate support for the of valence shifters. Negators flip the meaning of a po- president based upon the percentage of bills in which larized word. “Not” would be an example of a negator. the Senate voted in accordance with the position of The word “good” would typically be seen as positive; the president. A higher percentage of bills where the however, “not good” is negative. It should be noted Senate and president aligned in preferences indicates that the Lexicoder Sentiment Dictionary considers ne- greater Senate support for the president. Greater gators, but that is the only valence shifter it considers. support for the president in the Senate is typically Amplifiers increase the intensity of a polarized word. associated with a more seamless appointment Saying “really good” would be measured as more posi- process. tive than just “good.” De-amplifiers decrease the intensity of a polarized word. For instance, “barely good” is Public opinion polls from services like Gallup less positive than “good.” The final valence shifter that provide insight into the popularity of the president Sentimentr considers is adversative conjunctions. This among the general public. If the president is shifter looks for conjunctions that negate the previous unpopular with the public, it is more likely that the clause. For example, in the sentence, “It was good, but president’s nominees will face greater opposition. I wouldn't recommend it,” although the first phrase is positive, the second phrase is not, so Sentimentr would Vacancy attributes are difficult to define, as they overrule the first part with the second. encompass various factors that may impact the

10 Journal of Purdue Undergraduate Research: Volume 10, Fall 2020 Hot and Cold 11 Once the Haynsworth, Carswell, Bork, and Haynsworth, Carswell, Once the in a plain text format, transcripts were Kavanaugh to the transcript additions various nondialogue “(CROSSTALK),” This includes were removed. “(OFF-MIKE),” “(inaudible),” “(LAUGHTER),” were there Additionally, and “(APPLAUSE).” speaker is marked as occasions where the these were also removed. “(UNKNOWN)”; were removed, although Statements from protestors for the Kavanaugh hearings. this was only an issue cleaned of irregularities, they Once the data were script that separates the passed through a Python Python script converted The statements by speaker. into an Excel sheet, where column the text document the statement. For is the speaker and column B is A statements the purposes of this project, opening nominee were and senators directly questioning the evaluated. an updatedOn October 7, 2019, R Street released Bork hearingdataset that included both the Robert Thomas. and the second questioning of Clarence the otherThis provided the opportunity to compare the for data collection methodology with R Street’s wereBork hearing. It appears that the methodologies LSD sentimentfairly consistent when comparing the Bork data scoredscores for the entire hearing text. My For was 0.01381106. 0.01378956, while R Street’s Bork R Street’s the purposes of increased consistency, it appears that the data data were utilized; however, should befor Carswell, Haynsworth, and Kavanaugh hearings. sufficiently comparable to the other using R. Much of the data processing takes place files to The Excel sheets are exported as CSV OCRs ensure better compatibility with R. Since are not perfect at reading characters, many speaker names needed to be corrected to ensure accuracy sometimes the OCR reads Additionally, in the data. “replacement characters,” which are not valid for All non-ASCII characters are sentiment analysis. replaced with a space. The other major aspect of data processing was assigning each of the senators their respective “bioguide_id.” Every member of Congress has a unique identifier assigned to them, and each ID had to be assigned to their statements senator’s The primary purpose of this is to during the hearings. be able to easily associate each of the senators with scores, a measure their corresponding NOMINATE of political ideology (Lewis et al., 2020). The ultimate objective for this R script was to combine the data from each of the files into a single,

DATA COLLECTION PROCESS COLLECTION DATA nomination. Watson and Stookey provide several provide several and Stookey Watson nomination. vacancy attributes: would be considered factors that not successfully vacancy vacancy, “chief justice of swing-vote status an earlier nominee, filled with status representative justice, and special the vacating 1995, & Stookey, (Watson of vacating justice” The PDF data from the Haynsworth, Carswell, Blackmun, and Bork hearings needed to be converted into a format that could be processed these PDFs for sentiment analysis. Unfortunately, were scans from physical copies of the transcripts, Adobe which are difficult to use for textual analysis. Acrobat Pro DC was leveraged to convert these Acrobat generally While PDFs into plain text files. did a good job of converting these documents, the results are imperfect. For instance, due to a lack of clarity in the scans, the optical character reader (OCR) will sometimes read “Senator HRUSKA” (i.e., “U” is and sometimes “Senator HRTJSKA” sometimes mistaken for “TJ”). Mistakes like this one were corrected in an R script. It was also necessary to remove documents that were added to the record in the middle of the transcripts. A few additional hearings were important to include few additional hearings were important to include A Harrold for analysis: Clement Haynsworth, George and Brett Carswell, Harry Blackmun, Robert Bork, appointments Kavanaugh. It is essential that failed hearings might are also studied, as sentiment in these The PDF ones. differ from sentiment in successful can easily documents for many of these hearings website. be found on the Library of Congress’s hearing the transcript for the Kavanaugh However, sourced via was not yet available, so it had to be Advance as a series of text files. Lexis The majority of the transcript data for this project The majority of the transcript data for & Marcum, were sourced from R Street (Weissmann early stages, the When this project was in its 2019). 2019. April 22, latest update to the dataset was from of Lewis It provided the transcripts for the hearings 1987 hearing Powell to Neil Gorsuch, although the hearings from for Robert Bork and the second set of from the Thomas were notable omissions Clarence dataset. These measures of nomination setting will be These measures of the variation in sentiment useful when explaining Court nomination hearings. It is between Supreme worse nomination settings will hypothesized that more negative sentiment in the be correlated with hearings. p. 50). consistent CSV file. It was especially important to 1. Removes contractions (e.g., “isn’t” → “is not”) ensure consistency between the R Street data and 2. Removes words that should not be recognized the data that were processed separately. Once all the by the dictionary via punctuation. For example, hearings were combined and the NOMINATE scores “well” should not be counted as positive when added, the CSV file could be exported for analysis. it occurs at the beginning of a sentence In addition to the transcript data, each of the 3. Creates spaces around punctuation marks independent variables analyzed during this project had 4. Converts negations to read “not” (e.g., “not to also be collected and added to the dataset. In total, very” → “not”) data on 15 independent variables were collected: 5. Removes variations of words that should not be 1. Most recent election vote percentage recognized by the dictionary (e.g., “may very well” → “may very xwell”) 2. Years until next election 6. Removes punctuation from capital letter 3. Former House of Representatives member acronyms (e.g., “U.S.A.” → “USA”) 4. NOMINATE ideology score 7. Removes punctuation from abbreviations (e.g., 5. Chief justice nomination “Dr.” → “Dr”) 6. Past unsuccessful nomination 8. Removes proper nouns (e.g., “Ginsburg” → “G_insburg”) 7. Departing swing justice These functions should improve the accuracy of the 8. Departing justice has special representative sentiment analysis. After Luxon’s processing scripts status have completed, the text is converted to lowercase 9. Presidential approval status and all symbols, numbers, and punctuation, including hyphens, are removed. At this point, the 10. Senate party split text is ready for analysis. 11. Senate support for the president 12. Percentage of vote the president received in the MEASURING SENTIMENT senator’s state during most recent election The LSD comes in two varieties. The standard 13. Hearing on television LSD2015 dictionary is composed only of words or phrases that have either positive or negative 14. Senator is a member of the same party as the sentiment associated with them. LSD2015_NEG president is an extension of LSD2015 that includes negated 15. Senator tenure terms, so instead of having only positive and negative categories it also has neg_positive and neg_negative. The neg_positive category is a word PREPARING FOR SENTIMENT ANALYSIS pattern consisting of a positive word that is preceded There are a few steps that need to occur before by a negation, while neg_negative is a negative word sentiment analysis can take place. To better ensure preceded by a negation. Utilizing this extension the accuracy of sentiment analysis, the data must be to the LSD should improve the accuracy of the preprocessed. Emily Luxon (2017) of the University sentiment measured. For instance, the phrase “not of Michigan compiled an R script that provides a good” would receive a sentiment score of 0.5 by plethora of functions that are used to preprocess using the basic LSD2015 dictionary, which would textual data before conducting sentiment analysis indicate positive sentiment. However, extending the using the LSD. The LSD was selected for this dictionary with LSD2015_NEG results in “not good” project, primarily because it was designed with having a more appropriate score of −0.5, indicating political communication in mind. Additionally, in negative sentiment. Young and Soroka (2012) found their in-depth comparison, Young and Soroka (2012) that using LSD2015_NEG resulted in a “non- showed that it outperforms other popular sentiment negligible increase in performance” when compared dictionaries for political communications. Eight to just using the standard LSD2015, so it is being preprocessing functions from Luxon are used with utilized for this project. the transcripts:

12 Journal of Purdue Undergraduate Research: Volume 10, Fall 2020 Hot and Cold 13 Pr(>|t|) <2e-16 *** 7.26e-06 *** 0.002030 ** 0.000783 *** t-Value 9.439 4.552 −3.108 −3.387 SAME PARTY AS THE PRESIDENT THE AS PARTY SAME RESULTS When the senator is of the same party as the nominating president, then their sentiment during the hearings tends to be more positive. I theorize that this is due to the senator wanting to support their papers, I wanted to make sure that there was not that there was not wanted to make sure papers, I using LSD and difference between a substantial I selected Sentimentr dictionary. another sentiment R, as well as its more compatibility with due to its will sentiment. It technique in measuring advanced secondary measure to ensure essentially act as a The same text cleansing results. the accuracy of my with Sentimentr. techniques were utilized achieved with this project. I Notable success was statistically significant linear found that there were several independent variables correlations between I created two models: of a senator. and the sentiment Sentiment Dictionary and one using the Lexicoder process of eliminating The another with Sentimentr. I started off with a attributes was done manually. then removed model that contained all the attributes, with attributes from the model based on those I only the highest p-value. For these two models, less than included independent variables that had 0.05 for their p-value. of the overall a relatively small proportion However, predicted variance in senator sentiment can be each of using these independent variables, despite The adjusted them being statistically significant. and 0.1041 R-squared value is 0.088 for Lexicoder 1 and 2, Tables for Sentimentr (as shown in This means only about 9% or 10% of respectively). be explained the variation in senator sentiment can R-squared with these independent variables. Low of this nature. values are to be expected with studies Human behaviors are not easily predicted. Std. Error Std. 0.027356 0.000541 0.015242 0.014820 Estimate 0.258218 0.002463 −0.047376 −0.050203 TotalWords Note thatNote is best the suited analyzing LSD for large text. bodies All of tests conducted were on large the and at samples article-level reliability with improves the analyzed. number words of We recommenddo not analyzing text the at sentence (p. 3) to less likely be reliable. as is much it level, (Intercept) Vote % State President Hearing on TV Departing Swing of freedom 0.1096 on 364 degrees error: Residual standard 0.08842 R-squared: 0.09587; Adjusted Multiple R-squared: p-value: 5.239e-08 12.87 on 3 and 364 DF; F-statistic: codes: 0.001 ‘**’ Significance 0 ‘***’, (Positive + NegNegative) - (Negative + NegPositive) - (Negative + NegNegative) (Positive Additionally, it is important to note that whenever Additionally, senator, sentiment is measured for an individual or across an entire hearing, the text from a party, into a all the respective statements are combined as a whole. single block of text and then analyzed Young from This is based upon the recommendation and Soroka (2012) in their Lexicoder Sentiment They are quoted saying: Dictionary Codebook. package In addition to the LSD, the Sentimentr LSD As described earlier, for R was utilized. While only considers the negator valence shifter. this might be completely sufficient in accurately measuring sentiment, especially since LSD has been used in many other political science research The formula results in a single score between The formula results scores associated with −1 and 1, with negative and positive scores with negative sentiment Scores that are closer to the positive sentiment. are more negative or lower or upper bounds to 0 are that score close Texts positive, respectively. considered neutral. The LSD is a relatively simple dictionary as is a relatively simple The LSD A any weighting. words do not hold polarized for the number only needs to account researcher categories. for each of the respective of matches for the sentiment formula is utilized The following calculation. Table 1. R model summary Dictionary. Sentiment using exicoder Table Estimate Std. Error t-Value Pr(>|t|) (Intercept) 0.1967064 0.0273113 7.202 3.44e-12 *** President State % Vote 0.0021433 0.0021433 3.778 0.000185 *** Hearing on TV −0.037245 0.015044 −2.476 0.013754 * Same Party 0.0298179 0.0121204 2.460 0.014353 * Departing Swing −0.050203 0.0146321 −3.630 0.000324 *** Residual standard error: 0.1079 on 363 degrees of freedom Multiple R-squared: 0.1139; Adjusted R-squared: 0.1041 F-statistic: 11.66 on 4 and 363 DF; p-value: 6.39e-09 Signifi cance codes: 0 ‘***’, 0.01 ‘*’

Table 2. R model summary using Sentimentr.

Figure 1. President and senator belong to same party Figure 2. President and senator belong to same party (Lexicoder).* (Sentimentr).*

Figure 3. Hearing was aired on television (Lexicoder).* Figure 4. Hearing was aired on television (Sentimentr).*

*Black bars on the bar charts represent 1 standard deviation away from the average.

14 Journal of Purdue Undergraduate Research: Volume 10, Fall 2020 Hot and Cold 15 Figure 6. Departing (Sentimentr).* a swing voter was justice Figure *Black bars on the bar charts represent 1 standard deviation away from the average. from *Black bars on the bar charts deviation away 1 standard represent Percentage of vote the president received in the senator’s state (Lexicoder). state in the senator’s received the president of vote 7. Percentage Figure Figure 5. Departing (Lexicoder).* a swing voter was Figure justice Figure 8. Percentage of the vote the president received in the senator’s state (Sentimentr).

party’s president in an eff ort to show party unity. It DEPARTING SWING JUSTICE is important to note that signifi cance was only found If the departing justice from the Court was known for this independent variable with the Sentimentr to be a swing vote, then senators tend to be more methodology. The plots shown in Figures 1 and 2 negative (see Figures 5 and 6). This might be represent mean sentiment, with the black error bars attributed to the senators being more critical of the representing one standard deviation. appointed justice in an eff ort to get the president to appoint a more moderate replacement. HEARING AIRED ON TELEVISION Since the hearing for Justice Sandra Day O’Connor in 1981, every Supreme Court nominee’s confi r- SENATOR’S STATE PRESIDENTIAL mation has been aired on television. I found that VOTE PERCENTAGE senators tend to be more negative in hearings that In my opinion, the percentage of the vote the are aired on television (see Figures 3 and 4). This is president received in the senator’s state is the most likely due a cultural shift in the hearings. Bringing interesting contributing attribute in the model. This television cameras into the hearing room made the variable essentially means that senators tend to be hearings themselves much more public, and sena- more positive during the confi rmation hearings if the tors could be scrutinized for their actions during the nominating president performed well in their home hearings. Before the hearings were televised, it was state during the most recent election (see Figures 7 commonplace for only a few senators to ask ques- and 8). I believe this correlation could potentially tions (Farganis & Wedeking, 2014, p. 23). Today, be explained by the senators trying to either better however, it would be considered odd if a senator did represent their state’s political preferences or simply not ask a question during a confi rmation hearing. As improve their reelection chances. a result of this cultural shift in 1981, senators might be more critical during the hearings in an eff ort to show that they are properly vetting the nominees.

16 Journal of Purdue Undergraduate Research: Volume 10, Fall 2020 Hot and Cold

Political

Longman U.S. Senate Trinker/sentimentr: Trinker/sentimentr: . https://www Times New York Voteview: Congressional roll-call roll-call Congressional Voteview: . https://voteview.com accused Brett Kavanaugh. accused Brett .nytimes.com/2018/09/26/us/politics/brett-kavanaugh -confirmation-hearing-transcripts-as-data from 1828 to contains every presidential election by state 2016. Daily Kos. https://www.dailykos.com/stories /2016/11/25/1601042/-Nerd-Alert-This-spreadsheet -contains-every-presidential-election-by-state-from -1828-to-2016 automated coding of sentiment in political texts. Communication, 29, 205–231. https://doi.org/10.1080/105 84609.2012.671234 -accusers-women.html L. (2020). & Sonnet, votes database /nominations/SupremeCourtNominations1789present.htm Court appointments. politics of Supreme Publishing. confirmation hearing transcripts, as data. R Street Institute. https://www.rstreet.org/2019/04/04/supreme-court . http://www.lexicoder.com Dictionary pre-processing /lsdprep_jan2018-r/ V4. https://doi.org 1976–2018. Harvard Dataverse, /10.7910/DVN/PEJ5QU Project. University of California, American Presidency Santa Barbara. https://www.presidency.ucsb.edu/statistics /data/presidential-job-approval 0.4.0. Zenodo. https://doi.org/10.5281/zenodo Version .222103 . https://www.senate.gov/legislative (1789–present) Hauser, C. (2018, September 26). The women who have women who have The (2018, September 26). C. Hauser, This spreadsheet Alert! S. (2016, November 25). Nerd Wolf, The Affective news: L., & Soroka, S. (2012). Young, Lewis, J. B., Poole, K., Rosenthal, H., Boche, A., Rudkin, A., A., Rudkin, H., Boche, Poole, K., Rosenthal, Lewis, J. B., A. (1995). Shaping America: The J. G., & Stookey, Watson, Court A. (2019, October 7). Supreme S., & Marcum, Weissmann, R functions for Lexicoder Sentiment Luxon, E. (2017). R functions for Lexicoder (2017). Election Data and Science Lab. MIT . The job approval 14). Presidential Peters, G. (2018, October (2016, December). V. & Spinu, T., Rinker, Court nominations Supreme U.S. Senate. (2019, June 11).

. CNN. https://www.cnn Vital statistics on Vital . https://www.brookings.edu/multi-chapter Congress .com/2018/09/04/politics/brett-kavanaugh-supreme-court -report/vital-statistics-on-congress/ Kavanaugh confirmation hearing -senate-hearing/index.html confirmation hearings in the U.S. Senate: Reconsidering MI: University of Michigan Press. Arbor, Ann, the charade. REFERENCES ACKNOWLEDGMENTS DISCUSSION Brookings Institute. (2019, March 4). Partisan shouting marks first day of Brett Collinson, S. (2018). Partisan shouting marks first day of Brett Court J. (2014). Supreme Wedeking, & D., Farganis, I want to personally thank Dr. Eric Waltenburg and Waltenburg Eric I want to personally thank Dr. Mintao Nie for their incredible guidance with Dr. encouragement, I would their Without this project. I would research. not have pursued undergraduate the College family, Wilke also like to thank the Arts, and the Office of Undergraduate of Liberal for allowing me Research for their financial support to work on this project. The results from this project imply that senators imply that senators from this project The results Senate during the in their sentiment may be strategic for Supreme Court Committee hearings Judiciary and between sentiment The correlation nominees. vote that a president received the percentage of the election in the senator’s during the most recent This project fascinating. home state is particularly consider various factors may indicate that senators they will question the nominee. in determining how to investigate this, as I am just I plan to continue of what can be learned from scratching the surface with this topic. using sentiment analysis