Effect of Sunlight Levels on Taxicab Tipping

Abstract

Does level of sunlight affect tipping percentage in New York City (NYC) taxicab rides? NYC cab ride context is desirable for testing sunlight-tipping association because of random sorting between riders and drivers, direct exposure to sunlight, and lower confounding from variation in evaluations of non-differentiated cab ride service experiences. Combining hourly levels of solar radiation data with a sample of 13.82 million cab rides from January to October in 2009 in New York City, tipping increases by 0.63 percentage points when transitioning from dark overcast sky to fully sunny skies. The findings are robust to two-way clustering of standard errors based on hour-of-the-day and day-of-the-year and controlling for day-of-the-year, month-of-the-year, cab-driver fixed-effects, weather conditions, and ride characteristics.

Introduction

Tipping is an area of increasing interest in economics and social sciences [1-3]. Tipping represents a significant portion of the US economy. US Food industry alone generated $46.6 billion in tips in 2011 [4]. Both economic and non-economic rationales for tipping are proposed. The economic rationale for tipping is to reward service quality and receive better future service [1, 5]. Tipping also has positive externalities on other customers. However, tips are not legally required and represent an unnecessary cost to customers. Supporting non-economic rationale for tipping, Lynn and McCall [6] found near-zero correlation between tip size and service evaluations, and variability of less than 2 percent in tipping amounts. Lower correlation between tip size and service evaluation implies that customers could be primed by non-economic motives for tipping. Social psychology related rationales for tipping include ‘doing a good thing,’ reducing guilt, abiding by social norms, empathy and compassion towards service workers, among others [7]. While personality (e.g. individuals with greater empathy are likely to tip higher in different service contexts) and social conditioning (e.g. willingness to follow tipping norms) of an individual are generally stable, tipping could also be influenced by more transient factors such as weather-induced moods.

Weather explains significant variance in the mood, and among the weather-related factors, sunlight has the most immediate and more lasting impact on mood [8, 9]. The seemingly benign influence of weather is well documented in behavioral economics. Weather is shown to influence stock market returns [10], prices paid at art auctions in London [11], car purchases [12], returns of products purchased from catalogs [13], worker productivity [14], loan decisions [9], assessment of company earnings reports [15], among others. Sunny days induce positive mood and positive mood could potentially increase tipping [3, 10, 16, 17]. Some studies have found support for the relationship between sunlight and tipping [3, 18, 19] while others have found no association [20]. In a quantitative aggregation of two studies on sunshine and tipping, Lynn and McCall [21] found small to medium effect size between bill-adjusted tips and sunny weather.

To address the mixed findings on the sunlight-tipping association, the purpose of this paper is to assess the association between hourly variation in solar radiation and tipping percentage in New York City (NYC) taxicab rides. NYC cab ride tipping context is both practically and empirically meaningful. Practically, NYC taxicabs represent a significant portion of NYC’s economy – 13,437 medallions worth about $800,000 each, 485,000 rides per day, 50,000 active drivers working in 9.5 hour shifts on average who transport 600,000 passengers per day, or 236 million passengers per year. Empirically, taxicab tipping context allows for a robust testing for sunlight and tipping association. Cab ride experience is generally standardized (versus in other service settings where type of service, décor, and status of service establishment confound with tipping); service evaluations are less likely to vary across cab riders experiencing standardized cab service (lowering confound between variations in service evaluation and tipping); exposure to sunlight is direct in a cab (compared to in enclosed spaces such as restaurants); and sorting between drivers and riders is limited (compared to when higher quality service workers sort into establishments where higher tips are likely) [22]. These advantages allow for a quasi-natural experiment type setup where drivers and riders are almost randomly matched. The influence of treatment (sunlight) is exogenously varying on an hourly basis, and is less confounded by variations in service quality across encounters. Below, we further elaborate on the benefits of NYC taxicab tipping context in testing sunlight-tipping association.

First, past work on tipping shows that service providers play an important role in altering tipping behavior. Gestures, tone, affect, and actions of restaurant servers have a strong effect on tips [23, 24]. In taxicabs, the effects of such behaviors and actions are quite limited except for possible tone and nature of the occasional conversation between cab drivers and customers. Taxicab services are standardized and cab drivers do not have a significant control over enhancing the service experience. Anecdotal evidence suggests that cab riders focus on baseline hygiene factors and are not particularly motivated to go above and beyond the basic service levels [25].  In a service setting where a customer visits often, there is a stronger economic incentive to tip with the expectation of receiving better future services [5]. However, a cab rider is less likely to encounter the same cab driver again, therefore, the expectation of a future meeting would not drive the tipping percentage. Nor there are positive externalities from tipping, as tipping cab drivers do not improve overall service delivery of other cab drivers. Thus, taxicab rides lower confounding from variation in service experiences. While ‘friendliness’ of a cab driver is unobserved and could influence tipping, we control for the fixed-effects of cab drivers.

Second, past studies on weather and tipping behavior have tested effects of weather on tipping in enclosed spaces such as restaurants [18, 20] where customers are not directly exposed to sunlight. Whether customers in enclosed spaces pay attention to outside weather during service encounters is also affected by customers’ unobservables, such as their attention span. The ‘treatment’ of sunlight is present throughout the cab ride and does not vary from one cab to the other as NYC cabs are not allowed to be tinted for passenger and cab driver safety. The current study is the first large-scale field-study to assess the direct effects of hourly variation in sunlight on tipping behavior.

Third, and most importantly, the key confound biasing tipping estimates could be sorting between customers and service workers. Higher quality servers sort into higher paying restaurants and clients with greater willingness to spend more money (and, therefore tip) are more likely to visit high-end restaurants [22]. Sorting from demand and supply side are less plausible for cab rides. According to the 2014 NYC Taxicab Factbook, an average trip distance is 2.6 miles and the cost of an average cab ride, except for longer distances, is not substantially different from other modes of transportation (e.g. bus, or subway). Almost 90 percent of the NYC taxicab pickups are in Manhattan where most trips are less than a few blocks. Due to shorter distances and a majority of cab rides concentrated in Manhattan, where the easiest mode of transportation is a cab, selection effects from customers based on the cost of transportation are likely to be lower. Riders also cannot request cabs based on a driver’s ability or the distance they want to travel or ‘friendliness’ of cab drivers.

There also is limited sorting on the supply side. Except for anecdotal evidence in studies on cab drivers in New York City (NYC) who are less willing to stop for African American or other minority riders, cab drivers pick up passengers through street hails or e-hails, when notified by dispatch, or based on their number in the pickup queue at the airport. Cab drivers also cannot sort themselves in picking higher fare customers as they cannot deny fare based on the distance of travel.  Thus, cab drivers do not actively manage the selection of customers. Overall, due to quasi-randomness in rider-driver matching, we do not expect the estimates to be significantly biased to the possibility of two-sided sorting.

Finally, Azar [1] found that tipping behaviors could change based on changing social norms. Such changes in tipping norms are not present in the current sample as NYC Taxicab Factbook [26] states that “tipping by taxi passengers has remained rather con­stant for the last few years, holding at an average tip of 18%” (page 7). The previous NYC Taxicab Factbook published in 2006 assumes average tips at 15% during 1990 and 2005 [27]. Based on these two reports, the tipping norm for NYC cab rides has ranged from 15% to 18% and the data in this study shows the average tips to be around 19%, indicating limited, if any, change in tipping norms for taxicabs.

Overall, the proposed empirical setup allows for a quasi-natural experiment type setup to study the influence of hourly levels of sunlight on tipping in individual cab rides. The rest of the paper is organized as follows. We start by discussing past work on tipping followed by the influence of sunlight on mood. Thereafter we provide plausible pathways for a positive association between sunlight and tipping. Next, we elaborate on the cab-ride level sample and present the analyses. Finally, we summarize the results and discuss directions for future research.

Conceptual framework

Tipping literature

Tips are voluntary payments made by customers to service workers who perform services for them [22]. Numerous factors such as service quality, bill size, party size (e.g. number of customers at a restaurant table), social benefits of service after the encounter (e.g. haircuts), time of the day (tips tend to be larger in the evening), among others influence tipping. Some customers leave a flat dollar tip, others tip based on a percentage and while others leave a minimum tip irrespective of the bill amount [28]. In the United States, the norm of tipping is between 15 and 20 percent and tipping represents a significant portion of the informal economy [29]. Understanding tipping behavior has been an area of academic interest over the past three decades. Below we briefly review this literature, and for a more detailed review of tipping literature we refer readers to [5, 6, 21, 28].

Tipping could be motivated by both economic and non-economic reasons. Economic motivations for tipping are studied using rational choice theory and equity theory [30]. Tipping is counterintuitive to the rational choice theory as voluntary payments (tips) after receiving services do not directly influence utility received from services. However, others have demonstrated that higher utility from tipping is driven by service quality or the expectation of receiving better future service [21]. Equity theory explains tipping based on service quality, higher the ratio of service experiences of customers (outputs [e.g., service quality]) to the inputs provided by service workers, higher the tip [31]. Additional economic rationales for tipping include prices of services availed [21], bill size [32], patronage frequency [21], risk-sharing with the customer [33], paying by credit card [34], myopia in spending induced by alcohol consumption [35], among others.

Non-economic reasons for giving higher tips include abiding to social norms, increasing social esteem, and social status, enhancing server welfare, among others. Tipping is largely a norm-driven behavior [36]. Variations in cultural norms across countries also influence tipping [37]. Customers leave a tip to adhere to social norms or to lower guilt [2]. Customers also tip to maintain social status and improve feelings of self-esteem [38]. Tipping, as a tool to display individual status, allows customers to demonstrate their socioeconomic status to themselves, their guests, or to service workers [39]. Although there are differing accounts as to whether tipping started in 16^th century Europe or whether it was prevalent in the Roman times [40], scholars agree that tipping originated as a socioeconomic status feature [1], whereby individuals of economically higher status (i.e., wealthy) gave tips to people of economically lower status. As service workers are low-wage employees for whom tips are a major income source leaving higher tips could improve server welfare [7]. Additional non-economic reasons for tipping include service worker’s attractiveness [41], race [42], gender [21], clothing color [43], among others.

One aspect that also influences tipping is weather [21]. In tipping literature, on spring days, tips in Chicago area restaurant were higher [3], and the weather is positively related to tipping [18]. However, in a two-year longitudinal data of 11,766 credit-card receipts of a non-chain restaurant in Poughskeepsie, NY, there was no relationship between daily sunshine and daily tipping [20]. In their recent meta-analysis Lynn and McCall [21] found that “consumers left larger bill-adjusted tips under conditions known to elevate mood … when the weather was sunny (r = .20; z = 3.05, p < .01; n = 2)” (page  24).

Building on these mixed findings on the sunlight-tipping association, we test this association in NYC taxicab context with lower confounding from a variety of contextual factors and where two-sided sorting between customers and service providers is negligible. Next, we discuss the nature of the ordinal treatment of sunlight on human mood and behavior, followed by rationales for a positive association between sunlight and tipping.

Ordinal treatment of sunlight

Sunlight measured at the ground level is a combination of direct sunlight, diffused solar radiation in the sky, and reflected sunlight. Humans are exposed to both ultraviolet and visible solar radiation. Sunlight, with a visible radiation between wavelengths of 380 to 780 nanometers, operates from eye to brain through both visual and non-visual pathways [44]. Sunlight expressed in lux, a measure of illuminance, ranges from five lux with the sun at the horizon above thick storm clouds to 120,000 lux at noon. The human body does not respond to continuous ‘treatment’ of sunlight but to an ordinal ‘treatment’ [e.g., 28, 29, 30][1]. The visual pathway through the optic nerve affects “visual performance, perceptual judgments, and cognitions” (page 7) [44].

Impact of weather on mood and behavior

There is increasing evidence that weather influences affective states that in turn accumulate to influence decision making and behavior [45]. The mood and behavior induced by the weather could deviate from rational behaviors. The psychological and biological effects of sunlight on human behavior are documented in behavioral economics [e.g., 10, 46] and social psychology [e.g. 3]. In behavioral economics, weather is shown to influence a wide range of outcomes ranging from stock market returns [10] to prices paid at art auctions in London [11] and from car purchase [12] to college admission decisions [47, 48].

Sunlight induces positive mood as follows. Non-visual effects of sunlight through retinohypothalamic tract (RHT) are known to influence both limbic and endocrine system. Sunlight through its effects on both limbic and endocrine systems influences mood and cognition [30]. The dominant wavelength of sunlight (477 nanometers) through the retinal pathway modulates suprachiasmatic nuclei (SCN) that in turn regulates hormonal systems, blood pressure, and cognitive functioning [49]. SCN affects pineal glands and restricts conversion of serotonin to melatonin. Serotonin, the “feel good” hormone is associated with happiness, contentment, and relaxation, and melatonin, a derivative of serotonin, or the “wonder hormone,” affects sleep patterns and acts an antioxidant. Despite the presence of air-conditioning in certain service contexts (including NYC cab rides), sunlight also affects thermal sensation. The so-called thermic alliethesia refers to the physical response to manage body temperature. Exposure to sunlight increases body temperature. The perceptions of higher temperature could modulate body temperature and blood flow, which reduces stress and increase feelings of happiness and relaxation.

Positive association between sunlight and taxicab tipping

Based on the above discussion, we propose four possible pathways to explain the positive association between sunlight and tipping. While we do not observe the mechanisms listed below, to support our proposition, we connect literature on sunlight and positive mood to literature on antecedents to tipping.

First, increasing intensity of sunlight affects SCN functioning that increases serotonin levels and limits conversion of serotonin to melatonin, thus increasing total levels of happiness, contentment, and relaxation [18, 19]. Greater feelings of happiness increase helping behavior [50] and empathy [51], both outcomes are associated with tipping [5]. Past tipping studies have found that customers tip service workers with the desire to help [1, 22], and helping is the pre-dominant motive for tipping among US customers [52]. The positive mood primed by sunlight could actuate hard-wired neural wiring that triggers empathy [53] and generosity [3] that could, in turn, drive the motive to help cab drivers by tipping them more.

Second, a positive mood is also associated with optimism [54] that in turn could increase tipping. Tipping represents a significant (10-20%) portion of spending, and need for tipping could vary based on situational feelings towards savings, and thereby influencing the tendency to tip more (i.e. spend) or less (i.e. save). Lower concerns for saving for the future [55] and higher temporal discounting induced by optimism could lower need for saving and therefore tip more. Based on equity related explanations for tipping, customers could tip more as the perception of service quality (outputs) could be conflated by higher optimism [56, 57].

Third, a positive mood is positively associated with pro-social behaviors [58] and altruism [59]. Continuing from past work on pro-social behavior and tipping [60, 61], positive mood induced by sunlight could increase the tendency to pro-socially help service workers by giving higher tips. Customers exposed to songs with pro-social lyrics [62] or altruistic quotations on checks [63] left higher tip amounts. Pro-social behaviors primed through positive mood resulting from the sunshine could impel customers to meet social norms and avoid feelings of guilt [64] and motivate them to increase tips by abiding or exceeding social norms [1].

Fourth, mood maintenance hypothesis [65, 66], according to Handley, Lassiter, Nickell, and Herchenroeder [67], proposes that “individuals …  seek out positive activities while in a happy mood in order to maintain or elevate that mood … [a] tendency may become overlearned and, thus, automated” (page 106). When experiencing positive mood individuals prefer to avoid activities that ‘ruin’ positive mood. In an effort to maintain positive mood induced from sunlight customers may less stringently evaluate service outputs. Mood maintenance needs may also prime customers to avoid “losses” by lowering feelings of guilt resulting from not fulfilling their sense of duty or meeting the tipping norms.

Overall, we hypothesize that sunlight could influence tipping percentages in cab rides through multiple pathways. Next, we present the data description and the empirical test.

Methods

Data

We use ride-level data on licensed yellow cab rides in NYC available from the Taxi and Limousine Commission (TLC) and used in the study conducted by Haggag and Paci [68]. The full dataset of Haggag, K., Paci, G., 2014. Default Tips. American Economic Journal: Applied Economics 6, 1-19. is available in https://www.aeaweb.org/articles?id=10.1257/app.6.3.1 . Complete data is available for rides between January and October in 2009, 6 am to 4 pm on weekdays and 6 am to 8 pm on weekends. Each ride was paid only using a credit card, was for less than a 100 mile, the drivers’ shifts were longer than 30 minutes but less than 20 hours, and was used only if the base amount (comprising of fare, surcharge, toll, taxes) matched the fare. NYC TLC does not provide data on rides paid for in cash as there is no standard means to capture cash transactions.

The final sample aggregated based on driver × month × day × ride-level has 13,820,783 observations across 33,478 cab drivers. We matched the hourly solar data from National Solar Radiation Database.  We also matched the daily weather variables such as snowfall, rainfall, and average temperature obtained from National Climatic Data Center.

We converted hourly solar data available in Watts per meter-square (W/m²), or the intensity of solar radiation on a perpendicular surface, to lux, the measure of illumination [1 lux = 1 Watts per meter-squared / 0.0079]. The station for measuring solar radiation is located at John F. Kennedy (JFK) Airport. As a majority of cab rides occur in Manhattan, and due to the proximity of JFK airport to Manhattan, the data covers hourly sunlight variation for most cab rides in NYC.

Dependent variable – tipping

The outcome variable is the ratio of tip amount to total fare for the ride.

Predictor variable – lux category

We code the lux category as 0 if the lux value was 0. We code the category as 1, if lux value falls between 1 and 40; 2 [between 40 and 200]; 3 [between 200 and 400]; 4 [between 400 and 1000]; 5 [between 1000 and 2000]; 6 [between 2000 and 20000]; 7 [between 20000 and 110000]; 8 [between 110000 and 120000] and 9 [if value is above 120000].  [e.g., 30, 69, 70]

Variation in sunlight ‘treatment’ for each hour

To assess whether ordinal treatment of sunlight levels vary for each hour of the day, we plot the box-whiskers plot of lux during each hour (Figure 1). The variation in lux is downward biased because as the day progresses the mean sunlight increases, however varying sunlight due to changing cloud cover ‘shifts’ the lower extreme of a whisker of box-plot downwards. Significant variation in lux values around mean sunlight during each hour of the day (Figure 1), suggests that in the sample, sunlight varies during each hour of the day to plausibly influence tipping.

Figure 1. Hourly Variation in Lux

Next, to test our main proposition we explore the positive association between sunlight and tip fraction.

Association between sunlight and tip fraction

In Figure 2, we plot tip fractions across lux categories. At low lux levels (from dark to overcast skies, or lux categories 1 to 5) mean tipping levels are somewhat constant but have large confidence intervals. Thereafter, tipping levels rise with increasing brightness and with tighter confidence intervals. Increasing mean tip fraction in transitioning from overcast to sunny skies and tighter confidence intervals suggest a relative increase of average tips from overcast skies to bright sunlight.

Figure 2. Association between tip fraction and sunlight.

Lux Category	0	1	2	3	4	5	6	7	8	9
Average Tip Percentage	19.6946		19.6011	19.7766	19.6483	19.5931	19.7575	19.875	19.8272	20.161
Std Dev	14.9258		15.033	17.543	13.513	13.3426	14.556	13.2964	13.2544	14.098
Number of cab rides	528,006		21,224	49,667	35,406	62,091	758,678	6,562,780	1,227,778	4,575,153
95% CI – Low	19.6543		19.3989	19.6223	19.5075	19.4881	19.7248	19.8648	19.8038	20.1481
95% CI – High	19.7349		19.8034	19.9309	19.789	19.698	19.7903	19.8852	19.8507	20.1739

Note. There were no observations for lux category 1 in our sample.

However, interpretation in Figure 2 could be confounded by a wide range of factors. We explore pairwise correlations of plausible confounds with sunlight and tipping fraction (Table A, Appendix).

A variety of weather-related factors could influence tipping. It is widely documented that demand for cab rides increase significantly on rainy days or during bad weather [71]. Daily snowfall in millimeter (mm) and rainfall in tenths of a millimeter (mm) could increase demand of cab rides significantly, increase the propensity to tip, and also affect driving conditions. In S1 Table, snow and rainfall are positively associated with tipping fraction. Snow and rainfall are negatively correlated with shorter rides (negative correlation with distance) and extreme temperatures are positively correlated with longer rides.

In addition to the weather variables correlated with tipping, ride distance in miles, ride duration in minutes could also be correlated with tipping. Longer duration and distance could add to rider frustration, and therefore, could be negatively correlated with tipping fraction. In S1 Table, these two variables are negatively correlated with tip fraction.

Passenger count, or the number of passengers in a ride could prime the need for social esteem (tipping more), or for shared rides, customers could perceive more ‘saved’ money as fare could be shared among customers. In S1 Table, tip fraction is positively correlated with a higher passenger count.

During the period represented in the sample, two tipping systems were used in NYC cabs [68]. Although both systems provide credit card processing systems, during the study period, the tips were computed differently. Vendor system offers default options of $2/$3/$4 for fares below $15 and 20%, 25%, and 30% for fares above $15. Competitor system offered default tipping options of 15%, 20%, and 25% on the sum of fare, surcharge, taxes, and tolls. Default tip option (10%, 15%, and 20% tip) used at the end of the ride is positively correlated with a higher tip fraction (S1 Table).

Finally, demand for rides, willingness to tip a higher percentage, and the ability to drive could be affected by rush hour, weekday, day-of-the-year (e.g., generosity during Christmas holidays or frustration during traffic from New York Marathon), and month-of-the-year. These factors are correlated with tip fraction at varying levels in Table A. Finally, although driver effects cannot be meaningfully captured in a correlation using driver id, driver’s conversational abilities, driving skills, among others could be associated with tip and ride experience.

Empirical specification

Based on the above discussion on S1 Table, we propose the following specification that controls for a variety of confounds:

Tijhdm=αijhdm+βhLuxhd+βdDd+βjCij+βrRi+Driverj+Dayd+Monthm+εijhdm

Where

Tijhdmis the percentage tip received relative to total fare during ride

i, by driver

jin hour h on day

din  month

mduring the year.

Luxhdis the lux category during the hour on day d of the year.

Ddis the vector of day-level predictors—snowfall, rainfall, average temperature during the day, and the square of average temperature during the day. Cab-level predictors (

Cj)include whether the cab uses vendor or competitor passenger information system. Ride-level predictors (Ri) include ride duration, ride distance, number of passengers, whether default menu tip was used, ride during rush hour, and ride during weekday. We also include driver (driver id), day-of-the-year, and month-of-the-year fixed-effects.

εijhdm

is the error term of the regression model. The standard errors are clustered by the combined day-of-the-year and hour-of-the-day. This allows for within-cluster error arbitrary correlation among our observations.

Variable definitions and descriptive statistics of the sample are presented S2 Table. Average tip fraction was 19.95% with an inter-quartile range of 14.90 to 23.29%. The average distance of a ride is 2.57 miles with an inter-quartile range from 1.18 to 3.10 miles. An average ride duration is 12.77 minutes and the median ride duration is 11 minutes. An average ride has 1.61 passengers and 73.9% of rides have a single passenger. While NYC taxicabs have a default tipping option of 10%, 15%, and 20%, only 52.38% times do the passengers use the default tip option (note that the sample includes only the rides paid by credit card)  or 47.62% times they make a tipping decision that is not based on set norms of tipping 10%, 15%, or 20%). A majority of the rides occur during weekdays (71.74%), and the average lux is between 20,000 and 120,000 lux across different hours of the day.

Results

Table 1 presents the estimation results of various covariates on tipping percentage. Model 1 shows our baseline results without any fixed effects. Model 2 presents baseline results with driver fixed effects. Model 3 presents results with driver and day-of-the-year fixed effects. Model 4, our preferred specification, includes driver, day-of-the-year, and month fixed effects.

Table 1. Estimation results for impact of lux on tipping

	(1)	(2)	(3)	(4)
VARIABLES	Tip percentage	Tip percentage	Tip percentage	Tip percentage
Lux category [ 0 to 9]	0.0429***	0.0580***	0.0702***	0.0702***
	(0.0051)	(0.0050)	(0.0051)	(0.0051)
Snowfall	0.0029***	0.0027***	-0.0174	-0.0188
	(0.0008)	(0.0008)	(0.0193)	(0.0192)
Rainfall	0.0006***	0.0006***	0.0070	0.0076
	(0.0001)	(0.0001)	(0.0058)	(0.0058)
Average daily temperature	0.0008	0.0013	-0.5325**	-0.5906**
	(0.0024)	(0.0023)	(0.2501)	(0.2550)
Average daily temperature-squared	-0.0000	-0.0000	0.0087**	0.0094**
	(0.0000)	(0.0000)	(0.0037)	(0.0038)
Ride duration	-0.3329***	-0.3427***	-0.3456***	-0.3456***
	(0.0017)	(0.0016)	(0.0016)	(0.0016)
Ride distance	-0.1321***	-0.1310***	-0.1224***	-0.1224***
	(0.0044)	(0.0043)	(0.0043)	(0.0043)
Passenger count	0.0069**	0.0354***	0.0376***	0.0376***
	(0.0032)	(0.0078)	(0.0078)	(0.0078)
Vendor	3.6663***	3.7046***	3.7006***	3.7006***
	(0.0109)	(0.0255)	(0.0255)	(0.0255)
Default tip option used	6.6614***	6.6567***	6.6566***	6.6566***
	(0.0112)	(0.0111)	(0.0111)	(0.0111)
Ride during rush hour	-0.5445***	-0.5009***	-0.4887***	-0.4887***
	(0.0185)	(0.0184)	(0.0171)	(0.0171)
Weekday	0.5910***	0.6096***	1.4416	1.4758*
	(0.0161)	(0.0156)	(0.8796)	(0.8839)
Driver fixed effects	No	Yes	Yes	Yes
Day of the year fixed effects	No	No	Yes	Yes
Month fixed effects	No	No	No	Yes
R-squared	0.107	0.115	0.115	0.115

Notes.

Standard errors clustered by day-of-the-year and hour-of-the-day in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01

As small p-values can be an artifact of large sample size, we took four random samples of 5% of the sample to further reduce the possibility of Type 1 error. The results showed comparable estimates in direction and magnitude.

To ensure that the outcome measure was not mis-specified the two alternate measures of solar radiation — a continuous measure of hourly lux and hourly solar radiation on a perpendicular surface (Watts per meter-square) at JFK airport showed similar inferences.

An increase in lux category increases the tips that driver receive by 0.07 percentage point and is statistically significant at 1% level [t(stat)=27.86, p < .01]. An increase in lux category from zero (overcast skies) to 9 (fully sunny) increases tips by 0.6318 percentage points.

Based on the 2014 NYC Taxicab Factbook, an average ride is 2.56 miles long and with 485,000 trips per day, using the conservative fare structure (initial fare $2.50, $0.40 per 1/5^th mile, assuming no idle time and no surcharge) fare per ride without tax would be $5.70 ($2.50 for the first mile + [$0.40 for each 1/5^th mile for the remaining 1.6 miles of the 2.6 mile ride]). Indicating an increase in tips by $17,466.10 per day for the transition from 0 to 9, or in the order of several million dollars per year.

Discussion

Tipping remains an area of research interest in both economics and social psychology. Studies have focused on variations in tipping at the individual, establishment, and national levels [22]. Complementing the rational economic theory that does not fully explain tipping, researchers have proposed a variety of explanations including improving status, gaining approval, striving for equality in economic transactions, helping others, moods and behavior, among others [28].

We show that the tipping percentage increases by 0.63 percent when transitioning from overcast conditions to fully sunny skies. This result is based on a tipping context where confounders from variations in service-related factors are limited, exposure to sunlight is direct, and sorting between riders and cab drivers is random. In the current context, while the causal chain of the physio-psychological response of a customer to sunlight is unobservable and neither is a service encounter, we controlled for month-of-the-year, day-of-the-year, driver fixed effects, and additional ride-, cab-, and day-level characteristics. The standard errors were also clustered using both hour-of-the-day and day-of-the-year. The present study is one of the very few field studies drawing on a large sample from an unobtrusive context to study sunlight-tipping association.

Our findings provide an important extension to prior studies on the weather-tipping association.  In one of the earliest studies, Cunningham [3] collated data on daily sunlight levels during 13 spring days to assess the association between tipping across 130 customers (10 customers per day). While Cunningham [3] found a positive association, the inferences were based on a small sample with limited control for confounders. Other studies in this area have relied on actual or fake weather reported by the service workers [19] or the effects of future weather expectations on current restaurant tipping [18]. Recently, using two-year data from a single restaurant Flynn and Greenberg [20] found no association between sunlight and tipping. Despite extensions from Cunningham [3], Rind [19] and Rind and Strohmetz [18],  Flynn and Greenberg [20] study tipping levels at the daily level. Compared to these four studies, our sample is not only the largest ever to study the sunlight-tipping association, but also measures tipping for each transaction, uses a reliable hourly measure of solar radiation, controls for a variety of confounders, and is less influenced by sorting among customers and employees.

Related to practical implications, while tipping could be influenced by interventions such as normative messages, persuasion or influences to align individual tipping behaviors with service levels, unobtrusive factors such as weather could also impact tipping behavior. Related to direct effects of weather, sunny days could subconsciously prime positive mood that increases the need to tip. Messages identifying, guiding, or norming individual behaviors could be better received on sunny days, and customer interactions could be tailored to prime such behaviors when they could be in more positive moods on sunny days.

The finding must be interpreted in light of its limitations that also provide directions for future research. First, while the current findings from a quasi-natural experimental setting are robust to alternate specifications and a variety of controls, the nature of service interactions are not captured in the data. As such, future studies could focus on lab experiments to further understand the behavioral drivers in sunlight-tipping association. Such lab studies could also help further shed light on the continuous treatment effect of sunlight. Furthermore, field studies could be conducted to understand specific moods and motivations of customers. As such, the micro-dynamics of sunlight, associated behavioral changes, and the resulting effect on tipping could be studied in lab settings.

Second, the findings cannot be generalized beyond NYC cab rides during the period of observation, as the weather patterns, urbanization levels and other factors (e.g., the introduction of Uber) could influence the nature of cab services and competition. The sample is from 2009, when ride-sharing services such as Uber were not present in NYC. Therefore, temporal generalization may also not be made after 2009 and is left for future work. Cab data available for Chicago[2] and Singapore [72], and the recently released Uber data could also add to the generalizability of the finding. Continuing from the lab and field studies described above, cultural and situational characteristics varying across regional and global contexts could further explain the tipping variations.

Third, the findings cannot be generalized to other tipping contexts such as restaurants. The service setting for taxicabs is rather unique where richer service interactions do not occur between customers and service providers. Despite this limitation, as discussed earlier, the cab context has several unique advantages that facilitate robust inference for the sunlight-tipping association. Quasi-random matching between riders and drivers, limited need to tip higher to get better service in the future, limited need to be ‘nice’ based on patronage, limited variation in service provision ability across service providers are some of the strengths of the studied context. We call on future research to combine the experimental research on tipping with field setting to study tipping behavior in alternate service contexts.

An additional limitation could also be that the findings are a result of ‘noise’, as in noise around the standard tipping norms could be driving the results. While we control for the default tipping option, it is possible that the identified relationship could be an artifact of noise in a steady-state system consisting of tipping variations across millions of cab rides. However, as a variety of factors such as month and day of the year fixed effects, including driver fixed effects, are controlled for, the findings may not be a result of model mis-specification, however, we do not completely rule this out.

Future research could also draw on machine learning methodologies to understand complex associations among the included variables in the study. While the purpose of our study was to address a simple question on the sunlight-tipping association, the supervised learning method used here (regression) could be further augmented using unsupervised learning methods to create latent groups of customers to develop a richer understanding of the possible sub-classes in the current data. Machine learning approaches could also be used to understand decision making by the cab drivers based on past tips, ride performance – whether poorer ride performance or lower aggregate tips affect pick-up locations or driving performance. For such analysis, random forest techniques would be particularly helpful in learning behaviors that could increase fares and tips. Such decision tree methodologies could include pick up locations, day and hour, and ride distance and duration characteristics. With the availability of richer data on driver and passengers, methods such as Bayesian Additive Regression would also be particularly useful in identifying key treatment effects from a variety of cab driver and passenger attributes.

Tipping is an important part of the informal economy. Parameters ranging from service quality, service context, and service characteristics have been studied; yet, quantitative aggregation has found limited variation in tipping levels despite variations in perception of service quality [6]. Moving to factors that directly influence the tipping behavior in a NYC Taxicab context, lowers the confounding of service quality and service customization with weather-induced tendency to increase tipping. The findings show that the effects of sunlight on tipping are significant when transitioning from overcast to sunny skies.

Acknowledgement

We thank Michael Lynn and Michael Hicks for their comments on earlier versions of the manuscript. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

References

1. Azar OH. What sustains social norms and how they evolve?: The case of tipping. Journal of Economic Behavior & Organization. 2004;54(1):49-64.

2. Conlin M, Lynn M, O’Donoghue T. The norm of restaurant tipping. Journal of Economic Behavior & Organization. 2003;52(3):297-321.

3. Cunningham MR. Weather, mood, and helping behavior: Quasi experiments with the sunshine samaritan. Journal of Personality and Social Psychology. 1979;37(11):1947-56.

4. Azar OH. Business strategy and the social norm of tipping. Journal of Economic psychology. 2011;32(3):515-25.

5. Lynn M. Service gratuities and tipping: A motivational framework. Journal of Economic Psychology. 2015;46:74-88.

6. Lynn M, McCall M. Gratitude and gratuity: a meta-analysis of research on the service-tipping relationship. The Journal of Socio-Economics. 2000;29(2):203-14.

7. Azar OH. The social norm of tipping: does it improve social welfare? Journal of Economics. 2005;85(2):141-73.

8. Persinger MA, Levesque B. Geophysical variables and behavior: XII. The weather matrix accommodates large portions of variance of measured daily mood. Perceptual and motor skills. 1983;57(3):868-70.

9. Cortés K, Duchin R, Sosyura D. Clouded judgment: The role of sentiment in credit origination. Journal of Financial Economics. 2016;121(2):392-413.

10. Hirshleifer D, Shumway T. Good day sunshine: Stock returns and the weather. The Journal of Finance. 2003;58(3):1009-32.

11. De Silva DG, Pownall RA, Wolk L. Does the sun ‘shine’on art prices? Journal of Economic Behavior & Organization. 2012;82(1):167-78.

12. Busse MR, Pope DG, Pope JC, Silva-Risso J. The psychological effect of weather on car purchases. The Quarterly Journal of Economics. 2014;130(1):371-414.

13. Conlin M, O’Donoghue T, Vogelsang TJ. Projection bias in catalog orders. The American Economic Review. 2007;97(4):1217-49.

14. Lee JJ, Gino F, Staats BR. Rainmakers: Why bad weather means good productivity. Journal of Applied Psychology. 2014;99(3):504-13.

15. DeHaan E, Madsen J, Piotroski JD. Do Weather‐Induced Moods Affect the Processing of Earnings News? Journal of Accounting Research. 2017.

16. Boubekri M, Hull RB, Boyer LL. Impact of window size and sunlight penetration on office workers’ mood and satisfaction a novel way of assessing sunlight. Environment and Behavior. 1991;23(4):474-93.

17. Denissen JJ, Butalid L, Penke L, Van Aken MA. The effects of weather on daily mood: A multilevel approach. Emotion. 2008;8(5):662-7.

18. Rind B, Strohmetz D. Effect of beliefs about future weather conditions on restaurant tipping. Journal of Applied Social Psychology. 2001;31(10):2160-4.

19. Rind B. Effect of beliefs about weather conditions on tipping. Journal of Applied Social Psychology. 1996;26(2):137-47.

20. Flynn S, Greenberg AE. Does Weather Actually Affect Tipping? An Empirical Analysis of Time‐Series Data1. Journal of Applied Social Psychology. 2012;42(3):702-16.

21. Lynn M, McCall M. Beyond gratitude and gratuity: A meta-analytic review of the predictors of restaurant tipping. 2016.

22. Schotter A. The economics of tipping and gratuities: An essay in institution assisted micro-economics. 1979.

23. Crusco AH, Wetzel CG. The Midas touch the effects of interpersonal touch on restaurant tipping. Personality and Social Psychology Bulletin. 1984;10(4):512-7.

24. Garrity K, Degelman D. Effect of Server Introduction on Restaurant Tipping1. Journal of Applied Social Psychology. 1990;20(2):168-72.

25. Sheahan M, Smith P. Deviance and marginal occupations: the case of taxi drivers. Deviant Behavior. 2003;24(5):449-66.

26. NYC Taxicab Factbook 2014. NYC Taxi and Limousine Commission. New York City: 2014.

27. NYC Taxicab Factbook 2006. The New York City Taxicab Fact Book. Brooklyn, NY: Schaller Consulting, 2006.

28. Lynn M. Tipping in restaurants and around the globe: An interdisciplinary review. 2006.

29. Pearl RB. Tipping Practices of American Households: 1984. final report to the Internal Revenue Service under contract TIR. 1985;82(21):82-21.

30. Wright KP, McHill AW, Birks BR, Griffin BR, Rusterholz T, Chinoy ED. Entrainment of the human circadian clock to the natural light-dark cycle. Current Biology. 2013;23(16):1554-8.

31. Lynn M, Grassman A. Restaurant tipping: an examination of three ‘rational’explanations. Journal of Economic Psychology. 1990;11(2):169-81.

32. Carlson M, Charlin V, Miller N. Positive mood and helping behavior: a test of six hypotheses. Journal of Personality and Social Psychology. 1988;55(2):211-29.

33. Holland SJ. Tipping as risk sharing. The Journal of Socio-Economics. 2009;38(4):641-7.

34. McCall M, Belmont HJ. Credit card insignia and restaurant tipping: Evidence for an associative link. Journal of Applied Psychology. 1996;81(5):609.

35. Lynn M. The effects of alcohol consumption on restaurant tipping. Personality and Social Psychology Bulletin. 1988;14(1):87-91.

36. Azar OH. The Social Norm of Tipping: A Review1. Journal of Applied Social Psychology. 2007;37(2):380-402.

37. Lynn M, Zinkhan GM, Harris J. Consumer tipping: A cross-country study. Journal of Consumer Research. 1993;20(3):478-88.

38. Lynn M. Tipping customs and status seeking: a cross-country study. International Journal of Hospitality Management. 1997;16(2):221-4.

39. Shamir B. Between gratitude and gratuity an analysis of tipping. Annals of Tourism Research. 1984;11(1):59-78.

40. Azar OH. Why pay extra? Tipping and the importance of social norms and feelings in economic theory. The Journal of Socio-Economics. 2007;36(2):250-65.

41. Parrett M. Beauty and the feast: Examining the effect of beauty on earnings using restaurant tipping data. Journal of Economic Psychology. 2015;49:34-46.

42. Brewster ZW, Lynn M. Black–white earnings gap among restaurant servers: A replication, extension, and exploration of consumer racial discrimination in tipping. Sociological Inquiry. 2014;84(4):545-69.

43. Guéguen N, Jacob C. Clothing color and tipping: Gentlemen patrons give more tips to waitresses with red clothes. Journal of Hospitality & Tourism Research. 2014;38(2):275-80.

44. Veitch JA, Galasiu AD. The physiological and psychological effects of windows, daylight, and view at home: Review and research agenda. NRC-IRC Research Report RR-325. 2012.

45. Schwarz N, Clore GL. Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states. Journal of personality and social psychology. 1983;45(3):513.

46. Kamstra MJ, Kramer LA, Levi MD. Winter blues: A SAD stock market cycle. American Economic Review. 2003;93(1):324-43.

47. Simonsohn U. Clouds make nerds look good: Field evidence of the impact of incidental factors on decision making. Journal of Behavioral Decision Making. 2007;20(2):143-52.

48. Simonsohn U. Weather to go to college. The Economic Journal. 2010;120(543):270-80.

49. Buijs RM, Kalsbeek A. Hypothalamic integration of central and peripheral clocks. Nature Reviews Neuroscience. 2001;2(7):521-6.

50. Isen AM. Positive affect, cognitive processes, and social behavior. Advances in experimental social psychology. 1987;20:203-53.

51. Dunn EW, Aknin LB, Norton MI. Spending money on others promotes happiness. Science. 2008;319(5870):1687-8.

52. Lynn M. Individual differences in self-attributed motives for tipping: Antecedents, consequences, and implications. International Journal of Hospitality Management. 2009;28(3):432-8.

53. Keysers C. The empathic brain: How the discovery of mirror neurons changes our understanding of human nature: Lulu. com; 2011.

54. Segerstrom SC, Taylor SE, Kemeny ME, Fahey JL. Optimism is associated with mood, coping, and immune change in response to stress. Journal of personality and social psychology. 1998;74(6):1646.

55. Vuchelen J. Consumer sentiment and macroeconomic forecasts. Journal of economic psychology. 2004;25(4):493-506.

56. Augustine AA, Larsen RJ. Affect regulation and temporal discounting: Interactions between primed, state, and trait affect. Emotion. 2011;11(2):403.

57. Brown GW, Cliff MT. Investor sentiment and asset valuation. The Journal of Business. 2005;78(2):405-40.

58. George JM. State or trait: Effects of positive mood on prosocial behaviors at work. Journal of applied Psychology. 1991;76(2):299.

59. Manucia GK, Baumann DJ, Cialdini RB. Mood influences on helping: Direct effects or side effects? Journal of Personality and Social Psychology. 1984;46(2):357.

60. Bénabou R, Tirole J. Incentives and prosocial behavior. The American economic review. 2006;96(5):1652-78.

61. Carlson M, Charlin V, Miller N. Positive mood and helping behavior: a test of six hypotheses. American Psychological Association; 1988.

62. Jacob C, Guéguen N, Boulbry G. Effects of songs with prosocial lyrics on tipping behavior in a restaurant. International Journal of Hospitality Management. 2010;29(4):761-3.

63. Jacob C, Guéguen N, Ardiccioni R, Sénémeaud C. Exposure to altruism quotes and tipping behavior in a restaurant. International Journal of Hospitality Management. 2013;32:299-301.

64. Batson CD, Powell AA. Altruism and prosocial behavior. Handbook of psychology. 2003.

65. Kliger D, Kudryavtsev A. Out of the blue: mood maintenance hypothesis and seasonal effects on investors’ reaction to news. Quantitative Finance. 2014;14(4):629-40.

66. Forgas JP, Goldenberg L, Unkelbach C. Can bad weather improve your memory? An unobtrusive field study of natural mood effects on real-life memory. Journal of Experimental Social Psychology. 2009;45(1):254-7.

67. Handley IM, Lassiter GD, Nickell EF, Herchenroeder LM. Affect and automatic mood maintenance. Journal of Experimental Social Psychology. 2004;40(1):106-12.

68. Haggag K, Paci G. Default Tips. American Economic Journal: Applied Economics. 2014;6(3):1-19.

69. Hockberger PE. A history of ultraviolet photobiology for humans, animals and microorganisms. Photochemistry and photobiology. 2002;76(6):561-79.

70. Kent ST, McClure LA, Judd SE, Howard VJ, Crosson WL, Al‐Hamdan MZ, et al. Short‐and long‐term sunlight radiation and stroke incidence. Annals of neurology. 2013;73(1):32-7.

71. Ayres I, Vars FE, Zakariya N. To insure prejudice: Racial disparities in taxicab tipping. Yale Law Journal. 2005:1613-74.

72. Ho T-H, Chong JK, Xia X. Yellow taxis have fewer accidents than blue taxis because yellow is more visible than blue. Proceedings of the National Academy of Sciences. 2017;114(12):3074-8.

Supporting Information

S1 Table. Pairwise correlations.

S2 Table. Descriptive Statistics (n = 13,820,783 observations across 33,478 drivers).

S1 Table. Pairwise correlations

	1	2	3	4	5	6	7	8	9	10	11	12	13
1. Tip percentage	1
2. Lux category	0.0087	1
3. Snowfall	0.0077	-0.0844	1
4. Rainfall	0.0008	0.0527	0.0851	1
5. Average daily temperature	-0.0096	0.3562	-0.2306	0.0951	1
6. Average daily temperature-squared	-0.0082	0.3453	-0.2031	0.0785	0.9866	1
7. Ride distance	-0.1453	-0.0428	-0.0049	-0.0138	0.009	0.0094	1
8. Ride duration	-0.1837	0.0513	-0.0195	0.0207	0.0389	0.0323	0.7128	1
9. Passenger count	0.0384	-0.0015	0.0014	-0.0016	-0.0005	0.0001	0.0117	0.0123	1
10. Vendor	0.1163	0.0059	0.0021	-0.0013	0.0042	0.0043	0.0138	0.0112	0.3252	1
11. Default tip option used	0.2138	0.0219	0.0027	-0.0007	-0.0074	-0.0063	0.0659	0.1016	-0.0111	-0.061	1
12. Ride during rush hour	-0.0209	-0.1933	-0.0024	0.0058	-0.0088	-0.0093	0.0022	-0.0006	-0.0358	-0.0024	-0.0387	1
13. Weekday	0.0045	0.1714	0.0076	0.0171	-0.018	-0.019	-0.0611	0.0464	-0.0474	-0.0037	-0.0043	0.3843	1
14. Month	-0.0086	0.1406	-0.1705	0.0963	0.6621	0.6268	-0.005	0.043	-0.0044	0.0025	-0.0084	0.0038	0.0202

Notes.

Bold and italicized estimates is not significant at 10% level.

All remaining estimates are significant at 1% level, except bold typeface estimates are significant at 5% level or 10% level.

S2 Table. Descriptive Statistics (n = 13,820,783 observations across 33,478 drivers)

Variable	Description	Mean	Std Dev	Min	Max
Outcome variable
Tip percentage	Ratio of tip received to total fare	19.9495	13.7200	0	3900
Independent Variables
Day-level predictors
Lux Category	Lux category [0 to 9]; lux=0  lux value <1   lux=1  between 1 & 40 lux=2  between 40 & 200 lux=3  200 &400 lux=4  400 & 1000 lux=5  1000 & 2000 lux=6  2000 & 20000 lux=7  20000 & 110000 lux=8  110000 & 120000 lux=9  >120000	7.3899	1.8234	0	9
Snowfall	Snowfall in millimeter	1.4108	11.3087	0	165
Rainfall	Precipitation in tenths of millimeter	40.8984	91.5003	0	584
Average daily temperature	Average daily temperature in degrees Fahrenheit	57.0860	15.9027	12.47	83.48
Average daily temperature—squared	Square of average temperature	3511.7020	1674.7030	155.5009	6968.9110
Cab-level predictor
Vendor	Reference category   Vendor = 1; Competitor = 0	0.5266	0.4993	0	1
Ride-level predictors
Ride duration (minutes)	Ride duration in minutes	12.7692	8.0291	0.0167	180
Ride distance (mile)	Ride distance in miles	2.5655	2.3586	0.01	49.6
Passenger count	Number of passengers in a ride	1.6065	1.2480	0	7
Default tip option used	Whether the passenger used default tip option.	0.5238	0.4994	0	1
Ride during rush hour	Ride occurred during weekdays between 7 am and10 am and between 4 pm and 7 pm	0.2726	0.4453	0	1
Weekday	Ride occurred during the weekday	0.7174	0.4503	0	1

Note. The included fixed effects are driver (driver id),  month of the year, and day of the year. Only a couple of rides had 7 passengers.

---

[1] 120,000 lux: Brightest sunlight; 110,000 lux: Bright sunlight; 20,000 lux:  Shade illuminated by entire clear blue sky, midday; 1,000 – 2,000 lux: Typical overcast day, midday; <200 lux: Extreme of darkest storm clouds, midday; 400 lux: Sunrise or sunset on a clear day (ambient illumination); 40 lux: Fully overcast, sunset/sunrise; <1 lux:  Extreme of darkest storm clouds, sunset/rise. Source: Wikipedia.

[2] Source: http://digital.cityofchicago.org/index.php/chicago-taxi-data-released/