American Payments System: Examining Consumer Payment Behavior

Abstract

This paper examines the relationship between U.S. consumers’ subjective evaluations of payment method attributes and their use of that payment method for an in-person transaction. We use a two-stage methodological approach with conditional logistic regression in Stage 1 and probit regression in Stage 2. The study analyzes pooled 2021-2024 Survey and Diary of Consumer Payment Choice data covering 6,902 unique individuals and 50,707 unique transactions.

The results of Stage 1 show that payment choice is systematically associated with consumers’ subjective evaluations of payment method attributes. Convenience has the largest estimated association with payment choice: a one-unit increase in Convenience rating is associated with a 0.63 percentage point increase in the probability of choosing that payment method, while Security has the smallest estimated association at 0.11 percentage points. Stage 2 shows that a higher Convenience Advantage, how much more convenient consumers find their preferred payment method than the alternatives, is associated with a lower probability of deviating from the preferred payment method across cash-, credit-, and debit-preferring consumers. Transaction size also has an asymmetric relationship with deviation across preference groups.

Overall, the findings suggest that convenience and transaction context, such as the type of merchant, are important demand-side factors in consumer payment behavior. The study may be useful to policymakers and financial companies seeking to better understand consumer adoption, payment choice, and preference persistence.

Section 1: Introduction

1.1 Motivation

Millions of payments happen every day across the U.S. and serve as the backbone of our economic activity as an avenue for transactions of goods and services. In a highly sophisticated and complex economy, the American payment system ensures a regulated and efficient process for these transactions to occur or support the economic activity in the country. The payments system supports billion-dollar industries that process, develop, and transact these purchases. Recently, payment system modernization efforts have been put in place to benefit a number of American stakeholders, including the government, businesses and consumers.

In 2025, a presidential executive order, for instance, began transitioning the federal payments system towards digital payments, reducing the use of paper checks for federal payments and promoting electronic payment methods (The White House, 2025b). This was further supported by legislative efforts aimed at setting a regulatory framework for the modernized payments system. In July 2025, the GENIUS Act was signed into law and established the first federal regulatory framework for payment stablecoins (The White House, 2025a). In addition, the House-passed CLARITY Act proposed a broader regulatory framework for digital asset markets by establishing clearer requirements for market participants and regulatory responsibilities (U.S. House Committee on Financial Services, 2025). In addition, the Federal Reserve also contributed to building this transition by developing the FedNow payment system.

FedNow was released in 2023 in response to customer and business interest in faster transactions. The system is an instant payment rail that operates 24/7 and is available to eligible U.S. depository institutions, with potential benefits for small businesses through faster payments and improved cash flow (O’Neill, 2023). All of these developments create a supply-side infrastructure that aims to enable Americans to use digital-based methods for safer, cheaper, and faster payments and make government payments more efficient.

The U.S. has historically lagged in modernization of its payment system and widespread adoption of new payment technology. Comparatively, other developed economies such as Canada and Sweden have also pursued substantial payment modernization efforts and achieved broader adoption of digital payments. Canada’s payments modernization efforts could contribute between $3 and $7 billion in economic output annually (Meridian International Center, 2025), while Sweden now has an almost entirely digital payments market with widespread use of mobile payments (Sveriges Riksbank, 2025).

However, the U.S. adoption rates of new payment technology are much lower than Sweden’s, for instance. Han and Wang (2021) propose that the incremental benefits of this technology did not outweigh the costs of switching for most Americans, leading to a mostly card-complementing payment technology that still saw sluggish growth rates of use. This demonstrates that American payment modernization efforts are mostly facing demand-side challenges of “sticky” habits as consumers tend to continue using familiar technology even though supply-side infrastructure for new technology is built but waiting for widespread consumer demand.

Recent efforts by the government and the Fed to modernize the payment system are potentially very beneficial, as Birigozzi et al. (2025) show that a 1% increase in the rates of adoption and use of digital payments is associated with an increase in GDP growth equal to roughly 6%-8% of the existing growth rate. However, in order to ensure a successful transition to an efficient and modern digital economy, the U.S. needs to ensure that it bridges the gap between the benefits of innovation and the costs of certainty of legacy systems to align both supply-side and demand-side stakeholders in the payments ecosystem. Thus, to ensure an effective modernization transition, it is crucial for policymakers and businesses to understand what influences consumers’ payment choices from the demand-side perspective and analyze what forces already exist that incentivize consumers to deviate from their preferences to an alternative payment method.

1.2 Research Question

As such, the research question is:

The research question aims to provide a more comprehensive insight on consumer payment behavior from two perspectives. First, the research examines the association between payment method choice and consumers’ subjective assessments of payment attributes. Attribute evaluations are used as a key variable in this paper because they capture how consumers evaluate different payment methods from five dimensions (convenience, speed, cost, security, and acceptance), rather than treating payment valuations as an aggregate measure. Second, the research examines how consistent ex ante preferences are in payment methods based on consumers’ attribute evaluations for convenience and based on different transaction contexts, including changes in the size of the purchase and the type of the purchase based on the merchant category.

1.3 Structure of This Paper

This paper uses a two-stage methodological approach and contributes to the literature by conceptually consolidating these two attribute-based models of consumer choice, which are currently treated as largely independent in the literature. The structure of this paper is as follows: Section 2 provides an overview of prior work in the field and explains the existing theoretical and methodological background on which this paper’s empirical approach is built. Section 3 provides a detailed overview of the data sources and dataset construction used in the paper. Section 4 gives a comprehensive overview of the methodological approach for the two stages of analysis.

Mainly, Stage 1 utilizes a conditional logistic regression to estimate whether the attributes predict consumer choice, while Stage 2 uses probit regression to measure the probability of deviation from a preferred payment method among consumers in the sample. The results of the two stages are presented sequentially in Section 5. These findings are further discussed and interpreted in Section 6. Section 7 concludes the paper and presents the policy implications and methodological limitations.

Section 2: Literature Review

The existing research on payments offers a broad literature on the history, technology, and adoption of payments in the US, with a particular focus on the effect that changes in policy, price incentives, and supply-side factors have on the larger trends of consumer payment use and behavior. Given the focus of the paper, understanding these facets from a scholarly standpoint is crucial for a well-rounded and insightful analysis. A more general research conversation, for instance, offers a broader examination of the trends of the American payments industry and payments system, as well as their relative growth compared to developing and developed nations that directly ties back to the current modernization efforts (e.g., Han and Wang, 2026; Catalini and Lilley, 2021).

Moreover, a frequent and important focus for the economists in the field is cross-sectional research documenting the role of demographic factors like income, education, and payment attributes in shaping consumer payment choice (e.g., Schuh and Stavins, 2015; Sari et al., 2022). In addition, a large number of studies also investigate the influence of transaction-level contexts such as prices, surcharges, cashless policies, cash discounts, and merchant categories on payment behaviors and payment method choices (e.g., Cohen et al., 2022; Greene et al., 2026). This work is especially beneficial as it establishes the scholarly consensus on the topic that our paper aims to contribute to and extend methodologically and conceptually.

The following literature review separates the scholarly landscape into five main sections. It first examines current trends and challenges in the American payments system, followed by research on demographic characteristics and payment instrument attributes, prior work on supply-side and merchant factors, the theoretical background used in the methodological framework, and finally this paper’s contribution to the literature.

2.1 Current Trends and Challenges in the American Payments System

Existing literature in the space of American payment systems offers valuable insights into how consumer payment behavior evolved with the development of new technologies and payment features. Catalini and Lilley (2021), for instance, find that the lag of the U.S. payment systems compared to the global trends is due to structural reasons such as legacy infrastructure, market fragmentation, and lack of competition. Mainly, they show that there is still a heavy reliance on cash in the U.S., as roughly a quarter of domestic payments rely on cash, creating costs in terms of time, as well as ATM fees, theft risk, and other factors. Fragmentation of the payment system in the U.S. also impedes the existence of widespread fast and convenient payment rails, since currently payments may take days to process and incur a fee, while such services are free and continuously available in the UK. Ultimately, this paper adds conceptual grounding to the motivation of our study, as it outlines the vulnerabilities and structural gaps in the American payment system due to fragmentation.

A similar argument was put forth by Han and Wang (2026) in their discussion on the historically slow adoption rates of mobile payments in the U.S. compared to China and Kenya. Though the paper mostly focuses on mobile payments, our perspective is that mobile payments are a proxy for overall modernization efforts, since they are a case of introduction of new technology. Thus, this work is an important addition to our literature review, as it further aligns with the purpose of our research study by providing it with a structural scholarly background on the American stakeholders’ response to previous modernization efforts. As such, one of the factors that they highlight is that the U.S. and other developed countries have had successful experiences as early adopters of card payment technologies compared to developing countries but are now lagging in the adoption of what they state as the next step of the payment technology.

Their main findings are that the past success of adoption of card-payments is holding the U.S. back in adopting new payment technologies like mobile payments. Furthermore, they provide a supporting argument for our view that payment modernization can provide economic and social benefits. They find that alternative payment methods may have positive social effects but are not taken advantage of by the U.S. through widespread adoption of newer payment technologies. Finally, they also reinforce the idea that there is a misalignment of incentives in American stakeholders compared to those in Kenya and China. They state that for Americans, choosing digital payments over cards does not bring substantial incremental benefit yet, while in developing countries payment externalities benefit social welfare, boosting adoption and use rates.

Moreover, a paper by Murat (2025), produced as part of the JIW at Princeton University, also extends a similar argument of economic benefits of digital payment technology. The author uses DCPC to build a longitudinal individual-level panel dataset that tracks the same consumers over 3 years. The key focus of the paper is on mobile payments like QR code payments, digital wallet payments (tap-to-pay), and in-app payments (e.g., Starbucks Mobile Order). Particularly, the paper examines the effects of mobile payment adoption on whether consumers spend more for discretionary purchases like entertainment, restaurants, etc.

The author uses within-person OLS and fixed effects regression to compare discretionary expenditures of consumers before adopting mobile-payment and after. The paper finds a statistically significant association between the adoption of mobile payments and an increase in discretionary expenditure, as consumers spend almost 50% more on discretionary expenditures after adopting the new technology than before adopting mobile payments. Although it uses a similar data source and belongs to the same body of literature as our paper, its main contribution is in outlining associational benefits and payment behavior changes with mobile payments. Thus, this is particularly useful for our study, as it supports the argument that payment modernization through introduction of new technologies can be beneficial for the economy, as it is associated with higher spending and may contribute to economic growth.

2.2 Research on demographic characteristics and payment instrument attributes

The literature on demographic characteristics and payment attributes directly connects to our study, as they use a similar demand-side perspective on payment behavior. Papers by Koulayev et al. (2016) and Schuh and Stavins (2015) are especially valuable, as they present the existing scholarly findings on the role of payment attributes in payment choice. Our study extends the analysis to the most recent data with a modified methodological approach. Notably, these studies find that not all attributes have the same importance in how consumers adopt and use payment instruments. Both papers look at payment behavior from a two-stage approach, first examining adoption and then continued use of payment.

Our paper, similarly, uses a two-stage approach but one that identifies the role of attributes in the use of payment instruments and also its persistence. Schuh and Stavins (2015) provide valuable results, as they find that Speed and Security, although often highlighted as the most important payment features, were not the best predictors of adoption and use. In their sample on consumers from 2013 DCPC they test the relationship of all 5 characteristics on payment instrument adoption but only consider credit and debit cards. They state that speed and security have a small likelihood of substantially changing consumer payment behavior. They use simulation of increases in attribute perceptions and find that a 10% increase in security perception, for example, increases the transaction share of credit cards by 0.22 percentage points. Mainly, they conceptually outline that speed and security should matter because they can incentivize greater trust and reliance on payment methods but in reality convenience and cost appear as stronger determinants of payment instrument adoption.

In addition, an important paper by Koulayev et al. (2016) looks at the consumer payment behavior change such as substitution patterns in response to structural payment system changes like fees and regulations, specifically looking at interchange fees and surcharging by merchants. The scholars relied on older SCPC data from 2008 and had a methodological constraint, since they did not observe each transaction separately but only as an aggregate of all transactions for each payment method per respondent. Thus, our study extends the analysis of this literature by using more detailed transaction-level data and updates it both temporally and methodologically.

Notably, the scholars also use attributes such as security, speed, acceptance, etc., in their regressions and find that they are important predictors of use of payment methods. Additionally, they point to a stakeholder misalignment in the US, as policy focuses on work around security while empirical evidence shows that consumers put higher value in attributes like cost, ease of use, and others. Although their main focus is on how regulations on card payments can increase substitution towards cash and checks, they touch on a similar idea of deviation, but one that is policy-driven rather than payment-context-driven.

Furthermore, a study by Sari et al. (2022) explores the connection between the intent to use mobile payments and possible factors such as ease of use and others. The source is valuable because it introduces the behavioral reasons behind the use of mobile payments, including ease of use, convenience, social benefit on intention, etc., which is an important consideration when reviewing the adoption levels of mobile payments. This links back to the network effects paper, as the authors describe that there is rising competition in the mobile payment market, which creates incentives to attract and retain customers. The researchers conducted a cross-sectional empirical study in Indonesia. The main finding was that there was no significant direct effect of ease of use, convenience, etc. on the intention to use mobile payments. This work is useful for our research as it offers an insight into the role of payment attributes in adoption reasons for innovative technologies as part of efforts for modernization of payment systems.

2.3 Prior Work on supply-side and merchant factors in payment behavior

The most relevant and valuable literature for our research is a working paper by Greene et al. (2026) that offers an important methodological and conceptual contribution for the Stage 2 of our analysis. They estimate the probability of deviation from a preferred payment method with transaction size and merchant group acting as main regressors while looking at demographic and income relationships as well. Greene et al. use probit estimates of from-cash-to-card transactions and from-card-to-cash transactions. The authors highlight that acceptance of card payment is statistically significant but has a minor effect on deviation from the preferred payment method, showing an existing but relatively small effect of supply-side constraints, which serves as an underlying assumption in our analysis.

Greene et al. focus on deviations as a response to cash discounts and card surcharges. They found that even with the presence of cash discounts, credit preferrers do not deviate, which might show that merchant steering is not always effective. They also find that larger transactions are associated with continued use of cards and not switching to cash. For merchant analysis, they find that consumers are less likely to deviate to cash when paying for food, gas, and the grocery store and highlight that this could be due to common use of credit in these types of settings. Notably, the scholars look at surcharges that apply to both credit and debit, which leads them to group these payment methods into a single instrument called card payments. Our analysis, on the other hand, attempts to methodologically extend their paper by separating credit card and debit card as different instruments, since card surcharges are not a key regressor in our analysis. This allows our analysis to also consider the role that card payment features like rewards and credit line may play in consumer behavior when separating credit cards and debit cards.

Crucially, our paper adopts their methodology for our analysis by using probit and constructing the binary deviation variable in a similar way. However, the contribution of our analysis is in a novel approach by using attribute ratings as a regressor, examining debit cards as a separate payment method from credit cards, and using a larger temporal scope of 2021-2024 while controlling for year fixed effects.

Furthermore, a different study by Greene and Stavins (2020) uses 2017 DCPC with transaction-level analysis to focus on what determines consumer payment behavior for bill payments. Notably, the results confirm that as the amount increases people switch away from cash and that low income and unbanked status lead to more cash use. They also use probit to estimate deviation probability and use transaction size and merchant group as main regressors, similarly to some aspects of our methodology. A key finding from their study is that transaction size had an insignificant relationship with deviation, while merchant groups for utilities, credit card bills, and entertainment bundles (Netflix, etc.) had a negative relationship with deviation from the preferred instrument for bill payments.

However, their study differs in its approach from our research in the main transaction type of interest, as their focus centers around bill payments, while our focus centers around in-person purchases. Thus, the set of available alternatives is also different for Greene and Stavins, since bill payments innately have a restriction in the type of payment method available (e.g., online through banking apps, in-person through cash, etc.). Our analysis, instead, focuses on in-person payments and a common choice set of cash, credit, and debit, allowing us to examine deviation across these three instruments at the transaction level. This provides a broader setting for studying realized consumer payment behavior, although the data do not observe the full set of payment methods available at each transaction and therefore cannot completely rule out supply-side constraints.

Moreover, Cohen et al. (2022) similarly conducted a study on SCPC data between 2017 and 2020. Primarily, they track the trend of decline in the use of paper payments like cash and checks towards card and electronic payments and analyze how these payment behaviors correlate with merchant groups. Importantly, they identify a trend from in-person to remote payments that naturally, by design, favors electronic payments more for all merchants. They highlight that merchants that see most cash payments are person-to-person, food, donations, entertainment, and services.

They used probit marginal effects to estimate the probability of paying with paper-based methods, and they found that a tenfold increase in transaction value is associated with a 39.3-percentage-point decrease in the probability of using cash. Consistent with other literature in the space, they show that cash use is more associated with older, lower income, and less educated families, as well as in the Black and Latino demographic groups. However, they show a growing trend of online shopping and a moderate decline in use of cash at all merchant locations, but this connects to the pandemic shock as regulations and people’s health concerns forced them to switch to online ordering, which the authors hypothesized would continue even after the pandemic.

2.4 Theoretical Background

McFadden (1974) represents the backbone of the theoretical and methodological framework of this paper. Specifically, McFadden outlines the RUM framework of revealed preferences. The ideas of this study are relevant to our methodology because McFadden’s paper outlines the conditional logit framework for choices made from a set of available alternatives. This approach is used to understand how consumer choices can be empirically estimated in terms of probabilities of selecting one of the alternatives among others.

Moreover, random utility is explained as an economic concept of utility maximization. However, the utility-maximizing choices are influenced by both a systematic factor, which is observed, and a random factor that is unobserved but represented statistically in the model. The unobserved term represents idiosyncratic factors that are specific to individual i’s preferences and choices based on the available options. Most importantly, this source provides one of the central assumptions of our methodology: Independence of Irrelevant Alternatives (IIA). It states that the probability of choosing Option 1 over Option 2 is driven only by those two options and is not influenced by other alternatives like Option 3.

2.5 Contribution To The Literature

The literature review provides an insightful look at the current body of empirical studies and makes important contributions to the methodological and conceptual framework of my paper. However, the review also reveals that the emerging literature on consumer payment choice focuses mainly on external factors such as acceptance, policy changes, and pricing incentives. However, few papers incorporate consumers’ subjective payment method evaluations from the lens of the methods’ attributes. The existing studies that take that approach, such as Koulayev and Schuh, mostly look at attributes as mechanisms of adoption and use of payments rather than the mechanisms for consistency.

As such, our empirical study aims to bridge these two separate sets of literature: those that explain choice and those that explain deviation. Namely, this research demonstrates that consumer deviation from preferences is systematically associated with subjective attribute ratings and, through a two-stage analysis, provides insight into the connection between stated preferences and realized choices. Finally, this conceptual contribution is also supported by the utilization of a pooled cross-sectional dataset at the individual-transaction-level that allows a more detailed analysis of within-transaction and within-individual variation in payment choice and preference consistency in different contexts than is possible in existing literature.

Section 3: Data and Descriptive Statistics

3.1 Data Set Sources

The core data sources of our study are the Diary of Consumer Payment Choice (DCPC) and the Survey of Consumer Payment Choice (SCPC) published by the Federal Reserve Bank of Atlanta (2025). This is an extensive dataset used in most of the literature on consumer payment behavior in the US. Given that data from banks and merchants on the payments behavior and transactions of their customers are most often proprietary and rarely available for public use, these datasets offer a substantive survey-based resource suitable for empirical studies. The dataset is developed from the survey collection by USC’s “Understanding America Study” and published by Kevin Foster, Claire Greene, and Joanna Stavins from the Federal Reserve Banks of Atlanta and Boston.

The sample for the dataset is selected through probability-based sampling and is designed to be nationally representative of U.S. households across the 50 states. The SCPC collects individual-level demographic variables like age, gender, income, etc.; behavioral variables like preferences in payments, ratings of payment attributes; and attitudinal variables like views towards cashless society and banking. The DCPC offers a transaction-level extension to the SCPC’s sample.

It collects variables from the payment records of the same survey respondents’ daily transactions such as payment instrument used, the size of the purchase, the type of merchant, etc. It is important to note that both SCPC and DCPC capture data for the same individuals, who are assigned a unique respondent ID. As a result, we integrated the SCPC and DCPC into a single merged dataset that includes both the individual-level and transaction-level variables for these respondents. This is important, because collectively analyzing consumers’ stated subjective preferences and evaluations and their revealed payment choices is crucial for the purposes of this study.

3.2 Data Set Construction

As such, this study covers the integrated SCPC/DCPC datasets for the 2021-2024 time period. The raw dataset was constructed by pooling annual datasets from SCPC and DCPC over the 4 years. The two pooled cross-sectional datasets were formed into a transaction-level merged file that contains 101,163 observations with each individual row corresponding to a transaction recorded by the consumer. To achieve the correct format for the data required for the empirical methods of this study, redundant duplicate observations were removed and the file was reshaped into long format.

This is done to construct an alternative panel for each individual transaction that fits the structure requirements for conditional logistic regression. After restricting the sample to cash, credit card, and debit card payments, focusing on in-person purchases, and applying the final cleaning steps, the dataset contained 50,707 unique transactions. The data were then restructured into long format, with three rows for each transaction representing the alternative choices of cash, credit card, and debit card. As such, the final analytical dataset contained 152,121 observations.

This study focuses specifically on cash, credit card, and debit card because these payment methods are the most widely used instruments for in-person purchases. This is reflected in Table 1, which shows the breakdown of frequencies of each payment method used as reported by the consumers in the raw merged dataset. Payment methods like Check, Prepaid cards, and Money Orders are less than 5% of the total responses in the sample, which restricts their application in the study due to low variation. Instruments with higher use frequencies like Bank account number payment and others are not applicable because they are not typically used for in-person purchases, which serve as the main focus of the analysis.

The final cleaned dataset is in cross-sectional format pooled from years 2021 through 2024 and has 6,902 unique individuals with 50,707 unique transactions. Overall, there were 10,055 unique transactions in 2021, 11,805 unique transactions in 2022, 13,019 unique transactions in 2023, and 15,828 unique transactions in 2024. Since most of the literature relied on single-year cross-sectional analysis, the use of pooled data across the 4 years selected for the sample is a meaningful addition to this study’s findings and the literature. Namely, a larger sample and temporal scope may allow for higher statistical power due to improved estimation precision and lower standard errors.

3.3 Variables

The dataset focuses on transaction-level observations and implements both transaction-level and individual-level variables in the analysis. The empirical analysis of our study consists of two stages with differing regressors and outcome variables.

Stage 1:

Dependent Variable: Chosen Payment Instrument.

The original variable for the payment instrument used for a recorded transaction was categorical and identified one of the payment methods shown in Table 1. It is important to note that since the transaction-level dataset was expanded into a long (stacked) format, each transaction has 3 rows that represent the alternative choices for each consumer: cash, credit, or debit. Due to this format required for the conditional logit, the dependent variable was transformed into a binary variable that is equal to 1 for the row that indicates the payment instrument that was actually used in the transaction, with the other two alternative choices equal to zero. For instance, if an actual transaction was recorded by the consumer to have been paid with cash, the dependent variable equals 1 (Chosen = 1) for the cash row and equals 0 for the debit row and credit row (Chosen = 0).

Independent Variable: Subjective Rating of Payment Instrument Attributes

These variables are respondent-level evaluations recorded separately for each payment method and are provided by the original raw datasets. They include the consumer’s subjective ratings of payment attributes gathered prior to consumers making and recording their payment behavior. These attributes are: Acceptance, Cost, Convenience, Security, and Speed. The attribute on Acceptance measures subjective belief on how likely a payment is to be accepted in different settings. The attribute on Cost measures the subjective belief on the cost of owning the payment instrument, taking into account fees, subscriptions, interest, discounts, etc.

The attribute on Security measures the subjective belief on the protection of the payment method against loss of personal information and financial damages and losses. The attribute on Convenience measures the subjective belief on the ease of use and frictionlessness of payment methods in a transaction setting, including how easy it is to handle and pay with. The attribute on Speed measures the subjective belief on time spent at the counter when paying with a specific payment method. Each payment method is rated separately by the consumers for all attribute evaluations.

The attributes are used as the variables of interest and main regressors, as the study attempts to estimate the prediction consistency of subjective ratings with the likelihood of payment choice. They are presented in categorical format as rated from 1 to 5, where 1 is the lowest and 5 is the highest. For instance, a consumer rating of Speed as 1 for cash indicates their subjective belief that payments with cash are sluggish (expecting slow processing of payments), while rating it as a 5 indicates their belief that they are swift (transactions are fast).

It is important to note that the attribute for Cost is coded counterintuitively, as rating cash as 1 indicates subjective belief that the method is expensive or burdensome, while rating it as 5 indicates the belief that it is affordable or inexpensive. Although the variable is essentially categorical, in the analysis we make an assumption of treating it as a continuous variable, where an increase from a rating of a 2 to 3 is the same as an increase in the rating of 3 to 4. This is done for the purpose of easier interpretation as it allows for more convenient estimation of a single coefficient rather than multiple dummy variables.

Control Variables:

1) Transaction Size (Amount) interaction with Credit

2) Transaction Size (Amount) interaction with Debit

These variables were constructed by multiplying the log of transaction size by dummy variables that equal 1 for rows representing (1) the credit card alternative and (2) the debit card alternative, respectively, and 0 otherwise. The methodological structure of conditional logit relies on variation between alternatives within the same transaction choice set. As such, variables that do not vary across the three payment alternatives cannot be separately estimated unless they are interacted with alternative-specific indicators. However, in order to provide a more precise estimate for the analysis, transaction size interactions are added to control for the fact that larger purchases may have a systematic constraint on the type of payment used due to physical constraints for cash or feature constraints for debit (absence of a credit line). Thus, including the control variable separates the relationship between transaction size and probability of payment choice that may be absorbed by the attribute variables otherwise.

3) Credit Card Rewards interaction with Credit

Credit cards often offer rewards that may also confound the relationship of attributes with the payment choice probability, since they can be perceived as having extra value due to cash back, reward redemption, and other factors. Thus, controlling for rewards allows for a more precise estimate of the attribute relationship that is not confounded with the features and advantages of credit cards.

Stage 2:

Dependent Variable: Deviation in Payment Method Preference

The variable was constructed from an individual-level variable in the original dataset that identified the preferred in-person payment method of a consumer. It was a categorical variable indicating a single initial payment method (such as cash, credit, etc.) that the consumer prefers for all in-person transactions. For the purposes of this study’s methodology, only those individuals whose preferred payment method was either cash, credit, or debit were observed. The dependent variable is a binary variable that indicates whether the payment method used in a transaction matches the payment method stated as preferred prior to the transaction. For instance, Deviated = 1 if a consumer used cash for an in-person purchase but initially indicated a preference for credit cards for all in-person transactions.

Independent Variables:

1) Transaction Size (Amount)

The original Transaction Amount variable was presented in the dataset as a record of each transaction’s total sum that was paid. The original variable was skewed to the right with large outliers, which made the variable hard to interpret. In order to make the variable more normally distributed and avoid the influence of large outliers, the log of the variable was taken.

2) Convenience Advantage Rating

It was constructed by subtracting the average convenience rating of the two non-preferred payment methods from the convenience rating of the consumer’s stated preferred in-person payment method. A cash preferrer might rate cash convenience as 5, credit convenience as 2 and debit convenience as 2. The convenience advantage would measure that their preferred method is 3 points more convenient than the alternatives, since 5 - ((2+2)/2) = 3.

Control Variables

1) Merchant Group

The variable for Merchant Group is constructed from the original dataset’s merchant group variable, which included 21 separate merchant groups such as sit-down restaurants and bars, general merchandise and department stores, hairdressers, auto repair, etc. Given the large number of merchant groups and the distribution of the frequency of transactions at each separate merchant, a Merchant Group variable was consolidated by grouping similar vendors like restaurants and cafes into Food, and professional services (accounting, veterinarians) with personal services (auto repair, barbershops) into Services, and similarly for other merchant groups.

2) Demographic and Temporal Variables

The main demographic variables are age, gender, race, and highest level of education, and a high-income indicator for households with annual income of $60,000 or more, and are included to control for individual-level characteristics that may influence the relationship between the regressor variables and the probability of deviation. In addition to that, year fixed effects are used to control for temporal trends over the sample time period of 2021 through 2024.

3.4 Descriptive statistics

Transaction size serves as a control variable and a variable of interest in our analysis. Figure 1 is a histogram that reflects the distribution of log of transaction size in the sample. The original amount variable was skewed to the right, as most observed transactions are under $1000 but there are some big outliers in the range of $10,000s, which shows that smaller purchase amounts dominate. Thus, log-transforming Transaction Size substantially reduces its right skew and the influence of large outliers, producing a more symmetric distribution for the regression analysis.

Figure 2 is a time series plot that visualizes the trend in the use of the 3 payment methods of interest across the 4 years. As such, it demonstrates that from 2021 to 2024, cash use fell from 30.7% to 24.3%, debit card use rose slightly from 32.9% to 33.2%, while credit card use increased moderately from 36.5% to 42.5%. Overall, the graph shows the relative increase in popularity of card payments compared to cash, which has seen steadily declining rates.

The bar chart demonstrates the distribution of shares of in-person transactions done by consumers in the sample at each merchant group, which are a control variable and a variable of interest in the analysis. Notably, there are two merchant groups that dominate: grocery and food, which account for 28% of transactions each. This is consistent with the nature of these transactions, as they are routine and purchases at these vendors may be done multiple times a week. Department Stores and Gas Stations have the next-largest transaction shares at 15.5% and 12.6%, respectively. However, Travel and Utilities account for less than 3% of transactions combined, which is expected with their infrequent incidence, since utility bills are commonly paid monthly, while travel expenditures are often occasional and non-systematic.

Figure 4 shows a time series plot of payment method preferences by year. Credit and debit cards are the most preferred methods throughout the sample period. Credit preference increased from 41.0% of respondents in 2021 to 44.8% in 2024, while debit preference decreased from 43.3% to 40.4%. Cash-preferring respondents remained a minority throughout the period, accounting for approximately 15%-17% of respondents. Overall, the figure suggests a gradual shift in preference from debit toward credit over the sample period, while cash preference remained relatively stable at a much lower level.

Table 2 presents the average rating of the Convenience Advantage Variable that is central to Stage 2 of the Analysis. As such, the numbers show that those consumers that prefer credit and debit cards have a higher convenience advantage rating than consumers that prefer cash. Credit-preferring consumers have the highest mean convenience advantage rating of 0.836, showing that, on average, those consumers see their method of preference as more convenient than others. Debit Cards rank second with a 0.497 mean rating for convenience advantage, while consumers that usually prefer cash have a fairly low mean convenience advantage of 0.041, which may show that those that prefer cash may not have a high belief in its relative convenience.

Section 4: Methodology

4.1 Theoretical Framework and Assumptions

This paper takes an empirical approach to investigate consumer payment choice and behavior from the lens of behavioral and transaction-contextual factors. Using the transaction-level datasets, we attempt to understand the relationship between consumers’ ex ante subjective beliefs about payment methods and their realized payment choices for in-person transactions. The analysis uses a two-stage strategy, which is integral for examining this relationship from two perspectives: the first stage examines the association of payment attribute ratings with payment method use and the second stage identifies the association of the attribute ratings with deviation from payment method preferences.

Since we’re modeling discrete choice behavior, the methodological framework used in this study applies the Random Utility Model. The RUM approach allows Stage 1 of the analysis to measure the utility Uijt that individual i derives from the set of alternative choices. In the methodology of this study, there are three payment methods j, which include cash, credit, and debit, and are used for a realized in-person transaction t. The utility derived from the use of the methods for transactions comprises an error term εijt, which is the idiosyncratic unobserved component, and an observed systematic component shaped by, in the baseline specification, payment method attributes and, in the control specification, extended by a transaction context term.

As such, the methodology for Stage 1 adopts a similar approach to Koulayev et al. (2016) in using conditional logit to examine the relationship between payment attributes and utility. However, this paper differs from their study in the data structure as it uses transaction-level data pooled across 4 years and focuses on attribute evaluations as main regressors and variables of interest. The Koulayev study, on the other hand, uses aggregate transaction data for 2008 only and focuses mainly on the effects of transaction fees policy on payment instrument choice.

This study’s Stage 1 specifically uses conditional logit because the data structure includes 5 attribute evaluations that vary across payment method types within the same alternative choice set for the same consumer. This produces the appropriate variation across payment alternatives within each transaction choice set that is required for conditional logit. Notably, according to RUM, a central assumption of the conditional logit model is Independence of Irrelevant Alternatives (IIA). This assumption implies that the relative probability of choosing any two payment methods is determined only by those two options and is not influenced by the presence of a third alternative payment method.

4.2 Stage 1 Analysis: Conditional Logistic Regression

The first stage of the analysis estimates the relationship between subjective ratings of five attributes of payment methods and the probability of using a payment method for an in-person transaction. The regression is done in two parts, which are aimed at examining the stability of the coefficients on the regressor variables of subjective attribute rating. First, the baseline conditional logistic regression includes no control variables and estimates the raw relationship between the two variables. The inclusion of interaction controls is aimed at isolating the relationship of attribute rating from potential confounding with rewards and transaction size.

Also, individual fixed effects are not included in the Stage 1 analysis because the conditional logit estimates variation between alternatives within each transaction choice set. As such, demographic variables and other characteristics that do not vary across alternatives within the same transaction cannot be separately estimated unless they are interacted with an alternative-specific indicator. Standard errors are clustered at the respondent level to account for repeated transactions made by the same individual.

Baseline Conditional Logistic Regression:

This equation directly ties back to the RUM’s framework assumption of a utility-maximizing consumer. As such, this expression, which includes observable attributes and an unobserved error term, demonstrates the way the model estimates the theoretical utility Uijt derived by individual i from their use of payment method j in a transaction t. This latent utility is a core assumption of the theoretical model and, although unobserved as an actual result, allows observation of the probability in the conditional logit.

This expression is the observable component of our analysis and estimates the probability of individual i using payment method j during a transaction t. The regressors are the attribute variables, including Acceptance, Cost, Convenience, Security, and Speed. The relationship between the probability and attribute rating is measured by the beta, as an association of a one-unit increase in a consumer’s subjective rating of an attribute for a given payment method and the relative log odds of choosing that method. Marginal effects are reported separately to provide a more intuitive interpretation in terms of payment choice probability.

Conditional Logistic Regression With Interacted Controls:

The utility function for the regression that includes controls behaves similarly to the utility function in the baseline model by estimating the latent utility derived from the use of the payment method j. Further, it also reflects the main assumption of a utility-maximizing consumer choosing the payment method that provides the most utility among the alternative options. Importantly, the interacted control variables include Transaction Size interacted with Debit Card and Credit Card dummy variables. In addition, the Credit Card Rewards are interacted with the Credit Card dummy variable.

Transaction size interactions are added to control for the fact that larger purchases may have a systematic constraint on the type of payment used due to physical constraints for cash or feature constraints for debit (absence of credit line). Thus, including the control variable separates the relationship between transaction size and probability of payment choice that may be absorbed by the attribute variables otherwise. Moreover, credit cards often offer rewards that may also confound the relationship of attributes with the payment choice probability, since they can be perceived as having extra value due to cash back, reward redemption, and other factors. Thus, controlling for rewards allows for a more precise estimate of the attribute relationship that is not confounded with the features and advantages of credit cards.

This is a simplified expression of the conditional logistic model that measures the probability of using payment method j in a transaction t by individual i. It includes the attribute rating variables Acceptance, Cost, Convenience, Security, and Speed, as well as the interacted controls TransactionSize x Credit, TransactionSize x Debit, and CardRewards x Credit. The term β'xij represents the linear index formed by the regressor variables and their coefficients, where beta measures the associational relationship between a one-unit increase in attribute rating and the relative log odds of choosing payment method j. For control variables, it measures the relationship between a one-unit increase in the log of transaction size and changes in log odds of choosing credit or debit relative to cash. A one-unit increase in the log of transaction size corresponds to an approximately 2.72-times larger transaction amount.

4.3 Stage 2: Deviation Analysis Using Probit Regression

The second stage of the analysis estimates the relationship of convenience advantage variables and transaction-contextual variables with the probability of deviating from a preferred payment method. The second stage, similarly to the first, is done in multiple parts, which are aimed at examining not just the deviation probability, but also the directional deviation from one payment method to another, which offers more detailed and meaningful insight into consumer payment behavior. The methodology for Stage 2 uses probit because the outcome variable Deviated is binary. Our study’s methodological approach is similar to the methodology used by Greene et al. (2026), who also adopt probit regression in order to estimate the probability of deviation. Greene et al. mainly focus on how merchant steering through cash discounts and credit surcharges drives deviation. However, this study aims to examine the relationship primarily from the demand-side perspective.

The expression estimates the probability of a consumer deviating from their preferred payment method, measured separately for each method of preference: consumers that prefer cash, consumers that prefer credit cards and consumers that prefer debit. This is a probit regression, where Φ represents the cumulative distribution function (CDF) of the standard normal distribution that converts linear index, which is the expression inside the parentheses, into a probability; term α is the intercept; terms β, γ, δ, and λ denote the coefficients associated with key explanatory variables, merchant controls, time effects, and demographic variables, respectively.

The outcome variable is a binary indicator Deviated, which is equal to 1 if a consumer used a payment method for an in-person purchase that is not the same as their preferred method that they recorded initially before making any transactions. The variables of interest are Log of Transaction Size and Convenience Advantage. Transaction Size is included because understanding variation in purchase value is important for transaction-level payment behavior. As the size of the transactions varies, the costs and feasibility of the alternative payment methods vary as well. As such, the amount of the purchase serves as an element that defines the context of payment choices.

In addition, Convenience Advantage is included to observe the relationship between subjective evaluation of payment method and the probability of deviation. It measures the difference in convenience rating between the stated payment method of preference and the average of the other two methods. It captures the subjective understanding of how much more convenient the preferred in-person payment method is than the alternatives. Given the assumption that within-individual attribute ratings have little variation over time, the variable allows for a valid estimation of the relationship between preference consistency and convenience advantage.

Similar to Greene et al. (2026), Merchant Type is also used as a variable of interest for the analysis, although it is primarily a control variable in their paper. In our analysis, the variable is used to control for the context of transactions recorded in the dataset which may have a confounding relationship with the regressors and the outcome variable. However, it is also a variable of interest because it represents an important element of consumer payment behavior in the context of deviation and payment choice, since the type of vendor where a transaction is made sets the place, type, and the relative importance of that payment. The demographic and year variables control for temporal trends and idiosyncratic characteristics that may have a confounding association with the outcome and regressor variables.

4.3.1 Probit of Deviation to Credit Card and Debit Card use for cash-preferring consumers

The following parts of Stage 2 analysis cover the equations for directional deviation from cash among consumers who prefer cash but make purchases with either a credit card or a debit card. This is expressed in Equation 6 as the probability of deviation From Cash equaling 1. The second part of the directional deviation model estimates the probability of consumers who prefer credit cards and consumers who prefer debit cards actually using cash in an in-person transaction. This model is expressed in Equation 7 as the probability of deviation From Card equaling 1. It is important to note that although Equation 6 and Equation 7 have the same specification for both estimates, they measure the probability of deviation for debit and credit separately.

Similar to Equation 5, the model estimates the probability of deviation, but specifies the direction of this deviation. The estimations for deviation from cash to credit card and from cash to debit cards are measured separately but are specified in the equation similarly due to the same variable structure. As such, From Cash equals 1 if a cash-preferring consumer i uses the specific card payment method being estimated, credit or debit, for an in-person transaction t, and 0 otherwise. The Cash to Credit and Cash to Debit outcomes are estimated separately. Additionally, the regressor and control variables are the same as in Equation 5 and serve the same purpose. The only differentiating variable is From Cash which is a binary variable that specifies the direction of deviation.

4.3.2 Probit of Deviation to Cash use for Credit-preferring and Debit-preferring consumers

The model in Equation 7 also estimates the probability of deviation but, in particular, estimates deviation from a credit or debit card preference to cash. Importantly, the estimates for consumers that prefer credit but paid with cash and for consumers that prefer debit but paid with cash are measured separately, although they are specified in Equation 7 similarly due to the same variable structure. As such, From Card equals 1 if a consumer who prefers the specific card type being estimated, credit or debit, uses cash for an in-person transaction t, and 0 otherwise. The Credit to Cash and Debit to Cash outcomes are estimated separately. Similar to Equation 6, the regressor and control variables are the same as in Equation 5 and serve the same purpose. The only differentiating variable is From Card which is a binary variable that specifies the direction of deviation. The results for the above-listed regressions are presented in the following Section 5.

Section 5: Results

5.1 Stage 1: Conditional Logistic Regression Analysis

The following results follow the two-stage methodological approach described in Section 4. Tables 3 and 4 present the results for Stage 1 of the analysis using conditional logistic regression. Results for the probit regression of Stage 2 are described in Table 6, as well as Table 7 and Table 8, which present the results for directional deviation From Cash and From Card, respectively.

For both Table 3 and Table 4, the conditional logistic expression measures the results in log odds (utility scale), which are conditional logit coefficients and visible in column 1. In contrast to research studies like Koulayev et al. (2016) that observe coefficients and elasticities, our study reports marginal effects in column 2 for ease of interpretation of the results. As such, marginal effects represent the estimated relationship between a one-unit increase in consumer’s subjective rating of an attribute for a given payment method and the probability that the given payment method is used in an in-person transaction.

The results in Table 3 show that all attributes for the three payment methods have positive and statistically significant coefficients. Holding all other attributes constant, convenience has the largest association. A one-unit increase in an individual’s rating of convenience for a payment method is consistent with a roughly 0.71 percentage point increase in the probability of using that payment method in a transaction. The attribute with the second largest coefficient is Speed. A one-unit increase in Speed rating for a payment method correlates with an increase in the probability of using that payment method by roughly 0.26 percentage points.

Further, Cost is also positively associated with an increase in choice probability, as a one-unit increase in the rating is consistent with a 0.23-percentage-point higher likelihood of payment method use. On the other hand, Acceptance and Security had the lowest correlations with the propensity to use the higher-rated payment method. As such, a one-point increase in Acceptance rating coincides with a roughly 0.19 percentage point increase in likelihood of payment method use. Finally, Security has the weakest correlation out of all the attributes, with an increase in payment method use probability by 0.16 percentage points.

The results in Table 3 demonstrate a baseline conditional logit that associates an isolated relationship of attribute rating increase with payment method usage probability. Importantly, it supports the theoretical framework that the initial subjective ratings of a payment method predict the probability of its use in an actual transaction. For instance, for a consumer who rated cash’s convenience 3 out of 5, a one-point increase in the rating to 4 out of 5 would be associated with an approximately 0.71-percentage-point increase in the probability of using cash.

The baseline model identifies the relationship between attribute rating and payment method use probability in accordance with the RUM framework. However, other individual-level and transaction-level factors may contribute to this relationship and distort or bias the estimates in this regression. Thus, it is important to include relevant alternative-specific controls for these factors, given that each consumer faces three alternative payment methods for every transaction.

As such, inclusion of interaction control variables accounts for the difference that the transaction size can have when making a purchase with a credit card and debit card, while taking cash as the reference point. We also include the interaction control variable of credit card rewards as an instrument-specific variable for credit cards, since rewards can serve as an external incentive when choosing between payment methods. Table 4 demonstrates the results of the conditional logistic regression that adds these control variables to the baseline conditional logistic regression from Table 3.

The inclusion of interaction variables as controls confirms the results in Table 3 and supports the main findings from the baseline conditional logistic regression. As such, Table 4 also demonstrates positive and statistically significant coefficients for all variables. However, the overall trend in Table 4 is that the inclusion of control variables attenuated log-odds coefficients and marginal effects for most attributes except Cost, although the positive and statistically significant relationship for all attributes persists.

For attributes that attenuated, for instance, the coefficient for Convenience decreased from 0.496 to 0.3717, while the marginal effect decreased to 0.00629, indicating that a one-unit increase in the Convenience rating is associated with a 0.63-percentage-point increase in payment method choice likelihood. Conversely, Cost is the only attribute that increased in marginal effects and log-odds coefficient after controlling for non-attribute factors. Accordingly, the marginal effect for Cost rose to 0.00507 from 0.00231 in the baseline model. This suggests that a one-unit increase in the Cost rating is associated with a 0.51-percentage-point higher propensity to use the higher-rated payment method.

The control variables are also statistically significant and have a positive relationship with payment choice likelihood. The interactions of Log Transaction Amount with Credit Card and Debit Card are positive, indicating that larger transaction amounts are associated with approximately a 0.25-percentage-point higher probability of choosing credit and a 0.15-percentage-point higher probability of choosing debit relative to cash, respectively. Additionally, the Credit-specific interaction for Credit Card Rewards is associated with approximately a 0.55-percentage-point increase in the likelihood of using credit relative to cash, which is the largest marginal effect among the control variables.

Thus, the results in Table 4 demonstrate that the addition of interacted control variables adds a more robust specification to the regression by controlling for transaction-specific characteristics, which helps account for potential omitted-variable bias. The estimates from Table 3 and Table 4 directly connect to the research question by demonstrating that payment method choices have a systematic positive relationship with consumers’ subjective ratings of payment method attributes.

Crucially, the results show that Convenience has the largest estimated association with payment choice. Moreover, the overall finding is that payment choice is systematically associated with consumers’ subjective evaluations of payment method attributes. This directly connects to the topic of payment system modernization in the US. Understanding how Americans form their payment choices is essential, and in order to gain insight into these choices, examining demand-side factors that are associated with payment behavior is important for policy and businesses to form the right incentives for the consumer use of the new payments system. Thus, Stage 2 of the analysis observes the instances of consumer deviation from their ex ante preferred payment methods and identifies the factors that are associated with this behavior.

5.2 Stage 2: Probit Regression Analysis

Table 5 shows a cross-tabulation of payment methods that were stated as ex ante preferred and payment methods that were actually used in a transaction. The table highlights in grey the percentage of transactions that were made with the payment method that was initially stated as preferred. In addition, it demonstrates that there is roughly a 25% or greater level of deviation away from the stated preferred method across the three payment instruments.

Results from Table 5 show that consumers that prefer credit have the highest level of preference persistence. Overall, preference persistence is highest among credit-preferring consumers at 75.1%, followed by debit-preferring consumers at 66.6% and cash-preferring consumers at 62.4%. The corresponding deviation rates are 24.9%, 33.4%, and 37.6%, respectively. These descriptive patterns directly motivate the directional deviation analysis in Tables 7 and 8.

Moreover, Table 5 demonstrates the direction of these deviations. Among transactions made by cash-preferring consumers, 14.6% were paid with credit and 23.1% with debit, indicating that deviation toward debit was approximately 8.5 percentage points more common than deviation toward credit. Among transactions made by credit-preferring consumers, 18.9% were paid with cash and 6.0% with debit, meaning that deviation toward cash was more than three times as common as deviation toward debit. Among transactions made by debit-preferring consumers, 23.5% were paid with cash and 9.9% with credit, showing that deviation toward cash was more than twice as common as deviation toward credit.

Out of all the preference groups in the sample, the highest deviation level is among cash-preferring consumers, with 37.6% of their transactions made using a method other than cash. The lowest deviation level is among credit-preferring consumers, with 24.9% of their transactions made using a method other than credit.

Notably, card-preferring consumers generally showed higher preference persistence, as 75.1% of transactions by credit-preferring consumers were made with credit and 66.6% of transactions by debit-preferring consumers were made with debit. These statistics directly relate to Stage 2 of our analysis that looks at the directional deviation probability. Table 6 presents the results for the probit regression in Stage 2 of the analysis. Each column reports marginal effects for the three types of transactions: column 1 for transactions made by consumers who prefer cash, column 2 for transactions made by consumers who prefer credit cards, and column 3 for transactions made by consumers who prefer debit cards.

The results in Table 6 reveal that most of the variables are statistically significant with the exception of Travel and Food for transactions made by consumers that prefer cash, as well as Gas stations for both those who prefer cash and those who prefer credit. Transaction Size is positively correlated with the probability of deviation for transactions made by cash-preferring consumers, which suggests that a one-unit increase in the log of transaction size is consistent with a 9.1 percentage point increase in the likelihood of deviation for those individuals.

On the other hand, transaction size is negatively correlated with the probability of deviating for transactions made by credit-preferring and debit-preferring consumers, indicating that a one-unit increase in the log of transaction size is associated with decreases in deviation probability of 8.6 percentage points for credit-preferring consumers and 4.5 percentage points for debit-preferring consumers. Moreover, convenience advantage is negatively associated with the deviation probability for transactions made by all consumers regardless of their payment method preference. A one-unit increase in the subjective Convenience Advantage rating is associated with decreases in deviation probability of 4.7 percentage points for cash-preferring consumers, 4.9 percentage points for credit-preferring consumers, and roughly 4.7 percentage points for debit-preferring consumers.

Importantly, marginal effects on merchant group variables suggest heterogeneous relationships between deviation probability and merchant group. For instance, transactions at Department Stores have a positive relationship with the probability of deviation for cash-preferring and debit-preferring individuals, since a purchase at the department store is consistent with an 8 percentage point increase in propensity to deviate for cash-preferring consumers and roughly a 6.4 percentage point increase in the deviation propensity among transactions by debit-preferring consumers, while this relationship is negative for transactions by credit-preferring consumers.

Furthermore, services are positively related with transactions by both credit-preferring and debit-preferring consumers, associating with a 23.2-percentage-point and 27.0-percentage-point increase in deviation probabilities, respectively. However, for cash transactions, this relationship is negative, demonstrating that services are associated with a decrease in deviation probability by 9.7 percentage points. Additionally, transactions in Travel are consistent with a decrease in deviation probability for credit-preferring consumers by 4.8 percentage points but, conversely, an increase in deviation probability for debit-preferring consumers by almost 19 percentage points. Finally, transactions involving Utilities are associated with a 23.4 percentage point decrease in deviation probability for cash-preferring consumers. In contrast, Utilities transactions are associated with increases in deviation probability of 22.9 percentage points for credit-preferring consumers and 11.3 percentage points for debit-preferring consumers.

Moreover, purchases in Entertainment are associated with an 11.6-percentage-point decrease in deviation propensity for cash-preferring consumers, but increases of 26.7 percentage points for credit-preferring consumers and 38.8 percentage points for debit-preferring consumers. Notably, transactions at all Other merchant groups (which include person-to-person transactions, gifts, medical visits, etc.) have the highest magnitude of increase in deviation likelihood by 47.5 percentage points in credit-preferring consumers and 45.2 percentage points in debit-preferring consumers, but are consistent with lower propensity to deviate for cash-preferring consumers by 25.9 percentage points.

Stage 2 analysis includes two parts, where the first part examines the probabilities of deviation for each payment method ex ante preferred by the individuals in the sample. The second part examines the two directions of deviation: from cash to card payments and from card payments to cash. The structure of the second part is as follows: Table 7 shows deviation from cash to card payment, which includes credit (in column 1) and debit (in column 2). Table 8 shows deviation from card payments to cash, which includes deviation from credit (in column 1) and deviation from debit (in column 2).

Note: Observation counts differ between the two directional regressions because some categorical-variable groups perfectly predict the outcome and are omitted during probit estimation.

First, the propensity to deviate to credit and debit is negatively associated with Convenience Advantage for transactions made by cash-preferring consumers. A one-unit increase in Convenience Advantage is associated with a 1.72-percentage-point decrease in the probability of deviating to credit and a 3.04-percentage-point decrease in the probability of deviating to debit. In contrast, Transaction Size is positively associated with deviation towards card payments, with an increase in the deviation likelihood towards credit by 5.7 percentage points and towards debit by a smaller amount of 3.7 percentage points among cash-preferring consumers.

Transactions involving Services had a negative relationship with deviation probability to debit, while transactions relating to Travel and Department Stores had a positive relationship with likelihood of deviation to credit cards among cash-preferring consumers. The Merchant Groups that had statistically significant results for both deviation directions include Utilities and Other. For instance, purchases relating to Utilities are consistent with a decrease in deviation likelihood to use of credit cards by 10 percentage points and to use of debit cards by 15.5 percentage points among cash-preferring consumers. Purchases at Other merchant categories similarly decrease the deviation probability by 7.61 percentage points from cash to credit cards and by 18.6 percentage points from cash to debit cards.

Furthermore, Table 8 presents the results for the probit regression of Convenience Advantage, Log of Transaction Amount, and Merchant Group categories on the probability of deviation to cash from credit and debit cards.

First, Convenience Advantage has a negative relationship with deviation probability, consistent with the direction of the results in Table 7. A one-unit increase in consumers’ subjective Convenience Advantage rating is associated with a 3.24-percentage-point decrease in the probability of deviating to cash for credit-preferring consumers and a 3.89-percentage-point decrease for debit-preferring consumers. However, in contrast to the results from Table 7, the Log of Transaction Amount is negatively associated with deviation to cash for card-preferring consumers. For instance, a one-unit increase in the log of Transaction Amount is associated with an 8.37-percentage-point decrease in the probability of deviation to cash among both credit-preferring and debit-preferring consumers.

It is important to note that Grocery Stores serve as the reference category in the results for merchant groups in all tables. The results for Merchant Categories provide especially valuable insight into the relationship between consumer payment behavior and transaction context. Services, for instance, are positively related to deviation probability to cash, as a transaction involving Services is correlated with an increase in the deviation likelihood among credit-preferring consumers by 26.8 percentage points and by 23.5 percentage points among debit-preferring consumers. Utilities display a similar positive relationship, with transactions in this category associated with a 17.7-percentage-point increase in deviation probability for those who usually prefer credit and a 14.2-percentage-point increase for those who usually prefer debit.

Finally, purchases involving Entertainment suggest a strong positive relationship with deviations to cash. A transaction involving Entertainment is associated with a 30.2-percentage-point increase in deviation probability for credit-preferring consumers and a 36.6-percentage-point increase for debit-preferring consumers. Finally, purchases relating to Other merchants had the highest magnitude of deviation probability increase in percentage points by roughly 50 percentage points among both credit-preferring and debit-preferring consumers.

Section 6: Discussion

6.1 Discussion of Stage 1: conditional logistic regression

6.1.1 Baseline Conditional Logistic Regression

The results in Section 5 present important findings as part of the Stage 1 analysis and reveal meaningful insights into the relationship between attribute rating and consumer payment choice. Primarily, utilizing the conditional logistic regression of payment method attributes and probability of use, we find that consumers’ payment behavior is structurally consistent with the subjective rating of payment methods. As such, the central finding is that, out of all the attributes, the consumer’s subjective rating of Convenience for a payment method is the largest predictor of the probability of using that payment method in an in-person transaction.

This is consistent with the findings of Schuh and Stavins (2015) and Koulayev et al. (2016), who also find that convenience (defined as ease of use in the Koulayev et al. paper) has the strongest relationship with consumer choice of payment instrument adoption and use. Moreover, this finding is consistent with the Random Utility Model adopted for the theoretical framework of this study. The RUM suggests that consumers have a higher likelihood of using a payment instrument from which they derive the highest utility. The strong relationship of convenience rating and the probability of payment method use indicates that individuals may place higher emphasis on convenience when choosing a payment method for a transaction.

6.1.2 Conditional Logistic Regression With Controls

Compared to the baseline regression, Table 4 with interacted control variables saw multiple insightful changes in the coefficients. As such, the magnitude of the marginal effects for Cost increased, while the magnitude on other attributes decreased. Conceptually, a one-unit increase in the rating for the attribute Cost means that the higher-rated payment method is subjectively evaluated as having better economic value. The magnitude of the Cost effect may be smaller in the baseline because credit card rewards can change the effective economic value of using credit through cash back, reward redemption, and other benefits. Without controlling for rewards separately, part of this relationship may also be reflected in consumers’ subjective Cost ratings. As such, the inclusion of the Credit Card Rewards interaction may separate the relationship associated with rewards from the relationship between the Cost rating and payment choice.

In addition to that, the inclusion of the Transaction Size interactions with Credit and Debit also may contribute to the coefficient increase. Consumers may perceive higher transaction amounts as having higher costs. For instance, they may consider card-related fees or, in the case of credit cards, higher credit statements and potential interest. Thus, controlling for the size of the transaction in the sample may separate the effect of consumers’ consideration about fees and interest to isolate the relationship of attribute rating with probability of payment method use. All coefficients for the other attributes, on the other hand, have slightly attenuated compared to the baseline. Controlling for transaction size and rewards similarly isolates the relationship of the main variables of interest and helps address possible omitted-variable bias.

The findings also align with the conclusions of the Schuh and Stavins (2015) paper, as the results in Table 4 also find that Security is the weakest predictor of consumer payment method use. This may seem counterintuitive because the regulators of payment systems like the Fed, as well as developers of payment instruments, often emphasize security as a priority (Federal Reserve Financial Services, 2022). However, given that this is the subjective rating, people may often discount the importance of security in transactions as it is either assumed by default or taken as a given and a known risk.

Notably, security also decreases after controlling for transaction size and rewards control. One possible explanation is that in the baseline model the variable could have been accounting for larger transactions, where consumers might worry more about the security of their finances than about smaller routine purchases. After including the transaction size and rewards controls, the Security coefficient decreased substantially from 0.113 to roughly 0.067, which further supports the possibility that some consumers may see security as a default expectation rather than a differentiating factor between payment methods.

Moreover, the relationship between the Convenience rating and payment method use probability is also notable and aligns with findings of Koulayev et al. (2016) and Schuh and Stavins (2015). A possible explanation for the relatively stronger association of convenience with payment method use is that convenience is comparatively more variable by payment method than other attributes. Different payment instruments have different use cases, different features, and different physical characteristics. Cash, for instance, may be harder to carry in large sums while credit and debit are easier. However, credit and debit may take more steps to process during a purchase due to a higher number of procedures needed for a purchase, such as swiping or inserting, entering the card PIN, authorizing the payment, etc.

6.2 Discussion of Stage 2: Probit of deviation from the preferred payment method

6.2.1 Probit of Deviation From Ex Ante Preferred Payment Method

The second stage of the analysis provides an important insight into the changes in the consumer payment choice through the lens of deviation from an ex ante preferred in-person instrument in association with factors such as subjective attribute rating, and transaction context (including where the transaction is made and the size of that transaction). A notable finding is that a one-unit increase in the log of transaction size is associated with a 9.1 percentage point increase in deviation probability for cash-preferring consumers, but conversely is associated with a decrease in that probability for credit-preferring and debit-preferring consumers by 8.6 percentage points and 4.5 percentage points, respectively. This is consistent with the results reported by Greene et al. (2026) who also find that likelihood of deviation to cash falls as transaction size increases.

This asymmetry in the direction of deviation may be related to differences in the physical constraints and features pertaining to these different types of payments. First of all, the increased likelihood of deviation for cash-preferring consumers is consistent with physical constraints of cash. Naturally, it becomes harder to carry cash and store it safely and effectively as the amount of cash needed for a transaction increases. Consumers who usually prefer cash may still be incentivized to use a different payment method due to general tradeoffs of carrying and transacting cash in large quantities. Thus, consumers might deviate to card payments given the physical constraints of cash, since cash may offer lower practical purchasing capacity than credit or debit cards for the same level of effort and physical space.

On the other hand, credit-preferring consumers showed a larger percentage point decrease in deviation likelihood as Transaction Size increased than debit-preferring consumers. This may be related to the additional features that credit cards offer over debit cards. As such, credit cards offer credit lines, essentially allowing people to borrow money. In contrast to cash and debit, consumers that use credit increase their purchasing power when making transactions, leading to credit cards becoming more frictionless and useful for larger-size transactions. Debit-preferring consumers have lower overall preference persistence relative to credit-preferrers, but Transaction Size is still negatively associated with deviation for them, which may reflect the practical advantage of debit cards over cash for larger purchases.

Beyond Transaction Size, the Convenience Advantage results provide a second connection between Stage 1 and Stage 2. In Stage 1 of the analysis, we identified that Convenience was the strongest predictor of the payment method use likelihood. This was important because Convenience had the strongest estimated relationship with payment choice compared to other attributes like Cost and Acceptance, which may indicate that consumers put greater weight on effortlessness and ease of use. The second stage of the analysis reinforces the importance of this result by showing that a higher subjective rating of convenience for a payment method to alternatives is associated not only with its use but also with lower deviation from it.

Most importantly, the relationship between Convenience Advantage and deviation probability is roughly the same across the three payment method alternatives. Mainly, the rating of the relative benefit of the preferred method over the alternatives and the likelihood of deviation has a similar relationship for all three consumer types: cash-preferring, credit-preferring, and debit-preferring. A one-unit increase in convenience advantage rating is associated with a roughly 5 percentage point decrease in deviation likelihood for all transactions regardless of the consumer’s ex ante preference. This demonstrates that consumers’ rating of convenience advantage has an important relationship with preference consistency in consumers. This suggests that when people form a subjective opinion about an instrument as user-friendly, efficient, and high-velocity, they exhibit stability in preferences, as convenience becomes important similarly across individuals irrespective of the physical characteristics and features of the preferred payment method.

The analysis of merchant context also offers meaningful insights on deviation from consumer preferences. Table 6 reflects the asymmetry in the relationship between transaction context and the deviation likelihood for cash and card payments. For instance, deviation among cash-preferrers had a positive and statistically significant relationship with Department Store transactions. Travel and Gas Stations also had positive coefficients, but these estimates were not statistically significant. One possible explanation for some of these patterns may be the rewards and perks offered through card-linked programs, which are commonly associated with merchant categories such as department stores and travel services (The CardLinx Association, 2019).

The opposite pattern appears for several merchant categories among card-preferring consumers. One possible explanation is that some Food and Entertainment transactions may involve smaller or more routine purchases where cash remains practical. Moreover, the value of rewards may be less important relative to other payment considerations for some of these smaller and more routine transactions. Given the size of the transaction, the need for a credit line is also less relevant, which suggests why card-preferring consumers might deviate to cash. On the other hand, Utilities and Other transactions, which include gifts/donations and person-to-person payments, may involve different acceptance conditions or transaction formats that make card use less convenient in some settings.

6.2.2 Probit Analysis for Deviations From Cash to Credit and Debit Card Payments

Furthermore, a detailed look at the directional deviation in Table 7 from cash to credit and debit is especially valuable because it separates the relationships between deviations to debit and credit, allowing for a more meaningful interpretation of the results. As such, the coefficients on transaction size confirm the conclusions drawn from Table 6 that cash constraints may incentivize consumers to switch to card payment, but interestingly Table 7 shows that the positive relationship between Transaction Size and deviation is larger for credit than for debit. This can be interpreted as credit cards having multiple advantages over debit such as offering a credit line, rewards, and other perks that may influence its relative utility for larger transactions.

Beyond Transaction Size, Table 7 also provides a more detailed look at the relationship between Convenience Advantage, merchant context, and the direction of deviation. As such, the findings reveal that transaction size has a positive relationship with deviation for cash-preferring consumers to credit and debit, showing that increases in the size of the purchase are consistent with deviation. Notably, the relationship is larger for deviation to credit than to debit, which may be related to the additional advantages of a credit line, cash back, and rewards that credit cards often offer. The Convenience Advantage coefficient is negative for both To Credit and To Debit deviations, but the relationship is larger in magnitude for deviation to debit. This suggests that deviation to credit may be less sensitive to the Convenience Advantage of cash, potentially reflecting additional features of credit cards such as rewards and access to credit.

For merchant context factors, we find that cash-preferring consumers deviate to credit cards in settings where card-linked offers and loyalty rewards may be more relevant, such as Travel and Department Stores. One possible explanation is the additional incentives that rewards and loyalty programs provide, such as discounts and frequent-flyer points. However, for merchant groups like Utilities and Other (which includes P2P and Donations), there is a negative relationship with both directions of deviation from cash. Food also has a negative relationship with deviation to debit, while its relationship with deviation to credit is not statistically significant. This may suggest that the practical advantages of card payments do not necessarily outweigh cash preference in these settings, as cash-preferring consumers may place greater weight on potential card-related costs and interest when the value of rewards is relatively less important.

6.2.3 Probit Analysis for Deviations From Card-Based Payments to Cash

The second part of the Stage 2 analysis includes estimation of the deviation probability for credit-preferring and debit-preferring consumers to cash. In particular, the marginal effects reveal a similar negative relationship between transaction size and deviation to cash for both credit-preferring and debit-preferring individuals: a one-unit increase in the log of transaction size is associated with an approximately 8.4 percentage point lower propensity to deviate to cash, regardless of the preferred card type. This follows a general trend seen in the Stage 2 results that higher transaction sizes favor card-based payments, so that the amount of the purchase has a similar relationship with preference consistency for those who prefer credit cards or for those who prefer debit cards. This also reinforces the idea of the importance of the card-based payments’ relative purchasing capacity and frictionlessness for larger transactions.

Similarly, the relationship of convenience advantage for card-preferring consumers behaves similarly to that of cash-preferring individuals, and thus confirms the finding from Table 6 that Convenience Advantage has a similar relationship with preference consistency across consumers regardless of their preferred payment method. In particular, for those individuals that prefer cards, convenience advantage is related to preference consistency. This may reflect that a higher subjective rating of convenience compared to cash is associated with lower deviation. Within the RUM framework, this is consistent with utility-based models, as the payment method that is ex ante preferred for its seamlessness, transaction velocity, and user-friendliness increases the relative utility of credit or debit cards for their respective consumers compared to the alternatives. Thus, the subjective rating of frictionlessness and ease of use may diminish the need for choosing cash in most transactions for these consumers, when controlling for demographic and transaction context factors.

Although the relationship between Convenience Advantage and deviation likelihood is similar across all three preference groups, the relationship between Merchant Group and deviation likelihood shows an overall pattern of asymmetry in the Stage 2 results. Mainly, for merchant categories where cash-preferring consumers show preference consistency, consumers that usually prefer credit or debit tend to deviate to cash. Moreover, these card-preferring consumers tend to deviate to cash at the same merchant categories. This payment behavior is consistent with the broader pattern in the results that deviation to cash occurs in settings where card use may offer fewer advantages, such as lower-value or more routine transactions, fewer opportunities for rewards, or greater concerns about fees and interest. These settings include Food, Services, Utilities, Entertainment, and Other merchant groups.

The pattern differs across the two card-preferring groups. Credit-preferring consumers show greater preference consistency in categories such as Travel and Department Stores, while debit-preferring consumers display a different relationship in these settings. Overall, the results show that merchant context has a distinct and asymmetric relationship with preference consistency depending on the consumer’s preferred payment method. The finding suggests that the context of the transaction has a distinct relationship with consumer payment method choice and preference consistency behavior.

Section 7: Conclusion

7.1 Summary of Findings

From a broader perspective, this study pooled cross-sectional data from the Survey and Diary of Consumer Payment Choice for 2021-2024 to analyze how consumer payment method choice is predicted by consumers’ subjective evaluations of the payment method attributes. These attributes measure the subjective consumer ratings on cost, security, and other characteristics that suggest the assessed benefit and practical value that consumers get from each payment method. Through a two-stage analysis, the study examines how these attribute evaluations predict consumer payment choice in Stage 1, and whether the propensity to deviate from the preferred payment method has a systematic relationship with the attribute evaluations and transaction context like its size or merchant category in Stage 2.

The methodological approach uses conditional logistic regression in the first stage and probit in the second stage of the empirical analysis to offer results that directly connect to the research question and answer it in two parts. The research question asks 1) whether U.S. consumers’ subjective evaluations of payment method attributes predict the use of that payment method for an in-person transaction? 2) whether the evaluations of payment method attributes, as well as transaction context like merchant category and transaction size, have a systematic relationship with the probability of deviation from the consumer’s ex ante preferred in-person payment method?

Taken together, the two stages provide evidence on both the formation of realized payment choice and the persistence of stated payment preferences. The results for Stage 1 suggest that all attribute ratings are statistically significant and positive predictors of payment method choice, which is consistent with the Random Utility Model that was used as the conceptual framework of the analysis. The rating of Convenience was found to have the largest estimated association with payment choice out of all the attributes, while Security had the smallest estimated association, which is consistent with Koulayev et al. (2016). The results indicate that consumers put greater weight on user-friendliness of payment methods but may perceive security as standard.

The analysis in Stage 2 shows that convenience advantage, which measures to what extent consumers rate convenience of their preferred method higher than the alternatives, is systematically associated with a lower propensity to deviate from ex ante preferences for in-person purchases similarly for cash-preferring, credit-preferring, and debit-preferring consumers. Moreover, the two-part analysis also reveals that transaction size is systematically associated with an increase in deviation for cash-preferring consumers but a decrease in deviation for credit-preferring and debit-preferring consumers, a pattern that may be consistent with the physical constraints of cash in larger transactions and aligns with Greene et al. (2026). Finally, merchant context followed an asymmetric pattern, as Food and Services were associated with deviation to cash among card-preferring consumers, while Department Store transactions were associated with higher deviation among cash-preferring consumers. The directional analysis further shows positive relationships between Travel and Department Store transactions and deviation from cash to credit. These patterns may reflect differences in rewards, transaction costs, interest, and other features associated with different merchant contexts.

Ultimately, this paper contributes to the literature in three ways. First, it links the two mechanisms of payment behavior: the predictors of payment method choice and factors of persistence in payment choice preferences. This is done by focusing on demand-side factors like subjective attribute ratings as main variables of interest, including a novel variable that connects attribute ratings with preference consistency, that are treated as behavior-side factors influencing the selection and persistence in use of a payment method. Second, it separates the card-based payments into two distinct groups of credit-preferring consumers and debit-preferring consumers, which allows for more granular analysis of preference deviation likelihood. Third, this study uses pooled SCPC/DCPC data across 2021-2024 that offers an individual-transaction-level analysis with a more comprehensive sample and larger temporal scope than prior literature that mostly used single-year and aggregate transaction samples.

7.2 Policy Implications

The results and analysis in this study reveal multiple insights that may have implications for the future of American payment system modernization. Most importantly, the results show that Convenience has the largest estimated association with payment choice among the attributes in Stage 1, while Convenience Advantage is consistently associated with preference persistence in Stage 2. This is useful to policymakers and financial companies, as it indicates the approach the modern payment system might take to align the American stakeholders’ incentives for effective adoption of the system.

Mainly, the results suggest that new payment methods may need to offer convenience that is competitive with existing payment instruments in order to encourage consumer use and preference persistence. This can be done by emphasizing improvements in characteristics that define convenience, such as payment velocity, ease of use, lower effort to carry, and better ability to keep and store these payment instruments. On the other hand, Security has a weaker estimated relationship with payment choice than Convenience, suggesting that improvements in security communication alone may have a smaller influence on observed consumer choice.

In addition, the role of the context of transactions in payment behavior can also be insightful for policy decisions. The most important finding is understanding the merchant categories that show a break from credit card loyalty, which was shown to be the most preference-consistent payment method in the sample. The results suggest that Food, Entertainment, Services, and Other transactions (a category that includes person-to-person payments) see a higher rate of deviation away from credit cards to cash, identifying merchant settings where consumers are more likely to break from their stated card preference. This may reflect differences in rewards, transaction costs, acceptance, or other characteristics of these transactions. As such, these results may point to an opportunity for payment systems that address some of the physical constraints of cash and costs associated with card-based payments. For instance, QR-Code payments linked directly to a consumer’s bank account could provide an alternative worth testing in merchant settings where consumers are more likely to deviate from card payments to cash.

7.3 Limitations and Suggestions for Future Studies

Given the nature of the raw DCPC data, the cross-sectional data structure does not provide causal interpretation of the results but rather a suggestive, correlational analysis. Future studies can benefit from utilizing a quasi-experimental setup to identify causal determinants of consumer payment choice. Moreover, the variables on consumers’ ex ante payment method preference are recorded at the individual level while their payment behavior is recorded at the transaction level. This suggests that the analysis does not capture the variation in preferences of consumers by type of transaction but rather aggregates the preference for all transactions. Future papers that observe preferences for each transaction at the point-of-sale may allow for a more robust estimate of preference deviation.

In addition, the raw dataset does not capture what payment methods are readily available to consumers at each transaction, which limits the precision of the estimate on choice between alternatives. An empirical study that captures each consumer’s available options for every transaction could offer more precise measurements of payment choice. In addition, there is an aspect of endogeneity in the measurement of attribute ratings for each payment method. The attribute ratings are recorded before transactions are made, although, in reality, they may capture the influence of payment habits, as consumers that use cash regularly may rate cash attributes higher because of their frequent use rather than true evaluations. Future work could benefit from an experimental design or panel data to isolate attribute ratings from prior payment behavior.

Finally, the conditional logit model relies on the Independence of Irrelevant Alternatives assumption (IIA), which means that the relative probability of choosing between two payment methods is not influenced by the presence of the third alternative. This assumption presents a possible limitation, because consumers may see credit and debit cards as more similar substitutes for each other than either is to cash. Future research could address this by using nested or mixed logit models that allow for more flexible substitution between payment methods.

Section 8: Bibliography

1. Birigozzi, A., De Silva, C., & Luitel, P. (2025). Digital payments and GDP growth: A behavioural quantitative analysis. Research in International Business and Finance, 75, 102768. doi:10.1016/j.ribaf.2025.102768.
2. The CardLinx Association. (2019). Card-Linking and O2O Industry Survey 2019. https://digcomall.org/wp-content/uploads/2024/12/2019-survey-cardlinx_v8.pdf
3. Catalini, C., & Lilley, A. (2021). Why is the United States lagging behind in payments? SSRN. doi:10.2139/ssrn.3893937.
4. Cohen, R., Shy, O., & Stavins, J. (2022). Payments evolution from paper to electronic payments by merchant type (Research Department Working Paper No. 22-6). Federal Reserve Bank of Boston. doi:10.29412/res.wp.2022.06.
5. Federal Reserve Bank of Atlanta. (2025). Survey and Diary of Consumer Payment Choice [Data set]. https://www.atlantafed.org/research-and-data/surveys/survey-and-diary-of-consumer-payment-choice
6. Federal Reserve Financial Services. (2022, May 16). Federal Reserve survey: Consumers demanding faster payment options to fit their digital, always-on lives. https://www.frbservices.org/news/press-releases/051622-survey-consumers-faster-payment
7. Greene, C., & Stavins, J. (2020). Consumer payment choice for bill payments (Research Department Working Paper No. 20-9). Federal Reserve Bank of Boston. doi:10.29412/res.wp.2020.09.
8. Greene, C., Shy, O., & Stavins, J. (2026). Merchant steering of consumer payment choice (Research Department Working Paper No. 26-2). Federal Reserve Bank of Boston. doi:10.29412/res.wp.2026.02.
9. Han, P., & Wang, Z. (2021). Why is the U.S. lagging in adopting mobile payments? Economic Brief, 21-21. Federal Reserve Bank of Richmond.
10. Han, P., & Wang, Z. (2026). Technology adoption and leapfrogging: Racing for mobile payments (Working Paper No. 21-05R; revised February 2026). Federal Reserve Bank of Richmond. doi:10.21144/wp21-05.
11. Koulayev, S., Rysman, M., Schuh, S., & Stavins, J. (2016). Explaining adoption and use of payment instruments by U.S. consumers. The RAND Journal of Economics, 47(2), 293–325. doi:10.1111/1756-2171.12129.
12. McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Frontiers in econometrics (pp. 105–142). Academic Press.
13. Meridian International Center. (2025, October). The future of payments modernization and the impact of economic innovation for global prosperity. https://meridian.org/project/the-future-of-payments-modernization-and-the-impact-of-economic-innovation-for-global-prosperity/
14. Murat, Z. (2025). Investigating the post-pandemic effect of the use of mobile payments technology on household discretionary expenditure in the U.S. [Junior independent work]. Princeton University, Department of Economics.
15. O’Neill, K. (2023, September 6). FedNow: A once-in-a-generation payments innovation for the Fed. The Regional Economist. Federal Reserve Bank of St. Louis.
16. Sari, R. L., Habibi, A. B., & Hayuningputri, E. P. (2022). Impact of attitude, perceived ease of use, convenience, and social benefit on intention to use mobile payment. Asia-Pacific Management and Business Application, 11(2), 143–156. doi:10.21776/ub.apmba.2022.011.02.2.
17. Schuh, S., & Stavins, J. (2015). How do speed and security influence consumers’ payment behavior? Current Policy Perspectives, 2015-1. Federal Reserve Bank of Boston.
18. Sveriges Riksbank. (2025). The Swedish payments market is almost entirely digital. https://www.riksbank.se/en-gb/payments--cash/payments-in-sweden/payments-report-2025/trends-on-the-payments-market/the-swedish-payments-market-is-almost-entirely-digital/
19. The White House. (2025a, July 18). Fact sheet: President Donald J. Trump signs GENIUS Act into law. https://www.whitehouse.gov/fact-sheets/2025/07/fact-sheet-president-donald-j-trump-signs-genius-act-into-law/
20. The White House. (2025b, March 25). Modernizing payments to and from America’s bank account (Executive Order 14247). https://www.whitehouse.gov/presidential-actions/2025/03/modernizing-payments-to-and-from-americas-bank-account/
21. U.S. House Committee on Financial Services. (2025, July 17). Chairman Hill commends passage of landmark digital asset bills. https://financialservices.house.gov/news/documentsingle.aspx?DocumentID=410815

Acknowledgements

I am deeply grateful to Princeton University for giving me this incredible opportunity. I really appreciate the administration, the faculty and the community for their trust and support.

Many thanks to my family for their unconditional love and encouragement. I love you, Mom, Dad, and Kani! Your care and guidance always give me positivity and hope.

I would like to sincerely thank my advisor, Juan Camilo, for his helpful feedback, advice, and guidance during the writing of this paper.

I am grateful to Oscar Torres-Reyna for his time and help, and I greatly appreciate his patience and support during this process.

I wish to express my heartfelt gratitude to the Department of Economics for their work and support. Thank you to Andrea Wilson, Smita Brunnermeier, Gina Holland, and the department community for their time and effort to make this process clear and accessible.

I am sincerely thankful to all my friends from my Zee group, at Tower, across campus, and back home. You are a great addition to my life, and I will always cherish the memories we shared.

Thank you to Dean Wagner and Dean Luijendijk for their support and help throughout my years on campus.

I would like to thank Kevin Foster, Claire Greene, and Joanna Stavins for their insights and help with the DCPC data.

Senior thesis submitted to Princeton University, Department of Economics, in partial fulfillment of the requirements for the A.B. degree, April 9, 2026. Advised by Juan Camilo Castillo.