participatory-inequality.qmd

---
journal:
  blinded: false
wordcount: 9965
title: "Inequality in Agency Rulemaking"
format: 
  #docx
  cmc-article-pdf:
    fig-pos: 'h'
    fig-env: "figure"
    keep-tex: false
    filters: [citeproc.lua, wordcount.lua] # https://github.com/andrewheiss/quarto-wordcount
  # For the sideways table in appendix.
    include-in-header:
        text: |
          \usepackage{lscape}
          \newcommand{\blandscape}{\begin{landscape}}
          \newcommand{\elandscape}{\end{landscape}}
linestretch: 1.5
indent: true
number-sections: true
cap-location: top
runningtitle: Inequality in Agency Rulemaking
runningauthor: Carpenter et al.
author:
  - name: Daniel P. Carpenter
    affiliations: 
     - name: Harvard University
       department: Department of Government and Radcliffe Institute
  - name: Angelo Dagonel
    affiliations:  
     - name: Harvard University
       department: Department of Government
  - name: Devin Judge-Lord
    affiliations:  
     - name: University of Michigan
       department: Ford School of Public Policy
    email: "judgelor@umich.edu"
    attributes:
      corresponding: true
  - name: Christopher T. Kenny
    affiliations:  
     - name: Harvard University
       department: Department of Government
       # email: christopherkenny@fas.harvard.edu
  - name: Brian Libgober
    affiliations:  
     - name: Northwestern University
       department: Department of Political Science
       #email: brian.libgober@northwestern.edu
  - name: Steven Rashin
    affiliations:  
     - name: University of Texas, Austin 
       department: McCombs School of Business
  - name: Jacob Waggoner
    affiliations:  
     - name: Harvard University
       department: Government and Social Policy
  - name: Susan Webb Yackee
    affiliations:  
     - name: University of Wisconsin, Madison
       department: La Follette School of Public Affairs
thanks: |
    Authors are listed alphabetically; the listing does not imply the significance of contribution. We will be extremely grateful for any comments. The most recent draft is here: [judgelord.github.io/finreg](https://judgelord.github.io/finreg/participatory-inequality.pdf). We thank Larry Bartels, Devin Caughey, Christine Desan, Sanford Gordon, Alexander Hertel-Fernandez, Howell Jackson, David Moss, Eric Talley, Noah Greifer, and participants in the American Political Economy Conference and Harvard American Politics Research Workshop for comments and James Chen for excellent research assistance. We acknowledge the Russell Sage Foundation, the Washington Center for Equitable Growth, and the Radcliffe Institute for Advanced Study for research support. Judge-Lord acknowledges support from the APSA DDRIG (NSF Grant #2000500). Please do not cite without permission.
bibliography: "assets/finreg.bib"
execute:
  echo: false
  cache: false
  message: false
  warning: false
abstract: |
  Most U.S. law is now made by executive-branch agencies under pressure from vast flows of money, lobbying, and political mobilization. Yet, research on inequality overlooks administrative policymaking. Analyzing a new dataset of over 260,000 comments on draft agency rules implementing the Dodd-Frank Act, we identify the lobbying activities of over 6,000 organizations. Leveraging measures of organizations’ wealth, participation in administrative politics, lobbying sophistication, and lobbying success, we provide the first large-scale study of wealth-based inequality in agency rulemaking. We find that wealthier organizations are more likely to participate in rulemaking and enjoy more success in shifting the content of policy documents, while organizations with more members do not enjoy more success. More profit-driven organizations are also more likely to participate and enjoy more lobbying success. Wealthier organizations’ ability to marshal legal and technical expertise appears to be a key mechanism by which wealth leads to lobbying success.
keywords: "Inequality, Bureaucratic Policymaking, Interest groups, Lobbying, Rulemaking, Financial Regulation"
date: today
---

```{r}
source(here::here("code/setup.R"))

# overide defaults 
knitr::opts_chunk$set(echo = FALSE, cache = FALSE) 

# asset data from https://judgelord.github.io/finreg/participation#Save_asset_data
load(here::here("data", "commenter_assets.Rdata"))

# comment data from https://judgelord.github.io/finreg/efficacy#Save_data
load(here::here("data", "efficacyXsophistication.Rdata"))
``` 

\clearpage 

# Introduction {#sec-intro}

<!-- Steve 2024-06-17: below line "executive-branch agency policymaking" is confusing.  
I changed it to policymaking at executive branch agencies. 
-->

Studies of political inequality have revealed profound and durable patterns where wealthier groups have a disproportionate influence on legislative processes. Work in American politics by @Bartels2008, @Baumgartner2009, @Hacker2010, @Gilens2012, @Skocpol2004, @Schlozman2012, and others shows ties between economic and political inequality. In contrast to the large literature on inequality in legislative lawmaking, research on inequality in policymaking at executive branch agencies is sparse. Fundamental questions about economic and political power have yet to be addressed systematically: Does wealth-based inequality drive differential participation during administrative policymaking? Are agency officials more likely to address concerns raised by wealthier organizations? If so, why?


Scholars have focused on inequalities in legislative lobbying and influence partly because quantitative data exist in the form of legislative Lobbying Disclosure Act reports and congressional voting records.  A major barrier to scholarship on inequality has been the lack of parallel quantitative data on lobbying and policy outcomes for agency policymaking. This paper introduces such measures of wealth inequality, participation, and influence of organizations in agency rulemaking, enabling new tests of inequality in American policymaking. 

<!-- Angelo 2024-06-14: The paragraph below is a good place to signpost in response to R1's "Why Dodd-Frank" concern.  "Way forward" suggests "selling" the Dodd-Frank study as a reasonable scope around "broad but related policy". 
Steve 2024-06-17 added a sentence beginning with ... Dodd-Frank is an ideal venue
https://link.springer.com/article/10.1007/s11149-019-09379-8
-->

To investigate the link between economic and political inequality, we focus on agency rules implementing the Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010 (hereafter Dodd-Frank). Post-Dodd-Frank rulemaking is an ideal context to study inequality in administrative policymaking because the legislation delegated considerable authority to executive branch agencies to re-regulate the financial system.  The re-regulation sought changes to many important aspects of the financial system, such as ensuring the stability of systemically important banks and establishing new agencies and offices to protect consumers. The economic stakes were massive; agencies proposed rules that sought to increase compliance costs by billions of dollars.  Organized interests spent hundreds of millions on lobbyists and lawyers attempting to alter the proposed rules.  

## Summary of Contributions

First, we create a new database of all `r n_comments` public comments on proposed rules implementing Dodd-Frank, focusing especially on comments from companies and other organizations. Our data cover over eight hundred regulatory actions (in `r n_rules` rulemaking processes) across seven agencies.

<!-- Angelo 2024-06-14: R2 was concerned the paper is suggesting a contribution of new text analysis methods. "Way forward" suggests specifying what already existing methods are being used/combined, e.g., "dictionary-based methods". -->

Second, we develop a suite of new measurement and analytic tools to study who lobbies during rulemaking, how sophisticated their lobbying efforts are, and which organizations have their concerns addressed in the final rules (and which do not).  

Third, we leverage these data and tools to provide the first large-scale assessment of the effects of wealth inequality on agency policymaking. In doing so, we answer questions regarding inequality and lobbying participation which were, up to now, only answerable in the legislative process. Beyond prior studies showing differences between business and non-business groups, we can compare lobbying behavior among similar organizations. For example, we compare commenting behavior among banks. In doing so, we control for many known sources of variance in commenting behavior, which yields cleaner tests of the relationship between organizational wealth and policy influence.

## Summary of Findings

We find that wealthier and more profit-motivated organizations are more likely to participate in administrative policymaking, and even when the less wealthy organizations participate, wealthy organizations are more likely to have their concerns addressed.

First, we find that wealthier organizations participate in agency rulemaking at higher rates than less wealthy organizations. We replicate this result within and across various types of for-profit firms and non-profit organizations.

Second, we find that for-profit banks are more likely to participate than non-profit banks such as credit unions and savings associations.

Third, we find that organizations that spend more money on political campaigns and legislative lobbying are also more likely to participate in rulemaking.

<!--MOVED TO APPENDIX 
Fourth, among organizations that participate in rulemaking, we show that organizations that participate frequently are wealthier than those that participate infrequently.
-->

Fourth, wealthier organizations marshal more technically and legally sophisticated comments than less wealthy organizations.

<!-- Steve 6/17 merged the two paragraphs.  Before there was a break before "Using causal mediation analysis..."  That paragraph was related to the sixth finding, and it was weird that we had six findings and then a seventh paragraph that looked like a finding  
-->

Finally, text from the comments of wealthier organizations is more likely to be incorporated into final rule documents, suggesting that wealthier organizations' concerns are more salient to policymakers and more effective in shaping the policy agenda of the rulemaking process. Using causal mediation analysis, we find that the ability of wealthy organizations to marshal legal and technical expertise appears to be a key mechanism by which wealth leads to lobbying success. Lobbying sophistication explains much of the relationship between wealth and lobbying success. Money buys technical and legal sophistication, and sophistication appears to buy changes to policy documents. In contrast, campaign donations and total lobbying spending do not appear to explain a significant share of the relationship between wealth and lobbying success.

These results hold implications for our understanding of public participation and voice in the policymaking process and possible policy reforms. For example, scholars have long known that the high barriers to participation in agency rulemaking tended to amplify the voices of those who have resources over those who don’t. We show that this inequality runs much deeper than previously appreciated: our findings suggest that, even among organizations with some wealth and sophistication, those with the most wealth and sophistication tend to participate in agency rulemaking and enjoy more lobbying success. As a result, these findings imply a new understanding of privilege among interest groups and its impact on American policymaking.

Regarding policy reforms, our findings suggest that limiting campaign contributions would have little effect on the lobbying success of wealthy organizations at this stage of the policy process. To the extent that the lobbying sophistication mechanism is causal, our results highlight that reform efforts targeting inequalities in access to legal and technical expertise---such as those giving legal assistance and resource subsidies to poorer organizations to write more sophisticated comments and policy recommendations---may be effective in moderating the disproportionate influence of wealth inequality in administrative policymaking.



# Theory

<!-- Angelo 2024-06-14: Responding to R3's confusion about organization vs. individual wealth, it might be good to tailor the citations below to research on *organizational* wealth and inequality. They may be reading the Hacker Pierson and Gilens citations as the paper being concerned with individual wealth, not organizational wealth which is the actual focus. -->
<!-- 
Steve's unfunded mandate thought it would be nice to organize this section more with subheadings.  However, given that it's just an idea and I can't figure out a good way to implement it, it is probably ok to ignore this comment. 
-->

## Inequality in Agency Rulemaking 

The past two decades have witnessed an outpouring of political science research on how economic inequality shapes policy outcomes that generate further economic and social inequality. @Bartels2008 shows that legislative voting patterns in the U.S. Senate disproportionately reflect the preferences of those individuals at the highest levels of the income distribution. @Hacker2010 describe a "winner-take-all politics" by which wealthier Americans improved and secured their economic prospects under both liberal and conservative political leadership while the prospects for middle- and working-class Americans stagnated. @Gilens2012 further systematized these findings using survey data and legislative voting records. Many studies support and refine these observations [e.g., @Baumgartner2009; @Winters2009; @Kelly2010; @Schlozman2012; @Page2013; @Gilens2014; @Witko2021].

However, empirical portraits of the relationship between wealth and political inequality in the U.S. remain incomplete. The (relative) exclusion of administrative processes from the study of inequality is a major omission, as bureaucracies are "an essential site of political contestation" [@SoRelle2023], especially over policies with diffuse beneficiaries and concentrated costs [@Lowi1964]. 


Policymaking does not stop when Congress passes a law. Many critical policy decisions are made by administrative agencies, in part because the legislature delegates significant policymaking authority and discretion to these agencies to make public policy [@Epstein1999; @Huber2002; @Haeder2020]. Legislation almost always requires federal agencies to write the legally binding standards and procedures (i.e., rules) that give statutes practical effect [@West1995; @Kerwin2018]. 

Agency rulemaking has become the primary mode of policymaking in the United States. In 2023, federal agencies finalized over 3,000 legally binding rules, while only 30 bills were passed by Congress and signed into law.
<!--, 10 of which were appropriations, appointments, and honorific naming. https://docs.google.com/spreadsheets/d/1wvK7_0DydAsgpz_IAEz8Op5oWj_tuDYhyTLp04rhWdo/edit?gid=0#gid=0

These dynamics are often studied under the concept of regulatory capture [@Carpenter2013]. Still, few regulatory capture projects speak to questions of wealth-based or economic inequality. Likewise, few studies of political inequality address bureaucratic policymaking.-->
Given the scale and importance of agency policymaking and the large volume of data on business and interest group lobbying, rulemaking presents a unique opportunity to study the relationship between organizational wealth and policy influence. 

The rulemaking process creates opportunities for voice and influence. The Administrative Procedure Act of 1946 (APA) requires federal agencies to solicit public comments on their draft policy proposals and to consider any substantive comments before issuing a legally-binding final rule. Agency officials have discretion to change the rule text based on public comments. The firms and other organizations most affected often attempt to influence regulatory policy content by submitting public comments.^[Federal agency restrictions on ex parte (or "off the public record") lobbying after the issuance of a proposed rule generally allow researchers to use comments during notice and comment rulemaking to study lobbying [@Yackee2012].] 

<!--DIFFERENTIAL LOBBYING AND CAMPAIGN SPENDING-->
Because agencies make policy, moneyed interests spend considerable resources to influence administrative and executive decision-making [@Haeder2015; @You2017].
Firms collectively spend hundreds of millions of dollars lobbying after a bill becomes law, including lobbying the agencies tasked with writing the implementing rules [@You2017; @Ban2019], often spending more on lobbying agency officials than legislators [@Libgober2024]. Moreover, corporate lobbying of legislators often aims to enlist them in efforts to lobby agency officials, and legislators who receive more corporate Political Action Committee (PAC) money from companies are much more likely to lobby federal agencies on behalf of those companies [@Powell2022].

<!--DIFFERENTIAL PARTICIPATION HIGH COST--> 
Several factors suggest that inequalities observed in legislative lawmaking persist in administrative policymaking. Business interests are the main lobbying participants in most agency rulemaking [@GoldenJPART1998; @YackeeJOP2006]. Past research theorizes that the high costs of commenting on proposed rules are a primary reason for this bias. Knowing when and how to participate as agencies make policy requires an organization to monitor the bureaucracy's rulemaking activities, which can be complex and arcane [@Kerwin2018; @Rossi1997]. Recent research on local administrative policymaking finds that public commenters tend to be unrepresentative of the general public along several common demographic dimensions, including wealth (i.e., homeownership), and that these unrepresentative commenters are more likely to sway the policy decision-making of bureaucratic commissions [@Sahn2024].


<!--DIFFERENTIAL SOPHISTICATION-->
Business interests also tend to submit more technical comments suggesting they can better pay the high participation costs than other participants [@JewellJPART2006]. @Krawiec2013 studied public participation patterns early in the rulemaking process for section 619 of Dodd-Frank (commonly known as the Volcker Rule). She found that comments from financial industry firms were more detailed, complex, and lengthy than those from non-financial firms. 

<!--DIFFERENTIAL POWER-->
Another reason for this bias may be the type of information conveyed in comments. @Acs2023 demonstrate that agency lobbying often conveys "political information," as well as policy information, suggesting that wealthier organizations can better shape regulatory outcomes by signaling their political strength. Thus, even among what might be considered “wealthy” businesses or firms, inequality may still be determinative because more wealthy organizations may be able to better signal their political power to agency officials than less-wealthy ones. 

A related strand of recent research has suggested a mechanism by which traditionally disadvantaged interests may curb business influence during rulemaking: band together to lobby in diverse coalitions [@Dwidar2021APSR; @Dwidar2021PSJ]. These studies also point to continued inequalities because only certain types of coalitions appear to hold policy influence over agency rules, including those with greater resources. This research suggests that inequalities *among* non-profit interest groups demand scholarly attention, in addition to the relative influence of business *versus* non-profit groups.

<!--DIFFERENTIAL INFLUENCE--> 
Research suggests that business interests are influential in rulemaking. Comments from businesses on proposed transportation and labor regulations better predicted policy changes than non-business comments [@YackeeJOP2006]. Similarly, regulatory policy is more likely to change during the U.S. Office of Management and Budget's (OMB) review when more business interests lobby OMB [@Haeder2015]. However, the extent to which this bias toward business interests results from inequalities in the resources for lobbying agencies among businesses and non-business interests has yet to be studied. 
Because scholars measured neither the wealth of interest groups nor the sophistication of comments, we do not know if businesses enjoy greater influence *because* of wealth and sophisticated lobbying. The effect of wealth inequality on administrative policymaking remains an open question. 


<!-- Angelo 2024-06-14: The paragraph below is a good place to emphasize research on participation and *voice* in response to R2 and R3's encouragement. Doesn't Sahn cite that?-->

<!-- Angelo 2024-06-14: By "unrepresentative commenters", is the sentence referring to organizations?  If so, it may be good to clarify that to avoid R3's confusion about being concerned with individual wealth. Steve - I find this ok, so no change.
-->

<!-- Sahn cite needs to be fixed-->


<!-- Angelo 2024-06-14: Paragraph below is a good place to explicitly respond to R1's "why Dodd-Frank" concern.  Perhaps move from the opening "Financial regulation is a particularly important area" into a brief, explicit justification of why Dodd-Frank is an important case study in this important area. -->
<!-- If we're going to justify Dodd-Frank explicitly, we should do it in the intro, not in the middle of a section on theory.  I added a graf in the beginning about why Dodd-Frank -->

## Why Study Financial Regulation

Financial regulation is particularly important for studying inequality because the highly balkanized institutional landscape advantages large financial companies and frustrates citizen and consumer advocates [@SoRelle2023]. Financial deregulation since the 1970s has been a bipartisan project, with Democrats advancing policies despite their relationships with labor unions that have strenuously opposed deregulation [@Barton2022]. Policy entrepreneurs in both parties have also tended to advance legislation that delegates decisions to regulators. Likewise, the Dodd-Frank Act --- which aimed to re-regulate the financial sector --- handed considerable authority to federal financial agencies, including over 300 provisions authorizing new rulemaking [@Copeland2010]. Dodd-Frank sought (and in many cases succeeded) in strengthening consumer protection and regulation of risk to the banking system [@Engel2011]. However, it did little to change the key structural advantages financial corporations enjoy [see, e.g., @Young2012; @Pagliari2017; @Braun2018; @Heilman2017; @James2021].

<!--SHORTEN THIS--> 

<!-- Angelo 2024-06-14: This doesn't respond to a reviewer, but from my read -- it might be good to use an example or examples where there seem to be more clear-cut consequences of influence.
Steve 2024-06-17: what is the point of this paragraph? 
-->

The handful of existing studies that focus on financial rulemaking also present mixed findings about the policy impact of wealth inequality. These studies, which tend to focus on a single agency or rule, raise important questions for future scholarship. For example, there are mixed findings regarding the extent of bias in who is able to participate in financial rulemaking. @Gordon2020 found that a diverse coalition of actors came together to counter the role of larger and more established regulated entities in credit risk retention regulation [see also @Ziegler2016]. Agencies may benefit from the participation of interest groups with different preferences and may try to induce this diversity [@Hirsch2018]. However, @YoungPaglari2017 found that stakeholders beyond affected firms are much less likely to mobilize in the financial sector, especially when a rule is technically complex. 

There are also mixed findings about the relationship between organizational wealth and influence. Studying a sample of Securities and Exchange Commission rules following Dodd-Frank, @Ban2019 concluded that the resources an organization devotes to lobbying appeared to influence the likelihood that the SEC would cite an organization’s name in the preamble to its final rule. In contrast, @Rashin2020 examined thousands of public comments on SEC rules and found that organizational resources did not appear to correlate with a commenter's ability to secure policy changes.


<!-- Angelo 2024-06-14: The paragraph below might be a good place to distinguish influence-based participation from information-based participation conceptually, the latter of which R33 raised a concern about. 
Steve 2024-06-17: I don't think we can show that in the data, so it might raise a red flag if we make a conceptual distinction, but our results can't distinguish between the two
-->



<!-- Steve: change because to since as we have the word "because" twice-->

## The Divergent Interests of More and Less Wealthy Organizations: An Example from Debit Card Fee Regulations


The structural advantages of large, high-resource organizations mean that they often enjoy greater "voice" over policy than smaller, less wealthy, and less powerful organized interests. This power asymmetry may lead to unequal policy impacts, imposing costs or risks on consumers or less wealthy organizations. 

Smaller businesses and less wealthy organizations often have different interests than larger and wealthier organizations. For example, compliance costs may be more burdensome for organizations with fewer resources. At the same time, smaller organizations may pose less systemic risk. For these reasons, policymakers may give smaller or less wealthy organizations exemptions from regulation. However, just because they are not subject to regulation does not mean that smaller businesses and consumer groups do not still have an interest in regulatory policy aimed at shaping markets or preventing financial system collapse.  

To illustrate the diverging interests of wealthy and less wealthy organizations, consider Section 1075 of the Dodd-Frank Act, regulating debit card fees. While it provided an exemption for banks with less than 10 billion dollars of assets, small banks still had an interest in the policy outcomes and sought a voice in the policy process. Smaller banks voiced concerns during rule development that the regulations would have substantial downstream effects on them because they participate in the same market. In particular, comments from smaller banks noted that they depend more on revenue from fees than large banks and do not have the economies of scale enabling them to reduce costs. A price limit that, in theory, applies to the top-tier set of firms could create a market norm. To prevent this, the Independent Community Bankers of America argued that the Federal Reserve should make a rule “requiring the networks to adopt tiered rate schedules (one for exempt institutions at existing market rates and another for regulated institutions).” Federal Reserve officials did not implement this suggestion. One possible explanation is that a few large banks dominated the lobbying. Federal Reserve officials had multiple individual meetings during rule development with large companies, including Bank of America, JP Morgan, Wells Fargo, and larger regional banks. Regulators did not have a single individual meeting with a small bank. They also had fewer meetings with the organized representatives of small banks than with the peak associations dominated by larger banks.

Debit card regulations are but one example of conflicting interests and unequal voice between wealthier and less wealthy organizations. Below, we develop and systematically test hypotheses about wealth inequality in financial rulemaking.


<!-- Angelo 2024-06-14: R1 encouraged greater emphasis on the financial collapse's cost. The last sentence below might be a good place to emphasize diffuse, "macro-public" interests further. 

Steve 2024-06-17: merged the two paragraphs into one. Previously were split at "For example".  We need to fix Hirsch and Shotts 2018 citation.  I am not sure how to implement Angelo's comment, so I am leaving it as is
-->
 



## Wealth Inequality Hypotheses

We investigate the role that wealth inequality may play during the development of agency rules, focusing on two potential biases^[Here we refer to bias in the descriptive, Schattschneider-ian, sense of a system favoring the wealthy [@Schnattschneider1960]]: (1) potential biases in who participates and (2) potential biases in who has influence. We develop several hypotheses about each form of bias.

### Differential Lobbying Participation

 Wealthier organizations, such as businesses, are more likely to participate in agency rulemaking by submitting comments than less wealthy organized interests, such as labor and public interest organizations [@YackeeJOP2006]. 
Wealthy organizations are better able to pay the up-front costs of lobbying. While past research (e.g., YackeeJOP2006) focused on differences in lobbying participation *across* different organization types (i.e., businesses versus public interest groups), we go a step further to also address the effects of wealth differentials *within* organizations of a similar type. For example, we theorize that, even among banks, wealthier banks will participate in rulemaking more often than banks with fewer assets. By comparing similar organizations, we can better isolate whether wealth inequality drives differential lobbying participation in rulemaking.

> *Differential Participation Hypothesis (H1):* Organizations that comment on proposed rules are wealthier than organizations that do not comment on proposed rules.

<!-- Angelo 2024-06-14: The paragraph below is a good place to emphasize participation costs in response to R2.  Currently, this paragraph speaks more to participatory incentives rather than costs. 
Steve 2023-06-17: Not sure this is the place to do that; here, we just want to list hypotheses and where they came from
-->

Differential participation may also be driven by the concentration of the costs and benefits of regulatory lobbying [see broadly, @Lowi1964; @Olson1965; @Wilson1989]. For-profit organizations---especially regulated firms---tend to have concentrated stakes in regulations. Wealthy profit-seekers have especially strong incentives and the ability to lobby in rulemaking [@Libgober2020; @LibgoberQJPS2020]. Thus, we expect for-profit businesses and the industry associations that represent them to participate more often than other non-profit organizations. 

> *Profit-motivated Participation Hypothesis (H2):* Profit-seeking organizations and industry associations are more likely to comment than other non-profit organizations.

<!-- Angelo 2024-06-14: The paragraph below is maybe a good place to "address head-on" distinct interests of different-sized banks in response to R3's comment about small vs. big bank alignment. 
Steve 2024-06-17: I am not sure we can do that since we don't have a measure of banks' distinct interests. 
-->


### Differential Lobbying Success

<!---FIXME: THIS PARAGRAPH REPEATS A BIT FROM THE THEORY SECTION-->

<!-- Angelo 2024-06-14: Should the first sentence say "differential lobbying success" rather than benefit? -->

<!-- Angelo 2024-06-14: Signpost what "shifting the content" exactly means in response to R2. -->
<!-- Steve 2024-07-14: Do we mean changing the content? -->

Existing research hints at a differential lobbying benefit attached to wealth during rulemaking. For instance, @Haeder2015 find more policy change during rulemaking when business interests are more active than other types of organizations, such as public interest groups. Yet, such research does not provide a clean test of wealth inequality. After all, some businesses are large while some are small; some non-profits hold major financial assets while others do not. We thus seek to understand whether wealth inequality drives lobbying influence during rulemaking and whether wealthier organizations see greater lobbying success during rulemaking.

<!-- Steve 06-17-2024 changed shifting to changing. Here's the original hypothesis in case this change is the wrong shift  *Differential Lobbying Success Hypothesis (H3):* Wealthier organizations are more successful in changing the content of agency rules.-->

> *Differential Lobbying Success Hypothesis (H3):* Wealthier organizations are more successful in changing the content of agency rules.

<!-- Angelo 2024-06-14: The paragraph below can expand on what "sophisticated" means in response to R2. "Way forward" also references an added appendix figure on "Top 10 most sophisticated comments". 
Steve—I added "technical and legal" in front of sophistication to give greater color to sophistication. I think this is the wrong place to reference an appendix. We can do that when discussing the results. I think we want to keep this section relatively clean as just a hypothesis with minimal justification.
-->

Research suggests wealthier organizations are more influential because they are disproportionately able to marshal the technical expertise necessary to write sophisticated comments for rules [@WagnerALR2011].  Moreover, agency officials pay greater attention to abstract and technical arguments, such as those in comments from business organizations, while often minimizing the moral and personal arguments found in less sophisticated comments from individuals [@JewellJPART2006; @Mendelson2011]. Additionally, non-industry comments often lack the specificity and detail that agencies need to change policy [@Krawiec2013]. Comments from wealthier organizations may thus provide useful information to regulators and thus subsidize agencies as they seek to create technical and legally sophisticated regulations (see broadly, @Hall2006; @Schnakenberg2024). Consequently, we hypothesize that wealthier entities utilize their resources to produce comments with greater technical and legal sophistication than less well-resourced groups and that these more sophisticated comments will be more impactful.

> *Differential Sophistication Hypothesis (H4):* Wealthier organizations use more technical and sophisticated language when commenting on proposed rules.

> *Dividends of Sophistication Hypothesis (H5):* Comments from wealthier organizations are more successful in affecting the content of agency rules because of comment sophistication.  

<!--SUSAN DELETED: 
Together, these hypotheses assess wealth inequality's role in creating biases in who participates and who has influence: (1) wealthy organizations are better able to participate, and (2) even when the less wealthy participate, wealthy organizations are more likely to have their demands met. The hypotheses also identify a major theorized mechanism of lobbying influence: that wealthy organizations achieve regulatory policy influence via the legal and technical sophistication of their comments on proposed rules.
--> 

# Data and Methods {#data}

To assess the extent of inequality in financial rulemaking, we assembled data on draft and final rules, comments on those rules, the wealth of various organizations, political spending, and lobbying spending. Data sources included the Federal Register, Regulations.gov, Wharton Research Data Services, the Center for Responsive Politics, Federal Financial Institutions Examinations Council, and the Internal Revenue Service. Using comment text and metadata, we link comments to the organizations that submitted them and metadata about each organization's resources. 

## Agency Rules & Public Comments

<!-- Steve's pedantic note, isn't the Federal Reserve the FRB, not FRS?-->

From the Federal Register, we collected the text of all rules promulgated under authorities granted by Dodd-Frank between its enactment on July 20, 2010, and July 8, 2018, by the seven primary financial regulators tasked with writing rules under the Dodd-Frank Act: the Consumer Financial Protection Bureau (CFPB), the Commodity Futures Trading Commission (CFTC), the Federal Deposit Insurance Corporation (FDIC), Federal Reserve (FRS), National Credit Union Administration (NCUA), the Office of the Comptroller of the Currency (OCC), the Securities and Exchange Commission (SEC). We also collected all public comments and comment metadata on these rules from each agency's website or Regulations.gov. In doing so, we gathered key information, including the name of the entity submitting the comment and the comment submission date. We also collected the text of all comments from comment submission forms and file attachments. These data include `r n_comments` comments on `r n_rules` separate rulemaking dockets, covering `r n_actions` regulatory actions issued by one or more of these seven agencies.^[The law firm Davis Polk LLP maintains a list of Dodd-Frank-related rules. Each rule in our sample may be considered a set of connected regulatory actions, generally including a proposed and final rule connected by a Regulation Identifier Number (RIN). We count jointly-issued rules as two rules because agencies collected comments separately.]

@fig-actions shows significant variation in regulatory activity across these agencies. The largest agency in our sample by regulatory volume is the CFPB, while the smallest is NCUA. The figure also shows considerable variation in the range of regulatory actions, including advanced notices of proposed rulemaking (ANPRMs), proposed rules (also called "Notice of Proposed Rulemakings" or NPRMs), interim final rules, and final rules. 

<!-- (Chris 2022-07-23: I agree with Angelo. Maybe this is a generic table for the next version.) 
DJL: figures often need to be self-contained for journals-->

```{r}
#| label: fig-actions
#| out-width: 100%
#| fig-cap: "Dodd-Frank Act Implementing Actions by Agency. Counts of regulatory acts by agency and by year. Regulatory acts include ANPRM, NPRM, Interim Rule, and Final Rules. Note that one complete rulemaking process typically has two and sometimes more associated actions."
knitr::include_graphics(here::here("figs/actions-1.png"))
```

## The Wealth of Organizations

Our wealth inequality hypotheses focus on the lobbying behavior of organizations during rulemaking. As a result, we developed a suite of new measurement and analytic tools designed to capture measures of wealth for organizations and then linked these measures to lobbying activities. The final dataset is the subset of all comments on Dodd-Frank rules that match an organization with some form of wealth data. This dataset allows us to compare the wealth of organizations that commented on financial rules to the wealth of similar organizations that did not comment on these rules.

<!--TODO SUSAN SUGGESTED DELETING, BUT I THINK THIS PARAGRAPH IS NEEDED SOMEWHERE-->

We created the dataset by first collecting and digitizing the texts of all public comments on Dodd-Frank rules. We then extracted entity names and matched them to organizations in databases that yield information on wealth. No single database provides information on wealth for all types of organizations. We thus cast a wide net and identified multiple databases of organizations that might participate in financial rulemaking. The databases below contain nearly 500,000 banks, credit unions, publicly traded companies, and non-profits. We identify `r n_assets` comments submitted by organizations that appear in one or more of the databases described below. These databases are:

1. Financial data, including market capitalization, for all publicly traded companies listed on U.S. exchanges during our analysis time frame from the Wharton Research Data Service’s Compustat database.  Market capitalization is a common measure of firm size.

<!-- Angelo 2024-06-14: Responding to R1 -- Justify using different wealth measures for items 2-4. 
Steve 2024-06-17: Added sentence "Market capitalization is a common measure of firm size."
-->

<!-- Angelo 2024-06-14: Responding to R1 -- clarify when this data is accessed/what time frame or years they represent. -->

2. Separately, market capitalization for all corporations that filed disclosures with the SEC and thus listed in the SEC’s Central Index Key (CIK) database.

3. Assets under management for all bank and bank-like entities covered by the FDIC (as reported to the FDIC).

4. Assets under management for all U.S. credit unions from consolidated call reports published by the NCUA.

5. Total assets and annual revenue for all non-profit organizations as reported by Internal Revenue Service 990 forms.<!---TODO: VOLUNTEERS STEVE - HAVEN'T WE DONE THIS ALREADY?-->

6. Political Action Committee (PAC) donations from all organizations that file campaign disclosure reports with the Federal Election Commission, as compiled by the Center for Responsive Politics. These reports allow us to calculate each organization's average annual PAC contributions.

7. Lobbying expenditures, as compiled by the Center for Responsive Politics from Lobbying Disclosure Act reports. We then calculate the average annual lobbying expenditures for each organization.

Next, we used an iterative matching procedure to match organizations in these six databases to those organizations that commented on one or more Dodd-Frank rules. This step took considerable innovation because the names organizations use to submit comments and their names in various databases often differ. Our matching procedure involved several steps. We first identified comments that were likely from an organization, excluding those that were from individuals.^[Our study design purposefully sets aside comments from individuals, most of which are form comments, because previous research establishes that form comments are almost always part of a larger “campaign” orchestrated by an organization, and that the organizations that mobilize mass comment campaigns also submit technical comments on the same rules [@JudgeLord2021]. Our data include these technical comments.] 
We then linked these comments to the organization with the best matching name or to no organization when our matching algorithm did not identify a high-probability match in any of the databases. We spot-checked our processes for false positive matches by inspecting organizations that matched many comments and false negatives by inspecting especially long or sophisticated comments that did not match a known organization. We improved the matching algorithm through dozens of iterations and post hoc corrections, including hand-validating matches for over 30,000 comments, including all comments from entities that submitted ten or more comments.

These procedures resulted in a dataset of `r n_orgs` distinct organizations that submitted  `r n_assets` unique comments on one or more Dodd-Frank rules. Below, we use these data to compare the wealth of commenting organizations to the `r n_orgs_total` similar organizations in one of the above wealth databases that did not comment on a Dodd-Frank regulation.

```{r}
#| label: fig-org-types
#| fig-cap: "Number of Organizations by Type and Agency to which they Commented. These counts reflect only those comments matched to an organization. The lowest match rate (at the CFPB) still exceeds 20%. See the appendix."
#| out.width: 80%

# Agencies include the Consumer Financial Protection Bureau (CFPB), Commodity Futures Trading Commission (CFTC), Federal Reserve (FRS), National Credit Union Administration (NCUA), and Securities and Exchange Commission (SEC).
knitr::include_graphics(here::here("figs/org_count_type-2.png"))
```

```{r}
#| label: fig-comment-types
#| fig-cap: "Number of Comments by Organization Type and Agency. Counts of comments from organizations that have submitted comments to each financial regulator by organizational type. These counts reflect only those comments matched to an organization. Details about the match rate are in the appendix."
#| out-width: 80%
knitr::include_graphics(here::here("figs/org_count_type-1.png"))
```

<!--Steve 06-17-24 Since we have agency acronyms above, I changed this paragraph to just acronyms 
Consumer Financial Protection Bureau (CFPB), Commodity Futures Trading Commission (CFTC), Federal Deposit Insurance Corporation (FDIC), Federal Reserve (FRS), National Credit Union Administration (NCUA), Office of the Comptroller of the Currency (OCC), and Securities and Exchange Commission (SEC)
-->

@fig-org-types shows the number of unique commenting organizations matched to each database by the agency or agencies to which they submitted comments.<!--^[These counts reflect only those comments matched to an organization. Descriptives and details about the matching process are in the appendix.]-->
<!--FACT CHECK
Across all agencies except for the FRS, most commenting organizations were non-profits. The next most common type of commenter was federally-insured (FDIC-insured) banks (hereafter "banks"). Organizations that filed with the SEC and donors to PACs were less common. 
--> 
@fig-comment-types shows the number of comments submitted to each agency by these organizations. <!--There was considerable variation in the number of comments from organizations per rule. We matched at least two organizations to asset data for all rules that received more than 25 comments. <!--The agency with the largest median number of comments from organizations was the CFPB at 21 comments per rule. --> 

## Profit Motives 

We use an organization's legal incorporation status to infer profit motivations. Some 501(c)(3) non-profits, such as industry associations, are formed to advance narrow private interests. While our data on non-profits does not perfectly capture the extent to which organizations advance public or private interests, we classify an organization as representing "profit-seeking" interests if it is incorporated as either a for-profit company or an industry association. 

We also leverage variation in types of banking institutions to infer profit motivations. Commercial banks' legal and organizational structures make them more profit-oriented than credit unions and savings associations. Commercial banks are often large corporations managed by a board selected by shareholders and tend to serve corporations and wealthier, profit-motivated clients. In contrast, savings associations are chartered with the narrow purpose of providing affordable residential mortgages. Both types of banks may hold significant assets, but they have very different clients. 

## Comment Sophistication {#methods-sophistication}

We measure comment sophistication by counting the technical terms in each comment. To capture technical sophistication with respect to the use of finance and banking jargon, we use the Oxford Dictionary of Finance and Banking, which includes 5,260 finance and banking terms. To measure legal sophistication, we count legal citations (for details, see the appendix). When an organization submits a comment with multiple attachments, we measure sophistication by summing up the technical terms and legal citations across all submitted documents. This approach follows the intuition that attachments with additional technical language reflect additional sophistication.
<!--DJL LIKES THIS SENTENCE, BUT SUSAN CUT
For example, the most sophisticated organizations often submit a cover letter, a marked-up version of the proposed rules, and studies supporting their arguments.
--> 

## Lobbying Success {#methods-success}

After reviewing an agency's proposed rule, organizations typically use their comments to articulate the policy changes they want the agency to make in the final rule.^[Final rules include preamble and rule text. We include changes to both in the lobbying success measure.] To approximate the extent to which commenters’ requested policy changes are made, we measure the overlap between the text of each organization’s comment and the text added to the final rule. Our measure of lobbying success follows the intuition that an organization whose comment text is repeated by the agency in the text of the final rule is more influential in shifting regulatory content in its desired direction than an organization whose comment text is not reflected in changes in the final rule. Stated differently, more text reuse ---from comment to final rule---suggests greater lobbying success.

To construct this measure, we first link proposed rules to final rules by their Docket or Regulatory Identification Numbers. We then match comments to proposed rules by publication date. 
We then tokenize each draft and final rule and comment in groups of ten words. Ten-word phrases are long enough that they rarely co-occur by chance and are thus a well-validated measure of textual similarity [@Wilkerson2015; @Casas2019; @Rashin2020]. <!--@JudgeLord2017-->
Finally, we count the number of words in phrases of ten or more that appear in the comment and final rule but do not appear in the draft rule.^[
We exclude any text from the agency’s proposed rule in this calculation to ensure that we do not include phrases in an organization’s comment that simply quote the proposed rule. Excluding the proposed rule text in our calculations also guards against the possibility that an organization’s decision to include particular phrases in their comments is endogenous. By excluding the text of the proposed rule in our lobbying success measure, we remove the phrases and text that are most likely to be naturally repeated.]
For rules with multiple final rules, we take the sum of the comment's alignment with both final rules. When an organization submits a comment with multiple attachments, we include the highest-scoring document as the primary comment. This choice aligns with typical commenter behavior because organizations that submit multiple attachments almost always have a primary comment articulating their lobbying demands.

Our measure of lobbying success captures the idea that organizations desire policy change in line with their lobbying demands [@Mahoney2007JPP]. It captures “success” by measuring the alignment between specific requests made in an organization’s comment and subsequent policy changes. However, lobbying success, as we measure it, does not necessarily prove causality. For example, the organization’s comment and the agency may have copied the repeated text from a third source. Thus, we cannot definitively say that the comment caused the policy change, but we can say whether or not the organization achieved its stated lobbying objectives.

<!--THE CORRELATION BETWEEN SOPHISTICATION AND SUCCESS: This figure is here rather than in the results section to facilitate discussion of the data, but I'm open to suggestions--> 

```{r}
#| label: fig-efficacyXsophistication
#| fig-cap: "Lobbying Success by Comment Sophistication. Each point represents a comment. The horizontal axis bins comments by the order of magnitude of the number of technical terms per comment."
#| out-width: 70%
knitr::include_graphics(here::here("figs/boxplot-efficacyXsophistication-2.png"))
```

Descriptively, our measures of lobbying sophistication and lobbying success are highly correlated. Our measure of commenter lobbying success increases with the wealth of the commenting organization. @fig-efficacyXsophistication shows that the number of words from the comment added to the final rule (the y-axis) correlates with the number of technical words in the comment (the x-axis, binned on a log scale). Box plots show the middle two (25%-75%) quartiles and whiskers extending to 1.5 times the inter-quartile range (the distance between the first and third quartiles).

In Appendix @sec-efficacy, we validate our measure through case studies of the comments that scored highest on our measure of lobbying success. In each case, the text reuse measure picks up text that shows the agency officials taking commenter concerns seriously as they revised regulations, often leading to substantive policy change in the final text of the rule as well as significant engagement with commenters' arguments over statutory interpretation and legislative intent in the preamble to the final rule.^[Case C in particular shows that our measure picks up the case of policy deletion from proposed to final rule, which is difficult for text reuse approaches to capture.]  The top five highest-scoring commenters include: 

- a comment from a consumer rights advocacy organization, Occupy the SEC, that proposed closing an insider-trading loophole by applying regulations not only to banks’ investments but also to the investments of their top employees. The SEC modified the rule to extend regulations to employees exactly as the commenter suggested (Case A in Appendix @sec-efficacy);
- a comment from the Securities Industry and Financial Markets Association (SIFMA) that directed the agency to harmonize the regulation of swaps across two agencies, followed by the agency’s agreement to harmonize treatment, in which the agency expressly used the SIFMA comment’s phrase in the final rule text (Case B);
- a comment from Standard and Poor’s calling for the SEC to drop a requirement for securities raters to track whether debts are paid off, particularly in the case of securities that had had their ratings withdrawn, followed by the SEC’s decision to drop this requirement and the SEC’s specific citation of the comment in justifying its deletion (Case C).

@fig-efficacyXsophistication highlights the very highest-scoring comment: a comment to the SEC prepared by the law firm White & Case, LLP for the U.S. Chamber of Commerce, Americans for Limited Government, Ryder Systems, Inc., the Financial Services Institute, Inc., and Verizon. This highly-sophisticated comment included a 19-page cover letter with many technical citations underscoring the Chamber's "very serious concerns on the impact [that the rule's] whistleblower requirements will have on... companies' responsibilities to act in the best interests of their shareholders." This comment also included a marked-up draft of the SEC's proposed rule, suggesting specific changes, several of which were adopted by the SEC.^[See more details on this and other comments that score high on our measure of lobbying success in Appendix @sec-efficacy.] 

Other comments with high lobbying success scores include an 84-page comment from Standard & Poor's Global Ratings credit rating agency to the SEC, a 59-page comment from the Futures Industry Association to the CFTC, and several marked-up versions of proposed SEC rules from investment companies.
Overall, @fig-efficacyXsophistication shows a positive correlation between the number of technical banking terms in a comment and the amount of text it shares with the final rule. 
Using these data (comments, their sophistication, and their lobbying success), the following section assesses our hypotheses about the relationship between wealth, political participation, lobbying sophistication, and lobbying success. Notably, section @sec-mediation further explores the correlation between sophistication and lobbying success by assessing comment sophistication as a mediator in the relationship between wealth and success.


## Methods {#methods}

We assess our hypotheses about the relationship between wealth inequality and policy influence using descriptive statistics, regression, and causal mediation analysis. <!--We assess our hypotheses about the relationship between wealth inequality and policy influence using descriptive and statistical analyses, including t-tests, multivariate Logit, OLS, and Poisson models, and causal mediation analyses. --> 
We use Welch t-tests to assess differences between commenters and non-commenters (H1) and for-profit and non-profit organizations (H2).
We use regression analyses to assess whether wealth predicts various outcomes of interest. We employ logit regression to model the binary outcome of commenting as a function of wealth (H1) and organization type (while controlling for wealth). We model differences between non-profits with for-profits overall and, separately, between for-profit and non-profit types of banks (H2). The resulting model coefficients allow us to estimate how changes in an organization’s assets and organizational form produce changes in the odds that the organization will comment on a rule.
We use Poisson regression to model the count of words from a comment added to the final rule (H3) and the number of technical terms used in a comment (H4) as a function of wealth.
Finally, we employ causal mediation analysis to assess the extent to which campaign donations, lobbying expenditures, and comment sophistication mediate the relationship between wealth and lobbying success (H5).     

<!--SUSAN DELETED: 
Together, these hypotheses assess the role of wealth inequality in creating biases both in who participates and who has influence: (1) wealthy organizations are better able to participate, and (2) even when the less wealthy participate, wealthy organizations are more likely to have their demands met. The hypotheses also identify a major theorized mechanism of lobbying influence: that wealthy organizations achieve regulatory policy influence via the legal and technical sophistication of their comments on proposed rules.
--> 

   
# Results

In this section, we investigate each of our hypotheses in turn. First, we focus on inequalities in which organizations participate in financial rulemaking. Second, we focus on inequalities in lobbying influence among organizations that participate. In doing so, we test our hypotheses about wealth, access, and influence in the policy process using two types of variation: (1) variation among organizations that did comment and similar organizations that did not comment on rules implementing the Dodd-Frank Act and (2) variation in lobbying sophistication and success among organizations that did comment.

## Wealth Inequality in Lobbying Participation

First, we compare levels of resources among commenting organizations and similar organizations that did not comment.  Because our data included the full population of similar organizations (e.g., all banks or all non-profits), only some of which did submit comments, we can assess the relationship between wealth inequality and participation in the policy process.

### Wealthier organizations are more likely to participate

The *Differential Participation Hypothesis (H1)* posits that organizations that participate in policymaking are wealthier than organizations that do not. Overall, we find strong support for this hypothesis for all types of organizations. @fig-commenters-noncommenters shows distributions of wealth for organizations that commented on any Dodd-Frank rule and those that did not. The x-axes show measures of wealth (assets or market capitalization). Because the x-axes are logged, small differences on the right side of these distributions represent large substantive differences in wealth. Statistical tests show that differences in the wealth of commenters and non-commenters are significant at the 0.05 level for all types of organizations. Logistic regression results (@tbl-mp-all-table) likewise show that the odds of commenting increase with an organization's wealth.

```{r}
#| label: fig-commenters-noncommenters
#| fig-cap: "Financial Resources of Organizations that Did and Did Not Comment"
#| fig-subcap: 
#| - "Non-profits"
#| - "Credit Unions"
#| - "Industry Associations"
#| - "Banks"
#| - "Publicly-traded Companies"
#| layout-ncol: 2
knitr::include_graphics(here::here(c(
  "figs/nonprofit-density-1.png", 
  "figs/creditunion-density-1.png", 
  "figs/ind-assoc-density-1.png",
  "figs/FDIC-density-select-1.png",
  "figs/compustat-density-1.png")))
```

**Non-profits.** Panel (a) in @fig-commenters-noncommenters shows that non-profits that comment on proposed financial regulations tend to be significantly better resourced. The average assets of non-profits participating in Dodd-Frank rulemaking were about eleven times larger than non-profits that did not participate; the average assets of non-profits that did comment had approximately \$98 million, whereas the average assets of non-profits that did not comment was about \$9 million.

**Credit unions.** Similarly, panel (b) in @fig-commenters-noncommenters shows that credit unions that comment on proposed financial regulations have more assets, on average, than those that do not participate. The average credit union that did not comment has about \$183 million in assets, whereas the average credit union that did comment had about \$675 million. <!--That is, the average commenting credit union is more than three times as large as the average credit union that did not comment.-->

**Industry associations.** Industry associations that participate in rulemaking also tend to have more resources, almost five times more, than those that do not. Panel (c) in @fig-commenters-noncommenters shows that the average non-commenting industry association had about \$2 million in assets, whereas the average commenting industry association had about \$9 million.

**Banks.** Panel (d) in @fig-commenters-noncommenters shows that, on average, banks that comment on proposed financial regulations are better resourced than we would expect from a random sample of banks. Banks that participated in financial rulemaking had over three times the average assets of banks that did not participate. 

**Publicly-traded companies.** Panel (e) in @fig-commenters-noncommenters shows similar distributions over market capitalization (the total value of a company's stock) among publicly-traded companies. Companies that comment on proposed financial regulations are wealthier than those that do not, as measured by the total value of their stock. The median capitalization of companies that commented was about double that of the median company that did not comment. Logit models predicting the odds of commenting (the first column in @tbl-mp-all-table and @fig-mpcompustat) show the same result: companies with higher capitalization are more likely to comment.

### Organizations that spend more on political campaigns are more likely to comment

```{r}
#| label: fig-opensecrets-density
#| fig-cap: "Political Spending of Organizations that Did and Did Not Comment"
#| fig-subcap: 
#| - "Political campaign donations"
#| - "Disclosed lobbying spending"
#| layout-ncol: 2
knitr::include_graphics(here::here(c(
  "figs/opensecrets-density-1.png",
  "figs/opensecrets-lobbying-density-1.png")))
```

Panel (a) of @fig-opensecrets-density shows that organizations that commented on Dodd-Frank rules also donate more to political campaigns via PACs than organizational PAC donors that did not comment, further supporting the *Differential Participation* Hypothesis (H1). Among organizations that donate to PACs, the average campaign spending per two-year cycle was \$54,000 for those that did not comment, while the average for those that did comment on a Dodd-Frank rule was \$85,000 (p < 0.01). Logit models in the Appendix also show that PAC donations strongly predict commenting behavior. 
Panel (b) of @fig-opensecrets-density shows that organizations that commented on Dodd-Frank rules may also spend more on traditional lobbying than those that did not comment, but these differences are not significant at the 0.05 level. 

### Profit-driven organizations are more likely to comment than non-profits {#sec-results-profit}

The *Profit-Motivated Participation* Hypothesis (H2) posited that for-profit organizations are more likely to participate in rulemaking than non-profit organizations. We find strong support for this hypothesis overall (i.e., comparing for-profit companies with non-profits) and when we compare for-profit banks (commercial banks) to those that are non-profit (credit unions and savings associations). 12% of commercial banks commented on Dodd-Frank rules. In contrast, only 3% of non-profit savings associations, 2% of non-profit credit unions, and 0.2% of other non-profits commented. Commercial banks were six times more likely to comment on a Dodd-Frank rule than credit unions and 60 times more likely to comment than the average non-profit organization. 
 
Banks were more likely to comment than credit unions and other types of non-profits, even when controlling for differences in assets. @tbl-mp-all-table shows the results of logit models predicting the log odds of commenting by organization type (bank, credit union, industry association, or other non-profit organization) and total assets. Based on model 2 from @tbl-mp-all-table, @fig-mp-all-predict shows that the predicted probability of commenting increases as all types of organizations gain more assets. Wealthier banks, credit unions, and other non-profits are more likely to comment than less wealthy ones, further supporting the *Differential Participation* Hypothesis (H1). 

The relationship between wealth and commenting behavior is strongest for industry associations, further supporting the *Profit-Motivated Participation* Hypothesis (H2). Compared to other non-profits, those representing profit-seeking businesses (industry associations) are much more likely to deploy resources toward influencing public policy as they gain assets (see @fig-mp-all-predict and appendix @tbl-mp-all-table2).
 
```{r}
#| label: fig-mp-all-predict
#| fig-cap: "Predicted Probability of Participating in Dodd-Frank Rulemaking by Assets and Type of Organization"
#| out-width: 65%
knitr::include_graphics(here::here("figs/mp-all-predict-log-1.png"))
```

```{r}
#| label: tbl-mp-all-table
#| tbl-cap: "Log Odds of Commenting on Any Dodd-Frank Rule"

# Preferred models of commenting by wealth measures
load(here::here("models", "mpCompustat.Rdata"))
mpCompustat <- models[[2]]

load(here::here("models", "mpAll.Rdata"))
mpAll <- models[[4]]

load(here::here("models", "mpFDIC2.Rdata"))
mpFDIC <- models[[4]]

models <- list(mpCompustat, mpAll, mpFDIC)

rows <- tibble(
  term = c("Dependent Variable"),
  `1` = "Commented",
  `2` = "Commented",
  `3` = "Commented"
)

attr(rows, 'position') <- c(0)

modelsummary(models, notes = "Reference catagory = Banks for model 2, commercial banks for model 3")
```

To further test the *Profit-motivated Participation* Hypothesis (H2), we subset our data to banks and estimate the odds of commenting across different types of banks. @fig-mp-FDIC2 (based on model 3 from @tbl-mp-all-table) shows that for-profit banks (i.e., commercial banks) are significantly more likely to comment than non-commercial banks (i.e., non-profit savings associations and savings associations), further supporting the link between profit motives and lobbying activity. For example, among banks with a mean asset amount of approximately 1 billion USD, model 3 predicts a commercial bank to have a 29% probability of commenting on proposed financial regulation. Meanwhile, a non-profit savings association with the same assets has only a 12% probability of commenting. Moreover, assets remain a significant predictor of whether an organization comments even controlling for differences in the type of bank institution, providing additional evidence for the *Differential Participation* Hypothesis (H1). 

```{r}
#| label: fig-mp-FDIC2
#| fig-cap: Predicted Probability of Participating in Dodd-Frank Rulemaking by Type of Bank
#| out-width: 65%
#TODO CREDIT UNIONS SHOULD BE IN THIS model 
knitr::include_graphics(here::here("figs/mp-FDIC2-predict-1.png"))
```

Finally, we estimate the probability of commenting among publicly traded companies based on their market capitalization. This alternative measure of corporate wealth yields the same conclusion: companies with greater wealth are more likely to comment. <!--This is true even among large publicly-traded companies.--> @fig-mpcompustat (based on the results shown in model 1 of  @tbl-mp-all-table) shows that the predicted probability of commenting increases 57% (from about 7% to about 11%) as a company goes from having a market capitalization of one billion to 100 billion. 


```{r}
#| label: fig-mpcompustat
#| fig-cap: "Predicted Probability of Participating in Dodd-Frank Rulemaking by Market Capitalization"
#| out-width: 35%
knitr::include_graphics(here::here(                 "figs/mp-compustat-predict-1.png"))
```



The main takeaway thus far is that resources correlate with commenting behavior; wealthier organizations are more likely to participate in regulatory lobbying than less wealthy organizations. If representation is largely about who shows up to participate when the government makes important policy decisions, companies with high market capitalization, organizations that give more to political campaigns, and banks, credit unions, and non-profits with more assets are better represented than similar organizations with lower market capitalization, less political spending, and fewer assets to deploy. Both within and across different types of organizations, wealthier organizations have disprportionate "voice" in policymaking. 


## Wealth Inequality Among Organizations that Lobby

We now investigate wealth inequalities within the subset of organizations that do participate in rulemaking.
By focusing on variation *among* organizations that commented on at least one Dodd-Frank rule, we can have even more confidence that we are comparing similar organizations.

### Wealthier commenters have greater lobbying success 

Our final three hypotheses focus on the association between wealth inequality and lobbying success. The *Differential Lobbying Success* Hypothesis (H3)  posits that wealthy organizations will be more successful in regulatory lobbying. The remaining two hypotheses address why this pattern may emerge. For instance, is it because wealthy organizations spend more on political campaigns and lobbying targeting Congress? Or because they deploy more legal and technical expertise when they comment on proposed rules?

<!--model?--> 
@fig-repeated-text-assets provides descriptive support for Hypothesis 4. For banks and other companies, we see a positive correlation between an organization’s wealth and its lobbying success. The pattern is less clear for non-profits. In other words, wealthier companies enjoy more lobbying success than similar---but less wealthy---companies. The y-axes of plots in  @fig-repeated-text-assets indicate the number of words that appear in 10-word phrases in both an organization's comment and the final rule (but were not present in the proposed rule). The x-axes of each plot in @fig-repeated-text-assets represent indicators of wealth binned on a log scale. 
Differences in means capture the extent to which wealth is correlated with lobbying success (as measured by the amount of text added to an agency’s final policy documents containing exact phrases used by or suggested by an organization’s comment).

```{r}
#| label: fig-repeated-text-assets
#| fig-cap: "Amount of Text Repeated in Final Rules by Commenter Resources. Each point represents a comment. The horizontal axis bins comments by the order of magnitude of the assets of the commenting organization in millions of dollars. Axis labels represent the lower bound of the bin."
#| layout-ncol: 2
#| fig-subcap: 
#| - "Non-profits"
#| - "Credit Unions"
#| - "Industry Associations"
#| - "Banks"
#| - "Publicly-traded Companies"
knitr::include_graphics(here::here(c(
  "figs/boxplot-assets-efficacy-2.png",
  "figs/boxplot-assets-efficacy-4.png",
  "figs/boxplot-assets-efficacy-6.png",
  "figs/boxplot-assets-efficacy-10.png",
  "figs/boxplot-assets-efficacy-8.png")))
#TODO ADD CORRELATION COEFFICIENTS TO THESE PLOTS 
##| - "501(c)(3) Nonprofits"
##| - "501(c)(1) Credit Unions"
##| - "501(c)(6) Industry Associations"
##| - "FDIC-Insured Banks"
```

### Wealthier companies are more sophisticated at lobbying

We now consider possible explanations for the positive relationship between wealth and lobbying success. 
The *Differential Sophistication* Hypothesis  (H4) suggests that wealthier organizations submit more sophisticated comments than less wealthy entities. @fig-assets-terms provides evidence of just such a relationship. It shows that the comments from wealthier organizations tend to include more technical language specific to finance and banking. This pattern is especially strong for banks and publicly traded companies. For example, the average comment from a company with market capitalization over \$10 billion contained around 1000 technical terms (bins \$10-100 Billion and >\$100 Billion), while companies with lower market capitalization tended to submit less sophisticated comments (closer to 100 technical terms). 

```{r}
#| label: fig-assets-terms
#| fig-cap: "Amount of Technical Language by Assets. Each point represents a comment. The horizontal axis bins comments by orders of magnitude of the assets of the commenting organization in millions of dollars. Axis labels represent the lower bound of the bin."
#| layout-ncol: 2
#| fig-subcap: 
#| - "Non-profits"
#| - "Credit Unions"
#| - "Industry Associations"
#| - "Banks"
#| - "Publicly-traded Companies"
knitr::include_graphics(here::here(c(
  ## Nonprofits
  "figs/boxplot-assets-tech-2.png",
  ## Credit unions 
  "figs/boxplot-assets-tech-4.png",
  ## Industry Associations 
  "figs/boxplot-assets-tech-6.png",
  ## FDIC-insured banks 
  "figs/boxplot-assets-tech-10.png",
  ## Market cap (publicly-traded companies)
  "figs/boxplot-assets-tech-8.png")))
```



### More sophisticated comments correlate with greater lobbying success 

In the *Dividends of Sophistication* Hypothesis (H5), we theorize that comments from wealthier organizations are more successful in shifting the content of financial rules *because* wealthier organizations submit more sophisticated comments.
We investigate this proposed mechanism for unequal influence by assessing the relationship between legal and technical sophistication and lobbying success.

We find that comments that use more sophisticated technical language are more likely to contain text that was added to the final rule. @tbl-efficacy-sophistication shows estimates of lobbying success from regression models where the key predictor variable is the number of technical terms or legal citations in a comment. Both models suggest a statistically significant relationship. Substantively, using ten additional technical finance or banking terms in an organization's comment is associated with an additional word from that comment being added to the text of the final rule. Each legal citation in a comment is associated with about 14 additional words from that comment added to the final rule.^[By "additional words," we mean words from a ten-word phrase that appears in the organization's comment and in the final rule but was not present in the draft rule.]

```{r}
#| label: tbl-efficacy-sophistication
#| tbl-cap: "OLS Models of Lobbying Success by Comment Sophistication"
#| out-width: 80%
#knitr::include_graphics("figs/mes-1.png")
#knitr::include_graphics("figs/mes-2.png")

load(here::here("models", "mes.Rdata"))

#FIXME in efficacy.rmd 
rows[,2] <- "Lobbying Success"
rows[,3] <- "Lobbying Success"

modelsummary(models,  notes = "Poisson regression yields similar results
             (see appendix)") #%>% kableExtra::kable_styling(latex_options="scale_down")
```

### Legal and technical sophistication explains the lobbying success of wealthy companies {#sec-mediation} 

Finally, to further evaluate the *Dividends of Sophistication* Hypothesis (H5), we use mediation analysis to examine the extent to which the sophistication of the comments may explain the relationship between wealth and lobbying success. Mediation analysis can take various forms, from structural equation models and sequential testing of additional variables to more modern techniques rooted in causal inference [@mackinnon2012introduction]. Causal analysis of mediator variables requires strong assumptions [@imai2011unpacking] that may not be satisfied in complex observational settings such as this study. Still, we consider observational analysis of mediation hypotheses worthwhile for the same reason we regard other non-identified, descriptive studies as valuable: they permit the repeated, principled testing of hypotheses that can inform important scholarly and policy debates where randomized studies are problematic or impossible.  

We follow the estimation strategy of the more modern literature in causal inference [@tingley2014mediation]. In this literature, causal mediation analysis aims to decompose an Average Treatment Effect --- also called the Total Effect --- into its parts. This Total Effect is the sum of the direct effect and indirect effect(s) through mediator(s). <!--A mediator is often thought of as a mechanism by which a treatment transforms into an outcome.--> An effect may have multiple, potentially interrelated mediators. 

Here, we focus on the publicly-traded companies that submitted comments to our Dodd-Frank rules. Because the correlation between wealth and lobbying success (the "total effect" of wealth in this analysis) was largest for these companies (see @fig-repeated-text-assets), we use this subset to examine how sophisticated lobbying may mediate the relationship between wealth and lobbying success. In this analysis, market capitalization is the key predictor variable, lobbying success is the dependent variable in the main model, and comment sophistication is the proposed mediator (i.e., the dependent variable in the mediator model). 

<!-- Selection on observables, not selection on observable in footnote-->

We examine four causal pathways^[We use the term causal pathways to refer to the fact that this analysis assumes a causal model, in the same way that OLS with selection on observables assumes a causal model. To interpret the findings and point estimates of mediation analysis as causally identified requires stronger assumptions than we think are justified. To do that, we would need to assume that the relationship between wealth and lobbying success is causal and that the mediator(s) examined are the only possible causal pathways between wealth and lobbying success. That is, we must assume there is only a direct effect of wealth and a mediated effect through the proposed mediator. We think this is unlikely because of inter-relationships between the mediators and the possibility of omitted variables.] between wealth and lobbying success: (1) donating to political campaigns via PACs, (2) spending on lobbying covered by the Lobbying Disclosure Act, (3) using more technical language in public comments, or (4) using more legally-sophisticated language in public comments. We test each mediator individually because they are not causally sequential [@Imai2013]. 

Mediation analysis conducted in this fashion suggests that the bulk of the relationship between wealth and lobbying success is attributable to wealthier organizations submitting more sophisticated comments. The Average Conditional Marginal Effect (ACME) estimates in @fig-mediation-acme show that both technical and legal sophistication appear to help explain the relationship between wealth and lobbying success (p < 0.05). Moreover, Appendix @fig-mediation shows that the ACME for technical sophistication is a large share of the Total Effect of wealth on lobbying success. Thus, we see support for Hypothesis 5: much of the relationship between market capitalization and our measure of lobbying success results from wealthier organizations submitting more sophisticated comments. This conclusion is robust to focusing on technical or legal sophistication, but technical sophistication explains a larger share of the relationship between wealth and lobbying success than legal sophistication. 

  

```{r fig-mediation-acme, out.width="70%",  fig.cap= "Political Spending, Lobbying, Technical Sophistication, and Legal Sophistication as Proposed Mediators between Wealth and Lobbying Success"} 
# | label: fig-mediation-acme
# | out-width: 70%
#| fig-cap: "Political Spending, Lobbying, Technical Sophistication, and Legal Sophistication as Proposed Mediators between Wealth and Lobbying Success"

knitr::include_graphics(here::here("figs/mediation-4wayNo0-acme-1.png")) 
``` 

Mediation analysis allows us to compare alternative influence mechanisms. One alternative mechanism goes through campaign contributions and power in Congress. If organizational wealth enables greater political contributions, if political contributions buy power in Congress, and if agency officials are concerned about congressional sanction when revising rules, campaign contributions may drive lobbying success. This argument is similar to research by @Gordon2005 suggesting that large organizations exert influence through repeated political contributions. We use PAC contributions as the proposed mediator to test this pathway. 

As a second alternative mechanism, we use lobbying expenditures as the mediator. Since disclosed lobbying expenditures target both Congress and agency officials, this causal pathway could operate via congressional sanction (as with the campaign spending via PACs) or more directly through lobbyists persuading agency officials to adopt their client's preferred policy language.   

In both cases, the ACME is small and not statistically significant. This implies that increasing a corporation’s wealth increases its spending on candidates and lobbyists, but this does not then explain lobbying success in changes between draft and final rules .^[This does not mean that political spending and lobbying do not have large effects on earlier stages of the policy process.] 
 
### Alternative Interpretation: Wealth Indicates Larger Membership  
<!-- 2024-06-18 Steve Removed quotes from most likely -->
Another explanation for our findings is that wealthy organizations
have more influence because they represent more people. If true, then organizational wealth is merely a stand-in for organizational membership, affecting the normative implications of our findings. Upon investigation, however, a wealth-membership association cannot explain the patterns in the data.  We find that organizations with larger numbers of active members are no more likely to comment (Appendix @fig-assets-vol) or have lobbying success (Appendix @fig-efficacyXvol). The coefficient on assets is unaffected by including an organization’s number of volunteers in the model of lobbying success.

<!--- CUTTING PER SUSAN'S SUGGESTION and replacing with the above
Like previous research [@Bartels2008; @Gilens2012; @Witko2021], we make descriptive inferences from associations between wealth and policymaking influence. Unlike previous research, we focus on the wealth of organizations, not individuals. However, if organizational wealth is merely a stand-in for organizational membership, our findings would not necessarily contradict an account of individual-level political equality in the administrative process. If wealthy organizations have more influence *because* they represent more people rather than *because* they have more resources, our findings would be consistent with pluralist conceptions of democracy. 
<!---
Since nearly all comments come from organizations [@JudgeLord2021], organizational wealth is the right conceptual target for this study
--->

<!-- SUSAN DELETED 
 Second, our findings in section @sec-results-profit suggest that net of wealth, for-profit organizations still enjoy greater advantages than non-profit organizations. To be consistent with pluralist democracy, for-profit organizations would need to represent the interests of more people, on average, than non-profit organizations of similar size.
--> 

<!---CUTTING THIS BECAUSE I'M NOT SURE IT MAKES SENSE AND ALSO NOT SURE WE NEED IT 
Third, our relationships hold within and across a set of categories of banks. This is relevant because, for some financial institutions, different assets are driven far more by the size of deposits than the number of depositors. 

Alternative Explanation #2: Regulatory Applicability. Another reading of our results, especially those on participation, is that wealth-based inequalities point to the differential applicability of many financial regulations to larger institutions (such as the Dodd-Frank requirement that certain restrictions apply only to institutions with more than $50 billion in capitalization). We show (Appendix ____) that the relationships reported here are monotonic across the distribution of assets and wealth and are not driven by discontinuities associated with the applicability of regulations to financial institutions above a certain asset threshold.
 --> 

# Conclusion

<!-- SUMMARY --> 
Capital-based wealth inequality increased dramatically over the twentieth century, especially in the United States [e.g., @Piketty2014; @Saez2020]. While political science research documented profound and durable patterns where wealth inequality in the United States leads to disproportionate influence in congressional policymaking, inequality in administrative policymaking has largely escaped systematic study. 

We provide data and tools to study the relationship between wealth, representation, and inequality in administrative policymaking for the first time. <!--Specifically, we collect the most comprehensive data ever assembled on one of the most sweeping regulatory statutes ever enacted in the United States: the Dodd-Frank Act of 2010. This legislation delegated significant powers to federal agencies to flesh out statutory policies, restrictions, meaning, and standards in rulemaking. The degree of administrative discretion was vast, and our new dataset permits a direct examination of the regulatory policy content created by government agencies in response, as well as attempts to impact those policies by outside organizations.-->
Our systematic approach---covering all rules across multiple agencies implementing the same landmark piece of legislation---allows unique comparisons within and across agencies and types of organizations. <!--It is the first study of which we are aware to systematically measure the wealth of those participating in agency rulemaking.--> By combining data about changes in rules with data about comments and their authors, we assess the relative level of lobbying access and lobbying effectiveness that different types of organizations enjoy.

<!---BIG EFFECTS--> 
We find support for our hypotheses predicting that disparities in wealth lead to inequality in administrative policymaking, including both kinds of theorized bias in rulemaking: bias in participation and bias in influence. 

First, wealthy organizations are more likely to participate in regulatory lobbying than less wealthy organizations. These findings hold even when comparing similar organizations---such as when comparing wealthy banks to less wealthy banks. If representation is primarily shaped by who shows up, these results suggest that wealthy organizations are better represented during financial rulemaking.

Second, we find suggestive evidence that inequalities in wealth drive lobbying influence. For example, market capitalization is strongly correlated with lobbying success among publicly-traded firms. Market capitalization is also highly correlated with comment sophistication, which, in turn, is associated with lobbying success. 
Mediation analysis suggests that much of the association between organizational wealth and lobbying success results from the technical and legal sophistication of the organization’s comment, not political power gained through campaign contributions or spending on lobbying firms.

Our empirical study also provides stylized facts for some of the formal literature on administrative policymaking. Of the literature summarized in @Schnakenberg2024, our results cohere most with subsidy-based theories of special interest influence [@Hall2006] and with models of policy development monopolies [@Hirsch2018]. Specifically, our findings offer a rationale for agencies to follow Hirsch and Shotts’ recommendation that agencies induce information provision by policy developers with different preferences. Meanwhile, our findings that wealthier organizations are more likely to participate and, conditional on participating, have more lobbying success provide the basis for further theoretical work. Our novel measures of comment sophistication also permit those interested in formal modeling to test theories related to  costly signaling and the mobilization of expertise. That said, this study is a large-sample empirical study, and we do not claim to adjudicate among claims in the formal literature. In focusing on rulemaking, our analysis also leaves aside important issues about venue choice raised by @BoehmkeJPP2013 and others. Future research might consider the simultaneous empirical modeling of legislative lobbying and commenting as potential complements or substitutes.

Future work could extend our findings, following @Ban2019, to make explicit comparisons between the legislative and regulatory policymaking processes. Future work could also assess the relationship between wealth inequality in other areas of agency decision-making, such as spending, permitting, and enforcement decisions. 

<!--POLICY IMPLICATIONS --> 
These results hold important implications for reforms aimed at ameliorating the effects of wealth inequality on government policy [@OIRA2023]. For instance, reforms that provide resources to organizations to develop more sophisticated comments and policy recommendations may level the playing field between differentially resourced groups. Such a strategy would resemble subsidized legal representation. Efforts similar to this are already underway at some agencies, including the U.S. Federal Energy Regulatory Commission.^[See: <https://www.ferc.gov/equity>] Our findings suggest that such reforms merit close study. 

<!-- Angelo 2024-06-14: Add to the paragraph below an encouragement to study the organizational representation of interests to respond to R3. -->
 


<!-- SCALEABLE (cut for now)--> 
<!--The methods we used to document inequality in financial rulemaking are scalable and reproducible so that future work can compare levels of inequality across other policy domains. Agency rulemaking dominates many policy domains. Companies and other interest groups spend much of their lobbying budgets on agency rulemaking. The methods we develop here open up the biases in participation and influence in agency rulemaking to systematic study.
--> 

Most importantly, this study offers a model for quantitatively studying inequality in U.S. policymaking. 
With the rise of the administrative state, scholars have documented the importance of agency rulemaking [@Kerwin2018], institutional bias toward businesses [@YackeeJOP2006], and the massive value businesses gain from lobbying agencies [@LibgoberQJPS2020]. Our novel data and methods enable a new view of biases in participation and influence in agency rulemaking. The consistent patterns of wealth-based inequalities we uncover advance our understanding of lobbying, money in politics, and how these pressures shape democracy in the modern administrative state.


\clearpage

\singlespacing

## References

::: {#refs}
:::

\clearpage

\small

\setcounter{section}{0}
\renewcommand{\thesection}{\Alph{section}}

\setcounter{table}{0}
\renewcommand{\thetable}{A\arabic{table}}

\setcounter{figure}{0}
\renewcommand{\thefigure}{A\arabic{figure}}

{{< include participatory-inequality-appendix.qmd >}}