ABSTRACT
We document gender differences in the booking of business air travel among similar workers within a firm. Women pay consistently less per ticket than men after accounting for a large set of covariates. A large proportion of the lower fares paid are explained by women booking earlier. We find that gender differences increase with age but find no deviation from this trend during the childbearing years. We also find that country-level gender differences in reciprocity are associated with the documented gender differences. The documented gender differences have important monetary implications for firms and suggest an important role for workers’ morale.
I. Introduction
Despite robust experimental evidence about differences in the preferences of men and women, less is understood about gender differences within real-world firms. Measuring and comparing the outcomes of individual workers within a firm is challenging. For many firms, data are only available at the plant or firm level. One approach to understanding individual differences involves estimating a production function using plant-level or firm-level output and using the structure to identify gender differences.1 Alternatively, it is possible to directly measure the outcomes of workers in the small fraction of occupations where data are recorded at the individual level, such as lawyers, real estate agents, and salespeople.2
This article takes a complementary approach by studying gender differences in the booking of business air travel. We directly observe worker-specific outcomes and include workers from many occupations, firms, and countries. Our data set contains information about the business travel of around two million unique travelers working in more than 8,000 unique firms, in more than 60 countries, for the year 2014. About 25 percent of the workers in our data set are women. The data allow us to account for a large set of covariates that includes the characteristics of the trips, the employers, and the employees (more than 40,000 fixed effects in our full specification). These covariates explain more than 90 percent of the variation in the price paid for an air travel booking.
We document significant gender differences in the booking of business air travel among similar workers within a firm. Women pay consistently less per ticket than men. The covariates included explain 96 percent of the male–female gap in fare paid. In particular, we find that gender differences in advance booking explain a substantial part of the gender gap in paid fares. Women are 5 percent more likely to book at least two weeks in advance compared to men after accounting for our covariates. Advance booking behavior by women explains 18 U.S. dollars per trip of the male–female fare paid gap or 3.6 (2.3) percent of the median (mean) price of a plane ticket. Although the documented gender gaps are suggestive of gender differences in booking behavior, there may be nonbehavioral explanations (for example, how trips are assigned or accepted). To explore this further, we estimate models investigating heterogeneity in the gender gaps.
We explore heterogeneity by estimating models that include gender interactions with age, length of stay, traveler type, and region of the world. We report four main sets of results: (i) The gender differences increase with age. Interestingly, we do not find any deviation from this trend during the childbearing years. (ii) Overall, the female–male paid fare gap and the female–male gap in days booked in advance increase with the length of stay. (iii) The female–male paid fare gap is flat in the number of trips made per year. (iv) Finally, we find significant variation by world region. Workers in the United States and Europe exhibit the largest gender differences. Differences are smaller in South America, nonsignificant in Australia, and inverted in Asia. Women book tickets that are on average US$10 more expensive in Asia compared to men. The regional heterogeneity suggests that cultural influences may play an important role in understanding gender differences.3 We complement the business travel data with information about gender differences in economic preferences in each country.4 On the one hand, we do not find evidence that gender differences in patience, risk-taking, or altruism correlate with cross-country variation in the paid fare gender gap. On the other hand, we find that gender differences in positive/negative reciprocity and trust are associated with gender differences in the fare paid.
Although not conclusive, these results are consistent with gender differences in behavioral responses given the same set of tasks. For example, if women are less likely to be assigned to, or to accept, short-notice trips, one might expect the effect to be more pronounced during prime childbearing years or for travelers who travel less frequently. We do not find any such relationships, making the behavioral interpretation more compelling. The preference results suggest that men may be more willing to trade the firms’ money for their own utility when they feel they have been treated unfairly. This issue may be exacerbated in a context of incomplete contracts, whereby the firm cannot specify every possible contingency regarding the air bookings performed by its workers, increasing the scope to spend the firms’ money by the employee.5 Experimental or exogenous variation is needed to more conclusively establish a causal link between behaviors/preferences and the gender gaps that we observe.
The literature on gender differences in economic experiments has studied several traits that may help explain our results.6 Women have been documented to be more risk averse than men in the vast majority of studies that select members of the general population (as in, for example, Sunden and Surette 1998; Finucane et al. 2000; Bernasek and Shwiff 2001; Croson and Gneezy 2009; Niederle 2016).7 If women are more risk averse about a price increase or not finding a seat on their preferred flight, they may book earlier. For managers and professional populations like ours, however, gender differences in risk aversion have been found to be small or nonexistent (for example, Masters and Meier 1988; Birley 1989; Johnson and Powell 1994; Atkinson, Baird, and Frye 2003). Several articles document that women are more generous than men in specific contexts. Women have been shown to be more altruistic (for example, Eckel and Grossman 1998; Güth, Schmidt, and Sutter 2007) and more cooperative (for example, Frank, Gilovich, and Regan 1993; Seguino, Stevens, and Lutz 1996; Ortmann and Tichy 1999; Chermak and Krause 2002) than men.8 Women may book earlier flights to save the firm money, even if they do not receive a direct benefit or recognition for doing so. In Subsection IV.B, we do not find evidence that altruism is a significant driver of the gender difference in the cross-country analysis. Our result on altruism is consistent with the economics experimental literature, where there are no robust differences in average contributions in public-good games between men and women (for example, Ledyard 1995; Eckel and Grossman 2008; Croson and Gneezy 2009; Niederle 2016).9
Our article also relates to the literature on gender performance gaps in real-world labor markets. This literature is quite small due to the difficulties of measuring the output of individual workers within firms. There are two articles (Hellerstein, Neumark, and Troske 1999; Gallen 2018) that study gender productivity gaps by estimating production functions using data on value-added and the labor force of firms. These articles estimate the labor input as the sum of different types of labor, including, among other things, gender, race, age/experience, education, and occupation. Hellerstein, Neumark, and Troske (1999) use U.S. survey data on firms from the manufacturing sector. They find a gender productivity gap, where men are more productive than women. Most of the difference is driven by nonmanagerial, nonprofessional, and younger workers. Gallen (2018) uses data on the entire Danish economy and finds that, on the one hand, women with children are less productive than men, and on the other hand, women without children are more productive than men. An alternative approach is to focus on a particular occupation/industry, where individual output can be directly measured. Azmat and Ferrer (2017) study the performance of young lawyers in the United States. They find that male lawyers bill 10 percent more hours and bring in more than twice as much client revenue as female lawyers. Matsa and Miller (2013) study the behavior of firms affected by a change in gender quotas for corporate board seats in Norway. They find that affected firms undertake fewer workforce reductions, leading to increased labor costs and reduced short-term profits. Cook et al. (2020) study the performance of Uber drivers in the United States. They document that the 7 percent gender earnings gap can be explained by experience on the platform, location preference, and preference for driving speed. The goal of most of these studies is to measure the full output of workers and compare the gender productivity gap to the gender wage gap. Although this article does not attempt to explain the gender wage gap, it provides new insights into gender differences within a firm. We observe the air bookings at the worker level for a broad variety of firms, industries, countries, occupations, and employee types within the firm. In addition, we document an association in the gender differences to differences in economic preferences across countries. Studying business travel bookings is also of interest as it is likely not sensitive to biological explanations (for example, physical strength or bearing children) and more sensitive to other sources of gender differences (for example, preferences).
In summary, we make two main contributions. First, we document robust gender differences in the outcomes of working professionals, using a large data set spanning a wide variety of industries, firms, countries, occupations, and employee types within the firm. Women pay consistently less per ticket and book flights earlier than men. A large proportion of the lower fares paid by women can be explained by women booking flights earlier than men. Regardless of whether women end up with different travel assignments than men or behave differently, both findings imply that men and women contribute differently to firm outcomes. Second, we investigate heterogeneity in the observed gender gaps. We find that the gender differences in paid fares increase with the length of stay, are independent of the number of trips made per year, and increase with age with no deviation from the trend during the childbearing years. These results are consistent with women behaving differently. We also document significant variation in the documented gender differences by region of the world. Finally, by complementing the business travel data with information about economic preferences in each country, we report that average cross-country gender differences in positive/negative reciprocity and trust are correlated with the observed gender differences in paid fares.
Section II describes the data. Section III presents the main empirical results. Section IV describes potential mechanisms that may be consistent with our findings. Section V discusses the robustness of our results. Section VI presents the conclusions. Details about the preference data, definitions of variables, additional results, and additional robustness analyses are in the Appendix.
II. Data Description
We combine data from two sources. The primary data contain information about the business travel bookings of workers. The business travel data are an administrative data set obtained from a large multinational travel management company. We complement these data with information about economic preferences in each country. Preference data are obtained from the Global Preference Survey as presented by Falk et al. (2018). Below we describe these sources. We devote more space to the business travel data, which are novel. Details about the preference data are in Falk et al. (2018).
A. Business Travel Data
We collected business travel data from a large multinational travel management company. This company carries out the business travel needs of corporate clients in North America, South America, Europe, Africa, Asia, and Australia. In a given year, this company fulfills tens of millions of transactions across all travel categories (air, hotel, rental car, rail, etc.). The geographical scope comprises more than 45 countries in which this company has wholly owned operations, joint ventures, and minority holdings, plus more than 15 countries in their partner network.
For the analysis, we use a unique administrative data set, which includes information on travelers and their business air bookings in 2014. We observe detailed information about the bookings: the ticket price, dates of travel, origin and destination airports, ticket class, whether or not the flight is direct, the date booked, and the booking location. Travelers perform the booking using the website of the travel management company. The booking website may be customized for the firm where the employee works. Although the list of quotes provided by the travel agency’s website may fulfill some pre-established criteria (for example, cheaper flights being displayed first), all options available for the traveler are displayed for the booking. The information on travelers is anonymous and is based on the information provided by the travelers to the airlines needed to make the booking. It includes the gender and age of the traveler. We also have anonymous identifiers of the firms and the divisions within the firms where the travelers work. In our data set, there are more than 8,000 unique firms and more than 25,000 unique division–firm pairs. We also have information about the position of the employee within the division–firm for some firms.
To obtain the final sample used in our analysis, we applied the following selection criteria:
Only original transactions are included; refunds or ticket modifications are not considered.
Only round-trip tickets are selected.
Only routes with 100 tickets or more are included.
The top 1 percent of the tickets with the highest fares are excluded.
The resulting panel data set has approximately 7.4 million airline transactions corresponding to around two million unique travelers. Based on the information in the data set, we constructed the following variables: length of the trip in days, number of trips per traveler per year, and the number of days in advance that the trip was booked.
Table 1 displays summary statistics for the paid fare, days booked in advance, and the share of bookings made two weeks or more in advance. The fare paid varies considerably as expected, given the heterogeneity in destinations, ticket class, and the number of days booked in advance. The mean paid fare is US$791.24, and the standard deviation is US$1,021.00 (pooling together women and men). The mean paid fare for women is US$713.16, the mean paid fare for men is US$817.12, and the raw gender mean difference in paid fare is US$103.97. There is also substantial variation in the number of days booked in advance, with a mean of 18.65 days and a standard deviation of 21.05 days.
Summary Statistics of Business Travel Data: Dependent Variables
To get a sense of how the distribution of paid fares looks, the top panel in Figure 1 displays a kernel density estimate of the probability distribution function of the paid fare by gender. There is considerable variation in the paid fare, reflecting the wide variety of trips made in different industries, firms, and countries in the data set. The bottom panel in Figure 1 shows that the empirical cumulative distribution function of the paid fare for men first-order stochastically dominates the one for women. This feature indicates that women paid lower fares than men consistently throughout the observed range of paid fares.

Notes: The figure displays the kernel density estimate (Panel A) and empirical cumulative distribution (Panel B) of the paid fare in U.S. dollars by gender. See Online Appendix Section I for details on the kernel density estimation.
Table 2 displays summary statistics of selected covariates in our data set. It can be seen that 25 percent of the trips are booked by female travelers. Although there is considerable variation in the age of the traveler making the booking, 65 percent of the trips are booked by travelers aged 35–54 years old. There is also considerable variation in the number of trips per year made by travelers. The majority of the trips (89 percent) are booked without a connection (that is, “direct” flights) and are booked in the “economy” ticket class (89 percent). In terms of the length of the stay, 13 percent of the round trips last less than 24 hours, 58 percent last more than one day and less than four days, and the remaining 29 percent last five days or more. Regarding the destinations, 63 percent of the trips are domestic (origin and destination airports are within the same country), 25 percent are continental (origin and destination airports are within the same continent), and 13 percent are intercontinental. Finally, the trips originating from North America or the European Union constitute 85 percent of the booked flights.
Summary Statistics of Business Travel Data: Independent Variables
B. Preference Data
We complement the previous data with information about economic preferences in each country. Preference data are obtained from the Global Preference Survey (GPS) as presented by Falk et al. (2018). The GPS is an experimentally validated survey data set of time preference (patience), risk preference (risk-taking), positive and negative reciprocity, altruism, and trust from 80,000 individuals in 76 countries. Falk et al. (2018) standardize each preference measure at the individual level so that, by construction, each preference has a mean of zero and a standard deviation of one in the individual-level world sample. Table 3 from Falk et al. (2018) summarizes the survey items for each preference. See Falk et al. (2018) for a thorough discussion.
Survey Items of the Global Preference Survey
For each preference item in Table 3, we obtain the average gender difference at the country level reported by Falk et al. (2018, their Online Appendix EB). Then we merge the gender difference preferences to the business travel data using the country where the traveler works. See Appendix 2 for details. Panel B in Table 2 displays the summary statistics of the preference data. See Falk et al. (2018) for a detailed description and interpretation.
III. Empirical Results
This section presents our main empirical analysis in two steps. First, we document the female–male paid fare gap. We show that gender differences in advance booking alone explain 80 percent of the residual paid fare gap after accounting for trip, firm, and employee characteristics. Second, we document a robust gender gap in advance booking. The trip, firm, and employee characteristics account for between 34 and 39 percent of the female–male gap in advance booking. In the next section, we discuss potential mechanisms that could explain these gender gaps.
A. Female–Male Paid Fare Gap
We begin by analyzing the female–male paid fare gap for business travel. On average women pay US$103.97 less per ticket than men (Column 1 of Table 4A). The difference in paid fare by women and men could be due to a number of factors that include the characteristics of the trip, employer/firm characteristics, and employee characteristics. We take advantage of our rich data set to develop multiple covariates for each of these factors.
Femal–Male Business Travel Gaps
Table 4A displays the results of several hedonic regressions of the paid fare on a female indicator, trip characteristics, employer characteristics, and employee characteristics.10 First, we estimate a hedonic regression, including the characteristics of the trip. Trip characteristics include interactions of the origin–destination route and ticket class fixed effects, direct flight, length of stay dummy variables, and week of the year fixed effects. Column 2 in Table 4A shows that adding the trip characteristics increases the adjusted R2 from 0.2 percent to 89.6 percent. In Column 3, we add the characteristics of the employer. They include interactions of division and firm fixed effects and country fixed effects, for a total of 23,668 additional fixed effects. The adjusted R2 increases only modestly from 89.6 percent to 90.1 percent. In Column 4, we add the characteristics of the employees, which include age dummy variables, number of trips per traveler dummy variables, and employee type fixed effects.11 Adding the characteristics of the employees does not noticeably change the goodness of fit. The main conclusion from Columns 1–4 in Table 4A is that the characteristics of the trip can account for almost 90 percent of the variance in the fare paid, while additional characteristics of the employer and employee do not add much to the goodness of fit. This result is consistent with prior work in the industrial organization literature.12
We next decompose the female–male paid fare gap following Gelbach (2016). Gelbach (2016) develops a conditional decomposition to account for the role of groups of covariates that may exhibit sequence sensitivity when these groups are added progressively and are intercorrelated. The Gelbach decomposition nests the Oaxaca-Blinder decomposition and, because it is based on estimates from the full specification of the model, is order-invariant. Table 5 displays the Gelbach decomposition of the female–male paid fare gap into the following three components: (i) characteristics of the trips and employers, (ii) characteristics of the employees, and (iii) days booked in advance fixed effects. The latter component captures the advance booking gap between women and men. Columns 1 and 2 in Table 5 compare the unconditional female–male paid fare gap (Column 1 in Table 4A) to the female–male paid fare gap after accounting for the complete set of trip, employer, employee, and advance booking characteristics (Column 5 in Table 4A). Columns 3 and 4 of Table 5 show the amount of the female–male paid fare gap explained by the characteristics in dollars and percentage terms. The full specification explains about 95.7 percent of the raw female–male paid fare gap. Columns 3 and 4 show that the characteristics of the trips and employers explain about 73 percent of the raw female–male paid fare gap, while employee characteristics explain about 6 percent. With the final component, days booked in advance fixed effects, we seek to understand the share of the female–male paid fare gap explained by the relative difference in advance booking between men and women.13 Interestingly, advance booking explains a relatively large fraction, about 17 percent, of the raw female-male paid fare gap. In other words, after accounting for trip, employer, and employee characteristics, advance booking explains about 80 percent of the residual female-male paid fare gap [−18.003/(−4.46 − 18.003)]. To put the results into context, US$18.00 per trip is about 3.6 (2.3) percent of the median (mean) price of a plane ticket in our sample. Similarly, it represents a mean (median) of 2.22 (2.03) percent of the total annual expenditure of the firm on air tickets, or US$12,328 (US$558) per year for the mean (median) firm, in terms of the firms’ flight expenditures.
Gelbach Decomposition of the Female-Male Paid Fare Gap
B. Female–Male Advance Booking Gap
We now report the gender gap in advance booking. In Table 4B, we regress the days booked in advance on the characteristics of the trips, employers, and employees. In the base specification, on average, women book 2.73 days earlier than men (Column 1 of Table 4B). The full specification shows that women book on average 1.81 days earlier than men (Column 4 of Table 4B), after accounting for the characteristics of the trips, employers, and employees. Overall, the included characteristics in the full specification explain about 34 percent [1 – (1.809/2.728) = 0.3369] of the female–male advance booking gap.
Table 4C reports the female–male probability gap for booking two weeks or more in advance. In the base specification (Column 1 of Table 4C), the probability of booking two weeks or more in advance is 9 percent higher for women than for men. In the full specification (Column 4 of Table 4C), women are 5 percent more likely than men to book two weeks or more in advance. The probability of a man booking two weeks or more in advance is 44.2 percent. Thus, the gaps represent a substantial increase. The included characteristics explain 39 percent [1 – (0.053/0.087) = 0.3908] of the female–male advance booking gap, consistent with the results in Table 4B.
IV. Potential Mechanisms
Why do women pay lower fares and book earlier than men in the firm≥ We now discuss potential mechanisms that could explain the observed gender differences. In the subsections below, we report results from two types of interactions with the indicator for females and discuss which mechanisms may be consistent with the correlational evidence and the documented heterogeneity results.
A. Gender Interactions I
1. Age
Table 5 shows that a large fraction (17 percent) of the female–male paid fare gap is explained by responses in advance booking. We argued that such cost differences might represent behavioral responses given the same set of tasks. Call this hypothesis the behavioral-differences hypothesis. An alternative explanation is that women are assigned different tasks—different travel assignments. For instance, if women are less likely to be assigned to, or to accept, short-notice trips, one might expect the effect to be more pronounced during the prime childbearing years. Column 1 in Table 6 investigates this possibility; it displays female interactions with age, using Specification 4 from Table 4A. We find that the female–male paid fare gap increases with age. The gap is US$11.75 for workers less than 25 years old and US$18.89 for workers between 55 and 64 years old. Nevertheless, consistent with the behavioral-differences hypothesis, we do not find any deviation from this trend during the childbearing years. Also consistent with the paid fare gap increasing with age, Table 6 Column 3 shows that the female–male advance booking gap increases with age for workers younger than 65 years old.
Female–Male Business Travel Gaps: Female Interactions (Part I)
2. Length of stay
Next, we explore gender interactions with the length of stay. If the female–male paid fare gap were driven by the task assignment/acceptance, we may see that the paid fare gap is driven by certain kinds of trips. For example, women are given or are more likely to accept longer trips that are assigned earlier. Columns 4 and 6 in Table 6 show that the female–male gap in days booked in advance increases with the length of stay and that the female–male paid fare gap increases with the length of stay for trips spanning less than five days (Table 6, Column 2). Although there is some heterogeneity, women book earlier, cheaper tickets for trips of all durations. This evidence is also consistent with the behavioral-differences hypothesis.
3. Trips per year
We now explore female interactions with the number of trips made per year. If the female–male paid fare gap were driven by task assignment/acceptance, the fare paid gap may be larger among less frequent travelers. On the contrary, Column 1 in Table 7 shows that the female–male paid fare gap is essentially flat in the number of trips made per year, even when the female–male gap in days booked in advance decreases with the number of trips (Table 7, Column 3).
Female–Male Paid Fare Gap: Female Interactions (Part II)
4. Region of the world
Finally, Columns 2, 4, and 6 in Table 7 show significant variation by region of the world in the female–male paid fare gap and in the female–male days booked in advance gap. Workers in North America and Europe exhibit the largest gender differences in both gaps. Paid fare gaps are smaller in South America, nonsignificant in Australia, and inverted in Asia. On average, women book tickets that are US$9 more expensive in Asia compared to men. On average, women book tickets later than men in Australia and the Middle East.
B. Gender Interactions II: Preference Data by Country
In this section, we explore the potential role of gender differences in economic preferences. In the previous section, we found that the female–male paid fare gap differed by geographic region. Regional heterogeneity suggests that cultural influences may play an important role in understanding the documented gender differences. First, we consider how country-level differences in economic preferences may help explain the paid fare gap in Table 8. Table 8 displays the female–preference interactions in the paid fare model using Specification 4 from Table 4A. Column 1 in Table 8 repeats Specification 4 in Table 4A using the sample of countries with preference data.14 Second, we consider how the advance booking gap covaries with economic preferences in Tables 9 and 10.15
Female–Male Paid Fare Gap: Female Interactions with Preference Data
Female–Male Days Booked in Advance Gap: Female Interactions with Preference Data
Female–Male Probability Gap for Booking Two Weeks or More in Advance: Female Interactions with Preference Data
1. Female–male paid fare gap
We first consider positive reciprocity, where someone who has higher reciprocity is someone who is more likely to give a “gift in exchange for help” and “to return a favor.” Column 5 in Table 8 shows that the interaction between female and positive reciprocity is statistically different from zero and negatively correlated with the paid fare. However, the coefficient on female is similar in magnitude to the one in Column 1 and statistically different from zero. This result indicates that although gender differences in positive reciprocity are associated with gender differences in the fare paid, it does not seem to explain the average gender differences in paid fares.
As regards to negative reciprocity, the average female–male difference tends to be negative, where men are more “willing to take revenge and to punish unfair behavior towards self/others.”16 A positive interaction term would imply that countries with larger negative reciprocity differences have larger paid fare gaps. In the context of our empirical setting, the paid fare gap may be due to men being more willing to trade the firms’ money for their own utility if they feel that they have been treated unfairly. This issue may be exacerbated in a context of incomplete contracts, whereby the firm cannot specify every possible contingency regarding the air bookings performed by its workers; it may increase the scope to spend firms’ money by the employee. Consistent with that, Column 6 in Table 8 shows that the interaction between female and negative reciprocity is positive, statistically different from zero at the 5 percent level, and large in magnitude. Interestingly, the coefficient on female in Column 6 in Table 8 is the only one that is not statistically different from zero. Taken together, these results suggest that the paid fare gap may be explained by women being less willing to trade the firms’ money for their own utility compared to men.
Column 7 in Table 8 investigates the interaction with trust, in that “people have only the best intentions,” according to Table 3. The results are mixed. On the one hand, the interaction between female and trust is statistically different from zero. On the other, although the magnitude of the female coefficient is reduced, it is still large in magnitude and statistically different from zero. In other words, trust may explain part of the gender difference in paid fares, but not all of it. Gender differences in trust at the country level are highly correlated with gender differences in negative reciprocity (pairwise correlation coefficient of −0.932 in Table A2). Due to this collinearity, when both coefficients’ interactions are included, neither is statistically significant. The null hypothesis that both are zero is rejected. One explanation for the mixed results may be that trust is partially capturing the effect of negative reciprocity, which has a clearer interpretation in our empirical context. However, we cannot accept or reject this hypothesis with our data.
Lastly, we do not find significant evidence that patience, risk-taking, or altruism play a role explaining the paid fare gap (Columns 2–4 in Table 8).
2. Female–male advance booking gap
Tables 9 and 10 display the female–preference interactions in the days booked in advance model and in a linear probability model of booking at least 14 days in advance, respectively. Column 1 in Tables 9 and 10 repeat Specification 4 in Tables 4B and 4C using the sample of countries that have preference data. Three main results stand out from adding female–preference interactions to the advance booking models. First, the female coefficient is large and statistically significant in all specifications in both models. These results indicate no evidence that the preference differences considered can explain the advance booking gap. Second, the positive reciprocity, negative reciprocity, and trust interactions are statistically different from zero for the probability of booking two weeks or more in advance (Table 10), although none of the interactions are significant for days booked in advance (Table 9).
Finally, taken together, the results in Tables 8, 9, and 10 indicate that the interaction with preferences data may explain the female–male paid fare gap, but not through differences in advance booking. As emphasized above, our data do not allow us to accept or reject the behavioral-differences hypothesis conclusively. Further clarifying the mechanisms at play in this section on gender differences is an avenue of further research.
V. Robustness and Additional Results
We test the robustness of the empirical results in several ways. First, we obtain similar results using linear probability models for booking: (i) one week or more in advance, (ii) three weeks or more in advance, and (iii) four weeks or more in advance. In Online Appendix Section III.A, Tables A12–A23, we report similar results to the ones in Tables 4, 6, 7, and 10 for (i), (ii), and (iii). Second, we repeated the empirical analysis using several subsamples: (iv) the subset of the 25 percent most popular routes, (v) the subset of trips in the United States only, (vi) the subset of trips in the United States only without Thanksgiving week, and (vii) the subsample with all countries without end of the year holiday weeks. We obtained similar results in all cases. The subsample in (iv) addresses a potential concern about gender selection in popular cities (for example, women being less likely to be employed at firms/divisions in the most popular cities). The subsamples in (v)–(vii) address a concern about the gender differences in ticket costs being driven by differences in preferences for travel during holiday weeks (for example, women flying back earlier during the Thanksgiving week in the United States or during the end of the year holiday weeks). We summarize these results in Online Appendix Section III.B in Tables A24–A27, which show similar coefficients as the ones in Specifications 4 and 5 in Table 4A using, respectively, the subsamples in (iv)–(vii). Third, we also repeated the analysis in Section 3 using the subsample (viii) of countries that have preference data (7,011,259 observations in Appendix 2 Table A1) and obtained similar results. Fourth, we obtained similar results using (ix) other specifications for the “days booked in advance fixed effects.” In Online Appendix Section III.C Table A28, we report a summary of these results using a more saturated model that includes a set of 91 dummy variables for the days booked in advance fixed effects (rather than 26 dummy variables in Column 5 in Table 4A), one for each day booked in advance before the departure for the first 90 days and one additional dummy variable for more than 90 days. Fifth, we repeated the analysis clustering the standard errors: (x) at the firm and (xi) at the firm–division level, and we obtained similar results to the ones reported in the main text. This robustness check addresses the concern that employees within firms may travel in teams or to the same event, thus introducing correlation in their booking of business travel. We report a summary of these results in Online Appendix Section III.D in Tables A29–A30 and A31–A32 that repeat Table 4 adjusting the standard errors for 7,783 firm and 23,609 firm–division clusters. Sixth, similar results to the ones in Tables 4 and 5 were obtained (xii) using a log specification for the fare paid. Finally, from a computational perspective, (xiii) we performed the empirical analysis in R and Stata, using the numerical procedure described in Footnote 9, and obtained identical results. We conclude that the implications discussed in the article are robust in the cases examined.
In terms of additional results, we also find that women are: (xiv) less likely to book a flight in first class, business class, or premium economy (Online Appendix Section II.A) although this result is not economically important once the full set of controls are included; (xv) more likely to book a direct flight (Online Appendix Section II.B) and (xvi) slightly less likely to book a flight that spans over a weekend (Online Appendix Section II.C), although this result is not important in magnitude. For completeness, in Online Appendix Section II, we repeated Tables 4, 6, 7, and 10 for (xiv) and (xv).
VI. Concluding Remarks
We documented gender differences in the booking of business air travel for similar workers within a firm. Women pay consistently less per ticket and book flights earlier than men, after accounting for a large set of covariates that include the characteristics of the trips, the employers, and the employees—a total of more than 40,000 fixed effects. A significant and large proportion of the lower fares paid by women is explained by women booking flights earlier than men. We performed a wide range of robustness checks; the implications are robust to these alternative specifications.
We also investigate heterogeneity in the observed gender gaps. Gender differences in paid fare increase with the length of stay, are flat in the number of trips made per year, and increase with age with no deviation from the trend during the childbearing years. These sets of results are consistent with the cost differences representing behavioral responses given the same set of tasks. We also found significant variation by region of the world, suggesting cultural influences may play an important role in understanding these gender differences. Finally, by complementing the business travel data with information about economic preferences in each country, we found that positive/negative reciprocity and trust are correlated with the documented gender differences in paid fares. In particular, negative reciprocity can explain the observed geographic variation in gender differences in paid fares. The observed gender differences in business travel could result in substantial monetary savings for firms. Our findings also suggest a potentially important role of workers’ morale within a firm.
Appendix 1
Definitions of Variables and Fixed Effects
Below we present the definitions of the variables and fixed effects used in the regressions in Section III and Online Appendix Sections II and III. See Table 3 for a summary of the survey items for the preference data.
Paid fare: The price of the flight ticket in U.S. dollars.
Days booked in advance: The number of days booked in advance, as measured by the difference between the day where the booking was made and the day of departure of the flight.
Female: A dummy variable that equals one if the traveler’s gender is female, and zero otherwise.
Direct flight: A dummy variable that equals one if the flight is a direct flight, and zero otherwise. A direct flight is defined as a flight between two destinations with no change in flight numbers nor stops.
Age: The age of the individual who performs the flight in years. In the regressions we use “age dummy variables” using the following six groups for age: (1) “24 or less,” (2) “(24,34],” (3) “(34,44],” (4) “(44,54],” (5) “(54,64],” and (6) “greater than 65.” For each individual, each group represents a dummy variable equal to one if the age of the individual belongs to that group, and zero otherwise.
Length of stay: The length of the trip in days, as measured by the difference between the day of departure and the day of return. In the regressions, we use “length of stay dummy variables” using the following five groups for the length of stay: (1) “less than 1 day (that is less than 24 hours),” (2) “(1,2],” (3) “(2,3],” (4) “(3,4],” and (5) “5 days or more.” For each individual, each group represents a dummy variable equal to one if the length of stay of the individual belongs to that group, and zero otherwise.
Number of trips per traveler: The number of trips per traveler per year. In the regressions, we use “number of trips per traveler dummy variables” using the following four groups for the number of trips per traveler: (1) “5 trips or less,” (2) (5,10],” (3) “(10,15],” (4) “(10,15],” and (5) “16 or more.” For each individual, each group represents a dummy variable equal to one if the number of trips of the individual belongs to that group, and zero otherwise.
Ticket class: The fare basis code (typically referred to as a fare basis) used by the airlines. This is a categorical variable that belongs to one of the following four groups: (1) “First Class,” (2) “Business Class,” (3) “Premium Economy,” and (4) “Economy Class.” In the regressions, we use “ticket class dummy variables,” where for each individual, each group represents a dummy variable equal to one if the ticket class of the individual belongs to that group, and zero otherwise.
Flight type: This is a categorical variable that belongs to one of the following three groups: (1) “Continental,” (2) “Domestic,” and (3) “International.” In the regressions, we use “flight type dummy variables,” where for each trip, each group represents a dummy variable equal to one if the flight type belongs to that group, and zero otherwise.
Region: A categorical variable that records the region of the world where the flight originates. The possible regions are: (1) “Africa,” (2) “Australia,” (3) “Europe,” (4) “Asia,” (5) “Middle East,” (6) “North America,” and (7) “South America.”
Origin–destination route fixed effects: A set of 8,192 dummy variables, each corresponding to the unique origin–destination route in our sample (for example, LAX-ORD is one origin–destination route). The round trip “from airport A to airport B” and “from airport B to airport A” are two different dummy variables.
(Origin–destination route · ticket class) fixed effects: A set of 18,172 dummy variables that result from the interaction of “Origin–destination fixed effects” and the variable “ticket class.”
Week of the year fixed effects: A set of 52 dummy variables, each corresponding to the week of the year when the flight is scheduled.
Country fixed effects: A set of 66 dummy variables, each corresponding to the country of origin of the flight.
Firm fixed effects: A set of 8,067 dummy variables, each corresponding to the firm where the individual works when booking the flight.
Employee type: A set of six dummy variables, with the classification of the employees by their position within the firm where they work.
(Division × Firm) fixed effects: A set of 25,167 dummy variables, each corresponding to the unique division–firm combination (the classification of the divisions are unique to each firm) where the employee works when booking the flight.
Days booked in advance fixed effects: A set of 26 dummy variables, where each of them equals one depending on how many days or weeks in advance the booking was made, defined as follows. A set of 15 dummy variables, one for each of the first 15 days before a flight. A set of ten dummy variables, one for each of the ten weeks following the first 15 days before a flight. An additional dummy variable for a booking that took place 85 days (85 = 15 + 10 × 7) before the flight.
Appendix 2
Details about Preference Data
Preference data are obtained from the Global Preference Survey (GPS) as presented by Falk et al. (2018). For each preference item in Table 3, we obtain the gender difference at the country level reported by Falk et al. (2018, their Online Appendix EB). Then we merge the gender difference preferences to the business travel data, using the country of the traveler.
There are 20, generally small, countries that have business travel data but do not have preference data. The number of observations from these countries in the business travel data is 415,342. Thus, the number of observations drops from 7,426,390 in Table 1 to 7,011,259 in Table A1. We obtained similar results to Section III using the latter sample.
In their Online Appendix EB (Tables 15 and 16), Falk et al. (2018) report gender coefficients of several regressions by country. For each country, they regress the respective preference on a woman indicator (a dummy variable that equals one if the person is a woman and zero otherwise), age and its square, and subjective math skills. Falk et al. (2018) report the coefficients of the woman indicator for each country. Thus, each coefficient is in the same unit as the original preference measure from the GPS. The coefficient represents the mean gender difference by country in the original preference. In other words, a coefficient of 0.1 means that women in a given country report, on average, having 0.1 standard deviations higher in the respective preference compared to men.
Appendix Table A2 shows the pairwise correlation coefficients between the gender differences in preferences of the merged GPS. The correlation between some of these coefficients (for example, between negative reciprocity and trust) is relatively large. For this reason, we do not simultaneously include all gender differences in preference measures in Tables 8–10. See Section III for details.
Summary Statistics of Preference Data
Pairwise Correlation Coefficients between Gender Differences in Preferences
Footnotes
The authors especially thank Katherine Baldiga Coffman and Lucas Coffman for their many helpful suggestions, as well as the anonymous referees. The authors thank Esteban Aucejo, Yana Gallen, Juanna Joensen, Johanna Mollerstrom, and seminar participants at Ohio State for discussions that greatly benefited the work, and they are grateful to Joseph Rossetti for outstanding research assistance and to J.A. Kearns for editing assistance. The usual disclaimers apply. First version: January 2017. The data used in this article are owned by a travel management company and therefore cannot be posted online. The travel management company had the right to review the results before their circulation. The authors hope that interested researchers will understand their code and, if needed, make requests for further analysis (gregory.veramendi{at}econ.lmu.de), which will be accommodated within the limits of the nondisclosure agreement.
Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/jhr-supplementary.html
↵1. See, for example, Hellerstein, Neumark, and Troske (1999, 2002) and Gallen (2018).
↵2. See, for example, Azmat and Ferrer (2017) and Cook et al. (2021) for studies using data on lawyers and Uber drivers, respectively.
↵3. Falk et al. (2018), using an experimentally validated survey data set from 80,000 individuals across 76 countries, report considerable gender differences in preferences. They show that positive reciprocity and altruism are more pronounced among women, while negative reciprocity is weaker among women (see Table 5 in Falk et al. 2018).
↵4. Preference data are obtained from the Global Preference Survey documented in Falk et al. (2018). See Pope and Sydnor (2010) for another example where geographic variation in cultural attitudes and gender stereotypes is used to understand gender disparities in standardized test scores in the United States.
↵5. The impact of incentives on the behavior of employees within firms has been investigated in field experiments by Nagin et al. (2002) and Bandiera, Barankay, and Rasul (2005). See Bandiera, Barankay, and Rasul (2011) for a review of field experiments in firms.
↵6. See Eckel and Grossman (2008), Croson and Gneezy (2009), and Niederle (2016) for comprehensive reviews of the literature examining gender differences in economics experiments. See Bertrand (2011) and Azmat and Petrongolo (2014) for comprehensive reviews of the literature examining the role of experimental findings on gender differences in labor economics.
↵7. This has sometimes been attributed to women experiencing emotions more strongly than men (for example, Harshman and Paivio 1987; Loewenstein et al. 2001) or to the overconfidence of men relative to women about their performance in a task (for example, Niederle and Vesterlund 2007). See also Dohmen et al. (2011).
↵8. These findings, however, do not hold universally (for example, Brown-Kruse and Hummels 1993; Sell and Wilson 1991; Solow and Kirkwood 2002; Ben-Ner, Kong, and Putterman 2004; Bolton and Katok 1995; Ortmann and Tichy 1999). Croson and Gneezy (2009, their Section 3) attribute the variation in the findings in these studies to a “differential sensitivity of men and women to the social conditions in the experiment.” They show evidence that women are more sensitive to the social context of the experiment by looking within and between a large number of studies investigating gender differences in social preferences. Andreoni and Vesterlund (2001) find that women (men) are more altruistic than men (women) when it is relatively expensive (cheap) to give.
↵9. Babcock et al. (2017) study gender differences in a task allocation that everyone prefers to be completed by someone else, such as writing a report or serving on a committee. They show that women are more likely to volunteer than men but find no evidence that the differential is explained by individual characteristics, such as risk and altruism. The result is driven by beliefs about who will perform the task (that is, the belief that women are more likely than men to volunteer).
↵10. All regressions are OLS regressions implemented using the numerical procedure from Gaure (2013). This is an iterative procedure that relies on the Frisch–Waugh–Lovell decomposition theorem (Frisch and Waugh 1933;Lovell 1963, 2008)to avoid the inversion of the matrix of fixed effects. This procedure results in savings of computing time when the number of fixed effects is as large as in our case. The statistical properties of this estimator are the same as the ones of standard OLS (Gaure 2013), whereby one inverts the matrix with all the fixed effects.
↵11. See Appendix 1 for the definitions of the variables and fixed effects.
↵12. For some recent applications, see, for example, Pakes (2003), Erickson and Pakes (2011), and the references there.
↵13. We account for advance booking by including 26 fixed effects for how many days the traveler booked in advance: a set of 15 dummy variables, one for each of the first 15 days prior to a flight; a set of ten dummy variables, one for each of the ten weeks following the first 15 days prior to a flight; and an additional dummy variable for a booking that took place more than 85 days (85 = 15 +10 × 7) before the flight.
↵14. See Appendix 2 for details about the countries without preference data.
↵15. Column 1 in these tables shows the base gender difference in fare paid and advance booking without accounting for gender differences in preferences. Similar results to Table 4 are obtained. Columns 2–7 in Tables 8, 9, and 10 add interactions between female and each preference item from Table 3. We include both the variable female and the interaction between female and the preference because we are interested in both the gender difference in fare paid and advance booking in a country with no gender difference in a preference, that is, the female coefficient, and how the gender difference in fare paid (advance booking) varies with gender differences in preferences, that is the female · preference coefficient.
↵16. For example, Falk et al. (2018) show that negative reciprocity is weaker among women (Table 5).
- Received August 2018.
- Accepted September 2020.
This open access article is distributed under the terms of the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: http://jhr.uwpress.org