Skip to content
Publicly Available Published by De Gruyter October 7, 2020

Restaurant Tipping Discrimination: Evidence from a Representative Sample of US Households

  • Nusrat Jahan , Andrea Leschewski and David E. Davis ORCID logo EMAIL logo

Abstract

Discrimination in tipping creates concerns of inequity in service quality for restaurant operators (Brewster 2017). We use the National Household Food Acquisition and Purchase Survey data to conduct one of the very few nationally representative examinations of tipping behavior at US restaurants. We focus on differences in tipping behavior between groups with identifiable characteristics and investigate whether tipping differences between groups are robust to inclusion of a variety of controls. We investigate tipping at the extensive and intensive margins. In contrast to earlier studies, we find little evidence that tipping varies by race and gender.

1 Introduction

Discrimination in tipping creates concerns of inequity in service quality for restaurant operators (Brewster and Brauer 2017). Because tipping, giving an amount of money to someone that has provided a service, is voluntary, patrons are free to choose their tip amount. Basing tip amount on dining experience and service quality is typically justifiable discrimination, and servers frequently dress or act in ways meant to increase tip amount (e.g. having a pleasant appearance or writing thank you notes on bills (Lynn and McCall 2009). Of greater concern to restaurateurs is the perception of servers that patrons belonging to certain groups are less likely to leave a tip or are less generous tippers. When there is tipping discrimination between groups with identifiable characteristics, servers are likely to change their behavior in response. Servers are likely to be more attentive to reputed generous tipping groups and less attentive to other reputedly less generous groups (McCall and Lynn 2009).

An extensive literature examines patrons’ tipping behaviors and the tipping/service relationship. Yet nearly all previous works are based on small, non-representative samples of primary-survey or experimental data, with frequently contradictory results. In contrast, there are few studies that cover a broad geographical area using a large sample, and there are no studies that use a carefully constructed nationally representative sample of US households. Because no studies use a representative sample, it is unclear whether the findings from previous work are generalizable to the US population or only represent the behavior of the patrons sampled. In this paper, we use a nationally representative sample of household’s purchases at US restaurants to investigate patron’s tipping based on demographic and other characteristics.

We use the National Household Food Acquisition and Purchase Survey (FoodAPS) data to conduct our analysis. These data include a representative sample of households’ food purchases at US restaurants. We examine whether tipping varies with demographic characteristics (race, gender, age, and income), whether tipping varies with payment method (cash, credit card, gift card), and whether tipping varies by other observable characteristics (e.g. number of guests).

We contribute to the literature on restaurant tipping in three important ways. First, we use a nationally representative sample of households to examine tipping behavior. No other tipping study has used a representative sample of restaurant patrons. It is important to establish representative benchmarks to judge whether results from smaller, non-representative samples of diners are typical or atypical. Second, we focus on differences in tipping behavior between groups with identifiable characteristics and investigate whether tipping differences between groups are robust to inclusion of a variety of controls. Finally, we investigate tipping at the extensive and intensive margins. While there is a large literature on tip size, the intensive margin, relatively few if any prior studies examine the decision whether or not to tip, the extensive margin. While patrons infrequently fail to leave a tip, we show there are differences in that behavior related to identifiable characteristics.

2 Previous Literature

The extant literature suggests differences in tipping behavior across racial groups. Lynn (2004) examines differences in tipping behavior for White and Black consumers across a number of service providers (restaurant waiter, bartender, barber, cab driver, food delivery, hotel maid, sykcap/bellhop, masseuse, and usher). The author used data from a random-digit dialing national survey that asked respondents about their tipping practices. Regression results that controlled for gender, age, education, income, household size, and metro location, found that Black respondents reported being about 5% more likely to “stiff” a waiter than are White respondents.[1] The results also suggested that Black respondents reported tips to waiters about 3.5% smaller than White respondents. Results for the other services were qualitatively similar. The study states that results can be “generalized safely to all those adults in the United States who have a telephone and who are willing to participate in a telephone survey.” Yet, this population is unlikely to coincide with the population of all US households and results may not represent the behavior of all US adults. The study provides no evidence that the sample is truly representative of the United States population. Instead, it appears to rely on the randomness of the survey, ignoring that only 83 of 894, 9.3%, respondents were Black while the US Black population was 13.4% in the 2010 Census. Furthermore, the results depict actual tipping behavior only to the extent that respondents’ recall of tipping activities was accurate and honest.

Lynn (2013) examined tipping by Asian, Hispanic, and White restaurant patrons using a small, non-representative sample of 1274 responses from a particular restaurant’s online customer service survey. The paper reported two sets of regression results; one set used dollar tip as the dependent variable while the other used tip percent. The regressions included controls for bill size and a measure of perceived service quality along with other restaurant and dining event characteristics. No demographic controls beyond respondent race were included. Results suggested, “Hispanics, but not Asians tipped less than Whites.” The marginal effect for Hispanic respondents ranged from −2.78 to −3.2%.

Lynn (2006) surveyed the literature on race and tipping and concluded that African-American patrons leave smaller tips than White patrons, controlling for social-economic status and that the difference does not seem to be determined by deficiencies in service quality. Instead, the difference may arise from unfamiliarity with tipping norms. Lynn reviews papers using small samples of patron surveys and server records from a variety of geographic locations along with two papers using national telephone surveys. Lynn contends that the consistency of results from a variety of methods suggest the results should not be the result of “method-specific bias.”

The economics literature on tipping typically focuses on explanations for tipping behavior. Many papers attempt to distinguish whether tipping is an efficient reward for better service quality, or whether patrons tip only in response to social pressure. Bodvarsson and Gibson (1999) argue that tipping is related to service quality. They found service quality is a very strong predictor for tipping behavior and that tipping was a reward for good service, rather than an obligation or norm. Barkan and Israeli (2004) detected moderate correlation between service quality and dollar tip. Nevertheless, they suggested that service quality is independent of servers’ tip prediction. Bodvarsson and Gibson (1997) used a supply function of tips including both service quality and service quantity. They found that a large table size does not result in smaller tips; rather the mean tip rate is statistically the same for small and large tables. Moreover, the results revealed that regular diners tip more than do irregular diners, suggesting a role for service quality. Lynn (2001) conducted a meta-analysis based on eight published and six unpublished papers on tipping, and concluded that though service quality evaluation by customers and tip size are positively correlated, the correlation is a weak predictor of consumer satisfaction. Conlin, Lynn, and O’Donoghue (2003) investigate tipping as a “behavioral norm” and conclude that tipping is partially efficient in that tips award better service. But, they conclude that tipping is not fully efficient and that tipping behavior is also a conformity to social norms. Lynn and Sturman (2010) analyzed 275 dinning events of 51 business students at a commuter college in a large southern city and found that when service rating on a five-point scale increases by an additional point, tip increases by 2% of the total bill. Parrett (2006) based on restaurant data from Virginia concluded that tip size decreases with the table size and people tip due to reciprocity and guilt aversion. The results of experimental data also suggested that males tip more than females.

Lin (2007) concludes tipping is primarily a response to social norms, but considers the economics of that motivation. Lin suggests that if consumers are motivated by social norms then tipping may create excess burden (deadweight loss). Lin (2015) expands on this idea and includes tipping in a model of consumer choice. Because consumers respond to social norms and tip to avoid guilt, the demand for restaurant meals is below the optimum, creating excess burden. Lin allows that consumers may have different attitudes toward tipping and shows how that heterogeneity affects utility and efficiency.

There seems to be a strong case that consumers tip as a reaction to social norms. The evidence that consumers tip primarily to reward service seems weaker, but the two motivations are not mutually exclusive. Some authors suggest that social norms lead consumers to tip some base amount (e.g. 15%) but then adjust that base up or down based on service quality. There is considerable evidence that tipping varies by demographic characteristics, including race and gender. Given that norms are influenced by culture, and culture varies with demographics, it is perhaps not surprising that tipping varies by race, income, and gender (Schaller and Crandall 2004).

3 Data and Variables

We use data from the National Household Food Acquisition and Purchase Survey (FoodAPS). Data in FoodAPS are from a survey conducted under the supervision of the Economic Research Service (ERS) with support of the Food and Nutrition Service (FNS) of the U.S. Department of Agriculture (USDA). The survey was designed to provide a representative sample of the noninstitutionalized households in the continental US.

FoodAPS data were collected on food acquisitions and purchases from 4826 households for nine and half months between April 2012 and mid-January 2013. Sampled households recorded information on all foods acquired by all household members over seven days. The survey uses a broad definition of “household.” In FoodAPS “households” are persons who live together and share food and that were at the sampled address for part of the week data were collected.

For FoodAPS, USDA collected data on food acquired for consumption at home and food acquired away from home. We restrict our attention to the food away from home data. These data include 39,120 food away from home events from 4352 households. Respondents were asked to record the total bill amount and tip amount from a purchase. Of the 39,120 events, 6082 had usable tip data. We eliminated events that occurred at limited-service restaurants (e.g. fast-food restaurants, bakeries, buffets, and vending machines) leaving 2301 observations.[2] We also eliminated a few households that refused to give race information. Some dining events did not include a count of the number of diners, leaving 2218 observations from 1385 households that are used in our regression models below.

4 Sampling

USDA surveyed households using a multi-stage sample design to arrive at a representative sample. In the first stage, 948 primary sampling units (counties or a group of contiguous counties) were selected using metropolitan statistical area boundaries. Then probability proportional to size (PPS) was used to select a stratified sample of 50 primary sampling units (PSU). Each PSU was composed of four target groups, 1) Supplemental Nutrition Assistance Program (SNAP) households, and 2) Non-SNAP households i) below 100% poverty guideline, ii) between 100 and 184% poverty guideline, and iii) equal to or above 185% poverty guideline. In the second stage, 395 secondary sampling units (SSU) were selected using PPS again.

Each household has a sampling weight. We used that weight, along with a stratum variable, and a PSU variable to derive statistics and Taylor series “linearized” standard errors using the tools available in Stata version 15.1. All statistics and regression results below represent results from appropriate weighting to derive estimates representative of the continental United States.[3]

5 Variables

Table 1 describes the response variables and Table 2 describes the explanatory variables used in this study.

Table 1:

Response variables.

Response variable Definition Calculation
Tip decision Was tip amount > 0? Yes = 1; No = 0
Tip Pct. Tip as a proportion of total bill Tip  amount Total paid Tip amount
Table 2:

Explanatory variables.

Variable names Type Description
Gender Binary Male = 1, Female = 0
Age1 Binary Respondent’s age <21 = 1, 0 otherwise
Age2 Binary Respondent’s age >20 and <36 = 1, 0 otherwise
Age3 Binary Respondent’s age >35 and <60 = 1, 0 otherwise
Age4 Binary Respondent’s age <59 and <70 = 1, 0 otherwise
Age5 Binary Respondent’s age > 69, 0 otherwise
Payment type Categorical
  1. Cash

  • (2)Credit/debit card

  • (3)Gift card

Number of guests Discrete A count of the number of people dining
Total bill Continuous Total amount paid minus tip amount
Race Categorical Individuals were asked to identify their race according to OMB standards:
  1. White

  • (2)Black

  • (3)Asian (Asian/Native Hawaiian, or pacific Islander)

  • (4)Other race (American Indian/Alaska Native, other race, or multiple races)

Hispanic Binary Hispanic = 1, 0 otherwise
Education Categorical
  1. Less than high school

  • (2)High school graduate

  • (3)College graduate

US born Binary Born in US = 1, 0 otherwise
Citizenship Binary US Citizen = 1, 0 otherwise
Region Categorical Census region
  1. Northeast

  • (2)Midwest

  • (3)South

  • (4)West

Rural Binary The household resides in a rural county = 1, 0 otherwise
Nonmetro Binary The Household’s county is not within a (CBSA) = 1, 0 otherwise
Household income Categorical
  1. Low income, HH income < $45,750

  • (2)Mid income, $45,750 ≥ HH income < $136,000

  • (3)High income HH income ≥ $136,000

Bar/Grill Binary Purchase occurred at a bar or grill = 1, 0 otherwise

The dataset contains the variable total paid, i.e. total payment including tip amount. We created a variable, tip decision, which takes a value of one if a nonzero tip was part of the total paid, and 0 otherwise. We created Tip Pct. by dividing tip amount by total paid minus tip amount. While we call this variable “Tip Pct.”, it is actually tip as a proportion of the amount paid excluding tip.

In Table 2, race is a categorical variable based on the characteristic of the person answering the survey. We grouped race into four categories, White, Black, Asian and other to assure we had a large number of responses in each category. Hispanic is a separate variable and is not mutually exclusive of the race variables.

Other notable variables in Table 2 include the three geographic variables. Region is the household’s census region. Rural is from the ERS Food Access Research Atlas and takes a value of one if the population-weighed centroid of the census tract of a household is in a rural area, where rural areas are areas with fewer than 2500 people. Meanwhile, NonMetro equals one if the household’s county of residence is in a Core Based Statistical Area (CBSA). Rural and NonMetro are not mutually exclusive.

Table 3 includes summary statistics for all variables. Diners left a nonzero tip in 89.7% of dining events and the average tip was 16.7% of the total non-tip bill. The average tip in these data are within the commonly suggested restaurant tip of 15–20%, and is similar to the amount quoted in most of the restaurant tipping literature (e.g. Lynn 2004) Interestingly, no tip was left in slightly over 10% of dining events. This is a larger incidence of “stiffing” a waiter/waitress than has been previously observed. For example, Lynn (2004) reported that respondents to a telephone survey stated that in only 2% of instances would they leave zero tip to a waiter.

Table 3:

Summary statistics.

  Mean Linearized standard error Subpopulation observations
Tip decision 0.897 0.009 2301
Tip Pct. 0.167 0.004 2301
Male 0.440 0.021 2301
Female 0.560 0.021 2301
Age1 0.014 0.005 2301
Age2 0.241 0.019 2301
Age3 0.437 0.031 2301
Age4 0.221 0.017 2301
Age5 0.087 0.015 2301
High income 0.149 0.020 2301
Medium income 0.582 0.022 2301
Low income 0.269 0.021 2301
White 0.847 0.015 2301
Black 0.061 0.009 2301
Asian 0.043 0.008 2301
Other race 0.049 0.013 2301
Hispanic 0.105 0.022 2301
No high school 0.045 0.007 2301
High school 0.486 0.031 2301
College 0.294 0.026 2301
Rural 0.331 0.041 2301
NonMetro 0.095 0.035 2301
Bar/Grill 0.071 0.012 2301
Total bill 26.248 0.931 2301
US citizen 0.931 0.017 2301
US born 0.861 0.027 2301
Number of guests 1.955 0.042 2218
Cash 0.448 0.031 2191
Credit/debit card 0.555 0.033 2191
Gift card 0.015 0.004 2191

Because a sizeable portion of the sample did not report a tip, we also calculated the mean tip percent conditioned on a non-zero tip. That conditional mean was 18.6% (0.004 linearized standard error). While almost 2% points larger than the unconditional mean, this tip size remains within the accustomed 15–20% range.

6 Methods

We investigate tipping behavior at the extensive margin using Tip Decision, yes = 1, no = 0 as the dependent variable. We estimate a linear probability model (LPM) using ordinary least squares (OLS) because we are most interested in the marginal effects of key independent variables. We considered estimating relationships using a Probit model, but chose LPM because it is easy to interpret results and because OLS provides consistent and unbiased estimates of marginal effects regardless of the limited-dependent variable.[4]

Following previous literature that models determinants of tipping percent, we use OLS to investigate the intensive margin, Tip Pct. As mentioned, Tip Pct is more accurately described as a proportion as it is a number between zero and one. Again, we are interested in marginal effects and are persuaded that OLS is capable of providing unbiased estimates, even though the dependent variable is limited.

7 Empirical Results

We are initially interested in whether tipping behavior is different for diners with observable characteristics and begin by conducting analysis using variables obviously observable to servers. Table 4 shows results from a series of regressions using Tip Decision as the dependent variable where we also include controls observable to the wait-staff, the number of guests, a Bar/Grill dummy, and geographic dummy variables (Rural, NonMetro, South, West).

Table 4:

Regression results, dependent variable tip decision (yes = 1).

Variable (1) Std. Error (2) Std. Error (3) Std. Error (4) Std. Error (5) Std. Error (6) Std. Error
Female −0.026 * 0.015 −0.028 * 0.015 −0.028 * 0.016 −0.020 0.018
Age1 −0.295 ** 0.115 −0.294 ** 0.115 −0.330 ** 0.113 −0.362 *** 0.105
Age2 −0.034 0.045 −0.027 0.044 −0.037 0.046 −0.093 *** 0.025
Age3 −0.025 0.045 −0.023 0.043 −0.038 0.046 −0.080 *** 0.030
Age4 0.039 0.044 0.037 0.042 0.028 0.044 −0.019 0.027
Black −0.071 0.047 −0.058 0.046 −0.046 0.039 −0.041 0.043
Asian −0.032 0.067 −0.030 0.066 −0.007 0.070 −0.072 0.082
Other race −0.057 0.040 −0.052 0.042 −0.032 0.039 −0.044 0.037
Hispanic −0.013 0.037 −0.004 0.039 0.011 0.045 0.022 0.044
Medium income −0.060 ** 0.023 −0.038 * 0.020
Low income −0.079 *** 0.025 −0.045 * 0.025
No high school 0.044 0.036 0.042 0.031
High school 0.023 0.019 0.016 0.017
US citizen −0.122 * 0.063 −0.154 ** 0.068
US born 0.111 * 0.061 0.112 ** 0.058
Number of guests 0.0106 0.009 0.009 0.010 0.010 0.010 0.011 0.010 0.010 0.010 −0.016 0.010
Bar/Grill 0.0609 ** 0.027 0.067 ** 0.025 0.057 ** 0.027 0.059 ** 0.026 0.059 * 0.027 0.034 0.022
Total bill 0.002 *** 0.000
Cash 0.027 0.020
Credit/debit card 0.096 *** 0.020
Gift card 0.036 0.067
Rural 0.0104 0.027 0.007 0.028 0.000 0.025 0.003 0.026 −0.035 0.023 0.009 0.021
NonMetro −0.02 0.040 −0.025 0.041 −0.020 0.041 −0.026 0.040 −0.064 0.022 −0.024 0.036
Midwest −0.033 0.025 −0.024 0.023 −0.034 0.024 −0.027 0.021 −0.084 0.024 −0.025 0.023
South −0.067 *** 0.024 −0.054 *** 0.022 −0.063 *** 0.021 −0.056 *** 0.020 0.004 *** 0.028 −0.066 *** 0.021
West −0.082 *** 0.024 −0.071 *** 0.025 −0.077 *** 0.021 −0.068 *** 0.022 −0.023 *** 0.041 −0.080 *** 0.025
Constant 0.9336 *** 0.026 0.927 *** 0.050 0.931 *** 0.024 0.948 *** 0.048 1.025 *** 0.061 1.016 *** 0.058
N 2218 2218 2218 2218 2218 2191
F 3.55 5.32 3.63 4.39 3.48 9.8
Prob > F 0.007 0.0004 0.0049 0.0021 0.0183 0.0023
R 2 0.0138 0.0324 0.017 0.0376 0.0524 0.0928
  1. *p < 0.10, **p < 0.05, ***p < 0.001 linearized standard errors account for sampling weights.

7.1 Tip Decision

The first column reports the results from including only a gender dummy variable that takes a value of one for Female. The coefficient suggests that females leave a tip 2.6% less frequently than males, but it is only marginally significant with a p-value of 0.082.

The control variables suggest the probability of leaving a tip is 6.1% higher at a Bar/Grill over a traditional restaurant. Coefficients also suggest the probability of leaving a tip is 6.7% lower in the South and 8.2% lower in the West, than in the Northeast (the base category). The coefficient for Rural and Nonmetro are not statistically significant.

The second column reports results when age dummy variables are included with the geographic and Bar/Grill controls. The base category is persons age 70 years or older. The coefficient for Age1 suggest that young diners, those younger than 21 years, are 30% less likely to leave a tip (p-value = 0.011) than the base category. The coefficient for Age2 suggests diners between age 21 and 35 are 3.4 less likely to leave a tip, but the standard error is too large to conclude the coefficient is statistically significant different from zero (p-value 0.373). We reach a similar conclusion for the other age categories. Their coefficients suggest diners between age 36 and 59 are 2.5% less likely to leave a tip and diners age 60–69 are 4% more likely to leave a tip than the base category. However, in both cases, we cannot conclude the variables are statistically significant (p-values are 0.586 and 0.328, respectively).

Column 3 shows results from including race variables, including the Hispanic variable. The base category is White diners. While the coefficients are large enough to suggest sizable differences in the probability of leaving a tip, the standard errors are too large to conclude statistical significance for any of them.

The first three columns suggest few statistically significant differences in the probability of leaving a tip for the observable characteristics in our data. But, there is evidence that a server approaching a table observing a female, or young persons under 21, might expect a higher than typical probability of being “stiffed.”

The fourth column includes all observable characteristics to better control for omitted variable bias in the first three columns, should excluded variables be correlated with any of the included variables. The results in column four are consistent with the results in columns 1–3. The Female coefficient is marginally statistically significant and negative, while the Age1 coefficient is negative and significant. The coefficient Bar/Grill is statistically significant positive, while the region coefficients for South and West are negative and statistically significant.

In column 5, we show results from including controls that are not obviously observable to wait-staff, income, education, and citizenship. None of the education coefficients is statistically significant. However, the coefficients on the income variables suggest low-income diners are 8% less likely to leave a tip, while medium-income diners are 6% less likely to leave a tip, than high-income diners are.

The citizenship variables are US Citizen and US Born. All diners that are US born are also US citizens, but diners not born in the US may, or may not, be US citizens. To examine the effect of a US born citizen, we sum the coefficients on US Born and US Citizen, 0.11–0.122 = −0.01. A Wald test provides a Chi-square statistic of 0.09, with a p-value of 0.77. We cannot conclude that a US born citizen is more, or less, likely to leave a tip relative to a non-Citizen not born in the US. In contrast, the marginal effect of a US citizen not born in the US is identified by only the US Citizen coefficient, which has a value of −0.12. In other words, a non-US born citizen is 12% less likely to leave a tip than a US born citizen and a non-Citizen not born in the US. This likely reflects international differences in tipping customs.

In column 6, we show results from including payment characteristics in the regression. Once controlling for payment characteristics, the Female coefficient is no longer statistically significant, but remains near −0.02. The Age1 coefficient remains large and negative, −0.362. Two other age coefficients are statistically significant; Age2 suggests diners between 21 and 36 are 9.3% less likely to leave a tip, while Age3 suggests diners between 36 and 60 are 8% less likely to leave a tip, compared to diners over age 70.

None of the race coefficients are statistically significant. The medium and low-income coefficients remain statistically significant, but are smaller in absolute value than those in column 5 are. The citizenship variables give the same qualitative results are those in column 5, US born citizens are no more nor less likely to leave a tip, but non-US born citizens are 15% less likely to leave a tip.

Of the payment characteristics coefficients, bill size (Total Bill) and paying by credit or debit card (Credit/Debit Card) are statistically significant and positive. It is interesting to note that using a credit/debit card increases the probability of leaving a tip by nearly 10%.

7.2 Tip Decision Discussion

Overall, our analysis of the decision to leave a nonzero tip provides evidence that females are about 2% less likely to leave a tip than males, although statistical significance fades once payment characteristics are included in the regression. More definitively, our results suggest that younger diners are less likely to leave a tip than those over age 60. We find no statistical evidence that Black, Asian, Hispanic, or diners of other races are less likely to leave tips than are White diners. However, the magnitude of the coefficients for the race variables are large in absolute value, suggesting for example, Black diners are 4.1% and Asian diners are 7.2% less likely to leave a tip than White diners. It is possible these coefficients are unbiased estimates of the true parameters (including the Female coefficient), but the data do not include enough information to precisely estimate standard errors. If so, then they are large enough to be economically meaningful to restaurant managers. Finally, although arguably difficult for servers to observe, we find that low-income and middle-income diners are less likely to leave tips than high-income diners are.

7.3 Tip Percent

Table 5 shows the results from a linear regression with Tip Percent as the dependent variable. We present results in the same order as in Table 4. The first column includes a gender dummy variable along with a limited number of controls. The coefficient suggests that female diners leave a tip 1.1 smaller than male diners do. Again, the coefficient is only marginally significant with a p-value of only 0.094.

Table 5:

Regression results, dependent variable tip percent.

Variable (1) Std. Error (2) Std. Error (3) Std. Error (4) Std. Error (5) Std. Error (6) Std. Error
Female −0.011 * 0.007 −0.012 * 0.007 −0.011 0.007 −0.010 0.007
Age1 0.004 0.043 0.004 0.044 0.000 0.043 −0.007 0.042
Age2 0.015 0.010 0.019 * 0.011 0.017 0.012 0.006 0.011
Age3 0.018 * 0.010 0.020 * 0.010 0.016 0.011 0.010 0.011
Age4 0.017 ** 0.008 0.015 * 0.008 0.013 0.009 0.005 0.009
Black −0.026 0.016 −0.028 * 0.015 −0.024 0.015 −0.023 0.017
Asian −0.030 ** 0.013 −0.032 ** 0.012 −0.027 * 0.014 −0.029 ** 0.012
Other race −0.017 0.015 −0.017 0.016 −0.011 0.015 −0.014 0.015
Hispanic −0.011 0.012 −0.012 0.013 −0.006 0.013 −0.006 0.013
Medium income −0.017 * 0.009 −0.014 0.010
Low income −0.021 ** 0.010 −0.018 * 0.010
No high school 0.001 0.017 −0.003 0.017
High school 0.000 0.011 −0.003 0.011
US citizen −0.021 0.023 −0.022 0.022
US born 0.023 0.019 0.022 0.019
Number of guests −0.014 *** 0.002 −0.015 *** 0.002 −0.014 *** 0.002 −0.014 *** 0.002 −0.015 *** 0.002 −0.013 *** 0.003
Bar/Grill 0.049 *** 0.015 0.050 *** 0.015 0.049 *** 0.014 0.046 *** 0.015 0.046 *** 0.015 0.045 *** 0.015
Total bill −0.0002 * 0.000
Cash 0.006 0.011
Credit/debit card 0.014 0.009
Gift card 0.002 0.017
Rural 0.016 * 0.008 0.015 0.009 0.011 0.009 0.012 0.009 0.012 0.009 0.011 0.009
NonMetro −0.033 ** 0.016 −0.031 * 0.016 −0.033 0.016 −0.032 ** 0.015 −0.030 * 0.016 −0.034 ** 0.013
Midwest 0.002 0.009 0.003 0.009 0.002 0.008 0.002 0.008 0.000 0.008 0.000 0.008
South −0.007 0.008 −0.007 0.008 −0.005 0.007 −0.006 0.007 −0.008 0.007 −0.009 0.008
West −0.025 ** 0.010 −0.025 ** 0.011 −0.021 ** 0.009 −0.022 ** 0.009 −0.025 ** 0.010 −0.025 ** 0.010
Constant 0.201 ** 0.008 0.182 *** 0.011 0.200 *** 0.006 0.191 *** 0.011 0.211 *** 0.015 0.212 ** 0.019
N 2218 2218 2218 2218 2218 2191
F 9.33 6.34 7.07 5.76 10.58 20.1
Prob > F 0 0.0001 0.0001 0.0004 0.0001 0.0002
R 2 0.0501 0.0497 0.0546 0.0595 0.0647 0.0701
  1. *p < 0.10, **p < 0.05, ***p < 0.001 linearized standard errors account for sampling weights.

Of note in column 1 are the coefficients on number of guests, Bar/Grill, rural, nonmetro, and the West region dummy. The Number of Guests coefficient is negative and statistically significant and suggests that each additional guest reduces the average tip by 1.4%. Dining at a Bar/Grill apparently increases the average tip, ceteris paribus, by 5%. Diners in Rural areas tip 1.6% more than Urban areas, but diners in nonMetro areas tip 3.3% less than do diners in metro areas. Most Rural areas are also nonMetro and the sum of the coefficients suggest diners in nonMetro, Rural areas tip 1.7% less than diners in Urban, Metro areas, although the net effect is not statistically significant (p-value = 0.23). The results in column 1 also suggest diners in the West tip 2.5% less than diners in the East.

Column 2 shows the results when we include only age dummy variables with our controls. Interestingly, the coefficients suggest the average tip increases with age, peaks, and then declines. While the Age1 coefficient is essentially zero and the Age2 coefficient is not statistically significant, it is interesting that the marginal effect increases from 0.4 to 1.50% as age increases from less than 21 years, to the age bracket 22–35 years. The effect then peaks at 1.8% for age bracket 36–59 years, before declining slightly to 1.7% for the age bracket 60–70 years. Regardless, the coefficients suggest that diners in age categories three and four tip statistically significantly more than young diners in age categories 1 and 2, and old diners in age category 5. This result likely reflects the effect of omitted (in this specification) variable bias. In the specification in column 6 discussed below, adding income and controls for payment method eliminates the economic and statistical significance of the age coefficients. None-the-less, servers observing very young or very old patrons likely expect lower tips as those characteristics signal the patron is lower income and is likely to use a payment method less conducive to tipping.

Column 3 shows the results when we include only race dummy variables in addition to our controls. Only the Asian coefficient is statistically significant suggesting the Asian diners tip 3% less than White dinners. The Black coefficient is economically meaningful at negative 2.6%, but not statistically significant (p-value 0.108).

Column 4 shows the results when all observable characteristics are included, with the other controls. The Female coefficient is negative 1.2%, about the same magnitude as in column 1. The age coefficients are also above the same magnitude as in column 2, and the pattern suggesting tip percent peaks for the 35–60 years age bracket. Of note, the Black coefficient suggests a negative and statistically significant at the 10% level (p-value = 0.074) 2.8% effect, and the Asian coefficient suggests a negative 3.2% and statistically significant at the 5% level (p-value = 0.013) effect.

Column 5 shows results when income dummies are added as independent variables. The results suggest that low-income diners leave tips 2.1% lower than high-income diners, and medium-income diners leave tips 1.7 lower than high-income diners. When income variables are included, none of the gender, race, or age coefficients (except for the Other-Race variable coefficient) is statistically significant. However, the size of the coefficients themselves remain relatively stable across specifications. This stability may suggest that the marginal effects are accurate, but imprecisely estimated with standard errors too large to conclude statistical significance.

Column 6 adds Total Bill and several methods of payment to the list of independent variables. The Total Bill coefficient suggest that a $100 increase in the total bill reduces the average tip by 1.9%, while the payment method coefficients are not statistically significant.

Once all controls are included, the Female coefficient is not statistically significant, but remains about minus 1% in magnitude. Similarly, Asian is the only race coefficient that is statistically significant. Black is economically meaningful at negative 2.3%, but not statistically significant.

Comparing the coefficients on the age variables in column 4 to those in column 6 suggests that the age variables may be proxying for other variables in column 4. While three of the four age coefficients are statistically significant in column 4, none of them are significant in column 6 and each of them declines in magnitude.

7.4 TipPct Discussion

The results in Table 5 provide some evidence for differences in the tip percentage between racial groups. The results suggest that Asian diners leave tips 2.7–3.2% lower than White diners, and that result is robust across specification and statistically significant. This contrasts with findings in other studies (for example, Lynn 2013 finds that Asian diners do not tip less than White diners).

However, the evidence that Black diners leave smaller tips than their White counterparts is weaker than is suggested by much of the extant literature. The marginal effect for Black diners ranges from −2.3 to −2.8%. While economically meaningful, the coefficient is rarely statistically significant (and it is only marginally significant in the one case where it is significant). The marginal effect for Hispanic is economically small, ranging from −0.6 to −1.2%, and is never statistically significant.

Tipping-percentage differences across genders seems a minor consideration. The marginal effect ranges from −1.0 to −1.2% across three specifications, and is marginally significant at the 10% level in two of those specifications.

Other factors seem to have larger implications for tipping behavior over race or gender. For example, larger tables appear to receive lower relative tips than smaller tables in that each additional diner lowers percent tip by between 1.3 and 1.5%. Therefore, a table with six diners would leave a tip about 4.5% smaller than a table with three diners, all else equal.

7.5 General Discussion

It is worthwhile to consider whether our results contribute to the literature on the economics of tipping. We include two variables in our regressions that are commonly thought to proxy for service quality, Total bill and Number of guests (Conlin, Lynn, and Donoghue 2003). Both are negatively related to Tip percent in our regressions suggesting that diners use larger tables or larger bills to lessen their tip burden, rather than to reward greater service with larger tips. Clearly, our results do not suggest an efficiency explanation for tipping.

But our results are consistent with Lin (2015) who identifies four different types of restaurant consumers based on their attitudes toward restaurant service and social norms. He shows that consumers affect their utility by tipping according to their preferences. Our contention is that cultures have different preferences and attitudes toward tipping which leads to different tipping behavior. For example, travel guides frequently advise on tipping practices in different cultures. Tripsavvy, an online travel advisor, published a “Guide to Tipping in Asia” in 2019 (Rodgers 2019) and contends that tipping is not expected in many Asian cultures. If different cultures have different social norms toward tipping, then it seems logical to expect those cultural influences to reveal themselves in data on tipping. That Asian diners in our data leave smaller tips than other diners is likely a reflection that many Asian cultures do not tip, or tip less than US patrons, in restaurants (Rodgers 2019). Similarly, that our regressions show patrons in the Western US tipping less than patrons in the East and that those in the South, and females, are more likely to leave zero tip likely reflects different attitudes towards tipping because of differences in cultural and social norms across groups.

8 Conclusions

We estimate regressions to examine the relationship between observable characteristics and tipping behavior. We find some evidence of differences in tipping between diners of different races. Asian diners seem to tip about 3% less than do White diners. The evidence for Black diners is somewhat mixed in that we find a marginal effect on tipping of about −2%, but the effect is rarely statistically significant in our data. We find no evidence that Hispanic diners tip less than White diners. Race does not play a statistically significant role in the decision to tip in our data, but the marginal effects are large (in absolute value). Age seems to be significant observable characteristic providing an important signal to servers of diners’ likely tipping decisions and tipping amounts. Diners younger than 21 years are about 30% less likely to leave a tip, and tip percent seems to peak with diners between ages 35 and 60 years. Gender appears to affect both the decision to tip and perhaps the amount of tip left.

Our results fail to confirm some earlier findings on tipping. Most of the extant literature suggests Black and Hispanic diners tip smaller amounts than White diners. For example, Lynn (2013) finds “Hispanics, but not Asians tipped less than Whites,” the opposite of our findings. Lynn (2004) finds that Black diners tipped waiters smaller amounts than White diners. A survey by Lynn in 2006 that African-American patrons leave smaller tips than White patrons.

We suggest that our data are superior to those used in earlier studies. The data are structured to be representative of US households, while previous studies used convenience samples or relied on randomly collected surveys without paying attention to representativeness. Furthermore, our data have been checked for accuracy and are based on actual purchases rather than being reliant on a survey respondent’s recall.


Corresponding author: David E. Davis, Professor, Ness School of Management and Economics, South Dakota State University, 246 Harding Hall, Brookings, SD 57007, USA, E-mail:

Nusrat Jahan: Currently a Ph.D. candidate at the Department of Economics, 501 Fletcher Argue Building, University of Manitoba, Winnipeg, Manitoba R3T 5V5, Canada.


Funding source: South Dakota State Experiment Station

Funding source: USDA National Institute of Food and Agriculture

Award Identifier / Grant number: SD00H675-18

Acknowledgment

This work was supported by the South Dakota State Experiment Station and USDA National Institute of Food and Agriculture, Hatch project number SD00H675-18.

References

Barkan, R., and A. Israeli. 2004. “Testing Servers’ Roles as Experts and Managers of Tipping Behavior.” Service Industries Journal 24: 1–18.10.1080/0264206042000299194Search in Google Scholar

Bodvarsson, O., and W. Gibson. 1997. “Economics and Restaurant Gratuities: Determining Tip Rates.” American Journal of Economics and Sociology 56: 187–204, https://doi.org/10.1111/j.1536-7150.1997.tb03460.x.Search in Google Scholar

Bodvarsson, O., and W. Gibson. 1999. “An Economic Approach to Tips and Service Quality.” The Social Science Journal 36: 137–47, https://doi.org/10.1016/s0362-3319(99)80009-3.Search in Google Scholar

Brewster, Z. W., and J. R. Brauer. 2017. “Tableside Justice: Racial Differences in Retributive Reactions to Dissatisfaction.” Sociology of Race and Ethnicity 3: 383–97, https://doi.org/10.1177/2332649216683966.Search in Google Scholar

Conlin, M., M. Lynn, and T. O’Donoghue. 2003. “The Norm of Restaurant Tipping.” Journal of Economic Behavior and Organization 52: 297–321, https://doi.org/10.1016/s0167-2681(03)00030-1.Search in Google Scholar

Economic Research Service (ERS), U.S. Department of Agriculture (USDA). National Household Food Acquisition and Purchase Survey (FoodAPS). Also available at https://www.ers.usda.gov/foodaps.Search in Google Scholar

Lin, T.-C. 2007. “Economic Behavior of Restaurant Tipping.” Economics Bulletin 4 (2): 1–10.Search in Google Scholar

Lin, T.-C. 2015. “Modeling the Relationship between Restaurant Tipping and Consumer Behavior.” Theoretical Economics Letters, 5: 389–95, https://doi.org/10.4236/tel.2015.53044.Search in Google Scholar

Lynn, M. 2001. “Restaurant Tipping and Service Quality: A Tenuous Relationship.” Cornell Hotel and Restaurant Administration Quarterly 42: 14–20, https://doi.org/10.1016/s0010-8804(01)90006-0.Search in Google Scholar

Lynn, M. 2004. “Black-White Differences in Tipping of Various Service Providers.” Journal of Applied Social Psychology 34: 2261–71, https://doi.org/10.1111/j.1559-1816.2004.tb01976.x.Search in Google Scholar

Lynn, M. 2006. Race Differences in Restaurant Tipping: A Literature Review and Discussion of Practical Implications. Electronic version. Cornell University School of Hospitality Administration. http://scholarship.sha.cornell.edu/articles/38 (accessed June 08, 2019).Search in Google Scholar

Lynn, M. 2013. “A Comparison of Asians’, Hispanics’, and Whites’ Restaurant Tipping,” Journal of Applied Social Psychology 43: 834–9, https://doi.org/10.1111/jasp.12009.Search in Google Scholar

Lynn, M., and M. McCall, 2009. “Techniques for Increasing Server’s Tips: How Generalizable Are They?” Cornell Hospitality Quarterly 50: 198–208, https://doi.org/10.1177/1938965509334221.Search in Google Scholar

Lynn, M., and M. C. Sturman. 2010. “Tipping and Service Quality: A Within-Subjects Analysis.” Journal of Hospitality and Tourism Research 34: 269–75, https://doi.org/10.1177/1096348009350637.Search in Google Scholar

McCall, M., and A. Lynn. 2009. “Restaurant Servers’ Perceptions of Customer Tipping Intentions.” International Journal of Hospitality Management 28: 594–6, https://doi.org/10.1016/j.ijhm.2009.03.010.Search in Google Scholar

Parrett, M. 2006. “An Analysis of the Determinants of Tipping Behavior: A Laboratory Experiment and Evidence from Restaurant Tipping.” Southern Economic Journal 73: 489–514, https://doi.org/10.2307/20111903.Search in Google Scholar

Rodgers, G. 2019. A Guide to Tipping in Asia: Who, when, and How Much Should You Tip in Asia? Tripsavvy. https://www.tripsavvy.com/tipping-in-asia-1458307 (accessed September 5, 2020).Search in Google Scholar

Schaller, M., and C. Crandall, eds. (2004). The Psychological Foundations of Culture. Mahwah, New Jersey: Lawrence Erlbaum Associates.Search in Google Scholar

Received: 2020-04-08
Accepted: 2020-09-21
Published Online: 2020-10-07

© 2020 Walter de Gruyter GmbH, Berlin/Boston

Downloaded on 29.3.2024 from https://www.degruyter.com/document/doi/10.1515/jafio-2020-0007/html
Scroll to top button