State Party Platform Project

A great deal has been made about the new Republican Party Platform and its position on abortion. Specifically, the party has dropped language calling for a federal ban on abortion, the first time the party’s national platform did not contain such language since the 1980s. The new plank states: “We believe that the 14th Amendment to the Constitution of the United States guarantees that no person can be denied Life or Liberty without Due Process, and that the States are, therefore, free to pass Laws protecting those Rights.”

Normally, party platforms change slowly, if not glacially. What will be interesting is to see if state parties follow suit. This begs the question as to how much conflict is there between the national platform and existing state party platforms?

Abortion has been a defining ideological issue for the GOP since the 1980s. It is a classic case of issue-evolution. Issue-evolution occurs is when a party’s stance on an issue changes rapidly, as single-issue activists flood into a party, upsetting the previous equilibrium (often a neutral or nominal position on a divisive issue). Political scientists have found evidence for issue-evolution on civil rights and abortion.

In my own research, I have found, consistent with previous research, that sharp changes in Republican abortion positions occurred during the mid-1970s (there is new research coming out soon that the Democratic shift was earlier). For example, between 1976 and 1980, the Minnesota Republican Party changed its plank on abortion, voting down a pro-life plank in 1976 and approving it in 1980. In 1976, the Minnesota GOP turned down a plank banning abortion (except when the mother’s life was in danger) by a vote of 836 to 782 (with 72 “not sure” votes), but by 1980, a “Human Life Amendment” succeeded, winning by a vote of 856 to 476.

Given the recent apparent shift towards moderation (although many are skeptical), I decided to examine how much conflict there is between the GOP’s “new” abortion language and existing party positions at the state level. I looked at the GOP platforms that were issued between 2018 and 2022 to see if any had planks specifically calling for a ban on abortion nationwide. I selected these platforms as they were drafted prior to the current cycle, in which many state parties issued platforms after the national platform was issued (in a future I will examine whether parties issuing platforms before and after the national meeting adjusted their language).

Among the platforms issued by state Republican parties that I was able to collect, all have some position on abortion that can be classified as pro-life. However, in this case, I am only looking for evidence in which state parties are explicitly calling for a national ban on abortion. I found eight states in which the Republican platform calls for an amendment to the Constitution banning abortion (see table below) or some national ban. In contrast, two states that have positions more similar to the current national platform.

The 2020 North Dakota platform is fairly lengthy and points out the GOP’s historic opposition to abortion. In this case, activists are more likely to be upset by the national platform as indicated by the efforts to establish the deep roots of the orthodox position on the issue. But, for most parties, there isn’t a direct conflict and at least two parties (Texas and Idaho) match the current party platform. So, yes, the new national platform language contradicts several state party’s positions, but this may not produce a groundswell of support for change.

In part, institutional rules and organizational norms mean that changes are hard to make. In 2022, for example, the Utah Republican Party did not move on a relatively minor change to remove the clause “except to preserve the life of the mother or in cases of rape or incest” from its abortion plank. Again, in 2024, the state party refused to vote on a proposed plank that would call for “equal protection laws for preborn children from the moment of fertilization.” Party leaders are often sensitive to disrupting coalitions; the Minnesota DFL had a fairly short-lived, but intense fight in 2004 over proposed changes to its abortion language. In these cases, the changes are not going from pro-life to pro-choice (or vice-versa), but rather, subtle changes in how this position is articulated. But, these small changes can upset vocal activists and keeping the status quo position tends to serve the organized party and the politicians on the ballot in November.

This is not a case of issue-evolution. There is no organized push to get the GOP to moderate its abortion position. Generally, as described by Sundquist, Karol, Carmines and Stimson or Aldrich, when party positions change, activists attempt organize where there are low barriers to entry and challenge some existing equilibrium which they oppose. In this case, however, the change is elite-driven and probably temporary and its purpose is political expediency. Activists still prefer the existing position (a national ban) and state parties, such as Texas and Iowa, are more concerned with state autonomy for the purpose of blocking liberal, not conservative, changes to abortion policy. Consequently, I don’t expect any revolution in the GOP’s abortion stance in the near future.

Nice interactive article in the New York Times about word use at the Republican convention.

What I particularly like about this article is that it includes topics that are missing, including words associated with abortion and health care. Too often, text analysis is overly descriptive and doesn’t make use of discriminant validity in comparing texts or speakers.

As a political scientist, there are a lot of great blogs that are academically inclined. Unfortunately, a few of the best seem to be shutting down. The Monkey Cage stopped publishing earlier this year. There were plans to revive it outside of the Washington Post, but I haven’t seen anything. Another great blog, Mischief of Factions, might be shutting down. It was a great blog, but there were fewer and fewer posts and now they are looking for a new editorial team. This seems to suggest that they will keep going, but the unique insights about parties from many of architects of the “UCLA School” were always great to read and the blog was very topical, so like fivethirtyeight, you could get a level-headed, academic take on the latest news (without the vapid “rent-a-quote” commentary that so many of my colleagues seem to prioritize).

I hope Mischief of Factions can keep going.

Here are a couple of articles forthcoming in Party Politics:

“Mapping issue salience divergence in Europe from 1945 to the present” by Jacob R Gunderson. Article doesn’t appear to be paywalled. Good evidence that seems to match, in Europe, of the US pattern of nationalization (or internationalization) of a geographical reduction in issue heterogeneity.

“Blurred positions: The ideological ambiguity of valence populist parties”
Mattia Zulianello and Erik Gahner Larsen. I really like the idea of measuring/analyzing how parties “deliberately take blurry positions” as I think this does not get enough attention in the text analysis literature. Interestingly, as well, they find that ambiguity occurs alongside anti-corruption appeals, “the most paradigmatic example of a non-positional dimension.” I have often felt that the issue ownership literature should take the tendency of parties to equivocate or use less declarative language as evidence of being on the losing side of an issue divide.

I am currently working on a bunch of blog posts (the recent Midwest paper among them), but I thought one thing readers of this blog might be interested in is what articles in text analysis  have been coming out. If you are like me, you subscribe to journal TOC alerts. I try to subscribe to as many as possible. BUT, it’s hard to keep up with new research!! So I thought one thing  could add to this blog is what have been reading lately (no, I can’t read as much as Tyler Cowen).

I am picking these out because they deal with text analysis or have an interesting methodological  contribution the research we all grapple with. My plan is to do this every few weeks or so. If you have any suggestions, please recommend them in the comments!!!

Political groups over national parties: Measuring the Europeanization of the political arena through MEPs Twitter interactions – Livia van Vliet, Juliana Chueri, Petter Törnberg, Justus Uitermark, 2023 (sagepub.com)

Topic Classification for Political Texts with Pretrained Language Models | Political Analysis | Cambridge Core

Mapping Literature with Networks: An Application to Redistricting | Political Analysis | Cambridge Core

One of the things that is frustrating but really important to understand about coding political (or any) texts is how word usage varies by context. This is why sentiment dictionaries often perform poorly or have questionable validity. Human coding still really is the gold standard in terms of validity, but NLP has come a long way (ChatGTP!)

A new paper from Pedro Rodriguez, Arthur Spirling, and Brandon M. Stewart, “Embedding Regression: Models for Context-Specific Description and Inference,” demonstrates a way to use word embeddings to distinguish between how words are used differently by different users, such as politicians or political parties. Their paper is coming out soon in the American Political Science Review. From the abstract:

Social scientists commonly seek to make statements about how word use varies over circumstances—including time, partisan identity, or some other document-level covariate. For example, researchers might wish to know how Republicans and Democrats diverge in their understanding of the term “immigration.” Building on the success of pretrained language models, we introduce the à la carte on text (conText) embedding regression model for this purpose. This fast and simple method produces valid vector representations of how words are used—and thus what words “mean”—in different contexts. We show that it outperforms slower, more complicated alternatives and works well even with very few documents. The model also allows for hypothesis testing and statements about statistical significance. We demonstrate that it can be used for a broad range of important tasks, including understanding US polarization, historical legislative development, and sentiment detection. We provide open-source software for fitting the model.

The program, conText, is easy to implement in R. I ran through their tutorial using my own dataset (about 30 state party platforms from 2020). Their example for immigration worked well on my own dataset. I don’t have a lot of experience with word embeddings, so I used their pre-trained embeddings. Again, I am a novice at this, but I think because the pre-trained feature-embeddings contained most of the words used in the state party platforms, the results made sense. I like the program and their example. For those who use Quanteda, it is very similar (by design) and this made it easier to adjust when I ran into problems. In fact, it only took me about an hour or so to run through the immigration example, and then to also try another issue, abortion. I am definitely impressed. conText doesn’t need all that much user input. Just a few words are enough for it to comb through the corpus and pick out issue-specific words and also disentangle the partisan differences. For example, using nearest neighbors for “abortion” and setting the top 20 nearest words, produced the following:

Democratic words
A tibble: 20 x 4
target feature rank value
1 a legal 1 0.639
2 a access 2 0.579
3 a services 3 0.536
4 a provide 4 0.535
5 a including 5 0.531
6 a provides 6 0.480
7 a ability 7 0.460
8 a law 8 0.454
9 a federal 9 0.441
10 a funding 10 0.440
11 a health 11 0.438
12 a also 12 0.428
13 a women 13 0.425
14 a without 14 0.423
15 a care 15 0.418
16 a decision 16 0.418
17 a support 17 0.416
18 a rights 18 0.415
19 a information 19 0.413
20 a individuals 20 0.412

Republican words

A tibble: 20 x 4
target feature rank value
1 b human 1 0.610
2 b support 2 0.556
3 b child 3 0.506
4 b act 4 0.497
5 b legisl 5 0.464
6 b urg 6 0.464
7 b includ 7 0.458
8 b feder 8 0.455
9 b law 9 0.455
10 b public 10 0.454
11 b servic 11 0.454
12 b fund 12 0.453
13 b life 13 0.439
14 b provid 14 0.434
15 b also 15 0.431
16 b right 16 0.428
17 b children 17 0.419
18 b effort 18 0.412
19 b women 19 0.411
20 b author 20 0.409

(the word stemmings/lemmatization seems not to have been perfect). The idea is that “we know a word’s meaning by the company it keeps” and so differences in the nearest or most frequent neighbors indicate differences in partisan meaning. That said, it still largely picks out word differences (or differences in associations), and can’t necessarily provide insight into how the same word really is different. This is a point that they make in similar publications – researchers needs to make sense of the output and therefore expert knowledge and a good theory is necessary. Interpretation is for the researcher, not the machine. We can understand differences in the meaning of shared words (like “health”, “state”, or “government”), but this will require even more knowledge about the issues and how local (state) parties frame issues to make meaningful inferences.

In the figure below, the Republican words associated with the use of abortion are on the left (I ran this quickly and didn’t get the chance to adjust the code to relabel the plot, hence the “a” and “b” in the plot), the Democratic words are on the right and the shared words (denoted by the triangle) are in the middle.

I’ve been reading Text as Data (Justin Grimmer, Margaret Roberts and Brandon Stewart, Princeton 2022) and I appreciate the clarity of their approach. I also appreciate the sentiment that they express repeatedly that the researcher needs theory to make sense of the text and they “emphasize throughout our book that text as data methods should not displace the careful and thoughtful humanist” (2022: 9). This is a point I will be coming back to over the next several posts: with so many options of NLP, why bother with human coding anymore? I think this is a question that needs attention and I this is something I will be focusing on during my fall sabbatical (it got approved!!!!) and on this blog.

Thanks for reading!

I have been very lucky to be part of a group of researchers (including Dan Hopkins, Dan Galvin, Gerald Gamm, John Henderson, Joel Paddock, and Eric Schickler) who have collected just about every available state party platform issued between 1846 and 2017. The platforms are available via a bulk download at  https://doi.org/10.7910/DVN/KNOSHL.

Again, I am grateful to be part of the this project. My contribution was, admittedly, small. Collecting state party platforms is a giant pain and I only wish I had access to this when I was in grad school working on my dissertation (it would have saved trips to North Carolina and Minnesota – although the interviews and the travelling were totally worth it). It’s a great example of open access/collaborative research.

I recently collected 28 state party platforms that were published in 2020. This is far less 100 if all  state parties issued a platform, but around 30 state parties never issue platforms and many others issue platforms every four years during midterm elections. So, while 28 is less than usual, it still represents a good sample of party debates that occurred during the last year.

To analyze the texts, I used Quanteda, which I highly recommend. It’s easy to use, flexible and I have found it slightly easier to use than tidytext. Of course, regardless of which package one uses, it is really important for users to clean the text. I created my own list of stopwords in addition to the set of base words included in stopwords using the command:

(remove = c(stopwords(“english”), mystopwords)

This is really important because when implementing Wordfish or analyzing differences in word usage, state names will greatly affect the calculations as the proper names will be tagged as highly unusual and confuse them for an ideological signal.

mystopwords <- c(“california”, “state”, “nebraska”, “democrat”,….

First off, here are the obligatory word clouds (wordclouds have been called the mullets of text analysis):

Looking closely, many words aren’t very meaningful: “support”, etc. This is why I think it’s important to really clean the text first, even for really basic analysis. One can think of this in two ways; frequent occurrence of neutral words can be a signal of a moderate platform filled with platitudes. On the other hand, as I have pointed out in some of the posts below, words that appear infrequently can communicate volumes, but they will be buried in pile of noise.

Instead, I have used a keyness plot from Quanteda to highlight the greatest differences in vocabulary between the parties. This seems to communicate the usual differences between the parties. Democrats are focused on social welfare, as nearly all of these words focus on policies and/or beliefs that call for government intervention to reduce inequality: “access”, “afford”, “invest”, “care”, “insure” and “sustain” along with more specific policy words such as “disability”, “housing”, “climate” and “infrastructure.”

In contrast, Republicans focus on freedom from”, using words such as “limit”, “resolve”, and “liberty” and “oppose” and policy words such as “illegal” and “property” . Interestingly, along with traditionally conservative words reflecting a belief in traditional authority (“god”, “parent”, and “authority”) Republicans also tend to focus on the legal structure of the state: “article”, “amend”, “federal”, “legislature”, “govern”, and “constitution.”

The 2020 platforms were quite polarized; although this is nothing new compared to previous years (see posts below). I used Wordfish to estimate the ideology of each platform. These scores should be taken as estimates and this estimation is completely unsupervised, as further text cleaning is necessary. But, not surprisingly, the platforms are quite polarized, as the most conservative Democratic platform (South Carolina) is still much more liberal than the most left Republican platform (Arkansas). The North Dakota platform stands out as unusually conservative; this is confirmed by reports over the summer that the ND GOP was largely disassociating itself from the platform due to fairly extreme planks.

Due to the pandemic, many state parties altered the process for drafting and approving their platforms. I haven’t had the chance yet to really explore the effects of this, but it may have impacted parties that draft planks from the precincts or through in-person caucuses. Many state parties did not hold in person caucuses this year (although caucus states had begun moving towards primaries after 2016) , and few parties had in-person conventions (although Democratic state parties were more likely to hold virtual conventions than Republicans).

In addition to my research on text analysis, I also conduct research on public opinion and political psychology. About a decade ago, I had a battery of questions about environmental attitudes in the 2008 CCES. Recently, the “Green New Deal” has been a subject of considerable political debate between the parties.

Among the questions, was a simple one: “What do you think of when you hear the word ‘green’”? Since the question appeared among a sample of environmental questions, respondents had been primed to think of the word in this sense (and not the political party or even the color). Responses, as I will cover below, are extremely revealing. What I like about this is that the open-ended nature of the question allows for variance in how citizens articulate their environmental beliefs. The responses illustrate many of the key findings in the subfield of political psychology.

Political scientists have shown that citizens draw from elite discussions to formulate their own opinions. We usually see this with regard to closed-ended survey questions, in which citizen attitudes on issues polarize mirroring elite movements on issues, or changes in the political agenda being reflected in the public’s sense of the most important problem.

In this case, reactions to the word “green” produce differences between partisan identifiers. Respondents can draw on a theoretically infinite pool of words, yet clear patterns emerge in the reaction to this question.  This can be shown in the word cloud of all responses. (Note: for most of this analysis, I used Quanteda package in R. It seems fairly easy to use (in the past, I have relied on the tm package and occasionally tidytext).

The dataset has about 4,000 words, about 1,000 of which are unique words. I’ve provided the obligatory wordcloud above. Wordclouds aren’t all that informative. Instead, simple bar charts of word frequencies tend to give a better sense of proportion in terms of how the words are used. The bar chart below is easier to read and we can see that the most commonly used words are fairly unsurprising; the term “green” generally is associated with environmentalism.

These words tend to be used by nearly all respondents. In fact, using a  measure of text similarity, it is clear that partisans across the spectrum tend to draw on similar vocabularies.

                                dem.txt        gopgreen.txt

demgreen.txt    1.0000000    0.8445287

gopgreen.txt   0.8445287    1.0000000

indgreen.txt   0.8255080    0.8188555

However, there are clear differences across parties and that is what I will be exploring in the rest of this post.  For example, Democratic respondents usually emphasize more positive language. The difference plot below shows the words with the largest differences using a “keyness” score. The plot below shows a striking difference in terms of how Democrats and Republicans conceptualize environmental issues. Remember, these are open-ended responses, so these words are the unfiltered expressions of what came to the respondents minds.

Partisan Semantic Differences in Responses to “Green”

The words most likely to appear among Democratic respondents compared to Republican respondents are largely positive. Again, these aren’t surprising as they focus on reducing pollution, protecting the environment. A word like “footprint” is associated with “carbon” and the words like “keep”, “conserve”, “preserve” and “save” all express anxiety about environmental degradation. I suppose one could even draw a larger psychological claim about the orientation towards fragility. I have noted in looking at state party platforms that Democratic discussions on environmental issues are more aligned with Haidt’s purity foundation than any of the other moral foundations (or least as much are harm/care).

The Republican words are far more likely to be adjectives than verbs and these are clearly pejorative. Al Gore takes up two of the most different words and the others, “hippie”, “hugger”, “idiot” and even “lie” make no mistake how this issue is perceived among those on the right. Interestingly, these terms aren’t  really ideological; they represent the affective nature of attitudes many political scientists how found characterizes mass partisan polarization.

Similarly, using sentiment analysis the same pattern emerges. Emotions are connected to the construction of opinions; apparent ideological differences seem to be rooted more in affect than policy. Sentiment analysis shows that the vocabulary of partisans differs markedly. So, these emotional differences paint a clear picture of how partisans conceptualize these issues in emotional terms. Democrats discuss environmental issues in largely positive terms. For Democrats, the environment is and there is optimism that government efforts will be successful. The negative words, such as “destroy” and “harm” are associated with the positive words, “clean” and “protect” as the same underlying sentiment is driving the use of both sets of words.

Independents are more like Democrats, but there is some evidence of emotional ambivalence (for more on how ambivalence affects political reasoning, see here and here). In this case, independents use many of the same positive words that Democrats use, but their negative affect vocabulary is much more distinct. The negative and positive words are orthogonal; these are distinct sets of ideas that are generating the words the respondents use (I did find evidence in an unpublished paper that environmental issues generate a high degree of ambivalence). Words such as “problem”, “restrict” and “slow” are largely used when discussing environmental protection. In contrast, obviously words like “fake”, “hoax” and even “bullshit” are not describing the costs of environmental protection, but reflect a  rejection of the legitimacy of the issue. In other words, this is not ambivalence; the responses are not illustrative of someone struggling with internal conflict over competing or inconsistent goals or beliefs.

Finally, the Republican sentiment is almost exclusively negative. What separates Republicans from independents in this regard is that they have a much larger vocabulary (partisan responses were longer than independent open-ended responses). “Bullshit” actually appears twice (“bs”) and the condemnation of the other side is more apparent indicating a threat response provoked by the use of “green” to describe environmental issues. The words such as “corrupt”, “extremist”, “greed”, “scam” and “idiot” all refer not to a preference for economic growth over environmental issues, but describe the perceived threat that Republicans identify coming from both the promoters and believers of climate change.

While I had asked this survey question over a decade ago, the issues are in some ways more relevant today. Partisan responses about the environment are often framed as policy debates about the merits of economic growth and versus environmental protection. These responses, however, show that while this is the case for some respondents, the divide has more to do with subjective perceptions about the motives of the other side or the gravity of the issue. Responses are emotional, tapping into anger, anxiety and even enthusiasm. These responses were generated with little more than an ink-blot of priming, illustrating the biased-nature of information processing. One final note is that there is a considerable value to using open-ended survey questions, especially when they are conducted online, as the unstructured nature of these responses can be quite revealing in showing how citizens construct attitudes and opinions.

The eighth edition of the State of the Parties just came out! It’s a great book with contributions from many of the leading political science researchers in American politics. As with previous editions, there are sections on public opinion, campaign finance and an expanded focus on the nature of the current party system in America and what changes it may be undergoing.

As with previous editions, I have a chapter on state party platforms. What is different about my chapter this edition is the use of automated techniques to evaluate platform ideology. I used Wordscores to measure state party ideology instead of manual coding. In a future, I will show how the results vary based on the method used for coding.

The main finding in my chapter is that state parties remain highly polarized. I did not find, however, much evidence of a divide within the parties. A lot has been made of the division between Establishment and Trump Republicans and between Establishment and Progressive Democrats. The state party platforms, however, do not reveal much evidence of this divide. It could be that state party platforms just aren’t a good measure of intra-party divisions (although my previous research does capture some meaningful diversity of agenda attention). Dan Hopkins recent book shows that parties are nationalizing and the  state party platforms over the last century show greater homogeneity within each party. I wonder, though, if this nationalization could also show when the parties are becoming internally divided. In other words, if the divisions are being fought at the national level, then factional struggles at the state level should reflect this (as seen in several recent primaries as well as control over state party organizations).

I didn’t find evidence that states won by Clinton or Sanders were ideologically different; ditto for Trump versus states won by Establishment GOP candidates (there were some differences using sentiment analysis). Again, it could be the fact that platforms, because they aren’t written in every state and revised every year, won’t capture these divisions. Still, one of my conclusion was that newly engaged party activists have not sought to institutionalize their ideals into formal party documents as they have in the past. I wonder, however, as the Democratic Party moves away from caucuses, that this indicates past processes of party agenda formation are anachronistic. In an age of social media and online fundraising, the party organization is large and cumbersome mechanism to move when other smaller organizational tools and vehicles are available. As political scientists, we devote a lot of our research to understanding how parties as organizations reflect the goals of politicians, social movements, and activists. I just wonder if, as with many other institutions, parties will become less important in the coming years. Sure, it’s a big leap from a single year of state party platforms to this conclusion. Still, I think its worth thinking about how parties formalize their agendas and what this means about party organizational strength.