The flexibility of the workweek in the United States: evidence from the FIFA World Cup.

Lozano, Fernando A.

I. INTRODUCTION

At any given time, an American worker should benefit from some flexibility in his workweek schedule. There are events whose timing overlaps with the workday and if these events are valued by the worker, consuming them will increase the quality of the worker's leisure. The possibility to reschedule his work hours, and consume a given event, will result in higher worker's utility. Perhaps the worker knows ahead of time when this event will occur and plans accordingly: for example during the National Collegiate Athletic Association (NCAA) tournaments in March, Saint Patrick's Day also in March, or the Presidential Inauguration in January. Alternatively, the worker will equally benefit from some work flexibility in the face of an event that is unpredictable, such as: bad weather, a natural disaster, or a sick child at home. Even though weekly schedule flexibility is important and valued by workers, it is not clear whether workers adjust their work schedule in the face of an event that overlaps with the workday, especially if consuming the event results in higher workers' utility.

While workers' weekly schedule flexibility has remained relatively unexplored, economists have paid considerable attention to the market hours of American workers. For example, Costa (2000) analyzes the evolution of the workday length of the average American male between 1890 and 1991, Coleman and Pencavel (1993a, 1993b) analyze the evolution of hours worked by American workers between 1940 and 1988, and Kuhn and Lozano (2008) analyze the evolution of the length of the workweek of American workers between 1979 and 2006. Similarly, economists have been interested in understanding the tradeoff between leisure and paid labor. Recent research includes Connolly (2008), who shows that weather determines a worker's day-to-day allocation of market hours, as workers work longer hours during rainy days; or Gonzalez-Chapela (2007), who shows that as prices of recreation goods (complements of leisure) increase, American workers' hours of work increase as well. A natural extension is to ask whether workers would decrease their hours of labor in the face of an event that overlaps with the workweek. If workers' value of leisure increases from consuming such event, then it is conceivable that they would choose to consume it, rather than work at the time of the event.

In this paper, I explore the workweek's schedule flexibility of American workers during the last 15 years using the exogenous variation that arises due to the format of FIFA soccer World Cup. (1) That is, unlike events that cater to the American public and may be scheduled at times when Americans reduce their work hours anyway, the World Cup game schedule and host country choice is determined mostly independently from the United States labor market or the preferences of American workers. Further, as each World Cup is hosted by a different country and played every 4 years, the time games are broadcast live in America may or may not overlap with the workday depending on the host country where the World Cup is played. This allows me to estimate the causal effect that the World Cup has on changes in a worker's weekly schedule flexibility from three independent sources of variation: across time--comparing year t with year t + i, i = 1, 2, 3; across space--the host country will determine the time games are broadcast live in the United States; lastly, the third source of variation is the time games are broadcast live in each U.S. time zone, as each match broadcast's scheduled time will differ across different time zones--potentially the same game will overlap with the workday in a time zone within the United States and not in other time zones.

Empirically, I compare deviations in a worker's weekly work hours during the World Cup from his usual work schedule, with the deviations from the usual workweek of a demographically equivalent worker at other times. Importantly, and as argued above, games played in different countries are televised live at different local times, and the times that these games are televised locally will determine whether the timing of the World Cup overlaps with the worker's regular work schedule in each U.S. time zone. For example, in France 1998 most of the games were played at 9:00 p.m. Central European Time; which is 4:00 p.m. on the U.S. East Coast and 1:00 p.m. on the Pacific Coast. In contrast, in Korea-Japan 2002 most of the games took place at 8:30 p.m. Eastern Asia time; this is 7:30 a.m. on the East Coast and 4:30 a.m. on the West Coast. I hypothesize that as the World Cup is played in different host countries, the worker's decision to supply less market hours than in a usual work week varies accordingly to the time games are televised in the local time zone: when the games are televised early in the morning or late in the afternoon, Americans will not reduce their hours of work as much as when games are televised between 9:00 a.m. and 5:00 p.m. locally. In a sense, my strategy is similar to Hamermesh et al. (2008), who analyze the timing and coordination between persons' activities and local television schedules.

My results show that after controlling for observable demographic characteristics, as well as year and monthly fixed effects, American workers reduce their weekly hours of work on average during the World Cup by up to 9 min. This suggests that roughly one out of every ten workers reduces his weekly hours by the time it takes to watch a complete soccer match (90 min + 15 min half time intermission) per week. Most of this change in the worker' s weekly work hours is concentrated among salary paid workers, who reduce on average their hours of work by 28 weekly minutes. Again, this is equivalent to one in three salary paid workers adjusting his hours of work by the time it takes to watch a game, or more likely one in nine salary paid workers adjusts his hours of work by the amount of time it takes to watch three weekly soccer matches. Interestingly, after controlling for demographic characteristics, year and month fixed effects, hourly paid workers do not adjust their hours of work during the soccer World Cup. This difference is significant because the short run opportunity cost of 1 hour worked less among salary paid workers is arguably zero or very small, while for hourly paid workers the short run opportunity cost of 1 hour less of work is the forgone hourly wage. To the extent that salary paid workers are associated with white collar jobs, and hourly paid workers are associated with blue collar jobs (Hamermesh 2002), this result highlights an important source in labor market differences among workers with different pay frequency in the United States.

Finally, it must be noted that the World Cup schedule is publicly available well before the matches are played, and workers with flexible weekly work schedules can plan to reschedule their market hours in order to view the World Cup matches accordingly. In this sense my paper differs from Connolly (2008), where weather is assumed to be an exogenous shock and cannot be predicted ahead of time. Also, note that the emphasis of this paper is on the workweek-following Kuhn and Lozano (2008)--rather than day-to-day (as Costa 2000 or Connolly 2008) or annual hours of work (Coleman and Pencavel 1993a, 1993b). This is important as weekly work schedules represent a different margin of the worker's labor/leisure tradeoff.

II. DATA

In this paper, I use data from the 1994-2007 National Bureau of Economic Research (NBER) Collection of the Current Population Survey Outgoing Rotations Groups (CPS ORG). In order to identify variations in hours of work between households that are surveyed during the World Cup and not, I estimate the difference between the respondent's hours worked last week and his usual hours of work, and I multiply this difference by 60 for ease of interpretation. (2) These two measures are consistent as they refer to the hours of work in the respondent's main job, and the only difference is that usual hours refers to the mode of all workweeks, and last week refers to the hours in the week prior to the CPS survey week, the reference week. (3) A negative difference between last week hours and usual hours means that the hours of work last week were less than the hours in the usual week--a positive difference means that hours in the usual week are less than hours last week. The null hypothesis to test is whether the difference between usual and last week hours is zero, and during the World Cup I expect these differences in hours of work to be negative. Hereafter, I will refer to this difference as the weekly working gap. Importantly, I concentrate on the worker's flexibility to change hours across weeks, and am unable to identify flexibility within weeks or days. Omitting the latter attenuates my results towards zero, as I am failing to capture another dimension of schedule flexibility. Further, I am unable to make any statement about changes in total hours worked over the long run due to the World Cup, or any change in the worker's productivity.

Figures 1 and 2 show the weekly working gap for 48 periods of time between 1994 and 2007. Figure 1 refers to salary paid workers, and Figure 2 refers to hourly paid workers. Each time period is composed as follows: Period 1 contains observations surveyed during February(t), March(t) and April(t); Period 2 contains observations surveyed during May(t), June(t) and July(t); Period 3 contains observations from the August(t), September(t), and October(t) surveys; and Period 4 from the November(t), December(t) and January(t + 1) surveys. (4) Periods when the World Cup is being played are signaled with a clear bar and a dashed line marks all summers. The data in these figures highlight three facts: First, differences in hours of work between last week's hours and a usual week's hours tend to be negative; this is not surprising as workers tend to take days off, holidays, sick leave, and vacation. Figure 1 also shows that American workers do tend to work less hours last week on average during the World Cup than at other periods, but these differences do not seem much greater than differences in other June/July periods. Finally, these data show that the variance in differences between last week and usual hours is greater among salaried paid workers than among hourly paid workers, which is not surprising given that salaried paid workers have more discretion over their hours of work in the short run than hourly paid workers do (Bureau of Labor Statistics 2005).

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

The sample in this paper includes all employed males and females living in the United States surveyed in the CPS between 1994 and 2007; notice that the sample includes both foreign- and native-born workers. During this time four World Cups were played: USA 1994, France 1998, Korea-Japan 2002 and Germany 2006. (5) I restrict teachers and professors from the sample, as they are likely to change their working routine during the summers. I also drop from the sample agricultural workers. To control for outliers, I drop observations whose weekly working gap is greater than the 99th percentile and observations that are smaller than the 1st percentile. I also drop all observations whose hourly wage is smaller than $2.00. All monetary units are in real dollars where the base period is January 1994. Basic summary statistics are presented in Table 1, the first column presents means for observations surveyed in months other than a World Cup month and the second column presents means for observations surveyed in months during the World Cup. The top panel presents means for all observations in the sample, and the bottom panel presents observations for workers who are salary paid. When analyzing the complete sample, there are two important points to note in these raw estimates: first, usual hours worked are 0.19 (12 min) hours greater when the World Cup is at play than during other months, while hours worked last week are shorter during World Cup months by 0.33 h (18 min)--this is most likely because the World Cup is played during summers. Second, the proportion of employed workers who did not work last week is greater during the World Cup months as well and this difference is 1.4 percentage points. If I restrict the sample exclusively to salary paid workers, usual hours of work are indistinguishable between World Cup and non-World Cup surveys, but the difference in hours worked last week is almost an hour worked shorter during the World Cup months. Importantly, the proportion of employed workers reporting zero hours last week is almost twice as big during World Cup months than at other times.

To further explore these differences among salary paid workers between World Cup and non-World Cup summers, Table 2 presents the reason why last week hours are less than usual hours on each row, and the proportion of workers whose last week hours are less than their usual workweek in the first two columns and the weekly working gap in the last two columns. Notice that the sample in this table is restricted to salary paid workers surveyed exclusively in June or July. The results in this table suggest the following: first, although the same proportion of workers take time off due to holidays or vacations during summers with a World Cup and summers without, the working hours gap is greater during summers when the World Cup is played. Second, a greater proportion of workers reported less hours last week than in the usual week during World Cup summers due to either illness or labor conditions (slack labor, change of job, labor disputes). This result is similar to Skogman-Thoursie's (2004), who shows that workers in Sweden are more likely to report sick on the Monday after the Calgary Winter Olympic Games and the skiing cross country championships.

III. THE FIFA WORLD CUP

Recently, economists have paid some attention to the World Cup. For example, Dohmen et al. (2006) use an opinion survey in Germany to argue that when the performance of the German national team improved in the 2006 World Cup, the economic sentiments and expectations of Germans improved, and each win is associated with even more positive sentiments and expectations. Similarly Edmans et al. (2007) use a cross-section of countries to show that when a country's national team loses in the World Cup, the country's stock market will observe on average a loss of 64 basis points in the following trading day. Hagn and Maennig (2008) use the World Cup as a natural experiment to compare employment in German cities who hosted the 1974 World Cup, and those cities who don't, and they fail to find any evidence of employment effects due to hosting the World Cup. Finally and closer to this paper, Tucker (2008) uses the 2002 World Cup as an exogenous instrument to analyze the benefits of the introduction of a communication technology in a financial institution.

The FIFA Soccer World Cup is played every 4 years. To play in the final round all countries that are members of FIFA must qualify in regional tournaments. Then the final round is played in the host country over the span of a month. Until the 1998 World Cup the final round consisted of 24 teams, and since then the number of teams playing in the final round has increased to 32. For example the 2006 finals were played by 5 African teams, 4 Asian teams (including the Middle East), 1 team from Oceania, 4 teams from Central America, North America or the Caribbean, 4 teams from South America and 14 teams from Europe. These 32 teams are divided into 8 groups of 4. The top two teams in each group (16 total) qualify to the second round, where the tournament takes a format of direct elimination. The second round is followed by the quarterfinals (third round) which consist of every winner of the second round (8 in total). The winners from each quarterfinal play in the semifinals, and finally the grand final and third place match are played a month after the tournament started.

Table 3 presents information on local times and dates in which games were played for each of the four World Cups covered by the sample. Importantly, note that even within World Cup years there is variation in the calendar dates that the World Cup is played, and hence the CPS will observe different stages of the World Cup in different years. In particular, USA 1994 was played from June 17 to July 17, and two CPS surveys capture World Cup games in these days. The June survey, whose reference week refers to the days 12-18, includes the inauguration and first round games and the July survey includes the semifinals, the third place game and the grand final. France 1998 was played from June 10 to July 12 and also includes observations on the June survey when the first round was played (reference week June 7-June 13) and the July survey as the tournament's final took place on Sunday of the reference week (July 12-July 18). In contrast, Korea-Japan was played from May 31 to June 30, and the June survey's reference week includes 23 first round games. Germany 2006 was played between June 9 and July 9, and the June survey includes first round games from the week of June 11-June 17 plus the final match which was played on Sunday of the July survey's reference week.

It is worth noting that during the time span this paper covers soccer has gained in popularity in the United States. The New York Times (Sandomil, 2006) reports that the World Cup final match in 2006 reached 16.9 million viewers, and 11.9 of them saw it on ABC while 5 million saw it on Spanish speaking Univision. This is a 152% increment increase from the 2002 World Cup and 31% increment increase from 1998. Also, viewership for the 2006 final match was higher than that year's National Basketball Association (NBA) finals and almost on par with the NCAA tournament. But not only the final match saw increased viewership: the average viewership for for each game televised on ABC, ESPN and ESPN2 was 1.7 million, 2.3 million and 1.1 million respectively. (6) For the 2006 World Cup 17 out of 20 ESPN's match telecasts reached a rating of 1.0 or better, and 7 reached 2.0 or better. In the 1998 World Cup only 7 match telecasts reached a rating of 1.0 and only 1 reached a rating of 2.0 or better.

The identification strategy on this paper assumes that FIFA's decision of where and when the World Cup is played is independent from the U.S. labor market. If the choice of host country is done to maximize television viewership in the United States then using variation in the time at which games are televised in the United States fails to identify workers' decision between market hours and time spent watching the World Cup. (7) If such is the case, then it is conceivable that the World Cup games are scheduled at times that maximize viewership, and in the absence of those games American workers will consume other types of leisure anyway, and observed decreases of hours of work during the World Cup will be spurious. This seems unlikely: First, the World Cup's host country is chosen 7 years in advance of each tournament by FIFA's executive committee. The host country is chosen by a single transferable vote system, and each candidate country must fulfill the requisite of not belonging to the regional federation that hosted any of the previous two World Cups. But even if the decision by FIFA's executive committee is made by forecasting the best time to maximize U.S. viewership, variation across different local time zones across different World Cups will identify the effect of the World Cup on hours of work as the World Cup will overlap with the workday in some parts of the United States and not in other parts. Further, given the length of 1 month that the World Cup lasts, I am assured that the CPS monthly survey will contain observations which were surveyed during a World Cup every 4 years.

IV. EMPIRICAL STRATEGY

The empirical strategy in this paper consists of comparing the difference between hours of work last week and usual hours of work--heuristically this is similar to a matched pairs estimation where we observe for the same observation's hours during the World Cup (hours last week) and hours at other times (usual hours of work). Specifically I estimate the following equation

(1) [G.sub.iyt] = [gamma] W [C.sub.iyt] + [x'.sub.iyt] [theta] + [[epsilon].sub.iyt]

where [G.sub.iyt] represents the gap between last week and usual hours, W [C.sub.iyt] is an indicator variable that takes a value of one during a World Cup month, and zero otherwise; [x'.sub.iyt] is a vector of demographic characteristics that may or may not vary with time (age, age squared, education, state, year dummies and occupation fixed effects). The subscript i represents each worker, y represents each year and t represents each month. The parameter of interest is [gamma] which represents the change in hours of work during the World Cup, [gamma] < 0 means that hours last week are shorter than usual hours in the main job. Equation (1) will estimate y consistently as long as cov (W [C.sub.iyt], [[epsilon].sub.iyt]) = 0 which is an implausible assumption. As suggested in Figure 1, it is quite possible that hours of work are lower during the World Cup because this event takes place during the summer, and hours of work decrease during summers anyway even in the absence of the World Cup. Alternatively it may be that hours of work are shorter during the World Cup because of some idiosyncratic macroeconomic phenomenon during 1994, 1998, 2002 and 2006--a time variant characteristic. To control for this I decompose [[epsilon].sub.iyt] = [[mu].sub.t] + [v.sub.y] + [[upsilon].sub.iyt] where [v.sub.y] is a year specific component, [[mu].sub.t] is a month specific component and [[upsilon].sub.iyt] is a random variable assumed to have mean zero. Estimating Equation (1) with year and month fixed effects estimates y consistently as long as cop(W [C.sub.iyt], [[upsilon].sub.iyt]) = 0. (8)

An alternative specification, that relaxes the time invariance assumption, is to take advantage of the scheduled times the World Cup is played. As mentioned above, the World Cup is played in a different country every 4 years, which generates variation in the time high profile games are broadcast in the United States. For example, during the USA '94 World Cup most games were played at 4:35 p.m. in the East Coast, which is 1:35 p.m. in the Pacific Coast. On the other hand, during the Korea-Japan '02 World Cup games that were played at 8:30 p.m. Asian Standard Time were televised at 7:30 a.m. on the United States' East Coast, and at 4:30 a.m. in the Pacific Coast. As argued in the previous section, the variation in the choice of host country and therefore times games are televised in the United States is assumed to be exogenous because FIFA's Executive Committee chooses the country where the World Cup will be played arbitrarily. To take advantage of this variation I estimate the following equation:

(2) [G.sub.ijyt] [[beta].sub.1] [T1.sub.iyt] + [[beta].sub.2] [T2.sub.iyt] + [[beta].sub.3] [T3.sub.iyt] + [x'.sub.iyt] [theta] + [[epsilon].sub.ijyt]

where [T1.sub.jyt] equals one if most of the high profile games during the CPS reference week were televised in region j between 12 a.m. and 6 a.m. and zero otherwise, [T2.sub.jyt] takes a value of one if most of the high profile games during the CPS reference week were televised in region j between 6 a.m. and 12 p.m., and [T3.sub.jyt] takes a value of one if most of the high profile games during the CPS reference week were between 12 p.m. and 6 p.m. in region j. Again, for this specification I use the time when most of the high profile games were played during the reference week of the World Cup month and the time is marked with a star in Table 3. Under this strategy, note that none of the games were played between 6 p.m. and 12 a.m. in the different U.S. local times, and the omitted category in Equation (2) is all observations surveyed in months when the World Cup is not played.

A third specification is:

(3) (2) [G.sub.ijyt] [[delta].sub.1] [M1.sub.jyt] + [[delta].sub.2] [M2.sub.jyt] + [[delta].sub.3] [M3.sub.jyt] + [[delta].sub.4] [M4.sub.jyt] + + [x'.sub.iyt] [theta] + [[epsilon].sub.ijyt]

where [M1.sub.jyt] is the number of minutes World Cup games were broadcast live between 12 a.m. and 6 a.m. in region j's time zone during year y. Similarly [M2.sub.jyt] is the number of minutes games were broadcast live in region j between 6 a.m. and 12 p.m., [M3.sub.jyt] is the number of minutes games were broadcast live in region j live between 12 p.m. and 6 p.m., and [M4.sub.jyt] is the number of minutes games were broadcast live in region j. The estimates of [[delta].sub.1]-[[delta].sub.4] are relative to changes in the hours of work during non-World Cup periods conditional on the variables in vector x.

Importantly, as it may be possible that workers show up in their workplace during the World Cup, and they may take a break during the workday to follow the matches but fail to report different hours of work in the data, my results are likely to estimate a lower bound in these differences of the change in hours of work during the World Cup. (9)

V. RESULTS

Unconditional estimates of Equations (1)-(3) are presented in Table 4, the top panel shows estimates from Equation (1), the second panel shows estimates from Equation (2), and the bottom panel shows estimates from Equation (3). The first column presents estimates for all workers, second column estimates for hourly paid workers and the third column estimates for salary paid workers only. The first row in each panel represents the difference between actual hours and usual hours in times when the World Cup is not in play. Across all specifications, this number is robust at values of 62 weekly minutes worked less for hourly paid workers and 58 weekly minutes less for salary paid workers. As in Figures 1 and 2, it makes intuitive sense that these differences are negative as hours last week are likely zero (or smaller than the usual weekly hours) sometimes because people take vacations, sick leave, and temporary separations from the job--and as long as the separation is temporary and the job is still the respondent's main job, usual hours will be non-zero. The second row in the top panel presents unconditional estimates of [gamma] for Equation (1), suggesting that all workers reduce work by 31 min/week during the World Cup before any type of controls are added, hourly paid workers reduce their paid work during the World Cup by 20 min/week, and salary paid workers decrease their market hours by almost 50 min/week.

The estimates for [[beta].sub.1], [[beta].sub.2] and [[beta].sub.3] in Equation (2) are presented in the middle panel of Table 4. These estimates suggest that all American workers do not reduce their hours of work when games are played between 12 a.m.

and 6 a.m., they do reduce their hours by 28 min/week when games are played between 6 a.m. and 12 p.m., and by 33 min/week when games are played between 12 p.m. and 6 p.m.

The magnitude of the estimates for salary paid workers suggests that any differences in market labor during the World Cup are greatest among this group: between 6 a.m. and 12 p.m.

salary paid workers reduce their weekly hours by 43 min, while hourly paid workers reduce their weekly hours by 20 min. When games are between 12 p.m. and 6 p.m., salary paid workers reduce their weekly hours by 54 min, and hourly paid workers only by 21 min. Finally, rows 2, 3, 4 and 5 in the bottom panel show the estimates of [[delta].sub.1], [[delta].sub.2], [[delta].sub.3] and [[delta].sub.4] when the explanatory variable is the number of minutes a World Cup game was being broadcast at different time zones. Note that before controlling for other demographic characteristics, and concentrating on the salary paid sample, each minute a game is being broadcast suggests reductions of the hours gap by 0.05 min, 0.02 min and 0.26 min at 6 a.m.-12 p.m., 12 p.m.-6 p.m., and 6 p.m.--12 a.m. respectively. The unusually high estimate for 6 p.m.-12 a.m. is due to the United States World Cup in 1994, when the games were played on TV's prime-time and in the month of July, where not working due to vacation is more common. Again, the estimates in Table 4 are before any demographic controls are added, and due to the fact that the World Cup is played during summers, these estimates confound the effect that the World Cup has on the weekly work schedule and the effect that vacations have on the weekly work schedules. Perhaps the one striking feature of this table is the difference in the World Cup coefficient's estimates between hourly and salary pay, especially as the last week--usual hours estimate which captures the hourly gap in non-World Cup periods is relatively close across the different pay groups. To compare estimates of changes in the workweek due to the World Cup within each month and controlling for other demographic characteristics, I turn to Tables 5-7.

Table 5 presents ordinary least squares estimates for y when controls are added to Equation (1). The first column presents estimates for all workers and includes controls for education, age, age squared, state, month fixed effects, and year fixed effects. The second column presents estimates for all workers and includes all regressors in column one, plus occupation-year cross product fixed effects. The third and fourth columns present estimates for hourly paid workers, with and without occupation-year fixed effects respectively. The last two columns replicate the same, but for salary paid workers. The estimates across these specifications suggest that even after controlling for month and year fixed effects American workers reduce their number of weekly hours of work during the World Cup, and that this is mostly due to salary paid workers. For example, column 2 suggests that after controlling for observable characteristics all American workers reduce their hours of work by an average of 9 weekly minutes during the World Cup. When the sample is restricted to hourly paid workers, the estimates are not statistically different than zero. When the sample is restricted to salary paid workers and include all controls, the change in hours of work during the World Cup is 28 min less. Again, these results highlight the importance of time-pay method on a worker's schedule flexibility, where hourly paid workers do not change their hours of work during the World Cup. This may be because either hourly paid workers' opportunity cost of watching a match is the foregone wage, or because they can reschedule their hours within each week.

Table 6 presents estimates for Equation (2), again using occupation-year fixed effects in the even numbered columns. The results in column 2 suggest that after controlling for demographic characteristics workers do not change their hours of work if the games are between 12 a.m. and 6 a.m., but if games are between 6 a.m. and 12 p.m. workers supply on average around 15 min less per week, and if games are between 12 p.m. and 6 p.m. workers supply on average 8 min less per week. None of these estimates is statistically significantly different than zero at the 5% confidence level, but the last two are at the 10% confidence level. Again, and as in Table 5, there are stark contrasts between hourly and salary paid workers, as the bulk of the changes in weekly hours of work are concentrated among the latter. That is, in column six, where I restrict the sample to salary paid workers, I find that when games are between 6 a.m. and 12 p.m. the hours of work decrease by more than 32 weekly minutes, and when games are between 12 p.m. and 6 p.m. the hours of work decrease by 28 weekly minutes. These estimates do not seem out of line, and make intuitive sense, as they suggest that one out of three salary paid American workers watch a World Cup game every week when the World Cup overlaps with the work day.

Table 7 presents estimates of Equation (3), and they indicate the change in weekly minutes of work during the World Cup for each extra minute a game is played in each time slot. The results suggest that after controlling for demographic characteristics as well as year-occupation fixed effects, weekly minutes of work among all workers decrease by 2/100 for each extra minute of match broadcasted when games are broadcast between 6 a.m. and 12 p.m. If the sample is restricted to salary paid workers then weekly minutes of work during the World Cup decrease by 3/100 of a minute during 6 a.m. and 12 p.m. These estimates are not different from those in Table 5, if one considers that the average World Cup week has approximately 10 games or 15 weekly hours in the 6 a.m.-12 p.m. time slot, then 900 min x 3/100 equals 27 weekly minutes on average. Unlike Table 6, the estimates for matches played between 12 p.m. and 6 p.m. are not statistically significantly different from zero. This difference is mostly due to the definitions of T1, T2, and T3 where most of the games after 6 p.m. in the USA '94 World Cup fell in the 12 p.m.-6 p.m. category, as 4:35 p.m. was the most common match time and hence T2 got a value of one. Also, the point estimates in this table for games played between 6 p.m. and 12 a.m. are big in magnitude, but measured with high standard errors, yet once month fixed effects are added these estimates attenuate significantly from those in Table 4.

How can one reconcile these results with those in Table 3 which shows that during the World Cup salary paid workers are more likely to take longer vacations, take illness or personal days, or not work due to diverse labor issues--such as job changes or temporary layoffs. It is possible that the dimension these results are capturing is that salary paid workers have more flexibility in scheduling their vacation days, their personal days, or even the days when they switch jobs. That is, if a worker values watching the World Cup enough, then this worker may rearrange his schedule to be able to do this. In addition, recall that these estimates include month fixed effects, and are effectively comparing summer of year t with a World Cup with summers in years t + 1, t + 2 and t + 3. That is, the results in the presence of month fixed effects suggest that salary paid workers are more willing to take their vacation and personal days during summers with World Cups than during other summers. (10)

Table 8 estimates Equation (1) for different subgroups (and their complements) of salary paid workers and they include year-occupation fixed effects: males, females, immigrants, hispanics, college graduates, married workers and workers with 35 or less years of age. The point estimates in this Table can be compared with a baseline for all salary paid workers of -28 weekly minutes in column 6 from Table 4. Not surprisingly males reduce their hours of work more than females during the World Cup, immigrants tend to reduce their hours more than native workers, and so do Hispanic workers. Surprisingly, college graduates tend to reduce their hours more than workers with less education; that is, they reduce their hours of work by more than 30 min (again, this is after all controls are added, including month and year fixed effects). Similarly, single or divorced workers reduce their hours more than married workers. Finally, young salary paid workers reduce their hours more than older workers do, by approximately more than 30 min. These differences should be taken with caution and rather as supporting evidence of the results in Tables 5-7, as I am only comparing point estimates. Because of the small samples that result from breaking up the sample it is hard to make any inference across estimates based on the relatively high standard errors.

Finally, it is not clear whether wages should be included in Equations (1) and (2), as wages may determine changes in the hours of work during the World Cup, and preferences for leisure may determine jointly the number of hours of work and wages for a given worker. Furthermore, when using CPS data wages are calculated by dividing usual weekly earnings over usual weekly hours of work, and the denominator of the explanatory variable will also be part of the response variable if wages are added to Equation (1). Nevertheless, and assuming that I can identify the role that wages have on the gap between hours last week and usual hours, I estimate Equation (1) to include up to a quartic term in log wages, plus interactions between the World Cup variables and log wages up to a quartic term. (11) The results are presented in Figure 3 which shows the estimated gap for Equation (1) for salary and hourly paid workers, and as in Table 4 the bulk of the differences are concentrated among salary paid workers. Importantly, as we move across the distribution of wages from lowest earners to highest earners the difference in the weekly hours gap between observations surveyed during the World Cup and all other observations attenuates. That is, salary paid American workers at the bottom of the distribution work on average 1 hour less during the World Cup, and this difference attenuates to 30 min for salary workers in the second quantile, and it continues diminishing as we move to the fight of the distribution. (12)

[FIGURE 3 OMITTED]

VI. DISCUSSION

in this section, I do different robustness checks to test whether the negative relationship between hours of work and the FIFA World Cup that the estimates discussed above indicate is spurious, as my estimates may be capturing some other event or phenomenon that is unobserved in the data. The first exercise I do consists of extending Equation (1) to add a dummy variable for the months before and after each World Cup. Specifically, I estimate the following equation for salaried workers exclusively:

(4) [G.su.iyt] = [[lambda].sub.B] [B.sub.iyt] + [[lambda].sub.WC] W [C.sub.iyt] + [[lambda].sub.A] [A.sub.iyt] + [x'.sub.iyt] [theta] + [[epsilon].sub.iyt]

where [B.sub.iyt] represents a dummy variable that takes a value of one if the observation was surveyed in the month before the World Cup started, and [A.sub.iyt] represents a dummy variable that takes a value of 1 if the observation was surveyed in the month after the World Cup. If my results above are capturing some unobserved phenomenon then it may be that [[lambda].sub.B] [not equal to] 0, [[lambda].sub.WC] [not equal to] 0 and [[lambda].sub.A] [not equal to] 0. Note that if salaried workers have a contract with their employers that specifies a fixed number of hours of work in a set period of time, as suggested by the results in Connolly (2008), then it is possible that due to the World Cup [[lambda].sub.B] > 0, [[lambda].sub.A] > 0, or [[lambda].sub.B] + [[lambda].sub.A] > 0. If this is the case, then it must be that reductions of hours of work during the World Cup will be accompanied by increases in hours of work in the periods before or after the World Cup. This last point is difficult to prove, as inter-temporal substitution of hours of work is more likely within a week or even within a month, and not across months. Estimates for Equation (4) are presented in the top panel of Table 9, the first column presents unconditional estimates and the second column includes estimates when all demographic controls are added. Importantly, the estimates of [[lambda].sub.B] and [[lambda].sub.A] are zero, while the estimate of [[lambda].sub.wc] is similar to the estimate of [gamma] in Table 4. Similarly, I estimate whether changes in labor supply during the World Cup are compensated with changes in the hours of work in the months after and before the World Cup across different times when the World Cup is played. To do so I estimate the following extension of Equation (2) for salaried workers:

(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The interpretation of Equation (6) is analogous to Equation (2) and< [BT1.sub.iyt] takes a value of one if the observation was surveyed during the month before a World Cup, in a time zone where most of the matches were played between 12 a.m. and 6 a.m. local time; [WCT1.sub.iyt] takes a value of one if the observation was surveyed during the month when the World Cup was played, in a time zone where most of the games were between 12 a.m. and 6 a.m. local time; [AT1.sub.iyt] takes a value of one if the observation was surveyed during the month after a World Cup, in a time zone where most of the matches were played between 12 a.m. and 6 a.m. local time. Similarly [BT2.sub.iyt], [WCT2.sub.iyt], [AT2.sub.iyt] are defined for observations surveyed during the World Cup living in a time zone where games are telecast between 6 a.m. and 12 p.m. local time, and [BT3.sub.iyt], [WCT3.sub.iyt], [AT3.sub.iyt] are defined for observations surveyed during the World Cup living in a time zone where games are telecast between 12 p.m. and 6 p.m. The estimates are presented in the bottom panel of Table 9. Again, the estimates for months before and after the World Cup are zero once demographic controls are added. These results suggest that whatever is driving the reduction in hours of work during the World Cup is absent in the months before or after the World Cup is played.

Another robustness check to test whether the results in this paper are spurious is to test whether randomly chosen month-year combinations can replicate the results above. To do this I randomly choose 4 pairs on month-year combinations--recall there are four World Cup periods in the sample--and I generate the indicator variable NWC-(Not World Cup), which takes a value of one if the observation is surveyed in a chosen year-month combination. I then estimate the following equation analogous to Equation (1):

(6) [G.sub.iyt] = [psi] [NWC.sub.iyt] + [x'.sub.iyt] [theta] + [[epsilon].sub.iyt]

and I replicate this exercise 500 times. That is, I randomly generate 500 placebo World Cup experiments where the month and the year are randomly selected. Figure 4 presents the histogram of the 500 different coefficients [psi] estimated in this exercise. Note that the average mean of these coefficients is zero, further note that only 19% of these coefficients are statistically significantly smaller than zero (one sided test), that only 5% of these coefficients fall to the left of the upper bound of the 95% confidence interval of the estimate of [gamma] in Equation (1), and that less than 2% of these coefficients have a baseline estimate of magnitude as big (in absolute value) as that of the estimate of [gamma] in Equation (1). (13)

[FIGURE 4 OMITTED]

It is possible that the difference between the estimates of the coefficients in Figure 4 and the estimates in Table 5 are due to the fact that the World Cup is played during summers (although in Equations (1)-(3), I control for month fixed effects and this should take care of it). To test this hypothesis, I replicate the above exercise but I constrained the survey month to be either June or July, and randomly choose 4 years (different than the World Cup years), in such a way that the new NWC variable is composed of observations from four different June/July-year combinations. Again, I estimate Equation (6) 200 times and the histogram of estimates is presented in Figure 5. The average mean of these estimates is 2.14 min (not statistically significant different than zero) more during the placebo June/July World Cup observations, than at other times. Only six of these estimates are statistically significantly smaller than zero. None of these placebo estimates reaches the upper limit of the baseline estimate 95% confidence interval. This result is intuitive as what I am doing is comparing June/July from the World Cup, with all other June/July, and is suggestive that whatever is driving the results for July 1994, June 1998, June-July 2002 and June 2006 is different than all other summers.

Finally, soccer in the United States is often associated either with immigrants or with white-collar middle class workers, for example see Markovits and Hellerman (2001). If workers with strong preferences for soccer are more likely to be salary paid, then it is possible that the differences in weekly work schedules between salary paid workers and hourly paid workers during the World Cup are due to stronger tastes for soccer among the salary paid. The American Time Use Survey (ATUS) provides information on the amount of time a worker spends either watching soccer (live, not on TV) or playing soccer during his time use day. In addition, among a detailed set of labor market variables, the ATUS provides information on whether the worker is hourly or salary paid as well. The unconditional averages using ATUS data suggest that there is not a statistical difference between salary and hourly paid workers in the amount of time spent watching or playing soccer. On average, 0.3% of all hourly workers played or watched soccer on any given day, while 0.2% of salary paid workers did. If I restrict observations to those surveyed exclusively on Sundays, the unconditional estimates show that 0.8% of all hourly paid workers played or watched soccer, and 0.4% of all salary paid workers did. This difference is statistically different from zero, yet once I control for the same demographic characteristics as in Equation (1), these differences disappear, perhaps because younger workers are more likely to be hourly paid and more likely to watch a World Cup match. (14) Hence I find no support, using data on whether a worker watched or played soccer in the ATUS, that the results in this paper are being driven by difference in preferences for soccer between hourly and salary paid workers.

[FIGURE 5 OMITTED]

VII. SUMMARY

In today's labor market, workers benefit from having some flexibility in their work schedules, and the ability to periodically supply different hours of work in any given week than during the usual week will result in higher utility for the worker. In this paper, I use the natural variation that arises from the FIFA World Cup to explore the flexibility of the schedule of hours of work of American workers. The FIFA soccer World Cup presents a good natural experiment to test workers' schedule flexibility and take advantage of three exogenous sources of variation: the place where the World Cup is played, which determines that the games will be televised in the United States at different hours during the day, differences in the times games are broadcast in different time zones, and the fact that the World Cup is played every 4 years, which allows for inter-annual comparisons. I hypothesize that American workers are likely to reduce their hours of work when the timing of the games' live broadcast in the United States overlaps with their market labor.

My results suggest that, after controlling for demographic characteristics, year and month fixed effects, American workers supply 9 min of work less during the World Cup, a result that indicates that 1 in 10 American workers watch a weekly game. My results also show that the magnitude of these estimates is greater among salary paid workers, who reduce their hours of work on average by half an hour per week; this indicates that roughly 1 in 3 of all salary paid workers watch a World Cup game per week. Importantly, I show that my estimates for males, immigrants and highly educated workers are greater than for their complement demographic groups. Further, my results do not present any evidence that reductions of hours of work during the World Cup are accompanied by changes in the number of hours of work in the month before or after the World Cup, or that the results are driven by factors that are unobserved in the data. These results highlight the importance of pay frequency in a worker's schedule flexibility, as salaried paid workers have more discretion over the number of hours they work, highlighting an important difference between workers with different pay methods in the labor market.

ABBREVIATIONS

CPS ORG: Current Population Survey Outgoing Rotations Groups

ATUS: American Time Use Survey

NWC: Not World Cup

WC: World Cup

doi: 10.1111/j.1465-7295.2010.00321.x

REFERENCES

Bureau of Labor Statistics. "Workers on Flexible and Shift Schedules in May 2004, USDL 05-1198." 2005, Mimeo.

Coleman, M. T., and J. Pencavel. "Changes in Work Hours of Male Employees, 1940-1988." Industrial & Labor Relations Review, 46(2), 1993a, 262-83.

Coleman, M. T., and Pencavel, J. "Trends in Market Work Behavior of Women Since 1940." Industrial & Labor Relations Review, 46(3), 1993b, 653-76.

Connolly, M. "Here Comes the Rain Again: Weather and the Intertemporal Substitution of Leisure." Journal of Labor Economics, 26(1), 2008, 73-100.

Costa, D. "The Wage and the Length of the Work Day: From the 1890s to 1991 ." Journal of Labor Economics, 18(1), 2000, 156-81.

Dohmen, T., Falk, A., Huffman, D., and Sunde, U. "Seemingly Irrelevant Events Affect Economic Perceptions and Expectations: The FIFA World Cup 2006 as a Natural Experiment." IZA Working paper No. 2275, 2006.

Edmans, A., Garcia, D., and Norli, O. "Sports Sentiment and Stock Returns." Journal of Finance, 62(3), 2007, 1967-98.

Gonzalez-Chapela, J. "On the Price of Recreation Goods as a Determinant of Male Labor Supply." Journal of Labor Economics, 25(4), 2007, 795-24.

Hagn, F., and Maennig, W. "Employment Effects of the Football World Cup 1974 in Germany." Labour Economics, 15(5), 2008, 1062-75.

Hamermesh, D. S. "12 Million Salary Workers Missing." Industrial & Labor Relations Review, 55(4), 2002, 549-66.

Hamermesh, D. S., Myers, C. K., and Pocock, M. L. "Cues for Timing and Coordination: Latitude, Letterman and Longitude." Journal of Labor Economics, 26(2), 2008, 223 -46.

Kuhn, P. J., and Lozano, F. A. "The Expanding Workweek?

Understanding Trends in Long Work Hours Among US Men 1979-2006." Journal of Labor Economics, 26(2), 2008, 311-44.

Markovits, A. S., and Hellerman, S. L. Offside: Soccer and American Exceptionalism. Princeton, NJ: Princeton University Press, 2001.

Sandomil, R. "Cup Ratings are Up, But Fans Deserve Better." New York Times, July 11, 2006.

Skogman-Thoursie, P. "Reporting Sick: Are Sporting Events Contagious." Journal of Applied Econometrics, 19(3), 2004, 809-923.

Tucker, C. "Identifying Formal and Informal Influence in Technology Adoption with Network Externalities." Management Science, 54(12), 2008, 2024-38.

(1.) My definition of workweek schedule flexibility is different from that commonly used by the Bureau of Labor Statistics (BLS). While the traditional definition of schedule flexibility refers to the worker's ability to choose his work start and end dates within a day, in this paper I focus on weekly work schedules, which allows me to include when the workers choose to take vacation days and personal days.

(2.) Multiplying by 60 allows the results to be interpreted in minutes, instead of in fractions of an hour.

(3.) The question for usual hours is HRUSL1: How many hours per week (do/does) (name/you) USUALLY work at (your/his/her) (job?/main job? By main job we mean) (the one at which (you/he/she) usually) ((work/works) the most hours.) and the question for hours last week is HRACT1: ((LAST WEEK/THE WEEK BEFORE LAST)/So, for (LAST WEEK/THE WEEK BEFORE LAST)), how many hours did (you/he/she) ACTUALLY work at (your/his/her) (job?/MAIN job?). One difference between these two questions in that the universe for usual hours all employed people (lfsr94 = 1 or lfsr94 = 2) while for hours last week is only employed currently at work (lfsr94 = 1). For respondents whose labor force status is employed, not at work (lfsr = 2) the hours" last week response is missing as they work zero hours last week and are out of the question's universe. The reason why these workers were absent from work last week are (proportion): vacation (53%); illness (25%); child care problems, family or personal problems, paternity/maternity leave (12%); all other (10%). As these workers worked zero hours last week. t change the value of hours last week accordingly.

(4.) In these figures, I collapse the data into periods of 3 months just to facilitate the visual representation of the data, in the rest of the paper each period of time is one calendar month.

(5.) The results are robust to selecting only males, and robust to the inclusion of the 1994 World Cup or not.

(6.) Notice that 21 matches were played during the CPS reference week that overlaps with the 2006 World Cup. A back-of-the-envelope calculation which assumes a viewership of one and a half million viewers for each game implies total weekly viewership of 31,500,000 non-unique viewers of all matches during the CPS reference week. Assuming that each soccer fan watched on average 4 games during the reference week yields 7,875,000 (= 31,500, 000/4) unique viewers of the World Cup.

(7.) This may be true for the 1994 U.S. World Cup, and the results in this paper are consistent with adding 1994 to the sample or not.

(8.) To account for the CPS's two-step sample design 1 clustered the standard errors on each Primary Sample Unit (PSU).

(9.) While using television ratings for the World Cup is an attractive idea, I am unable to do this as weekly ratings by region are not available from Nielsen Media (conversation with Ms. Carly Litzenberg, Client Service Executive, The Nielsen Company) for the World Cups before 2002, thus using TV ratings will take away either the variation across space if I use all four World Cups, or across time if I use the regional data.

(10.) If I drop from the sample all workers reporting negative gaps because of illness, labor issues, childcare problems, bad weather and other causes my results are still robust although smaller in magnitude; these results are available from the author.

(11.) As usual hours enters in the left side of the equation with a negative sign and in the denominator on the right side of the equation, my intuition is that a higher hourly wage due to relatively low usual hours will bias the estimate of y towards zero as higher earnings will be associated with longer last week worked hours.

(12.) A similar exercise for Equation (2) estimates a similar relationship, but the estimated differences between World Cup and other periods in the 6 a.m.-12 p.m. time slot, and the 12 p.m.-6 p.m. time slot are only statistically significantly different from zero at the 10% confidence level.

(13.) Interestingly, 6 of the 9 estimates with a coefficient of magnitude -28 or smaller include the survey week of January 1996 when the Atlantic Coast lived through one of the worst blizzards ever recorded. While in January 1995 0.4% of salaried paid workers reported an absence for work due to weather and 2.0% did in January 1997, 12.0% did in January 1996.

(14.) Results are available from the author.

FERNANDO A LOZANO *

* I am grateful to Tahir Andrabi, Charlie Brown, Sheldon Danziger, Dan Hamermesh, Peter Kuhn, Greg Price, Gary Smith and Michael Steinberger for their comments and suggestions, to seminar participants at UC Riverside, The University of Michigan Ann Arbor, The University of Wisconsin Madison, Duke University, San Diego State University, SUNY Stony Brook and to participants in the 2008 Meetings of the Society of Labor Economics, the 2008 DITE Conference in Santa Barbara, CA, and the 2008 European Association of Labor Economics Meetings. This research was supported in part by a grant from the Ford Foundation to the Research and Training Program on Poverty and Public Policy at the University of Michigan. All errors remain mine.

Lozano: Department of Economics, Pomona College, 425 N. College Avenue, Claremont, CA 91711. Phone 909-621-8985, Fax 909-621-8576, E-mail fernando.lozano@pomona.edu

TABLE 1
Summary Statistics: Weekly Hours of Work

 (1) (2)
 Not World Cup World Cup

Usual hours 39.031 39.223
 (0.007) (0.045)
Hours last week 38.020 37.688
 (0.010) (0.063)
Not working--employed 0.028 0.042
 (0.000) (0.001)
Proportion salary paid 0.368 0.364
 (0.000) (0.002)
N 1,821,466 46,309
Conditional on being salary paid
Usual hours 42.869 42.863
 (0.011) (0.072)
Hours last week 41.901 41.047
 (0.015) (0.105)
Not working--employed 0.021 0.039
 (0.000) (0.001)
N 663,396 16,671

Sample: All workers in the 1994-2007 CPS ORG.
Standard errors in parentheses.

TABLE 2
Differences in Weekly Hours of Work

 Change in Minutes-
 Proportion Working Less Working Less

 (1) (2) (3) (4)
 Not World Not World
 Cup World Cup Cup World Cup

Vacation-holiday 0.066 0.066 -24.449 -25.806 *
 (0.001) (0.002) (0.166) (0.411)
Childcare problems 0.007 0.007 -22.583 -22.158
 (0.000) (0.001) (0.535) (1.267)
Illness 0.012 0.016 * -20.703 -22.730
 (0.000) (0.001) (0.391) (0.861)
Labor Issues 0.003 0.004 * -15.466 -17.439
 (0.000) (0.001) (0.603) (1.225)
Bad weather 0.001 0.000 -16.952 -11.317
 (0.000) (0.000) (1.491) (2.244)
Other 0.005 0.006 -22.314 -23.024
 (0.000) (0.001) (0.633) (1.507)
N 96,740 16,671

Note: Standard errors in parentheses.

Sample: All salary paid workers surveyed in June or July
in the 1994-2007 CPS ORG.

* World Cup estimate is statistically significant different
from the non World Cup estimate at 95% confidence level.

TABLE 3
Minutes Played in Each Time Zone During
CPS Reference Week

All World Cups

Standard
Time Zone Eastern Central Mountain Pacific

12 a.m.-6 a.m. 1260 1260 1530 1830
6 a.m.-12 p.m. 1435 1945 1705 2190
12 p.m.-6 p.m. 2705 2270 2420 1380
6 p.m.-12 a.m. 270 195 15 270

United States 1994: June 17 to July 17

12 a.m.-6 a.m. 0 0 0 0
6 a.m.-12 p.m. 25 145 175 180
12 p.m.-6 p.m. 695 * 650 * 800 * 810 *
6 p.m.-12 a.m. 270 195 15 0

France 1998: June 10 to July 12

12 a.m.-6 a.m. 0 0 0 30
6 a.m.-12 p.m. 180 360 360 330
12 p.m.-6 p.m. 540 * 360 * 360 * 360 *
6 p.m.-12 a.m. 0 0 0 0

Korea-Japan 2002: May 31 to June 30

12 a.m.-6 a.m. 1260 1260 1530 * 1800 *
6 a.m.-12 p.m. 810 * 810 * 540 0
12 p.m.-6 p.m. 0 0 0 0
6 p.m.-12 a.m. 0 0 0 270

Germany 2006: June 9 to July 9

12 a.m.-6 a.m. 0 0 0 0
6 a.m.-12 p.m. 420 630 630 1680 *
12 p.m.-6 p.m. 1470 * 1260 * 1260 * 210
6 p.m.-12 a.m. 0 0 0 270

Source: FIFA.com.

* Denotes World Cup estimate is statistically significant
different from the non World Cup estimate at 95% confidence
level.

TABLE 4
Ordinary Least Squares (OLS) Minutes of Work
on World Cup Unconditional Estimates

 (1) (2) (3)
 All Workers Hourly Paid Salary Paid

(a) World Cup indicator variable

Last week--usual hours -60.644 * -62.139 * -58.074 *
 (0.459) (0.583) (0.695)
During World Cup -31.437 * -20.274 * -50.892 *
 (3.398) (4.101) (5.194)
N 1,867,775 1,187,708 680,067

(b) Most common time of game indicator variable

Last week--usual hours -60.644 * -62.139 * -58.074 *
 (0.459) (0.583) (0.695)
World Cup match 0-6 -16.490 -9.735 -28.193
 (11.279) (14.656) (19.923)
World Cup match 6-12 -28.792 * -20.156 * -43.258 *
 (7.614) (9.877) (10.481)
World Cup match 12-18 -33.223 * -21.049 * -54.698 *
 (3.988) (4.703) (6.238)
N 1,867,775 1,187,708 680,067

(c) Minutes played at each time variable

Last week--usual hours -60.704 * -62.161 * -58.202 *
 (0.460) (0.583) (0.695)
Minutes 0-6 0.006 0.008 0.004
 (0.006) (0.008) (0.009)
Minutes 6-12 -0.034 * -0.027 * -0.046 *
 (0.008) (0.011) (0.012)
Minutes 12-18 -0.006 0.002 -0.020 *
 (0.005) (0.007) (0.008)
Minutes 18-24 -0.217 * -0.194 * -0.259 *
 (0.034) (0.041) (0.055)
N 1,867,775 1,187,708 680,067

Note: Robust standard errors in parentheses clustered at PSU level.
Sample: All workers in the 1994-2007 CPS ORG.

* Statistically significant different to zero at 5% confidence level.

TABLE 5
OLS Minutes of Work on World Cup Conditional Estimates

 All Workers Hourly Paid

 (1) (2) (3) (4)

World Cup Indicator -8.982 * -8.973 * 2.207 1.968
 (3.732) (3.734) (4.566) (4.570)
Occ-Year Fixed Effects No Yes No Yes
Constant -16.392 * -4.914 -18.930 * -8.015
 (6.159) (5.998) (7.456) (7.214)
[R.sup.2] 0.006 0.007 0.005 0.007
N 1,867,775 1,867.775 1,187,708 1,187,708

 Salary Paid

 (5) (6)

World Cup Indicator -28.388 * -28.115 *
 (5.793) (5.782)
Occ-Year Fixed Effects No Yes
Constant -16.375 0.816
 (11.877) (11.707)
[R.sup.2] 0.008 0.010
N 680,067 680,067

Note: Robust standard errors in parentheses clustered at PSU level.
Sample: All workers in the 1994-2007 CPS ORG. Other regressors
included are month fixed effects, year fixed effects, state fixed
effects, years of education, age, age squared hispanic, black,
immigrant, marital status, and female indicator.

* Statistically significant different than zero at 5% confidence level.

TABLE 6
OLS Minutes of Work on World Cup by Hour of Match

 All Workers Hourly Paid

 (1) (2) (3) (4)

World Cup match 0-6 -4.324 -1.829 3.014 6.178
 (11.491) (11.600) (14.848) (14.898)
World Cup match 6-12 -17.668 * -15.471 -8.168 -5.660
 (7.805) (8.037) (10.115) (10.248)
World Cup match 12-18 -11.036 * -7.774 0.925 3.614
 (4.086) (4.196) (4.910) (5.074)
Occ-Year Fixed Effects No Yes No Yes
Constant -6.770 -4.906 -10.230 -8.017
 (5.839) (5.998) (7.045) (7.215)
[R.sup.2] 0.006 0.007 0.005 0.007
N 1,867,775 1,867,775 1,187,708 1,187,708

 Salary Paid

 (5) (6)

World Cup match 0-6 -16.550 -14.269
 (20.405) (20.583)
World Cup match 6-12 -33.580 * -32.138 *
 (10.781) (11.120)
World Cup match 12-18 -32.118 * -27.973 *
 (6.499) (6.658)
Occ-Year Fixed Effects No Yes
Constant -5.081 0.859
 (11.550) (11.707)
[R.sup.2] 0.008 0.010
N 680,067 680,067

Note: Robust standard errors in parenthesis clustered at PSU level.
Sample: All workers in the 1994-2007 CPS ORG. Other regressors
included are month fixed effects, year fixed effects, state fixed
effects, years of education, age, age squared hispanic, black,
immigrant, marital status and female indicator.

* Statistically significant at 5% confidence level.

TABLE 7
OLS Hours of Work on World Cup by Minutes Played in Survey Week

 All Workers Hourly Paid

 (1) (2) (3) (4)

World Cup match 0-6 0.004 0.004 0.007 0.007
 (0.006) (0.006) (0.008) (0.008)
World Cup match 6-12 -0.023 * -0.022 * -0.017 -0.015
 (0.008) (0.008) (0.011) (0.011)
World Cup match 12-18 0.004 0.005 0.012 0.013
 (0.006) (0.006) (0.007) (0.007)
World Cup match 18-24 -0.079 * -0.050 -0.064 -0.037
 (0.035) (0.036) (0.042) (0.043)
Occ-Year fixed effects No Yes No Yes
Constant -6.835 -4.968 -10.315 -8.081
 (5.838) (5.998) (7.045) (7.213)
[R.sup.2] 0.006 0.007 0.005 0.007
N 1,867,775 1,867,775 1,187,708 1,187,708

 Salary Paid

 (5) (6)

World Cup match 0-6 0.000 -0.001
 (0.009) (0.010)
World Cup match 6-12 -0.035 * -0.033 *
 (0.012) (0.012)
World Cup match 12-18 -0.010 -0.009
 (0.008) (0.009)
World Cup match 18-24 -0.109 * -0.073
 (0.056) (0.057)
Occ-Year fixed effects No Yes
Constant -5.080 0.838
 (11.550) (11.708)
[R.sup.2] 0.007 0.010
N 680,067 680,067

Note: Robust standard errors in parentheses clustered at PSU level.
Sample: All workers in the 1994-2007 CPS ORG. Other regressors
included are month fixed effects, year fixed effects, state fixed
effects, years of education, age, age squared hispanic, black,
immigrant, marital status and female indicator.

* Statistically significant different than zero at 5% confidence level.

TABLE 8
OLS Regressions: Different Subgroups

 (1) (3) (5)
 Male Immigrant Hispanic

World Cup indicator -36.800 * -39.947 * -37.069
 (9.063) (12.154) (18.975)
Constant -43.774 * -45.319 -61.600
 (17.685) (27.902) (112.378)
[R.sup.2] 0.012 0.013 0.023
N 296,330 118,268 47,484

 (2) (4) (6)

 Women US Born Non-Hispanic

World Cup indicator -21.164 * -25.475 * -27.265 *
 (6.943) (6.522) (6.077)
Constant -88.941 * 18.713 4.537
 (14.853) (13.084) (12.224)
[R.sup.2] 0.007 0.011 0.010
N 383,737 561,799 632,583

 (7) (9) (11)
 College Grad Married Young

World Cup indicator -31.467 * -18.233 * -32.472 *
 (8.452) (7.280) (9.441)
Constant 62.882 * -52.976 * 58.622
 (18.744) (17.531) 42.374
[R.sup.2] 0.012 0.012 0.011
N 31.0274 44.4159 229,397

 (8) (10) (12)
 Not College
 Grad Not Married Not Young

World Cup indicator -25.359 * -45.462 * -26.383 *
 (7.771) (9.816) (7.444)
Constant -21.256 -10.721 -58.090
 (14.802) (18.234) (37.426)
[R.sup.2] 0.010 0.009 0.011
N 369,793 235,908 450,670

Note: Robust standard errors in parentheses clustered at PSU level.
Sample: All salary paid workers in the 1994-2007 CPS ORG. Other
regressors included are month fixed effects, year fixed effects,
state fixed effects, years of education, age, age squared
hispanic, black, immigrant, marital status and female indicator.

* Statistically significant different than zero at 5% confidence level.

TABLE 9
OLS Hours of Work on World Cup, Before
and After

 (1) (2)

(a) World Cup Indicator Variable
Month before World Cup 14.305 * 5.397
 (4.126) (4.830)
During World Cup -51.736 * -26.140 *
 (5.201) (5.911)
Month after World Cup -50.442 * 8.108
 (5.293) (5.950)
Controls No Yes
Occ-year fixed effects No Yes
Constant -57.230 * 0.827
 (0.712) (11.707)
[R.sup.2] 0.001 0.010
N 680,067 680,067

(b) Most common time of game indicator
Month before WC match 0-6 5.739 -10.245
 (16.197) (17.142)
World Cup match 0-6 -29.037 -14.023
 (19.924) (20.652)
Month after WC match 0-6 -69.463 * -12.885
 (20.131) (20.701)
Month before WC match 6-12 17.428 * -1.830
 (7.847) (8.721)
World Cup match 6-12 -44.101 * -31.737 *
 (10.481) (11.224)
Month after WC match 6-12 -50.822 * 3.923
 (11.073) (11.939)
Month before WC match 12-18 13.993 * 8.330
 (4.919) (5.528)
World Cup match 12-18 -55.542 * -25.263 *
 (6.245) (6.802)
Month after WC match 12-18 -49.038 * 10.673
 (6.297) (6.792)
Controls No Yes
Occ-Year fixed effects No Yes
Constant -57.230 * 0.786
 (0.712) (11.708)
[R.sup.2] 0.001 0.010
N 680,067 680,067

Note: Robust standard errors in parentheses clustered at
PSU level.

Sample: All salary paid workers in the 1994-2007 CPS
ORG. Other regressors included are month fixed effects. year
fixed effects, state fixed effects, years of education, age,
age squared hispanic, black, immigrant, marital status and
female indicator.

* Statistically significant different than zero at 5%
confidence level.