From Washington to Obama: The Evolution of Presidential Speech
Comparing the Sentiment, Language Use, and Readability of Presidential Inaugural Speeches Over Time
Abstract
Analyzing political rhetoric in inaugural addresses made by U.S. presidents from Washington (1789) to Obama (2009). Specific methods of analysis include tone, word choice, and readability and how their values have shifted over time.
Image source: The New Yorker
Introduction
From calls for unity to speeches delivered in times of crisis, presidential inaugural addresses show how leaders shape their language to reflect the moment. In this project, we explore how tone, word choice, and readability have shifted from Washington to Obama using sentiment analysis (AFINN) and readability tools like Flesch-Kincaid. We’re interested in the patterns — how presidential speeches and political rhetoric have changed over time, and what that says about leadership, politics, and the public they’re speaking to.
Methods
The inaugural speeches were scraped from Avalon Project at Yale Law School. For more information on the data and methodology, please see the About page.
Sentiment Analysis Over Time
To assess how inaugural speeches have changed over time, we first looked at average sentiment over time. This interactive bar chart displays the average AFINN sentiment score of each U.S. presidential inaugural address from 1789 to 2009 (Nielsen 2011). The chart was created using ggplot2 with a custom blue–gray–red gradient and made interactive using plotly, allowing readers to hover over each bar for details like president name, year, and sentiment score (“Plot (plotly)” n.d.; Wickham 2024). Please see the About page for more information on methodology and why we used average AFINN sentiment score.
Bars are color-coded by sentiment strength, with blue tones representing more positive language and red tones more negative sentiment. This helps visualize how the emotional tone of inaugural speeches has shifted over time.
While sentiment has remained mostly positive throughout history, we observe notably lower average sentiment during the mid-19th century, especially surrounding the Civil War era under Abraham Lincoln. A brief dip also appears in the early 20th century. More recently, sentiment peaked with George H. W. Bush’s 1989 inaugural address — the highest score in the dataset. Since then, sentiment scores have generally declined, suggesting a rhetorical shift toward more neutral or somber tones in modern speeches.
Note: Lincoln’s relatively low sentiment score isn’t a reflection of the purpose or urgency of his speech, but likely a result of the AFINN lexicon assigning negative values to many war-related terms. For more detail on how AFINN works and why this matters, see our About page.
While sentiment helps us understand the emotional tone of inaugural speeches, it doesn’t tell us what presidents were actually talking about. To dive deeper into the substance of presidential rhetoric, we analyzed the most common two-word phrases, or bigrams, used across six major historical eras.
Normalized Bigram Frequency by Era
Bigrams marked with an asterisk appear in more than one era.
This interactive tabset highlights the most frequently used bigrams (two-word phrases) in inaugural speeches across six major historical eras. Users can click through the tabs to view the top bigrams for each era. Frequencies are normalized to account for era and speech length, so each bar reflects that bigram’s share of all bigrams within its era. Normalization helps adjust for the fact that different eras had varying numbers of inaugural speeches and speech lengths, allowing for fairer comparisons of which biwords were most prominent within each period. For more information on data wrangling, please see the About page.
Bigrams that appear in multiple eras (marked with an asterisk) suggest recurring rhetorical themes like “federal government” or “foreign nations,” underscoring the lasting importance of governance and international relations in presidential rhetoric.
While some language remains consistent over time, the top bigrams also shift to reflect the issues and values most relevant to each era:
In the Revolution & Early Republic era (1789–1815), frequent bigrams like “foreign nations” and “public opinion” reflect the new republic’s focus on diplomacy and democratic legitimacy. The appearance of “human nature” and “human mind” reflect Enlightenment ideals, emphasizing reason, morality, and the philosophical foundations of governance.
During the Expansion, Reform & Civil War era (1816–1865), phrases like “federal government” and “public money” signal increased federal responsibility, while “naval force” and “indian tribes” point to growing military power and westward expansion.
In the Reconstruction to WWI era (1866–1919), phrases like “civil service” and “public expenditures” reflect growing attention to internal administration and government reform.
The Interwar & WWII period (1920–1945) includes bigrams like “eighteenth amendment” and “illegal liquor”, capturing the national focus on Prohibition.
In the Postwar & Civil Rights era (1946–1969), phrases like “united nations”, “armed attack”, “free nations,” and “free world’s” reflect Cold War tensions and the defense of global democracy. Figure 6
During the Post-Watergate to Obama era (1970–2009), bigrams like “nuclear weapons”, “god bless”, and “vice president” suggest a shift toward national security concerns, as well with moral and religious appeals in presidential rhetoric.
These changes underscore how presidential language not only reflects shared values, but also adapts to shifting historical and political priorities.
TL;DR: While some themes persist, each era’s top bigrams reveal shifting national concerns, from Enlightenment ideals to Cold War tensions and modern appeals to faith.
Readability Analysis
Readability analysis helps us understand how easy (or difficult) a text is to read. Two common measures are the Flesch Reading Ease (FRE) and Flesch-Kincaid Grade Level (FKGL)
Understanding the relationship: Flesch Reading Ease (FRE) & Flesch-Kincaid Grading Level (FKGL)
To get a better understanding of the relationship between the Flesch Reading Ease (FRE) and the Flesch-Kincaid Grade Level (FKGL), it is helpful to plot them together on a scatterplot.
`geom_smooth()` using formula = 'y ~ x'The FRE and the FKGL have an inverse relationship because they are based on similar textual features—sentence length and word complexity—but use opposing scales to express readability.
\[ \text{FRE} = 206.835 - 1.015 \times \left( \frac{\text{Total Words}}{\text{Total Sentences}} \right) - 84.6 \times \left( \frac{\text{Total Syllables}}{\text{Total Words}} \right) \] \[ \text{FKGL} = 0.39 \times \left( \frac{\text{Total Words}}{\text{Total Sentences}} \right) + 11.8 \times \left( \frac{\text{Total Syllables}}{\text{Total Words}} \right) - 15.59 \]
As sentence length and syllables per word increase, FRE scores decrease (the text becomes harder to read), while FKGL scores increase (the grade level rises). This opposing behavior is why they have an inverse relationship: when one goes up, the other tends to go down.
Flesch Reading Ease (FRE)
\[ \text{FRE} = 206.835 - 1.015 \times \left( \frac{\text{Total Words}}{\text{Total Sentences}} \right) - 84.6 \times \left( \frac{\text{Total Syllables}}{\text{Total Words}} \right) \]
This score rates text on a 0 to 100 scale. A higher score means the text is easier to read.
- 100 - 90: 5th Grade
- 90 - 80: 6th Grade
- 80 - 70: 7th Grade
- 70 - 60: 8th & 9th Grade
- 60 - 50: 10th to 12th Grade
- 50 - 30: College
- 30 - 10: College Graduate
- 10 - 0: Professional
FRE vs. Sentiment
To better understand the relationship between writing style and emotional tone in the presidential inaugural addresses, we compare the FRE scores with sentiment scores calculated for each speech.
The FRE measures how easy a text is to read, while the sentiment score captures the overall positivity or negativity of the language. By analyzing these two metrics together, we can apply clustering techniques to group speeches with similar readability and sentiment profiles. The number of clusters was determined using an elbow plot and diminishing returns in reducing within-cluster variance.
Using the clustering analysis, some light interpretation can be made from the groupings.
Cluster 1: Presidents have average sentiment score with low readability.
Cluster 2: Presidents have higher sentiment score with low readability.
Cluster 3: Presidents have higher sentiment score with high readability.
Cluster 4: Presidents have low sentiment scores with high readability.
Cluster 5: Presidents on the lower end of the sentiment score spectrum. Readability ranges from standard to easy.
Cluster 6: Presidents in this cluster have the highest sentiment score spectrum. Readability ranges from 30-40.
Flesch-Kincaid Grade Level (FKGL)
\[ \text{FKGL} = 0.39 \times \left( \frac{\text{Total Words}}{\text{Total Sentences}} \right) + 11.8 \times \left( \frac{\text{Total Syllables}}{\text{Total Words}} \right) - 15.59 \]
This score estimates the U.S. school grade level needed to understand the text. Due to this formulas construction, the score does not have an upper bound.
For example:
- A score of 8.0 means the average 8th grader should be able to understand it.
- A score of 12.0 suggests it matches a high school senior’s reading level.
FKGL vs. Sentiment
In addition to examining readability through the Flesch score, we also analyze the FKGL in relation to sentiment scores for each inaugural address. By comparing FKGL to sentiment scores, we explore how the sophistication of presidential language interacts with the emotional tone conveyed while using clustering methods.The number of clusters was determined using an elbow plot and diminishing returns in reducing within-cluster variance.
In this example, clustering analysis using FKGL vs Sentiment values does not yield any meaningful results. Clustering appears to be driven entirely by the sentiment axis and form vertical “layers” that reflect sentiment values alone.
This suggests that the algorithm is not detecting any significant structure in the Flesch-Kincaid axis, and as a result, the clustering lacks any interpretive value.
FKGL vs. Time
We also examine the relationship between the FKGL and time to track how the complexity of presidential inaugural speeches has changed. Overall, we find a clear trend that speech complexity has decreased over time, with more recent presidents using simpler, more accessible language compared to earlier eras.
`geom_smooth()` using formula = 'y ~ x'Possible explanations include:
- Initially, the audience of the inaugural speech was composed of congressional leaders.
- The growing emphasis on mass communication, as presidents began speaking to a broader and more diverse audience.
- The influence of modern media, which favors shorter, clearer messaging that resonates quickly with the public.
| President | Year | Flesch-Kincaid Score |
|---|---|---|
| George Bush | 1989 | 7.15 |
| Lyndon Baines Johnson | 1965 | 7.54 |
| Dwight D. Eisenhower | 1957 | 8.09 |
| Franklin D. Roosevelt | 1945 | 8.89 |
| Richard Milhous Nixon | 1969 | 9.24 |
| President | Year | Flesch-Kincaid Score |
|---|---|---|
| John Adams | 1797 | 28.74 |
| George Washington | 1789 | 28.43 |
| James Madison | 1809 | 25.57 |
| Thomas Jefferson | 1805 | 24.19 |
| Zachary Taylor | 1849 | 23.77 |
Figure 12 displays a table ranking the five presidents with the lowest Flesch-Kincaid Grade Level scores between 1789 and 2009. Similarly, Figure 13 displays a table ranking the five presidents with the highest Flesch-Kincaid Grade Level scores. By looking at the years, we can see how the highest scores correspond to the presidents in the first two centuries while the lowest scores correspond to the presidents in the more recent centuries. This indicates that the language used in inaugural speeches has become simpler and more accessible over time (Liberman 2011). The tables’ features were accomplished from the kabbleExtra package by following the documentation provided by CRAN (“Create awesome HTML table with knitr” 2024) and the colors were taken from the hex color codes documentation (“Colors in r” n.d.)
Sentiment and Flesch-Kincaid Scores Change among Presidents that Served Two Terms
Figure 14 displays an interactive scatter plot showing the sentiment scores of U.S. presidents’ first and second inaugural addresses for those who served two terms. Similarly, Figure 15 shows presidents’ first term Flesch-Kincaid scores and second term Flesch-Kincaid scores. By hovering over each data point, the user can view the president’s name as well as their respective scores. Each president is colored differently.
By examining the sentiment and Flesch-Kincaid scores across different terms, we can explore how presidential leadership and political rhetoric evolve after serving a first term. This is important because it may reveal a correlation between a president’s experience and changes in language complexity or emotional tone
Difference Score
| President | Sentiment Score Difference | Flesch-Kincaid Score Difference |
|---|---|---|
| George Washington | -0.74 | -11.28 |
| Thomas Jefferson | -0.10 | 4.93 |
| James Madison | -0.99 | -7.53 |
| James Monroe | -0.39 | 2.03 |
| Andrew Jackson | -0.23 | -2.43 |
| Abraham Lincoln | 0.06 | -1.88 |
| Ulysses S. Grant | -0.38 | 1.03 |
| Grover Cleveland | -0.36 | -0.82 |
| William McKinley | -0.19 | -3.13 |
| Woodrow Wilson | 0.65 | 0.12 |
| Franklin D. Roosevelt | 0.09 | -1.01 |
| Dwight D. Eisenhower | 0.02 | -2.25 |
| Richard Milhous Nixon | 0.15 | 3.06 |
| Ronald Reagan | -0.04 | 0.64 |
| William Clinton | 0.33 | 0.45 |
| George W. Bush | 0.09 | 1.29 |
Sentiment Score difference:
- Positive Difference Value: The second term inaugural speech was more positive than their first term inaugural speech.
- Negative Difference Value: The second term inaugural speech was less positive or more negative than the first term inaugural speech.
- Zero or close to Zero Difference Value: The sentiment/tone of the inaugural speech remained pretty consistent.
Flesch-Kincaid Score Difference:
- Positive Difference Value: The second term inaugural speech was more complex than the first term inaugural speech.
- Negative Difference Value: The second term inaugural speech was simpler and easier to understand than the first term inaugural speech.
- Zero or close to Zero Difference Value: The complexity of the inaugural speech remained pretty consistent.
To see this more clearly, Figure 16 displays a table showing the changes in sentiment scores and Flesch-Kincaid scores for all presidents that served two terms. This will facilitate determining whether their tone became more positive or negative and whether their language became more complex or simpler across different terms. As we can see from Figure 16, the emotional tone of inaugural speeches across these presidents’ terms didn’t vary as much since the difference of all sentiment scores approximate 0. Most of the speeches across both terms remained pretty consistent in terms of sentiment, with about half of the second-term inaugural speeches having a sliglty less positive or more negative tone than their first term speech while the other half were slightly more positive than their first term inaugural speech. On the other hand, the complexity of most of the second term inaugural speeches became simpler and easier than their first term inaugural speech in terms of readability, while only a few second-term inaugural speeches either remained consistent or became slightly more complex than that president’s first term inaugural speech.
Conclusion
From sentiment and word choice to readability and rhetorical complexity, our analysis reveals that presidential inaugural speeches have evolved in ways that reflect both changing political priorities and shifts in public communication. Early speeches often featured higher complexity and formal tone, while more recent addresses have grown simpler and more emotionally neutral, possibly in response to the broader audiences that presidents now address (see Figure 1, Figure 11, Figure 12, and Figure 13).
Repeated bigrams such as “foreign nations” and “federal government” highlight consistent institutional themes (see Figure 2 through Figure 7), while new phrases like “united nations” and “god bless” signal emerging concerns and values. Readability metrics show a clear decline in linguistic complexity over time (see Figure 11), and clustering analyses suggest that tone and readability often vary together (see Figure 9 and Figure 10).
Exploring the difference in emotional tone and readability scores across presidents that served for two terms revealed that a president’s experience can change the language they use while emotional tone can remain consistent, but lean toward either a slightly more positive or negative tone. This shows that leadership and political rhetoric can evolve after serving a first term (see Figure 14, Figure 15, and Figure 16).
These patterns show that inaugural speeches aren’t just about outlining plans. They also reflect the tone, priorities, and challenges of the moment in which they were delivered.
References
“Colors in r” (n.d.). R Charts, Online documentation, Available at https://r-charts.com/colors/.
“Create awesome HTML table with knitr::kable and kableExtra” (2024), CRAN, Online tutorial, Available at https://cran.r-project.org/web/packages/kableExtra/vignettes/awesome_table_in_html.html.
Liberman (2011), “Real trends in word and sentence length,” Language Log, Electronic article,.
Nielsen, F. Å. (2011), “AFINN sentiment lexicon.”
“Plot (plotly)” (n.d.). Shiny, vignette, Available at https://shiny.posit.co/r/components/outputs/plot-plotly/.
Wickham, H. (2024), “Scale gradient functions in ggplot2,” Online documentation, Available at https://ggplot2.tidyverse.org/reference/scale_gradient.html.