The Venmo Effect: Digital Payment Platforms (Part 3)

5.1 Participants Participation was limited to undergraduate students who currently attend a college or university in New England. The specific institutions represented are Bates College, Boston College, Bowdoin College, Colby College, Hamilton College, Middlebury College, Mount Holyoke College, Northeastern University, Smith College, Trinity College, Tufts University, University of Massachusetts-Amherst, University of Vermont, Wesleyan University, and Williams College. Participants were sourced through faculty at their institution’s economics and/or psychology departments. Prior to the experiment, all participants answered a series of screening questions, which allows us to screen out non-college students and non-Venmo users. To qualify as a Venmo user, each participant needed to have a Venmo account, have the Venmo app downloaded, and either have a nonzero Venmo balance or an alternate payment form linked to their Venmo account. The other disqualifying factors were being under 18 years old and not possessing either a debit card or a credit card. Researchers often use college students for lab experiments, both in economics and psychology research. This choice is often out of convenience; college students are located near the lab and generally willing to participate in exchange for low compensation. While using college students can sometimes be limiting for experimenters, in this case college students are the demographic of interest for testing the Venmo Effect. Most of the existing literature on willingness to pay with cash and credit cards is at least 20 years old, meaning that participants were from a different generation than today’s college students. Over time, young people’s relationship with and views of various payment forms may have changed, especially for digital payment platforms like Venmo. A 2017 survey finds that 65% of millennial respondents had used a mobile payment and 44% had used Venmo (Acker and 18 Murthy 2018); similarly, Pew Research (2018) finds that people in Generation X or younger made up 74% of respondents who use mobile payments. Additionally, the social effects of Venmo may be particularly relevant for college students as young people are more susceptible to influences of conspicuous consumption and peer pressure. For example, Gardner and Steinberg (2005) find that college-aged participants are 50% more likely to engage in risky behavior after being exposed to peers while they find no peer impact on adults. In the context of consumption decisions, research shows that college students see spending money as necessary to build friendships and feel like they belong (McClure and Ryder 2017). Overall, we use college students as we expect them to be most susceptible to the Venmo Effect due to their technological literacy and social impressionability. 5.2 Treatments Participants are assigned to one of five payment form treatment groups: 1) debit card, 2) credit card, 3) Venmo on the Private setting, 4) Venmo on the Friends Only setting, and 5) Venmo on the Public setting. Participants are told that they can only use their assigned form of payment for the rest of the experiment. The Venmo groups are asked to open the app and change their privacy settings to their assigned setting (Public, Friends Only, or Private). Privacy settings can easily be changed back after the experiment. Debit and credit are our comparison groups; debit is a proxy for cash, which would not have been feasible to exchange in an online experiment. Additionally, for participants in any of the three Venmo payment groups, there is a second treatment: priming. Regardless of assigned privacy setting, all the Venmo assignees are randomly assigned to either priming or no priming. Participants assigned to priming are asked to browse through the Public Venmo feed for 2 minutes; 19 participants assigned to no priming are not informed of this assignment and simply proceed with the experiment. Limiting our analysis to only Venmo, credit, and debit would only allow us to identify whether a Venmo Effect exists; we would not be able to examine the mechanisms behind it. Thus, we include the additional privacy setting and priming variations to determine whether social factors on Venmo contribute to a Venmo Effect. We believe this insight is worth the reduction in power caused by the additional treatments, though we recognize that a larger sample size would be ideal. 5.3 Experiment The experiment consists of ten incentive-compatible Becker, DeGroot, and Marschak (1964) lotteries for ten different low-cost household goods. As is standard in a Becker, DeGroot, and Marschak (henceforth BDM) lottery, participants first indicate their willingness to pay for a given item. Then, a price is randomly selected from a reasonable range; this range is not shared with participants in order to avoid anchoring effects (Wertenbroch and Skiera 2002). If the random price is less than or equal to the price the participant indicated as their maximum price, then they must purchase the good at the randomly drawn price. If the randomly drawn price is greater than the participant’s indicated maximum price, then they are not given an opportunity to buy the good. In our experiment, we conduct ten BDM lotteries for ten different common low-cost items: a notebook, a pen, an adhesive phone wallet, a can of soda, a pack of gum, a granola bar, a bottle of coffee, a bar of soap, a tube of toothpaste, and a cloth face mask. In order to avoid any “winner” or “loser” effects, we have participants list their maximum price for all ten items before drawing any prices; essentially, we run ten BDM lotteries simultaneously 20 (rather than back-to-back, as the previous lottery outcomes could influence the subsequent WTP estimates). Thus, only after the participants provide their WTP for all ten items is a random price between $1.00 and $5.00 drawn for each good. Participants are then shown all the random prices followed by whether they won (price ≤ WTP) or lost (price > WTP) each lottery. Appendix Figures 11 and 12 demonstrate how lottery results are communicated. There are three potential outcomes for each participant: 1) win zero lotteries, 2) win one lottery, or 3) win multiple lotteries. Those who do not win any lotteries are not given the opportunity to transact and thus proceed to final demographic questions. Those who win one lottery are asked to purchase the item they won at the randomly drawn price via their assigned payment form. For budget purposes and simplicity, those who win multiple lotteries are only asked to purchase one of their winning items, which is chosen randomly. After the random selection, they pay for their item at the randomly drawn price via their assigned payment form, just like those who win only one lottery. All transactions are completely secure for participants. Those assigned to Venmo are asked to pay a Venmo account created for the experiment (@WillingnessToPayResearch). Those assigned to pay with debit or credit are asked to pay through Square, a secure 3rd party credit card processing service, which opens in a new tab from the online experiment. If participants do not feel comfortable, they are allowed to abstain from paying; only 1.6% of participants who won at least one lottery chose not to pay. If a participant does not pay, they are asked why; the provided options are 1) feel uncomfortable providing payment information online, 2) do not think the item was worth the randomly drawn price, 3) never planned to submit a payment, or 4) do not possess the assigned payment form. The only response that does not compromise the participant’s WTP estimates is the first option, that 21 they did not feel comfortable entering their payment information online. The other answers suggest that the participant either misreported personal information or inaccurately estimated their WTP, both of which question the integrity of the data they provided. Thus, if a participant selects anything other than feeling uncomfortable submitting a payment online, the associated observation is dropped from the dataset. After paying (or, in the case that the participant had zero wins, directly after the lotteries), participants answer a series of follow-up questions. We ask participants about their normal spending habits, payment preferences, Venmo usage (how often they use Venmo, how many friends they have on Venmo, how much money is in their Venmo balance, etc.), and social media usage. Next, we ask demographic questions and what participants thought the purpose of the experiment was. Lastly, participants are debriefed, the complete text of which is shown in Appendix Figure 13. After completing the experiment, each participant receives a $5 Amazon e-gift card via email. If they purchased an item during the experiment, it is mailed to them the following business day. 5.4 Pilot Prior to running the full experiment, we executed a pilot to assess levels of attrition and payment compliance. We had originally planned to use Amazon Mechanical Turk to source participants, so the pilot consisted of 49 participants on mTurk. Out of the 49 total participants, 25 made it through the screening questions and won at least one lottery. All of those 25 claimed that they paid for their winning item, but only one participant actually submitted a payment. More importantly, we found that 21 of these 25 responses were completed by one person who created 21 different mTurk accounts to avoid the site’s attempt 22 limits. Given that this data was essentially unusable, we modified our experimental design prior to running the full experiment. First, we needed a new source of participants who are less dishonest and more naïve than those on mTurk. Thus, to collect the rest of our data, we decided to reach out directly to college students in New England in hopes that a more personal connection would foster sincerity in respondents. Prior to the pilot, while we were still designing the experiment, we conducted focus groups with Amherst College students to test whether the survey instructions were clear. During these sessions, most students said they would have submitted a payment if they were participating in the real experiment. Therefore, it seems that students who have a more personal connection to the researchers (e.g., are a part of the same institution or academic network) are more trusting and less dishonest. Since many students at Amherst College were already familiar with the experiment, we decided to use students at other colleges and universities in New England. Second, we added more attention checks and verification questions at the beginning of the experiment to screen out participants who will not pay. Immediately following the informed consent page, we added a question that asks participants if they would be willing to make a payment during the experiment. During each lottery, we added a verification question that would appear if a participant listed a willingness to pay over $20. We also added followup questions during the payment stage, so a participant can indicate why they did not pay; this way, we can tell whether the data is still usable even if the participant did not pay, like if they were simply concerned about the security of an online transaction. Adding these verification questions and attention checks increased compliance and led to more accurate WTP estimates than in the pilot. 23 5.5 Empirical Strategy The empirical strategy outlined below aims to answer three questions: 1) Does our experimental data show evidence of the Venmo Effect? 2) Which, if any, of our three proposed mechanisms are supported by the experimental data? 3) Does the Venmo Effect vary depending on the type of item being purchased? To answer the first two questions, we compare mean WTP across the eight treatment groups. For the general Venmo Effect, we compare mean willingness to pay for participants assigned to Venmo and those assigned to the debit or credit treatments; we expect participants to have a higher mean WTP with Venmo than with debit or credit cards. To assess the mental accounting mechanism, we measure the impact of Venmo in the absence of its social features by comparing mean WTP of participants assigned to the Venmo Private No Priming treatment to the mean WTP of those assigned to debit. The Venmo Private No Priming treatment group is the closest we can get to isolating the mental accounting mechanism of Venmo; participants in that group should not experience any social influences because they know that no one will see their transaction and are not instructed to view any other transactions prior to estimating their WTP. For the signaling mechanism, we compare mean WTP across the three privacy setting treatments: Private, Friends Only, and Public. This allows us to test whether users who know their friends will see their transactions have a higher WTP than those who know that strangers or no one will see their Venmo payments; we expect participants who use the Friends setting on Venmo to have higher WTP estimates than those who use the Public or Private settings. And, finally, to test the priming mechanism, we compare mean WTP for participants who were assigned to priming to mean WTP for those who were not assigned to priming; we expect participants who scroll through 24 the Venmo social feed to have higher WTP estimates than those who do not. As the impact of priming may depend on the privacy setting, we also examine how mean WTP varies within each of the privacy settings; for example, the difference between mean WTP for Venmo Public No Priming and Venmo Public Priming represents the impact of priming when using the Public setting. We expect the largest impact of priming to be for the Friends Only group. In order to account for variation in WTP caused by differences across the ten items and the potential influence of differing levels of experience with Venmo, we also conduct the following ordinary least squares regression: WTPi,j = 0 + 1 Crediti + 2 VenmoPrivateNoPrimingi + 3 VenmoPrivatePrimingi + 4 VenmoFriendsNoPrimingi + 5 VenmoFriendsPrimingi + 6 VenmoPublicNoPrimingi + 7 VenmoPublicPrimingi + 8 Itemj + 9 Controlsi + i,j where i represents an individual respondent and j represents the lottery item. WTPi,j, the outcome variable, is the maximum price listed by the participant. The variables Crediti, VenmoPrivateNoPrimingi, VenmoPrivatePrimingi, VenmoFriendsNoPrimingi, VenmoFriendsPrimingi, VenmoPublicNoPrimingi, and VenmoPublicPrimingi are binary indicators for whether the individual was assigned to that given treatment arm; the debit treatment is omitted. Itemj represents item-level fixed effects and Controlsi are control variables that measure Venmo usage, specifically age of Venmo account, Venmo balance, number of friends on Venmo, percentage of spend via Venmo, number of linked payments on Venmo, monthly transactions, monthly inflow, and monthly outflow. This regression estimation allows us to answer our empirical questions through postestimation testing. For the general Venmo Effect, the difference between the Venmo WTP estimates and the debit WTP estimates are measured by 2, 3, 4, 5, 6, and 7. The 25 existence of the Venmo Effect would be supported if these coefficients are positive and statistically significant. For the mental accounting mechanism, the coefficient of interest is 2. For the signaling mechanism, we compare jointly 2, 3 with 4, 5 and with 6, 7. For the priming mechanism, we compare jointly 2, 4, 6 with 3, 5, 7. For the within-privacy setting analysis, we compare the respective priming and no priming coefficients; for example, we compare 2 to 3 to see the impact of priming on WTP for the Private setting. To test whether different types of items may have different Venmo Effects, we split the ten items into four categories: 1. Food/Drinks (the can of Coca Cola, granola bar, pack of gum, and bottle of coffee), 2. Office Supplies (the notebook, pen, and phone wallet), 3. Toiletries (the toothpaste and soap), and 4. COVID-Related (the cloth face mask). We repeat the previous regression estimation omitting the fixed effects and controls but introducing indicator variables for each category and interaction terms between the treatment and category variables. We compare the coefficients for the interaction terms across payment types within a given category of items. We expect that items for which Venmo is typically used, like food and drinks, will have a larger Venmo Effect than items less commonly purchased on Venmo, like office supplies or toiletries. 6. Results and Discussion 6.1 Summary Statistics Summary statistics for the participants in our experiment are presented in Table 1. A total of 533 participants attempted the experiment. We keep observations only from the 261 participants who completed the experiment; this excludes anyone who was screened out, failed attention check questions, or quit partway through. We drop observations from any participants who misreported paying for their winning item (n = 6) or who did not pay for 26 any reason other than security concerns (n = 1). We also drop observations from participants who put zero for all of their WTP estimates (n = 11) as they may have been understating their true willingness to pay to avoid being asked to purchase an item. To further eliminate potentially dishonest respondents, we drop observations of any participants whose answers to the initial screening questions were inconsistent with their answers to final demographic questions (n = 6); specifically, we drop observations of participants who initially said they have one or more debit cards but later said they do not have a debit card (or a similar inconsistency for credit card or Venmo usage). Additionally, we drop both observations from one participant who completed the survey twice (n = 2). We are left with 2,350 observations from 235 unique participants. The mean WTP across all treatment arms is $1.13 (s.e. ($0.051)) and the median is $0.50. We drop one outlier WTP estimate of $100, resulting in 2,349 observations with a mean WTP of $1.09 (s.e. ($0.029)) and median WTP of $0.50. Table 1 shows mean values of selected variables across the eight treatment groups. The selected variables include demographic information, measures of Venmo usage, measures of social media usage, and mean willingness to pay estimates. In general, there are not statistically significant differences in observables across the eight treatment groups but there are a few exceptions that do not seem to follow any trends across treatment groups or variables. For a complete set of pairwise t-tests, see Appendix Table 1. The average participant from our sample is more female and wealthier than the average college student in the United States. The gender composition of the sample is 62.6% female, 35.3% male, 2.1% nonbinary. A majority of participants have a household income greater than $100,000, while the remaining participants are roughly evenly distributed between $0 and $100,000. Participants represent 15 different colleges and universities in 27 New England, but three institutions account for a majority (61.3%) of participants; University of Massachusetts-Amherst comprises 32.8% of the total sample, Williams College comprises 14.9%, and Colby College comprises 13.6%. Participants represent all undergraduate years of study, with the average year of study being roughly between a sophomore and junior. The average participant is approximately 20 years old. Participants are majority white; 68.9% of participants indicated that they identify as white, compared to 29.8% identifying as Asian, 3.4% identifying as Black, and 0.4% identifying as American Indian or Alaska Native. Only 7.2% of all participants indicated that they are of Hispanic, Latino, or Spanish origin. Table 1: Summary Statistics Across Treatment Groups N Demographic Variables Age Year Of Study Female White Household Income Venmo Usage Venmo Balance Years with Venmo Account Number of Friends on Venmo App Venmo Transactions Per Month Percentage of Money Spent via Venmo Social Media Usage Number Of Social Media Sites Used Average Number of Hours Spent on Social Media Per Day (2) Mean/SE 50 Venmo Private No Priming (3) Mean/SE 27 Venmo Private Priming (4) Mean/SE 22 Venmo Friends No Priming (5) Mean/SE 30 Venmo Friends Priming (6) Mean/SE 23 Venmo Public No Priming (7) Mean/SE 24 Venmo Public Priming (8) Mean/SE 26 20.424 [0.238] 2.758 [0.180] 0.545 [0.088] 0.818 [0.068] 3.759 [0.332] 20.180 [0.166] 2.720 [0.149] 0.700 [0.065] 0.640 [0.069] 4.000 [0.229] 20.222 [0.235] 2.630 [0.221] 0.667 [0.092] 0.778 [0.082] 3.520 [0.417] 19.909 [0.217] 2.364 [0.224] 0.500 [0.109] 0.500 [0.109] 3.429 [0.313] 19.900 [0.241] 2.367 [0.222] 0.800 [0.074] 0.633 [0.089] 4.333 [0.305] 20.087 [0.288] 2.696 [0.255] 0.609 [0.104] 0.739 [0.094] 3.750 [0.383] 20.458 [0.282] 2.875 [0.243] 0.542 [0.104] 0.708 [0.095] 4.048 [0.348] 20.038 [0.232] 2.615 [0.193] 0.538 [0.100] 0.692 [0.092] 3.957 [0.330] 147.302 [98.291] 2.970 [0.236] 104.182 [17.282] 4.970 [2.062] 12.970 [1.476] 83.029 [23.943] 3.080 [0.173] 96.640 [10.882] 2.718 [0.305] 14.651 [2.031] 40.069 [12.291] 3.370 [0.234] 109.407 [15.281] 2.926 [0.420] 15.222 [2.137] 139.936 [51.672] 2.571 [0.289] 79.727 [14.672] 1.977 [0.306] 17.955 [3.953] 48.050 [14.647] 2.724 [0.253] 69.000 [10.986] 3.267 [0.758] 17.600 [2.312] 153.994 [89.637] 2.909 [0.254] 135.522 [24.341] 4.261 [0.801] 21.913 [4.099] 81.218 [45.619] 3.000 [0.314] 105.708 [17.329] 2.792 [0.420] 15.229 [2.856] 105.323 [44.660] 2.808 [0.266] 112.615 [17.174] 3.269 [0.655] 17.327 [3.786] 5.303 [0.321] 5.300 [0.262] 5.407 [0.407] 4.773 [0.431] 4.733 [0.321] 4.727 [0.510] 4.792 [0.390] 4.500 [0.404] 2.864 [0.471] 2.925 [0.242] 2.778 [0.371] 3.114 [0.431] 2.667 [0.219] 2.774 [0.288] 3.219 [0.292] 3.000 [0.365] Credit Debit (1) Mean/SE 33 WTP Estimates Average WTP for All Items 0.955 1.068 1.058 1.712 0.997 0.792 1.110 1.530 [0.147] [0.123] [0.185] [0.491] [0.136] [0.131] [0.193] [0.252] 0.820 0.819 0.910 0.968 0.752 0.573 0.846 1.230 [0.113] [0.111] [0.143] [0.162] [0.114] [0.108] [0.145] [0.199] Average WTP for Office Supply Items 0.845 1.001 1.059 1.213 0.995 0.657 1.008 1.565 [0.168] [0.151] [0.234] [0.286] [0.179] [0.149] [0.258] [0.340] Average WTP for Toiletry Items 0.273 0.327 0.346 0.376 0.295 0.302 0.367 0.483 [0.043] [0.042] [0.066] [0.059] [0.048] [0.061] [0.066] [0.077] Average WTP for COVID-Related Items 1.553 1.784 0.994 6.602 1.617 1.234 1.759 1.823 [0.472] [0.235] [0.227] [4.465] [0.397] [0.283] [0.324] [0.381] Note: This table reports the mean values for selected variables across the eight treatment and control groups in our experiment. Pairwise t-test values can be found in Appendix Table 1. Household income is categorical; 0 represents income under $20,000, 1 represents $20,000-$34,999, 2 represents $35,000-$49,999, 3 represents $50,000-$74,999, 4 represents $75,000-$99,999, and 5 represents $100,000+ Average WTP for Food or Drink Items 28 The average participant in our experiment is an active Venmo user but uses credit or debit cards as their primary payment method. At the time of the experiment, participants had an average of 100 friends on the Venmo app and $97 in their Venmo balance (median of $11). Participants use Venmo for an average of 3.3 transactions per month and estimate that they spend $23 and receive $27 on Venmo in an average month. The average participant uses their credit card the most often, followed by debit, Venmo, cash, and any other digital payment platforms, in that order. Similarly, Venmo transactions account for 16% of participants’ total spending, which is less than credit or debit but more than cash or other digital payment platforms. Participants use an average of 1.4 other digital payment platforms in addition to Venmo, the most popular of which is PayPal. 6.2 Main Results Between the Venmo, debit, and credit treatment arms, participants assigned to the Venmo treatments have the highest mean WTP ($1.12; s.e. ($0.037)), followed by debit ($1.07; s.e. ($0.058)), and then credit ($0.96; s.e. ($0.074)). This difference in means is only statistically significant at the 12.5% level; all other differences in means reported in this section are significant at the 5% level. It is important to note that while these differences might seem low in magnitude, they represent large percentage changes in WTP; given that the mean WTP for the credit group is $0.96 (s.e. ($0.074)), an increase of $0.16 represents a 17% increase in WTP. Within Venmo, the highest mean WTP is for the Public setting ($1.33; s.e. ($0.072)), followed by the Private setting ($1.15; s.e. ($0.065)). The Friends Only setting has a much lower mean WTP ($0.91; s.e. ($0.052)), 32% less than Public and 21% less than Private. Also within Venmo, those assigned to priming have a 15% higher mean WTP ($1.21; s.e. 29 ($0.057)) than those assigned to no priming ($1.05, s.e. ($0.047)). The impact of priming on WTP depends on privacy setting; within those assigned to the Venmo Private treatment, WTP is 19% higher for those who underwent priming ($1.26; s.e. ($0.104)) than those who were not primed ($1.06; s.e. ($0.082)). Similarly, for the Venmo Public treatment, the mean WTP of participants who were primed is 38% greater ($1.53; s.e. ($0.108)) than the mean WTP of those who were not primed ($1.11; s.e. ($0.088)). In contrast, within those assigned to the Friends treatment, those who were primed have a 21% lower mean WTP ($0.79; s.e. ($0.068)) than those who did not undergo priming ($1.00; s.e. ($0.076)). Thus, priming is associated with a higher WTP for the Private and Public settings but a lower WTP for the Friends Only setting. Across all eight treatment groups, we find that mean WTP is highest for participants assigned to Venmo Public Priming ($1.53; s.e. ($0.108)) and lowest for participants who were assigned to Venmo Friends Priming ($0.79; s.e. ($0.068)). The first row of the “WTP Estimates” section in Table 1 shows these differences in means across the eight treatment groups. The subsequent 4 rows show the differences in means across the groups for specific categories of goods; these results are discussed in greater detail in Section 6.3. Despite removing the one outlier of $100, the WTP estimates are still highly skewed (skewness of 2.32). Thus, we examine the differences in median WTP across treatment groups to determine whether these differences in means are driven by a few high WTP estimates. In this analysis of median WTP estimates, we find the same trends as in the differences in means. Median WTP for the Venmo group ($0.70) is higher than debit ($0.50) and credit ($0.50). The median WTP for Venmo Public ($1.00) is the highest, Venmo Friends ($0.50) is the lowest, and Venmo Private ($0.75) falls between the two. Priming has 30 a higher median WTP ($0.75) than no priming ($0.50). For the Public and Private groups, priming increases median WTP ($0.63 to $0.75 for Private; $0.75 to $1.00 for Public) but priming has no impact on median WTP for Friends Only ($0.50 regardless of priming assignment). Given that we find the same trends in median WTP as mean WTP across treatment groups, we do not find the skewness of the WTP data to be concerning. Table 2 gives the results of our OLS regression analyses, which we conduct to account for variation in WTP caused by differences across the ten items. Column (1) contains the estimates of a basic OLS regression with item-level fixed effects in which indicators for the credit and Venmo treatments (excluding debit) are regressed against willingness to pay. The specification in column (2) includes controls for Venmo usage. Columns (3) and (4) repeat the same specifications as (1) and (2) but replace the single Venmo indicator with an indicator for each of the three privacy setting assignments. Columns (5) and (6) replace the privacy setting variables with six indicators, one for each combination of privacy setting and priming assignment; column (6) is the regression estimation we described in the experimental design section. Column (7) contains the estimate of a OLS regression with item-level fixed effects in which an indicator for the priming treatment is regressed against willingness to pay. Column (8) repeats the same specification with controls for Venmo usage. Each of these specifications addresses a different hypothesis regarding the Venmo Effect. For the general Venmo Effect, Column (2) shows that Venmo increases mean WTP by $0.02 in comparison to debit while credit decreases mean WTP by $0.11, though these coefficients are not statistically significant at the 5% level. For the mental accounting mechanism, Column (6) finds that the Venmo Private No Priming treatment has a statistically insignificant $0.02 lower WTP than the debit group. For the signaling mechanism, Column 31 Table 2: Main Results -- The Effect of Payment Form on Willingness to Pay VARIABLES (1) WTP Payment Forms (Debit omitted) Credit Card -0.113* [0.051] Venmo 0.0566 [0.055] Venmo Privacy Settings Venmo Private Venmo Friends Venmo Public Venmo Treatment Groups Venmo Private No Priming Venmo Private Priming Venmo Friends No Priming Venmo Friends Priming Venmo Public No Priming Venmo Public Priming Priming Treatment Priming (2) WTP (3) WTP (4) WTP (5) WTP (6) WTP -0.113 [0.096] 0.0230 [0.069] -0.113* [0.051] -0.114 [0.096] -0.113* [0.051] -0.115 [0.096] 0.0832 [0.077] -0.160*** [0.036] 0.265** [0.092] 0.0657 [0.088] -0.160** [0.079] 0.180** [0.091] -0.00967 [0.108] 0.198** [0.083] -0.0714 [0.046] -0.277*** [0.062] 0.0420 [0.071] 0.462*** [0.125] -0.0197 [0.101] 0.179 [0.122] -0.0508 [0.098] -0.299*** [0.092] 0.0454 [0.114] 0.289** [0.116] (7) WTP (8) WTP 0.159*** [0.036] 0.103 [0.077] Constant 1.068*** 1.182*** 1.068*** 1.185*** 1.068*** 1.151*** 1.050*** 1.130*** [0.039] [0.107] [0.039] [0.107] [0.039] [0.107] [0.017] [0.124] Item-Level FE Yes Yes Yes Yes Yes Yes Yes Yes Controls No Yes No Yes No Yes No Yes Observations 2,339 2,289 2,339 2,289 2,339 2,289 1,509 1,459 R-squared 0.002 0.010 0.013 0.017 0.021 0.021 0.003 0.022 Note: This table shows the results of our regression estimations. The outcome variable is mean WTP indicated by participants and each of the variables is a binary indicator for whether the participant was assigned to that treatment group. The Debit indicator is omitted, meaning the constants for columns (1), (3), and (5) should be interpreted as the difference between mean WTP for debit and mean WTP for that respective treatment group. Columns (7) and (8) are limited to participants assigned to Venmo; the constant in column (7) is the mean WTP for participants assigned to Venmo without priming. Columns (2), (4), (6), and (8) include controls for Venmo uage variables, specifically age of account, number of friends, number of linked payments, percentage of spend via Venmo, monthly transactions, monthly inflow, and monthly outflow. All coefficients are nearly identical without fixed effects and exactly identical with additional individual-level fixed effects. (4) estimates coefficients for each of the Venmo privacy settings; we find that the Public privacy setting increases mean WTP by $0.18, the Private setting increases mean WTP by $0.07 (though not at a statistically significant level), and the Friends privacy setting decreases mean WTP by $0.16. Wald postestimation tests show that the coefficients for each of the privacy settings statistically differ from each other at the 1% level. For the priming 32 mechanism, Column (8) finds that the priming treatment increases mean WTP by $0.10. Additionally, Column (6) allows us to examine the effect of priming within each privacy setting. The Friends setting has a larger decrease in WTP when paired with priming and the Public and Private settings have larger increases in WTP when paired with priming. Wald postestimation tests find that the coefficients between the priming and no priming treatment groups differ significantly for Friends (Prob > F = 0.0213) and Public (Prob > F = 0.0830), but not for Private (Prob > F = 0.1468). 6.3 Results by Item Type As Venmo is typically used for specific types of purchases, like food and drinks, we also investigate whether the impact of payment form depends on the type of item being purchased. We divide the ten items in our experiment into four categories: Food/Drinks (the can of Coca Cola, granola bar, pack of gum, and bottle of coffee), Office Supplies (the notebook, pen, and phone wallet), Toiletries (the toothpaste and soap), and COVID-Related Items (the cloth face mask). Table 3 shows the results of the fixed effects regressions we run. Column (1) contains the estimates of a basic OLS regression with dummy variables for the item categories, omitting the Food/Drinks category. While all the coefficients are statistically significant, it could be argued that these differences in willingness to pay are simply reflections of the different items’ values. Column (2) adds variables for seven of the eight treatment groups, omitting debit. Column (3) adds interaction terms between the treatment groups and the item categories; again, the debit and Food/Drink groups are omitted to avoid multicollinearity. In examining the estimates of (3), we find little evidence that the Venmo Effect depends on the type of item purchased. The only statistically significant interaction variable is for Venmo Private No Priming * COVID, which means that the mean WTP for 33 Table 3: Willingness to Pay by Type of Item VARIABLES Categories (Food/Drinks omitted) Office Items Toiletry Items COVID-Related Items Treatment Groups (Debit omitted) Credit Card Venmo Private No Priming Venmo Private Priming Venmo Friends No Priming Venmo Friends Priming Venmo Public No Priming Venmo Public Priming (1) WTP (2) WTP (3) WTP 0.175*** [0.065] 0.504*** [0.075] 0.760*** [0.129] 0.175*** [0.065] 0.504*** [0.074] 0.761*** [0.128] 0.181 [0.133] 0.490*** [0.156] 0.965*** [0.246] -0.113 [0.093] -0.00967 [0.100] 0.198* [0.117] -0.0714 [0.094] -0.277*** [0.087] 0.0420 [0.103] 0.462*** [0.122] 0.000473 [0.110] 0.0911 [0.130] 0.149 [0.155] -0.0675 [0.115] -0.246** [0.105] 0.0267 [0.130] 0.411*** [0.153] Interaction Terms CreditCard * Office -0.156 [0.194] Venmo Private No Priming * Office -0.0323 [0.239] Venmo Private Priming * Office 0.0632 [0.283] Venmo Friends No Priming * Office 0.0623 [0.214] Venmo Friends Priming * Office -0.0973 [0.191] Venmo Public No Priming * Office -0.0195 [0.243] Venmo Public Priming * Office 0.154 [0.301] Credit Card * Toiletry -0.218 [0.230] Venmo Private No Priming * Toiletry -0.0149 [0.283] Venmo Private Priming * Toiletry 0.0451 [0.301] Venmo Friends No Priming * Toiletry -0.0631 [0.236] Venmo Friends Priming * Toiletry 0.146 [0.261] Venmo Public No Priming * Toiletry 0.132 [0.291] Venmo Public Priming * Toiletry 0.211 [0.317] Credit Card * COVID -0.231 [0.535] Venmo Private No Priming * COVID -0.881** [0.349] Venmo Private Priming * COVID 0.222 [0.496] Venmo Friends No Priming * COVID -0.0998 [0.472] Venmo Friends Priming * COVID -0.303 [0.379] Venmo Public No Priming * COVID -0.0520 [0.417] Venmo Public Priming * COVID -0.372 [0.469] Constant 0.859*** 0.839*** 0.819*** [0.035] [0.062] [0.074] Observations 2,339 2,339 2,339 R-squared 0.034 0.053 0.059 Note: Note: This table shows the results of our regression estimations by type of item. The outcome variable is mean WTP indicated by participants. The category variables are binary indicators for whether the item is from that category. The treatment group variables are binary indicators for whether the participant was assigned to that treatment group. The interaction terms are simple interaction variables between the categories and the treatment groups. 34 COVID-Related Items when using the Venmo Private setting without priming is $0.88 less than the mean WTP for items in the Food/Drinks category when using a debit card, approximately a 100% decrease. Wald tests within each category do not find significant differences in coefficients for each of the seven payment groups. 6.4 Discussion Our difference in means analysis finds that a Venmo Effect exists as participants had higher willingness to pay estimates with Venmo than with credit or debit cards, though the corresponding coefficient in our regression estimation is not statistically significant. We are not particularly concerned by the lack of statistical significance as we believe it is a function of sample size limitations. As expected, we find that priming overall increases willingness to pay when using Venmo. However, for the Friends privacy setting, we find that priming actually decreases WTP. In contrast, priming increases WTP for the Private and Public settings. In the absence of priming, the Friends treatment decreases WTP while the Private and Public treatments increase WTP. Thus, the priming treatment heightens any existing influences on WTP like the privacy settings rather than having a uniformly positive effect on its own. This finding that the Venmo Friends treatment had a lower mean WTP than debit or the other Venmo privacy settings is inconsistent with our initial hypothesis about the signaling mechanism. This seems to suggest that spending money on Venmo is shameful, rather than a positive signal of wealth or social status as we had expected. This could be a result of the specific experimental context, since participants knew that they would have to pay a third party Venmo account named “Willingness to Pay Research.” College students may be embarrassed to show their peers that they made a Venmo payment as a part of an 35 experiment; they could be concerned that paying a stranger makes them seem naïve to their peers or that participating in research makes it seem as if they are in need of money. While this result is the opposite direction of what we expected, it shows that the privacy setting on Venmo does have an impact on willingness to pay, reinforcing the importance of Venmo’s social factors on spending. Surprisingly, we find that credit cards yield lower WTP estimates than debit cards. This challenges the existing literature (Hirschman 1979; Feinberg 1986) that finds higher willingness to pay with credit than debit. This unexpected finding may suggest a generational change; today’s youth may feel less comfortable with credit (or the accrual of interest) than their parents and grandparents. It could also simply be a factor of inexperience due to the young age of our participants. Given that 98% of participants have a debit card but only 72% have a credit card, we test whether this difference in ownership is driving the WTP gap by limiting the sample to participants who have both a debit and credit card. In this analysis, we find that credit cards yield a $0.16 (s.e. ($0.067)) lower mean WTP than debit cards, a larger and more statistically significant effect than those found in our regression estimations with the full sample (see Table 2). This suggests that the difference in WTP between debit and credit is not driven by a lack of owning credit cards. Although the difference in WTP cannot be attributed to differential credit card possession, it could be a result of the length of time that participants have had their credit and debit cards. We do not have data on length of possession, but most participants have probably had a credit card at most for a few years, which could mean that they are less familiar with credit than debit and are thus more hesitant to spend with credit. 36